Assigning "attr" to "attr" does not have any effect, but was caught by
Coverity, so let's remove this.
Reported-by: coverity (CID 1354720)
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Tested-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Fixes: 1b76c13e4b ("bpf tools: Introduce 'bpf' library and add bpf feature check")
Link: http://lkml.kernel.org/r/1461551694-5512-2-git-send-email-f.fainelli@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Skylake processor supports a new set of RAPL registers for controlling
entire SoC instead of just CPU package. This is useful for thermal
and power control when source of power/thermal is not just CPU/GPU.
This change adds a new platform domain (AKA PSys) to the current
power capping Intel RAPL driver.
PSys also supports PL1 (long term) and PL2 (short term) control like
package domain. This also follows same MSRs for energy and time
units as package domain.
Unlike package domain, PSys support requires more than just processor
level implementation. The other parts in the system need additional
implementation, which OEMs needs to support. So not all Skylake
systems will support PSys.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: bp@alien8.de
Cc: hpa@zytor.com
Cc: jacob.jun.pan@linux.intel.com
Cc: rjw@rjwysocki.net
Link: http://lkml.kernel.org/r/1460930581-29748-3-git-send-email-srinivas.pandruvada@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This patch fixes a bug which was introduced by:
b16a5b52eb ("perf/x86: Add option to disable reading branch flags/cycles")
In this patch, lbr_sel_mask is used to mask the lbr_select. But LBR_SEL_MASK
doesn't include the bit for LBR_CALL_STACK. So LBR call stack will never be
set in lbr_select.
This patch corrects the LBR_SEL_MASK by including all valid bits in
LBR_SELECT. Also, the LBR_CALL_STACK bit is different as other bit in
LBR_SELECT. It does not operate in suppress mode, so it needs to be
specially handled in intel_pmu_setup_hw_lbr_filter.
Signed-off-by: Kan Liang <kan.liang@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Link: http://lkml.kernel.org/r/1461231010-4399-1-git-send-email-kan.liang@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Some versions of Intel PT do not support tracing across VMXON, more
specifically, VMXON will clear TraceEn control bit and any attempt to
set it before VMXOFF will throw a #GP, which in the current state of
things will crash the kernel. Namely:
$ perf record -e intel_pt// kvm -nographic
on such a machine will kill it.
To avoid this, notify the intel_pt driver before VMXON and after
VMXOFF so that it knows when not to enable itself.
Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Gleb Natapov <gleb@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: hpa@zytor.com
Link: http://lkml.kernel.org/r/87oa9dwrfk.fsf@ashishki-desk.ger.corp.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Jann reported that the ptrace_may_access() check in
find_lively_task_by_vpid() is racy against exec().
Specifically:
perf_event_open() execve()
ptrace_may_access()
commit_creds()
... if (get_dumpable() != SUID_DUMP_USER)
perf_event_exit_task();
perf_install_in_context()
would result in installing a counter across the creds boundary.
Fix this by wrapping lots of perf_event_open() in cred_guard_mutex.
This should be fine as perf_event_exit_task() is already called with
cred_guard_mutex held, so all perf locks already nest inside it.
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The entry for PERF_COUNT_HW_REF_CPU_CYCLES is not used on AMD, but is
referenced by filter_events() which expects undefined events to have a
value of 0.
Found via KASAN:
UBSAN: Undefined behaviour in arch/x86/events/amd/core.c:132:30
index 9 is out of range for type 'u64 [9]'
UBSAN: Undefined behaviour in arch/x86/events/amd/core.c:132:9
load of address ffffffff81c021c8 with insufficient space for an object of type 'const u64'
Signed-off-by: Adam Borowski <kilobyte@angband.pl>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Link: http://lkml.kernel.org/r/1461749731-30979-1-git-send-email-kilobyte@angband.pl
Signed-off-by: Ingo Molnar <mingo@kernel.org>
User visible:
- perf trace --pf maj/min/all works with --call-graph: (Arnaldo Carvalho de Melo)
Tracing write syscalls and major page faults with callchains while starting
firefox, limiting the stack to 5 frames:
# perf trace -e write --pf maj --max-stack 5 firefox
589.549 ( 0.014 ms): firefox/15377 write(fd: 4, buf: 0x7fff80acc898, count: 151) = 151
[0xfaed] (/usr/lib64/libpthread-2.22.so)
fire_glxtest_process+0x5c (/usr/lib64/firefox/libxul.so)
InstallGdkErrorHandler+0x41 (/usr/lib64/firefox/libxul.so)
XREMain::XRE_mainInit+0x12c (/usr/lib64/firefox/libxul.so)
XREMain::XRE_main+0x1e4 (/usr/lib64/firefox/libxul.so)
760.704 ( 0.000 ms): firefox/15332 majfault [gtk_tree_view_accessible_get_type+0x0] => /usr/lib64/libgtk-3.so.0.1800.9@0xa0850 (x.)
gtk_tree_view_accessible_get_type+0x0 (/usr/lib64/libgtk-3.so.0.1800.9)
gtk_tree_view_class_intern_init+0x1a54 (/usr/lib64/libgtk-3.so.0.1800.9)
g_type_class_ref+0x6dd (/usr/lib64/libgobject-2.0.so.0.4600.2)
[0x115378] (/usr/lib64/libgnutls.so.30.6.3)
This automagically selects "--call-graph dwarf", use "--call-graph fp" on systems
where -fno-omit-frame-pointer was used to built the components of interest, to
incur in less overhead, or tune "--call-graph dwarf" appropriately, see 'perf record --help'.
- Allow /proc/sys/kernel/perf_event_max_stack, that defaults to the old hard coded value
of PERF_MAX_STACK_DEPTH (127), useful for huge callstacks for things like Groovy, Ruby, etc,
and also to reduce overhead by limiting it to a smaller value, upcoming work will allow
this to be done per-event (Arnaldo Carvalho de Melo)
- Make 'perf trace --min-stack' be honoured by --pf and --event (Arnaldo Carvalho de Melo)
- Make 'perf evlist -v' decode perf_event_attr->branch_sample_type (Arnaldo Carvalho de Melo)
# perf record --call lbr usleep 1
# perf evlist -v
cycles:ppp: ... sample_type: IP|TID|TIME|CALLCHAIN|PERIOD|BRANCH_STACK, ...
branch_sample_type: USER|CALL_STACK|NO_FLAGS|NO_CYCLES
#
- Clear dummy entry accumulated period, fixing such 'perf top/report' output
as: (Kan Liang)
4769.98% 0.01% 0.00% 0.01% tchain_edit [kernel] [k] update_fast_timekeeper
- System calls with pid_t arguments gets them augmented with the COMM event
more thoroughly:
# trace -e perf_event_open perf stat -e cycles -p 15608
6.876 ( 0.014 ms): perf_event_open(attr_uptr: 0x2ae20d8, pid: 15608 (hexchat), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 3
6.882 ( 0.005 ms): perf_event_open(attr_uptr: 0x2ae20d8, pid: 15639 (gmain), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 4
6.889 ( 0.005 ms): perf_event_open(attr_uptr: 0x2ae20d8, pid: 15640 (gdbus), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 5
^^^^^^^^^^^^^^^^^^
^C
- Fix offline module name mismatch issue in 'perf probe' (Ravi Bangoria)
- Fix module probe issue if no dwarf support in (Ravi Bangoria)
Assorted fixes:
- Fix off-by-one in write_buildid() (Andrey Ryabinin)
- Fix segfault when printing callchains in 'perf script' (Chris Phlipot)
- Replace assignment with comparison on assert check in 'perf test' entry (Colin Ian King)
- Fix off-by-one comparison in intel-pt code (Colin Ian King)
- Close target file on error path in 'perf probe' (Masami Hiramatsu)
- Set default kprobe group name if not given in 'perf probe' (Masami Hiramatsu)
- Avoid partial perf_event_header reads (Wang Nan)
Infrastructure:
- Update x86's syscall_64.tbl copy, adding preadv2 & pwritev2 (Arnaldo Carvalho de Melo)
- Make the x86 clean quiet wrt syscall table removal (Jiri Olsa)
Cleanups:
- Simplify wrapper for LOCK_PI in 'perf bench futex' (Davidlohr Bueso)
- Remove duplicate const qualifier (Eric Engestrom)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2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=MzXA
-----END PGP SIGNATURE-----
Merge tag 'perf-core-for-mingo-20160427' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core
Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:
User visible changes:
- perf trace --pf maj/min/all works with --call-graph: (Arnaldo Carvalho de Melo)
Tracing write syscalls and major page faults with callchains while starting
firefox, limiting the stack to 5 frames:
# perf trace -e write --pf maj --max-stack 5 firefox
589.549 ( 0.014 ms): firefox/15377 write(fd: 4, buf: 0x7fff80acc898, count: 151) = 151
[0xfaed] (/usr/lib64/libpthread-2.22.so)
fire_glxtest_process+0x5c (/usr/lib64/firefox/libxul.so)
InstallGdkErrorHandler+0x41 (/usr/lib64/firefox/libxul.so)
XREMain::XRE_mainInit+0x12c (/usr/lib64/firefox/libxul.so)
XREMain::XRE_main+0x1e4 (/usr/lib64/firefox/libxul.so)
760.704 ( 0.000 ms): firefox/15332 majfault [gtk_tree_view_accessible_get_type+0x0] => /usr/lib64/libgtk-3.so.0.1800.9@0xa0850 (x.)
gtk_tree_view_accessible_get_type+0x0 (/usr/lib64/libgtk-3.so.0.1800.9)
gtk_tree_view_class_intern_init+0x1a54 (/usr/lib64/libgtk-3.so.0.1800.9)
g_type_class_ref+0x6dd (/usr/lib64/libgobject-2.0.so.0.4600.2)
[0x115378] (/usr/lib64/libgnutls.so.30.6.3)
This automagically selects "--call-graph dwarf", use "--call-graph fp" on systems
where -fno-omit-frame-pointer was used to built the components of interest, to
incur in less overhead, or tune "--call-graph dwarf" appropriately, see 'perf record --help'.
- Allow /proc/sys/kernel/perf_event_max_stack, that defaults to the old hard coded value
of PERF_MAX_STACK_DEPTH (127), useful for huge callstacks for things like Groovy, Ruby, etc,
and also to reduce overhead by limiting it to a smaller value, upcoming work will allow
this to be done per-event (Arnaldo Carvalho de Melo)
- Make 'perf trace --min-stack' be honoured by --pf and --event (Arnaldo Carvalho de Melo)
- Make 'perf evlist -v' decode perf_event_attr->branch_sample_type (Arnaldo Carvalho de Melo)
# perf record --call lbr usleep 1
# perf evlist -v
cycles:ppp: ... sample_type: IP|TID|TIME|CALLCHAIN|PERIOD|BRANCH_STACK, ...
branch_sample_type: USER|CALL_STACK|NO_FLAGS|NO_CYCLES
#
- Clear dummy entry accumulated period, fixing such 'perf top/report' output
as: (Kan Liang)
4769.98% 0.01% 0.00% 0.01% tchain_edit [kernel] [k] update_fast_timekeeper
- System calls with pid_t arguments gets them augmented with the COMM event
more thoroughly:
# trace -e perf_event_open perf stat -e cycles -p 15608
6.876 ( 0.014 ms): perf_event_open(attr_uptr: 0x2ae20d8, pid: 15608 (hexchat), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 3
6.882 ( 0.005 ms): perf_event_open(attr_uptr: 0x2ae20d8, pid: 15639 (gmain), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 4
6.889 ( 0.005 ms): perf_event_open(attr_uptr: 0x2ae20d8, pid: 15640 (gdbus), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 5
^^^^^^^^^^^^^^^^^^
^C
- Fix offline module name mismatch issue in 'perf probe' (Ravi Bangoria)
- Fix module probe issue if no dwarf support in (Ravi Bangoria)
Assorted fixes:
- Fix off-by-one in write_buildid() (Andrey Ryabinin)
- Fix segfault when printing callchains in 'perf script' (Chris Phlipot)
- Replace assignment with comparison on assert check in 'perf test' entry (Colin Ian King)
- Fix off-by-one comparison in intel-pt code (Colin Ian King)
- Close target file on error path in 'perf probe' (Masami Hiramatsu)
- Set default kprobe group name if not given in 'perf probe' (Masami Hiramatsu)
- Avoid partial perf_event_header reads (Wang Nan)
Infrastructure changes:
- Update x86's syscall_64.tbl copy, adding preadv2 & pwritev2 (Arnaldo Carvalho de Melo)
- Make the x86 clean quiet wrt syscall table removal (Jiri Olsa)
Cleanups:
- Simplify wrapper for LOCK_PI in 'perf bench futex' (Davidlohr Bueso)
- Remove duplicate const qualifier (Eric Engestrom)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
There is an upper limit to what tooling considers a valid callchain,
and it was tied to the hardcoded value in the kernel,
PERF_MAX_STACK_DEPTH (127), now that this can be tuned via a sysctl,
make it read it and use that as the upper limit, falling back to
PERF_MAX_STACK_DEPTH for kernels where this sysctl isn't present.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-yjqsd30nnkogvj5oyx9ghir9@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The default remains 127, which is good for most cases, and not even hit
most of the time, but then for some cases, as reported by Brendan, 1024+
deep frames are appearing on the radar for things like groovy, ruby.
And in some workloads putting a _lower_ cap on this may make sense. One
that is per event still needs to be put in place tho.
The new file is:
# cat /proc/sys/kernel/perf_event_max_stack
127
Chaging it:
# echo 256 > /proc/sys/kernel/perf_event_max_stack
# cat /proc/sys/kernel/perf_event_max_stack
256
But as soon as there is some event using callchains we get:
# echo 512 > /proc/sys/kernel/perf_event_max_stack
-bash: echo: write error: Device or resource busy
#
Because we only allocate the callchain percpu data structures when there
is a user, which allows for changing the max easily, its just a matter
of having no callchain users at that point.
Reported-and-Tested-by: Brendan Gregg <brendan.d.gregg@gmail.com>
Reviewed-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: David Ahern <dsahern@gmail.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: Zefan Li <lizefan@huawei.com>
Link: http://lkml.kernel.org/r/20160426002928.GB16708@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Propagate the error instead.
Cc: David Ahern <dsahern@gmail.com>
Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>
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-z6erjg35d1gekevwujoa0223@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Introduced in commit 4babf2c5ef ("x86: wire up preadv2 and pwritev2").
This will make 'perf trace' aware of them.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-vojoylgce2cetsy36446s5ny@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Perf is not able to register probe in kernel module when dwarf supprt
is not there(and so it goes for symtab). Perf passes full path of
module where only module name is required which is causing the problem.
This patch fixes this issue.
Before applying patch:
$ dpkg -s libdw-dev
dpkg-query: package 'libdw-dev' is not installed and no information is...
$ sudo ./perf probe -m /linux/samples/kprobes/kprobe_example.ko kprobe_init
Added new event:
probe:kprobe_init (on kprobe_init in /linux/samples/kprobes/kprobe_example.ko)
You can now use it in all perf tools, such as:
perf record -e probe:kprobe_init -aR sleep 1
$ sudo cat /sys/kernel/debug/tracing/kprobe_events
p:probe/kprobe_init /linux/samples/kprobes/kprobe_example.ko:kprobe_init
$ sudo ./perf record -a -e probe:kprobe_init
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.105 MB perf.data ]
$ sudo ./perf script # No output here
After applying patch:
$ sudo ./perf probe -m /linux/samples/kprobes/kprobe_example.ko kprobe_init
Added new event:
probe:kprobe_init (on kprobe_init in kprobe_example)
You can now use it in all perf tools, such as:
perf record -e probe:kprobe_init -aR sleep 1
$ sudo cat /sys/kernel/debug/tracing/kprobe_events
p:probe/kprobe_init kprobe_example:kprobe_init
$ sudo ./perf record -a -e probe:kprobe_init
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.105 MB perf.data (2 samples) ]
$ sudo ./perf script
insmod 13990 [002] 5961.216833: probe:kprobe_init: ...
insmod 13995 [002] 5962.889384: probe:kprobe_init: ...
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.vnet.ibm.com>
Acked-by: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1461680741-12517-1-git-send-email-ravi.bangoria@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Perf can add a probe on kernel module which has not been loaded yet.
The current implementation finds the module name from path. But if the
filename is different from the actual module name then perf fails to
register a probe while loading module because of mismatch in the names.
For example, samples/kobject/kobject-example.ko is loaded as
kobject_example.
Before applying patch:
$ sudo ./perf probe -m /linux/samples/kobject/kobject-example.ko foo_show
Added new event:
probe:foo_show (on foo_show in kobject-example)
You can now use it in all perf tools, such as:
perf record -e probe:foo_show -aR sleep 1
$ cat /sys/kernel/debug/tracing/kprobe_events
p:probe/foo_show kobject-example:foo_show
$ insmod kobject-example.ko
$ lsmod
Module Size Used by
kobject_example 16384 0
Generate read to /sys/kernel/kobject_example/foo while recording data
with below command
$ sudo ./perf record -e probe:foo_show -a
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.093 MB perf.data ]
$./perf report --stdio -F overhead,comm,dso,sym
Error:
The perf.data.old file has no samples!
After applying patch:
$ sudo ./perf probe -m /linux/samples/kobject/kobject-example.ko foo_show
Added new event:
probe:foo_show (on foo_show in kobject_example)
You can now use it in all perf tools, such as:
perf record -e probe:foo_show -aR sleep 1
$ sudo cat /sys/kernel/debug/tracing/kprobe_events
p:probe/foo_show kobject_example:foo_show
$ insmod kobject-example.ko
$ lsmod
Module Size Used by
kobject_example 16384 0
Generate read to /sys/kernel/kobject_example/foo while recording data
with below command
$ sudo ./perf record -e probe:foo_show -a
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.097 MB perf.data (8 samples) ]
$ sudo ./perf report --stdio -F overhead,comm,dso,sym
...
# Samples: 8 of event 'probe:foo_show'
# Event count (approx.): 8
#
# Overhead Command Shared Object Symbol
# ........ ....... ................. ............
#
100.00% cat [kobject_example] [k] foo_show
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.vnet.ibm.com>
Acked-by: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1461680741-12517-2-git-send-email-ravi.bangoria@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
We get notifications for threads that gets created while we're tracing,
but for preexisting threads we may end not having synthesized them, like
when tracing a 'perf trace' session that will use '--pid' to trace some
other thread.
And besides we should probably stop synthesizing those records and
instead read thread information in a lazy way, i.e. just when we need,
like done in this patch:
Now the 'pid_t' argument in 'perf_event_open' gets translated to a COMM:
# perf trace -e perf_event_open perf stat -e cycles -p 31601
0.027 ( 0.027 ms): perf/23393 perf_event_open(attr_uptr: 0x2fdd0d8, pid: 31601 (abrt-dump-journ), cpu: -1, group_fd: -1, flags: FD_CLOEXEC)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
= 3
^C
And in other syscalls containing pid_t without thread->comm_set at the
time of the formatting.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-ioeps6dlwst17d6oozc9shtk@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Will be used for lazy comm loading in 'perf trace'.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-7ogbkuoka1y2qsmcckqxvl5m@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
To read things like /proc/self/comm.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-ztpkbmseidt0hq2psr46o0h9@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Leave it alone so that it ends up assigned to SCA_PID via its type,
'pid_t', that will look up the pid on the machine thread rb_tree and
possibly find its COMM.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-r7dujgmhtxxfajuunpt1bkuo@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
To reduce the size of builtin-trace.c.
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-8r3gmymyn3r0ynt4yuzspp9g@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Set kprobe group name as "probe" if it is not given.
Signed-off-by: Masami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Ananth N Mavinakayanahalli <ananth@linux.vnet.ibm.com>
Cc: Hemant Kumar <hemant@linux.vnet.ibm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20160426090413.11891.95640.stgit@devbox
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Since other methods return 0 if succeeded (or filedesc), let
probe_file__add_event() return 0 instead of the length of written bytes.
Signed-off-by: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Ananth N Mavinakayanahalli <ananth@linux.vnet.ibm.com>
Cc: Hemant Kumar <hemant@linux.vnet.ibm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20160426090303.11891.18232.stgit@devbox
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
As a utility function, add lsdir() which reads given directory and store
entry name into a strlist. lsdir accepts a filter function so that user
can filter out unneeded entries.
Signed-off-by: Masami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Ananth N Mavinakayanahalli <ananth@linux.vnet.ibm.com>
Cc: Hemant Kumar <hemant@linux.vnet.ibm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20160426090242.11891.79014.stgit@devbox
[ Do not use the 'dirname' it is used in some distros ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Fix a bug to close target elf file in get_text_start_address().
Signed-off-by: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20160426064737.1443.44093.stgit@devbox
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Don't read broken data after 'head' pointer.
Following commits will feed perf_evlist__mmap_read() with some 'head'
pointers not maintained by kernel. If 'head' pointer breaks an event, we
should avoid reading from the broken event. This can happen in backward
ring buffer.
For example:
old head
| |
V V
+---+------+----------+----+-----+--+
|..E|D....D|C........C|B..B|A....|E.|
+---+------+----------+----+-----+--+
'old' pointer points to the beginning of 'A' and trying read from it,
but 'A' has been overwritten. In this case, don't try to read from 'A',
simply return NULL.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1461637738-62722-2-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The check for the maximum code is off-by-one; the current comparison of
a code that is INTEL_PT_ERR_MAX will cause the strlcpy to perform an out
of bounds array access on the intel_pt_err_msgs array.
Fix this with a >= comparison.
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1461524203-10224-1-git-send-email-colin.king@canonical.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Given that the 'val' parameter is ignored for FUTEX_LOCK_PI, get rid of
the bogus deadlock detection flag in the wrapper code and avoid the
extra argument, making it resemble its unlock counterpart. And if
nothing else, we already only pass 0 anyway.
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Cc: Davidlohr Bueso <dbueso@suse.de>
Link: http://lkml.kernel.org/r/1461208447-29328-1-git-send-email-dave@stgolabs.net
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The current assert check is checking an assignment, which will always be
true. Instead, the assert should be checking if scale is equal to 0.122
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1461419154-16918-1-git-send-email-colin.king@canonical.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Fix perf_clean target to follow the same logic as perf target.
Fixes the following make invokation:
$ cd <kernelsrc> && make tools/perf_clean
Reported-by: TJ <linux@iam.tj>
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>
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=116411
Link: http://lkml.kernel.org/r/1461615438-27894-2-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
To check deeply nested page fault callchains.
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-wuji34xx003kr88nmqt6jkgf@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-shj0fazntmskhjild5i6x73l@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This fixes a bug caused by an unitialized callchain cursor. The crash
frist appeared in:
6f736735e3 ("perf evsel: Require that callchains be resolved before
calling fprintf_{sym,callchain}")
The callchain cursor is a struct that contains pointers, that when
uninitialized will cause unpredictable behavior (usually a crash)
when trying to append to the callchain.
The existing implementation has the following issues:
1. The callchain cursor used is not initialized, resulting in
unpredictable behavior when used.
2. The cursor is declared on the stack. Even if it is properly initalized,
the implmentation will leak memory when the function returns,
since all the references to the callchain_nodes allocated by
callchain_cursor_append will be lost when the cursor goes out of
scope.
3. Storing the cursor on the stack is inefficient. Even if memory is
properly freed when it goes out of scope, a performance penalty
will be incurred due to reallocation of callchain nodes.
callchain_cursor_append is designed to avoid these reallocations
when an existing cursor is reused.
This patch fixes the crash by replacing cursor_callchain with a reference
to the global callchain_cursor which also resolves all 3 issues mentioned
above.
How to reproduce the crash:
$ perf record --call-graph=dwarf stress -t 1 -c 1
$ perf script > /dev/null
Segfault
Signed-off-by: Chris Phlipot <cphlipot0@gmail.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Fixes: 6f736735e3 ("perf evsel: Require that callchains be resolved before calling fprintf_{sym,callchain}")
Link: http://lkml.kernel.org/r/1461119531-2529-1-git-send-email-cphlipot0@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Prep work for next patches, where we'll need access to the created
evsels, to possibly configure callchains.
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-2pcgsgnkgellhlcao4aub8tu@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
write_buildid() increments 'name_len' with intention to take into
account trailing zero byte. However, 'name_len' was already incremented
in machine__write_buildid_table() before. So this leads to
out-of-bounds read in do_write():
$ ./perf record sleep 0
[ perf record: Woken up 1 times to write data ]
=================================================================
==15899==ERROR: AddressSanitizer: global-buffer-overflow on address 0x00000099fc92 at pc 0x7f1aa9c7eab5 bp 0x7fff940f84d0 sp 0x7fff940f7c78
READ of size 19 at 0x00000099fc92 thread T0
#0 0x7f1aa9c7eab4 (/usr/lib/gcc/x86_64-pc-linux-gnu/5.3.0/libasan.so.2+0x44ab4)
#1 0x649c5b in do_write util/header.c:67
#2 0x649c5b in write_padded util/header.c:82
#3 0x57e8bc in write_buildid util/build-id.c:239
#4 0x57e8bc in machine__write_buildid_table util/build-id.c:278
...
0x00000099fc92 is located 0 bytes to the right of global variable '*.LC99' defined in 'util/symbol.c' (0x99fc80) of size 18
'*.LC99' is ascii string '[kernel.kallsyms]'
...
Shadow bytes around the buggy address:
0x00008012bf80: f9 f9 f9 f9 00 00 00 00 00 00 03 f9 f9 f9 f9 f9
=>0x00008012bf90: 00 00[02]f9 f9 f9 f9 f9 00 00 00 00 00 05 f9 f9
0x00008012bfa0: f9 f9 f9 f9 00 03 f9 f9 f9 f9 f9 f9 00 00 00 00
Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1461053847-5633-1-git-send-email-aryabinin@virtuozzo.com
[ Remove the off-by one at the origin, to keep len(s) == strlen(s) assumption ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:
Build fixes:
- Fix 'perf trace' build when DWARF unwind isn't available (Arnaldo Carvalho de Melo)
- Remove x86 references from arch-neutral Build, fixing it in !x86 arches,
reported as breaking the build for powerpc64le in linux-next (Arnaldo Carvalho de Melo)
Infrastructure changes:
- Do memset() variable 'st' using the correct size in the jit code (Colin Ian King)
- Fix postgresql ubuntu 'perf script' install instructions (Chris Phlipot)
- Use callchain_param more thoroughly when checking how callchains were
configured, eventually will be the only way to look for callchain parameters
(Arnaldo Carvalho de Melo)
- Fix some issues in the 'perf test kallsyms' entry (Arnaldo Carvalho de Melo)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
With the array aligned as per events/intel/core.c it was fairly
obvious we missed one, add it in.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Re-order the model array to match the order in events/intel/core.c,
to easier spot gaps and such.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Add Skylake client support for RAPL domains. In addition to RAPL domains
in Broadwell clients, it has support for platform domain (aka PSys). The
PSys domain controls the entire SoC instead of just a CPU package. Unlike
package domain, PSys support requires more than just processor level
implementation. The other parts in the system need additional HW level
signaling, which OEMs need to support. When not supported, the energy
counter register in PSys domain returns 0.
Also corrected error in comment for GPU counter, which previously was
DRAM counter.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com
[ Cnverted to model_match stuff. ]
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: bp@alien8.de
Cc: hpa@zytor.com
Cc: jacob.jun.pan@linux.intel.com
Cc: rjw@rjwysocki.net
Link: http://lkml.kernel.org/r/1460930581-29748-2-git-send-email-srinivas.pandruvada@linux.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This patch introduces 'write_backward' bit to perf_event_attr, which
controls the direction of a ring buffer. After set, the corresponding
ring buffer is written from end to beginning. This feature is design to
support reading from overwritable ring buffer.
Ring buffer can be created by mapping a perf event fd. Kernel puts event
records into ring buffer, user tooling like perf fetch them from
address returned by mmap(). To prevent racing between kernel and tooling,
they communicate to each other through 'head' and 'tail' pointers.
Kernel maintains 'head' pointer, points it to the next free area (tail
of the last record). Tooling maintains 'tail' pointer, points it to the
tail of last consumed record (record has already been fetched). Kernel
determines the available space in a ring buffer using these two
pointers to avoid overwrite unfetched records.
By mapping without 'PROT_WRITE', an overwritable ring buffer is created.
Different from normal ring buffer, tooling is unable to maintain 'tail'
pointer because writing is forbidden. Therefore, for this type of ring
buffers, kernel overwrite old records unconditionally, works like flight
recorder. This feature would be useful if reading from overwritable ring
buffer were as easy as reading from normal ring buffer. However,
there's an obscure problem.
The following figure demonstrates a full overwritable ring buffer. In
this figure, the 'head' pointer points to the end of last record, and a
long record 'E' is pending. For a normal ring buffer, a 'tail' pointer
would have pointed to position (X), so kernel knows there's no more
space in the ring buffer. However, for an overwritable ring buffer,
kernel ignore the 'tail' pointer.
(X) head
. |
. V
+------+-------+----------+------+---+
|A....A|B.....B|C........C|D....D| |
+------+-------+----------+------+---+
Record 'A' is overwritten by event 'E':
head
|
V
+--+---+-------+----------+------+---+
|.E|..A|B.....B|C........C|D....D|E..|
+--+---+-------+----------+------+---+
Now tooling decides to read from this ring buffer. However, none of these
two natural positions, 'head' and the start of this ring buffer, are
pointing to the head of a record. Even the full ring buffer can be
accessed by tooling, it is unable to find a position to start decoding.
The first attempt tries to solve this problem AFAIK can be found from
[1]. It makes kernel to maintain 'tail' pointer: updates it when ring
buffer is half full. However, this approach introduces overhead to
fast path. Test result shows a 1% overhead [2]. In addition, this method
utilizes no more tham 50% records.
Another attempt can be found from [3], which allows putting the size of
an event at the end of each record. This approach allows tooling to find
records in a backward manner from 'head' pointer by reading size of a
record from its tail. However, because of alignment requirement, it
needs 8 bytes to record the size of a record, which is a huge waste. Its
performance is also not good, because more data need to be written.
This approach also introduces some extra branch instructions to fast
path.
'write_backward' is a better solution to this problem.
Following figure demonstrates the state of the overwritable ring buffer
when 'write_backward' is set before overwriting:
head
|
V
+---+------+----------+-------+------+
| |D....D|C........C|B.....B|A....A|
+---+------+----------+-------+------+
and after overwriting:
head
|
V
+---+------+----------+-------+---+--+
|..E|D....D|C........C|B.....B|A..|E.|
+---+------+----------+-------+---+--+
In each situation, 'head' points to the beginning of the newest record.
From this record, tooling can iterate over the full ring buffer and fetch
records one by one.
The only limitation that needs to be considered is back-to-back reading.
Due to the non-deterministic of user programs, it is impossible to ensure
the ring buffer keeps stable during reading. Consider an extreme situation:
tooling is scheduled out after reading record 'D', then a burst of events
come, eat up the whole ring buffer (one or multiple rounds). When the
tooling process comes back, reading after 'D' is incorrect now.
To prevent this problem, we need to find a way to ensure the ring buffer
is stable during reading. ioctl(PERF_EVENT_IOC_PAUSE_OUTPUT) is
suggested because its overhead is lower than
ioctl(PERF_EVENT_IOC_ENABLE).
By carefully verifying 'header' pointer, reader can avoid pausing the
ring-buffer. For example:
/* A union of all possible events */
union perf_event event;
p = head = perf_mmap__read_head();
while (true) {
/* copy header of next event */
fetch(&event.header, p, sizeof(event.header));
/* read 'head' pointer */
head = perf_mmap__read_head();
/* check overwritten: is the header good? */
if (!verify(sizeof(event.header), p, head))
break;
/* copy the whole event */
fetch(&event, p, event.header.size);
/* read 'head' pointer again */
head = perf_mmap__read_head();
/* is the whole event good? */
if (!verify(event.header.size, p, head))
break;
p += event.header.size;
}
However, the overhead is high because:
a) In-place decoding is not safe.
Copying-verifying-decoding is required.
b) Fetching 'head' pointer requires additional synchronization.
(From Alexei Starovoitov:
Even when this trick works, pause is needed for more than stability of
reading. When we collect the events into overwrite buffer we're waiting
for some other trigger (like all cpu utilization spike or just one cpu
running and all others are idle) and when it happens the buffer has
valuable info from the past. At this point new events are no longer
interesting and buffer should be paused, events read and unpaused until
next trigger comes.)
This patch utilizes event's default overflow_handler introduced
previously. perf_event_output_backward() is created as the default
overflow handler for backward ring buffers. To avoid extra overhead to
fast path, original perf_event_output() becomes __perf_event_output()
and marked '__always_inline'. In theory, there's no extra overhead
introduced to fast path.
Performance testing:
Calling 3000000 times of 'close(-1)', use gettimeofday() to check
duration. Use 'perf record -o /dev/null -e raw_syscalls:*' to capture
system calls. In ns.
Testing environment:
CPU : Intel(R) Core(TM) i7-4790 CPU @ 3.60GHz
Kernel : v4.5.0
MEAN STDVAR
BASE 800214.950 2853.083
PRE1 2253846.700 9997.014
PRE2 2257495.540 8516.293
POST 2250896.100 8933.921
Where 'BASE' is pure performance without capturing. 'PRE1' is test
result of pure 'v4.5.0' kernel. 'PRE2' is test result before this
patch. 'POST' is test result after this patch. See [4] for the detailed
experimental setup.
Considering the stdvar, this patch doesn't introduce performance
overhead to the fast path.
[1] http://lkml.iu.edu/hypermail/linux/kernel/1304.1/04584.html
[2] http://lkml.iu.edu/hypermail/linux/kernel/1307.1/00535.html
[3] http://lkml.iu.edu/hypermail/linux/kernel/1512.0/01265.html
[4] http://lkml.kernel.org/g/56F89DCD.1040202@huawei.com
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Cc: <acme@kernel.org>
Cc: <pi3orama@163.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: Zefan Li <lizefan@huawei.com>
Link: http://lkml.kernel.org/r/1459865478-53413-1-git-send-email-wangnan0@huawei.com
[ Fixed the changelog some more. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
LBR filtering is also supported on the Silvermont and Airmont
microarchitectures. The layout of MSR_LBR_SELECT is the same as Nehalem.
Signed-off-by: Kan Liang <kan.liang@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Link: http://lkml.kernel.org/r/1460706825-46163-1-git-send-email-kan.liang@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Add perf core PMU support for Intel Goldmont CPU cores:
- The init code is based on Silvermont.
- There is a new cache event list, based on the Silvermont cache event list.
- Goldmont has 32 LBR entries. It also uses new LBRv6 format, which
report the cycle information using upper 16-bit of the LBR_TO.
- It's recommended to use CPU_CLK_UNHALTED.CORE_P + NPEBS for precise cycles.
For details, please refer to the latest SDM058:
http://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-software-developer-vol-3b-part-2-manual.pdf
Signed-off-by: Kan Liang <kan.liang@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Link: http://lkml.kernel.org/r/1460706167-45320-1-git-send-email-kan.liang@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Markus reported that 0 should also disable the throttling we per
Documentation/sysctl/kernel.txt.
Reported-by: Markus Trippelsdorf <markus@trippelsdorf.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Fixes: 91a612eea9 ("perf/core: Fix dynamic interrupt throttle")
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Everything the same as base Skylake, just a new model number.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Link: http://lkml.kernel.org/r/1460751933-2264-1-git-send-email-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
One of the branches leading to an error had no debug message emitted,
fix it, the new lines are:
# perf test -v kallsyms
<SNIP>
0xffffffff81001000: diff name v: xen_hypercall_set_trap_table k: hypercall_page
0xffffffff810691f0: diff name v: try_to_free_pud_page k: try_to_free_pmd_page
<SNIP>
0xffffffff8150bb20: diff name v: wakeup_expire_count_show.part.5 k: wakeup_active_count_show.part.7
0xffffffff816bc7f0: diff name v: phys_switch_id_show.part.11 k: phys_port_name_show.part.12
0xffffffff817bbb90: diff name v: __do_softirq k: __softirqentry_text_start
<SNIP>
This in turn exercises another bug, still under investigation, because those
aliases _are_ in kallsyms, with the same name...
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>
Fixes: ab414dcda8 ("perf test: Fixup aliases checking in the 'vmlinux matches kallsyms' test")
Link: http://lkml.kernel.org/n/tip-5fhea7a54a54gsmagu9obpr4@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Florian Fainelli