- 06 Nov, 2013 4 commits
-
-
Peter Zijlstra authored
There's no point in re-doing the memory-barrier when we fail the cmpxchg(). Also placing it after the space reservation loop makes it clearer it only separates the userpage->tail read from the data stores. Signed-off-by:
Peter Zijlstra <peterz@infradead.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca> Cc: Michael Ellerman <michael@ellerman.id.au> Cc: Michael Neuling <mikey@neuling.org> Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> Cc: james.hogan@imgtec.com Cc: Vince Weaver <vince@deater.net> Cc: Victor Kaplansky <VICTORK@il.ibm.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Anton Blanchard <anton@samba.org> Link: http://lkml.kernel.org/n/tip-c19u6egfldyx86tpyc3zgkw9@git.kernel.orgSigned-off-by:
Ingo Molnar <mingo@kernel.org>
-
Peter Zijlstra authored
Add unlikely() annotations to 'slow' paths: When having a sampling event but no output buffer; you have bigger issues -- also the bail is still faster than actually doing the work. When having a sampling event but a control page only buffer, you have bigger issues -- again the bail is still faster than actually doing work. Optimize for the case where you're not loosing events -- again, not doing the work is still faster but make sure that when you have to actually do work its as fast as possible. The typical watermark is 1/2 the buffer size, so most events will not take this path. Shrinks perf_output_begin() by 16 bytes on x86_64-defconfig. Signed-off-by:
-
Peter Zijlstra authored
By using CIRC_SPACE() we can obviate the need for perf_output_space(). Shrinks the size of perf_output_begin() by 17 bytes on x86_64-defconfig. Signed-off-by:
-
Ingo Molnar authored
Merge tag 'perf-core-for-mingo' 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: * Check maximum frequency rate for record/top, emitting better error messages, from Jiri Olsa. * Disable live kvm command if timerfd is not supported, from David Ahern. * Add usage to 'perf list', from David Ahern. * Fix detection of non-core features, from David Ahern. * Consolidate __hists__add_*entry(), cleanup from Namhyung Kim. Signed-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by:
-
- 05 Nov, 2013 9 commits
-
-
Arnaldo Carvalho de Melo authored
They convey no information, perhaps I was bitten by some snake at some point, complete the detox by naming the last of those arguments more sensibly. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-u1r0dnjoro08dgztiy2g3t2q@git.kernel.orgSigned-off-by:
-
Jiri Olsa authored
Adding the check for maximum allowed frequency rate defined in following file: /proc/sys/kernel/perf_event_max_sample_rate When we cross the maximum value we fail and display detailed error message with advise. $ perf record -F 3000 ls Maximum frequency rate (2000) reached. Please use -F freq option with lower value or consider tweaking /proc/sys/kernel/perf_event_max_sample_rate. In case user does not specify the frequency and the default value cross the maximum, we display warning and set the frequency value to the current maximum. $ perf record ls Lowering default frequency rate to 2000. Please consider tweaking /proc/sys/kernel/perf_event_max_sample_rate. Same messages are used for 'perf top'. Signed-off-by:
Jiri Olsa <jolsa@redhat.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> 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/1383660887-1734-4-git-send-email-jolsa@redhat.comSigned-off-by:
-
Jiri Olsa authored
Adding procfs support into fs class. The interface function: const char *procfs__mountpoint(void); provides existing mountpoint path for procfs. Signed-off-by:
-
Arnaldo Carvalho de Melo authored
Shorten it, "finding" it is an implementation detail, what callers want is the pathname, not to ask for it to _always_ do the lookup. And the existing implementation already caches it, i.e. it doesn't "finds" it on every call. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-r24wa4bvtccg7mnkessrbbdj@git.kernel.orgSigned-off-by:
-
Jiri Olsa authored
Moving sysfs code into generic fs object and preparing it to carry procfs support. This should be merged with tools/lib/lk/debugfs.c at some point in the future. Signed-off-by:
-
David Ahern authored
Currently 'perf list' is not very helpful if you forget the syntax: $ perf list -h List of pre-defined events (to be used in -e): After: $ perf list -h usage: perf list [hw|sw|cache|tracepoint|pmu|event_glob] Signed-off-by:
David Ahern <dsahern@gmail.com> Acked-by:
-
David Ahern authored
With a return after the if check an indentation level can be removed. Indentation shift only; no functional changes. Signed-off-by:
-
David Ahern authored
feature_check needs to be invoked through call, and LDFLAGS may not be set so quotes are needed. Thanks to Jiri for spotting the quotes around LDFLAGS; that one was driving me nuts with the upcoming timerfd feature detection. Signed-off-by:
-
David Ahern authored
If the OS does not have timerfd support (e.g., older OS'es like RHEL5) disable perf kvm stat live. Signed-off-by:
-
- 04 Nov, 2013 27 commits
-
-
Namhyung Kim authored
The __hists__add_{branch,mem}_entry() does almost the same thing that __hists__add_entry() does. Consolidate them into one. Signed-off-by:Namhyung Kim <namhyung@kernel.org> Acked-by:
-
Ingo Molnar authored
Merge tag 'perf-core-for-mingo' 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: * Add new COMM infrastructure, further improving histogram processing, from Frédéric Weisbecker, one fix from Namhyung Kim. * Enhance option parse error message, showing just the help lines of the options affected, from Namhyung Kim. * Fixup PERF_SAMPLE_TRANSACTION handling in sample synthesizing and 'perf test', from Adrian Hunter. * Set up output options for in-stream attributes, from Adrian Hunter. * Fix 32-bit cross build, from Adrian Hunter. * Fix libunwind build and feature detection for 32-bit build, from Adrian Hunter. * Always use perf_evsel__set_sample_bit to set sample_type, from Adrian Hunter. perf evlist: Add a debug print if event buffer mmap fails * Add missing data.h into LIB_H headers, fix from Jiri Olsa. * libtraceevent updates from upstream trace-cmd repo, from Steven Rostedt. Signed-off-by:
-
Steven Rostedt authored
Add the pevent_print_func_field() that will look up a field that is expected to be a function pointer, and it will print the function name and offset of the address given by the field. Signed-off-by:
Steven Rostedt <rostedt@goodmis.org> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/20131101215501.869542711@goodmis.orgSigned-off-by:
-
Steven Rostedt authored
Add the flags EVENT_FL_NOHANDLE and EVENT_FL_PRINTRAW to the event flags to have the event either ignore the register handler or to ignore the handler and also print the raw format respectively. This allows a tool to force a raw format or non handle for an event. Signed-off-by:
-
Steven Rostedt (Red Hat) authored
Currently when using the raw format for fields, when looking at a character array, to determine if it is a string or not, we make sure all characters are "isprint()". If not, then we consider it a numeric array, and print the hex numbers of the characters instead. But it seems that '\n' fails the isprint() check! Add isspace() to the check as well, such that if all characters pass isprint() or isspace() it will assume the character array is a string. Reported-by:
Xenia Ragiadakou <burzalodowa@gmail.com> Signed-off-by:
-
Steven Rostedt (Red Hat) authored
The trace_bprintk() in the kernel looks like: ring_buffer_producer_thread: Missed: 0 ring_buffer_producer_thread: Hit: 62174350 ring_buffer_producer_thread: Entries per millisec: 6296 ring_buffer_producer_thread: 158 ns per entry ring_buffer_producer_thread: Sleeping for 10 secs ring_buffer_producer_thread: Starting ring buffer hammer ring_buffer_producer_thread: End ring buffer hammer But the current output looks like this: ring_buffer_producer_thread : Time: 9407018 (usecs) ring_buffer_producer_thread : Overruns: 43285485 ring_buffer_producer_thread : Read: 4405365 (by events) ring_buffer_producer_thread : Entries: 0 ring_buffer_producer_thread : Total: 47690850 ring_buffer_producer_thread : Missed: 0 ring_buffer_producer_thread : Hit: 47690850 Remove the space between the function and the colon. Signed-off-by:
-
Howard Cochran authored
The kernel has a few events with a format similar to this excerpt: field:unsigned int len; offset:12; size:4; signed:0; field:__data_loc unsigned char[] data_array; offset:16; size:4; signed:0; print fmt: "%s", __print_hex(__get_dynamic_array(data_array), REC->len) trace-cmd could already parse that arg correctly, but print_str_arg() was unable to handle the first parameter being a dynamic array. (It just printed a "field not found" warning). Teach print_str_arg's PRINT_HEX case to handle the nested PRINT_DYNAMIC_ARRAY correctly. The output now matches the kernel's own formatting for this case. Signed-off-by:Howard Cochran <hcochran@lexmark.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/1381503349-12271-1-git-send-email-hcochran@lexmark.com [ Removed "polish compare", we don't do that here ] Signed-off-by:
-
Steven Rostedt (Red Hat) authored
If the format string of TP_printk() contains a %s, and the argument is not a string, check if the argument is a pointer that might match the printk_formats that were stored. Signed-off-by:
-
Steven Rostedt (Red Hat) authored
Instead of cropping off the '"' and '\n"' from a printk format every time it is referenced, do it when it's added. This makes it easier to reference a printk_map and should speed things up a little. Signed-off-by:
-
Yoshihiro YUNOMAE authored
If trace-cmd extracts trace_clock, trace-cmd reads trace_clock data from the trace.dat and switches outputting format of timestamp for each trace_clock. Signed-off-by:
Yoshihiro YUNOMAE <yoshihiro.yunomae.ez@hitachi.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Link: http://lkml.kernel.org/r/20130424231305.14877.86147.stgit@yunodevelSigned-off-by:
-
Namhyung Kim authored
Print related option help messages only when it failed to process options. While at it, modify parse_options_usage() to skip usage part so that it can be used for showing multiple option help messages naturally like below: $ perf stat -Bx, ls -B option not supported with -x usage: perf stat [<options>] [<command>] -B, --big-num print large numbers with thousands' separators -x, --field-separator <separator> print counts with custom separator Signed-off-by: -
Namhyung Kim authored
The -s (--sort) option was processed after normal option parsing so that it cannot call the parse_options_usage() automatically. Currently it calls usage_with_options() which shows entire help messages for event option. Fix it by showing just -s options. $ perf top -s help Error: Unknown --sort key: `help' usage: perf top [<options>] -s, --sort <key[,key2...]> sort by key(s): pid, comm, dso, symbol, ... Signed-off-by: -
Namhyung Kim authored
The -s (--sort) option was processed after normal option parsing so that it cannot call the parse_options_usage() automatically. Currently it calls usage_with_options() which shows entire help messages for event option. Fix it by showing just -s options. $ perf report -s help Error: Unknown --sort key: `help' usage: perf report [<options>] -s, --sort <key[,key2...]> sort by key(s): pid, comm, dso, symbol, ... Signed-off-by: -
Namhyung Kim authored
If setup_browser() called earlier than option parsing, the actual error message can be discarded during the terminal reset. So move it after setup_sorting() checks whether the sort keys are valid. Signed-off-by:
-
Namhyung Kim authored
Current option parser outputs whole option help string when it failed to parse an option. However this is not good for user if the command has many option, she might feel hard which one is related easily. Fix it by just showing the help message of the given option only. Signed-off-by:
-
Adrian Hunter authored
Add missing PERF_SAMPLE_TRANSACTION to perf_event__synthesize_sample() and perf_event__sample_event_size(). This makes the "sample parsing" test pass. Signed-off-by:
Adrian Hunter <adrian.hunter@intel.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.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/1383313899-15987-11-git-send-email-adrian.hunter@intel.comSigned-off-by:
-
Adrian Hunter authored
In fact the "sample parsing" test does not automatically check new sample type bits - they must be added to the comparison logic. Doing that shows that the test fails because the functions perf_event__synthesize_sample() and perf_event__sample_event_size() have not been updated with PERF_SAMPLE_TRANSACTION either. Signed-off-by:
-
Adrian Hunter authored
Add missing overflow check for PERF_SAMPLE_TRANSACTION in perf_evsel__parse_sample(). Signed-off-by:
-
Adrian Hunter authored
Always use perf_evsel__set_sample_bit() rather than just setting the bit. Signed-off-by:
-
Adrian Hunter authored
Add a debug print if mmap of the perf event ring buffer fails. Signed-off-by:
-
Adrian Hunter authored
Use -lunwind-x86 instead of -lunwind-x86_64 for 32-bit build. Signed-off-by:
-
Adrian Hunter authored
Setting EXTRA_CFLAGS=-m32 did not work because it was not passed around. Signed-off-by:
-
Adrian Hunter authored
Attributes (struct perf_event_attr) are recorded separately in the perf.data file. perf script uses them to set up output options. However attributes can also be in the event stream, for example when the input is a pipe (i.e. live mode). This patch makes perf script process in-stream attributes in the same way as on-file attributes. Here is an example: Before this patch: $ perf record uname | perf script Linux [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.015 MB (null) (~655 samples) ] :4220 4220 [-01] 2933367.838906: cycles: :4220 4220 [-01] 2933367.838910: cycles: :4220 4220 [-01] 2933367.838912: cycles: :4220 4220 [-01] 2933367.838914: cycles: :4220 4220 [-01] 2933367.838916: cycles: :4220 4220 [-01] 2933367.838918: cycles: uname 4220 [-01] 2933367.838938: cycles: uname 4220 [-01] 2933367.839207: cycles: After this patch: $ perf record uname | perf script Linux [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.015 MB (null) (~655 samples) ] :4582 4582 2933425.707724: cycles: ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms]) :4582 4582 2933425.707728: cycles: ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms]) :4582 4582 2933425.707730: cycles: ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms]) :4582 4582 2933425.707732: cycles: ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms]) :4582 4582 2933425.707734: cycles: ffffffff81043ffa native_write_msr_safe ([kernel.kallsyms]) :4582 4582 2933425.707736: cycles: ffffffff81309a24 memcpy ([kernel.kallsyms]) uname 4582 2933425.707760: cycles: ffffffff8109c1c7 enqueue_task_fair ([kernel.kallsyms]) uname 4582 2933425.707978: cycles: ffffffff81308457 clear_page_c ([kernel.kallsyms]) Signed-off-by: -
Adrian Hunter authored
There is a debug print (at verbose level 2) for each call to perf_event_open. Add another debug print if the call fails, and print the error number. Signed-off-by:
-
Namhyung Kim authored
At insert time, a hist entry should reference comm at the time otherwise it'll get the last comm anyway. Signed-off-by:
Frederic Weisbecker <fweisbec@gmail.com> Tested-by:
-
Frederic Weisbecker authored
Now that comm strings are allocated only once and refcounted to be shared among threads, these can now be safely compared by addresses. This should remove most hists collapses on post processing. Signed-off-by:
-
Frederic Weisbecker authored
This new COMM infrastructure provides two features: 1) It keeps track of all comms lifecycle for a given thread. This way we can associate a timeframe to any thread COMM, as long as PERF_SAMPLE_TIME samples are joined to COMM and fork events. As a result we should have more precise COMM sorted hists with seperated entries for pre and post exec time after a fork. 2) It also makes sure that a given COMM string is not duplicated but rather shared among the threads that refer to it. This way the threads COMM can be compared against pointer values from the sort infrastructure. Signed-off-by:
-
