- 20 Dec, 2011 5 commits
-
-
Nelson Elhage authored
On failure, perf_evlist__mmap_per_{cpu,thread} will try to munmap() every map that doesn't have a NULL base. This will fail with EINVAL if one of them has base == MAP_FAILED, clobbering errno, so that perf_evlist__map will return EINVAL on any failure regardless of the root cause. Fix this by resetting failed maps to a NULL base. Acked-by:Namhyung Kim <namhyung@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1324301972-22740-2-git-send-email-nelhage@nelhage.comSigned-off-by:
Nelson Elhage <nelhage@nelhage.com> Signed-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Namhyung Kim authored
The '--call-graph' command line option can receive undocumented optional print_limit argument. Besides, use strtoul() to parse the option since its type is u32. Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1323703017-6060-2-git-send-email-namhyung@gmail.comSigned-off-by:
-
Robert Richter authored
Memory in struct perf_sample is not fully initialized during parsing. Depending on sampling data some parts may left unchanged. Zero out struct perf_sample first to avoid access to uninitialized memory. Cc: Ingo Molnar <mingo@elte.hu> Link: http://lkml.kernel.org/r/1323966762-8574-2-git-send-email-robert.richter@amd.comSigned-off-by:
Robert Richter <robert.richter@amd.com> Signed-off-by:
-
Andrew Vagin authored
The problem is that when SAMPLE_PERIOD is not set, the kernel generates a number of samples in proportion to an event's period. Number of these samples may be too big and the kernel throttles all samples above a defined limit. E.g.: I want to trace when a process sleeps. I created a process which sleeps for 1ms and for 4ms. perf got 100 events in both cases. swapper 0 [000] 1141.371830: sched_stat_sleep: comm=foo pid=1801 delay=1386750 [ns] swapper 0 [000] 1141.369444: sched_stat_sleep: comm=foo pid=1801 delay=4499585 [ns] In the first case a kernel want to send 4499585 events and in the second case it wants to send 1386750 events. perf-reports shows that process sleeps in both places equal time. Instead of this we can get only one sample with an attribute period. As result we have less data transferring between kernel and user-space and we avoid throttling of samples. The patch "events: Don't divide events if it has field period" added a kernel part of this functionality. Acked-by:
Arun Sharma <asharma@fb.com> Cc: Arun Sharma <asharma@fb.com> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: devel@openvz.org Link: http://lkml.kernel.org/r/1324391565-1369947-1-git-send-email-avagin@openvz.orgSigned-off-by:
Andrew Vagin <avagin@openvz.org> Signed-off-by:
-
Ingo Molnar authored
Merge branch 'for-tip' of git://git.kernel.org/pub/scm/linux/kernel/git/rric/oprofile into perf/core
-
- 12 Dec, 2011 3 commits
-
-
Andrew Vagin authored
It's the counterpart of perf_session__parse_sample. v2: fixed mistakes found by David Ahern. v3: s/data/sample/ s/perf_event__change_sample/perf_event__synthesize_sample Reviewed-by:David Ahern <dsahern@gmail.com> Cc: Arun Sharma <asharma@fb.com> Cc: David Ahern <dsahern@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: devel@openvz.org Link: http://lkml.kernel.org/r/1323266161-394927-3-git-send-email-avagin@openvz.orgSigned-off-by:
-
Robert Richter authored
The option is documented in man perf-script but was not yet implemented: -a Force system-wide collection. Scripts run without a <command> normally use -a by default, while scripts run with a <command> normally don't - this option allows the latter to be run in system-wide mode. As with perf record you now can profile in system-wide mode for the runtime of a given command, e.g.: # perf script -a syscall-counts sleep 2 Cc: Ingo Molnar <mingo@elte.hu> Link: http://lkml.kernel.org/r/1322229925-10075-1-git-send-email-robert.richter@amd.comSigned-off-by: -
Robert Richter authored
Fix mem leaks and missing NULL pointer checks after strdup(). And get_script_path() did not free __script_root in case of continue. Introduce a helper function get_script_root(). Cc: Ingo Molnar <mingo@elte.hu> Link: http://lkml.kernel.org/r/1322217520-3287-1-git-send-email-robert.richter@amd.comSigned-off-by:
-
- 07 Dec, 2011 2 commits
-
-
Andreas Krebbel authored
With this patch the OProfile Basic Mode Sampling support for System z is enhanced with a counter file system. That way hardware sampling can be configured using the user space tools with only little modifications. With the patch by default new cpu_types (s390/z10, s390/z196) are returned in order to indicate that we are running a CPU which provides the hardware sampling facility. Existing user space tools will complain about an unknown cpu type. In order to be compatible with existing user space tools the `cpu_type' module parameter has been added. Setting the parameter to `timer' will force the module to return `timer' as cpu_type. The module will still try to use hardware sampling if available and the hwsampling virtual filesystem will be also be available for configuration. So this has a different effect than using the generic oprofile module parameter `timer=1'. If the basic mode sampling is enabled on the machine and the cpu_type=timer parameter is not used the kernel module will provide the following virtual filesystem: /dev/oprofile/0/enabled /dev/oprofile/0/event /dev/oprofile/0/count /dev/oprofile/0/unit_mask /dev/oprofile/0/kernel /dev/oprofile/0/user In the counter file system only the values of 'enabled', 'count', 'kernel', and 'user' are evaluated by the kernel module. Everything else must contain fixed values. The 'event' value only supports a single event - HWSAMPLING with value 0. The 'count' value specifies the hardware sampling rate as it is passed to the CPU measurement facility. The 'kernel' and 'user' flags can now be used to filter for samples when using hardware sampling. Additionally also the following file will be created: /dev/oprofile/timer/enabled This will always be the inverted value of /dev/oprofile/0/enabled. 0 is not accepted without hardware sampling. Signed-off-by:
Andreas Krebbel <krebbel@linux.vnet.ibm.com> Signed-off-by:
-
Robert Richter authored
Removing remainings of oprofile_timer_exit() completly. Signed-off-by:
-
- 06 Dec, 2011 14 commits
-
-
Gleb Natapov authored
KVM needs to know perf capability to decide which PMU it can expose to a guest. Signed-off-by:
Gleb Natapov <gleb@redhat.com> Signed-off-by:
Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1320929850-10480-8-git-send-email-gleb@redhat.comSigned-off-by:
Ingo Molnar <mingo@elte.hu>
-
Peter Zijlstra authored
Implement the disabling of arch events as a quirk so that we can print a message along with it. This creates some visibility into the problem space and could allow us to work on adding more work-around like the AAJ80 one. Requested-by:
-
Gleb Natapov authored
Intel CPUs report non-available architectural events in cpuid leaf 0AH.EBX. Use it to disable events that are not available according to CPU. Signed-off-by:
-
Peter Zijlstra authored
Provide two initializers for jump_label_key that initialize it enabled or disabled. Also modify all jump_label code to allow for jump_labels to be initialized enabled. Signed-off-by:
-
Peter Zijlstra authored
WARNING: arch/x86/kernel/built-in.o(.text+0x4c71): Section mismatch in reference from the function arch_jump_label_transform_static() to the function .init.text:text_poke_early() The function arch_jump_label_transform_static() references the function __init text_poke_early(). This is often because arch_jump_label_transform_static lacks a __init annotation or the annotation of text_poke_early is wrong. Signed-off-by:
-
Ingo Molnar authored
Merge branch 'tip/perf/core' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace into perf/core
-
Gleb Natapov authored
jump_lable patching is very expensive operation that involves pausing all cpus. The patching of perf_sched_events jump_label is easily controllable from userspace by unprivileged user. When te user runs a loop like this: "while true; do perf stat -e cycles true; done" ... the performance of my test application that just increments a counter for one second drops by 4%. This is on a 16 cpu box with my test application using only one of them. An impact on a real server doing real work will be worse. Performance of KVM PMU drops nearly 50% due to jump_lable for "perf record" since KVM PMU implementation creates and destroys perf event frequently. This patch introduces a way to rate limit jump_label patching and uses it to fix the above problem. I believe that as jump_label use will spread the problem will become more common and thus solving it in a generic code is appropriate. Also fixing it in the perf code would result in moving jump_label accounting logic to perf code with all the ifdefs in case of JUMP_LABEL=n kernel. With this patch all details are nicely hidden inside jump_label code. Signed-off-by:
Jason Baron <jbaron@redhat.com> Signed-off-by:
-
Peter Zijlstra authored
Deng-Cheng Zhu reported that sibling events that were created disabled with enable_on_exec would never get enabled. Iterate all events instead of the group lists. Reported-by:
Deng-Cheng Zhu <dczhu@mips.com> Tested-by:
-
Peter Zijlstra authored
Signed-off-by:
-
Peter Zijlstra authored
This avoids a scheduling failure for cases like: cycles, cycles, instructions, instructions (on Core2) Which would end up being programmed like: PMC0, PMC1, FP-instructions, fail Because all events will have the same weight. Signed-off-by:
-
Robert Richter authored
The current x86 event scheduler fails to resolve scheduling problems of certain combinations of events and constraints. This happens if the counter mask of such an event is not a subset of any other counter mask of a constraint with an equal or higher weight, e.g. constraints of the AMD family 15h pmu: counter mask weight amd_f15_PMC30 0x09 2 <--- overlapping counters amd_f15_PMC20 0x07 3 amd_f15_PMC53 0x38 3 The scheduler does not find then an existing solution. Here is an example: event code counter failure possible solution 0x02E PMC[3,0] 0 3 0x043 PMC[2:0] 1 0 0x045 PMC[2:0] 2 1 0x046 PMC[2:0] FAIL 2 The event scheduler may not select the correct counter in the first cycle because it needs to know which subsequent events will be scheduled. It may fail to schedule the events then. To solve this, we now save the scheduler state of events with overlapping counter counstraints. If we fail to schedule the events we rollback to those states and try to use another free counter. Constraints with overlapping counters are marked with a new introduced overlap flag. We set the overlap flag for such constraints to give the scheduler a hint which events to select for counter rescheduling. The EVENT_CONSTRAINT_OVERLAP() macro can be used for this. Care must be taken as the rescheduling algorithm is O(n!) which will increase scheduling cycles for an over-commited system dramatically. The number of such EVENT_CONSTRAINT_OVERLAP() macros and its counter masks must be kept at a minimum. Thus, the current stack is limited to 2 states to limit the number of loops the algorithm takes in the worst case. On systems with no overlapping-counter constraints, this implementation does not increase the loop count compared to the previous algorithm. V2: * Renamed redo -> overlap. * Reimplementation using perf scheduling helper functions. V3: * Added WARN_ON_ONCE() if out of save states. * Changed function interface of perf_sched_restore_state() to use bool as return value. Signed-off-by: -
Robert Richter authored
This patch introduces x86 perf scheduler code helper functions. We need this to later add more complex functionality to support overlapping counter constraints (next patch). The algorithm is modified so that the range of weight values is now generated from the constraints. There shouldn't be other functional changes. With the helper functions the scheduler is controlled. There are functions to initialize, traverse the event list, find unused counters etc. The scheduler keeps its own state. V3: * Added macro for_each_set_bit_cont(). * Changed functions interfaces of perf_sched_find_counter() and perf_sched_next_event() to use bool as return value. * Added some comments to make code better understandable. V4: * Fix broken event assignment if weight of the first event is not wmin (perf_sched_init()). Signed-off-by:
-
Peter Zijlstra authored
Gleb writes: > Currently pmu is disabled and re-enabled on each timer interrupt even > when no rotation or frequency adjustment is needed. On Intel CPU this > results in two writes into PERF_GLOBAL_CTRL MSR per tick. On bare metal > it does not cause significant slowdown, but when running perf in a virtual > machine it leads to 20% slowdown on my machine. Cure this by keeping a perf_event_context::nr_freq counter that counts the number of active events that require frequency adjustments and use this in a similar fashion to the already existing nr_events != nr_active test in perf_rotate_context(). By being able to exclude both rotation and frequency adjustments a-priory for the common case we can avoid the otherwise superfluous PMU disable. Suggested-by:
-
Ingo Molnar authored
Merge reason: Add these cherry-picked commits so that future changes on perf/core don't conflict. Signed-off-by:
-
- 05 Dec, 2011 16 commits
-
-
Steven Rostedt authored
If the set_ftrace_filter is cleared by writing just whitespace to it, then the filter hash refcounts will be decremented but not updated. This causes two bugs: 1) No functions will be enabled for tracing when they all should be 2) If the users clears the set_ftrace_filter twice, it will crash ftrace: ------------[ cut here ]------------ WARNING: at /home/rostedt/work/git/linux-trace.git/kernel/trace/ftrace.c:1384 __ftrace_hash_rec_update.part.27+0x157/0x1a7() Modules linked in: Pid: 2330, comm: bash Not tainted 3.1.0-test+ #32 Call Trace: [<ffffffff81051828>] warn_slowpath_common+0x83/0x9b [<ffffffff8105185a>] warn_slowpath_null+0x1a/0x1c [<ffffffff810ba362>] __ftrace_hash_rec_update.part.27+0x157/0x1a7 [<ffffffff810ba6e8>] ? ftrace_regex_release+0xa7/0x10f [<ffffffff8111bdfe>] ? kfree+0xe5/0x115 [<ffffffff810ba51e>] ftrace_hash_move+0x2e/0x151 [<ffffffff810ba6fb>] ftrace_regex_release+0xba/0x10f [<ffffffff8112e49a>] fput+0xfd/0x1c2 [<ffffffff8112b54c>] filp_close+0x6d/0x78 [<ffffffff8113a92d>] sys_dup3+0x197/0x1c1 [<ffffffff8113a9a6>] sys_dup2+0x4f/0x54 [<ffffffff8150cac2>] system_call_fastpath+0x16/0x1b ---[ end trace 77a3a7ee73794a02 ]--- Link: http://lkml.kernel.org/r/20111101141420.GA4918@debianReported-by:
Rabin Vincent <rabin@rab.in> Signed-off-by:
Steven Rostedt <rostedt@goodmis.org>
-
Steven Rostedt authored
A update is made to the sched:sched_switch event that adds some logic to the first parameter of the __print_flags() that shows the state of tasks. This change cause perf to fail parsing the flags. A simple fix is needed to have the parser be able to process ops within the argument. Cc: stable@vger.kernel.org Reported-by:
-
Gleb Natapov authored
If cpu A calls jump_label_inc() just after atomic_add_return() is called by cpu B, atomic_inc_not_zero() will return value greater then zero and jump_label_inc() will return to a caller before jump_label_update() finishes its job on cpu B. Link: http://lkml.kernel.org/r/20111018175551.GH17571@redhat.com Cc: stable@vger.kernel.org Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by:
-
Steven Rostedt authored
A forced undef of a config value was used for testing and was accidently left in during the final commit. This causes x86 to run slower than needed while running function tracing as well as causes the function graph selftest to fail when DYNMAIC_FTRACE is not set. This is because the code in MCOUNT expects the ftrace code to be processed with the config value set that happened to be forced not set. The forced config option was left in by: commit 6331c28c ftrace: Fix dynamic selftest failure on some archs Link: http://lkml.kernel.org/r/20111102150255.GA6973@debian Cc: stable@vger.kernel.org Reported-by: -
Li Zefan authored
Though not all events have field 'prev_pid', it was allowed to do this: # echo 'prev_pid == 100' > events/sched/filter but commit 75b8e982 (tracing/filter: Swap entire filter of events) broke it without any reason. Link: http://lkml.kernel.org/r/4EAF46CF.8040408@cn.fujitsu.comSigned-off-by:
Li Zefan <lizf@cn.fujitsu.com> Signed-off-by:
-
Ilya Dryomov authored
Fix a bug introduced by e9dbfae5, which prevents event_subsystem from ever being released. Ref_count was added to keep track of subsystem users, not for counting events. Subsystem is created with ref_count = 1, so there is no need to increment it for every event, we have nr_events for that. Fix this by touching ref_count only when we actually have a new user - subsystem_open(). Cc: stable@vger.kernel.org Signed-off-by:
Ilya Dryomov <idryomov@gmail.com> Link: http://lkml.kernel.org/r/1320052062-7846-1-git-send-email-idryomov@gmail.comSigned-off-by:
-
Masami Hiramatsu authored
Add instruction dump mode to insn_sanity tool for checking decoder really decoded instructions. This mode is enabled when passing double -v (-vv) to insn_sanity. It is useful for who wants to check whether the decoder can decode some instructions correctly. e.g. $ echo 0f 73 10 11 | ./insn_sanity -y -vv -i - Instruction = { .prefixes = { .value = 0, bytes[] = {0, 0, 0, 0}, .got = 1, .nbytes = 0}, .rex_prefix = { .value = 0, bytes[] = {0, 0, 0, 0}, .got = 1, .nbytes = 0}, .vex_prefix = { .value = 0, bytes[] = {0, 0, 0, 0}, .got = 1, .nbytes = 0}, .opcode = { .value = 29455, bytes[] = {f, 73, 0, 0}, .got = 1, .nbytes = 2}, .modrm = { .value = 16, bytes[] = {10, 0, 0, 0}, .got = 1, .nbytes = 1}, .sib = { .value = 0, bytes[] = {0, 0, 0, 0}, .got = 1, .nbytes = 0}, .displacement = { .value = 0, bytes[] = {0, 0, 0, 0}, .got = 1, .nbytes = 0}, .immediate1 = { .value = 17, bytes[] = {11, 0, 0, 0}, .got = 1, .nbytes = 1}, .immediate2 = { .value = 0, bytes[] = {0, 0, 0, 0}, .got = 0, .nbytes = 0}, .attr = 44800, .opnd_bytes = 4, .addr_bytes = 8, .length = 4, .x86_64 = 1, .kaddr = 0x7fff0f7d9430} Success: decoded and checked 1 given instructions with 0 errors (seed:0x0) Signed-off-by:Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: yrl.pp-manager.tt@hitachi.com Link: http://lkml.kernel.org/r/20111205120603.15475.91192.stgit@cloudSigned-off-by:
-
Masami Hiramatsu authored
Since new Intel software developers manual introduces new format for AVX instruction set (including AVX2), it is important to update x86-opcode-map.txt to fit those changes. Signed-off-by:
-
Masami Hiramatsu authored
Fix x86 instruction decoder test to dump all error messages to stderr and others to stdout. Signed-off-by:
-
Masami Hiramatsu authored
Fix instruction decoder test (insn_sanity), so that it doesn't show both info and error messages twice on same instruction. (In that case, show only error message) Signed-off-by:
-
Masami Hiramatsu authored
For reducing memory usage of attribute table, x86 instruction decoder puts "Group" attribute only on "no-last-prefix" attribute table (same as vex_p == 0 case). Thus, the decoder should look no-last-prefix table first, and then only if it is not a group, move on to "with-last-prefix" table (vex_p != 0). However, current implementation, inat_get_avx_attribute() looks with-last-prefix directly. So, when decoding a grouped AVX instruction, the decoder fails to find correct group because there is no "Group" attribute on the table. This ends up with the mis-decoding of instructions, as Ingo reported in http://thread.gmane.org/gmane.linux.kernel/1214103 This patch fixes it to check no-last-prefix table first even if that is an AVX instruction, and get an attribute from "with last-prefix" table only if that is not a group. Reported-by:
-
Masami Hiramatsu authored
Fix arch/x86/tools/Makefile to compile both test tools correctly. This bug leads build error. Signed-off-by:
-
Ingo Molnar authored
Merge branch 'tip/perf/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace into perf/urgent
-
git://github.com/acmel/linuxIngo Molnar authored
-
Peter Zijlstra authored
When you do: $ perf record -e cycles,cycles,cycles noploop 10 You expect about 10,000 samples for each event, i.e., 10s at 1000samples/sec. However, this is not what's happening. You get much fewer samples, maybe 3700 samples/event: $ perf report -D | tail -15 Aggregated stats: TOTAL events: 10998 MMAP events: 66 COMM events: 2 SAMPLE events: 10930 cycles stats: TOTAL events: 3644 SAMPLE events: 3644 cycles stats: TOTAL events: 3642 SAMPLE events: 3642 cycles stats: TOTAL events: 3644 SAMPLE events: 3644 On a Intel Nehalem or even AMD64, there are 4 counters capable of measuring cycles, so there is plenty of space to measure those events without multiplexing (even with the NMI watchdog active). And even with multiplexing, we'd expect roughly the same number of samples per event. The root of the problem was that when the event that caused the buffer to become full was not the first event passed on the cmdline, the user notification would get lost. The notification was sent to the file descriptor of the overflowed event but the perf tool was not polling on it. The perf tool aggregates all samples into a single buffer, i.e., the buffer of the first event. Consequently, it assumes notifications for any event will come via that descriptor. The seemingly straight forward solution of moving the waitq into the ringbuffer object doesn't work because of life-time issues. One could perf_event_set_output() on a fd that you're also blocking on and cause the old rb object to be freed while its waitq would still be referenced by the blocked thread -> FAIL. Therefore link all events to the ringbuffer and broadcast the wakeup from the ringbuffer object to all possible events that could be waited upon. This is rather ugly, and we're open to better solutions but it works for now. Reported-by:Stephane Eranian <eranian@google.com> Finished-by:
-
Robert Richter authored
On AMD family 10h we see firmware bug messages like the following: [Firmware Bug]: cpu 6, try to use APIC500 (LVT offset 0) for vector 0x10400, but the register is already in use for vector 0xf9 on another cpu [Firmware Bug]: cpu 6, IBS interrupt offset 0 not available (MSRC001103A=0x0000000000000100) [Firmware Bug]: using offset 1 for IBS interrupts [Firmware Bug]: workaround enabled for IBS LVT offset perf: AMD IBS detected (0x00000007) We always see this, since the offsets are not assigned by the BIOS for this family. Force LVT offset assignment in this case. If the OS assignment fails, fallback to BIOS settings and try to setup this. The fallback to BIOS settings weakens the family check since force_ibs_eilvt_setup() may fail e.g. in case of virtual machines. But setup may still succeed if BIOS offsets are correct. Other families don't have a workaround implemented that assigns LVT offsets. It's ok, to drop calling force_ibs_eilvt_setup() for that families. With the patch the [Firmware Bug] messages vanish. We see now: IBS: LVT offset 1 assigned perf: AMD IBS detected (0x00000007) Signed-off-by:
-
