- 31 Mar, 2016 26 commits
-
-
Arnaldo Carvalho de Melo authored
We catch this record to provide a visual indication that events are getting lost, then call the default method to allow extra logging shared with the other tools to take place. This extra logging was done twice because we were continuing to the "default" clause where machine__process_event() will end up calling machine__process_lost_event() again, fix it. 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-wus2zlhw3qo24ye84ewu4aqw@git.kernel.orgSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Wang Nan authored
Convert perf_output_begin() to __perf_output_begin() and make the later function able to write records from the end of the ring-buffer. Following commits will utilize the 'backward' flag. This is the core patch to support writing to the ring-buffer backwards, which will be introduced by upcoming patches to support reading from overwritable ring-buffers. In theory, this patch should not introduce any extra performance overhead since we use always_inline, but it does not hurt to double check that assumption: When CONFIG_OPTIMIZE_INLINING is disabled, the output object is nearly identical to original one. See: http://lkml.kernel.org/g/56F52E83.70409@huawei.com When CONFIG_OPTIMIZE_INLINING is enabled, the resuling object file becomes smaller: $ size kernel/events/ring_buffer.o* text data bss dec hex filename 4641 4 8 4653 122d kernel/events/ring_buffer.o.old 4545 4 8 4557 11cd kernel/events/ring_buffer.o.new Performance testing results: 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 PRE 2253846.700 9997.014 POST 2257495.540 8516.293 Where 'BASE' is pure performance without capturing. 'PRE' is test result of pure 'v4.5.0' kernel. 'POST' is test result after this patch. Considering the stdvar, this patch doesn't hurt performance, within noise margin. For testing details, see: http://lkml.kernel.org/g/56F89DCD.1040202@huawei.comSigned-off-by:
Wang Nan <wangnan0@huawei.com> Signed-off-by:
Peter Zijlstra (Intel) <peterz@infradead.org> Cc: <pi3orama@163.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexei Starovoitov <ast@kernel.org> 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/1459147292-239310-4-git-send-email-wangnan0@huawei.comSigned-off-by:
Ingo Molnar <mingo@kernel.org>
-
Wang Nan authored
Set a default event->overflow_handler in perf_event_alloc() so don't need to check event->overflow_handler in __perf_event_overflow(). Following commits can give a different default overflow_handler. Initial idea comes from Peter: http://lkml.kernel.org/r/20130708121557.GA17211@twins.programming.kicks-ass.net Since the default value of event->overflow_handler is not NULL, existing 'if (!overflow_handler)' checks need to be changed. is_default_overflow_handler() is introduced for this. No extra performance overhead is introduced into the hot path because in the original code we still need to read this handler from memory. A conditional branch is avoided so actually we remove some instructions. Signed-off-by:
-
Wang Nan authored
Add new ioctl() to pause/resume ring-buffer output. In some situations we want to read from the ring-buffer only when we ensure nothing can write to the ring-buffer during reading. Without this patch we have to turn off all events attached to this ring-buffer to achieve this. This patch is a prerequisite to enable overwrite support for the perf ring-buffer support. Following commits will introduce new methods support reading from overwrite ring buffer. Before reading, caller must ensure the ring buffer is frozen, or the reading is unreliable. Signed-off-by:
-
Jiri Olsa authored
Currently we check sample type for ftrace:function events even if it's not created as a sampling event. That prevents creating ftrace_function event in counting mode. Make sure we check sample types only for sampling events. Before: $ sudo perf stat -e ftrace:function ls ... Performance counter stats for 'ls': <not supported> ftrace:function 0.001983662 seconds time elapsed After: $ sudo perf stat -e ftrace:function ls ... Performance counter stats for 'ls': 44,498 ftrace:function 0.037534722 seconds time elapsed Suggested-by:Namhyung Kim <namhyung@kernel.org> Signed-off-by:
Jiri Olsa <jolsa@kernel.org> Signed-off-by:
Steven Rostedt <rostedt@goodmis.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> 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/1458138873-1553-2-git-send-email-jolsa@kernel.orgSigned-off-by:
-
Alexander Shishkin authored
As per AUX buffer management requirement, AUX output has to happen between pmu::start and pmu::stop calls so that perf_event_stop() actually stops it and therefore perf can free the AUX data after it has called pmu::stop. This patch moves perf_aux_output_{begin,end} from bts_event_{add,del} to bts_event_{start,stop}. As a bonus, we get rid of bts_buffer_is_full(), which is already taken care of by perf_aux_output_begin() anyway. Signed-off-by:Alexander Shishkin <alexander.shishkin@linux.intel.com> Signed-off-by:
-
Alexander Shishkin authored
As per AUX buffer management requirement, AUX output has to happen between pmu::start and pmu::stop calls so that perf_event_stop() actually stops it and therefore perf can free the AUX data after it has called pmu::stop. This patch moves perf_aux_output_{begin,end} from pt_event_{add,del} to pt_event_{start,stop}. As a bonus, we get rid of pt_buffer_is_full(), which is already taken care of by perf_aux_output_begin() anyway. Signed-off-by: -
Alexander Shishkin authored
In order to ensure safe AUX buffer management, we rely on the assumption that pmu::stop() stops its ongoing AUX transaction and not just the hw. This patch documents this requirement for the perf_aux_output_{begin,end}() APIs. Signed-off-by: -
Alexander Shishkin authored
Now that we can ensure that when ring buffer's AUX area is on the way to getting unmapped new transactions won't start, we only need to stop all events that can potentially be writing aux data to our ring buffer. Having done that, we can safely free the AUX pages and corresponding PMU data, as this time it is guaranteed to be the last aux reference holder. This partially reverts: 57ffc5ca ("perf: Fix AUX buffer refcounting") ... which was made to defer deallocation that was otherwise possible from an NMI context. Now it is no longer the case; the last call to rb_free_aux() that drops the last AUX reference has to happen in perf_mmap_close() on that AUX area. Signed-off-by:
-
Alexander Shishkin authored
When ring buffer's AUX area is unmapped and rb->aux_mmap_count drops to zero, new AUX transactions into this buffer can still be started, even though the buffer in en route to deallocation. This patch adds a check to perf_aux_output_begin() for rb->aux_mmap_count being zero, in which case there is no point starting new transactions, in other words, the ring buffers that pass a certain point in perf_mmap_close will not have their events sending new data, which clears path for freeing those buffers' pages right there and then, provided that no active transactions are holding the AUX reference. Signed-off-by:
-
Peter Zijlstra authored
There should (and can) only be a single PMU for perf_hw_context events. This is because of how we schedule events: once a hardware event fails to schedule (the PMU is 'full') we stop trying to add more. The trivial 'fix' would break the Round-Robin scheduling we do. Signed-off-by:
-
Peter Zijlstra authored
Ingo says: "If we do a separate file we should have it in arch/x86/events/Kconfig (not in arch/x86/Kconfig.perf), and also move some of the other bits, such as PERF_EVENTS_AMD_POWER?" Suggested-by:
-
Huang Rui authored
AMD Zeppelin (Family 17h, Model 00h) introduces an instructions retired performance counter which is indicated by CPUID.8000_0008H:EBX[1]. A dedicated Instructions Retired MSR register (MSR 0xC000_000E9) increments once for every instruction retired. Signed-off-by:
Huang Rui <ray.huang@amd.com> Signed-off-by:
-
Huang Rui authored
AMD Carrizo (Family 15h, Model 60h) introduces a time-stamp counter which is indicated by CPUID.8000_0001H:ECX[27]. It increments at a 100 MHz rate in all P-states, and C states, S0, or S1. The frequency is about 100MHz. This counter will be used to calculate processor power and other parts. So add an interface into the MSR PMU to get the PTSC counter value. Signed-off-by:
-
Thomas Gleixner authored
Add the exit function and allow the driver to be built as a module. Signed-off-by:
Thomas Gleixner <tglx@linutronix.de> Signed-off-by:
-
Thomas Gleixner authored
There is no point in WARN_ON() inside of a well known init function. We already know the call stack and it's really not of critical importance whether the registration of a PMU fails. Aside of that for consistency reasons it's just pointless to try to register another PMU if the first register attempt failed. There is also no value in keeping one PMU if the second one can not be registered. Make it consistent so we can finaly modularize the driver. Signed-off-by:
-
Thomas Gleixner authored
The whole probing functionality can simply be expressed with model matching and a bunch of structures describing the variants. This is a first step to make that driver modular. While at it, get rid of completely pointless comments and name the enums so they are self explaining. Signed-off-by:
-
Thomas Gleixner authored
The current implementation aside of being an incomprehensible mess is broken. # cat /sys/bus/event_source/devices/cstate_core/cpumask 0-17 That's on a quad socket machine with 72 physical cores! Qualitee stuff. So it's not a surprise that event migration in case of CPU hotplug does not work either. # perf stat -e cstate_core/c6-residency/ -C 1 sleep 60 & # echo 0 >/sys/devices/system/cpu/cpu1/online Tracing cstate_pmu_event_update gives me: [001] cstate_pmu_event_update <-event_sched_out After the fix it properly moves the event: [001] cstate_pmu_event_update <-event_sched_out [073] cstate_pmu_event_update <-__perf_event_read [073] cstate_pmu_event_update <-event_sched_out The migration of pkg events does not work either. Not that I'm surprised. I really could not be bothered to decode that loop mess and simply replaced it by querying the proper cpumasks which give us the answer in a comprehensible way. This also requires to direct the event to the current active reader CPU in cstate_pmu_event_init() otherwise the hotplug logic can't work. Signed-off-by:
-
Kan Liang authored
By default, the RAPL driver will be built into the kernel. If it is configured as a module, the supported CPU model can be auto loaded. Also clean up the code of rapl_pmu_init(). Based-on-a-patch-by:
Kan Liang <kan.liang@intel.com> Signed-off-by:
-
Kan Liang authored
By default, the uncore driver will be built into the kernel. If it is configured as a module, the supported CPU model can be auto loaded. This patch also cleans up the code of uncore_cpu_init() and uncore_pci_init(). Based-on-a-patch-by:
-
Ingo Molnar authored
Signed-off-by: -
Peter Zijlstra authored
Patch 5a50f529 ("perf/x86/ibs: Fix race with IBS_STARTING state") closed a big hole while opening another, smaller hole. Signed-off-by:
-
Alexander Shishkin authored
In the error path, event_file not being NULL is used to determine whether the event itself still needs to be free'd, so fix it up to avoid leaking. Reported-by:
Leon Yu <chianglungyu@gmail.com> Signed-off-by:
-
Peter Zijlstra authored
Stephane reported that commit: 3cbaa590 ("perf: Fix ctx time tracking by introducing EVENT_TIME") introduced a regression wrt. time tracking, as easily observed by: > This patch introduce a bug in the time tracking of events when > multiplexing is used. > > The issue is easily reproducible with the following perf run: > > $ perf stat -a -C 0 -e branches,branches,branches,branches,branches,branches -I 1000 > 1.000730239 652,394 branches (66.41%) > 1.000730239 597,809 branches (66.41%) > 1.000730239 593,870 branches (66.63%) > 1.000730239 651,440 branches (67.03%) > 1.000730239 656,725 branches (66.96%) > 1.000730239 <not counted> branches > > One branches event is shown as not having run. Yet, with > multiplexing, all events should run especially with a 1s (-I 1000) > interval. The delta for time_running comes out to 0. Yet, the event > has run because the kernel is actually multiplexing the events. The > problem is that the time tracking is the kernel and especially in > ctx_sched_out() is wrong now. > > The problem is that in case that the kernel enters ctx_sched_out() with the > following state: > ctx->is_active=0x7 event_type=0x1 > Call Trace: > [<ffffffff813ddd41>] dump_stack+0x63/0x82 > [<ffffffff81182bdc>] ctx_sched_out+0x2bc/0x2d0 > [<ffffffff81183896>] perf_mux_hrtimer_handler+0xf6/0x2c0 > [<ffffffff811837a0>] ? __perf_install_in_context+0x130/0x130 > [<ffffffff810f5818>] __hrtimer_run_queues+0xf8/0x2f0 > [<ffffffff810f6097>] hrtimer_interrupt+0xb7/0x1d0 > [<ffffffff810509a8>] local_apic_timer_interrupt+0x38/0x60 > [<ffffffff8175ca9d>] smp_apic_timer_interrupt+0x3d/0x50 > [<ffffffff8175ac7c>] apic_timer_interrupt+0x8c/0xa0 > > In that case, the test: > if (is_active & EVENT_TIME) > > will be false and the time will not be updated. Time must always be updated on > sched out. Fix this by always updating time if EVENT_TIME was set, as opposed to only updating time when EVENT_TIME changed. Reported-by:
Stephane Eranian <eranian@google.com> Tested-by:
-
Ingo Molnar authored
Merge tag 'perf-core-for-mingo-20160330' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core Pull perf/core improvements and fixes: User visible changes: - Add support for skipping itrace instructions, useful to fast forward processor trace (Intel PT, BTS) to right after initialization code at the start of a workload (Andi Kleen) - Add support for backtraces in perl 'perf script's (Dima Kogan) - Add -U/-K (--all-user/--all-kernel) options to 'perf mem' (Jiri Olsa) - Make -f/--force option documentation consistent across tools (Jiri Olsa) Infrastructure changes: - Add 'perf test' to check for event times (Jiri Olsa) - 'perf config' cleanups (Taeung Song) Signed-off-by:
-
Ingo Molnar authored
Merge tag 'perf-urgent-for-mingo-20160330' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/urgent Pull perf/urgent fixes from Arnaldo Carvalho de Melo: - Fix determination of a callchain node's childlessness in the top/report TUI, which was preventing navigating some callchains, --stdio unnaffected (Andres Freund) - Fix jitdump's genelf assumption that PowerPC is big endian only (Anton Blanchard) Signed-off-by:
-
- 30 Mar, 2016 12 commits
-
-
Anton Blanchard authored
Commit 9b07e27f ("perf inject: Add jitdump mmap injection support") incorrectly assumed that PowerPC is big endian only. Simplify things by consolidating the define of GEN_ELF_ENDIAN and checking for __BYTE_ORDER == __BIG_ENDIAN. The PowerPC checks were also incorrect, they do not match what gcc emits. We should first look for __powerpc64__, then __powerpc__. Signed-off-by:
Anton Blanchard <anton@samba.org> Acked-by:
Michael Ellerman <mpe@ellerman.id.au> Cc: Carl Love <cel@us.ibm.com> Cc: Stephane Eranian <eranian@google.com> Cc: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com> Cc: linuxppc-dev@lists.ozlabs.org Fixes: 9b07e27f ("perf inject: Add jitdump mmap injection support") Link: http://lkml.kernel.org/r/20160329175944.33a211cc@krytenSigned-off-by:
-
Andres Freund authored
The 4b3a3212 ("perf hists browser: Support flat callchains") commit over-aggressively tried to optimize callchain_node__init_have_children(). That lead to --tui mode not allowing to expand call chain elements if a call chain element had only one parent. That's why --inverted callgraphs looked halfway sane, but plain ones didn't. Revert that individual optimization, it wasn't really related to the rest of the commit. Signed-off-by:
Andres Freund <andres@anarazel.de> Acked-by:
-
Andi Kleen authored
When using 'perf script' to look at PT traces it is often useful to ignore the initialization code at the beginning. On larger traces which may have many millions of instructions in initialization code doing that in a pipeline can be very slow, with perf script spending a lot of CPU time calling printf and writing data. This patch adds an extension to the --itrace argument that skips 'n' events (instructions, branches or transactions) at the beginning. This is much more efficient. v2: Add support for BTS (Adrian Hunter) Document in itrace.txt Fix branch check Check transactions and instructions too Committer note: To test intel_pt one needs to make sure VT-x isn't active, i.e. stopping KVM guests on the test machine, as described by Andi Kleen at http://lkml.kernel.org/r/20160301234953.GD23621@tassilo.jf.intel.comSigned-off-by:
Andi Kleen <ak@linux.intel.com> Tested-by:
-
Dima Kogan authored
We have some infrastructure to use perl or python to analyze logs generated by perf. Prior to this patch, only the python tools had access to backtrace information. This patch makes this information available to perl scripts as well. Example: Let's look at malloc() calls made by the seq utility. First we create a probe point: $ perf probe -x /lib/x86_64-linux-gnu/libc.so.6 malloc Added new events: ... Now we run seq, while monitoring malloc() calls with perf $ perf record --call-graph=dwarf -e probe_libc:malloc seq 5 1 2 3 4 5 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.064 MB perf.data (6 samples) ] We can use perf to look at its log to see the malloc calls and the backtrace $ perf script seq 14195 [000] 1927993.748254: probe_libc:malloc: (7f9ff8edd320) bytes=0x22 7f9ff8edd320 malloc (/lib/x86_64-linux-gnu/libc-2.22.so) 7f9ff8e8eab0 set_binding_values.part.0 (/lib/x86_64-linux-gnu/libc-2.22.so) 7f9ff8e8eda1 __bindtextdomain (/lib/x86_64-linux-gnu/libc-2.22.so) 401b22 main (/usr/bin/seq) 7f9ff8e82610 __libc_start_main (/lib/x86_64-linux-gnu/libc-2.22.so) 402799 _start (/usr/bin/seq) ... We can also use the scripting facilities. We create a skeleton perl script that simply prints out the events $ perf script -g perl generated Perl script: perf-script.pl We can then use this script to see the malloc() calls with a backtrace. Prior to this patch, the backtrace was not available to the perl scripts. $ perf script -s perf-script.pl probe_libc::malloc 0 1927993.748254260 14195 seq __probe_ip=140325052863264, bytes=34 [7f9ff8edd320] malloc [7f9ff8e8eab0] set_binding_values.part.0 [7f9ff8e8eda1] __bindtextdomain [401b22] main [7f9ff8e82610] __libc_start_main [402799] _start ... Tested-by:Dima Kogan <dima@secretsauce.net>
-
Taeung Song authored
Change the variable name 'v' to 'home' to make it more readable. Signed-off-by:
Taeung Song <treeze.taeung@gmail.com> Acked-by:
-
Taeung Song authored
To avoid repeated calling perf_config() remove buildid_dir_command_config() and add new perf_buildid_config into perf_default_config. Because perf_config() is already called with perf_default_config at main(). Signed-off-by:
-
Taeung Song authored
Signed-off-by:
-
Jiri Olsa authored
This test creates software event 'cpu-clock' attaches it in several ways and checks that enabled and running times match. Committer notes: Testing it: [acme@jouet linux]$ perf test -v times 44: Test events times : --- start --- test child forked, pid 27170 attaching to spawned child, enable on exec OK : ena 307328, run 307328 attaching to current thread as enabled OK : ena 7826, run 7826 attaching to current thread as disabled OK : ena 738, run 738 attaching to CPU 0 as enabled SKIP : not enough rights attaching to CPU 0 as enabled SKIP : not enough rights test child finished with -2 ---- end ---- Test events times: Skip [acme@jouet linux]$ [root@jouet ~]# perf test times 44: Test events times : Ok [root@jouet ~]# perf test -v times 44: Test events times : --- start --- test child forked, pid 27306 attaching to spawned child, enable on exec OK : ena 479290, run 479290 attaching to current thread as enabled OK : ena 11356, run 11356 attaching to current thread as disabled OK : ena 987, run 987 attaching to CPU 0 as enabled OK : ena 3717, run 3717 attaching to CPU 0 as enabled OK : ena 2323, run 2323 test child finished with 0 ---- end ---- Test events times: Ok [root@jouet ~]# Signed-off-by: -
Jiri Olsa authored
Signed-off-by:
-
Jiri Olsa authored
No need to export hists__collapse_insert_entry function. Signed-off-by:
-
Jiri Olsa authored
Add -U/-K (--all-user/--all-kernel) options to use the perf record --all-user/--all-kernel options. Signed-off-by:
-
Ingo Molnar authored
Merge tag 'perf-urgent-for-mingo-20160329' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/urgent Pull perf/urgent fix from Arnaldo Carvalho de Melo: - Add missing initialization of perf_sample.cpumode in synthesized samples, affects jitdump, records for pre-existing threads and records synthesized from processor trace data, noticed while testing intel_pt events with 'perf script' (Arnaldo Carvalho de Melo) Signed-off-by:
-
- 29 Mar, 2016 2 commits
-
-
Arnaldo Carvalho de Melo authored
In 473398a2 ("perf tools: Add cpumode to struct perf_sample"), I missed some places where perf_sample fields are directly initialized in addition to what is done in perf_evsel__parse_sample(), namely when synthesizing PERF_RECORD_{MMAP*,COMM,FORK,EXIT} for pre-existing threads and also in intel_pt and intel_bts when synthesizing events from processor trace, the jitdump code also was affected, fix it. The problem was noticed with running: # perf record -e intel_pt//u true # perf script Where the samples wouldn't get resolved because perf_sample.cpumode would be left as zero, i.e. PERF_RECORD_MISC_CPUMODE_UNKNOWN, not resolving as kernel, hypervisor or user cpu modes. Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Jiri Olsa <jolsa@redhat.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> Fixes: 473398a2 ("perf tools: Add cpumode to struct perf_sample") Link: http://lkml.kernel.org/n/tip-n5sdauxgk24d5nun8kuuu2mh@git.kernel.orgSigned-off-by:
-
Ingo Molnar authored
Merge tag 'perf-urgent-for-mingo-20160328' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/urgent Pull perf/urgent fix from Arnaldo Carvalho de Melo: - Fix build break on PowerPC due to missing headers with prototypes for functions defined in tools/perf/arch/powerpc/util/header.c (Sukadev Bhattiprolu) Signed-off-by:
-
