The move of blocked readers to the ring buffer left out the
init of the wait queue that is used. Tests missed this due to running
stress tests against the buffers, which didn't allow for any
readers to end up waiting. Running a simple read and wait triggered
a bug.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
As we've added __init annotation to field-defining functions, we should
add __refdata annotation to event_call variables, which reference those
functions.
Link: http://lkml.kernel.org/r/51343C1F.2050502@huawei.com
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The new multi-buffers added a descriptor that kept track of module
events, and the directories they use, with struct ftace_module_file_ops.
This is used to add a ref count to keep modules from unloading while
their files are being accessed.
As the descriptor is only needed when CONFIG_MODULES is enabled, it
is only declared when the config is enabled. But that struct is
dereferenced in a few areas outside the #ifdef CONFIG_MODULES.
By adding some helper routines and moving code around a little,
events can be compiled again without modules.
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
With the conversion of the data array to per cpu, sparse now complains
about the use of per_cpu_ptr() on the variable. But The variable is
allocated with alloc_percpu() and is fine to use. But since the structure
that contains the data variable does not annotate it as such, sparse
gives out a lot of false warnings.
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
These two functions are called during kernel boot only.
Link: http://lkml.kernel.org/r/51258796.7020704@huawei.com
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Move duplicate code in event print functions to a helper function.
This shrinks the size of the kernel by ~13K.
text data bss dec hex filename
6596137 1743966 10138672 18478775 119f6b7 vmlinux.o.old
6583002 1743849 10138672 18465523 119c2f3 vmlinux.o.new
Link: http://lkml.kernel.org/r/51258746.2060304@huawei.com
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Move the logic to wake up on ring buffer data into the ring buffer
code itself. This simplifies the tracing code a lot and also has the
added benefit that waiters on one of the instance buffers can be woken
only when data is added to that instance instead of data added to
any instance.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
If the ring buffer is empty, a read to trace_pipe_raw wont block.
The tracing code has the infrastructure to wake up waiting readers,
but the trace_pipe_raw doesn't take advantage of that.
When a read is done to trace_pipe_raw without the O_NONBLOCK flag
set, have the read block until there's data in the requested buffer.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The trace_pipe_raw never implemented polling and this was casing
issues for several utilities. This is now implemented.
Blocked reads still are on the TODO list.
Reported-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Tested-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Currently only the splice NONBLOCK flag is checked to determine if
the splice read should block or not. But the file descriptor NONBLOCK
flag also needs to be checked.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The names used to display the field and type in the event format
files are copied, as well as the system name that is displayed.
All these names are created by constant values passed in.
If one of theses values were to be removed by a module, the module
would also be required to remove any event it created.
By using the strings directly, we can save over 100K of memory.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The event structures used by the trace events are mostly persistent,
but they are also allocated by kmalloc, which is not the best at
allocating space for what is used. By converting these kmallocs
into kmem_cache_allocs, we can save over 50K of space that is
permanently allocated.
After boot we have:
slab name active allocated size
--------- ------ --------- ----
ftrace_event_file 979 1005 56 67 1
ftrace_event_field 2301 2310 48 77 1
The ftrace_event_file has at boot up 979 active objects out of
1005 allocated in the slabs. Each object is 56 bytes. In a normal
kmalloc, that would allocate 64 bytes for each object.
1005 - 979 = 26 objects not used
26 * 56 = 1456 bytes wasted
But if we used kmalloc:
64 - 56 = 8 bytes unused per allocation
8 * 979 = 7832 bytes wasted
7832 - 1456 = 6376 bytes in savings
Doing the same for ftrace_event_field where there's 2301 objects
allocated in a slab that can hold 2310 with 48 bytes each we have:
2310 - 2301 = 9 objects not used
9 * 48 = 432 bytes wasted
A kmalloc would also use 64 bytes per object:
64 - 48 = 16 bytes unused per allocation
16 * 2301 = 36816 bytes wasted!
36816 - 432 = 36384 bytes in savings
This change gives us a total of 42760 bytes in savings. At least
on my machine, but as there's a lot of these persistent objects
for all configurations that use trace points, this is a net win.
Thanks to Ezequiel Garcia for his trace_analyze presentation which
pointed out the wasted space in my code.
Cc: Ezequiel Garcia <elezegarcia@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
With the new descriptors used to allow multiple buffers in the
tracing directory added, the kernel command line parameter
trace_events=... no longer works. This is because the top level
(global) trace array now has a list of descriptors associated
with the events and the files in the debugfs directory. But in
early bootup, when the command line is processed and the events
enabled, the trace array list of events has not been set up yet.
Without the list of events in the trace array, the setting of
events to record will fail because it would not match any events.
The solution is to set up the top level array in two stages.
The first is to just add the ftrace file descriptors that just point
to the events. This will allow events to be enabled and start tracing.
The second stage is called after the filesystem is set up, and this
stage will create the debugfs event files and directories associated
with the trace array events.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Add a method to the hijacked dentry descriptor of the
"instances" directory to allow for rmdir to remove an
instance of a multibuffer.
Example:
cd /debug/tracing/instances
mkdir hello
ls
hello/
rmdir hello
ls
Like the mkdir method, the i_mutex is dropped for the instances
directory. The instances directory is created at boot up and can
not be renamed or removed. The trace_types_lock mutex is used to
synchronize adding and removing of instances.
I've run several stress tests with different threads trying to
create and delete directories of the same name, and it has stood
up fine.
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Add the interface ("instances" directory) to add multiple buffers
to ftrace. To create a new instance, simply do a mkdir in the
instances directory:
This will create a directory with the following:
# cd instances
# mkdir foo
# ls foo
buffer_size_kb free_buffer trace_clock trace_pipe
buffer_total_size_kb set_event trace_marker tracing_enabled
events/ trace trace_options tracing_on
Currently only events are able to be set, and there isn't a way
to delete a buffer when one is created (yet).
Note, the i_mutex lock is dropped from the parent "instances"
directory during the mkdir operation. As the "instances" directory
can not be renamed or deleted (created on boot), I do not see
any harm in dropping the lock. The creation of the sub directories
is protected by trace_types_lock mutex, which only lets one
instance get into the code path at a time. If two tasks try to
create or delete directories of the same name, only one will occur
and the other will fail with -EEXIST.
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Currently the syscall events record into the global buffer. But if
multiple buffers are in place, then we need to have syscall events
record in the proper buffers.
By adding descriptors to pass to the syscall event functions, the
syscall events can now record into the buffers that have been assigned
to them (one event may be applied to mulitple buffers).
This will allow tracing high volume syscalls along with seldom occurring
syscalls without losing the seldom syscall events.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The global and max-tr currently use static per_cpu arrays for the CPU data
descriptors. But in order to get new allocated trace_arrays, they need to
be allocated per_cpu arrays. Instead of using the static arrays, switch
the global and max-tr to use allocated data.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Pass the struct ftrace_event_file *ftrace_file to the
trace_event_buffer_lock_reserve() (new function that replaces the
trace_current_buffer_lock_reserver()).
The ftrace_file holds a pointer to the trace_array that is in use.
In the case of multiple buffers with different trace_arrays, this
allows different events to be recorded into different buffers.
Also fixed some of the stale comments in include/trace/ftrace.h
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The global_trace variable in kernel/trace/trace.c has been kept 'static' and
local to that file so that it would not be used too much outside of that
file. This has paid off, even though there were lots of changes to make
the trace_array structure more generic (not depending on global_trace).
Removal of a lot of direct usages of global_trace is needed to be able to
create more trace_arrays such that we can add multiple buffers.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Both RING_BUFFER_ALL_CPUS and TRACE_PIPE_ALL_CPU are defined as
-1 and used to say that all the ring buffers are to be modified
or read (instead of just a single cpu, which would be >= 0).
There's no reason to keep TRACE_PIPE_ALL_CPU as it is also started
to be used for more than what it was created for, and now that
the ring buffer code added a generic RING_BUFFER_ALL_CPUS define,
we can clean up the trace code to use that instead and remove
the TRACE_PIPE_ALL_CPU macro.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The trace events for ftrace are all defined via global variables.
The arrays of events and event systems are linked to a global list.
This prevents multiple users of the event system (what to enable and
what not to).
By adding descriptors to represent the event/file relation, as well
as to which trace_array descriptor they are associated with, allows
for more than one set of events to be defined. Once the trace events
files have a link between the trace event and the trace_array they
are associated with, we can create multiple trace_arrays that can
record separate events in separate buffers.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The latency tracers require the buffers to be in overwrite mode,
otherwise they get screwed up. Force the buffers to stay in overwrite
mode when latency tracers are enabled.
Added a flag_changed() method to the tracer structure to allow
the tracers to see what flags are being changed, and also be able
to prevent the change from happing.
Cc: stable@vger.kernel.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Changing the overwrite mode for the ring buffer via the trace
option only sets the normal buffer. But the snapshot buffer could
swap with it, and then the snapshot would be in non overwrite mode
and the normal buffer would be in overwrite mode, even though the
option flag states otherwise.
Keep the two buffers overwrite modes in sync.
Cc: stable@vger.kernel.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Seems that the tracer flags have never been protected from
synchronous writes. Luckily, admins don't usually modify the
tracing flags via two different tasks. But if scripts were to
be used to modify them, then they could get corrupted.
Move the trace_types_lock that protects against tracers changing
to also protect the flags being set.
Cc: stable@vger.kernel.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Because function tracing is very invasive, and can even trace
calls to rcu_read_lock(), RCU access in function tracing is done
with preempt_disable_notrace(). This requires a synchronize_sched()
for updates and not a synchronize_rcu().
Function probes (traceon, traceoff, etc) must be freed after
a synchronize_sched() after its entry has been removed from the
hash. But call_rcu() is used. Fix this by using call_rcu_sched().
Also fix the usage to use hlist_del_rcu() instead of hlist_del().
Cc: stable@vger.kernel.org
Cc: Paul McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Although the swap is wrapped with a spin_lock, the assignment
of the temp buffer used to swap is not within that lock.
It needs to be moved into that lock, otherwise two swaps
happening on two different CPUs, can end up using the wrong
temp buffer to assign in the swap.
Luckily, all current callers of the swap function appear to have
their own locks. But in case something is added that allows two
different callers to call the swap, then there's a chance that
this race can trigger and corrupt the buffers.
New code is coming soon that will allow for this race to trigger.
I've Cc'd stable, so this bug will not show up if someone backports
one of the changes that can trigger this bug.
Cc: stable@vger.kernel.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
To use the tracing snapshot feature, writing a '1' into the snapshot
file causes the snapshot buffer to be allocated if it has not already
been allocated and dose a 'swap' with the main buffer, so that the
snapshot now contains what was in the main buffer, and the main buffer
now writes to what was the snapshot buffer.
To free the snapshot buffer, a '0' is written into the snapshot file.
To clear the snapshot buffer, any number but a '0' or '1' is written
into the snapshot file. But if the file is not allocated it returns
-EINVAL error code. This is rather pointless. It is better just to
do nothing and return success.
Acked-by: Hiraku Toyooka <hiraku.toyooka.gu@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When cat'ing the snapshot file, instead of showing an empty trace
header like the trace file does, show how to use the snapshot
feature.
Also, this is a good place to show if the snapshot has been allocated
or not. Users may want to "pre allocate" the snapshot to have a fast
"swap" of the current buffer. Otherwise, a swap would be slow and might
fail as it would need to allocate the snapshot buffer, and that might
fail under tight memory constraints.
Here's what it looked like before:
# tracer: nop
#
# entries-in-buffer/entries-written: 0/0 #P:4
#
# _-----=> irqs-off
# / _----=> need-resched
# | / _---=> hardirq/softirq
# || / _--=> preempt-depth
# ||| / delay
# TASK-PID CPU# |||| TIMESTAMP FUNCTION
# | | | |||| | |
Here's what it looks like now:
# tracer: nop
#
#
# * Snapshot is freed *
#
# Snapshot commands:
# echo 0 > snapshot : Clears and frees snapshot buffer
# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.
# Takes a snapshot of the main buffer.
# echo 2 > snapshot : Clears snapshot buffer (but does not allocate)
# (Doesn't have to be '2' works with any number that
# is not a '0' or '1')
Acked-by: Hiraku Toyooka <hiraku.toyooka.gu@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The prompt to enable DYNAMIC_FTRACE (the ability to nop and
enable function tracing at run time) had a confusing statement:
"enable/disable ftrace tracepoints dynamically"
This was written before tracepoints were added to the kernel,
but now that tracepoints have been added, this is very confusing
and has confused people enough to give wrong information during
presentations.
Not only that, I looked at the help text, and it still references
that dreaded daemon that use to wake up once a second to update
the nop locations and brick NICs, that hasn't been around for over
five years.
Time to bring the text up to the current decade.
Cc: stable@vger.kernel.org
Reported-by: Ezequiel Garcia <elezegarcia@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Pull scheduler changes from Ingo Molnar:
"Main changes:
- scheduler side full-dynticks (user-space execution is undisturbed
and receives no timer IRQs) preparation changes that convert the
cputime accounting code to be full-dynticks ready, from Frederic
Weisbecker.
- Initial sched.h split-up changes, by Clark Williams
- select_idle_sibling() performance improvement by Mike Galbraith:
" 1 tbench pair (worst case) in a 10 core + SMT package:
pre 15.22 MB/sec 1 procs
post 252.01 MB/sec 1 procs "
- sched_rr_get_interval() ABI fix/change. We think this detail is not
used by apps (so it's not an ABI in practice), but lets keep it
under observation.
- misc RT scheduling cleanups, optimizations"
* 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (24 commits)
sched/rt: Add <linux/sched/rt.h> header to <linux/init_task.h>
cputime: Remove irqsave from seqlock readers
sched, powerpc: Fix sched.h split-up build failure
cputime: Restore CPU_ACCOUNTING config defaults for PPC64
sched/rt: Move rt specific bits into new header file
sched/rt: Add a tuning knob to allow changing SCHED_RR timeslice
sched: Move sched.h sysctl bits into separate header
sched: Fix signedness bug in yield_to()
sched: Fix select_idle_sibling() bouncing cow syndrome
sched/rt: Further simplify pick_rt_task()
sched/rt: Do not account zero delta_exec in update_curr_rt()
cputime: Safely read cputime of full dynticks CPUs
kvm: Prepare to add generic guest entry/exit callbacks
cputime: Use accessors to read task cputime stats
cputime: Allow dynamic switch between tick/virtual based cputime accounting
cputime: Generic on-demand virtual cputime accounting
cputime: Move default nsecs_to_cputime() to jiffies based cputime file
cputime: Librarize per nsecs resolution cputime definitions
cputime: Avoid multiplication overflow on utime scaling
context_tracking: Export context state for generic vtime
...
Fix up conflict in kernel/context_tracking.c due to comment additions.
Pull perf changes from Ingo Molnar:
"There are lots of improvements, the biggest changes are:
Main kernel side changes:
- Improve uprobes performance by adding 'pre-filtering' support, by
Oleg Nesterov.
- Make some POWER7 events available in sysfs, equivalent to what was
done on x86, from Sukadev Bhattiprolu.
- tracing updates by Steve Rostedt - mostly misc fixes and smaller
improvements.
- Use perf/event tracing to report PCI Express advanced errors, by
Tony Luck.
- Enable northbridge performance counters on AMD family 15h, by Jacob
Shin.
- This tracing commit:
tracing: Remove the extra 4 bytes of padding in events
changes the ABI. All involved parties (PowerTop in particular)
seem to agree that it's safe to do now with the introduction of
libtraceevent, but the devil is in the details ...
Main tooling side changes:
- Add 'event group view', from Namyung Kim:
To use it, 'perf record' should group events when recording. And
then perf report parses the saved group relation from file header
and prints them together if --group option is provided. You can
use the 'perf evlist' command to see event group information:
$ perf record -e '{ref-cycles,cycles}' noploop 1
[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 0.385 MB perf.data (~16807 samples) ]
$ perf evlist --group
{ref-cycles,cycles}
With this example, default perf report will show you each event
separately.
You can use --group option to enable event group view:
$ perf report --group
...
# group: {ref-cycles,cycles}
# ========
# Samples: 7K of event 'anon group { ref-cycles, cycles }'
# Event count (approx.): 6876107743
#
# Overhead Command Shared Object Symbol
# ................ ....... ................. ..........................
99.84% 99.76% noploop noploop [.] main
0.07% 0.00% noploop ld-2.15.so [.] strcmp
0.03% 0.00% noploop [kernel.kallsyms] [k] timerqueue_del
0.03% 0.03% noploop [kernel.kallsyms] [k] sched_clock_cpu
0.02% 0.00% noploop [kernel.kallsyms] [k] account_user_time
0.01% 0.00% noploop [kernel.kallsyms] [k] __alloc_pages_nodemask
0.00% 0.00% noploop [kernel.kallsyms] [k] native_write_msr_safe
0.00% 0.11% noploop [kernel.kallsyms] [k] _raw_spin_lock
0.00% 0.06% noploop [kernel.kallsyms] [k] find_get_page
0.00% 0.02% noploop [kernel.kallsyms] [k] rcu_check_callbacks
0.00% 0.02% noploop [kernel.kallsyms] [k] __current_kernel_time
As you can see the Overhead column now contains both of ref-cycles
and cycles and header line shows group information also - 'anon
group { ref-cycles, cycles }'. The output is sorted by period of
group leader first.
- Initial GTK+ annotate browser, from Namhyung Kim.
- Add option for runtime switching perf data file in perf report,
just press 's' and a menu with the valid files found in the current
directory will be presented, from Feng Tang.
- Add support to display whole group data for raw columns, from Jiri
Olsa.
- Add per processor socket count aggregation in perf stat, from
Stephane Eranian.
- Add interval printing in 'perf stat', from Stephane Eranian.
- 'perf test' improvements
- Add support for wildcards in tracepoint system name, from Jiri
Olsa.
- Add anonymous huge page recognition, from Joshua Zhu.
- perf build-id cache now can show DSOs present in a perf.data file
that are not in the cache, to integrate with build-id servers being
put in place by organizations such as Fedora.
- perf top now shares more of the evsel config/creation routines with
'record', paving the way for further integration like 'top'
snapshots, etc.
- perf top now supports DWARF callchains.
- Fix mmap limitations on 32-bit, fix from David Miller.
- 'perf bench numa mem' NUMA performance measurement suite
- ... and lots of fixes, performance improvements, cleanups and other
improvements I failed to list - see the shortlog and git log for
details."
* 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (270 commits)
perf/x86/amd: Enable northbridge performance counters on AMD family 15h
perf/hwbp: Fix cleanup in case of kzalloc failure
perf tools: Fix build with bison 2.3 and older.
perf tools: Limit unwind support to x86 archs
perf annotate: Make it to be able to skip unannotatable symbols
perf gtk/annotate: Fail early if it can't annotate
perf gtk/annotate: Show source lines with gray color
perf gtk/annotate: Support multiple event annotation
perf ui/gtk: Implement basic GTK2 annotation browser
perf annotate: Fix warning message on a missing vmlinux
perf buildid-cache: Add --update option
uprobes/perf: Avoid uprobe_apply() whenever possible
uprobes/perf: Teach trace_uprobe/perf code to use UPROBE_HANDLER_REMOVE
uprobes/perf: Teach trace_uprobe/perf code to pre-filter
uprobes/perf: Teach trace_uprobe/perf code to track the active perf_event's
uprobes: Introduce uprobe_apply()
perf: Introduce hw_perf_event->tp_target and ->tp_list
uprobes/perf: Always increment trace_uprobe->nhit
uprobes/tracing: Kill uprobe_trace_consumer, embed uprobe_consumer into trace_uprobe
uprobes/tracing: Introduce is_trace_uprobe_enabled()
...
Commit: c1bf08ac "ftrace: Be first to run code modification on modules"
changed ftrace module notifier's priority to INT_MAX in order to
process the ftrace nops before anything else could touch them
(namely kprobes). This was the correct thing to do.
Unfortunately, the ftrace module notifier also contains the ftrace
clean up code. As opposed to the set up code, this code should be
run *after* all the module notifiers have run in case a module is doing
correct clean-up and unregisters its ftrace hooks. Basically, ftrace
needs to do clean up on module removal, as it needs to know about code
being removed so that it doesn't try to modify that code. But after it
removes the module from its records, if a ftrace user tries to remove
a probe, that removal will fail due as the record of that code segment
no longer exists.
Nothing really bad happens if the probe removal is called after ftrace
did the clean up, but the ftrace removal function will return an error.
Correct code (such as kprobes) will produce a WARN_ON() if it fails
to remove the probe. As people get annoyed by frivolous warnings, it's
best to do the ftrace clean up after everything else.
By splitting the ftrace_module_notifier into two notifiers, one that
does the module load setup that is run at high priority, and the other
that is called for module clean up that is run at low priority, the
problem is solved.
Cc: stable@vger.kernel.org
Reported-by: Frank Ch. Eigler <fche@redhat.com>
Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The tracing of ia32 compat system calls has been a bit of a pain as they
use different system call numbers than the 64bit equivalents.
I wrote a simple 'lls' program that lists files. I compiled it as a i686
ELF binary and ran it under a x86_64 box. This is the result:
echo 0 > /debug/tracing/tracing_on
echo 1 > /debug/tracing/events/syscalls/enable
echo 1 > /debug/tracing/tracing_on ; ./lls ; echo 0 > /debug/tracing/tracing_on
grep lls /debug/tracing/trace
[.. skipping calls before TS_COMPAT is set ...]
lls-1127 [005] d... 936.409188: sys_recvfrom(fd: 0, ubuf: 4d560fc4, size: 0, flags: 8048034, addr: 8, addr_len: f7700420)
lls-1127 [005] d... 936.409190: sys_recvfrom -> 0x8a77000
lls-1127 [005] d... 936.409211: sys_lgetxattr(pathname: 0, name: 1000, value: 3, size: 22)
lls-1127 [005] d... 936.409215: sys_lgetxattr -> 0xf76ff000
lls-1127 [005] d... 936.409223: sys_dup2(oldfd: 4d55ae9b, newfd: 4)
lls-1127 [005] d... 936.409228: sys_dup2 -> 0xfffffffffffffffe
lls-1127 [005] d... 936.409236: sys_newfstat(fd: 4d55b085, statbuf: 80000)
lls-1127 [005] d... 936.409242: sys_newfstat -> 0x3
lls-1127 [005] d... 936.409243: sys_removexattr(pathname: 3, name: ffcd0060)
lls-1127 [005] d... 936.409244: sys_removexattr -> 0x0
lls-1127 [005] d... 936.409245: sys_lgetxattr(pathname: 0, name: 19614, value: 1, size: 2)
lls-1127 [005] d... 936.409248: sys_lgetxattr -> 0xf76e5000
lls-1127 [005] d... 936.409248: sys_newlstat(filename: 3, statbuf: 19614)
lls-1127 [005] d... 936.409249: sys_newlstat -> 0x0
lls-1127 [005] d... 936.409262: sys_newfstat(fd: f76fb588, statbuf: 80000)
lls-1127 [005] d... 936.409279: sys_newfstat -> 0x3
lls-1127 [005] d... 936.409279: sys_close(fd: 3)
lls-1127 [005] d... 936.421550: sys_close -> 0x200
lls-1127 [005] d... 936.421558: sys_removexattr(pathname: 3, name: ffcd00d0)
lls-1127 [005] d... 936.421560: sys_removexattr -> 0x0
lls-1127 [005] d... 936.421569: sys_lgetxattr(pathname: 4d564000, name: 1b1abc, value: 5, size: 802)
lls-1127 [005] d... 936.421574: sys_lgetxattr -> 0x4d564000
lls-1127 [005] d... 936.421575: sys_capget(header: 4d70f000, dataptr: 1000)
lls-1127 [005] d... 936.421580: sys_capget -> 0x0
lls-1127 [005] d... 936.421580: sys_lgetxattr(pathname: 4d710000, name: 3000, value: 3, size: 812)
lls-1127 [005] d... 936.421589: sys_lgetxattr -> 0x4d710000
lls-1127 [005] d... 936.426130: sys_lgetxattr(pathname: 4d713000, name: 2abc, value: 3, size: 32)
lls-1127 [005] d... 936.426141: sys_lgetxattr -> 0x4d713000
lls-1127 [005] d... 936.426145: sys_newlstat(filename: 3, statbuf: f76ff3f0)
lls-1127 [005] d... 936.426146: sys_newlstat -> 0x0
lls-1127 [005] d... 936.431748: sys_lgetxattr(pathname: 0, name: 1000, value: 3, size: 22)
Obviously I'm not calling newfstat with a fd of 4d55b085. The calls are
obviously incorrect, and confusing.
Other efforts have been made to fix this:
https://lkml.org/lkml/2012/3/26/367
But the real solution is to rewrite the syscall internals and come up
with a fixed solution. One that doesn't require all the kluge that the
current solution has.
Thus for now, instead of outputting incorrect data, simply ignore them.
With this patch the changes now have:
#> grep lls /debug/tracing/trace
#>
Compat system calls simply are not traced. If users need compat
syscalls, then they should just use the raw syscall tracepoints.
For an architecture to make their compat syscalls ignored, it must
define ARCH_TRACE_IGNORE_COMPAT_SYSCALLS (done in asm/ftrace.h) and also
define an arch_trace_is_compat_syscall() function that will return true
if the current task should ignore tracing the syscall.
I want to stress that this change does not affect actual syscalls in any
way, shape or form. It is only used within the tracing system and
doesn't interfere with the syscall logic at all. The changes are
consolidated nicely into trace_syscalls.c and asm/ftrace.h.
I had to make one small modification to asm/thread_info.h and that was
to remove the include of asm/ftrace.h. As asm/ftrace.h required the
current_thread_info() it was causing include hell. That include was
added back in 2008 when the function graph tracer was added:
commit caf4b323 "tracing, x86: add low level support for ftrace return tracing"
It does not need to be included there.
Link: http://lkml.kernel.org/r/1360703939.21867.99.camel@gandalf.local.home
Acked-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
uprobe_perf_open/close call the costly uprobe_apply() every time,
we can avoid it if:
- "nr_systemwide != 0" is not changed.
- There is another process/thread with the same ->mm.
- copy_proccess() does inherit_event(). dup_mmap() preserves the
inserted breakpoints.
- event->attr.enable_on_exec == T, we can rely on uprobe_mmap()
called by exec/mmap paths.
- tp_target is exiting. Only _close() checks PF_EXITING, I don't
think TRACE_REG_PERF_OPEN can hit the dying task too often.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Change uprobe_trace_func() and uprobe_perf_func() to return "int". Change
uprobe_dispatcher() to return "trace_ret | perf_ret" although this is not
needed, currently TP_FLAG_TRACE/TP_FLAG_PROFILE are mutually exclusive.
The only functional change is that uprobe_perf_func() checks the filtering
too and returns UPROBE_HANDLER_REMOVE if nobody wants to trace current.
Testing:
# perf probe -x /lib/libc.so.6 syscall
# perf record -e probe_libc:syscall -i perl -e 'fork; syscall -1 for 1..10; wait'
# perf report --show-total-period
100.00% 10 perl libc-2.8.so [.] syscall
Before this patch:
# cat /sys/kernel/debug/tracing/uprobe_profile
/lib/libc.so.6 syscall 20
A child process doesn't have a counter, but still it hits this breakoint
"copied" by dup_mmap().
After the patch:
# cat /sys/kernel/debug/tracing/uprobe_profile
/lib/libc.so.6 syscall 11
The child process hits this int3 only once and does unapply_uprobe().
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Finally implement uprobe_perf_filter() which checks ->nr_systemwide or
->perf_events to figure out whether we need to insert the breakpoint.
uprobe_perf_open/close are changed to do uprobe_apply(true/false) when
the new perf event comes or goes away.
Note that currently this is very suboptimal:
- uprobe_register() called by TRACE_REG_PERF_REGISTER becomes a
heavy nop, consumer->filter() always returns F at this stage.
As it was already discussed we need uprobe_register_only() to
avoid the costly register_for_each_vma() when possible.
- uprobe_apply() is oftenly overkill. Unless "nr_systemwide != 0"
changes we need uprobe_apply_mm(), unapply_uprobe() is almost
what we need.
- uprobe_apply() can be simply avoided sometimes, see the next
changes.
Testing:
# perf probe -x /lib/libc.so.6 syscall
# perl -e 'syscall -1 while 1' &
[1] 530
# perf record -e probe_libc:syscall perl -e 'syscall -1 for 1..10; sleep 1'
# perf report --show-total-period
100.00% 10 perl libc-2.8.so [.] syscall
Before this patch:
# cat /sys/kernel/debug/tracing/uprobe_profile
/lib/libc.so.6 syscall 79291
A huge ->nrhit == 79291 reflects the fact that the background process
530 constantly hits this breakpoint too, even if doesn't contribute to
the output.
After the patch:
# cat /sys/kernel/debug/tracing/uprobe_profile
/lib/libc.so.6 syscall 10
This shows that only the target process was punished by int3.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Introduce "struct trace_uprobe_filter" which records the "active"
perf_event's attached to ftrace_event_call. For the start we simply
use list_head, we can optimize this later if needed. For example, we
do not really need to record an event with ->parent != NULL, we can
rely on parent->child_list. And we can certainly do some optimizations
for the case when 2 events have the same ->tp_target or tp_target->mm.
Change trace_uprobe_register() to process TRACE_REG_PERF_OPEN/CLOSE
and add/del this perf_event to the list.
We can probably avoid any locking, but lets start with the "obvioulsy
correct" trace_uprobe_filter->rwlock which protects everything.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Move tu->nhit++ from uprobe_trace_func() to uprobe_dispatcher().
->nhit counts how many time we hit the breakpoint inserted by this
uprobe, we do not want to loose this info if uprobe was enabled by
sys_perf_event_open().
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
trace_uprobe->consumer and "struct uprobe_trace_consumer" add the
unnecessary indirection and complicate the code for no reason.
This patch simply embeds uprobe_consumer into "struct trace_uprobe",
all other changes only fix the compilation errors.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
probe_event_enable/disable() check tu->consumer != NULL to avoid the
wrong uprobe_register/unregister().
We are going to kill this pointer and "struct uprobe_trace_consumer",
so we add the new helper, is_trace_uprobe_enabled(), which can rely
on TP_FLAG_TRACE/TP_FLAG_PROFILE instead.
Note: the current logic doesn't look optimal, it is not clear why
TP_FLAG_TRACE/TP_FLAG_PROFILE are mutually exclusive, we will probably
change this later.
Also kill the unused TP_FLAG_UPROBE.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
probe_event_enable/disable() check tu->inode != NULL at the start.
This is ugly, if igrab() can fail create_trace_uprobe() should not
succeed and "postpone" the failure.
And S_ISREG(inode->i_mode) check added by d24d7dbf is not safe.
Note: alloc_uprobe() should probably check igrab() != NULL as well.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
probe_event_enable() does uprobe_register() and only after that sets
utc->tu and tu->consumer/flags. This can race with uprobe_dispatcher()
which can miss these assignments or see them out of order. Nothing
really bad can happen, but this doesn't look clean/safe.
And this does not allow to use uprobe_consumer->filter() we are going
to add, it is called by uprobe_register() and it needs utc->tu.
Change this code to initialize everything before uprobe_register(), and
reset tu->consumer/flags if it fails. We can't race with event_disable(),
the caller holds event_mutex, and if we could the code would be wrong
anyway.
In fact I think uprobe_trace_consumer should die, it buys nothing but
complicates the code. We can simply add uprobe_consumer into trace_uprobe.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
create_trace_uprobe() does kern_path() to find ->d_inode, but forgets
to do path_put(). We can do this right after igrab().
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Change handle_swbp() to set regs->ip = bp_vaddr in advance, this is
what consumer->handler() needs but uprobe_get_swbp_addr() is not
exported.
This also simplifies the code and makes it more consistent across
the supported architectures. handle_swbp() becomes the only caller
of uprobe_get_swbp_addr().
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
uprobe_consumer->filter() is pointless in its current form, kill it.
We will add it back, but with the different signature/semantics. Perhaps
we will even re-introduce the callsite in handler_chain(), but not to
just skip uc->handler().
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Move rt scheduler definitions out of include/linux/sched.h into
new file include/linux/sched/rt.h
Signed-off-by: Clark Williams <williams@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/20130207094707.7b9f825f@riff.lan
Signed-off-by: Ingo Molnar <mingo@kernel.org>
On early boot up, when the ftrace ring buffer is initialized, the
static variable current_trace is initialized to &nop_trace.
Before this initialization, current_trace is NULL and will never
become NULL again. It is always reassigned to a ftrace tracer.
Several places check if current_trace is NULL before it uses
it, and this check is frivolous, because at the point in time
when the checks are made the only way current_trace could be
NULL is if ftrace failed its allocations at boot up, and the
paths to these locations would probably not be possible.
By initializing current_trace to &nop_trace where it is declared,
current_trace will never be NULL, and we can remove all these
checks of current_trace being NULL which never needed to be
checked in the first place.
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Hiraku Toyooka <hiraku.toyooka.gu@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Steven Rostedt (Red Hat)