linux/tools/perf/util/machine.c

2258 lines
54 KiB
C
Raw Normal View History

#include <dirent.h>
#include <errno.h>
#include <inttypes.h>
#include <regex.h>
#include "callchain.h"
#include "debug.h"
#include "event.h"
#include "evsel.h"
#include "hist.h"
#include "machine.h"
#include "map.h"
#include "sort.h"
#include "strlist.h"
#include "thread.h"
#include "vdso.h"
#include <stdbool.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include "unwind.h"
perf callchain: Support handling complete branch stacks as histograms Currently branch stacks can be only shown as edge histograms for individual branches. I never found this display particularly useful. This implements an alternative mode that creates histograms over complete branch traces, instead of individual branches, similar to how normal callgraphs are handled. This is done by putting it in front of the normal callgraph and then using the normal callgraph histogram infrastructure to unify them. This way in complex functions we can understand the control flow that lead to a particular sample, and may even see some control flow in the caller for short functions. Example (simplified, of course for such simple code this is usually not needed), please run this after the whole patchkit is in, as at this point in the patch order there is no --branch-history, that will be added in a patch after this one: tcall.c: volatile a = 10000, b = 100000, c; __attribute__((noinline)) f2() { c = a / b; } __attribute__((noinline)) f1() { f2(); f2(); } main() { int i; for (i = 0; i < 1000000; i++) f1(); } % perf record -b -g ./tsrc/tcall [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ] % perf report --no-children --branch-history ... 54.91% tcall.c:6 [.] f2 tcall | |--65.53%-- f2 tcall.c:5 | | | |--70.83%-- f1 tcall.c:11 | | f1 tcall.c:10 | | main tcall.c:18 | | main tcall.c:18 | | main tcall.c:17 | | main tcall.c:17 | | f1 tcall.c:13 | | f1 tcall.c:13 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:12 | | f1 tcall.c:12 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:11 | | | --29.17%-- f1 tcall.c:12 | f1 tcall.c:12 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:11 | f1 tcall.c:10 | main tcall.c:18 | main tcall.c:18 | main tcall.c:17 | main tcall.c:17 | f1 tcall.c:13 | f1 tcall.c:13 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:12 The default output is unchanged. This is only implemented in perf report, no change to record or anywhere else. This adds the basic code to report: - add a new "branch" option to the -g option parser to enable this mode - when the flag is set include the LBR into the callstack in machine.c. The rest of the history code is unchanged and doesn't know the difference between LBR entry and normal call entry. - detect overlaps with the callchain - remove small loop duplicates in the LBR Current limitations: - The LBR flags (mispredict etc.) are not shown in the history and LBR entries have no special marker. - It would be nice if annotate marked the LBR entries somehow (e.g. with arrows) v2: Various fixes. v3: Merge further patches into this one. Fix white space. v4: Improve manpage. Address review feedback. v5: Rename functions. Better error message without -g. Fix crash without -b. v6: Rebase v7: Rebase. Use NO_ENTRY in memset. v8: Port to latest tip. Move add_callchain_ip to separate patch. Skip initial entries in callchain. Minor cleanups. Signed-off-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1415844328-4884-3-git-send-email-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-13 10:05:20 +08:00
#include "linux/hash.h"
perf tools: Add PERF_RECORD_NAMESPACES to include namespaces related info Introduce a new option to record PERF_RECORD_NAMESPACES events emitted by the kernel when fork, clone, setns or unshare are invoked. And update perf-record documentation with the new option to record namespace events. Committer notes: Combined it with a later patch to allow printing it via 'perf report -D' and be able to test the feature introduced in this patch. Had to move here also perf_ns__name(), that was introduced in another later patch. Also used PRIu64 and PRIx64 to fix the build in some enfironments wrt: util/event.c:1129:39: error: format '%lx' expects argument of type 'long unsigned int', but argument 6 has type 'long long unsigned int' [-Werror=format=] ret += fprintf(fp, "%u/%s: %lu/0x%lx%s", idx ^ Testing it: # perf record --namespaces -a ^C[ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 1.083 MB perf.data (423 samples) ] # # perf report -D <SNIP> 3 2028902078892 0x115140 [0xa0]: PERF_RECORD_NAMESPACES 14783/14783 - nr_namespaces: 7 [0/net: 3/0xf0000081, 1/uts: 3/0xeffffffe, 2/ipc: 3/0xefffffff, 3/pid: 3/0xeffffffc, 4/user: 3/0xeffffffd, 5/mnt: 3/0xf0000000, 6/cgroup: 3/0xeffffffb] 0x1151e0 [0x30]: event: 9 . . ... raw event: size 48 bytes . 0000: 09 00 00 00 02 00 30 00 c4 71 82 68 0c 7f 00 00 ......0..q.h.... . 0010: a9 39 00 00 a9 39 00 00 94 28 fe 63 d8 01 00 00 .9...9...(.c.... . 0020: 03 00 00 00 00 00 00 00 ce c4 02 00 00 00 00 00 ................ <SNIP> NAMESPACES events: 1 <SNIP> # Signed-off-by: Hari Bathini <hbathini@linux.vnet.ibm.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexei Starovoitov <ast@fb.com> Cc: Ananth N Mavinakayanahalli <ananth@linux.vnet.ibm.com> Cc: Aravinda Prasad <aravinda@linux.vnet.ibm.com> Cc: Brendan Gregg <brendan.d.gregg@gmail.com> Cc: Daniel Borkmann <daniel@iogearbox.net> Cc: Eric Biederman <ebiederm@xmission.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sargun Dhillon <sargun@sargun.me> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/148891930386.25309.18412039920746995488.stgit@hbathini.in.ibm.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-03-08 04:41:43 +08:00
#include "asm/bug.h"
#include "sane_ctype.h"
#include <symbol/kallsyms.h>
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock);
static void dsos__init(struct dsos *dsos)
{
INIT_LIST_HEAD(&dsos->head);
dsos->root = RB_ROOT;
pthread_rwlock_init(&dsos->lock, NULL);
}
int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
{
memset(machine, 0, sizeof(*machine));
map_groups__init(&machine->kmaps, machine);
RB_CLEAR_NODE(&machine->rb_node);
dsos__init(&machine->dsos);
machine->threads = RB_ROOT;
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
pthread_rwlock_init(&machine->threads_lock, NULL);
machine->nr_threads = 0;
INIT_LIST_HEAD(&machine->dead_threads);
machine->last_match = NULL;
machine->vdso_info = NULL;
machine->env = NULL;
machine->pid = pid;
machine->id_hdr_size = 0;
machine->kptr_restrict_warned = false;
machine->comm_exec = false;
machine->kernel_start = 0;
memset(machine->vmlinux_maps, 0, sizeof(machine->vmlinux_maps));
machine->root_dir = strdup(root_dir);
if (machine->root_dir == NULL)
return -ENOMEM;
if (pid != HOST_KERNEL_ID) {
struct thread *thread = machine__findnew_thread(machine, -1,
pid);
char comm[64];
if (thread == NULL)
return -ENOMEM;
snprintf(comm, sizeof(comm), "[guest/%d]", pid);
thread__set_comm(thread, comm, 0);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
thread__put(thread);
}
machine->current_tid = NULL;
return 0;
}
struct machine *machine__new_host(void)
{
struct machine *machine = malloc(sizeof(*machine));
if (machine != NULL) {
machine__init(machine, "", HOST_KERNEL_ID);
if (machine__create_kernel_maps(machine) < 0)
goto out_delete;
}
return machine;
out_delete:
free(machine);
return NULL;
}
struct machine *machine__new_kallsyms(void)
{
struct machine *machine = machine__new_host();
/*
* FIXME:
* 1) MAP__FUNCTION will go away when we stop loading separate maps for
* functions and data objects.
* 2) We should switch to machine__load_kallsyms(), i.e. not explicitely
* ask for not using the kcore parsing code, once this one is fixed
* to create a map per module.
*/
if (machine && __machine__load_kallsyms(machine, "/proc/kallsyms", MAP__FUNCTION, true) <= 0) {
machine__delete(machine);
machine = NULL;
}
return machine;
}
static void dsos__purge(struct dsos *dsos)
{
struct dso *pos, *n;
pthread_rwlock_wrlock(&dsos->lock);
list_for_each_entry_safe(pos, n, &dsos->head, node) {
perf symbols: Improve DSO long names lookup speed with rbtree With workload that spawns and destroys many threads and processes, it was found that perf-mem could took a long time to post-process the perf data after the target workload had completed its operation. The performance bottleneck was found to be the lookup and insertion of the new DSO structures (thousands of them in this case). In a dual-socket Ivy-Bridge E7-4890 v2 machine (30-core, 60-thread), the perf profile below shows what perf was doing after the profiled AIM7 shared workload completed: - 83.94% perf libc-2.11.3.so [.] __strcmp_sse42 - __strcmp_sse42 - 99.82% map__new machine__process_mmap_event perf_session_deliver_event perf_session__process_event __perf_session__process_events cmd_record cmd_mem run_builtin main __libc_start_main - 13.17% perf perf [.] __dsos__findnew __dsos__findnew map__new machine__process_mmap_event perf_session_deliver_event perf_session__process_event __perf_session__process_events cmd_record cmd_mem run_builtin main __libc_start_main So about 97% of CPU times were spent in the map__new() function trying to insert new DSO entry into the DSO linked list. The whole post-processing step took about 9 minutes. The DSO structures are currently searched linearly. So the total processing time will be proportional to n^2. To overcome this performance problem, the DSO code is modified to also put the DSO structures in a RB tree sorted by its long name in additional to being in a simple linked list. With this change, the processing time will become proportional to n*log(n) which will be much quicker for large n. However, the short name will still be searched using the old linear searching method. With that patch in place, the same perf-mem post-processing step took less than 30 seconds to complete. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Douglas Hatch <doug.hatch@hp.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Scott J Norton <scott.norton@hp.com> Link: http://lkml.kernel.org/r/1412098575-27863-3-git-send-email-Waiman.Long@hp.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-10-01 01:36:15 +08:00
RB_CLEAR_NODE(&pos->rb_node);
perf symbols: Fix dso lookup by long name and missing buildids Commit 4598a0a6d22f ("perf symbols: Improve DSO long names lookup speed with rbtree") Added a tree to lookup dsos by long name. That tree gets corrupted whenever a dso long name is changed because the tree is not updated. One effect of that is buildid-list does not work with the 'with-hits' option because dso lookup fails and results in two structs for the same dso. The first has the buildid but no hits, the second has hits but no buildid. e.g. Before: $ tools/perf/perf record ls arch certs CREDITS Documentation firmware include ipc Kconfig lib Makefile net REPORTING-BUGS scripts sound usr block COPYING crypto drivers fs init Kbuild kernel MAINTAINERS mm README samples security tools virt [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.012 MB perf.data (11 samples) ] $ tools/perf/perf buildid-list 574da826c66538a8d9060d393a8866289bd06005 [kernel.kallsyms] 30c94dc66a1fe95180c3d68d2b89e576d5ae213c /lib/x86_64-linux-gnu/libc-2.19.so $ tools/perf/perf buildid-list -H 574da826c66538a8d9060d393a8866289bd06005 [kernel.kallsyms] 0000000000000000000000000000000000000000 /lib/x86_64-linux-gnu/libc-2.19.so After: $ tools/perf/perf buildid-list -H 574da826c66538a8d9060d393a8866289bd06005 [kernel.kallsyms] 30c94dc66a1fe95180c3d68d2b89e576d5ae213c /lib/x86_64-linux-gnu/libc-2.19.so The fix is to record the root of the tree on the dso so that dso__set_long_name() can update the tree when the long name changes. Signed-off-by: Adrian Hunter <adrian.hunter@intel.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Douglas Hatch <doug.hatch@hp.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Scott J Norton <scott.norton@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Fixes: 4598a0a6d22f ("perf symbols: Improve DSO long names lookup speed with rbtree") Link: http://lkml.kernel.org/r/1447408112-1920-2-git-send-email-adrian.hunter@intel.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-11-13 17:48:30 +08:00
pos->root = NULL;
list_del_init(&pos->node);
dso__put(pos);
}
pthread_rwlock_unlock(&dsos->lock);
}
static void dsos__exit(struct dsos *dsos)
{
dsos__purge(dsos);
pthread_rwlock_destroy(&dsos->lock);
}
void machine__delete_threads(struct machine *machine)
{
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
struct rb_node *nd;
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
pthread_rwlock_wrlock(&machine->threads_lock);
nd = rb_first(&machine->threads);
while (nd) {
struct thread *t = rb_entry(nd, struct thread, rb_node);
nd = rb_next(nd);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
__machine__remove_thread(machine, t, false);
}
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
pthread_rwlock_unlock(&machine->threads_lock);
}
void machine__exit(struct machine *machine)
{
machine__destroy_kernel_maps(machine);
map_groups__exit(&machine->kmaps);
dsos__exit(&machine->dsos);
machine__exit_vdso(machine);
zfree(&machine->root_dir);
zfree(&machine->current_tid);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
pthread_rwlock_destroy(&machine->threads_lock);
}
void machine__delete(struct machine *machine)
{
if (machine) {
machine__exit(machine);
free(machine);
}
}
void machines__init(struct machines *machines)
{
machine__init(&machines->host, "", HOST_KERNEL_ID);
machines->guests = RB_ROOT;
}
void machines__exit(struct machines *machines)
{
machine__exit(&machines->host);
/* XXX exit guest */
}
struct machine *machines__add(struct machines *machines, pid_t pid,
const char *root_dir)
{
struct rb_node **p = &machines->guests.rb_node;
struct rb_node *parent = NULL;
struct machine *pos, *machine = malloc(sizeof(*machine));
if (machine == NULL)
return NULL;
if (machine__init(machine, root_dir, pid) != 0) {
free(machine);
return NULL;
}
while (*p != NULL) {
parent = *p;
pos = rb_entry(parent, struct machine, rb_node);
if (pid < pos->pid)
p = &(*p)->rb_left;
else
p = &(*p)->rb_right;
}
rb_link_node(&machine->rb_node, parent, p);
rb_insert_color(&machine->rb_node, &machines->guests);
return machine;
}
void machines__set_comm_exec(struct machines *machines, bool comm_exec)
{
struct rb_node *nd;
machines->host.comm_exec = comm_exec;
for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
struct machine *machine = rb_entry(nd, struct machine, rb_node);
machine->comm_exec = comm_exec;
}
}
struct machine *machines__find(struct machines *machines, pid_t pid)
{
struct rb_node **p = &machines->guests.rb_node;
struct rb_node *parent = NULL;
struct machine *machine;
struct machine *default_machine = NULL;
if (pid == HOST_KERNEL_ID)
return &machines->host;
while (*p != NULL) {
parent = *p;
machine = rb_entry(parent, struct machine, rb_node);
if (pid < machine->pid)
p = &(*p)->rb_left;
else if (pid > machine->pid)
p = &(*p)->rb_right;
else
return machine;
if (!machine->pid)
default_machine = machine;
}
return default_machine;
}
struct machine *machines__findnew(struct machines *machines, pid_t pid)
{
char path[PATH_MAX];
const char *root_dir = "";
struct machine *machine = machines__find(machines, pid);
if (machine && (machine->pid == pid))
goto out;
if ((pid != HOST_KERNEL_ID) &&
(pid != DEFAULT_GUEST_KERNEL_ID) &&
(symbol_conf.guestmount)) {
sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
if (access(path, R_OK)) {
static struct strlist *seen;
if (!seen)
seen = strlist__new(NULL, NULL);
if (!strlist__has_entry(seen, path)) {
pr_err("Can't access file %s\n", path);
strlist__add(seen, path);
}
machine = NULL;
goto out;
}
root_dir = path;
}
machine = machines__add(machines, pid, root_dir);
out:
return machine;
}
void machines__process_guests(struct machines *machines,
machine__process_t process, void *data)
{
struct rb_node *nd;
for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
struct machine *pos = rb_entry(nd, struct machine, rb_node);
process(pos, data);
}
}
char *machine__mmap_name(struct machine *machine, char *bf, size_t size)
{
if (machine__is_host(machine))
snprintf(bf, size, "[%s]", "kernel.kallsyms");
else if (machine__is_default_guest(machine))
snprintf(bf, size, "[%s]", "guest.kernel.kallsyms");
else {
snprintf(bf, size, "[%s.%d]", "guest.kernel.kallsyms",
machine->pid);
}
return bf;
}
void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
{
struct rb_node *node;
struct machine *machine;
machines->host.id_hdr_size = id_hdr_size;
for (node = rb_first(&machines->guests); node; node = rb_next(node)) {
machine = rb_entry(node, struct machine, rb_node);
machine->id_hdr_size = id_hdr_size;
}
return;
}
static void machine__update_thread_pid(struct machine *machine,
struct thread *th, pid_t pid)
{
struct thread *leader;
if (pid == th->pid_ || pid == -1 || th->pid_ != -1)
return;
th->pid_ = pid;
if (th->pid_ == th->tid)
return;
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
leader = __machine__findnew_thread(machine, th->pid_, th->pid_);
if (!leader)
goto out_err;
if (!leader->mg)
leader->mg = map_groups__new(machine);
if (!leader->mg)
goto out_err;
if (th->mg == leader->mg)
return;
if (th->mg) {
/*
* Maps are created from MMAP events which provide the pid and
* tid. Consequently there never should be any maps on a thread
* with an unknown pid. Just print an error if there are.
*/
if (!map_groups__empty(th->mg))
pr_err("Discarding thread maps for %d:%d\n",
th->pid_, th->tid);
map_groups__put(th->mg);
}
th->mg = map_groups__get(leader->mg);
out_put:
thread__put(leader);
return;
out_err:
pr_err("Failed to join map groups for %d:%d\n", th->pid_, th->tid);
goto out_put;
}
/*
* Caller must eventually drop thread->refcnt returned with a successful
* lookup/new thread inserted.
*/
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
static struct thread *____machine__findnew_thread(struct machine *machine,
pid_t pid, pid_t tid,
bool create)
{
struct rb_node **p = &machine->threads.rb_node;
struct rb_node *parent = NULL;
struct thread *th;
/*
* Front-end cache - TID lookups come in blocks,
* so most of the time we dont have to look up
* the full rbtree:
*/
th = machine->last_match;
if (th != NULL) {
if (th->tid == tid) {
machine__update_thread_pid(machine, th, pid);
return thread__get(th);
}
machine->last_match = NULL;
}
while (*p != NULL) {
parent = *p;
th = rb_entry(parent, struct thread, rb_node);
if (th->tid == tid) {
machine->last_match = th;
machine__update_thread_pid(machine, th, pid);
return thread__get(th);
}
if (tid < th->tid)
p = &(*p)->rb_left;
else
p = &(*p)->rb_right;
}
if (!create)
return NULL;
th = thread__new(pid, tid);
if (th != NULL) {
rb_link_node(&th->rb_node, parent, p);
rb_insert_color(&th->rb_node, &machine->threads);
/*
* We have to initialize map_groups separately
* after rb tree is updated.
*
* The reason is that we call machine__findnew_thread
* within thread__init_map_groups to find the thread
* leader and that would screwed the rb tree.
*/
if (thread__init_map_groups(th, machine)) {
rb_erase_init(&th->rb_node, &machine->threads);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
RB_CLEAR_NODE(&th->rb_node);
thread__put(th);
return NULL;
}
/*
* It is now in the rbtree, get a ref
*/
thread__get(th);
machine->last_match = th;
++machine->nr_threads;
}
return th;
}
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
{
return ____machine__findnew_thread(machine, pid, tid, true);
}
struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
pid_t tid)
{
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
struct thread *th;
pthread_rwlock_wrlock(&machine->threads_lock);
th = __machine__findnew_thread(machine, pid, tid);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
pthread_rwlock_unlock(&machine->threads_lock);
return th;
}
struct thread *machine__find_thread(struct machine *machine, pid_t pid,
pid_t tid)
{
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
struct thread *th;
pthread_rwlock_rdlock(&machine->threads_lock);
th = ____machine__findnew_thread(machine, pid, tid, false);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
pthread_rwlock_unlock(&machine->threads_lock);
return th;
}
struct comm *machine__thread_exec_comm(struct machine *machine,
struct thread *thread)
{
if (machine->comm_exec)
return thread__exec_comm(thread);
else
return thread__comm(thread);
}
int machine__process_comm_event(struct machine *machine, union perf_event *event,
struct perf_sample *sample)
{
struct thread *thread = machine__findnew_thread(machine,
event->comm.pid,
event->comm.tid);
bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
int err = 0;
if (exec)
machine->comm_exec = true;
if (dump_trace)
perf_event__fprintf_comm(event, stdout);
if (thread == NULL ||
__thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
err = -1;
}
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
thread__put(thread);
return err;
}
perf tools: Add PERF_RECORD_NAMESPACES to include namespaces related info Introduce a new option to record PERF_RECORD_NAMESPACES events emitted by the kernel when fork, clone, setns or unshare are invoked. And update perf-record documentation with the new option to record namespace events. Committer notes: Combined it with a later patch to allow printing it via 'perf report -D' and be able to test the feature introduced in this patch. Had to move here also perf_ns__name(), that was introduced in another later patch. Also used PRIu64 and PRIx64 to fix the build in some enfironments wrt: util/event.c:1129:39: error: format '%lx' expects argument of type 'long unsigned int', but argument 6 has type 'long long unsigned int' [-Werror=format=] ret += fprintf(fp, "%u/%s: %lu/0x%lx%s", idx ^ Testing it: # perf record --namespaces -a ^C[ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 1.083 MB perf.data (423 samples) ] # # perf report -D <SNIP> 3 2028902078892 0x115140 [0xa0]: PERF_RECORD_NAMESPACES 14783/14783 - nr_namespaces: 7 [0/net: 3/0xf0000081, 1/uts: 3/0xeffffffe, 2/ipc: 3/0xefffffff, 3/pid: 3/0xeffffffc, 4/user: 3/0xeffffffd, 5/mnt: 3/0xf0000000, 6/cgroup: 3/0xeffffffb] 0x1151e0 [0x30]: event: 9 . . ... raw event: size 48 bytes . 0000: 09 00 00 00 02 00 30 00 c4 71 82 68 0c 7f 00 00 ......0..q.h.... . 0010: a9 39 00 00 a9 39 00 00 94 28 fe 63 d8 01 00 00 .9...9...(.c.... . 0020: 03 00 00 00 00 00 00 00 ce c4 02 00 00 00 00 00 ................ <SNIP> NAMESPACES events: 1 <SNIP> # Signed-off-by: Hari Bathini <hbathini@linux.vnet.ibm.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexei Starovoitov <ast@fb.com> Cc: Ananth N Mavinakayanahalli <ananth@linux.vnet.ibm.com> Cc: Aravinda Prasad <aravinda@linux.vnet.ibm.com> Cc: Brendan Gregg <brendan.d.gregg@gmail.com> Cc: Daniel Borkmann <daniel@iogearbox.net> Cc: Eric Biederman <ebiederm@xmission.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sargun Dhillon <sargun@sargun.me> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/148891930386.25309.18412039920746995488.stgit@hbathini.in.ibm.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-03-08 04:41:43 +08:00
int machine__process_namespaces_event(struct machine *machine __maybe_unused,
union perf_event *event,
struct perf_sample *sample __maybe_unused)
{
struct thread *thread = machine__findnew_thread(machine,
event->namespaces.pid,
event->namespaces.tid);
int err = 0;
WARN_ONCE(event->namespaces.nr_namespaces > NR_NAMESPACES,
"\nWARNING: kernel seems to support more namespaces than perf"
" tool.\nTry updating the perf tool..\n\n");
WARN_ONCE(event->namespaces.nr_namespaces < NR_NAMESPACES,
"\nWARNING: perf tool seems to support more namespaces than"
" the kernel.\nTry updating the kernel..\n\n");
if (dump_trace)
perf_event__fprintf_namespaces(event, stdout);
if (thread == NULL ||
thread__set_namespaces(thread, sample->time, &event->namespaces)) {
dump_printf("problem processing PERF_RECORD_NAMESPACES, skipping event.\n");
err = -1;
}
thread__put(thread);
return err;
}
int machine__process_lost_event(struct machine *machine __maybe_unused,
union perf_event *event, struct perf_sample *sample __maybe_unused)
{
dump_printf(": id:%" PRIu64 ": lost:%" PRIu64 "\n",
event->lost.id, event->lost.lost);
return 0;
}
perf tools: handle PERF_RECORD_LOST_SAMPLES This patch modifies the perf tool to handle the new RECORD type, PERF_RECORD_LOST_SAMPLES. The number of lost-sample events is stored in .nr_events[PERF_RECORD_LOST_SAMPLES]. The exact number of samples which the kernel dropped is stored in total_lost_samples. When the percentage of dropped samples is greater than 5%, a warning is printed. Here are some examples: Eg 1, Recording different frequently-occurring events is safe with the patch. Only a very low drop rate is associated with such actions. $ perf record -e '{cycles:p,instructions:p}' -c 20003 --no-time ~/tchain ~/tchain $ perf report -D | tail SAMPLE events: 120243 MMAP2 events: 5 LOST_SAMPLES events: 24 FINISHED_ROUND events: 15 cycles:p stats: TOTAL events: 59348 SAMPLE events: 59348 instructions:p stats: TOTAL events: 60895 SAMPLE events: 60895 $ perf report --stdio --group # To display the perf.data header info, please use --header/--header-only options. # # # Total Lost Samples: 24 # # Samples: 120K of event 'anon group { cycles:p, instructions:p }' # Event count (approx.): 24048600000 # # Overhead Command Shared Object Symbol # ................ ........... ................ .................................. # 99.74% 99.86% tchain_edit tchain_edit [.] f3 0.09% 0.02% tchain_edit tchain_edit [.] f2 0.04% 0.00% tchain_edit [kernel.vmlinux] [k] ixgbe_read_reg Eg 2, Recording the same thing multiple times can lead to high drop rate, but it is not a useful configuration. $ perf record -e '{cycles:p,cycles:p}' -c 20003 --no-time ~/tchain Warning: Processed 600592 samples and lost 99.73% samples! [perf record: Woken up 148 times to write data] [perf record: Captured and wrote 36.922 MB perf.data (1206322 samples)] [perf record: Woken up 1 times to write data] [perf record: Captured and wrote 0.121 MB perf.data (1629 samples)] Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: acme@infradead.org Cc: eranian@google.com Link: http://lkml.kernel.org/r/1431285195-14269-9-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-05-11 03:13:15 +08:00
int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
union perf_event *event, struct perf_sample *sample)
{
dump_printf(": id:%" PRIu64 ": lost samples :%" PRIu64 "\n",
sample->id, event->lost_samples.lost);
return 0;
}
static struct dso *machine__findnew_module_dso(struct machine *machine,
struct kmod_path *m,
const char *filename)
{
struct dso *dso;
pthread_rwlock_wrlock(&machine->dsos.lock);
dso = __dsos__find(&machine->dsos, m->name, true);
if (!dso) {
dso = __dsos__addnew(&machine->dsos, m->name);
if (dso == NULL)
goto out_unlock;
dso__set_module_info(dso, m, machine);
dso__set_long_name(dso, strdup(filename), true);
}
dso__get(dso);
out_unlock:
pthread_rwlock_unlock(&machine->dsos.lock);
return dso;
}
int machine__process_aux_event(struct machine *machine __maybe_unused,
union perf_event *event)
{
if (dump_trace)
perf_event__fprintf_aux(event, stdout);
return 0;
}
int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
union perf_event *event)
{
if (dump_trace)
perf_event__fprintf_itrace_start(event, stdout);
return 0;
}
int machine__process_switch_event(struct machine *machine __maybe_unused,
union perf_event *event)
{
if (dump_trace)
perf_event__fprintf_switch(event, stdout);
return 0;
}
perf machine: Adjust dso->long_name for offline module Something unexpected may happen if copy statically linked perf to a production environment: # ./perf probe -m ./mymodule.ko my_func [mymodule] with build id 326ab42550ef3d24944f53c817533728367effeb not found, continuing without symbols Failed to find symbol my_func in /home/wangnan/kmodule/mymodule.ko Error: Failed to add events. # ./perf buildid-cache -a ./mymodule.ko # ./perf probe -m ./mymodule.ko my_func Added new event: probe:my_func (on my_func in /home/wangnan/kmodule/mymodule.ko) You can now use it in all perf tools, such as: perf record -e probe:my_func -aR sleep 1 Where: # ldd ./perf not a dynamic executable # strace -e open ./perf probe -m ./mymodule.ko my_func ... open("/home/wangnan/kmodule/mymodule.ko", O_RDONLY) = 3 open("/home/wangnan/kmodule/../lib64/elfutils/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) ... open("/lib64/tls/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("/lib64/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("/usr/lib64/tls/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("/usr/lib64/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("[mymodule]", O_RDONLY) = -1 ENOENT (No such file or directory) open("/home/wangnan/.debug/.build-id/32/6ab42550ef3d24944f53c817533728367effeb", O_RDONLY) = -1 ENOENT (No such file or directory) open("[mymodule]", O_RDONLY) = -1 ENOENT (No such file or directory) In the above example, probe fails before we put the module into buildid-cache. However, user would expect it success in both case because perf is able to find probe points actually. The reason is because perf won't utilize module's full path if it failed to open debuginfo. In: convert_to_probe_trace_events -> find_probe_trace_events_from_map -> get_target_map -> kernel_get_module_map -> machine__findnew_module_map -> map_groups__find_by_name map_groups__find_by_name() is able to find the map of that module, but this information is found from /proc/module before it knows the real path of the offline module. Therefore, the map->dso->long_name is set to something like '[mymodule]', which prevent dso__load() find the real path of the module file. In another aspect, if dso__load() can get the offline module through buildid cache, it can read symble table from that ko. Even if debuginfo is not available, 'perf probe' can success if the '.symtab' can be found. This patch improves machine__findnew_module_map(): when dso->long_name is leading with '[' (doesn't find path of module when parsing /proc/modules), fixes it by dso__set_long_name(), so following dso__load() is possible to find the symbol table. This patch won't interfere with buildid matching. Here is the test result: # ./perf probe -m ./mymodule.ko my_func Added new event: probe:my_func (on my_func in /home/wangnan/kmodule/mymodule.ko) You can now use it in all perf tools, such as: perf record -e probe:my_func -aR sleep 1 # ./perf probe -d '*' Removed event: probe:my_func # mv ./mymodule.{ko,.bak} # mv ./moduleb.ko mymodule.ko # ./perf probe -m ./mymodule.ko my_func /home/wangnan/kmodule/mymodule.ko with build id 326ab42550ef3d24944f53c817533728367effeb not found, continuing without symbols Failed to find symbol my_func in /home/wangnan/kmodule/mymodule.ko Error: Failed to add events. # ./perf probe -v -m ./mymodule.ko my_func probe-definition(0): my_func symbol:my_func file:(null) line:0 offset:0 return:0 lazy:(null) 0 arguments Could not open debuginfo. Try to use symbols. symsrc__init: build id mismatch for /home/wangnan/kmodule/mymodule.ko. /home/wangnan/kmodule/mymodule.ko with build id 326ab42550ef3d24944f53c817533728367effeb not found, continuing without symbols Failed to find symbol my_func in /home/wangnan/kmodule/mymodule.ko Error: Failed to add events. Reason: No such file or directory (Code: -2) Signed-off-by: Wang Nan <wangnan0@huawei.com> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Zefan Li <lizefan@huawei.com> Cc: pi3orama@163.com Link: http://lkml.kernel.org/r/1448510397-187965-1-git-send-email-wangnan0@huawei.com [ Renamed adjust_dso_long_name() do dso__adjust_kmod_long_name() ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-11-26 11:59:57 +08:00
static void dso__adjust_kmod_long_name(struct dso *dso, const char *filename)
{
const char *dup_filename;
if (!filename || !dso || !dso->long_name)
return;
if (dso->long_name[0] != '[')
return;
if (!strchr(filename, '/'))
return;
dup_filename = strdup(filename);
if (!dup_filename)
return;
dso__set_long_name(dso, dup_filename, true);
perf machine: Adjust dso->long_name for offline module Something unexpected may happen if copy statically linked perf to a production environment: # ./perf probe -m ./mymodule.ko my_func [mymodule] with build id 326ab42550ef3d24944f53c817533728367effeb not found, continuing without symbols Failed to find symbol my_func in /home/wangnan/kmodule/mymodule.ko Error: Failed to add events. # ./perf buildid-cache -a ./mymodule.ko # ./perf probe -m ./mymodule.ko my_func Added new event: probe:my_func (on my_func in /home/wangnan/kmodule/mymodule.ko) You can now use it in all perf tools, such as: perf record -e probe:my_func -aR sleep 1 Where: # ldd ./perf not a dynamic executable # strace -e open ./perf probe -m ./mymodule.ko my_func ... open("/home/wangnan/kmodule/mymodule.ko", O_RDONLY) = 3 open("/home/wangnan/kmodule/../lib64/elfutils/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) ... open("/lib64/tls/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("/lib64/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("/usr/lib64/tls/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("/usr/lib64/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("[mymodule]", O_RDONLY) = -1 ENOENT (No such file or directory) open("/home/wangnan/.debug/.build-id/32/6ab42550ef3d24944f53c817533728367effeb", O_RDONLY) = -1 ENOENT (No such file or directory) open("[mymodule]", O_RDONLY) = -1 ENOENT (No such file or directory) In the above example, probe fails before we put the module into buildid-cache. However, user would expect it success in both case because perf is able to find probe points actually. The reason is because perf won't utilize module's full path if it failed to open debuginfo. In: convert_to_probe_trace_events -> find_probe_trace_events_from_map -> get_target_map -> kernel_get_module_map -> machine__findnew_module_map -> map_groups__find_by_name map_groups__find_by_name() is able to find the map of that module, but this information is found from /proc/module before it knows the real path of the offline module. Therefore, the map->dso->long_name is set to something like '[mymodule]', which prevent dso__load() find the real path of the module file. In another aspect, if dso__load() can get the offline module through buildid cache, it can read symble table from that ko. Even if debuginfo is not available, 'perf probe' can success if the '.symtab' can be found. This patch improves machine__findnew_module_map(): when dso->long_name is leading with '[' (doesn't find path of module when parsing /proc/modules), fixes it by dso__set_long_name(), so following dso__load() is possible to find the symbol table. This patch won't interfere with buildid matching. Here is the test result: # ./perf probe -m ./mymodule.ko my_func Added new event: probe:my_func (on my_func in /home/wangnan/kmodule/mymodule.ko) You can now use it in all perf tools, such as: perf record -e probe:my_func -aR sleep 1 # ./perf probe -d '*' Removed event: probe:my_func # mv ./mymodule.{ko,.bak} # mv ./moduleb.ko mymodule.ko # ./perf probe -m ./mymodule.ko my_func /home/wangnan/kmodule/mymodule.ko with build id 326ab42550ef3d24944f53c817533728367effeb not found, continuing without symbols Failed to find symbol my_func in /home/wangnan/kmodule/mymodule.ko Error: Failed to add events. # ./perf probe -v -m ./mymodule.ko my_func probe-definition(0): my_func symbol:my_func file:(null) line:0 offset:0 return:0 lazy:(null) 0 arguments Could not open debuginfo. Try to use symbols. symsrc__init: build id mismatch for /home/wangnan/kmodule/mymodule.ko. /home/wangnan/kmodule/mymodule.ko with build id 326ab42550ef3d24944f53c817533728367effeb not found, continuing without symbols Failed to find symbol my_func in /home/wangnan/kmodule/mymodule.ko Error: Failed to add events. Reason: No such file or directory (Code: -2) Signed-off-by: Wang Nan <wangnan0@huawei.com> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Zefan Li <lizefan@huawei.com> Cc: pi3orama@163.com Link: http://lkml.kernel.org/r/1448510397-187965-1-git-send-email-wangnan0@huawei.com [ Renamed adjust_dso_long_name() do dso__adjust_kmod_long_name() ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-11-26 11:59:57 +08:00
}
struct map *machine__findnew_module_map(struct machine *machine, u64 start,
const char *filename)
{
struct map *map = NULL;
perf machine: Fix machine__findnew_module_map to put dso Fix machine__findnew_module_map to drop the reference to the dso because it is already referenced by both machine__findnew_module_dso() and map__new2(). Refcnt debugger shows: ==== [1] ==== Unreclaimed dso: 0x1ffd980 Refcount +1 => 1 at ./perf(dso__new+0x1ff) [0x4a62df] ./perf(__dsos__addnew+0x29) [0x4a6e19] ./perf() [0x4b8b91] ./perf(modules__parse+0xfc) [0x4a9d5c] ./perf() [0x4b8460] ./perf(machine__create_kernel_maps+0x150) [0x4bb550] ./perf(machine__new_host+0xfa) [0x4bb75a] ./perf(init_probe_symbol_maps+0x93) [0x506623] ./perf() [0x455ffa] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7f1345a8eaf5] ./perf() [0x4220a9] This map_groups__insert(0x4b8b91) already gets a reference to the new dso: ---- eu-addr2line -e ./perf -f 0x4b8b91 map_groups__insert inlined at util/machine.c:586 in machine__create_module util/map.h:207 ---- So this dso refcnt will be released when map_groups gets released. [snip] Refcount +1 => 2 at ./perf(dso__get+0x34) [0x4a65f4] ./perf() [0x4b8b35] ./perf(modules__parse+0xfc) [0x4a9d5c] ./perf() [0x4b8460] ./perf(machine__create_kernel_maps+0x150) [0x4bb550] ./perf(machine__new_host+0xfa) [0x4bb75a] ./perf(init_probe_symbol_maps+0x93) [0x506623] ./perf() [0x455ffa] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7f1345a8eaf5] ./perf() [0x4220a9] Here, machine__findnew_module_dso(0x4b8b35) gets the dso (and stores it in a local variable): ---- # eu-addr2line -e ./perf -f 0x4b8b35 machine__findnew_module_dso inlined at util/machine.c:578 in machine__create_module util/machine.c:514 ---- Refcount +1 => 3 at ./perf(dso__get+0x34) [0x4a65f4] ./perf(map__new2+0x76) [0x4be1c6] ./perf() [0x4b8b4f] ./perf(modules__parse+0xfc) [0x4a9d5c] ./perf() [0x4b8460] ./perf(machine__create_kernel_maps+0x150) [0x4bb550] ./perf(machine__new_host+0xfa) [0x4bb75a] ./perf(init_probe_symbol_maps+0x93) [0x506623] ./perf() [0x455ffa] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7f1345a8eaf5] ./perf() [0x4220a9] But also map__new2() gets the dso which will be put when the map is released. So, we have to drop the constructor reference obtained in machine__findnew_module_dso(). Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/20151118064035.30709.58824.stgit@localhost.localdomain Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-11-18 14:40:35 +08:00
struct dso *dso = NULL;
struct kmod_path m;
if (kmod_path__parse_name(&m, filename))
return NULL;
map = map_groups__find_by_name(&machine->kmaps, MAP__FUNCTION,
m.name);
perf machine: Adjust dso->long_name for offline module Something unexpected may happen if copy statically linked perf to a production environment: # ./perf probe -m ./mymodule.ko my_func [mymodule] with build id 326ab42550ef3d24944f53c817533728367effeb not found, continuing without symbols Failed to find symbol my_func in /home/wangnan/kmodule/mymodule.ko Error: Failed to add events. # ./perf buildid-cache -a ./mymodule.ko # ./perf probe -m ./mymodule.ko my_func Added new event: probe:my_func (on my_func in /home/wangnan/kmodule/mymodule.ko) You can now use it in all perf tools, such as: perf record -e probe:my_func -aR sleep 1 Where: # ldd ./perf not a dynamic executable # strace -e open ./perf probe -m ./mymodule.ko my_func ... open("/home/wangnan/kmodule/mymodule.ko", O_RDONLY) = 3 open("/home/wangnan/kmodule/../lib64/elfutils/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) ... open("/lib64/tls/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("/lib64/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("/usr/lib64/tls/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("/usr/lib64/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("[mymodule]", O_RDONLY) = -1 ENOENT (No such file or directory) open("/home/wangnan/.debug/.build-id/32/6ab42550ef3d24944f53c817533728367effeb", O_RDONLY) = -1 ENOENT (No such file or directory) open("[mymodule]", O_RDONLY) = -1 ENOENT (No such file or directory) In the above example, probe fails before we put the module into buildid-cache. However, user would expect it success in both case because perf is able to find probe points actually. The reason is because perf won't utilize module's full path if it failed to open debuginfo. In: convert_to_probe_trace_events -> find_probe_trace_events_from_map -> get_target_map -> kernel_get_module_map -> machine__findnew_module_map -> map_groups__find_by_name map_groups__find_by_name() is able to find the map of that module, but this information is found from /proc/module before it knows the real path of the offline module. Therefore, the map->dso->long_name is set to something like '[mymodule]', which prevent dso__load() find the real path of the module file. In another aspect, if dso__load() can get the offline module through buildid cache, it can read symble table from that ko. Even if debuginfo is not available, 'perf probe' can success if the '.symtab' can be found. This patch improves machine__findnew_module_map(): when dso->long_name is leading with '[' (doesn't find path of module when parsing /proc/modules), fixes it by dso__set_long_name(), so following dso__load() is possible to find the symbol table. This patch won't interfere with buildid matching. Here is the test result: # ./perf probe -m ./mymodule.ko my_func Added new event: probe:my_func (on my_func in /home/wangnan/kmodule/mymodule.ko) You can now use it in all perf tools, such as: perf record -e probe:my_func -aR sleep 1 # ./perf probe -d '*' Removed event: probe:my_func # mv ./mymodule.{ko,.bak} # mv ./moduleb.ko mymodule.ko # ./perf probe -m ./mymodule.ko my_func /home/wangnan/kmodule/mymodule.ko with build id 326ab42550ef3d24944f53c817533728367effeb not found, continuing without symbols Failed to find symbol my_func in /home/wangnan/kmodule/mymodule.ko Error: Failed to add events. # ./perf probe -v -m ./mymodule.ko my_func probe-definition(0): my_func symbol:my_func file:(null) line:0 offset:0 return:0 lazy:(null) 0 arguments Could not open debuginfo. Try to use symbols. symsrc__init: build id mismatch for /home/wangnan/kmodule/mymodule.ko. /home/wangnan/kmodule/mymodule.ko with build id 326ab42550ef3d24944f53c817533728367effeb not found, continuing without symbols Failed to find symbol my_func in /home/wangnan/kmodule/mymodule.ko Error: Failed to add events. Reason: No such file or directory (Code: -2) Signed-off-by: Wang Nan <wangnan0@huawei.com> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Zefan Li <lizefan@huawei.com> Cc: pi3orama@163.com Link: http://lkml.kernel.org/r/1448510397-187965-1-git-send-email-wangnan0@huawei.com [ Renamed adjust_dso_long_name() do dso__adjust_kmod_long_name() ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-11-26 11:59:57 +08:00
if (map) {
/*
* If the map's dso is an offline module, give dso__load()
* a chance to find the file path of that module by fixing
* long_name.
*/
dso__adjust_kmod_long_name(map->dso, filename);
goto out;
perf machine: Adjust dso->long_name for offline module Something unexpected may happen if copy statically linked perf to a production environment: # ./perf probe -m ./mymodule.ko my_func [mymodule] with build id 326ab42550ef3d24944f53c817533728367effeb not found, continuing without symbols Failed to find symbol my_func in /home/wangnan/kmodule/mymodule.ko Error: Failed to add events. # ./perf buildid-cache -a ./mymodule.ko # ./perf probe -m ./mymodule.ko my_func Added new event: probe:my_func (on my_func in /home/wangnan/kmodule/mymodule.ko) You can now use it in all perf tools, such as: perf record -e probe:my_func -aR sleep 1 Where: # ldd ./perf not a dynamic executable # strace -e open ./perf probe -m ./mymodule.ko my_func ... open("/home/wangnan/kmodule/mymodule.ko", O_RDONLY) = 3 open("/home/wangnan/kmodule/../lib64/elfutils/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) ... open("/lib64/tls/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("/lib64/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("/usr/lib64/tls/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("/usr/lib64/libebl_x86_64.so", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory) open("[mymodule]", O_RDONLY) = -1 ENOENT (No such file or directory) open("/home/wangnan/.debug/.build-id/32/6ab42550ef3d24944f53c817533728367effeb", O_RDONLY) = -1 ENOENT (No such file or directory) open("[mymodule]", O_RDONLY) = -1 ENOENT (No such file or directory) In the above example, probe fails before we put the module into buildid-cache. However, user would expect it success in both case because perf is able to find probe points actually. The reason is because perf won't utilize module's full path if it failed to open debuginfo. In: convert_to_probe_trace_events -> find_probe_trace_events_from_map -> get_target_map -> kernel_get_module_map -> machine__findnew_module_map -> map_groups__find_by_name map_groups__find_by_name() is able to find the map of that module, but this information is found from /proc/module before it knows the real path of the offline module. Therefore, the map->dso->long_name is set to something like '[mymodule]', which prevent dso__load() find the real path of the module file. In another aspect, if dso__load() can get the offline module through buildid cache, it can read symble table from that ko. Even if debuginfo is not available, 'perf probe' can success if the '.symtab' can be found. This patch improves machine__findnew_module_map(): when dso->long_name is leading with '[' (doesn't find path of module when parsing /proc/modules), fixes it by dso__set_long_name(), so following dso__load() is possible to find the symbol table. This patch won't interfere with buildid matching. Here is the test result: # ./perf probe -m ./mymodule.ko my_func Added new event: probe:my_func (on my_func in /home/wangnan/kmodule/mymodule.ko) You can now use it in all perf tools, such as: perf record -e probe:my_func -aR sleep 1 # ./perf probe -d '*' Removed event: probe:my_func # mv ./mymodule.{ko,.bak} # mv ./moduleb.ko mymodule.ko # ./perf probe -m ./mymodule.ko my_func /home/wangnan/kmodule/mymodule.ko with build id 326ab42550ef3d24944f53c817533728367effeb not found, continuing without symbols Failed to find symbol my_func in /home/wangnan/kmodule/mymodule.ko Error: Failed to add events. # ./perf probe -v -m ./mymodule.ko my_func probe-definition(0): my_func symbol:my_func file:(null) line:0 offset:0 return:0 lazy:(null) 0 arguments Could not open debuginfo. Try to use symbols. symsrc__init: build id mismatch for /home/wangnan/kmodule/mymodule.ko. /home/wangnan/kmodule/mymodule.ko with build id 326ab42550ef3d24944f53c817533728367effeb not found, continuing without symbols Failed to find symbol my_func in /home/wangnan/kmodule/mymodule.ko Error: Failed to add events. Reason: No such file or directory (Code: -2) Signed-off-by: Wang Nan <wangnan0@huawei.com> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Zefan Li <lizefan@huawei.com> Cc: pi3orama@163.com Link: http://lkml.kernel.org/r/1448510397-187965-1-git-send-email-wangnan0@huawei.com [ Renamed adjust_dso_long_name() do dso__adjust_kmod_long_name() ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-11-26 11:59:57 +08:00
}
dso = machine__findnew_module_dso(machine, &m, filename);
if (dso == NULL)
goto out;
map = map__new2(start, dso, MAP__FUNCTION);
if (map == NULL)
goto out;
map_groups__insert(&machine->kmaps, map);
perf machine: Fix machine__findnew_module_map to put registered map Fix machine object to drop the reference to the map object after it inserted it into machine->kmaps. refcnt debugger shows what happened: ---- ==== [2] ==== Unreclaimed map: 0x346f750 Refcount +1 => 1 at ./perf(map__new2+0xb5) [0x4bdea5] ./perf() [0x4b8aaf] ./perf(modules__parse+0xfc) [0x4a9cbc] ./perf() [0x4b83c0] ./perf(machine__create_kernel_maps+0x148) [0x4bb208] ./perf(machine__new_host+0xfa) [0x4bb3fa] ./perf(init_probe_symbol_maps+0x93) [0x5062b3] ./perf() [0x455ffa] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7f5373899af5] ./perf() [0x4220a9] Refcount +1 => 2 at ./perf(maps__insert+0x9a) [0x4bfd4a] ./perf() [0x4b8acb] ./perf(modules__parse+0xfc) [0x4a9cbc] ./perf() [0x4b83c0] ./perf(machine__create_kernel_maps+0x148) [0x4bb208] ./perf(machine__new_host+0xfa) [0x4bb3fa] ./perf(init_probe_symbol_maps+0x93) [0x5062b3] ./perf() [0x455ffa] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7f5373899af5] ./perf() [0x4220a9] Refcount -1 => 1 at ./perf(map_groups__exit+0x94) [0x4bea54] ./perf(machine__delete+0x3d) [0x4b91ed] ./perf(exit_probe_symbol_maps+0x28) [0x506358] ./perf() [0x45628a] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7f5373899af5] ./perf() [0x4220a9] ---- This pattern clearly shows that the refcnt of the map is acquired twice by map__new2 and maps__insert but released onlu once at map_groups__exit, when we purge its maps rbtree. Since maps__insert already reference counted the map, we have to drop the constructor (map__new2) reference count right after inserting it. These happened in machine__findnew_module_map, as below. ---- # eu-addr2line -e ./perf -f 0x4b8aaf machine__findnew_module_map inlined at util/machine.c:1046 in machine__create_module util/machine.c:582 # eu-addr2line -e ./perf -f 0x4b8acb map_groups__insert inlined at util/machine.c:585 in machine__create_module util/map.h:208 ---- (note that both are at util/machine.c:58X which is machine__findnew_module_map) Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/20151118064020.30709.40499.stgit@localhost.localdomain Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-11-18 14:40:20 +08:00
/* Put the map here because map_groups__insert alread got it */
map__put(map);
out:
perf machine: Fix machine__findnew_module_map to put dso Fix machine__findnew_module_map to drop the reference to the dso because it is already referenced by both machine__findnew_module_dso() and map__new2(). Refcnt debugger shows: ==== [1] ==== Unreclaimed dso: 0x1ffd980 Refcount +1 => 1 at ./perf(dso__new+0x1ff) [0x4a62df] ./perf(__dsos__addnew+0x29) [0x4a6e19] ./perf() [0x4b8b91] ./perf(modules__parse+0xfc) [0x4a9d5c] ./perf() [0x4b8460] ./perf(machine__create_kernel_maps+0x150) [0x4bb550] ./perf(machine__new_host+0xfa) [0x4bb75a] ./perf(init_probe_symbol_maps+0x93) [0x506623] ./perf() [0x455ffa] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7f1345a8eaf5] ./perf() [0x4220a9] This map_groups__insert(0x4b8b91) already gets a reference to the new dso: ---- eu-addr2line -e ./perf -f 0x4b8b91 map_groups__insert inlined at util/machine.c:586 in machine__create_module util/map.h:207 ---- So this dso refcnt will be released when map_groups gets released. [snip] Refcount +1 => 2 at ./perf(dso__get+0x34) [0x4a65f4] ./perf() [0x4b8b35] ./perf(modules__parse+0xfc) [0x4a9d5c] ./perf() [0x4b8460] ./perf(machine__create_kernel_maps+0x150) [0x4bb550] ./perf(machine__new_host+0xfa) [0x4bb75a] ./perf(init_probe_symbol_maps+0x93) [0x506623] ./perf() [0x455ffa] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7f1345a8eaf5] ./perf() [0x4220a9] Here, machine__findnew_module_dso(0x4b8b35) gets the dso (and stores it in a local variable): ---- # eu-addr2line -e ./perf -f 0x4b8b35 machine__findnew_module_dso inlined at util/machine.c:578 in machine__create_module util/machine.c:514 ---- Refcount +1 => 3 at ./perf(dso__get+0x34) [0x4a65f4] ./perf(map__new2+0x76) [0x4be1c6] ./perf() [0x4b8b4f] ./perf(modules__parse+0xfc) [0x4a9d5c] ./perf() [0x4b8460] ./perf(machine__create_kernel_maps+0x150) [0x4bb550] ./perf(machine__new_host+0xfa) [0x4bb75a] ./perf(init_probe_symbol_maps+0x93) [0x506623] ./perf() [0x455ffa] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7f1345a8eaf5] ./perf() [0x4220a9] But also map__new2() gets the dso which will be put when the map is released. So, we have to drop the constructor reference obtained in machine__findnew_module_dso(). Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/20151118064035.30709.58824.stgit@localhost.localdomain Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-11-18 14:40:35 +08:00
/* put the dso here, corresponding to machine__findnew_module_dso */
dso__put(dso);
free(m.name);
return map;
}
size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
{
struct rb_node *nd;
size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
struct machine *pos = rb_entry(nd, struct machine, rb_node);
ret += __dsos__fprintf(&pos->dsos.head, fp);
}
return ret;
}
size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
bool (skip)(struct dso *dso, int parm), int parm)
{
return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
}
size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
bool (skip)(struct dso *dso, int parm), int parm)
{
struct rb_node *nd;
size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
struct machine *pos = rb_entry(nd, struct machine, rb_node);
ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
}
return ret;
}
size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
{
int i;
size_t printed = 0;
struct dso *kdso = machine__kernel_map(machine)->dso;
if (kdso->has_build_id) {
char filename[PATH_MAX];
if (dso__build_id_filename(kdso, filename, sizeof(filename),
false))
printed += fprintf(fp, "[0] %s\n", filename);
}
for (i = 0; i < vmlinux_path__nr_entries; ++i)
printed += fprintf(fp, "[%d] %s\n",
i + kdso->has_build_id, vmlinux_path[i]);
return printed;
}
size_t machine__fprintf(struct machine *machine, FILE *fp)
{
size_t ret;
struct rb_node *nd;
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
pthread_rwlock_rdlock(&machine->threads_lock);
ret = fprintf(fp, "Threads: %u\n", machine->nr_threads);
for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
struct thread *pos = rb_entry(nd, struct thread, rb_node);
ret += thread__fprintf(pos, fp);
}
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
pthread_rwlock_unlock(&machine->threads_lock);
return ret;
}
static struct dso *machine__get_kernel(struct machine *machine)
{
const char *vmlinux_name = NULL;
struct dso *kernel;
if (machine__is_host(machine)) {
vmlinux_name = symbol_conf.vmlinux_name;
if (!vmlinux_name)
vmlinux_name = DSO__NAME_KALLSYMS;
kernel = machine__findnew_kernel(machine, vmlinux_name,
"[kernel]", DSO_TYPE_KERNEL);
} else {
char bf[PATH_MAX];
if (machine__is_default_guest(machine))
vmlinux_name = symbol_conf.default_guest_vmlinux_name;
if (!vmlinux_name)
vmlinux_name = machine__mmap_name(machine, bf,
sizeof(bf));
kernel = machine__findnew_kernel(machine, vmlinux_name,
"[guest.kernel]",
DSO_TYPE_GUEST_KERNEL);
}
if (kernel != NULL && (!kernel->has_build_id))
dso__read_running_kernel_build_id(kernel, machine);
return kernel;
}
struct process_args {
u64 start;
};
static void machine__get_kallsyms_filename(struct machine *machine, char *buf,
size_t bufsz)
{
if (machine__is_default_guest(machine))
scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
else
scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
}
const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
/* Figure out the start address of kernel map from /proc/kallsyms.
* Returns the name of the start symbol in *symbol_name. Pass in NULL as
* symbol_name if it's not that important.
*/
perf symbols: Accept symbols starting at address 0 That is the case of _text on s390, and we have some functions that return an address, using address zero to report problems, oops. This would lead the symbol loading routines to not use "_text" as the reference relocation symbol, or the first symbol for the kernel, but use instead "_stext", that is at the same address on x86_64 and others, but not on s390: [acme@localhost perf-4.11.0-rc6]$ head -15 /proc/kallsyms 0000000000000000 T _text 0000000000000418 t iplstart 0000000000000800 T start 000000000000080a t .base 000000000000082e t .sk8x8 0000000000000834 t .gotr 0000000000000842 t .cmd 0000000000000846 t .parm 000000000000084a t .lowcase 0000000000010000 T startup 0000000000010010 T startup_kdump 0000000000010214 t startup_kdump_relocated 0000000000011000 T startup_continue 00000000000112a0 T _ehead 0000000000100000 T _stext [acme@localhost perf-4.11.0-rc6]$ Which in turn would make 'perf test vmlinux' to fail because it wouldn't find the symbols before "_stext" in kallsyms. Fix it by using the return value only for errors and storing the address, when the symbol is successfully found, in a provided pointer arg. Before this patch: After: [acme@localhost perf-4.11.0-rc6]$ tools/perf/perf test -v 1 1: vmlinux symtab matches kallsyms : --- start --- test child forked, pid 40693 Looking at the vmlinux_path (8 entries long) Using /usr/lib/debug/lib/modules/3.10.0-654.el7.s390x/vmlinux for symbols ERR : 0: _text not on kallsyms ERR : 0x418: iplstart not on kallsyms ERR : 0x800: start not on kallsyms ERR : 0x80a: .base not on kallsyms ERR : 0x82e: .sk8x8 not on kallsyms ERR : 0x834: .gotr not on kallsyms ERR : 0x842: .cmd not on kallsyms ERR : 0x846: .parm not on kallsyms ERR : 0x84a: .lowcase not on kallsyms ERR : 0x10000: startup not on kallsyms ERR : 0x10010: startup_kdump not on kallsyms ERR : 0x10214: startup_kdump_relocated not on kallsyms ERR : 0x11000: startup_continue not on kallsyms ERR : 0x112a0: _ehead not on kallsyms <SNIP warnings> test child finished with -1 ---- end ---- vmlinux symtab matches kallsyms: FAILED! [acme@localhost perf-4.11.0-rc6]$ After: [acme@localhost perf-4.11.0-rc6]$ tools/perf/perf test -v 1 1: vmlinux symtab matches kallsyms : --- start --- test child forked, pid 47160 <SNIP warnings> test child finished with 0 ---- end ---- vmlinux symtab matches kallsyms: Ok [acme@localhost perf-4.11.0-rc6]$ Reported-by: Michael Petlan <mpetlan@redhat.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-9x9bwgd3btwdk1u51xie93fz@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-04-28 08:21:09 +08:00
static int machine__get_running_kernel_start(struct machine *machine,
const char **symbol_name, u64 *start)
{
char filename[PATH_MAX];
perf symbols: Accept symbols starting at address 0 That is the case of _text on s390, and we have some functions that return an address, using address zero to report problems, oops. This would lead the symbol loading routines to not use "_text" as the reference relocation symbol, or the first symbol for the kernel, but use instead "_stext", that is at the same address on x86_64 and others, but not on s390: [acme@localhost perf-4.11.0-rc6]$ head -15 /proc/kallsyms 0000000000000000 T _text 0000000000000418 t iplstart 0000000000000800 T start 000000000000080a t .base 000000000000082e t .sk8x8 0000000000000834 t .gotr 0000000000000842 t .cmd 0000000000000846 t .parm 000000000000084a t .lowcase 0000000000010000 T startup 0000000000010010 T startup_kdump 0000000000010214 t startup_kdump_relocated 0000000000011000 T startup_continue 00000000000112a0 T _ehead 0000000000100000 T _stext [acme@localhost perf-4.11.0-rc6]$ Which in turn would make 'perf test vmlinux' to fail because it wouldn't find the symbols before "_stext" in kallsyms. Fix it by using the return value only for errors and storing the address, when the symbol is successfully found, in a provided pointer arg. Before this patch: After: [acme@localhost perf-4.11.0-rc6]$ tools/perf/perf test -v 1 1: vmlinux symtab matches kallsyms : --- start --- test child forked, pid 40693 Looking at the vmlinux_path (8 entries long) Using /usr/lib/debug/lib/modules/3.10.0-654.el7.s390x/vmlinux for symbols ERR : 0: _text not on kallsyms ERR : 0x418: iplstart not on kallsyms ERR : 0x800: start not on kallsyms ERR : 0x80a: .base not on kallsyms ERR : 0x82e: .sk8x8 not on kallsyms ERR : 0x834: .gotr not on kallsyms ERR : 0x842: .cmd not on kallsyms ERR : 0x846: .parm not on kallsyms ERR : 0x84a: .lowcase not on kallsyms ERR : 0x10000: startup not on kallsyms ERR : 0x10010: startup_kdump not on kallsyms ERR : 0x10214: startup_kdump_relocated not on kallsyms ERR : 0x11000: startup_continue not on kallsyms ERR : 0x112a0: _ehead not on kallsyms <SNIP warnings> test child finished with -1 ---- end ---- vmlinux symtab matches kallsyms: FAILED! [acme@localhost perf-4.11.0-rc6]$ After: [acme@localhost perf-4.11.0-rc6]$ tools/perf/perf test -v 1 1: vmlinux symtab matches kallsyms : --- start --- test child forked, pid 47160 <SNIP warnings> test child finished with 0 ---- end ---- vmlinux symtab matches kallsyms: Ok [acme@localhost perf-4.11.0-rc6]$ Reported-by: Michael Petlan <mpetlan@redhat.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-9x9bwgd3btwdk1u51xie93fz@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-04-28 08:21:09 +08:00
int i, err = -1;
const char *name;
u64 addr = 0;
machine__get_kallsyms_filename(machine, filename, PATH_MAX);
if (symbol__restricted_filename(filename, "/proc/kallsyms"))
return 0;
for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
perf symbols: Accept symbols starting at address 0 That is the case of _text on s390, and we have some functions that return an address, using address zero to report problems, oops. This would lead the symbol loading routines to not use "_text" as the reference relocation symbol, or the first symbol for the kernel, but use instead "_stext", that is at the same address on x86_64 and others, but not on s390: [acme@localhost perf-4.11.0-rc6]$ head -15 /proc/kallsyms 0000000000000000 T _text 0000000000000418 t iplstart 0000000000000800 T start 000000000000080a t .base 000000000000082e t .sk8x8 0000000000000834 t .gotr 0000000000000842 t .cmd 0000000000000846 t .parm 000000000000084a t .lowcase 0000000000010000 T startup 0000000000010010 T startup_kdump 0000000000010214 t startup_kdump_relocated 0000000000011000 T startup_continue 00000000000112a0 T _ehead 0000000000100000 T _stext [acme@localhost perf-4.11.0-rc6]$ Which in turn would make 'perf test vmlinux' to fail because it wouldn't find the symbols before "_stext" in kallsyms. Fix it by using the return value only for errors and storing the address, when the symbol is successfully found, in a provided pointer arg. Before this patch: After: [acme@localhost perf-4.11.0-rc6]$ tools/perf/perf test -v 1 1: vmlinux symtab matches kallsyms : --- start --- test child forked, pid 40693 Looking at the vmlinux_path (8 entries long) Using /usr/lib/debug/lib/modules/3.10.0-654.el7.s390x/vmlinux for symbols ERR : 0: _text not on kallsyms ERR : 0x418: iplstart not on kallsyms ERR : 0x800: start not on kallsyms ERR : 0x80a: .base not on kallsyms ERR : 0x82e: .sk8x8 not on kallsyms ERR : 0x834: .gotr not on kallsyms ERR : 0x842: .cmd not on kallsyms ERR : 0x846: .parm not on kallsyms ERR : 0x84a: .lowcase not on kallsyms ERR : 0x10000: startup not on kallsyms ERR : 0x10010: startup_kdump not on kallsyms ERR : 0x10214: startup_kdump_relocated not on kallsyms ERR : 0x11000: startup_continue not on kallsyms ERR : 0x112a0: _ehead not on kallsyms <SNIP warnings> test child finished with -1 ---- end ---- vmlinux symtab matches kallsyms: FAILED! [acme@localhost perf-4.11.0-rc6]$ After: [acme@localhost perf-4.11.0-rc6]$ tools/perf/perf test -v 1 1: vmlinux symtab matches kallsyms : --- start --- test child forked, pid 47160 <SNIP warnings> test child finished with 0 ---- end ---- vmlinux symtab matches kallsyms: Ok [acme@localhost perf-4.11.0-rc6]$ Reported-by: Michael Petlan <mpetlan@redhat.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-9x9bwgd3btwdk1u51xie93fz@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-04-28 08:21:09 +08:00
err = kallsyms__get_function_start(filename, name, &addr);
if (!err)
break;
}
perf symbols: Accept symbols starting at address 0 That is the case of _text on s390, and we have some functions that return an address, using address zero to report problems, oops. This would lead the symbol loading routines to not use "_text" as the reference relocation symbol, or the first symbol for the kernel, but use instead "_stext", that is at the same address on x86_64 and others, but not on s390: [acme@localhost perf-4.11.0-rc6]$ head -15 /proc/kallsyms 0000000000000000 T _text 0000000000000418 t iplstart 0000000000000800 T start 000000000000080a t .base 000000000000082e t .sk8x8 0000000000000834 t .gotr 0000000000000842 t .cmd 0000000000000846 t .parm 000000000000084a t .lowcase 0000000000010000 T startup 0000000000010010 T startup_kdump 0000000000010214 t startup_kdump_relocated 0000000000011000 T startup_continue 00000000000112a0 T _ehead 0000000000100000 T _stext [acme@localhost perf-4.11.0-rc6]$ Which in turn would make 'perf test vmlinux' to fail because it wouldn't find the symbols before "_stext" in kallsyms. Fix it by using the return value only for errors and storing the address, when the symbol is successfully found, in a provided pointer arg. Before this patch: After: [acme@localhost perf-4.11.0-rc6]$ tools/perf/perf test -v 1 1: vmlinux symtab matches kallsyms : --- start --- test child forked, pid 40693 Looking at the vmlinux_path (8 entries long) Using /usr/lib/debug/lib/modules/3.10.0-654.el7.s390x/vmlinux for symbols ERR : 0: _text not on kallsyms ERR : 0x418: iplstart not on kallsyms ERR : 0x800: start not on kallsyms ERR : 0x80a: .base not on kallsyms ERR : 0x82e: .sk8x8 not on kallsyms ERR : 0x834: .gotr not on kallsyms ERR : 0x842: .cmd not on kallsyms ERR : 0x846: .parm not on kallsyms ERR : 0x84a: .lowcase not on kallsyms ERR : 0x10000: startup not on kallsyms ERR : 0x10010: startup_kdump not on kallsyms ERR : 0x10214: startup_kdump_relocated not on kallsyms ERR : 0x11000: startup_continue not on kallsyms ERR : 0x112a0: _ehead not on kallsyms <SNIP warnings> test child finished with -1 ---- end ---- vmlinux symtab matches kallsyms: FAILED! [acme@localhost perf-4.11.0-rc6]$ After: [acme@localhost perf-4.11.0-rc6]$ tools/perf/perf test -v 1 1: vmlinux symtab matches kallsyms : --- start --- test child forked, pid 47160 <SNIP warnings> test child finished with 0 ---- end ---- vmlinux symtab matches kallsyms: Ok [acme@localhost perf-4.11.0-rc6]$ Reported-by: Michael Petlan <mpetlan@redhat.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-9x9bwgd3btwdk1u51xie93fz@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-04-28 08:21:09 +08:00
if (err)
return -1;
if (symbol_name)
*symbol_name = name;
perf symbols: Accept symbols starting at address 0 That is the case of _text on s390, and we have some functions that return an address, using address zero to report problems, oops. This would lead the symbol loading routines to not use "_text" as the reference relocation symbol, or the first symbol for the kernel, but use instead "_stext", that is at the same address on x86_64 and others, but not on s390: [acme@localhost perf-4.11.0-rc6]$ head -15 /proc/kallsyms 0000000000000000 T _text 0000000000000418 t iplstart 0000000000000800 T start 000000000000080a t .base 000000000000082e t .sk8x8 0000000000000834 t .gotr 0000000000000842 t .cmd 0000000000000846 t .parm 000000000000084a t .lowcase 0000000000010000 T startup 0000000000010010 T startup_kdump 0000000000010214 t startup_kdump_relocated 0000000000011000 T startup_continue 00000000000112a0 T _ehead 0000000000100000 T _stext [acme@localhost perf-4.11.0-rc6]$ Which in turn would make 'perf test vmlinux' to fail because it wouldn't find the symbols before "_stext" in kallsyms. Fix it by using the return value only for errors and storing the address, when the symbol is successfully found, in a provided pointer arg. Before this patch: After: [acme@localhost perf-4.11.0-rc6]$ tools/perf/perf test -v 1 1: vmlinux symtab matches kallsyms : --- start --- test child forked, pid 40693 Looking at the vmlinux_path (8 entries long) Using /usr/lib/debug/lib/modules/3.10.0-654.el7.s390x/vmlinux for symbols ERR : 0: _text not on kallsyms ERR : 0x418: iplstart not on kallsyms ERR : 0x800: start not on kallsyms ERR : 0x80a: .base not on kallsyms ERR : 0x82e: .sk8x8 not on kallsyms ERR : 0x834: .gotr not on kallsyms ERR : 0x842: .cmd not on kallsyms ERR : 0x846: .parm not on kallsyms ERR : 0x84a: .lowcase not on kallsyms ERR : 0x10000: startup not on kallsyms ERR : 0x10010: startup_kdump not on kallsyms ERR : 0x10214: startup_kdump_relocated not on kallsyms ERR : 0x11000: startup_continue not on kallsyms ERR : 0x112a0: _ehead not on kallsyms <SNIP warnings> test child finished with -1 ---- end ---- vmlinux symtab matches kallsyms: FAILED! [acme@localhost perf-4.11.0-rc6]$ After: [acme@localhost perf-4.11.0-rc6]$ tools/perf/perf test -v 1 1: vmlinux symtab matches kallsyms : --- start --- test child forked, pid 47160 <SNIP warnings> test child finished with 0 ---- end ---- vmlinux symtab matches kallsyms: Ok [acme@localhost perf-4.11.0-rc6]$ Reported-by: Michael Petlan <mpetlan@redhat.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-9x9bwgd3btwdk1u51xie93fz@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-04-28 08:21:09 +08:00
*start = addr;
return 0;
}
int __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
{
int type;
perf symbols: Accept symbols starting at address 0 That is the case of _text on s390, and we have some functions that return an address, using address zero to report problems, oops. This would lead the symbol loading routines to not use "_text" as the reference relocation symbol, or the first symbol for the kernel, but use instead "_stext", that is at the same address on x86_64 and others, but not on s390: [acme@localhost perf-4.11.0-rc6]$ head -15 /proc/kallsyms 0000000000000000 T _text 0000000000000418 t iplstart 0000000000000800 T start 000000000000080a t .base 000000000000082e t .sk8x8 0000000000000834 t .gotr 0000000000000842 t .cmd 0000000000000846 t .parm 000000000000084a t .lowcase 0000000000010000 T startup 0000000000010010 T startup_kdump 0000000000010214 t startup_kdump_relocated 0000000000011000 T startup_continue 00000000000112a0 T _ehead 0000000000100000 T _stext [acme@localhost perf-4.11.0-rc6]$ Which in turn would make 'perf test vmlinux' to fail because it wouldn't find the symbols before "_stext" in kallsyms. Fix it by using the return value only for errors and storing the address, when the symbol is successfully found, in a provided pointer arg. Before this patch: After: [acme@localhost perf-4.11.0-rc6]$ tools/perf/perf test -v 1 1: vmlinux symtab matches kallsyms : --- start --- test child forked, pid 40693 Looking at the vmlinux_path (8 entries long) Using /usr/lib/debug/lib/modules/3.10.0-654.el7.s390x/vmlinux for symbols ERR : 0: _text not on kallsyms ERR : 0x418: iplstart not on kallsyms ERR : 0x800: start not on kallsyms ERR : 0x80a: .base not on kallsyms ERR : 0x82e: .sk8x8 not on kallsyms ERR : 0x834: .gotr not on kallsyms ERR : 0x842: .cmd not on kallsyms ERR : 0x846: .parm not on kallsyms ERR : 0x84a: .lowcase not on kallsyms ERR : 0x10000: startup not on kallsyms ERR : 0x10010: startup_kdump not on kallsyms ERR : 0x10214: startup_kdump_relocated not on kallsyms ERR : 0x11000: startup_continue not on kallsyms ERR : 0x112a0: _ehead not on kallsyms <SNIP warnings> test child finished with -1 ---- end ---- vmlinux symtab matches kallsyms: FAILED! [acme@localhost perf-4.11.0-rc6]$ After: [acme@localhost perf-4.11.0-rc6]$ tools/perf/perf test -v 1 1: vmlinux symtab matches kallsyms : --- start --- test child forked, pid 47160 <SNIP warnings> test child finished with 0 ---- end ---- vmlinux symtab matches kallsyms: Ok [acme@localhost perf-4.11.0-rc6]$ Reported-by: Michael Petlan <mpetlan@redhat.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-9x9bwgd3btwdk1u51xie93fz@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-04-28 08:21:09 +08:00
u64 start = 0;
if (machine__get_running_kernel_start(machine, NULL, &start))
return -1;
/* In case of renewal the kernel map, destroy previous one */
machine__destroy_kernel_maps(machine);
for (type = 0; type < MAP__NR_TYPES; ++type) {
struct kmap *kmap;
struct map *map;
machine->vmlinux_maps[type] = map__new2(start, kernel, type);
if (machine->vmlinux_maps[type] == NULL)
return -1;
machine->vmlinux_maps[type]->map_ip =
machine->vmlinux_maps[type]->unmap_ip =
identity__map_ip;
map = __machine__kernel_map(machine, type);
kmap = map__kmap(map);
if (!kmap)
return -1;
kmap->kmaps = &machine->kmaps;
map_groups__insert(&machine->kmaps, map);
}
return 0;
}
void machine__destroy_kernel_maps(struct machine *machine)
{
int type;
for (type = 0; type < MAP__NR_TYPES; ++type) {
struct kmap *kmap;
struct map *map = __machine__kernel_map(machine, type);
if (map == NULL)
continue;
kmap = map__kmap(map);
map_groups__remove(&machine->kmaps, map);
if (kmap && kmap->ref_reloc_sym) {
/*
* ref_reloc_sym is shared among all maps, so free just
* on one of them.
*/
if (type == MAP__FUNCTION) {
zfree((char **)&kmap->ref_reloc_sym->name);
zfree(&kmap->ref_reloc_sym);
} else
kmap->ref_reloc_sym = NULL;
}
perf machine: Fix machine__destroy_kernel_maps to drop vmlinux_maps references Fix machine__destroy_kernel_maps() to drop vmlinux_maps references before filling it with NULL. Refcnt debugger shows ==== [1] ==== Unreclaimed map: 0x36b1070 Refcount +1 => 1 at ./perf(map__new2+0xb5) [0x4bdec5] ./perf(machine__create_kernel_maps+0x72) [0x4bb152] ./perf(machine__new_host+0xfa) [0x4bb41a] ./perf(init_probe_symbol_maps+0x93) [0x5062d3] ./perf() [0x455ffa] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7f1fc9fc4af5] ./perf() [0x4220a9] Refcount +1 => 2 at ./perf(maps__insert+0x9a) [0x4bfd6a] ./perf(machine__create_kernel_maps+0xc3) [0x4bb1a3] ./perf(machine__new_host+0xfa) [0x4bb41a] ./perf(init_probe_symbol_maps+0x93) [0x5062d3] ./perf() [0x455ffa] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7f1fc9fc4af5] ./perf() [0x4220a9] Refcount -1 => 1 at ./perf(map_groups__exit+0x94) [0x4bea74] ./perf(machine__delete+0x3d) [0x4b91fd] ./perf(exit_probe_symbol_maps+0x28) [0x506378] ./perf() [0x45628a] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7f1fc9fc4af5] ./perf() [0x4220a9] map__new2() returns map with refcnt = 1, and also map_groups__insert gets it again in__machine__create_kernel_maps(). machine__destroy_kernel_maps() calls map_groups__remove() to decrement the refcnt, but before decrement it again (corresponding to map__new2), it makes vmlinux_maps[type] = NULL. And this may cause a refcnt leak. Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/20151118064022.30709.3897.stgit@localhost.localdomain Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-11-18 14:40:22 +08:00
map__put(machine->vmlinux_maps[type]);
machine->vmlinux_maps[type] = NULL;
}
}
int machines__create_guest_kernel_maps(struct machines *machines)
{
int ret = 0;
struct dirent **namelist = NULL;
int i, items = 0;
char path[PATH_MAX];
pid_t pid;
char *endp;
if (symbol_conf.default_guest_vmlinux_name ||
symbol_conf.default_guest_modules ||
symbol_conf.default_guest_kallsyms) {
machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
}
if (symbol_conf.guestmount) {
items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
if (items <= 0)
return -ENOENT;
for (i = 0; i < items; i++) {
if (!isdigit(namelist[i]->d_name[0])) {
/* Filter out . and .. */
continue;
}
pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
if ((*endp != '\0') ||
(endp == namelist[i]->d_name) ||
(errno == ERANGE)) {
pr_debug("invalid directory (%s). Skipping.\n",
namelist[i]->d_name);
continue;
}
sprintf(path, "%s/%s/proc/kallsyms",
symbol_conf.guestmount,
namelist[i]->d_name);
ret = access(path, R_OK);
if (ret) {
pr_debug("Can't access file %s\n", path);
goto failure;
}
machines__create_kernel_maps(machines, pid);
}
failure:
free(namelist);
}
return ret;
}
void machines__destroy_kernel_maps(struct machines *machines)
{
struct rb_node *next = rb_first(&machines->guests);
machine__destroy_kernel_maps(&machines->host);
while (next) {
struct machine *pos = rb_entry(next, struct machine, rb_node);
next = rb_next(&pos->rb_node);
rb_erase(&pos->rb_node, &machines->guests);
machine__delete(pos);
}
}
int machines__create_kernel_maps(struct machines *machines, pid_t pid)
{
struct machine *machine = machines__findnew(machines, pid);
if (machine == NULL)
return -1;
return machine__create_kernel_maps(machine);
}
int __machine__load_kallsyms(struct machine *machine, const char *filename,
enum map_type type, bool no_kcore)
{
struct map *map = machine__kernel_map(machine);
int ret = __dso__load_kallsyms(map->dso, filename, map, no_kcore);
if (ret > 0) {
dso__set_loaded(map->dso, type);
/*
* Since /proc/kallsyms will have multiple sessions for the
* kernel, with modules between them, fixup the end of all
* sections.
*/
__map_groups__fixup_end(&machine->kmaps, type);
}
return ret;
}
int machine__load_kallsyms(struct machine *machine, const char *filename,
enum map_type type)
{
return __machine__load_kallsyms(machine, filename, type, false);
}
int machine__load_vmlinux_path(struct machine *machine, enum map_type type)
{
struct map *map = machine__kernel_map(machine);
int ret = dso__load_vmlinux_path(map->dso, map);
if (ret > 0)
dso__set_loaded(map->dso, type);
return ret;
}
static void map_groups__fixup_end(struct map_groups *mg)
{
int i;
for (i = 0; i < MAP__NR_TYPES; ++i)
__map_groups__fixup_end(mg, i);
}
static char *get_kernel_version(const char *root_dir)
{
char version[PATH_MAX];
FILE *file;
char *name, *tmp;
const char *prefix = "Linux version ";
sprintf(version, "%s/proc/version", root_dir);
file = fopen(version, "r");
if (!file)
return NULL;
version[0] = '\0';
tmp = fgets(version, sizeof(version), file);
fclose(file);
name = strstr(version, prefix);
if (!name)
return NULL;
name += strlen(prefix);
tmp = strchr(name, ' ');
if (tmp)
*tmp = '\0';
return strdup(name);
}
static bool is_kmod_dso(struct dso *dso)
{
return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
}
static int map_groups__set_module_path(struct map_groups *mg, const char *path,
struct kmod_path *m)
{
struct map *map;
char *long_name;
map = map_groups__find_by_name(mg, MAP__FUNCTION, m->name);
if (map == NULL)
return 0;
long_name = strdup(path);
if (long_name == NULL)
return -ENOMEM;
dso__set_long_name(map->dso, long_name, true);
dso__kernel_module_get_build_id(map->dso, "");
/*
* Full name could reveal us kmod compression, so
* we need to update the symtab_type if needed.
*/
if (m->comp && is_kmod_dso(map->dso))
map->dso->symtab_type++;
return 0;
}
static int map_groups__set_modules_path_dir(struct map_groups *mg,
const char *dir_name, int depth)
{
struct dirent *dent;
DIR *dir = opendir(dir_name);
int ret = 0;
if (!dir) {
pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
return -1;
}
while ((dent = readdir(dir)) != NULL) {
char path[PATH_MAX];
struct stat st;
/*sshfs might return bad dent->d_type, so we have to stat*/
snprintf(path, sizeof(path), "%s/%s", dir_name, dent->d_name);
if (stat(path, &st))
continue;
if (S_ISDIR(st.st_mode)) {
if (!strcmp(dent->d_name, ".") ||
!strcmp(dent->d_name, ".."))
continue;
/* Do not follow top-level source and build symlinks */
if (depth == 0) {
if (!strcmp(dent->d_name, "source") ||
!strcmp(dent->d_name, "build"))
continue;
}
ret = map_groups__set_modules_path_dir(mg, path,
depth + 1);
if (ret < 0)
goto out;
} else {
struct kmod_path m;
ret = kmod_path__parse_name(&m, dent->d_name);
if (ret)
goto out;
if (m.kmod)
ret = map_groups__set_module_path(mg, path, &m);
free(m.name);
if (ret)
goto out;
}
}
out:
closedir(dir);
return ret;
}
static int machine__set_modules_path(struct machine *machine)
{
char *version;
char modules_path[PATH_MAX];
version = get_kernel_version(machine->root_dir);
if (!version)
return -1;
snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
machine->root_dir, version);
free(version);
return map_groups__set_modules_path_dir(&machine->kmaps, modules_path, 0);
}
int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
const char *name __maybe_unused)
{
return 0;
}
perf record: Fix wrong size in perf_record_mmap for last kernel module During work on perf report for s390 I ran into the following issue: 0 0x318 [0x78]: PERF_RECORD_MMAP -1/0: [0x3ff804d6990(0xfffffc007fb2966f) @ 0]: x /lib/modules/4.12.0perf1+/kernel/drivers/s390/net/qeth_l2.ko This is a PERF_RECORD_MMAP entry of the perf.data file with an invalid module size for qeth_l2.ko (the s390 ethernet device driver). Even a mainframe does not have 0xfffffc007fb2966f bytes of main memory. It turned out that this wrong size is created by the perf record command. What happens is this function call sequence from __cmd_record(): perf_session__new(): perf_session__create_kernel_maps(): machine__create_kernel_maps(): machine__create_modules(): Creates map for all loaded kernel modules. modules__parse(): Reads /proc/modules and extracts module name and load address (1st and last column) machine__create_module(): Called for every module found in /proc/modules. Creates a new map for every module found and enters module name and start address into the map. Since the module end address is unknown it is set to zero. This ends up with a kernel module map list sorted by module start addresses. All module end addresses are zero. Last machine__create_kernel_maps() calls function map_groups__fixup_end(). This function iterates through the maps and assigns each map entry's end address the successor map entry start address. The last entry of the map group has no successor, so ~0 is used as end to consume the remaining memory. Later __cmd_record calls function record__synthesize() which in turn calls perf_event__synthesize_kernel_mmap() and perf_event__synthesize_modules() to create PERF_REPORT_MMAP entries into the perf.data file. On s390 this results in the last module qeth_l2.ko (which has highest start address, see module table: [root@s8360047 perf]# cat /proc/modules qeth_l2 86016 1 - Live 0x000003ff804d6000 qeth 266240 1 qeth_l2, Live 0x000003ff80296000 ccwgroup 24576 1 qeth, Live 0x000003ff80218000 vmur 36864 0 - Live 0x000003ff80182000 qdio 143360 2 qeth_l2,qeth, Live 0x000003ff80002000 [root@s8360047 perf]# ) to be the last entry and its map has an end address of ~0. When the PERF_RECORD_MMAP entry is created for kernel module qeth_l2.ko its start address and length is written. The length is calculated in line: event->mmap.len = pos->end - pos->start; and results in 0xffffffffffffffff - 0x3ff804d6990(*) = 0xfffffc007fb2966f (*) On s390 the module start address is actually determined by a __weak function named arch__fix_module_text_start() in machine__create_module(). I think this improvable. We can use the module size (2nd column of /proc/modules) to get each loaded kernel module size and calculate its end address. Only for map entries which do not have a valid end address (end is still zero) we can use the heuristic we have now, that is use successor start address or ~0. Signed-off-by: Thomas-Mich Richter <tmricht@linux.vnet.ibm.com> Reviewed-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Cc: Thomas-Mich Richter <tmricht@linux.vnet.ibm.com> Cc: Zvonko Kosic <zvonko.kosic@de.ibm.com> LPU-Reference: 20170803134902.47207-2-tmricht@linux.vnet.ibm.com Link: http://lkml.kernel.org/n/tip-nmoqij5b5vxx7rq2ckwu8iaj@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-08-03 21:49:02 +08:00
static int machine__create_module(void *arg, const char *name, u64 start,
u64 size)
{
struct machine *machine = arg;
struct map *map;
if (arch__fix_module_text_start(&start, name) < 0)
return -1;
map = machine__findnew_module_map(machine, start, name);
if (map == NULL)
return -1;
perf record: Fix wrong size in perf_record_mmap for last kernel module During work on perf report for s390 I ran into the following issue: 0 0x318 [0x78]: PERF_RECORD_MMAP -1/0: [0x3ff804d6990(0xfffffc007fb2966f) @ 0]: x /lib/modules/4.12.0perf1+/kernel/drivers/s390/net/qeth_l2.ko This is a PERF_RECORD_MMAP entry of the perf.data file with an invalid module size for qeth_l2.ko (the s390 ethernet device driver). Even a mainframe does not have 0xfffffc007fb2966f bytes of main memory. It turned out that this wrong size is created by the perf record command. What happens is this function call sequence from __cmd_record(): perf_session__new(): perf_session__create_kernel_maps(): machine__create_kernel_maps(): machine__create_modules(): Creates map for all loaded kernel modules. modules__parse(): Reads /proc/modules and extracts module name and load address (1st and last column) machine__create_module(): Called for every module found in /proc/modules. Creates a new map for every module found and enters module name and start address into the map. Since the module end address is unknown it is set to zero. This ends up with a kernel module map list sorted by module start addresses. All module end addresses are zero. Last machine__create_kernel_maps() calls function map_groups__fixup_end(). This function iterates through the maps and assigns each map entry's end address the successor map entry start address. The last entry of the map group has no successor, so ~0 is used as end to consume the remaining memory. Later __cmd_record calls function record__synthesize() which in turn calls perf_event__synthesize_kernel_mmap() and perf_event__synthesize_modules() to create PERF_REPORT_MMAP entries into the perf.data file. On s390 this results in the last module qeth_l2.ko (which has highest start address, see module table: [root@s8360047 perf]# cat /proc/modules qeth_l2 86016 1 - Live 0x000003ff804d6000 qeth 266240 1 qeth_l2, Live 0x000003ff80296000 ccwgroup 24576 1 qeth, Live 0x000003ff80218000 vmur 36864 0 - Live 0x000003ff80182000 qdio 143360 2 qeth_l2,qeth, Live 0x000003ff80002000 [root@s8360047 perf]# ) to be the last entry and its map has an end address of ~0. When the PERF_RECORD_MMAP entry is created for kernel module qeth_l2.ko its start address and length is written. The length is calculated in line: event->mmap.len = pos->end - pos->start; and results in 0xffffffffffffffff - 0x3ff804d6990(*) = 0xfffffc007fb2966f (*) On s390 the module start address is actually determined by a __weak function named arch__fix_module_text_start() in machine__create_module(). I think this improvable. We can use the module size (2nd column of /proc/modules) to get each loaded kernel module size and calculate its end address. Only for map entries which do not have a valid end address (end is still zero) we can use the heuristic we have now, that is use successor start address or ~0. Signed-off-by: Thomas-Mich Richter <tmricht@linux.vnet.ibm.com> Reviewed-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Cc: Thomas-Mich Richter <tmricht@linux.vnet.ibm.com> Cc: Zvonko Kosic <zvonko.kosic@de.ibm.com> LPU-Reference: 20170803134902.47207-2-tmricht@linux.vnet.ibm.com Link: http://lkml.kernel.org/n/tip-nmoqij5b5vxx7rq2ckwu8iaj@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-08-03 21:49:02 +08:00
map->end = start + size;
dso__kernel_module_get_build_id(map->dso, machine->root_dir);
return 0;
}
static int machine__create_modules(struct machine *machine)
{
const char *modules;
char path[PATH_MAX];
if (machine__is_default_guest(machine)) {
modules = symbol_conf.default_guest_modules;
} else {
snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
modules = path;
}
if (symbol__restricted_filename(modules, "/proc/modules"))
return -1;
if (modules__parse(modules, machine, machine__create_module))
return -1;
if (!machine__set_modules_path(machine))
return 0;
pr_debug("Problems setting modules path maps, continuing anyway...\n");
return 0;
}
int machine__create_kernel_maps(struct machine *machine)
{
struct dso *kernel = machine__get_kernel(machine);
perf symbols: Accept symbols starting at address 0 That is the case of _text on s390, and we have some functions that return an address, using address zero to report problems, oops. This would lead the symbol loading routines to not use "_text" as the reference relocation symbol, or the first symbol for the kernel, but use instead "_stext", that is at the same address on x86_64 and others, but not on s390: [acme@localhost perf-4.11.0-rc6]$ head -15 /proc/kallsyms 0000000000000000 T _text 0000000000000418 t iplstart 0000000000000800 T start 000000000000080a t .base 000000000000082e t .sk8x8 0000000000000834 t .gotr 0000000000000842 t .cmd 0000000000000846 t .parm 000000000000084a t .lowcase 0000000000010000 T startup 0000000000010010 T startup_kdump 0000000000010214 t startup_kdump_relocated 0000000000011000 T startup_continue 00000000000112a0 T _ehead 0000000000100000 T _stext [acme@localhost perf-4.11.0-rc6]$ Which in turn would make 'perf test vmlinux' to fail because it wouldn't find the symbols before "_stext" in kallsyms. Fix it by using the return value only for errors and storing the address, when the symbol is successfully found, in a provided pointer arg. Before this patch: After: [acme@localhost perf-4.11.0-rc6]$ tools/perf/perf test -v 1 1: vmlinux symtab matches kallsyms : --- start --- test child forked, pid 40693 Looking at the vmlinux_path (8 entries long) Using /usr/lib/debug/lib/modules/3.10.0-654.el7.s390x/vmlinux for symbols ERR : 0: _text not on kallsyms ERR : 0x418: iplstart not on kallsyms ERR : 0x800: start not on kallsyms ERR : 0x80a: .base not on kallsyms ERR : 0x82e: .sk8x8 not on kallsyms ERR : 0x834: .gotr not on kallsyms ERR : 0x842: .cmd not on kallsyms ERR : 0x846: .parm not on kallsyms ERR : 0x84a: .lowcase not on kallsyms ERR : 0x10000: startup not on kallsyms ERR : 0x10010: startup_kdump not on kallsyms ERR : 0x10214: startup_kdump_relocated not on kallsyms ERR : 0x11000: startup_continue not on kallsyms ERR : 0x112a0: _ehead not on kallsyms <SNIP warnings> test child finished with -1 ---- end ---- vmlinux symtab matches kallsyms: FAILED! [acme@localhost perf-4.11.0-rc6]$ After: [acme@localhost perf-4.11.0-rc6]$ tools/perf/perf test -v 1 1: vmlinux symtab matches kallsyms : --- start --- test child forked, pid 47160 <SNIP warnings> test child finished with 0 ---- end ---- vmlinux symtab matches kallsyms: Ok [acme@localhost perf-4.11.0-rc6]$ Reported-by: Michael Petlan <mpetlan@redhat.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Link: http://lkml.kernel.org/n/tip-9x9bwgd3btwdk1u51xie93fz@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-04-28 08:21:09 +08:00
const char *name = NULL;
u64 addr = 0;
perf tools: Fix machine__create_kernel_maps to put kernel dso refcount Fix machine__create_kernel_maps() to put kernel dso because the dso has been gotten via __machine__create_kernel_maps(). Refcnt debugger shows: ==== [0] ==== Unreclaimed dso: 0x3036ab0 Refcount +1 => 1 at ./perf(dso__new+0x1ff) [0x4a62df] ./perf(__dsos__addnew+0x29) [0x4a6e19] ./perf(dsos__findnew+0xd1) [0x4a7181] ./perf(machine__findnew_kernel+0x27) [0x4a5e17] ./perf() [0x4b8cf2] ./perf(machine__create_kernel_maps+0x28) [0x4bb428] ./perf(machine__new_host+0xfa) [0x4bb74a] ./perf(init_probe_symbol_maps+0x93) [0x506613] ./perf() [0x455ffa] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7ffa6809eaf5] ./perf() [0x4220a9] [snip] Refcount +1 => 2 at ./perf(dsos__findnew+0x7e) [0x4a712e] ./perf(machine__findnew_kernel+0x27) [0x4a5e17] ./perf() [0x4b8cf2] ./perf(machine__create_kernel_maps+0x28) [0x4bb428] ./perf(machine__new_host+0xfa) [0x4bb74a] ./perf(init_probe_symbol_maps+0x93) [0x506613] ./perf() [0x455ffa] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7ffa6809eaf5] ./perf() [0x4220a9] [snip] Refcount -1 => 1 at ./perf(dso__put+0x2f) [0x4a664f] ./perf(machine__delete+0xfe) [0x4b93ee] ./perf(exit_probe_symbol_maps+0x28) [0x5066b8] ./perf() [0x45628a] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7ffa6809eaf5] ./perf() [0x4220a9] Actually, dsos__findnew gets the dso before returning it, so the dso user (in this case machine__create_kernel_maps) has to put the dso after used. Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/20151118064033.30709.98954.stgit@localhost.localdomain Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-11-18 14:40:33 +08:00
int ret;
if (kernel == NULL)
return -1;
perf tools: Fix machine__create_kernel_maps to put kernel dso refcount Fix machine__create_kernel_maps() to put kernel dso because the dso has been gotten via __machine__create_kernel_maps(). Refcnt debugger shows: ==== [0] ==== Unreclaimed dso: 0x3036ab0 Refcount +1 => 1 at ./perf(dso__new+0x1ff) [0x4a62df] ./perf(__dsos__addnew+0x29) [0x4a6e19] ./perf(dsos__findnew+0xd1) [0x4a7181] ./perf(machine__findnew_kernel+0x27) [0x4a5e17] ./perf() [0x4b8cf2] ./perf(machine__create_kernel_maps+0x28) [0x4bb428] ./perf(machine__new_host+0xfa) [0x4bb74a] ./perf(init_probe_symbol_maps+0x93) [0x506613] ./perf() [0x455ffa] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7ffa6809eaf5] ./perf() [0x4220a9] [snip] Refcount +1 => 2 at ./perf(dsos__findnew+0x7e) [0x4a712e] ./perf(machine__findnew_kernel+0x27) [0x4a5e17] ./perf() [0x4b8cf2] ./perf(machine__create_kernel_maps+0x28) [0x4bb428] ./perf(machine__new_host+0xfa) [0x4bb74a] ./perf(init_probe_symbol_maps+0x93) [0x506613] ./perf() [0x455ffa] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7ffa6809eaf5] ./perf() [0x4220a9] [snip] Refcount -1 => 1 at ./perf(dso__put+0x2f) [0x4a664f] ./perf(machine__delete+0xfe) [0x4b93ee] ./perf(exit_probe_symbol_maps+0x28) [0x5066b8] ./perf() [0x45628a] ./perf(cmd_probe+0x6c) [0x4566bc] ./perf() [0x47abc5] ./perf(main+0x610) [0x421f90] /lib64/libc.so.6(__libc_start_main+0xf5) [0x7ffa6809eaf5] ./perf() [0x4220a9] Actually, dsos__findnew gets the dso before returning it, so the dso user (in this case machine__create_kernel_maps) has to put the dso after used. Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/20151118064033.30709.98954.stgit@localhost.localdomain Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-11-18 14:40:33 +08:00
ret = __machine__create_kernel_maps(machine, kernel);
dso__put(kernel);
if (ret < 0)
return -1;
if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
if (machine__is_host(machine))
pr_debug("Problems creating module maps, "
"continuing anyway...\n");
else
pr_debug("Problems creating module maps for guest %d, "
"continuing anyway...\n", machine->pid);
}
/*
* Now that we have all the maps created, just set the ->end of them:
*/
map_groups__fixup_end(&machine->kmaps);
perf machine: Fix segfault for kernel.kptr_restrict=2 Michael reported the segfault when kernel.kptr_restrict=2 is set. $ perf record ls ... perf: Segmentation fault Obtained 16 stack frames. ./perf(dump_stack+0x2d) [0x5068df] ./perf(sighandler_dump_stack+0x2d) [0x5069bf] ./perf() [0x43e47b] /lib64/libc.so.6(+0x3594f) [0x7f762004794f] /lib64/libc.so.6(strlen+0x26) [0x7f762009ef86] /lib64/libc.so.6(__strdup+0xd) [0x7f762009ecbd] ./perf(maps__set_kallsyms_ref_reloc_sym+0x4d) [0x51590f] ./perf(machine__create_kernel_maps+0x136) [0x50a7de] ./perf(perf_session__create_kernel_maps+0x2c) [0x510a81] ./perf(perf_session__new+0x13d) [0x510e23] ./perf() [0x43fd61] ./perf(cmd_record+0x704) [0x441823] ./perf() [0x4bc1a0] ./perf() [0x4bc40d] ./perf() [0x4bc55f] ./perf(main+0x2d5) [0x4bc939] Segmentation fault (core dumped) The reason is that with kernel.kptr_restrict=2, we don't get the symbol from machine__get_running_kernel_start, which we want to use in maps__set_kallsyms_ref_reloc_sym and we crash. Check the symbol name value before calling maps__set_kallsyms_ref_reloc_sym() and succeed without ref_reloc_sym being set. It's safe because we check its existence before we use it. Reported-by: Michael Petlan <mpetlan@redhat.com> Signed-off-by: Jiri Olsa <jolsa@kernel.org> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: David Ahern <dsahern@gmail.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/20170626095153.553-1-jolsa@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-06-26 17:51:53 +08:00
if (!machine__get_running_kernel_start(machine, &name, &addr)) {
if (name &&
maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps, name, addr)) {
machine__destroy_kernel_maps(machine);
return -1;
}
}
return 0;
}
static void machine__set_kernel_mmap_len(struct machine *machine,
union perf_event *event)
{
int i;
for (i = 0; i < MAP__NR_TYPES; i++) {
machine->vmlinux_maps[i]->start = event->mmap.start;
machine->vmlinux_maps[i]->end = (event->mmap.start +
event->mmap.len);
/*
* Be a bit paranoid here, some perf.data file came with
* a zero sized synthesized MMAP event for the kernel.
*/
if (machine->vmlinux_maps[i]->end == 0)
machine->vmlinux_maps[i]->end = ~0ULL;
}
}
static bool machine__uses_kcore(struct machine *machine)
{
struct dso *dso;
list_for_each_entry(dso, &machine->dsos.head, node) {
if (dso__is_kcore(dso))
return true;
}
return false;
}
static int machine__process_kernel_mmap_event(struct machine *machine,
union perf_event *event)
{
struct map *map;
char kmmap_prefix[PATH_MAX];
enum dso_kernel_type kernel_type;
bool is_kernel_mmap;
/* If we have maps from kcore then we do not need or want any others */
if (machine__uses_kcore(machine))
return 0;
machine__mmap_name(machine, kmmap_prefix, sizeof(kmmap_prefix));
if (machine__is_host(machine))
kernel_type = DSO_TYPE_KERNEL;
else
kernel_type = DSO_TYPE_GUEST_KERNEL;
is_kernel_mmap = memcmp(event->mmap.filename,
kmmap_prefix,
strlen(kmmap_prefix) - 1) == 0;
if (event->mmap.filename[0] == '/' ||
(!is_kernel_mmap && event->mmap.filename[0] == '[')) {
map = machine__findnew_module_map(machine, event->mmap.start,
event->mmap.filename);
if (map == NULL)
goto out_problem;
map->end = map->start + event->mmap.len;
} else if (is_kernel_mmap) {
const char *symbol_name = (event->mmap.filename +
strlen(kmmap_prefix));
/*
* Should be there already, from the build-id table in
* the header.
*/
perf tools: Fix build-id matching on vmlinux There's a problem on finding correct kernel symbols when perf report runs on a different kernel. Although a part of the problem was solved by the prior commit 0a7e6d1b6844 ("perf tools: Check recorded kernel version when finding vmlinux"), there's a remaining problem still. When perf records samples, it synthesizes the kernel map using machine__mmap_name() and ref_reloc_sym like "[kernel.kallsyms]_text". You can easily see it using 'perf report -D' command. After finishing record, it goes through the recorded events to find maps/dsos actually used. And then record build-id info of them. During this process, it needs to load symbols in a dso and it'd call dso__load_vmlinux_path() since the default value of the symbol_conf. try_vmlinux_path is true. However it changes dso->long_name to a real path of the vmlinux file (e.g. /lib/modules/3.16.4/build/vmlinux) if one is running on a custom kernel. It resulted in that perf report reads the build-id of the vmlinux, but cannot use it since it only knows about the [kernel.kallsyms] map. It then falls back to possible vmlinux paths by using the recorded kernel version (in case of a recent version) or a running kernel silently. Even with the recent tools, this still has a possibility of breaking the result. As the build directory is a symbolic link, if one built a new kernel in the same directory with different source/config, the old link to vmlinux will point the new file. So it's absolutely needed to use build-id when finding a kernel image. In this patch, it's now changed to try to search a kernel dso in the existing dso list which was constructed during build-id table parsing so it'll always have a build-id. If not found, search "[kernel.kallsyms]". Before: $ perf report # Children Self Command Shared Object Symbol # ........ ........ ....... ................. ............................... # 72.15% 0.00% swapper [kernel.kallsyms] [k] set_curr_task_rt 72.15% 0.00% swapper [kernel.kallsyms] [k] native_calibrate_tsc 72.15% 0.00% swapper [kernel.kallsyms] [k] tsc_refine_calibration_work 71.87% 71.87% swapper [kernel.kallsyms] [k] module_finalize ... After (for the same perf.data): 72.15% 0.00% swapper vmlinux [k] cpu_startup_entry 72.15% 0.00% swapper vmlinux [k] arch_cpu_idle 72.15% 0.00% swapper vmlinux [k] default_idle 71.87% 71.87% swapper vmlinux [k] native_safe_halt ... Signed-off-by: Namhyung Kim <namhyung@kernel.org> Acked-by: Ingo Molnar <mingo@kernel.org> Link: http://lkml.kernel.org/r/20140924073356.GB1962@gmail.com Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung.kim@lge.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1415063674-17206-8-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-04 09:14:33 +08:00
struct dso *kernel = NULL;
struct dso *dso;
pthread_rwlock_rdlock(&machine->dsos.lock);
list_for_each_entry(dso, &machine->dsos.head, node) {
perf tools: Deal with kernel module names in '[]' correctly Before patch ba92732e9808 ('perf kmaps: Check kmaps to make code more robust'), 'perf report' and 'perf annotate' will segfault if trace data contains kernel module information like this: # perf report -D -i ./perf.data ... 0 0 0x188 [0x50]: PERF_RECORD_MMAP -1/0: [0xffffffbff1018000(0xf068000) @ 0]: x [test_module] ... # perf report -i ./perf.data --objdump=/path/to/objdump --kallsyms=/path/to/kallsyms perf: Segmentation fault -------- backtrace -------- /path/to/perf[0x503478] /lib64/libc.so.6(+0x3545f)[0x7fb201f3745f] /path/to/perf[0x499b56] /path/to/perf(dso__load_kallsyms+0x13c)[0x49b56c] /path/to/perf(dso__load+0x72e)[0x49c21e] /path/to/perf(map__load+0x6e)[0x4ae9ee] /path/to/perf(thread__find_addr_map+0x24c)[0x47deec] /path/to/perf(perf_event__preprocess_sample+0x88)[0x47e238] /path/to/perf[0x43ad02] /path/to/perf[0x4b55bc] /path/to/perf(ordered_events__flush+0xca)[0x4b57ea] /path/to/perf[0x4b1a01] /path/to/perf(perf_session__process_events+0x3be)[0x4b428e] /path/to/perf(cmd_report+0xf11)[0x43bfc1] /path/to/perf[0x474702] /path/to/perf(main+0x5f5)[0x42de95] /lib64/libc.so.6(__libc_start_main+0xf4)[0x7fb201f23bd4] /path/to/perf[0x42dfc4] This is because __kmod_path__parse treats '[' leading names as kernel name instead of names of kernel module. If perf.data contains build information and the buildid of such modules can be found, the dso->kernel of it will be set to DSO_TYPE_KERNEL by __event_process_build_id(), not kernel module. It will then be passed to dso__load() -> dso__load_kernel_sym() -> dso__load_kcore() if --kallsyms is provided. The refered patch adds NULL pointer checker to avoid segfault. However, such kernel modules are still processed incorrectly. This patch fixes __kmod_path__parse, makes it treat names like '[test_module]' as kernel modules. kmod-path.c is also update to reflect the above changes. Signed-off-by: Wang Nan <wangnan0@huawei.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Zefan Li <lizefan@huawei.com> Link: http://lkml.kernel.org/r/1433321541-170245-1-git-send-email-wangnan0@huawei.com [ Fixed the merged with 0443f36b0de0 ("perf machine: Fix the search for the kernel DSO on the unified list" ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-06-03 16:52:21 +08:00
/*
* The cpumode passed to is_kernel_module is not the
* cpumode of *this* event. If we insist on passing
* correct cpumode to is_kernel_module, we should
* record the cpumode when we adding this dso to the
* linked list.
*
* However we don't really need passing correct
* cpumode. We know the correct cpumode must be kernel
* mode (if not, we should not link it onto kernel_dsos
* list).
*
* Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
* is_kernel_module() treats it as a kernel cpumode.
*/
if (!dso->kernel ||
is_kernel_module(dso->long_name,
PERF_RECORD_MISC_CPUMODE_UNKNOWN))
perf tools: Fix build-id matching on vmlinux There's a problem on finding correct kernel symbols when perf report runs on a different kernel. Although a part of the problem was solved by the prior commit 0a7e6d1b6844 ("perf tools: Check recorded kernel version when finding vmlinux"), there's a remaining problem still. When perf records samples, it synthesizes the kernel map using machine__mmap_name() and ref_reloc_sym like "[kernel.kallsyms]_text". You can easily see it using 'perf report -D' command. After finishing record, it goes through the recorded events to find maps/dsos actually used. And then record build-id info of them. During this process, it needs to load symbols in a dso and it'd call dso__load_vmlinux_path() since the default value of the symbol_conf. try_vmlinux_path is true. However it changes dso->long_name to a real path of the vmlinux file (e.g. /lib/modules/3.16.4/build/vmlinux) if one is running on a custom kernel. It resulted in that perf report reads the build-id of the vmlinux, but cannot use it since it only knows about the [kernel.kallsyms] map. It then falls back to possible vmlinux paths by using the recorded kernel version (in case of a recent version) or a running kernel silently. Even with the recent tools, this still has a possibility of breaking the result. As the build directory is a symbolic link, if one built a new kernel in the same directory with different source/config, the old link to vmlinux will point the new file. So it's absolutely needed to use build-id when finding a kernel image. In this patch, it's now changed to try to search a kernel dso in the existing dso list which was constructed during build-id table parsing so it'll always have a build-id. If not found, search "[kernel.kallsyms]". Before: $ perf report # Children Self Command Shared Object Symbol # ........ ........ ....... ................. ............................... # 72.15% 0.00% swapper [kernel.kallsyms] [k] set_curr_task_rt 72.15% 0.00% swapper [kernel.kallsyms] [k] native_calibrate_tsc 72.15% 0.00% swapper [kernel.kallsyms] [k] tsc_refine_calibration_work 71.87% 71.87% swapper [kernel.kallsyms] [k] module_finalize ... After (for the same perf.data): 72.15% 0.00% swapper vmlinux [k] cpu_startup_entry 72.15% 0.00% swapper vmlinux [k] arch_cpu_idle 72.15% 0.00% swapper vmlinux [k] default_idle 71.87% 71.87% swapper vmlinux [k] native_safe_halt ... Signed-off-by: Namhyung Kim <namhyung@kernel.org> Acked-by: Ingo Molnar <mingo@kernel.org> Link: http://lkml.kernel.org/r/20140924073356.GB1962@gmail.com Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung.kim@lge.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1415063674-17206-8-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-04 09:14:33 +08:00
continue;
perf tools: Deal with kernel module names in '[]' correctly Before patch ba92732e9808 ('perf kmaps: Check kmaps to make code more robust'), 'perf report' and 'perf annotate' will segfault if trace data contains kernel module information like this: # perf report -D -i ./perf.data ... 0 0 0x188 [0x50]: PERF_RECORD_MMAP -1/0: [0xffffffbff1018000(0xf068000) @ 0]: x [test_module] ... # perf report -i ./perf.data --objdump=/path/to/objdump --kallsyms=/path/to/kallsyms perf: Segmentation fault -------- backtrace -------- /path/to/perf[0x503478] /lib64/libc.so.6(+0x3545f)[0x7fb201f3745f] /path/to/perf[0x499b56] /path/to/perf(dso__load_kallsyms+0x13c)[0x49b56c] /path/to/perf(dso__load+0x72e)[0x49c21e] /path/to/perf(map__load+0x6e)[0x4ae9ee] /path/to/perf(thread__find_addr_map+0x24c)[0x47deec] /path/to/perf(perf_event__preprocess_sample+0x88)[0x47e238] /path/to/perf[0x43ad02] /path/to/perf[0x4b55bc] /path/to/perf(ordered_events__flush+0xca)[0x4b57ea] /path/to/perf[0x4b1a01] /path/to/perf(perf_session__process_events+0x3be)[0x4b428e] /path/to/perf(cmd_report+0xf11)[0x43bfc1] /path/to/perf[0x474702] /path/to/perf(main+0x5f5)[0x42de95] /lib64/libc.so.6(__libc_start_main+0xf4)[0x7fb201f23bd4] /path/to/perf[0x42dfc4] This is because __kmod_path__parse treats '[' leading names as kernel name instead of names of kernel module. If perf.data contains build information and the buildid of such modules can be found, the dso->kernel of it will be set to DSO_TYPE_KERNEL by __event_process_build_id(), not kernel module. It will then be passed to dso__load() -> dso__load_kernel_sym() -> dso__load_kcore() if --kallsyms is provided. The refered patch adds NULL pointer checker to avoid segfault. However, such kernel modules are still processed incorrectly. This patch fixes __kmod_path__parse, makes it treat names like '[test_module]' as kernel modules. kmod-path.c is also update to reflect the above changes. Signed-off-by: Wang Nan <wangnan0@huawei.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Zefan Li <lizefan@huawei.com> Link: http://lkml.kernel.org/r/1433321541-170245-1-git-send-email-wangnan0@huawei.com [ Fixed the merged with 0443f36b0de0 ("perf machine: Fix the search for the kernel DSO on the unified list" ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-06-03 16:52:21 +08:00
perf tools: Fix build-id matching on vmlinux There's a problem on finding correct kernel symbols when perf report runs on a different kernel. Although a part of the problem was solved by the prior commit 0a7e6d1b6844 ("perf tools: Check recorded kernel version when finding vmlinux"), there's a remaining problem still. When perf records samples, it synthesizes the kernel map using machine__mmap_name() and ref_reloc_sym like "[kernel.kallsyms]_text". You can easily see it using 'perf report -D' command. After finishing record, it goes through the recorded events to find maps/dsos actually used. And then record build-id info of them. During this process, it needs to load symbols in a dso and it'd call dso__load_vmlinux_path() since the default value of the symbol_conf. try_vmlinux_path is true. However it changes dso->long_name to a real path of the vmlinux file (e.g. /lib/modules/3.16.4/build/vmlinux) if one is running on a custom kernel. It resulted in that perf report reads the build-id of the vmlinux, but cannot use it since it only knows about the [kernel.kallsyms] map. It then falls back to possible vmlinux paths by using the recorded kernel version (in case of a recent version) or a running kernel silently. Even with the recent tools, this still has a possibility of breaking the result. As the build directory is a symbolic link, if one built a new kernel in the same directory with different source/config, the old link to vmlinux will point the new file. So it's absolutely needed to use build-id when finding a kernel image. In this patch, it's now changed to try to search a kernel dso in the existing dso list which was constructed during build-id table parsing so it'll always have a build-id. If not found, search "[kernel.kallsyms]". Before: $ perf report # Children Self Command Shared Object Symbol # ........ ........ ....... ................. ............................... # 72.15% 0.00% swapper [kernel.kallsyms] [k] set_curr_task_rt 72.15% 0.00% swapper [kernel.kallsyms] [k] native_calibrate_tsc 72.15% 0.00% swapper [kernel.kallsyms] [k] tsc_refine_calibration_work 71.87% 71.87% swapper [kernel.kallsyms] [k] module_finalize ... After (for the same perf.data): 72.15% 0.00% swapper vmlinux [k] cpu_startup_entry 72.15% 0.00% swapper vmlinux [k] arch_cpu_idle 72.15% 0.00% swapper vmlinux [k] default_idle 71.87% 71.87% swapper vmlinux [k] native_safe_halt ... Signed-off-by: Namhyung Kim <namhyung@kernel.org> Acked-by: Ingo Molnar <mingo@kernel.org> Link: http://lkml.kernel.org/r/20140924073356.GB1962@gmail.com Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung.kim@lge.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1415063674-17206-8-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-04 09:14:33 +08:00
kernel = dso;
break;
}
pthread_rwlock_unlock(&machine->dsos.lock);
perf tools: Fix build-id matching on vmlinux There's a problem on finding correct kernel symbols when perf report runs on a different kernel. Although a part of the problem was solved by the prior commit 0a7e6d1b6844 ("perf tools: Check recorded kernel version when finding vmlinux"), there's a remaining problem still. When perf records samples, it synthesizes the kernel map using machine__mmap_name() and ref_reloc_sym like "[kernel.kallsyms]_text". You can easily see it using 'perf report -D' command. After finishing record, it goes through the recorded events to find maps/dsos actually used. And then record build-id info of them. During this process, it needs to load symbols in a dso and it'd call dso__load_vmlinux_path() since the default value of the symbol_conf. try_vmlinux_path is true. However it changes dso->long_name to a real path of the vmlinux file (e.g. /lib/modules/3.16.4/build/vmlinux) if one is running on a custom kernel. It resulted in that perf report reads the build-id of the vmlinux, but cannot use it since it only knows about the [kernel.kallsyms] map. It then falls back to possible vmlinux paths by using the recorded kernel version (in case of a recent version) or a running kernel silently. Even with the recent tools, this still has a possibility of breaking the result. As the build directory is a symbolic link, if one built a new kernel in the same directory with different source/config, the old link to vmlinux will point the new file. So it's absolutely needed to use build-id when finding a kernel image. In this patch, it's now changed to try to search a kernel dso in the existing dso list which was constructed during build-id table parsing so it'll always have a build-id. If not found, search "[kernel.kallsyms]". Before: $ perf report # Children Self Command Shared Object Symbol # ........ ........ ....... ................. ............................... # 72.15% 0.00% swapper [kernel.kallsyms] [k] set_curr_task_rt 72.15% 0.00% swapper [kernel.kallsyms] [k] native_calibrate_tsc 72.15% 0.00% swapper [kernel.kallsyms] [k] tsc_refine_calibration_work 71.87% 71.87% swapper [kernel.kallsyms] [k] module_finalize ... After (for the same perf.data): 72.15% 0.00% swapper vmlinux [k] cpu_startup_entry 72.15% 0.00% swapper vmlinux [k] arch_cpu_idle 72.15% 0.00% swapper vmlinux [k] default_idle 71.87% 71.87% swapper vmlinux [k] native_safe_halt ... Signed-off-by: Namhyung Kim <namhyung@kernel.org> Acked-by: Ingo Molnar <mingo@kernel.org> Link: http://lkml.kernel.org/r/20140924073356.GB1962@gmail.com Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung.kim@lge.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1415063674-17206-8-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-04 09:14:33 +08:00
if (kernel == NULL)
kernel = machine__findnew_dso(machine, kmmap_prefix);
if (kernel == NULL)
goto out_problem;
kernel->kernel = kernel_type;
if (__machine__create_kernel_maps(machine, kernel) < 0) {
dso__put(kernel);
goto out_problem;
}
if (strstr(kernel->long_name, "vmlinux"))
dso__set_short_name(kernel, "[kernel.vmlinux]", false);
perf tools: Make vmlinux short name more like kallsyms short name The previous patch changed kernel dso name from '[kernel.kallsyms]' to vmlinux. However it might add confusion to old users accustomed to the old name. So change the short name to '[kernel.vmlinux]' to reduce such confusion. Before: # Overhead Command Shared Object Symbol # ........ .............. ....................... ............................... # 9.83% swapper vmlinux [k] intel_idle 4.10% awk libc-2.20.so [.] __strcmp_sse2 1.86% sed libc-2.20.so [.] __strcmp_sse2 1.78% netctl-auto libc-2.20.so [.] __strcmp_sse2 1.23% netctl-auto libc-2.20.so [.] __mbrtowc 1.21% firefox libxul.so [.] 0x00000000024b62bd 1.20% swapper vmlinux [k] cpuidle_enter_state 1.03% sleep vmlinux [k] copy_user_generic_unrolled After: # Overhead Command Shared Object Symbol # ........ .............. ....................... ............................... # 9.83% swapper [kernel.vmlinux] [k] intel_idle 4.10% awk libc-2.20.so [.] __strcmp_sse2 1.86% sed libc-2.20.so [.] __strcmp_sse2 1.78% netctl-auto libc-2.20.so [.] __strcmp_sse2 1.23% netctl-auto libc-2.20.so [.] __mbrtowc 1.21% firefox libxul.so [.] 0x00000000024b62bd 1.20% swapper [kernel.vmlinux] [k] cpuidle_enter_state 1.03% sleep [kernel.vmlinux] [k] copy_user_generic_unrolled Signed-off-by: Namhyung Kim <namhyung@kernel.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung.kim@lge.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1415063674-17206-9-git-send-email-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-04 09:14:34 +08:00
machine__set_kernel_mmap_len(machine, event);
/*
* Avoid using a zero address (kptr_restrict) for the ref reloc
* symbol. Effectively having zero here means that at record
* time /proc/sys/kernel/kptr_restrict was non zero.
*/
if (event->mmap.pgoff != 0) {
maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps,
symbol_name,
event->mmap.pgoff);
}
if (machine__is_default_guest(machine)) {
/*
* preload dso of guest kernel and modules
*/
dso__load(kernel, machine__kernel_map(machine));
}
}
return 0;
out_problem:
return -1;
}
int machine__process_mmap2_event(struct machine *machine,
union perf_event *event,
struct perf_sample *sample)
{
struct thread *thread;
struct map *map;
enum map_type type;
int ret = 0;
if (dump_trace)
perf_event__fprintf_mmap2(event, stdout);
if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
sample->cpumode == PERF_RECORD_MISC_KERNEL) {
ret = machine__process_kernel_mmap_event(machine, event);
if (ret < 0)
goto out_problem;
return 0;
}
thread = machine__findnew_thread(machine, event->mmap2.pid,
event->mmap2.tid);
if (thread == NULL)
goto out_problem;
if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
type = MAP__VARIABLE;
else
type = MAP__FUNCTION;
map = map__new(machine, event->mmap2.start,
event->mmap2.len, event->mmap2.pgoff,
event->mmap2.maj,
event->mmap2.min, event->mmap2.ino,
event->mmap2.ino_generation,
event->mmap2.prot,
event->mmap2.flags,
event->mmap2.filename, type, thread);
if (map == NULL)
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
goto out_problem_map;
ret = thread__insert_map(thread, map);
if (ret)
goto out_problem_insert;
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
thread__put(thread);
map__put(map);
return 0;
out_problem_insert:
map__put(map);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
out_problem_map:
thread__put(thread);
out_problem:
dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
return 0;
}
int machine__process_mmap_event(struct machine *machine, union perf_event *event,
struct perf_sample *sample)
{
struct thread *thread;
struct map *map;
enum map_type type;
int ret = 0;
if (dump_trace)
perf_event__fprintf_mmap(event, stdout);
if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
sample->cpumode == PERF_RECORD_MISC_KERNEL) {
ret = machine__process_kernel_mmap_event(machine, event);
if (ret < 0)
goto out_problem;
return 0;
}
thread = machine__findnew_thread(machine, event->mmap.pid,
event->mmap.tid);
if (thread == NULL)
goto out_problem;
if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
type = MAP__VARIABLE;
else
type = MAP__FUNCTION;
map = map__new(machine, event->mmap.start,
event->mmap.len, event->mmap.pgoff,
0, 0, 0, 0, 0, 0,
event->mmap.filename,
type, thread);
if (map == NULL)
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
goto out_problem_map;
ret = thread__insert_map(thread, map);
if (ret)
goto out_problem_insert;
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
thread__put(thread);
map__put(map);
return 0;
out_problem_insert:
map__put(map);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
out_problem_map:
thread__put(thread);
out_problem:
dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
return 0;
}
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock)
perf tools: Sample after exit loses thread correlation Occassionally events (e.g., context-switch, sched tracepoints) are losing the conversion of sample data associated with a thread. For example: $ perf record -e sched:sched_switch -c 1 -a -- sleep 5 $ perf script <selected events shown> ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... :30482 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... The last line lost the conversion from tid to comm. If you look at the events (perf script -D) you see why - a SAMPLE event is generated after the EXIT: 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: :30482:30482 When perf processes the EXIT event the thread is moved to the dead_threads list. When the SAMPLE event is processed no thread exists for the pid so a new one is created by machine__findnew_thread. This patch address the problem by delaying the move to the dead_threads list until the tid is re-used (per Adrian's suggestion). With this patch we get the previous example shows: ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... and 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: ls:30482 v4: per Arnaldo's request add dead flag to thread struct and set when task exits v3: re-do from a time based check to a delayed move to dead_threads list v2: Rebased to latest perf/core branch. Changed time comparison to use a macro which explicitly shows the time basis Signed-off-by: David Ahern <dsahern@gmail.com> Acked-by: Adrian Hunter <adrian.hunter@intel.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1376491767-84171-1-git-send-email-dsahern@gmail.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-08-14 22:49:27 +08:00
{
if (machine->last_match == th)
machine->last_match = NULL;
BUG_ON(refcount_read(&th->refcnt) == 0);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
if (lock)
pthread_rwlock_wrlock(&machine->threads_lock);
rb_erase_init(&th->rb_node, &machine->threads);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
RB_CLEAR_NODE(&th->rb_node);
--machine->nr_threads;
perf tools: Sample after exit loses thread correlation Occassionally events (e.g., context-switch, sched tracepoints) are losing the conversion of sample data associated with a thread. For example: $ perf record -e sched:sched_switch -c 1 -a -- sleep 5 $ perf script <selected events shown> ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... :30482 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... The last line lost the conversion from tid to comm. If you look at the events (perf script -D) you see why - a SAMPLE event is generated after the EXIT: 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: :30482:30482 When perf processes the EXIT event the thread is moved to the dead_threads list. When the SAMPLE event is processed no thread exists for the pid so a new one is created by machine__findnew_thread. This patch address the problem by delaying the move to the dead_threads list until the tid is re-used (per Adrian's suggestion). With this patch we get the previous example shows: ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... and 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: ls:30482 v4: per Arnaldo's request add dead flag to thread struct and set when task exits v3: re-do from a time based check to a delayed move to dead_threads list v2: Rebased to latest perf/core branch. Changed time comparison to use a macro which explicitly shows the time basis Signed-off-by: David Ahern <dsahern@gmail.com> Acked-by: Adrian Hunter <adrian.hunter@intel.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1376491767-84171-1-git-send-email-dsahern@gmail.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-08-14 22:49:27 +08:00
/*
* Move it first to the dead_threads list, then drop the reference,
* if this is the last reference, then the thread__delete destructor
* will be called and we will remove it from the dead_threads list.
perf tools: Sample after exit loses thread correlation Occassionally events (e.g., context-switch, sched tracepoints) are losing the conversion of sample data associated with a thread. For example: $ perf record -e sched:sched_switch -c 1 -a -- sleep 5 $ perf script <selected events shown> ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... :30482 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... The last line lost the conversion from tid to comm. If you look at the events (perf script -D) you see why - a SAMPLE event is generated after the EXIT: 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: :30482:30482 When perf processes the EXIT event the thread is moved to the dead_threads list. When the SAMPLE event is processed no thread exists for the pid so a new one is created by machine__findnew_thread. This patch address the problem by delaying the move to the dead_threads list until the tid is re-used (per Adrian's suggestion). With this patch we get the previous example shows: ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... and 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: ls:30482 v4: per Arnaldo's request add dead flag to thread struct and set when task exits v3: re-do from a time based check to a delayed move to dead_threads list v2: Rebased to latest perf/core branch. Changed time comparison to use a macro which explicitly shows the time basis Signed-off-by: David Ahern <dsahern@gmail.com> Acked-by: Adrian Hunter <adrian.hunter@intel.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1376491767-84171-1-git-send-email-dsahern@gmail.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-08-14 22:49:27 +08:00
*/
list_add_tail(&th->node, &machine->dead_threads);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
if (lock)
pthread_rwlock_unlock(&machine->threads_lock);
thread__put(th);
perf tools: Sample after exit loses thread correlation Occassionally events (e.g., context-switch, sched tracepoints) are losing the conversion of sample data associated with a thread. For example: $ perf record -e sched:sched_switch -c 1 -a -- sleep 5 $ perf script <selected events shown> ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... :30482 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... The last line lost the conversion from tid to comm. If you look at the events (perf script -D) you see why - a SAMPLE event is generated after the EXIT: 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: :30482:30482 When perf processes the EXIT event the thread is moved to the dead_threads list. When the SAMPLE event is processed no thread exists for the pid so a new one is created by machine__findnew_thread. This patch address the problem by delaying the move to the dead_threads list until the tid is re-used (per Adrian's suggestion). With this patch we get the previous example shows: ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... and 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: ls:30482 v4: per Arnaldo's request add dead flag to thread struct and set when task exits v3: re-do from a time based check to a delayed move to dead_threads list v2: Rebased to latest perf/core branch. Changed time comparison to use a macro which explicitly shows the time basis Signed-off-by: David Ahern <dsahern@gmail.com> Acked-by: Adrian Hunter <adrian.hunter@intel.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1376491767-84171-1-git-send-email-dsahern@gmail.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-08-14 22:49:27 +08:00
}
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
void machine__remove_thread(struct machine *machine, struct thread *th)
{
return __machine__remove_thread(machine, th, true);
}
int machine__process_fork_event(struct machine *machine, union perf_event *event,
struct perf_sample *sample)
{
struct thread *thread = machine__find_thread(machine,
event->fork.pid,
event->fork.tid);
struct thread *parent = machine__findnew_thread(machine,
event->fork.ppid,
event->fork.ptid);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
int err = 0;
if (dump_trace)
perf_event__fprintf_task(event, stdout);
/*
* There may be an existing thread that is not actually the parent,
* either because we are processing events out of order, or because the
* (fork) event that would have removed the thread was lost. Assume the
* latter case and continue on as best we can.
*/
if (parent->pid_ != (pid_t)event->fork.ppid) {
dump_printf("removing erroneous parent thread %d/%d\n",
parent->pid_, parent->tid);
machine__remove_thread(machine, parent);
thread__put(parent);
parent = machine__findnew_thread(machine, event->fork.ppid,
event->fork.ptid);
}
perf tools: Sample after exit loses thread correlation Occassionally events (e.g., context-switch, sched tracepoints) are losing the conversion of sample data associated with a thread. For example: $ perf record -e sched:sched_switch -c 1 -a -- sleep 5 $ perf script <selected events shown> ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... :30482 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... The last line lost the conversion from tid to comm. If you look at the events (perf script -D) you see why - a SAMPLE event is generated after the EXIT: 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: :30482:30482 When perf processes the EXIT event the thread is moved to the dead_threads list. When the SAMPLE event is processed no thread exists for the pid so a new one is created by machine__findnew_thread. This patch address the problem by delaying the move to the dead_threads list until the tid is re-used (per Adrian's suggestion). With this patch we get the previous example shows: ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... and 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: ls:30482 v4: per Arnaldo's request add dead flag to thread struct and set when task exits v3: re-do from a time based check to a delayed move to dead_threads list v2: Rebased to latest perf/core branch. Changed time comparison to use a macro which explicitly shows the time basis Signed-off-by: David Ahern <dsahern@gmail.com> Acked-by: Adrian Hunter <adrian.hunter@intel.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1376491767-84171-1-git-send-email-dsahern@gmail.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-08-14 22:49:27 +08:00
/* if a thread currently exists for the thread id remove it */
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
if (thread != NULL) {
perf tools: Sample after exit loses thread correlation Occassionally events (e.g., context-switch, sched tracepoints) are losing the conversion of sample data associated with a thread. For example: $ perf record -e sched:sched_switch -c 1 -a -- sleep 5 $ perf script <selected events shown> ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... :30482 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... The last line lost the conversion from tid to comm. If you look at the events (perf script -D) you see why - a SAMPLE event is generated after the EXIT: 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: :30482:30482 When perf processes the EXIT event the thread is moved to the dead_threads list. When the SAMPLE event is processed no thread exists for the pid so a new one is created by machine__findnew_thread. This patch address the problem by delaying the move to the dead_threads list until the tid is re-used (per Adrian's suggestion). With this patch we get the previous example shows: ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... and 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: ls:30482 v4: per Arnaldo's request add dead flag to thread struct and set when task exits v3: re-do from a time based check to a delayed move to dead_threads list v2: Rebased to latest perf/core branch. Changed time comparison to use a macro which explicitly shows the time basis Signed-off-by: David Ahern <dsahern@gmail.com> Acked-by: Adrian Hunter <adrian.hunter@intel.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1376491767-84171-1-git-send-email-dsahern@gmail.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-08-14 22:49:27 +08:00
machine__remove_thread(machine, thread);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
thread__put(thread);
}
perf tools: Sample after exit loses thread correlation Occassionally events (e.g., context-switch, sched tracepoints) are losing the conversion of sample data associated with a thread. For example: $ perf record -e sched:sched_switch -c 1 -a -- sleep 5 $ perf script <selected events shown> ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... :30482 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... The last line lost the conversion from tid to comm. If you look at the events (perf script -D) you see why - a SAMPLE event is generated after the EXIT: 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: :30482:30482 When perf processes the EXIT event the thread is moved to the dead_threads list. When the SAMPLE event is processed no thread exists for the pid so a new one is created by machine__findnew_thread. This patch address the problem by delaying the move to the dead_threads list until the tid is re-used (per Adrian's suggestion). With this patch we get the previous example shows: ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... and 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: ls:30482 v4: per Arnaldo's request add dead flag to thread struct and set when task exits v3: re-do from a time based check to a delayed move to dead_threads list v2: Rebased to latest perf/core branch. Changed time comparison to use a macro which explicitly shows the time basis Signed-off-by: David Ahern <dsahern@gmail.com> Acked-by: Adrian Hunter <adrian.hunter@intel.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1376491767-84171-1-git-send-email-dsahern@gmail.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-08-14 22:49:27 +08:00
thread = machine__findnew_thread(machine, event->fork.pid,
event->fork.tid);
if (thread == NULL || parent == NULL ||
thread__fork(thread, parent, sample->time) < 0) {
dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
err = -1;
}
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
thread__put(thread);
thread__put(parent);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
return err;
}
int machine__process_exit_event(struct machine *machine, union perf_event *event,
struct perf_sample *sample __maybe_unused)
{
struct thread *thread = machine__find_thread(machine,
event->fork.pid,
event->fork.tid);
if (dump_trace)
perf_event__fprintf_task(event, stdout);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
if (thread != NULL) {
perf tools: Sample after exit loses thread correlation Occassionally events (e.g., context-switch, sched tracepoints) are losing the conversion of sample data associated with a thread. For example: $ perf record -e sched:sched_switch -c 1 -a -- sleep 5 $ perf script <selected events shown> ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... :30482 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... The last line lost the conversion from tid to comm. If you look at the events (perf script -D) you see why - a SAMPLE event is generated after the EXIT: 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: :30482:30482 When perf processes the EXIT event the thread is moved to the dead_threads list. When the SAMPLE event is processed no thread exists for the pid so a new one is created by machine__findnew_thread. This patch address the problem by delaying the move to the dead_threads list until the tid is re-used (per Adrian's suggestion). With this patch we get the previous example shows: ls 30482 [000] 1379727.583037: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.586339: sched:sched_switch: prev_comm=ls prev_pid=30482 ... ls 30482 [000] 1379727.589462: sched:sched_switch: prev_comm=ls prev_pid=30482 ... and 0 1379727589449774 0x1540b0 [0x38]: PERF_RECORD_EXIT(30482:30482):(30482:30482) 0 1379727589462497 0x1540e8 [0x80]: PERF_RECORD_SAMPLE(IP, 1): 30482/30482: 0xffffffff816416f1 period: 1 addr: 0 ... thread: ls:30482 v4: per Arnaldo's request add dead flag to thread struct and set when task exits v3: re-do from a time based check to a delayed move to dead_threads list v2: Rebased to latest perf/core branch. Changed time comparison to use a macro which explicitly shows the time basis Signed-off-by: David Ahern <dsahern@gmail.com> Acked-by: Adrian Hunter <adrian.hunter@intel.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1376491767-84171-1-git-send-email-dsahern@gmail.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2013-08-14 22:49:27 +08:00
thread__exited(thread);
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
thread__put(thread);
}
return 0;
}
int machine__process_event(struct machine *machine, union perf_event *event,
struct perf_sample *sample)
{
int ret;
switch (event->header.type) {
case PERF_RECORD_COMM:
ret = machine__process_comm_event(machine, event, sample); break;
case PERF_RECORD_MMAP:
ret = machine__process_mmap_event(machine, event, sample); break;
perf tools: Add PERF_RECORD_NAMESPACES to include namespaces related info Introduce a new option to record PERF_RECORD_NAMESPACES events emitted by the kernel when fork, clone, setns or unshare are invoked. And update perf-record documentation with the new option to record namespace events. Committer notes: Combined it with a later patch to allow printing it via 'perf report -D' and be able to test the feature introduced in this patch. Had to move here also perf_ns__name(), that was introduced in another later patch. Also used PRIu64 and PRIx64 to fix the build in some enfironments wrt: util/event.c:1129:39: error: format '%lx' expects argument of type 'long unsigned int', but argument 6 has type 'long long unsigned int' [-Werror=format=] ret += fprintf(fp, "%u/%s: %lu/0x%lx%s", idx ^ Testing it: # perf record --namespaces -a ^C[ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 1.083 MB perf.data (423 samples) ] # # perf report -D <SNIP> 3 2028902078892 0x115140 [0xa0]: PERF_RECORD_NAMESPACES 14783/14783 - nr_namespaces: 7 [0/net: 3/0xf0000081, 1/uts: 3/0xeffffffe, 2/ipc: 3/0xefffffff, 3/pid: 3/0xeffffffc, 4/user: 3/0xeffffffd, 5/mnt: 3/0xf0000000, 6/cgroup: 3/0xeffffffb] 0x1151e0 [0x30]: event: 9 . . ... raw event: size 48 bytes . 0000: 09 00 00 00 02 00 30 00 c4 71 82 68 0c 7f 00 00 ......0..q.h.... . 0010: a9 39 00 00 a9 39 00 00 94 28 fe 63 d8 01 00 00 .9...9...(.c.... . 0020: 03 00 00 00 00 00 00 00 ce c4 02 00 00 00 00 00 ................ <SNIP> NAMESPACES events: 1 <SNIP> # Signed-off-by: Hari Bathini <hbathini@linux.vnet.ibm.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexei Starovoitov <ast@fb.com> Cc: Ananth N Mavinakayanahalli <ananth@linux.vnet.ibm.com> Cc: Aravinda Prasad <aravinda@linux.vnet.ibm.com> Cc: Brendan Gregg <brendan.d.gregg@gmail.com> Cc: Daniel Borkmann <daniel@iogearbox.net> Cc: Eric Biederman <ebiederm@xmission.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sargun Dhillon <sargun@sargun.me> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/148891930386.25309.18412039920746995488.stgit@hbathini.in.ibm.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-03-08 04:41:43 +08:00
case PERF_RECORD_NAMESPACES:
ret = machine__process_namespaces_event(machine, event, sample); break;
case PERF_RECORD_MMAP2:
ret = machine__process_mmap2_event(machine, event, sample); break;
case PERF_RECORD_FORK:
ret = machine__process_fork_event(machine, event, sample); break;
case PERF_RECORD_EXIT:
ret = machine__process_exit_event(machine, event, sample); break;
case PERF_RECORD_LOST:
ret = machine__process_lost_event(machine, event, sample); break;
case PERF_RECORD_AUX:
ret = machine__process_aux_event(machine, event); break;
case PERF_RECORD_ITRACE_START:
ret = machine__process_itrace_start_event(machine, event); break;
perf tools: handle PERF_RECORD_LOST_SAMPLES This patch modifies the perf tool to handle the new RECORD type, PERF_RECORD_LOST_SAMPLES. The number of lost-sample events is stored in .nr_events[PERF_RECORD_LOST_SAMPLES]. The exact number of samples which the kernel dropped is stored in total_lost_samples. When the percentage of dropped samples is greater than 5%, a warning is printed. Here are some examples: Eg 1, Recording different frequently-occurring events is safe with the patch. Only a very low drop rate is associated with such actions. $ perf record -e '{cycles:p,instructions:p}' -c 20003 --no-time ~/tchain ~/tchain $ perf report -D | tail SAMPLE events: 120243 MMAP2 events: 5 LOST_SAMPLES events: 24 FINISHED_ROUND events: 15 cycles:p stats: TOTAL events: 59348 SAMPLE events: 59348 instructions:p stats: TOTAL events: 60895 SAMPLE events: 60895 $ perf report --stdio --group # To display the perf.data header info, please use --header/--header-only options. # # # Total Lost Samples: 24 # # Samples: 120K of event 'anon group { cycles:p, instructions:p }' # Event count (approx.): 24048600000 # # Overhead Command Shared Object Symbol # ................ ........... ................ .................................. # 99.74% 99.86% tchain_edit tchain_edit [.] f3 0.09% 0.02% tchain_edit tchain_edit [.] f2 0.04% 0.00% tchain_edit [kernel.vmlinux] [k] ixgbe_read_reg Eg 2, Recording the same thing multiple times can lead to high drop rate, but it is not a useful configuration. $ perf record -e '{cycles:p,cycles:p}' -c 20003 --no-time ~/tchain Warning: Processed 600592 samples and lost 99.73% samples! [perf record: Woken up 148 times to write data] [perf record: Captured and wrote 36.922 MB perf.data (1206322 samples)] [perf record: Woken up 1 times to write data] [perf record: Captured and wrote 0.121 MB perf.data (1629 samples)] Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: acme@infradead.org Cc: eranian@google.com Link: http://lkml.kernel.org/r/1431285195-14269-9-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-05-11 03:13:15 +08:00
case PERF_RECORD_LOST_SAMPLES:
ret = machine__process_lost_samples_event(machine, event, sample); break;
case PERF_RECORD_SWITCH:
case PERF_RECORD_SWITCH_CPU_WIDE:
ret = machine__process_switch_event(machine, event); break;
default:
ret = -1;
break;
}
return ret;
}
perf report/top: Add option to collapse undesired parts of call graph For example, in an application with an expensive function implemented with deeply nested recursive calls, the default call-graph presentation is dominated by the different callchains within that function. By ignoring these callees, we can collect the callchains leading into the function and compactly identify what to blame for expensive calls. For example, in this report the callers of garbage_collect() are scattered across the tree: $ perf report -d ruby 2>- | grep -m10 ^[^#]*[a-z] 22.03% ruby [.] gc_mark --- gc_mark |--59.40%-- mark_keyvalue | st_foreach | gc_mark_children | |--99.75%-- rb_gc_mark | | rb_vm_mark | | gc_mark_children | | gc_marks | | |--99.00%-- garbage_collect If we ignore the callees of garbage_collect(), its callers are coalesced: $ perf report --ignore-callees garbage_collect -d ruby 2>- | grep -m10 ^[^#]*[a-z] 72.92% ruby [.] garbage_collect --- garbage_collect vm_xmalloc |--47.08%-- ruby_xmalloc | st_insert2 | rb_hash_aset | |--98.45%-- features_index_add | | rb_provide_feature | | rb_require_safe | | vm_call_method Signed-off-by: Greg Price <price@mit.edu> Tested-by: Jiri Olsa <jolsa@redhat.com> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/20130623031720.GW22203@biohazard-cafe.mit.edu Link: http://lkml.kernel.org/r/20130708115746.GO22203@biohazard-cafe.mit.edu Cc: Fengguang Wu <fengguang.wu@intel.com> [ remove spaces at beginning of line, reported by Fengguang Wu ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2012-12-07 13:48:05 +08:00
static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
{
if (!regexec(regex, sym->name, 0, NULL, 0))
return 1;
return 0;
}
static void ip__resolve_ams(struct thread *thread,
struct addr_map_symbol *ams,
u64 ip)
{
struct addr_location al;
memset(&al, 0, sizeof(al));
/*
* We cannot use the header.misc hint to determine whether a
* branch stack address is user, kernel, guest, hypervisor.
* Branches may straddle the kernel/user/hypervisor boundaries.
* Thus, we have to try consecutively until we find a match
* or else, the symbol is unknown
*/
thread__find_cpumode_addr_location(thread, MAP__FUNCTION, ip, &al);
ams->addr = ip;
ams->al_addr = al.addr;
ams->sym = al.sym;
ams->map = al.map;
}
static void ip__resolve_data(struct thread *thread,
u8 m, struct addr_map_symbol *ams, u64 addr)
{
struct addr_location al;
memset(&al, 0, sizeof(al));
thread__find_addr_location(thread, m, MAP__VARIABLE, addr, &al);
if (al.map == NULL) {
/*
* some shared data regions have execute bit set which puts
* their mapping in the MAP__FUNCTION type array.
* Check there as a fallback option before dropping the sample.
*/
thread__find_addr_location(thread, m, MAP__FUNCTION, addr, &al);
}
ams->addr = addr;
ams->al_addr = al.addr;
ams->sym = al.sym;
ams->map = al.map;
}
struct mem_info *sample__resolve_mem(struct perf_sample *sample,
struct addr_location *al)
{
struct mem_info *mi = zalloc(sizeof(*mi));
if (!mi)
return NULL;
ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
ip__resolve_data(al->thread, al->cpumode, &mi->daddr, sample->addr);
mi->data_src.val = sample->data_src;
return mi;
}
static int add_callchain_ip(struct thread *thread,
struct callchain_cursor *cursor,
struct symbol **parent,
struct addr_location *root_al,
u8 *cpumode,
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
u64 ip,
bool branch,
struct branch_flags *flags,
int nr_loop_iter,
int samples,
u64 branch_from)
{
struct addr_location al;
al.filtered = 0;
al.sym = NULL;
if (!cpumode) {
perf callchain: Support handling complete branch stacks as histograms Currently branch stacks can be only shown as edge histograms for individual branches. I never found this display particularly useful. This implements an alternative mode that creates histograms over complete branch traces, instead of individual branches, similar to how normal callgraphs are handled. This is done by putting it in front of the normal callgraph and then using the normal callgraph histogram infrastructure to unify them. This way in complex functions we can understand the control flow that lead to a particular sample, and may even see some control flow in the caller for short functions. Example (simplified, of course for such simple code this is usually not needed), please run this after the whole patchkit is in, as at this point in the patch order there is no --branch-history, that will be added in a patch after this one: tcall.c: volatile a = 10000, b = 100000, c; __attribute__((noinline)) f2() { c = a / b; } __attribute__((noinline)) f1() { f2(); f2(); } main() { int i; for (i = 0; i < 1000000; i++) f1(); } % perf record -b -g ./tsrc/tcall [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ] % perf report --no-children --branch-history ... 54.91% tcall.c:6 [.] f2 tcall | |--65.53%-- f2 tcall.c:5 | | | |--70.83%-- f1 tcall.c:11 | | f1 tcall.c:10 | | main tcall.c:18 | | main tcall.c:18 | | main tcall.c:17 | | main tcall.c:17 | | f1 tcall.c:13 | | f1 tcall.c:13 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:12 | | f1 tcall.c:12 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:11 | | | --29.17%-- f1 tcall.c:12 | f1 tcall.c:12 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:11 | f1 tcall.c:10 | main tcall.c:18 | main tcall.c:18 | main tcall.c:17 | main tcall.c:17 | f1 tcall.c:13 | f1 tcall.c:13 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:12 The default output is unchanged. This is only implemented in perf report, no change to record or anywhere else. This adds the basic code to report: - add a new "branch" option to the -g option parser to enable this mode - when the flag is set include the LBR into the callstack in machine.c. The rest of the history code is unchanged and doesn't know the difference between LBR entry and normal call entry. - detect overlaps with the callchain - remove small loop duplicates in the LBR Current limitations: - The LBR flags (mispredict etc.) are not shown in the history and LBR entries have no special marker. - It would be nice if annotate marked the LBR entries somehow (e.g. with arrows) v2: Various fixes. v3: Merge further patches into this one. Fix white space. v4: Improve manpage. Address review feedback. v5: Rename functions. Better error message without -g. Fix crash without -b. v6: Rebase v7: Rebase. Use NO_ENTRY in memset. v8: Port to latest tip. Move add_callchain_ip to separate patch. Skip initial entries in callchain. Minor cleanups. Signed-off-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1415844328-4884-3-git-send-email-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-13 10:05:20 +08:00
thread__find_cpumode_addr_location(thread, MAP__FUNCTION,
ip, &al);
} else {
if (ip >= PERF_CONTEXT_MAX) {
switch (ip) {
case PERF_CONTEXT_HV:
*cpumode = PERF_RECORD_MISC_HYPERVISOR;
break;
case PERF_CONTEXT_KERNEL:
*cpumode = PERF_RECORD_MISC_KERNEL;
break;
case PERF_CONTEXT_USER:
*cpumode = PERF_RECORD_MISC_USER;
break;
default:
pr_debug("invalid callchain context: "
"%"PRId64"\n", (s64) ip);
/*
* It seems the callchain is corrupted.
* Discard all.
*/
callchain_cursor_reset(cursor);
return 1;
}
return 0;
}
thread__find_addr_location(thread, *cpumode, MAP__FUNCTION,
ip, &al);
}
if (al.sym != NULL) {
if (perf_hpp_list.parent && !*parent &&
symbol__match_regex(al.sym, &parent_regex))
*parent = al.sym;
else if (have_ignore_callees && root_al &&
symbol__match_regex(al.sym, &ignore_callees_regex)) {
/* Treat this symbol as the root,
forgetting its callees. */
*root_al = al;
callchain_cursor_reset(cursor);
}
}
if (symbol_conf.hide_unresolved && al.sym == NULL)
return 0;
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
return callchain_cursor_append(cursor, al.addr, al.map, al.sym,
branch, flags, nr_loop_iter, samples,
branch_from);
}
struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
struct addr_location *al)
{
unsigned int i;
const struct branch_stack *bs = sample->branch_stack;
struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
if (!bi)
return NULL;
for (i = 0; i < bs->nr; i++) {
ip__resolve_ams(al->thread, &bi[i].to, bs->entries[i].to);
ip__resolve_ams(al->thread, &bi[i].from, bs->entries[i].from);
bi[i].flags = bs->entries[i].flags;
}
return bi;
}
perf callchain: Support handling complete branch stacks as histograms Currently branch stacks can be only shown as edge histograms for individual branches. I never found this display particularly useful. This implements an alternative mode that creates histograms over complete branch traces, instead of individual branches, similar to how normal callgraphs are handled. This is done by putting it in front of the normal callgraph and then using the normal callgraph histogram infrastructure to unify them. This way in complex functions we can understand the control flow that lead to a particular sample, and may even see some control flow in the caller for short functions. Example (simplified, of course for such simple code this is usually not needed), please run this after the whole patchkit is in, as at this point in the patch order there is no --branch-history, that will be added in a patch after this one: tcall.c: volatile a = 10000, b = 100000, c; __attribute__((noinline)) f2() { c = a / b; } __attribute__((noinline)) f1() { f2(); f2(); } main() { int i; for (i = 0; i < 1000000; i++) f1(); } % perf record -b -g ./tsrc/tcall [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ] % perf report --no-children --branch-history ... 54.91% tcall.c:6 [.] f2 tcall | |--65.53%-- f2 tcall.c:5 | | | |--70.83%-- f1 tcall.c:11 | | f1 tcall.c:10 | | main tcall.c:18 | | main tcall.c:18 | | main tcall.c:17 | | main tcall.c:17 | | f1 tcall.c:13 | | f1 tcall.c:13 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:12 | | f1 tcall.c:12 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:11 | | | --29.17%-- f1 tcall.c:12 | f1 tcall.c:12 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:11 | f1 tcall.c:10 | main tcall.c:18 | main tcall.c:18 | main tcall.c:17 | main tcall.c:17 | f1 tcall.c:13 | f1 tcall.c:13 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:12 The default output is unchanged. This is only implemented in perf report, no change to record or anywhere else. This adds the basic code to report: - add a new "branch" option to the -g option parser to enable this mode - when the flag is set include the LBR into the callstack in machine.c. The rest of the history code is unchanged and doesn't know the difference between LBR entry and normal call entry. - detect overlaps with the callchain - remove small loop duplicates in the LBR Current limitations: - The LBR flags (mispredict etc.) are not shown in the history and LBR entries have no special marker. - It would be nice if annotate marked the LBR entries somehow (e.g. with arrows) v2: Various fixes. v3: Merge further patches into this one. Fix white space. v4: Improve manpage. Address review feedback. v5: Rename functions. Better error message without -g. Fix crash without -b. v6: Rebase v7: Rebase. Use NO_ENTRY in memset. v8: Port to latest tip. Move add_callchain_ip to separate patch. Skip initial entries in callchain. Minor cleanups. Signed-off-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1415844328-4884-3-git-send-email-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-13 10:05:20 +08:00
#define CHASHSZ 127
#define CHASHBITS 7
#define NO_ENTRY 0xff
#define PERF_MAX_BRANCH_DEPTH 127
/* Remove loops. */
static int remove_loops(struct branch_entry *l, int nr)
{
int i, j, off;
unsigned char chash[CHASHSZ];
memset(chash, NO_ENTRY, sizeof(chash));
BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
for (i = 0; i < nr; i++) {
int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
/* no collision handling for now */
if (chash[h] == NO_ENTRY) {
chash[h] = i;
} else if (l[chash[h]].from == l[i].from) {
bool is_loop = true;
/* check if it is a real loop */
off = 0;
for (j = chash[h]; j < i && i + off < nr; j++, off++)
if (l[j].from != l[i + off].from) {
is_loop = false;
break;
}
if (is_loop) {
memmove(l + i, l + i + off,
(nr - (i + off)) * sizeof(*l));
nr -= off;
}
}
}
return nr;
}
perf tools: Construct LBR call chain LBR call stack only has user-space callchains. It is output in the PERF_SAMPLE_BRANCH_STACK data format. For kernel callchains, it's still in the form of PERF_SAMPLE_CALLCHAIN. The perf tool has to handle both data sources to construct a complete callstack. For the "perf report -D" option, both lbr and fp information will be displayed. A new call chain recording option "lbr" is introduced into the perf tool for LBR call stack. The user can use --call-graph lbr to get the call stack information from hardware. Here are some examples. When profiling bc(1) on Fedora 19: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph lbr bc -l < cmd If enabling LBR, perf report output looks like: 50.36% bc bc [.] bc_divide | --- bc_divide execute run_code yyparse main __libc_start_main _start 33.66% bc bc [.] _one_mult | --- _one_mult bc_divide execute run_code yyparse main __libc_start_main _start 7.62% bc bc [.] _bc_do_add | --- _bc_do_add | |--99.89%-- 0x2000186a8 --0.11%-- [...] 6.83% bc bc [.] _bc_do_sub | --- _bc_do_sub | |--99.94%-- bc_add | execute | run_code | yyparse | main | __libc_start_main | _start --0.06%-- [...] 0.46% bc libc-2.17.so [.] __memset_sse2 | --- __memset_sse2 | |--54.13%-- bc_new_num | | | |--51.00%-- bc_divide | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | |--30.46%-- _bc_do_sub | | bc_add | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | --18.55%-- _bc_do_add | bc_add | execute | run_code | yyparse | main | __libc_start_main | _start | --45.87%-- bc_divide execute run_code yyparse main __libc_start_main _start If using FP, perf report output looks like: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph fp bc -l < cmd 50.49% bc bc [.] bc_divide | --- bc_divide 33.57% bc bc [.] _one_mult | --- _one_mult 7.61% bc bc [.] _bc_do_add | --- _bc_do_add 0x2000186a8 6.88% bc bc [.] _bc_do_sub | --- _bc_do_sub 0.42% bc libc-2.17.so [.] __memcpy_ssse3_back | --- __memcpy_ssse3_back If using LBR, perf report -D output looks like: 3458145275743 0x2fd750 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x2): 9748/9748: 0x408ea8 period: 609644 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 0000000000408e50 ..... 2: 000000000040a458 ..... 3: 000000000040562e ..... 4: 0000000000408590 ..... 5: 00000000004022c0 ..... 6: 00000000004015dd ..... 7: 0000003d1cc21b43 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 0000000000408ea8 ... thread: bc:9748 ...... dso: /usr/bin/bc The LBR call stack has the following known limitations: - Zero length calls are not filtered out by the hardware - Exception handing such as setjmp/longjmp will have calls/returns not match - Pushing different return address onto the stack will have calls/returns not match - If callstack is deeper than the LBR, only the last entries are captured Tested-by: Jiri Olsa <jolsa@kernel.org> Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Simon Que <sque@chromium.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1420482185-29830-3-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-01-06 02:23:05 +08:00
/*
* Recolve LBR callstack chain sample
* Return:
* 1 on success get LBR callchain information
* 0 no available LBR callchain information, should try fp
* negative error code on other errors.
*/
static int resolve_lbr_callchain_sample(struct thread *thread,
struct callchain_cursor *cursor,
perf tools: Construct LBR call chain LBR call stack only has user-space callchains. It is output in the PERF_SAMPLE_BRANCH_STACK data format. For kernel callchains, it's still in the form of PERF_SAMPLE_CALLCHAIN. The perf tool has to handle both data sources to construct a complete callstack. For the "perf report -D" option, both lbr and fp information will be displayed. A new call chain recording option "lbr" is introduced into the perf tool for LBR call stack. The user can use --call-graph lbr to get the call stack information from hardware. Here are some examples. When profiling bc(1) on Fedora 19: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph lbr bc -l < cmd If enabling LBR, perf report output looks like: 50.36% bc bc [.] bc_divide | --- bc_divide execute run_code yyparse main __libc_start_main _start 33.66% bc bc [.] _one_mult | --- _one_mult bc_divide execute run_code yyparse main __libc_start_main _start 7.62% bc bc [.] _bc_do_add | --- _bc_do_add | |--99.89%-- 0x2000186a8 --0.11%-- [...] 6.83% bc bc [.] _bc_do_sub | --- _bc_do_sub | |--99.94%-- bc_add | execute | run_code | yyparse | main | __libc_start_main | _start --0.06%-- [...] 0.46% bc libc-2.17.so [.] __memset_sse2 | --- __memset_sse2 | |--54.13%-- bc_new_num | | | |--51.00%-- bc_divide | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | |--30.46%-- _bc_do_sub | | bc_add | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | --18.55%-- _bc_do_add | bc_add | execute | run_code | yyparse | main | __libc_start_main | _start | --45.87%-- bc_divide execute run_code yyparse main __libc_start_main _start If using FP, perf report output looks like: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph fp bc -l < cmd 50.49% bc bc [.] bc_divide | --- bc_divide 33.57% bc bc [.] _one_mult | --- _one_mult 7.61% bc bc [.] _bc_do_add | --- _bc_do_add 0x2000186a8 6.88% bc bc [.] _bc_do_sub | --- _bc_do_sub 0.42% bc libc-2.17.so [.] __memcpy_ssse3_back | --- __memcpy_ssse3_back If using LBR, perf report -D output looks like: 3458145275743 0x2fd750 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x2): 9748/9748: 0x408ea8 period: 609644 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 0000000000408e50 ..... 2: 000000000040a458 ..... 3: 000000000040562e ..... 4: 0000000000408590 ..... 5: 00000000004022c0 ..... 6: 00000000004015dd ..... 7: 0000003d1cc21b43 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 0000000000408ea8 ... thread: bc:9748 ...... dso: /usr/bin/bc The LBR call stack has the following known limitations: - Zero length calls are not filtered out by the hardware - Exception handing such as setjmp/longjmp will have calls/returns not match - Pushing different return address onto the stack will have calls/returns not match - If callstack is deeper than the LBR, only the last entries are captured Tested-by: Jiri Olsa <jolsa@kernel.org> Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Simon Que <sque@chromium.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1420482185-29830-3-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-01-06 02:23:05 +08:00
struct perf_sample *sample,
struct symbol **parent,
struct addr_location *root_al,
int max_stack)
{
perf tools: Construct LBR call chain LBR call stack only has user-space callchains. It is output in the PERF_SAMPLE_BRANCH_STACK data format. For kernel callchains, it's still in the form of PERF_SAMPLE_CALLCHAIN. The perf tool has to handle both data sources to construct a complete callstack. For the "perf report -D" option, both lbr and fp information will be displayed. A new call chain recording option "lbr" is introduced into the perf tool for LBR call stack. The user can use --call-graph lbr to get the call stack information from hardware. Here are some examples. When profiling bc(1) on Fedora 19: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph lbr bc -l < cmd If enabling LBR, perf report output looks like: 50.36% bc bc [.] bc_divide | --- bc_divide execute run_code yyparse main __libc_start_main _start 33.66% bc bc [.] _one_mult | --- _one_mult bc_divide execute run_code yyparse main __libc_start_main _start 7.62% bc bc [.] _bc_do_add | --- _bc_do_add | |--99.89%-- 0x2000186a8 --0.11%-- [...] 6.83% bc bc [.] _bc_do_sub | --- _bc_do_sub | |--99.94%-- bc_add | execute | run_code | yyparse | main | __libc_start_main | _start --0.06%-- [...] 0.46% bc libc-2.17.so [.] __memset_sse2 | --- __memset_sse2 | |--54.13%-- bc_new_num | | | |--51.00%-- bc_divide | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | |--30.46%-- _bc_do_sub | | bc_add | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | --18.55%-- _bc_do_add | bc_add | execute | run_code | yyparse | main | __libc_start_main | _start | --45.87%-- bc_divide execute run_code yyparse main __libc_start_main _start If using FP, perf report output looks like: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph fp bc -l < cmd 50.49% bc bc [.] bc_divide | --- bc_divide 33.57% bc bc [.] _one_mult | --- _one_mult 7.61% bc bc [.] _bc_do_add | --- _bc_do_add 0x2000186a8 6.88% bc bc [.] _bc_do_sub | --- _bc_do_sub 0.42% bc libc-2.17.so [.] __memcpy_ssse3_back | --- __memcpy_ssse3_back If using LBR, perf report -D output looks like: 3458145275743 0x2fd750 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x2): 9748/9748: 0x408ea8 period: 609644 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 0000000000408e50 ..... 2: 000000000040a458 ..... 3: 000000000040562e ..... 4: 0000000000408590 ..... 5: 00000000004022c0 ..... 6: 00000000004015dd ..... 7: 0000003d1cc21b43 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 0000000000408ea8 ... thread: bc:9748 ...... dso: /usr/bin/bc The LBR call stack has the following known limitations: - Zero length calls are not filtered out by the hardware - Exception handing such as setjmp/longjmp will have calls/returns not match - Pushing different return address onto the stack will have calls/returns not match - If callstack is deeper than the LBR, only the last entries are captured Tested-by: Jiri Olsa <jolsa@kernel.org> Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Simon Que <sque@chromium.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1420482185-29830-3-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-01-06 02:23:05 +08:00
struct ip_callchain *chain = sample->callchain;
int chain_nr = min(max_stack, (int)chain->nr), i;
u8 cpumode = PERF_RECORD_MISC_USER;
u64 ip, branch_from = 0;
perf tools: Construct LBR call chain LBR call stack only has user-space callchains. It is output in the PERF_SAMPLE_BRANCH_STACK data format. For kernel callchains, it's still in the form of PERF_SAMPLE_CALLCHAIN. The perf tool has to handle both data sources to construct a complete callstack. For the "perf report -D" option, both lbr and fp information will be displayed. A new call chain recording option "lbr" is introduced into the perf tool for LBR call stack. The user can use --call-graph lbr to get the call stack information from hardware. Here are some examples. When profiling bc(1) on Fedora 19: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph lbr bc -l < cmd If enabling LBR, perf report output looks like: 50.36% bc bc [.] bc_divide | --- bc_divide execute run_code yyparse main __libc_start_main _start 33.66% bc bc [.] _one_mult | --- _one_mult bc_divide execute run_code yyparse main __libc_start_main _start 7.62% bc bc [.] _bc_do_add | --- _bc_do_add | |--99.89%-- 0x2000186a8 --0.11%-- [...] 6.83% bc bc [.] _bc_do_sub | --- _bc_do_sub | |--99.94%-- bc_add | execute | run_code | yyparse | main | __libc_start_main | _start --0.06%-- [...] 0.46% bc libc-2.17.so [.] __memset_sse2 | --- __memset_sse2 | |--54.13%-- bc_new_num | | | |--51.00%-- bc_divide | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | |--30.46%-- _bc_do_sub | | bc_add | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | --18.55%-- _bc_do_add | bc_add | execute | run_code | yyparse | main | __libc_start_main | _start | --45.87%-- bc_divide execute run_code yyparse main __libc_start_main _start If using FP, perf report output looks like: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph fp bc -l < cmd 50.49% bc bc [.] bc_divide | --- bc_divide 33.57% bc bc [.] _one_mult | --- _one_mult 7.61% bc bc [.] _bc_do_add | --- _bc_do_add 0x2000186a8 6.88% bc bc [.] _bc_do_sub | --- _bc_do_sub 0.42% bc libc-2.17.so [.] __memcpy_ssse3_back | --- __memcpy_ssse3_back If using LBR, perf report -D output looks like: 3458145275743 0x2fd750 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x2): 9748/9748: 0x408ea8 period: 609644 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 0000000000408e50 ..... 2: 000000000040a458 ..... 3: 000000000040562e ..... 4: 0000000000408590 ..... 5: 00000000004022c0 ..... 6: 00000000004015dd ..... 7: 0000003d1cc21b43 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 0000000000408ea8 ... thread: bc:9748 ...... dso: /usr/bin/bc The LBR call stack has the following known limitations: - Zero length calls are not filtered out by the hardware - Exception handing such as setjmp/longjmp will have calls/returns not match - Pushing different return address onto the stack will have calls/returns not match - If callstack is deeper than the LBR, only the last entries are captured Tested-by: Jiri Olsa <jolsa@kernel.org> Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Simon Que <sque@chromium.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1420482185-29830-3-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-01-06 02:23:05 +08:00
for (i = 0; i < chain_nr; i++) {
if (chain->ips[i] == PERF_CONTEXT_USER)
break;
}
/* LBR only affects the user callchain */
if (i != chain_nr) {
struct branch_stack *lbr_stack = sample->branch_stack;
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
int lbr_nr = lbr_stack->nr, j, k;
bool branch;
struct branch_flags *flags;
perf tools: Construct LBR call chain LBR call stack only has user-space callchains. It is output in the PERF_SAMPLE_BRANCH_STACK data format. For kernel callchains, it's still in the form of PERF_SAMPLE_CALLCHAIN. The perf tool has to handle both data sources to construct a complete callstack. For the "perf report -D" option, both lbr and fp information will be displayed. A new call chain recording option "lbr" is introduced into the perf tool for LBR call stack. The user can use --call-graph lbr to get the call stack information from hardware. Here are some examples. When profiling bc(1) on Fedora 19: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph lbr bc -l < cmd If enabling LBR, perf report output looks like: 50.36% bc bc [.] bc_divide | --- bc_divide execute run_code yyparse main __libc_start_main _start 33.66% bc bc [.] _one_mult | --- _one_mult bc_divide execute run_code yyparse main __libc_start_main _start 7.62% bc bc [.] _bc_do_add | --- _bc_do_add | |--99.89%-- 0x2000186a8 --0.11%-- [...] 6.83% bc bc [.] _bc_do_sub | --- _bc_do_sub | |--99.94%-- bc_add | execute | run_code | yyparse | main | __libc_start_main | _start --0.06%-- [...] 0.46% bc libc-2.17.so [.] __memset_sse2 | --- __memset_sse2 | |--54.13%-- bc_new_num | | | |--51.00%-- bc_divide | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | |--30.46%-- _bc_do_sub | | bc_add | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | --18.55%-- _bc_do_add | bc_add | execute | run_code | yyparse | main | __libc_start_main | _start | --45.87%-- bc_divide execute run_code yyparse main __libc_start_main _start If using FP, perf report output looks like: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph fp bc -l < cmd 50.49% bc bc [.] bc_divide | --- bc_divide 33.57% bc bc [.] _one_mult | --- _one_mult 7.61% bc bc [.] _bc_do_add | --- _bc_do_add 0x2000186a8 6.88% bc bc [.] _bc_do_sub | --- _bc_do_sub 0.42% bc libc-2.17.so [.] __memcpy_ssse3_back | --- __memcpy_ssse3_back If using LBR, perf report -D output looks like: 3458145275743 0x2fd750 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x2): 9748/9748: 0x408ea8 period: 609644 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 0000000000408e50 ..... 2: 000000000040a458 ..... 3: 000000000040562e ..... 4: 0000000000408590 ..... 5: 00000000004022c0 ..... 6: 00000000004015dd ..... 7: 0000003d1cc21b43 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 0000000000408ea8 ... thread: bc:9748 ...... dso: /usr/bin/bc The LBR call stack has the following known limitations: - Zero length calls are not filtered out by the hardware - Exception handing such as setjmp/longjmp will have calls/returns not match - Pushing different return address onto the stack will have calls/returns not match - If callstack is deeper than the LBR, only the last entries are captured Tested-by: Jiri Olsa <jolsa@kernel.org> Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Simon Que <sque@chromium.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1420482185-29830-3-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-01-06 02:23:05 +08:00
/*
* LBR callstack can only get user call chain.
* The mix_chain_nr is kernel call chain
* number plus LBR user call chain number.
* i is kernel call chain number,
* 1 is PERF_CONTEXT_USER,
* lbr_nr + 1 is the user call chain number.
* For details, please refer to the comments
* in callchain__printf
*/
int mix_chain_nr = i + 1 + lbr_nr + 1;
for (j = 0; j < mix_chain_nr; j++) {
int err;
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
branch = false;
flags = NULL;
perf tools: Construct LBR call chain LBR call stack only has user-space callchains. It is output in the PERF_SAMPLE_BRANCH_STACK data format. For kernel callchains, it's still in the form of PERF_SAMPLE_CALLCHAIN. The perf tool has to handle both data sources to construct a complete callstack. For the "perf report -D" option, both lbr and fp information will be displayed. A new call chain recording option "lbr" is introduced into the perf tool for LBR call stack. The user can use --call-graph lbr to get the call stack information from hardware. Here are some examples. When profiling bc(1) on Fedora 19: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph lbr bc -l < cmd If enabling LBR, perf report output looks like: 50.36% bc bc [.] bc_divide | --- bc_divide execute run_code yyparse main __libc_start_main _start 33.66% bc bc [.] _one_mult | --- _one_mult bc_divide execute run_code yyparse main __libc_start_main _start 7.62% bc bc [.] _bc_do_add | --- _bc_do_add | |--99.89%-- 0x2000186a8 --0.11%-- [...] 6.83% bc bc [.] _bc_do_sub | --- _bc_do_sub | |--99.94%-- bc_add | execute | run_code | yyparse | main | __libc_start_main | _start --0.06%-- [...] 0.46% bc libc-2.17.so [.] __memset_sse2 | --- __memset_sse2 | |--54.13%-- bc_new_num | | | |--51.00%-- bc_divide | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | |--30.46%-- _bc_do_sub | | bc_add | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | --18.55%-- _bc_do_add | bc_add | execute | run_code | yyparse | main | __libc_start_main | _start | --45.87%-- bc_divide execute run_code yyparse main __libc_start_main _start If using FP, perf report output looks like: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph fp bc -l < cmd 50.49% bc bc [.] bc_divide | --- bc_divide 33.57% bc bc [.] _one_mult | --- _one_mult 7.61% bc bc [.] _bc_do_add | --- _bc_do_add 0x2000186a8 6.88% bc bc [.] _bc_do_sub | --- _bc_do_sub 0.42% bc libc-2.17.so [.] __memcpy_ssse3_back | --- __memcpy_ssse3_back If using LBR, perf report -D output looks like: 3458145275743 0x2fd750 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x2): 9748/9748: 0x408ea8 period: 609644 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 0000000000408e50 ..... 2: 000000000040a458 ..... 3: 000000000040562e ..... 4: 0000000000408590 ..... 5: 00000000004022c0 ..... 6: 00000000004015dd ..... 7: 0000003d1cc21b43 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 0000000000408ea8 ... thread: bc:9748 ...... dso: /usr/bin/bc The LBR call stack has the following known limitations: - Zero length calls are not filtered out by the hardware - Exception handing such as setjmp/longjmp will have calls/returns not match - Pushing different return address onto the stack will have calls/returns not match - If callstack is deeper than the LBR, only the last entries are captured Tested-by: Jiri Olsa <jolsa@kernel.org> Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Simon Que <sque@chromium.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1420482185-29830-3-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-01-06 02:23:05 +08:00
if (callchain_param.order == ORDER_CALLEE) {
if (j < i + 1)
ip = chain->ips[j];
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
else if (j > i + 1) {
k = j - i - 2;
ip = lbr_stack->entries[k].from;
branch = true;
flags = &lbr_stack->entries[k].flags;
} else {
perf tools: Construct LBR call chain LBR call stack only has user-space callchains. It is output in the PERF_SAMPLE_BRANCH_STACK data format. For kernel callchains, it's still in the form of PERF_SAMPLE_CALLCHAIN. The perf tool has to handle both data sources to construct a complete callstack. For the "perf report -D" option, both lbr and fp information will be displayed. A new call chain recording option "lbr" is introduced into the perf tool for LBR call stack. The user can use --call-graph lbr to get the call stack information from hardware. Here are some examples. When profiling bc(1) on Fedora 19: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph lbr bc -l < cmd If enabling LBR, perf report output looks like: 50.36% bc bc [.] bc_divide | --- bc_divide execute run_code yyparse main __libc_start_main _start 33.66% bc bc [.] _one_mult | --- _one_mult bc_divide execute run_code yyparse main __libc_start_main _start 7.62% bc bc [.] _bc_do_add | --- _bc_do_add | |--99.89%-- 0x2000186a8 --0.11%-- [...] 6.83% bc bc [.] _bc_do_sub | --- _bc_do_sub | |--99.94%-- bc_add | execute | run_code | yyparse | main | __libc_start_main | _start --0.06%-- [...] 0.46% bc libc-2.17.so [.] __memset_sse2 | --- __memset_sse2 | |--54.13%-- bc_new_num | | | |--51.00%-- bc_divide | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | |--30.46%-- _bc_do_sub | | bc_add | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | --18.55%-- _bc_do_add | bc_add | execute | run_code | yyparse | main | __libc_start_main | _start | --45.87%-- bc_divide execute run_code yyparse main __libc_start_main _start If using FP, perf report output looks like: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph fp bc -l < cmd 50.49% bc bc [.] bc_divide | --- bc_divide 33.57% bc bc [.] _one_mult | --- _one_mult 7.61% bc bc [.] _bc_do_add | --- _bc_do_add 0x2000186a8 6.88% bc bc [.] _bc_do_sub | --- _bc_do_sub 0.42% bc libc-2.17.so [.] __memcpy_ssse3_back | --- __memcpy_ssse3_back If using LBR, perf report -D output looks like: 3458145275743 0x2fd750 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x2): 9748/9748: 0x408ea8 period: 609644 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 0000000000408e50 ..... 2: 000000000040a458 ..... 3: 000000000040562e ..... 4: 0000000000408590 ..... 5: 00000000004022c0 ..... 6: 00000000004015dd ..... 7: 0000003d1cc21b43 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 0000000000408ea8 ... thread: bc:9748 ...... dso: /usr/bin/bc The LBR call stack has the following known limitations: - Zero length calls are not filtered out by the hardware - Exception handing such as setjmp/longjmp will have calls/returns not match - Pushing different return address onto the stack will have calls/returns not match - If callstack is deeper than the LBR, only the last entries are captured Tested-by: Jiri Olsa <jolsa@kernel.org> Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Simon Que <sque@chromium.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1420482185-29830-3-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-01-06 02:23:05 +08:00
ip = lbr_stack->entries[0].to;
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
branch = true;
flags = &lbr_stack->entries[0].flags;
branch_from =
lbr_stack->entries[0].from;
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
}
perf tools: Construct LBR call chain LBR call stack only has user-space callchains. It is output in the PERF_SAMPLE_BRANCH_STACK data format. For kernel callchains, it's still in the form of PERF_SAMPLE_CALLCHAIN. The perf tool has to handle both data sources to construct a complete callstack. For the "perf report -D" option, both lbr and fp information will be displayed. A new call chain recording option "lbr" is introduced into the perf tool for LBR call stack. The user can use --call-graph lbr to get the call stack information from hardware. Here are some examples. When profiling bc(1) on Fedora 19: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph lbr bc -l < cmd If enabling LBR, perf report output looks like: 50.36% bc bc [.] bc_divide | --- bc_divide execute run_code yyparse main __libc_start_main _start 33.66% bc bc [.] _one_mult | --- _one_mult bc_divide execute run_code yyparse main __libc_start_main _start 7.62% bc bc [.] _bc_do_add | --- _bc_do_add | |--99.89%-- 0x2000186a8 --0.11%-- [...] 6.83% bc bc [.] _bc_do_sub | --- _bc_do_sub | |--99.94%-- bc_add | execute | run_code | yyparse | main | __libc_start_main | _start --0.06%-- [...] 0.46% bc libc-2.17.so [.] __memset_sse2 | --- __memset_sse2 | |--54.13%-- bc_new_num | | | |--51.00%-- bc_divide | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | |--30.46%-- _bc_do_sub | | bc_add | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | --18.55%-- _bc_do_add | bc_add | execute | run_code | yyparse | main | __libc_start_main | _start | --45.87%-- bc_divide execute run_code yyparse main __libc_start_main _start If using FP, perf report output looks like: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph fp bc -l < cmd 50.49% bc bc [.] bc_divide | --- bc_divide 33.57% bc bc [.] _one_mult | --- _one_mult 7.61% bc bc [.] _bc_do_add | --- _bc_do_add 0x2000186a8 6.88% bc bc [.] _bc_do_sub | --- _bc_do_sub 0.42% bc libc-2.17.so [.] __memcpy_ssse3_back | --- __memcpy_ssse3_back If using LBR, perf report -D output looks like: 3458145275743 0x2fd750 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x2): 9748/9748: 0x408ea8 period: 609644 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 0000000000408e50 ..... 2: 000000000040a458 ..... 3: 000000000040562e ..... 4: 0000000000408590 ..... 5: 00000000004022c0 ..... 6: 00000000004015dd ..... 7: 0000003d1cc21b43 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 0000000000408ea8 ... thread: bc:9748 ...... dso: /usr/bin/bc The LBR call stack has the following known limitations: - Zero length calls are not filtered out by the hardware - Exception handing such as setjmp/longjmp will have calls/returns not match - Pushing different return address onto the stack will have calls/returns not match - If callstack is deeper than the LBR, only the last entries are captured Tested-by: Jiri Olsa <jolsa@kernel.org> Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Simon Que <sque@chromium.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1420482185-29830-3-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-01-06 02:23:05 +08:00
} else {
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
if (j < lbr_nr) {
k = lbr_nr - j - 1;
ip = lbr_stack->entries[k].from;
branch = true;
flags = &lbr_stack->entries[k].flags;
}
perf tools: Construct LBR call chain LBR call stack only has user-space callchains. It is output in the PERF_SAMPLE_BRANCH_STACK data format. For kernel callchains, it's still in the form of PERF_SAMPLE_CALLCHAIN. The perf tool has to handle both data sources to construct a complete callstack. For the "perf report -D" option, both lbr and fp information will be displayed. A new call chain recording option "lbr" is introduced into the perf tool for LBR call stack. The user can use --call-graph lbr to get the call stack information from hardware. Here are some examples. When profiling bc(1) on Fedora 19: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph lbr bc -l < cmd If enabling LBR, perf report output looks like: 50.36% bc bc [.] bc_divide | --- bc_divide execute run_code yyparse main __libc_start_main _start 33.66% bc bc [.] _one_mult | --- _one_mult bc_divide execute run_code yyparse main __libc_start_main _start 7.62% bc bc [.] _bc_do_add | --- _bc_do_add | |--99.89%-- 0x2000186a8 --0.11%-- [...] 6.83% bc bc [.] _bc_do_sub | --- _bc_do_sub | |--99.94%-- bc_add | execute | run_code | yyparse | main | __libc_start_main | _start --0.06%-- [...] 0.46% bc libc-2.17.so [.] __memset_sse2 | --- __memset_sse2 | |--54.13%-- bc_new_num | | | |--51.00%-- bc_divide | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | |--30.46%-- _bc_do_sub | | bc_add | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | --18.55%-- _bc_do_add | bc_add | execute | run_code | yyparse | main | __libc_start_main | _start | --45.87%-- bc_divide execute run_code yyparse main __libc_start_main _start If using FP, perf report output looks like: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph fp bc -l < cmd 50.49% bc bc [.] bc_divide | --- bc_divide 33.57% bc bc [.] _one_mult | --- _one_mult 7.61% bc bc [.] _bc_do_add | --- _bc_do_add 0x2000186a8 6.88% bc bc [.] _bc_do_sub | --- _bc_do_sub 0.42% bc libc-2.17.so [.] __memcpy_ssse3_back | --- __memcpy_ssse3_back If using LBR, perf report -D output looks like: 3458145275743 0x2fd750 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x2): 9748/9748: 0x408ea8 period: 609644 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 0000000000408e50 ..... 2: 000000000040a458 ..... 3: 000000000040562e ..... 4: 0000000000408590 ..... 5: 00000000004022c0 ..... 6: 00000000004015dd ..... 7: 0000003d1cc21b43 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 0000000000408ea8 ... thread: bc:9748 ...... dso: /usr/bin/bc The LBR call stack has the following known limitations: - Zero length calls are not filtered out by the hardware - Exception handing such as setjmp/longjmp will have calls/returns not match - Pushing different return address onto the stack will have calls/returns not match - If callstack is deeper than the LBR, only the last entries are captured Tested-by: Jiri Olsa <jolsa@kernel.org> Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Simon Que <sque@chromium.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1420482185-29830-3-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-01-06 02:23:05 +08:00
else if (j > lbr_nr)
ip = chain->ips[i + 1 - (j - lbr_nr)];
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
else {
perf tools: Construct LBR call chain LBR call stack only has user-space callchains. It is output in the PERF_SAMPLE_BRANCH_STACK data format. For kernel callchains, it's still in the form of PERF_SAMPLE_CALLCHAIN. The perf tool has to handle both data sources to construct a complete callstack. For the "perf report -D" option, both lbr and fp information will be displayed. A new call chain recording option "lbr" is introduced into the perf tool for LBR call stack. The user can use --call-graph lbr to get the call stack information from hardware. Here are some examples. When profiling bc(1) on Fedora 19: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph lbr bc -l < cmd If enabling LBR, perf report output looks like: 50.36% bc bc [.] bc_divide | --- bc_divide execute run_code yyparse main __libc_start_main _start 33.66% bc bc [.] _one_mult | --- _one_mult bc_divide execute run_code yyparse main __libc_start_main _start 7.62% bc bc [.] _bc_do_add | --- _bc_do_add | |--99.89%-- 0x2000186a8 --0.11%-- [...] 6.83% bc bc [.] _bc_do_sub | --- _bc_do_sub | |--99.94%-- bc_add | execute | run_code | yyparse | main | __libc_start_main | _start --0.06%-- [...] 0.46% bc libc-2.17.so [.] __memset_sse2 | --- __memset_sse2 | |--54.13%-- bc_new_num | | | |--51.00%-- bc_divide | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | |--30.46%-- _bc_do_sub | | bc_add | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | --18.55%-- _bc_do_add | bc_add | execute | run_code | yyparse | main | __libc_start_main | _start | --45.87%-- bc_divide execute run_code yyparse main __libc_start_main _start If using FP, perf report output looks like: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph fp bc -l < cmd 50.49% bc bc [.] bc_divide | --- bc_divide 33.57% bc bc [.] _one_mult | --- _one_mult 7.61% bc bc [.] _bc_do_add | --- _bc_do_add 0x2000186a8 6.88% bc bc [.] _bc_do_sub | --- _bc_do_sub 0.42% bc libc-2.17.so [.] __memcpy_ssse3_back | --- __memcpy_ssse3_back If using LBR, perf report -D output looks like: 3458145275743 0x2fd750 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x2): 9748/9748: 0x408ea8 period: 609644 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 0000000000408e50 ..... 2: 000000000040a458 ..... 3: 000000000040562e ..... 4: 0000000000408590 ..... 5: 00000000004022c0 ..... 6: 00000000004015dd ..... 7: 0000003d1cc21b43 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 0000000000408ea8 ... thread: bc:9748 ...... dso: /usr/bin/bc The LBR call stack has the following known limitations: - Zero length calls are not filtered out by the hardware - Exception handing such as setjmp/longjmp will have calls/returns not match - Pushing different return address onto the stack will have calls/returns not match - If callstack is deeper than the LBR, only the last entries are captured Tested-by: Jiri Olsa <jolsa@kernel.org> Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Simon Que <sque@chromium.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1420482185-29830-3-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-01-06 02:23:05 +08:00
ip = lbr_stack->entries[0].to;
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
branch = true;
flags = &lbr_stack->entries[0].flags;
branch_from =
lbr_stack->entries[0].from;
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
}
perf tools: Construct LBR call chain LBR call stack only has user-space callchains. It is output in the PERF_SAMPLE_BRANCH_STACK data format. For kernel callchains, it's still in the form of PERF_SAMPLE_CALLCHAIN. The perf tool has to handle both data sources to construct a complete callstack. For the "perf report -D" option, both lbr and fp information will be displayed. A new call chain recording option "lbr" is introduced into the perf tool for LBR call stack. The user can use --call-graph lbr to get the call stack information from hardware. Here are some examples. When profiling bc(1) on Fedora 19: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph lbr bc -l < cmd If enabling LBR, perf report output looks like: 50.36% bc bc [.] bc_divide | --- bc_divide execute run_code yyparse main __libc_start_main _start 33.66% bc bc [.] _one_mult | --- _one_mult bc_divide execute run_code yyparse main __libc_start_main _start 7.62% bc bc [.] _bc_do_add | --- _bc_do_add | |--99.89%-- 0x2000186a8 --0.11%-- [...] 6.83% bc bc [.] _bc_do_sub | --- _bc_do_sub | |--99.94%-- bc_add | execute | run_code | yyparse | main | __libc_start_main | _start --0.06%-- [...] 0.46% bc libc-2.17.so [.] __memset_sse2 | --- __memset_sse2 | |--54.13%-- bc_new_num | | | |--51.00%-- bc_divide | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | |--30.46%-- _bc_do_sub | | bc_add | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | --18.55%-- _bc_do_add | bc_add | execute | run_code | yyparse | main | __libc_start_main | _start | --45.87%-- bc_divide execute run_code yyparse main __libc_start_main _start If using FP, perf report output looks like: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph fp bc -l < cmd 50.49% bc bc [.] bc_divide | --- bc_divide 33.57% bc bc [.] _one_mult | --- _one_mult 7.61% bc bc [.] _bc_do_add | --- _bc_do_add 0x2000186a8 6.88% bc bc [.] _bc_do_sub | --- _bc_do_sub 0.42% bc libc-2.17.so [.] __memcpy_ssse3_back | --- __memcpy_ssse3_back If using LBR, perf report -D output looks like: 3458145275743 0x2fd750 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x2): 9748/9748: 0x408ea8 period: 609644 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 0000000000408e50 ..... 2: 000000000040a458 ..... 3: 000000000040562e ..... 4: 0000000000408590 ..... 5: 00000000004022c0 ..... 6: 00000000004015dd ..... 7: 0000003d1cc21b43 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 0000000000408ea8 ... thread: bc:9748 ...... dso: /usr/bin/bc The LBR call stack has the following known limitations: - Zero length calls are not filtered out by the hardware - Exception handing such as setjmp/longjmp will have calls/returns not match - Pushing different return address onto the stack will have calls/returns not match - If callstack is deeper than the LBR, only the last entries are captured Tested-by: Jiri Olsa <jolsa@kernel.org> Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Simon Que <sque@chromium.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1420482185-29830-3-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-01-06 02:23:05 +08:00
}
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
err = add_callchain_ip(thread, cursor, parent,
root_al, &cpumode, ip,
branch, flags, 0, 0,
branch_from);
perf tools: Construct LBR call chain LBR call stack only has user-space callchains. It is output in the PERF_SAMPLE_BRANCH_STACK data format. For kernel callchains, it's still in the form of PERF_SAMPLE_CALLCHAIN. The perf tool has to handle both data sources to construct a complete callstack. For the "perf report -D" option, both lbr and fp information will be displayed. A new call chain recording option "lbr" is introduced into the perf tool for LBR call stack. The user can use --call-graph lbr to get the call stack information from hardware. Here are some examples. When profiling bc(1) on Fedora 19: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph lbr bc -l < cmd If enabling LBR, perf report output looks like: 50.36% bc bc [.] bc_divide | --- bc_divide execute run_code yyparse main __libc_start_main _start 33.66% bc bc [.] _one_mult | --- _one_mult bc_divide execute run_code yyparse main __libc_start_main _start 7.62% bc bc [.] _bc_do_add | --- _bc_do_add | |--99.89%-- 0x2000186a8 --0.11%-- [...] 6.83% bc bc [.] _bc_do_sub | --- _bc_do_sub | |--99.94%-- bc_add | execute | run_code | yyparse | main | __libc_start_main | _start --0.06%-- [...] 0.46% bc libc-2.17.so [.] __memset_sse2 | --- __memset_sse2 | |--54.13%-- bc_new_num | | | |--51.00%-- bc_divide | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | |--30.46%-- _bc_do_sub | | bc_add | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | --18.55%-- _bc_do_add | bc_add | execute | run_code | yyparse | main | __libc_start_main | _start | --45.87%-- bc_divide execute run_code yyparse main __libc_start_main _start If using FP, perf report output looks like: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph fp bc -l < cmd 50.49% bc bc [.] bc_divide | --- bc_divide 33.57% bc bc [.] _one_mult | --- _one_mult 7.61% bc bc [.] _bc_do_add | --- _bc_do_add 0x2000186a8 6.88% bc bc [.] _bc_do_sub | --- _bc_do_sub 0.42% bc libc-2.17.so [.] __memcpy_ssse3_back | --- __memcpy_ssse3_back If using LBR, perf report -D output looks like: 3458145275743 0x2fd750 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x2): 9748/9748: 0x408ea8 period: 609644 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 0000000000408e50 ..... 2: 000000000040a458 ..... 3: 000000000040562e ..... 4: 0000000000408590 ..... 5: 00000000004022c0 ..... 6: 00000000004015dd ..... 7: 0000003d1cc21b43 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 0000000000408ea8 ... thread: bc:9748 ...... dso: /usr/bin/bc The LBR call stack has the following known limitations: - Zero length calls are not filtered out by the hardware - Exception handing such as setjmp/longjmp will have calls/returns not match - Pushing different return address onto the stack will have calls/returns not match - If callstack is deeper than the LBR, only the last entries are captured Tested-by: Jiri Olsa <jolsa@kernel.org> Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Simon Que <sque@chromium.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1420482185-29830-3-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-01-06 02:23:05 +08:00
if (err)
return (err < 0) ? err : 0;
}
return 1;
}
return 0;
}
static int thread__resolve_callchain_sample(struct thread *thread,
struct callchain_cursor *cursor,
perf tools: Construct LBR call chain LBR call stack only has user-space callchains. It is output in the PERF_SAMPLE_BRANCH_STACK data format. For kernel callchains, it's still in the form of PERF_SAMPLE_CALLCHAIN. The perf tool has to handle both data sources to construct a complete callstack. For the "perf report -D" option, both lbr and fp information will be displayed. A new call chain recording option "lbr" is introduced into the perf tool for LBR call stack. The user can use --call-graph lbr to get the call stack information from hardware. Here are some examples. When profiling bc(1) on Fedora 19: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph lbr bc -l < cmd If enabling LBR, perf report output looks like: 50.36% bc bc [.] bc_divide | --- bc_divide execute run_code yyparse main __libc_start_main _start 33.66% bc bc [.] _one_mult | --- _one_mult bc_divide execute run_code yyparse main __libc_start_main _start 7.62% bc bc [.] _bc_do_add | --- _bc_do_add | |--99.89%-- 0x2000186a8 --0.11%-- [...] 6.83% bc bc [.] _bc_do_sub | --- _bc_do_sub | |--99.94%-- bc_add | execute | run_code | yyparse | main | __libc_start_main | _start --0.06%-- [...] 0.46% bc libc-2.17.so [.] __memset_sse2 | --- __memset_sse2 | |--54.13%-- bc_new_num | | | |--51.00%-- bc_divide | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | |--30.46%-- _bc_do_sub | | bc_add | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | --18.55%-- _bc_do_add | bc_add | execute | run_code | yyparse | main | __libc_start_main | _start | --45.87%-- bc_divide execute run_code yyparse main __libc_start_main _start If using FP, perf report output looks like: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph fp bc -l < cmd 50.49% bc bc [.] bc_divide | --- bc_divide 33.57% bc bc [.] _one_mult | --- _one_mult 7.61% bc bc [.] _bc_do_add | --- _bc_do_add 0x2000186a8 6.88% bc bc [.] _bc_do_sub | --- _bc_do_sub 0.42% bc libc-2.17.so [.] __memcpy_ssse3_back | --- __memcpy_ssse3_back If using LBR, perf report -D output looks like: 3458145275743 0x2fd750 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x2): 9748/9748: 0x408ea8 period: 609644 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 0000000000408e50 ..... 2: 000000000040a458 ..... 3: 000000000040562e ..... 4: 0000000000408590 ..... 5: 00000000004022c0 ..... 6: 00000000004015dd ..... 7: 0000003d1cc21b43 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 0000000000408ea8 ... thread: bc:9748 ...... dso: /usr/bin/bc The LBR call stack has the following known limitations: - Zero length calls are not filtered out by the hardware - Exception handing such as setjmp/longjmp will have calls/returns not match - Pushing different return address onto the stack will have calls/returns not match - If callstack is deeper than the LBR, only the last entries are captured Tested-by: Jiri Olsa <jolsa@kernel.org> Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Simon Que <sque@chromium.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1420482185-29830-3-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-01-06 02:23:05 +08:00
struct perf_evsel *evsel,
struct perf_sample *sample,
struct symbol **parent,
struct addr_location *root_al,
int max_stack)
{
struct branch_stack *branch = sample->branch_stack;
struct ip_callchain *chain = sample->callchain;
perf report: Make --branch-history work without callgraphs(-g) option in perf record perf record -b -g <command> perf report --branch-history This merges the LBRs with the callgraphs. However it would be nice if it also works without callgraphs (-g) set in perf record, so that only the LBRs are displayed. But currently perf report errors in this case. For example, perf record -b <command> perf report --branch-history Error: Selected -g or --branch-history but no callchain data. Did you call 'perf record' without -g? This patch displays the LBRs only even if callgraphs(-g) is not enabled in perf record. Change log: v2: According to Milian Wolff's comment, change the obsolete error message. Now the error message is: ┌─Error:─────────────────────────────────────┐ │Selected -g or --branch-history. │ │But no callchain or branch data. │ │Did you call 'perf record' without -g or -b?│ │ │ │ │ │Press any key... │ └────────────────────────────────────────────┘ When passing the last parameter to hists__fprintf, changes "|" to "||". hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout, symbol_conf.use_callchain || symbol_conf.show_branchflag_count); Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Reviewed-by: Andi Kleen <ak@linux.intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/1494240182-28899-1-git-send-email-yao.jin@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-05-08 18:43:02 +08:00
int chain_nr = 0;
u8 cpumode = PERF_RECORD_MISC_USER;
int i, j, err, nr_entries;
perf callchain: Support handling complete branch stacks as histograms Currently branch stacks can be only shown as edge histograms for individual branches. I never found this display particularly useful. This implements an alternative mode that creates histograms over complete branch traces, instead of individual branches, similar to how normal callgraphs are handled. This is done by putting it in front of the normal callgraph and then using the normal callgraph histogram infrastructure to unify them. This way in complex functions we can understand the control flow that lead to a particular sample, and may even see some control flow in the caller for short functions. Example (simplified, of course for such simple code this is usually not needed), please run this after the whole patchkit is in, as at this point in the patch order there is no --branch-history, that will be added in a patch after this one: tcall.c: volatile a = 10000, b = 100000, c; __attribute__((noinline)) f2() { c = a / b; } __attribute__((noinline)) f1() { f2(); f2(); } main() { int i; for (i = 0; i < 1000000; i++) f1(); } % perf record -b -g ./tsrc/tcall [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ] % perf report --no-children --branch-history ... 54.91% tcall.c:6 [.] f2 tcall | |--65.53%-- f2 tcall.c:5 | | | |--70.83%-- f1 tcall.c:11 | | f1 tcall.c:10 | | main tcall.c:18 | | main tcall.c:18 | | main tcall.c:17 | | main tcall.c:17 | | f1 tcall.c:13 | | f1 tcall.c:13 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:12 | | f1 tcall.c:12 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:11 | | | --29.17%-- f1 tcall.c:12 | f1 tcall.c:12 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:11 | f1 tcall.c:10 | main tcall.c:18 | main tcall.c:18 | main tcall.c:17 | main tcall.c:17 | f1 tcall.c:13 | f1 tcall.c:13 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:12 The default output is unchanged. This is only implemented in perf report, no change to record or anywhere else. This adds the basic code to report: - add a new "branch" option to the -g option parser to enable this mode - when the flag is set include the LBR into the callstack in machine.c. The rest of the history code is unchanged and doesn't know the difference between LBR entry and normal call entry. - detect overlaps with the callchain - remove small loop duplicates in the LBR Current limitations: - The LBR flags (mispredict etc.) are not shown in the history and LBR entries have no special marker. - It would be nice if annotate marked the LBR entries somehow (e.g. with arrows) v2: Various fixes. v3: Merge further patches into this one. Fix white space. v4: Improve manpage. Address review feedback. v5: Rename functions. Better error message without -g. Fix crash without -b. v6: Rebase v7: Rebase. Use NO_ENTRY in memset. v8: Port to latest tip. Move add_callchain_ip to separate patch. Skip initial entries in callchain. Minor cleanups. Signed-off-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1415844328-4884-3-git-send-email-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-13 10:05:20 +08:00
int skip_idx = -1;
int first_call = 0;
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
int nr_loop_iter;
perf callchain: Support handling complete branch stacks as histograms Currently branch stacks can be only shown as edge histograms for individual branches. I never found this display particularly useful. This implements an alternative mode that creates histograms over complete branch traces, instead of individual branches, similar to how normal callgraphs are handled. This is done by putting it in front of the normal callgraph and then using the normal callgraph histogram infrastructure to unify them. This way in complex functions we can understand the control flow that lead to a particular sample, and may even see some control flow in the caller for short functions. Example (simplified, of course for such simple code this is usually not needed), please run this after the whole patchkit is in, as at this point in the patch order there is no --branch-history, that will be added in a patch after this one: tcall.c: volatile a = 10000, b = 100000, c; __attribute__((noinline)) f2() { c = a / b; } __attribute__((noinline)) f1() { f2(); f2(); } main() { int i; for (i = 0; i < 1000000; i++) f1(); } % perf record -b -g ./tsrc/tcall [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ] % perf report --no-children --branch-history ... 54.91% tcall.c:6 [.] f2 tcall | |--65.53%-- f2 tcall.c:5 | | | |--70.83%-- f1 tcall.c:11 | | f1 tcall.c:10 | | main tcall.c:18 | | main tcall.c:18 | | main tcall.c:17 | | main tcall.c:17 | | f1 tcall.c:13 | | f1 tcall.c:13 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:12 | | f1 tcall.c:12 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:11 | | | --29.17%-- f1 tcall.c:12 | f1 tcall.c:12 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:11 | f1 tcall.c:10 | main tcall.c:18 | main tcall.c:18 | main tcall.c:17 | main tcall.c:17 | f1 tcall.c:13 | f1 tcall.c:13 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:12 The default output is unchanged. This is only implemented in perf report, no change to record or anywhere else. This adds the basic code to report: - add a new "branch" option to the -g option parser to enable this mode - when the flag is set include the LBR into the callstack in machine.c. The rest of the history code is unchanged and doesn't know the difference between LBR entry and normal call entry. - detect overlaps with the callchain - remove small loop duplicates in the LBR Current limitations: - The LBR flags (mispredict etc.) are not shown in the history and LBR entries have no special marker. - It would be nice if annotate marked the LBR entries somehow (e.g. with arrows) v2: Various fixes. v3: Merge further patches into this one. Fix white space. v4: Improve manpage. Address review feedback. v5: Rename functions. Better error message without -g. Fix crash without -b. v6: Rebase v7: Rebase. Use NO_ENTRY in memset. v8: Port to latest tip. Move add_callchain_ip to separate patch. Skip initial entries in callchain. Minor cleanups. Signed-off-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1415844328-4884-3-git-send-email-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-13 10:05:20 +08:00
perf report: Make --branch-history work without callgraphs(-g) option in perf record perf record -b -g <command> perf report --branch-history This merges the LBRs with the callgraphs. However it would be nice if it also works without callgraphs (-g) set in perf record, so that only the LBRs are displayed. But currently perf report errors in this case. For example, perf record -b <command> perf report --branch-history Error: Selected -g or --branch-history but no callchain data. Did you call 'perf record' without -g? This patch displays the LBRs only even if callgraphs(-g) is not enabled in perf record. Change log: v2: According to Milian Wolff's comment, change the obsolete error message. Now the error message is: ┌─Error:─────────────────────────────────────┐ │Selected -g or --branch-history. │ │But no callchain or branch data. │ │Did you call 'perf record' without -g or -b?│ │ │ │ │ │Press any key... │ └────────────────────────────────────────────┘ When passing the last parameter to hists__fprintf, changes "|" to "||". hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout, symbol_conf.use_callchain || symbol_conf.show_branchflag_count); Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Reviewed-by: Andi Kleen <ak@linux.intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/1494240182-28899-1-git-send-email-yao.jin@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-05-08 18:43:02 +08:00
if (chain)
chain_nr = chain->nr;
if (perf_evsel__has_branch_callstack(evsel)) {
err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
perf tools: Construct LBR call chain LBR call stack only has user-space callchains. It is output in the PERF_SAMPLE_BRANCH_STACK data format. For kernel callchains, it's still in the form of PERF_SAMPLE_CALLCHAIN. The perf tool has to handle both data sources to construct a complete callstack. For the "perf report -D" option, both lbr and fp information will be displayed. A new call chain recording option "lbr" is introduced into the perf tool for LBR call stack. The user can use --call-graph lbr to get the call stack information from hardware. Here are some examples. When profiling bc(1) on Fedora 19: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph lbr bc -l < cmd If enabling LBR, perf report output looks like: 50.36% bc bc [.] bc_divide | --- bc_divide execute run_code yyparse main __libc_start_main _start 33.66% bc bc [.] _one_mult | --- _one_mult bc_divide execute run_code yyparse main __libc_start_main _start 7.62% bc bc [.] _bc_do_add | --- _bc_do_add | |--99.89%-- 0x2000186a8 --0.11%-- [...] 6.83% bc bc [.] _bc_do_sub | --- _bc_do_sub | |--99.94%-- bc_add | execute | run_code | yyparse | main | __libc_start_main | _start --0.06%-- [...] 0.46% bc libc-2.17.so [.] __memset_sse2 | --- __memset_sse2 | |--54.13%-- bc_new_num | | | |--51.00%-- bc_divide | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | |--30.46%-- _bc_do_sub | | bc_add | | execute | | run_code | | yyparse | | main | | __libc_start_main | | _start | | | --18.55%-- _bc_do_add | bc_add | execute | run_code | yyparse | main | __libc_start_main | _start | --45.87%-- bc_divide execute run_code yyparse main __libc_start_main _start If using FP, perf report output looks like: echo 'scale=2000; 4*a(1)' > cmd; perf record --call-graph fp bc -l < cmd 50.49% bc bc [.] bc_divide | --- bc_divide 33.57% bc bc [.] _one_mult | --- _one_mult 7.61% bc bc [.] _bc_do_add | --- _bc_do_add 0x2000186a8 6.88% bc bc [.] _bc_do_sub | --- _bc_do_sub 0.42% bc libc-2.17.so [.] __memcpy_ssse3_back | --- __memcpy_ssse3_back If using LBR, perf report -D output looks like: 3458145275743 0x2fd750 [0xd8]: PERF_RECORD_SAMPLE(IP, 0x2): 9748/9748: 0x408ea8 period: 609644 addr: 0 ... LBR call chain: nr:8 ..... 0: fffffffffffffe00 ..... 1: 0000000000408e50 ..... 2: 000000000040a458 ..... 3: 000000000040562e ..... 4: 0000000000408590 ..... 5: 00000000004022c0 ..... 6: 00000000004015dd ..... 7: 0000003d1cc21b43 ... FP chain: nr:2 ..... 0: fffffffffffffe00 ..... 1: 0000000000408ea8 ... thread: bc:9748 ...... dso: /usr/bin/bc The LBR call stack has the following known limitations: - Zero length calls are not filtered out by the hardware - Exception handing such as setjmp/longjmp will have calls/returns not match - Pushing different return address onto the stack will have calls/returns not match - If callstack is deeper than the LBR, only the last entries are captured Tested-by: Jiri Olsa <jolsa@kernel.org> Signed-off-by: Kan Liang <kan.liang@intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Borislav Petkov <bp@suse.de> Cc: David Ahern <dsahern@gmail.com> Cc: Don Zickus <dzickus@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Simon Que <sque@chromium.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1420482185-29830-3-git-send-email-kan.liang@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-01-06 02:23:05 +08:00
root_al, max_stack);
if (err)
return (err < 0) ? err : 0;
}
perf callchain: Support handling complete branch stacks as histograms Currently branch stacks can be only shown as edge histograms for individual branches. I never found this display particularly useful. This implements an alternative mode that creates histograms over complete branch traces, instead of individual branches, similar to how normal callgraphs are handled. This is done by putting it in front of the normal callgraph and then using the normal callgraph histogram infrastructure to unify them. This way in complex functions we can understand the control flow that lead to a particular sample, and may even see some control flow in the caller for short functions. Example (simplified, of course for such simple code this is usually not needed), please run this after the whole patchkit is in, as at this point in the patch order there is no --branch-history, that will be added in a patch after this one: tcall.c: volatile a = 10000, b = 100000, c; __attribute__((noinline)) f2() { c = a / b; } __attribute__((noinline)) f1() { f2(); f2(); } main() { int i; for (i = 0; i < 1000000; i++) f1(); } % perf record -b -g ./tsrc/tcall [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ] % perf report --no-children --branch-history ... 54.91% tcall.c:6 [.] f2 tcall | |--65.53%-- f2 tcall.c:5 | | | |--70.83%-- f1 tcall.c:11 | | f1 tcall.c:10 | | main tcall.c:18 | | main tcall.c:18 | | main tcall.c:17 | | main tcall.c:17 | | f1 tcall.c:13 | | f1 tcall.c:13 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:12 | | f1 tcall.c:12 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:11 | | | --29.17%-- f1 tcall.c:12 | f1 tcall.c:12 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:11 | f1 tcall.c:10 | main tcall.c:18 | main tcall.c:18 | main tcall.c:17 | main tcall.c:17 | f1 tcall.c:13 | f1 tcall.c:13 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:12 The default output is unchanged. This is only implemented in perf report, no change to record or anywhere else. This adds the basic code to report: - add a new "branch" option to the -g option parser to enable this mode - when the flag is set include the LBR into the callstack in machine.c. The rest of the history code is unchanged and doesn't know the difference between LBR entry and normal call entry. - detect overlaps with the callchain - remove small loop duplicates in the LBR Current limitations: - The LBR flags (mispredict etc.) are not shown in the history and LBR entries have no special marker. - It would be nice if annotate marked the LBR entries somehow (e.g. with arrows) v2: Various fixes. v3: Merge further patches into this one. Fix white space. v4: Improve manpage. Address review feedback. v5: Rename functions. Better error message without -g. Fix crash without -b. v6: Rebase v7: Rebase. Use NO_ENTRY in memset. v8: Port to latest tip. Move add_callchain_ip to separate patch. Skip initial entries in callchain. Minor cleanups. Signed-off-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1415844328-4884-3-git-send-email-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-13 10:05:20 +08:00
/*
* Based on DWARF debug information, some architectures skip
* a callchain entry saved by the kernel.
*/
skip_idx = arch_skip_callchain_idx(thread, chain);
perf callchain: Support handling complete branch stacks as histograms Currently branch stacks can be only shown as edge histograms for individual branches. I never found this display particularly useful. This implements an alternative mode that creates histograms over complete branch traces, instead of individual branches, similar to how normal callgraphs are handled. This is done by putting it in front of the normal callgraph and then using the normal callgraph histogram infrastructure to unify them. This way in complex functions we can understand the control flow that lead to a particular sample, and may even see some control flow in the caller for short functions. Example (simplified, of course for such simple code this is usually not needed), please run this after the whole patchkit is in, as at this point in the patch order there is no --branch-history, that will be added in a patch after this one: tcall.c: volatile a = 10000, b = 100000, c; __attribute__((noinline)) f2() { c = a / b; } __attribute__((noinline)) f1() { f2(); f2(); } main() { int i; for (i = 0; i < 1000000; i++) f1(); } % perf record -b -g ./tsrc/tcall [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ] % perf report --no-children --branch-history ... 54.91% tcall.c:6 [.] f2 tcall | |--65.53%-- f2 tcall.c:5 | | | |--70.83%-- f1 tcall.c:11 | | f1 tcall.c:10 | | main tcall.c:18 | | main tcall.c:18 | | main tcall.c:17 | | main tcall.c:17 | | f1 tcall.c:13 | | f1 tcall.c:13 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:12 | | f1 tcall.c:12 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:11 | | | --29.17%-- f1 tcall.c:12 | f1 tcall.c:12 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:11 | f1 tcall.c:10 | main tcall.c:18 | main tcall.c:18 | main tcall.c:17 | main tcall.c:17 | f1 tcall.c:13 | f1 tcall.c:13 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:12 The default output is unchanged. This is only implemented in perf report, no change to record or anywhere else. This adds the basic code to report: - add a new "branch" option to the -g option parser to enable this mode - when the flag is set include the LBR into the callstack in machine.c. The rest of the history code is unchanged and doesn't know the difference between LBR entry and normal call entry. - detect overlaps with the callchain - remove small loop duplicates in the LBR Current limitations: - The LBR flags (mispredict etc.) are not shown in the history and LBR entries have no special marker. - It would be nice if annotate marked the LBR entries somehow (e.g. with arrows) v2: Various fixes. v3: Merge further patches into this one. Fix white space. v4: Improve manpage. Address review feedback. v5: Rename functions. Better error message without -g. Fix crash without -b. v6: Rebase v7: Rebase. Use NO_ENTRY in memset. v8: Port to latest tip. Move add_callchain_ip to separate patch. Skip initial entries in callchain. Minor cleanups. Signed-off-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1415844328-4884-3-git-send-email-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-13 10:05:20 +08:00
/*
* Add branches to call stack for easier browsing. This gives
* more context for a sample than just the callers.
*
* This uses individual histograms of paths compared to the
* aggregated histograms the normal LBR mode uses.
*
* Limitations for now:
* - No extra filters
* - No annotations (should annotate somehow)
*/
if (branch && callchain_param.branch_callstack) {
int nr = min(max_stack, (int)branch->nr);
struct branch_entry be[nr];
if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
pr_warning("corrupted branch chain. skipping...\n");
goto check_calls;
}
for (i = 0; i < nr; i++) {
if (callchain_param.order == ORDER_CALLEE) {
be[i] = branch->entries[i];
perf report: Make --branch-history work without callgraphs(-g) option in perf record perf record -b -g <command> perf report --branch-history This merges the LBRs with the callgraphs. However it would be nice if it also works without callgraphs (-g) set in perf record, so that only the LBRs are displayed. But currently perf report errors in this case. For example, perf record -b <command> perf report --branch-history Error: Selected -g or --branch-history but no callchain data. Did you call 'perf record' without -g? This patch displays the LBRs only even if callgraphs(-g) is not enabled in perf record. Change log: v2: According to Milian Wolff's comment, change the obsolete error message. Now the error message is: ┌─Error:─────────────────────────────────────┐ │Selected -g or --branch-history. │ │But no callchain or branch data. │ │Did you call 'perf record' without -g or -b?│ │ │ │ │ │Press any key... │ └────────────────────────────────────────────┘ When passing the last parameter to hists__fprintf, changes "|" to "||". hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout, symbol_conf.use_callchain || symbol_conf.show_branchflag_count); Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Reviewed-by: Andi Kleen <ak@linux.intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/1494240182-28899-1-git-send-email-yao.jin@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-05-08 18:43:02 +08:00
if (chain == NULL)
continue;
perf callchain: Support handling complete branch stacks as histograms Currently branch stacks can be only shown as edge histograms for individual branches. I never found this display particularly useful. This implements an alternative mode that creates histograms over complete branch traces, instead of individual branches, similar to how normal callgraphs are handled. This is done by putting it in front of the normal callgraph and then using the normal callgraph histogram infrastructure to unify them. This way in complex functions we can understand the control flow that lead to a particular sample, and may even see some control flow in the caller for short functions. Example (simplified, of course for such simple code this is usually not needed), please run this after the whole patchkit is in, as at this point in the patch order there is no --branch-history, that will be added in a patch after this one: tcall.c: volatile a = 10000, b = 100000, c; __attribute__((noinline)) f2() { c = a / b; } __attribute__((noinline)) f1() { f2(); f2(); } main() { int i; for (i = 0; i < 1000000; i++) f1(); } % perf record -b -g ./tsrc/tcall [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ] % perf report --no-children --branch-history ... 54.91% tcall.c:6 [.] f2 tcall | |--65.53%-- f2 tcall.c:5 | | | |--70.83%-- f1 tcall.c:11 | | f1 tcall.c:10 | | main tcall.c:18 | | main tcall.c:18 | | main tcall.c:17 | | main tcall.c:17 | | f1 tcall.c:13 | | f1 tcall.c:13 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:12 | | f1 tcall.c:12 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:11 | | | --29.17%-- f1 tcall.c:12 | f1 tcall.c:12 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:11 | f1 tcall.c:10 | main tcall.c:18 | main tcall.c:18 | main tcall.c:17 | main tcall.c:17 | f1 tcall.c:13 | f1 tcall.c:13 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:12 The default output is unchanged. This is only implemented in perf report, no change to record or anywhere else. This adds the basic code to report: - add a new "branch" option to the -g option parser to enable this mode - when the flag is set include the LBR into the callstack in machine.c. The rest of the history code is unchanged and doesn't know the difference between LBR entry and normal call entry. - detect overlaps with the callchain - remove small loop duplicates in the LBR Current limitations: - The LBR flags (mispredict etc.) are not shown in the history and LBR entries have no special marker. - It would be nice if annotate marked the LBR entries somehow (e.g. with arrows) v2: Various fixes. v3: Merge further patches into this one. Fix white space. v4: Improve manpage. Address review feedback. v5: Rename functions. Better error message without -g. Fix crash without -b. v6: Rebase v7: Rebase. Use NO_ENTRY in memset. v8: Port to latest tip. Move add_callchain_ip to separate patch. Skip initial entries in callchain. Minor cleanups. Signed-off-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1415844328-4884-3-git-send-email-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-13 10:05:20 +08:00
/*
* Check for overlap into the callchain.
* The return address is one off compared to
* the branch entry. To adjust for this
* assume the calling instruction is not longer
* than 8 bytes.
*/
if (i == skip_idx ||
chain->ips[first_call] >= PERF_CONTEXT_MAX)
first_call++;
else if (be[i].from < chain->ips[first_call] &&
be[i].from >= chain->ips[first_call] - 8)
first_call++;
} else
be[i] = branch->entries[branch->nr - i - 1];
}
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
nr_loop_iter = nr;
perf callchain: Support handling complete branch stacks as histograms Currently branch stacks can be only shown as edge histograms for individual branches. I never found this display particularly useful. This implements an alternative mode that creates histograms over complete branch traces, instead of individual branches, similar to how normal callgraphs are handled. This is done by putting it in front of the normal callgraph and then using the normal callgraph histogram infrastructure to unify them. This way in complex functions we can understand the control flow that lead to a particular sample, and may even see some control flow in the caller for short functions. Example (simplified, of course for such simple code this is usually not needed), please run this after the whole patchkit is in, as at this point in the patch order there is no --branch-history, that will be added in a patch after this one: tcall.c: volatile a = 10000, b = 100000, c; __attribute__((noinline)) f2() { c = a / b; } __attribute__((noinline)) f1() { f2(); f2(); } main() { int i; for (i = 0; i < 1000000; i++) f1(); } % perf record -b -g ./tsrc/tcall [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ] % perf report --no-children --branch-history ... 54.91% tcall.c:6 [.] f2 tcall | |--65.53%-- f2 tcall.c:5 | | | |--70.83%-- f1 tcall.c:11 | | f1 tcall.c:10 | | main tcall.c:18 | | main tcall.c:18 | | main tcall.c:17 | | main tcall.c:17 | | f1 tcall.c:13 | | f1 tcall.c:13 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:12 | | f1 tcall.c:12 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:11 | | | --29.17%-- f1 tcall.c:12 | f1 tcall.c:12 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:11 | f1 tcall.c:10 | main tcall.c:18 | main tcall.c:18 | main tcall.c:17 | main tcall.c:17 | f1 tcall.c:13 | f1 tcall.c:13 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:12 The default output is unchanged. This is only implemented in perf report, no change to record or anywhere else. This adds the basic code to report: - add a new "branch" option to the -g option parser to enable this mode - when the flag is set include the LBR into the callstack in machine.c. The rest of the history code is unchanged and doesn't know the difference between LBR entry and normal call entry. - detect overlaps with the callchain - remove small loop duplicates in the LBR Current limitations: - The LBR flags (mispredict etc.) are not shown in the history and LBR entries have no special marker. - It would be nice if annotate marked the LBR entries somehow (e.g. with arrows) v2: Various fixes. v3: Merge further patches into this one. Fix white space. v4: Improve manpage. Address review feedback. v5: Rename functions. Better error message without -g. Fix crash without -b. v6: Rebase v7: Rebase. Use NO_ENTRY in memset. v8: Port to latest tip. Move add_callchain_ip to separate patch. Skip initial entries in callchain. Minor cleanups. Signed-off-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1415844328-4884-3-git-send-email-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-13 10:05:20 +08:00
nr = remove_loops(be, nr);
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
/*
* Get the number of iterations.
* It's only approximation, but good enough in practice.
*/
if (nr_loop_iter > nr)
nr_loop_iter = nr_loop_iter - nr + 1;
else
nr_loop_iter = 0;
perf callchain: Support handling complete branch stacks as histograms Currently branch stacks can be only shown as edge histograms for individual branches. I never found this display particularly useful. This implements an alternative mode that creates histograms over complete branch traces, instead of individual branches, similar to how normal callgraphs are handled. This is done by putting it in front of the normal callgraph and then using the normal callgraph histogram infrastructure to unify them. This way in complex functions we can understand the control flow that lead to a particular sample, and may even see some control flow in the caller for short functions. Example (simplified, of course for such simple code this is usually not needed), please run this after the whole patchkit is in, as at this point in the patch order there is no --branch-history, that will be added in a patch after this one: tcall.c: volatile a = 10000, b = 100000, c; __attribute__((noinline)) f2() { c = a / b; } __attribute__((noinline)) f1() { f2(); f2(); } main() { int i; for (i = 0; i < 1000000; i++) f1(); } % perf record -b -g ./tsrc/tcall [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ] % perf report --no-children --branch-history ... 54.91% tcall.c:6 [.] f2 tcall | |--65.53%-- f2 tcall.c:5 | | | |--70.83%-- f1 tcall.c:11 | | f1 tcall.c:10 | | main tcall.c:18 | | main tcall.c:18 | | main tcall.c:17 | | main tcall.c:17 | | f1 tcall.c:13 | | f1 tcall.c:13 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:12 | | f1 tcall.c:12 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:11 | | | --29.17%-- f1 tcall.c:12 | f1 tcall.c:12 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:11 | f1 tcall.c:10 | main tcall.c:18 | main tcall.c:18 | main tcall.c:17 | main tcall.c:17 | f1 tcall.c:13 | f1 tcall.c:13 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:12 The default output is unchanged. This is only implemented in perf report, no change to record or anywhere else. This adds the basic code to report: - add a new "branch" option to the -g option parser to enable this mode - when the flag is set include the LBR into the callstack in machine.c. The rest of the history code is unchanged and doesn't know the difference between LBR entry and normal call entry. - detect overlaps with the callchain - remove small loop duplicates in the LBR Current limitations: - The LBR flags (mispredict etc.) are not shown in the history and LBR entries have no special marker. - It would be nice if annotate marked the LBR entries somehow (e.g. with arrows) v2: Various fixes. v3: Merge further patches into this one. Fix white space. v4: Improve manpage. Address review feedback. v5: Rename functions. Better error message without -g. Fix crash without -b. v6: Rebase v7: Rebase. Use NO_ENTRY in memset. v8: Port to latest tip. Move add_callchain_ip to separate patch. Skip initial entries in callchain. Minor cleanups. Signed-off-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1415844328-4884-3-git-send-email-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-13 10:05:20 +08:00
for (i = 0; i < nr; i++) {
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
if (i == nr - 1)
err = add_callchain_ip(thread, cursor, parent,
root_al,
NULL, be[i].to,
true, &be[i].flags,
nr_loop_iter, 1,
be[i].from);
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
else
err = add_callchain_ip(thread, cursor, parent,
root_al,
NULL, be[i].to,
true, &be[i].flags,
0, 0, be[i].from);
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
perf callchain: Support handling complete branch stacks as histograms Currently branch stacks can be only shown as edge histograms for individual branches. I never found this display particularly useful. This implements an alternative mode that creates histograms over complete branch traces, instead of individual branches, similar to how normal callgraphs are handled. This is done by putting it in front of the normal callgraph and then using the normal callgraph histogram infrastructure to unify them. This way in complex functions we can understand the control flow that lead to a particular sample, and may even see some control flow in the caller for short functions. Example (simplified, of course for such simple code this is usually not needed), please run this after the whole patchkit is in, as at this point in the patch order there is no --branch-history, that will be added in a patch after this one: tcall.c: volatile a = 10000, b = 100000, c; __attribute__((noinline)) f2() { c = a / b; } __attribute__((noinline)) f1() { f2(); f2(); } main() { int i; for (i = 0; i < 1000000; i++) f1(); } % perf record -b -g ./tsrc/tcall [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ] % perf report --no-children --branch-history ... 54.91% tcall.c:6 [.] f2 tcall | |--65.53%-- f2 tcall.c:5 | | | |--70.83%-- f1 tcall.c:11 | | f1 tcall.c:10 | | main tcall.c:18 | | main tcall.c:18 | | main tcall.c:17 | | main tcall.c:17 | | f1 tcall.c:13 | | f1 tcall.c:13 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:12 | | f1 tcall.c:12 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:11 | | | --29.17%-- f1 tcall.c:12 | f1 tcall.c:12 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:11 | f1 tcall.c:10 | main tcall.c:18 | main tcall.c:18 | main tcall.c:17 | main tcall.c:17 | f1 tcall.c:13 | f1 tcall.c:13 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:12 The default output is unchanged. This is only implemented in perf report, no change to record or anywhere else. This adds the basic code to report: - add a new "branch" option to the -g option parser to enable this mode - when the flag is set include the LBR into the callstack in machine.c. The rest of the history code is unchanged and doesn't know the difference between LBR entry and normal call entry. - detect overlaps with the callchain - remove small loop duplicates in the LBR Current limitations: - The LBR flags (mispredict etc.) are not shown in the history and LBR entries have no special marker. - It would be nice if annotate marked the LBR entries somehow (e.g. with arrows) v2: Various fixes. v3: Merge further patches into this one. Fix white space. v4: Improve manpage. Address review feedback. v5: Rename functions. Better error message without -g. Fix crash without -b. v6: Rebase v7: Rebase. Use NO_ENTRY in memset. v8: Port to latest tip. Move add_callchain_ip to separate patch. Skip initial entries in callchain. Minor cleanups. Signed-off-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1415844328-4884-3-git-send-email-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-13 10:05:20 +08:00
if (!err)
err = add_callchain_ip(thread, cursor, parent, root_al,
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
NULL, be[i].from,
true, &be[i].flags,
0, 0, 0);
perf callchain: Support handling complete branch stacks as histograms Currently branch stacks can be only shown as edge histograms for individual branches. I never found this display particularly useful. This implements an alternative mode that creates histograms over complete branch traces, instead of individual branches, similar to how normal callgraphs are handled. This is done by putting it in front of the normal callgraph and then using the normal callgraph histogram infrastructure to unify them. This way in complex functions we can understand the control flow that lead to a particular sample, and may even see some control flow in the caller for short functions. Example (simplified, of course for such simple code this is usually not needed), please run this after the whole patchkit is in, as at this point in the patch order there is no --branch-history, that will be added in a patch after this one: tcall.c: volatile a = 10000, b = 100000, c; __attribute__((noinline)) f2() { c = a / b; } __attribute__((noinline)) f1() { f2(); f2(); } main() { int i; for (i = 0; i < 1000000; i++) f1(); } % perf record -b -g ./tsrc/tcall [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ] % perf report --no-children --branch-history ... 54.91% tcall.c:6 [.] f2 tcall | |--65.53%-- f2 tcall.c:5 | | | |--70.83%-- f1 tcall.c:11 | | f1 tcall.c:10 | | main tcall.c:18 | | main tcall.c:18 | | main tcall.c:17 | | main tcall.c:17 | | f1 tcall.c:13 | | f1 tcall.c:13 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:12 | | f1 tcall.c:12 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:11 | | | --29.17%-- f1 tcall.c:12 | f1 tcall.c:12 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:11 | f1 tcall.c:10 | main tcall.c:18 | main tcall.c:18 | main tcall.c:17 | main tcall.c:17 | f1 tcall.c:13 | f1 tcall.c:13 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:12 The default output is unchanged. This is only implemented in perf report, no change to record or anywhere else. This adds the basic code to report: - add a new "branch" option to the -g option parser to enable this mode - when the flag is set include the LBR into the callstack in machine.c. The rest of the history code is unchanged and doesn't know the difference between LBR entry and normal call entry. - detect overlaps with the callchain - remove small loop duplicates in the LBR Current limitations: - The LBR flags (mispredict etc.) are not shown in the history and LBR entries have no special marker. - It would be nice if annotate marked the LBR entries somehow (e.g. with arrows) v2: Various fixes. v3: Merge further patches into this one. Fix white space. v4: Improve manpage. Address review feedback. v5: Rename functions. Better error message without -g. Fix crash without -b. v6: Rebase v7: Rebase. Use NO_ENTRY in memset. v8: Port to latest tip. Move add_callchain_ip to separate patch. Skip initial entries in callchain. Minor cleanups. Signed-off-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1415844328-4884-3-git-send-email-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-13 10:05:20 +08:00
if (err == -EINVAL)
break;
if (err)
return err;
}
perf report: Make --branch-history work without callgraphs(-g) option in perf record perf record -b -g <command> perf report --branch-history This merges the LBRs with the callgraphs. However it would be nice if it also works without callgraphs (-g) set in perf record, so that only the LBRs are displayed. But currently perf report errors in this case. For example, perf record -b <command> perf report --branch-history Error: Selected -g or --branch-history but no callchain data. Did you call 'perf record' without -g? This patch displays the LBRs only even if callgraphs(-g) is not enabled in perf record. Change log: v2: According to Milian Wolff's comment, change the obsolete error message. Now the error message is: ┌─Error:─────────────────────────────────────┐ │Selected -g or --branch-history. │ │But no callchain or branch data. │ │Did you call 'perf record' without -g or -b?│ │ │ │ │ │Press any key... │ └────────────────────────────────────────────┘ When passing the last parameter to hists__fprintf, changes "|" to "||". hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout, symbol_conf.use_callchain || symbol_conf.show_branchflag_count); Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Reviewed-by: Andi Kleen <ak@linux.intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/1494240182-28899-1-git-send-email-yao.jin@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-05-08 18:43:02 +08:00
if (chain_nr == 0)
return 0;
perf callchain: Support handling complete branch stacks as histograms Currently branch stacks can be only shown as edge histograms for individual branches. I never found this display particularly useful. This implements an alternative mode that creates histograms over complete branch traces, instead of individual branches, similar to how normal callgraphs are handled. This is done by putting it in front of the normal callgraph and then using the normal callgraph histogram infrastructure to unify them. This way in complex functions we can understand the control flow that lead to a particular sample, and may even see some control flow in the caller for short functions. Example (simplified, of course for such simple code this is usually not needed), please run this after the whole patchkit is in, as at this point in the patch order there is no --branch-history, that will be added in a patch after this one: tcall.c: volatile a = 10000, b = 100000, c; __attribute__((noinline)) f2() { c = a / b; } __attribute__((noinline)) f1() { f2(); f2(); } main() { int i; for (i = 0; i < 1000000; i++) f1(); } % perf record -b -g ./tsrc/tcall [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.044 MB perf.data (~1923 samples) ] % perf report --no-children --branch-history ... 54.91% tcall.c:6 [.] f2 tcall | |--65.53%-- f2 tcall.c:5 | | | |--70.83%-- f1 tcall.c:11 | | f1 tcall.c:10 | | main tcall.c:18 | | main tcall.c:18 | | main tcall.c:17 | | main tcall.c:17 | | f1 tcall.c:13 | | f1 tcall.c:13 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:12 | | f1 tcall.c:12 | | f2 tcall.c:7 | | f2 tcall.c:5 | | f1 tcall.c:11 | | | --29.17%-- f1 tcall.c:12 | f1 tcall.c:12 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:11 | f1 tcall.c:10 | main tcall.c:18 | main tcall.c:18 | main tcall.c:17 | main tcall.c:17 | f1 tcall.c:13 | f1 tcall.c:13 | f2 tcall.c:7 | f2 tcall.c:5 | f1 tcall.c:12 The default output is unchanged. This is only implemented in perf report, no change to record or anywhere else. This adds the basic code to report: - add a new "branch" option to the -g option parser to enable this mode - when the flag is set include the LBR into the callstack in machine.c. The rest of the history code is unchanged and doesn't know the difference between LBR entry and normal call entry. - detect overlaps with the callchain - remove small loop duplicates in the LBR Current limitations: - The LBR flags (mispredict etc.) are not shown in the history and LBR entries have no special marker. - It would be nice if annotate marked the LBR entries somehow (e.g. with arrows) v2: Various fixes. v3: Merge further patches into this one. Fix white space. v4: Improve manpage. Address review feedback. v5: Rename functions. Better error message without -g. Fix crash without -b. v6: Rebase v7: Rebase. Use NO_ENTRY in memset. v8: Port to latest tip. Move add_callchain_ip to separate patch. Skip initial entries in callchain. Minor cleanups. Signed-off-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1415844328-4884-3-git-send-email-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2014-11-13 10:05:20 +08:00
chain_nr -= nr;
}
check_calls:
for (i = first_call, nr_entries = 0;
i < chain_nr && nr_entries < max_stack; i++) {
u64 ip;
if (callchain_param.order == ORDER_CALLEE)
perf tools powerpc: Adjust callchain based on DWARF debug info When saving the callchain on Power, the kernel conservatively saves excess entries in the callchain. A few of these entries are needed in some cases but not others. We should use the DWARF debug information to determine when the entries are needed. Eg: the value in the link register (LR) is needed only when it holds the return address of a function. At other times it must be ignored. If the unnecessary entries are not ignored, we end up with duplicate arcs in the call-graphs. Use the DWARF debug information to determine if any callchain entries should be ignored when building call-graphs. Callgraph before the patch: 14.67% 2234 sprintft libc-2.18.so [.] __random | --- __random | |--61.12%-- __random | | | |--97.15%-- rand | | do_my_sprintf | | main | | generic_start_main.isra.0 | | __libc_start_main | | 0x0 | | | --2.85%-- do_my_sprintf | main | generic_start_main.isra.0 | __libc_start_main | 0x0 | --38.88%-- rand | |--94.01%-- rand | do_my_sprintf | main | generic_start_main.isra.0 | __libc_start_main | 0x0 | --5.99%-- do_my_sprintf main generic_start_main.isra.0 __libc_start_main 0x0 Callgraph after the patch: 14.67% 2234 sprintft libc-2.18.so [.] __random | --- __random | |--95.93%-- rand | do_my_sprintf | main | generic_start_main.isra.0 | __libc_start_main | 0x0 | --4.07%-- do_my_sprintf main generic_start_main.isra.0 __libc_start_main 0x0 TODO: For split-debug info objects like glibc, we can only determine the call-frame-address only when both .eh_frame and .debug_info sections are available. We should be able to determin the CFA even without the .eh_frame section. Fix suggested by Anton Blanchard. Thanks to valuable input on DWARF debug information from Ulrich Weigand. Reported-by: Maynard Johnson <maynard@us.ibm.com> Tested-by: Maynard Johnson <maynard@us.ibm.com> Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com> Link: http://lkml.kernel.org/r/20140625154903.GA29607@us.ibm.com Signed-off-by: Jiri Olsa <jolsa@kernel.org>
2014-06-25 23:49:03 +08:00
j = i;
else
perf tools powerpc: Adjust callchain based on DWARF debug info When saving the callchain on Power, the kernel conservatively saves excess entries in the callchain. A few of these entries are needed in some cases but not others. We should use the DWARF debug information to determine when the entries are needed. Eg: the value in the link register (LR) is needed only when it holds the return address of a function. At other times it must be ignored. If the unnecessary entries are not ignored, we end up with duplicate arcs in the call-graphs. Use the DWARF debug information to determine if any callchain entries should be ignored when building call-graphs. Callgraph before the patch: 14.67% 2234 sprintft libc-2.18.so [.] __random | --- __random | |--61.12%-- __random | | | |--97.15%-- rand | | do_my_sprintf | | main | | generic_start_main.isra.0 | | __libc_start_main | | 0x0 | | | --2.85%-- do_my_sprintf | main | generic_start_main.isra.0 | __libc_start_main | 0x0 | --38.88%-- rand | |--94.01%-- rand | do_my_sprintf | main | generic_start_main.isra.0 | __libc_start_main | 0x0 | --5.99%-- do_my_sprintf main generic_start_main.isra.0 __libc_start_main 0x0 Callgraph after the patch: 14.67% 2234 sprintft libc-2.18.so [.] __random | --- __random | |--95.93%-- rand | do_my_sprintf | main | generic_start_main.isra.0 | __libc_start_main | 0x0 | --4.07%-- do_my_sprintf main generic_start_main.isra.0 __libc_start_main 0x0 TODO: For split-debug info objects like glibc, we can only determine the call-frame-address only when both .eh_frame and .debug_info sections are available. We should be able to determin the CFA even without the .eh_frame section. Fix suggested by Anton Blanchard. Thanks to valuable input on DWARF debug information from Ulrich Weigand. Reported-by: Maynard Johnson <maynard@us.ibm.com> Tested-by: Maynard Johnson <maynard@us.ibm.com> Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com> Link: http://lkml.kernel.org/r/20140625154903.GA29607@us.ibm.com Signed-off-by: Jiri Olsa <jolsa@kernel.org>
2014-06-25 23:49:03 +08:00
j = chain->nr - i - 1;
#ifdef HAVE_SKIP_CALLCHAIN_IDX
if (j == skip_idx)
continue;
#endif
ip = chain->ips[j];
if (ip < PERF_CONTEXT_MAX)
++nr_entries;
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
err = add_callchain_ip(thread, cursor, parent,
root_al, &cpumode, ip,
false, NULL, 0, 0, 0);
if (err)
return (err < 0) ? err : 0;
}
return 0;
}
static int unwind_entry(struct unwind_entry *entry, void *arg)
{
struct callchain_cursor *cursor = arg;
if (symbol_conf.hide_unresolved && entry->sym == NULL)
return 0;
return callchain_cursor_append(cursor, entry->ip,
perf report: Add branch flag to callchain cursor node Since the branch ip has been added to call stack for easier browsing, this patch adds more branch information. For example, add a flag to indicate if this ip is a branch, and also add with the branch flag. Then we can know if the cursor node represents a branch and know what the branch flag it has. The branch history code has a loop detection pass that removes loops. It would be nice for knowing how many loops were removed then in next steps, we can compute out the average number of iterations. For example: Before remove_loops(), entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x300, to = 0x250 entry3: from = 0x300, to = 0x250 entry4: from = 0x700, to = 0x800 After remove_loops() entry0: from = 0x100, to = 0x200 entry1: from = 0x300, to = 0x250 entry2: from = 0x700, to = 0x800 The original entry2 and entry3 are removed. So the number of iterations (from = 0x300, to = 0x250) is equal to removed number + 1 (2 + 1). iterations = removed number + 1; average iteractions = Sum(iteractions) / number of samples This formula ignores other cases, for example, iterations cross multiple buffers and one buffer contains 2+ loops. Because in practice, it's good enough. Signed-off-by: Yao Jin <yao.jin@linux.intel.com> Acked-by: Andi Kleen <ak@linux.intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@intel.com> Cc: Linux-kernel@vger.kernel.org Cc: Yao Jin <yao.jin@linux.intel.com> Link: http://lkml.kernel.org/n/1477876794-30749-2-git-send-email-yao.jin@linux.intel.com [ Renamed 'iter' to 'nr_loop_iter' for clarity ] Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-10-31 09:19:49 +08:00
entry->map, entry->sym,
false, NULL, 0, 0, 0);
}
perf callchain: Fix incorrect ordering of entries The existing implementation of thread__resolve_callchain, under certain circumstances, can assemble callchain entries in the incorrect order. The callchain entries are resolved incorrectly for a sample when all of the following conditions are met: 1. callchain_param.order is set to ORDER_CALLER 2. thread__resolve_callchain_sample is able to resolve callchain entries for the sample. 3. unwind__get_entries is also able to resolve callchain entries for the sample. The fix is accomplished by reversing the order in which thread__resolve_callchain_sample and unwind__get_entries are called when callchain_param.order is set to ORDER_CALLER. Unwind specific code from thread__resolve_callchain is also moved into a new static function to improve readability of the fix. How to Reproduce the Existing Bug: Modifying perf script to print call trees in the opposite order or applying the remaining patches from this series and comparing the results output from export-to-postgtresql.py are the easiest ways to see the bug, however it can still be seen in current builds using perf report. Here is how i can reproduce the bug using perf report: # perf record --call-graph=dwarf stress -c 1 -t 5 when i run this command: # perf report --call-graph=flat,0,0,callee This callchain, containing kernel (handle_irq_event, etc) and userspace samples (__libc_start_main, etc) is contained in the output, which looks correct (callee order): gen8_irq_handler handle_irq_event_percpu handle_irq_event handle_edge_irq handle_irq do_IRQ ret_from_intr __random rand 0x558f2a04dded 0x558f2a04c774 __libc_start_main 0x558f2a04dcd9 Now run this command using caller order: # perf report --call-graph=flat,0,0,caller It is expected to see the exact reverse of the above when using caller order (with "0x558f2a04dcd9" at the top and "gen8_irq_handler" at the bottom) in the output, but it is nowhere to be found. instead you see this: ret_from_intr do_IRQ handle_irq handle_edge_irq handle_irq_event handle_irq_event_percpu gen8_irq_handler 0x558f2a04dcd9 __libc_start_main 0x558f2a04c774 0x558f2a04dded rand __random Notice how internally the kernel symbols are reversed and the user space symbols are reversed, but the kernel symbols still appear above the user space symbols. if this patch is applied and perf script is re-run, you will see the expected output (with "0x558f2a04dcd9" at the top and "gen8_irq_handler" at the bottom): 0x558f2a04dcd9 __libc_start_main 0x558f2a04c774 0x558f2a04dded rand __random ret_from_intr do_IRQ handle_irq handle_edge_irq handle_irq_event handle_irq_event_percpu gen8_irq_handler Signed-off-by: Chris Phlipot <cphlipot0@gmail.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/1461831551-12213-2-git-send-email-cphlipot0@gmail.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-04-28 16:19:06 +08:00
static int thread__resolve_callchain_unwind(struct thread *thread,
struct callchain_cursor *cursor,
struct perf_evsel *evsel,
struct perf_sample *sample,
int max_stack)
{
/* Can we do dwarf post unwind? */
if (!((evsel->attr.sample_type & PERF_SAMPLE_REGS_USER) &&
(evsel->attr.sample_type & PERF_SAMPLE_STACK_USER)))
return 0;
/* Bail out if nothing was captured. */
if ((!sample->user_regs.regs) ||
(!sample->user_stack.size))
return 0;
return unwind__get_entries(unwind_entry, cursor,
thread, sample, max_stack);
perf callchain: Fix incorrect ordering of entries The existing implementation of thread__resolve_callchain, under certain circumstances, can assemble callchain entries in the incorrect order. The callchain entries are resolved incorrectly for a sample when all of the following conditions are met: 1. callchain_param.order is set to ORDER_CALLER 2. thread__resolve_callchain_sample is able to resolve callchain entries for the sample. 3. unwind__get_entries is also able to resolve callchain entries for the sample. The fix is accomplished by reversing the order in which thread__resolve_callchain_sample and unwind__get_entries are called when callchain_param.order is set to ORDER_CALLER. Unwind specific code from thread__resolve_callchain is also moved into a new static function to improve readability of the fix. How to Reproduce the Existing Bug: Modifying perf script to print call trees in the opposite order or applying the remaining patches from this series and comparing the results output from export-to-postgtresql.py are the easiest ways to see the bug, however it can still be seen in current builds using perf report. Here is how i can reproduce the bug using perf report: # perf record --call-graph=dwarf stress -c 1 -t 5 when i run this command: # perf report --call-graph=flat,0,0,callee This callchain, containing kernel (handle_irq_event, etc) and userspace samples (__libc_start_main, etc) is contained in the output, which looks correct (callee order): gen8_irq_handler handle_irq_event_percpu handle_irq_event handle_edge_irq handle_irq do_IRQ ret_from_intr __random rand 0x558f2a04dded 0x558f2a04c774 __libc_start_main 0x558f2a04dcd9 Now run this command using caller order: # perf report --call-graph=flat,0,0,caller It is expected to see the exact reverse of the above when using caller order (with "0x558f2a04dcd9" at the top and "gen8_irq_handler" at the bottom) in the output, but it is nowhere to be found. instead you see this: ret_from_intr do_IRQ handle_irq handle_edge_irq handle_irq_event handle_irq_event_percpu gen8_irq_handler 0x558f2a04dcd9 __libc_start_main 0x558f2a04c774 0x558f2a04dded rand __random Notice how internally the kernel symbols are reversed and the user space symbols are reversed, but the kernel symbols still appear above the user space symbols. if this patch is applied and perf script is re-run, you will see the expected output (with "0x558f2a04dcd9" at the top and "gen8_irq_handler" at the bottom): 0x558f2a04dcd9 __libc_start_main 0x558f2a04c774 0x558f2a04dded rand __random ret_from_intr do_IRQ handle_irq handle_edge_irq handle_irq_event handle_irq_event_percpu gen8_irq_handler Signed-off-by: Chris Phlipot <cphlipot0@gmail.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/1461831551-12213-2-git-send-email-cphlipot0@gmail.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-04-28 16:19:06 +08:00
}
perf callchain: Fix incorrect ordering of entries The existing implementation of thread__resolve_callchain, under certain circumstances, can assemble callchain entries in the incorrect order. The callchain entries are resolved incorrectly for a sample when all of the following conditions are met: 1. callchain_param.order is set to ORDER_CALLER 2. thread__resolve_callchain_sample is able to resolve callchain entries for the sample. 3. unwind__get_entries is also able to resolve callchain entries for the sample. The fix is accomplished by reversing the order in which thread__resolve_callchain_sample and unwind__get_entries are called when callchain_param.order is set to ORDER_CALLER. Unwind specific code from thread__resolve_callchain is also moved into a new static function to improve readability of the fix. How to Reproduce the Existing Bug: Modifying perf script to print call trees in the opposite order or applying the remaining patches from this series and comparing the results output from export-to-postgtresql.py are the easiest ways to see the bug, however it can still be seen in current builds using perf report. Here is how i can reproduce the bug using perf report: # perf record --call-graph=dwarf stress -c 1 -t 5 when i run this command: # perf report --call-graph=flat,0,0,callee This callchain, containing kernel (handle_irq_event, etc) and userspace samples (__libc_start_main, etc) is contained in the output, which looks correct (callee order): gen8_irq_handler handle_irq_event_percpu handle_irq_event handle_edge_irq handle_irq do_IRQ ret_from_intr __random rand 0x558f2a04dded 0x558f2a04c774 __libc_start_main 0x558f2a04dcd9 Now run this command using caller order: # perf report --call-graph=flat,0,0,caller It is expected to see the exact reverse of the above when using caller order (with "0x558f2a04dcd9" at the top and "gen8_irq_handler" at the bottom) in the output, but it is nowhere to be found. instead you see this: ret_from_intr do_IRQ handle_irq handle_edge_irq handle_irq_event handle_irq_event_percpu gen8_irq_handler 0x558f2a04dcd9 __libc_start_main 0x558f2a04c774 0x558f2a04dded rand __random Notice how internally the kernel symbols are reversed and the user space symbols are reversed, but the kernel symbols still appear above the user space symbols. if this patch is applied and perf script is re-run, you will see the expected output (with "0x558f2a04dcd9" at the top and "gen8_irq_handler" at the bottom): 0x558f2a04dcd9 __libc_start_main 0x558f2a04c774 0x558f2a04dded rand __random ret_from_intr do_IRQ handle_irq handle_edge_irq handle_irq_event handle_irq_event_percpu gen8_irq_handler Signed-off-by: Chris Phlipot <cphlipot0@gmail.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/1461831551-12213-2-git-send-email-cphlipot0@gmail.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-04-28 16:19:06 +08:00
int thread__resolve_callchain(struct thread *thread,
struct callchain_cursor *cursor,
struct perf_evsel *evsel,
struct perf_sample *sample,
struct symbol **parent,
struct addr_location *root_al,
int max_stack)
{
int ret = 0;
callchain_cursor_reset(&callchain_cursor);
if (callchain_param.order == ORDER_CALLEE) {
ret = thread__resolve_callchain_sample(thread, cursor,
evsel, sample,
parent, root_al,
max_stack);
if (ret)
return ret;
ret = thread__resolve_callchain_unwind(thread, cursor,
evsel, sample,
max_stack);
} else {
ret = thread__resolve_callchain_unwind(thread, cursor,
evsel, sample,
max_stack);
if (ret)
return ret;
ret = thread__resolve_callchain_sample(thread, cursor,
evsel, sample,
parent, root_al,
max_stack);
}
return ret;
}
int machine__for_each_thread(struct machine *machine,
int (*fn)(struct thread *thread, void *p),
void *priv)
{
struct rb_node *nd;
struct thread *thread;
int rc = 0;
for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
thread = rb_entry(nd, struct thread, rb_node);
rc = fn(thread, priv);
if (rc != 0)
return rc;
}
list_for_each_entry(thread, &machine->dead_threads, node) {
rc = fn(thread, priv);
if (rc != 0)
return rc;
}
return rc;
}
int machines__for_each_thread(struct machines *machines,
int (*fn)(struct thread *thread, void *p),
void *priv)
{
struct rb_node *nd;
int rc = 0;
rc = machine__for_each_thread(&machines->host, fn, priv);
if (rc != 0)
return rc;
for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
struct machine *machine = rb_entry(nd, struct machine, rb_node);
rc = machine__for_each_thread(machine, fn, priv);
if (rc != 0)
return rc;
}
return rc;
}
int __machine__synthesize_threads(struct machine *machine, struct perf_tool *tool,
struct target *target, struct thread_map *threads,
perf_event__handler_t process, bool data_mmap,
unsigned int proc_map_timeout)
{
if (target__has_task(target))
return perf_event__synthesize_thread_map(tool, threads, process, machine, data_mmap, proc_map_timeout);
else if (target__has_cpu(target))
return perf_event__synthesize_threads(tool, process, machine, data_mmap, proc_map_timeout);
/* command specified */
return 0;
}
pid_t machine__get_current_tid(struct machine *machine, int cpu)
{
if (cpu < 0 || cpu >= MAX_NR_CPUS || !machine->current_tid)
return -1;
return machine->current_tid[cpu];
}
int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
pid_t tid)
{
struct thread *thread;
if (cpu < 0)
return -EINVAL;
if (!machine->current_tid) {
int i;
machine->current_tid = calloc(MAX_NR_CPUS, sizeof(pid_t));
if (!machine->current_tid)
return -ENOMEM;
for (i = 0; i < MAX_NR_CPUS; i++)
machine->current_tid[i] = -1;
}
if (cpu >= MAX_NR_CPUS) {
pr_err("Requested CPU %d too large. ", cpu);
pr_err("Consider raising MAX_NR_CPUS\n");
return -EINVAL;
}
machine->current_tid[cpu] = tid;
thread = machine__findnew_thread(machine, pid, tid);
if (!thread)
return -ENOMEM;
thread->cpu = cpu;
perf machine: Protect the machine->threads with a rwlock In addition to using refcounts for the struct thread lifetime management, we need to protect access to machine->threads from concurrent access. That happens in 'perf top', where a thread processes events, inserting and deleting entries from that rb_tree while another thread decays hist_entries, that end up dropping references and ultimately deleting threads from the rb_tree and releasing its resources when no further hist_entry (or other data structures, like in 'perf sched') references it. So the rule is the same for refcounts + protected trees in the kernel, get the tree lock, find object, bump the refcount, drop the tree lock, return, use object, drop the refcount if no more use of it is needed, keep it if storing it in some other data structure, drop when releasing that data structure. I.e. pair "t = machine__find(new)_thread()" with a "thread__put(t)", and "perf_event__preprocess_sample(&al)" with "addr_location__put(&al)". The addr_location__put() one is because as we return references to several data structures, we may end up adding more reference counting for the other data structures and then we'll drop it at addr_location__put() time. Acked-by: David Ahern <dsahern@gmail.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Borislav Petkov <bp@suse.de> Cc: Don Zickus <dzickus@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-bs9rt4n0jw3hi9f3zxyy3xln@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-04-07 07:43:22 +08:00
thread__put(thread);
return 0;
}
int machine__get_kernel_start(struct machine *machine)
{
struct map *map = machine__kernel_map(machine);
int err = 0;
/*
* The only addresses above 2^63 are kernel addresses of a 64-bit
* kernel. Note that addresses are unsigned so that on a 32-bit system
* all addresses including kernel addresses are less than 2^32. In
* that case (32-bit system), if the kernel mapping is unknown, all
* addresses will be assumed to be in user space - see
* machine__kernel_ip().
*/
machine->kernel_start = 1ULL << 63;
if (map) {
err = map__load(map);
perf symbols: Accept zero as the kernel base address Which is the case in S/390, where symbols were not being resolved because machine__get_kernel_start was only setting machine->kernel_start when the just successfully loaded kernel symtab had its map->start set to !0, when it was left at (1ULL << 63) assuming a partitioning of the address space for user/kernel, which is not the case in S/390 nor in Sparc. So just check if map__load() was successfull and set machine->kernel_start to zero, fixing kernel symbol resolution on S/390. Test performed by Thomas: ---- I like this patch. I have done a new build and removed all my debug output to start from scratch. Without your patch I get this: # Samples: 4 of event 'cpu-clock' # Event count (approx.): 1000000 # # Children Self Command Shared Object Symbol # ........ ........ ....... ................ ........................ 75.00% 0.00% true [unknown] [k] 0x00000000004bedda | ---0x4bedda | |--50.00%--0x42693a | | | --25.00%--0x2a72e0 | 0x2af0ca | 0x3d1003fe4c0 | --25.00%--0x4272bc 0x26fa84 and with your patch (I just rebuilt the perf tool, nothing else and used the same perf.data file as input): # Samples: 4 of event 'cpu-clock' # Event count (approx.): 1000000 # # Children Self Command Shared Object Symbol # ........ ........ ....... .......................... .................................. 75.00% 0.00% true [kernel.vmlinux] [k] pgm_check_handler | ---pgm_check_handler do_dat_exception handle_mm_fault __handle_mm_fault filemap_map_pages | |--25.00%--rcu_read_lock_held | rcu_lockdep_current_cpu_online | 0x3d1003ff4c0 | --25.00%--lock_release Looks good to me.... ---- Reported-and-Tested-by: Thomas-Mich Richter <tmricht@linux.vnet.ibm.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Wang Nan <wangnan0@huawei.com> Cc: Zvonko Kosic <zvonko.kosic@de.ibm.com> Link: http://lkml.kernel.org/n/tip-dk0n1uzmbe0tbthrpfqlx6bz@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-07-12 03:21:40 +08:00
if (!err)
machine->kernel_start = map->start;
}
return err;
}
struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
{
return dsos__findnew(&machine->dsos, filename);
}
char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
{
struct machine *machine = vmachine;
struct map *map;
struct symbol *sym = map_groups__find_symbol(&machine->kmaps, MAP__FUNCTION, *addrp, &map);
if (sym == NULL)
return NULL;
*modp = __map__is_kmodule(map) ? (char *)map->dso->short_name : NULL;
*addrp = map->unmap_ip(map, sym->start);
return sym->name;
}