user/sven/linux.git/include/linux/perf_event.h, branch v3.3.5

perf: Fix broken interrupt rate throttling

2012-01-27T11:06:39Z

This patch fixes the sampling interrupt throttling mechanism. It was broken in v3.2. Events were not being unthrottled. The unthrottling mechanism required that events be checked at each timer tick. This patch solves this problem and also separates: - unthrottling - multiplexing - frequency-mode period adjustments Not all of them need to be executed at each timer tick. This third version of the patch is based on my original patch + PeterZ proposal (https://lkml.org/lkml/2012/1/7/87). At each timer tick, for each context: - if the current CPU has throttled events, we unthrottle events - if context has frequency-based events, we adjust sampling periods - if we have reached the jiffies interval, we multiplex (rotate) We decoupled rotation (multiplexing) from frequency-mode sampling period adjustments. They should not necessarily happen at the same rate. Multiplexing is subject to jiffies_interval (currently at 1 but could be higher once the tunable is exposed via sysfs). We have grouped frequency-mode adjustment and unthrottling into the same routine to minimize code duplication. When throttled while in frequency mode, we scan the events only once. We have fixed the threshold enforcement code in __perf_event_overflow(). There was a bug whereby it would allow more than the authorized rate because an increment of hwc->interrupts was not executed at the right place. The patch was tested with low sampling limit (2000) and fixed periods, frequency mode, overcommitted PMU. On a 2.1GHz AMD CPU: $ cat /proc/sys/kernel/perf_event_max_sample_rate 2000 We set a rate of 3000 samples/sec (2.1GHz/3000 = 700000): $ perf record -e cycles,cycles -c 700000 noploop 10 $ perf report -D | tail -21 Aggregated stats: TOTAL events: 80086 MMAP events: 88 COMM events: 2 EXIT events: 4 THROTTLE events: 19996 UNTHROTTLE events: 19996 SAMPLE events: 40000 cycles stats: TOTAL events: 40006 MMAP events: 5 COMM events: 1 EXIT events: 4 THROTTLE events: 9998 UNTHROTTLE events: 9998 SAMPLE events: 20000 cycles stats: TOTAL events: 39996 THROTTLE events: 9998 UNTHROTTLE events: 9998 SAMPLE events: 20000 For 10s, the cap is 2x2000x10 = 40000 samples. We get exactly that: 20000 samples/event. Signed-off-by: Stephane Eranian Cc: # v3.2+ Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/20120126160319.GA5655@quad Signed-off-by: Ingo Molnar

perf events: Add PERF_COUNT_HW_REF_CPU_CYCLES generic PMU event

2011-12-21T09:26:37Z

This event counts the number of reference core cpu cycles. Reference means that the event increments at a constant rate which is not subject to core CPU frequency adjustments. The event may not count when the processor is in halted (low power) state. As such, it may not be equivalent to wall clock time. However, when the processor is not halted state, the event keeps a constant correlation with wall clock time. Signed-off-by: Stephane Eranian Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/1323559734-3488-3-git-send-email-eranian@google.com Signed-off-by: Ingo Molnar

perf, core: Rate limit perf_sched_events jump_label patching

2011-12-06T07:34:02Z

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: Gleb Natapov Acked-by: Jason Baron Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/20111127155909.GO2557@redhat.com Signed-off-by: Ingo Molnar

perf: Avoid a useless pmu_disable() in the perf-tick

2011-12-06T07:33:52Z

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: Gleb Natapov Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/n/tip-515yhoatehd3gza7we9fapaa@git.kernel.org Signed-off-by: Ingo Molnar

perf: Fix loss of notification with multi-event

2011-12-05T08:33:03Z

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 Finished-by: Stephane Eranian Reviewed-by: Stephane Eranian Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/20111126014731.GA7030@quad Signed-off-by: Ingo Molnar

perf, core: Introduce attrs to count in either host or guest mode

2011-10-06T11:00:28Z

The two new attributes exclude_guest and exclude_host can bes used by user-space to tell the kernel to setup performance counter to either only count while the CPU is in guest or in host mode. An additional check is also introduced to make sure user-space does not try to exclude guest and host mode from counting. Signed-off-by: Joerg Roedel Signed-off-by: Gleb Natapov Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/1317816084-18026-2-git-send-email-gleb@redhat.com Signed-off-by: Ingo Molnar

perf events: Fix slow and broken cgroup context switch code

2011-08-29T10:28:33Z

The current cgroup context switch code was incorrect leading to bogus counts. Furthermore, as soon as there was an active cgroup event on a CPU, the context switch cost on that CPU would increase by a significant amount as demonstrated by a simple ping/pong example: $ ./pong Both processes pinned to CPU1, running for 10s 10684.51 ctxsw/s Now start a cgroup perf stat: $ perf stat -e cycles,cycles -A -a -G test -C 1 -- sleep 100 $ ./pong Both processes pinned to CPU1, running for 10s 6674.61 ctxsw/s That's a 37% penalty. Note that pong is not even in the monitored cgroup. The results shown by perf stat are bogus: $ perf stat -e cycles,cycles -A -a -G test -C 1 -- sleep 100 Performance counter stats for 'sleep 100': CPU1 cycles test CPU1 16,984,189,138 cycles # 0.000 GHz The second 'cycles' event should report a count @ CPU clock (here 2.4GHz) as it is counting across all cgroups. The patch below fixes the bogus accounting and bypasses any cgroup switches in case the outgoing and incoming tasks are in the same cgroup. With this patch the same test now yields: $ ./pong Both processes pinned to CPU1, running for 10s 10775.30 ctxsw/s Start perf stat with cgroup: $ perf stat -e cycles,cycles -A -a -G test -C 1 -- sleep 10 Run pong outside the cgroup: $ /pong Both processes pinned to CPU1, running for 10s 10687.80 ctxsw/s The penalty is now less than 2%. And the results for perf stat are correct: $ perf stat -e cycles,cycles -A -a -G test -C 1 -- sleep 10 Performance counter stats for 'sleep 10': CPU1 cycles test # 0.000 GHz CPU1 23,933,981,448 cycles # 0.000 GHz Now perf stat reports the correct counts for for the non cgroup event. If we run pong inside the cgroup, then we also get the correct counts: $ perf stat -e cycles,cycles -A -a -G test -C 1 -- sleep 10 Performance counter stats for 'sleep 10': CPU1 22,297,726,205 cycles test # 0.000 GHz CPU1 23,933,981,448 cycles # 0.000 GHz 10.001457237 seconds time elapsed Signed-off-by: Stephane Eranian Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/20110825135803.GA4697@quad Signed-off-by: Ingo Molnar

atomic: use

2011-07-26T23:49:47Z

This allows us to move duplicated code in (atomic_inc_not_zero() for now) to Signed-off-by: Arun Sharma Reviewed-by: Eric Dumazet Cc: Ingo Molnar Cc: David Miller Cc: Eric Dumazet Acked-by: Mike Frysinger Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds

perf: export perf_event_refresh() to modules

2011-07-01T09:06:40Z

KVM needs one-shot samples, since a PMC programmed to -X will fire after X events and then again after 2^40 events (i.e. variable period). Signed-off-by: Avi Kivity Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/1309362157-6596-4-git-send-email-avi@redhat.com Signed-off-by: Ingo Molnar

perf: Add context field to perf_event

2011-07-01T09:06:38Z

The perf_event overflow handler does not receive any caller-derived argument, so many callers need to resort to looking up the perf_event in their local data structure. This is ugly and doesn't scale if a single callback services many perf_events. Fix by adding a context parameter to perf_event_create_kernel_counter() (and derived hardware breakpoints APIs) and storing it in the perf_event. The field can be accessed from the callback as event->overflow_handler_context. All callers are updated. Signed-off-by: Avi Kivity Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/1309362157-6596-2-git-send-email-avi@redhat.com Signed-off-by: Ingo Molnar