linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: "Liang, Kan" <kan.liang@linux.intel.com>
To: Peter Zijlstra <peterz@infradead.org>
Cc: Ian Rogers <irogers@google.com>, Ingo Molnar <mingo@redhat.com>,
	Arnaldo Carvalho de Melo <acme@kernel.org>,
	Mark Rutland <mark.rutland@arm.com>,
	Alexander Shishkin <alexander.shishkin@linux.intel.com>,
	Jiri Olsa <jolsa@redhat.com>, Namhyung Kim <namhyung@kernel.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Masahiro Yamada <yamada.masahiro@socionext.com>,
	Kees Cook <keescook@chromium.org>,
	Catalin Marinas <catalin.marinas@arm.com>,
	Petr Mladek <pmladek@suse.com>,
	Mauro Carvalho Chehab <mchehab+samsung@kernel.org>,
	Qian Cai <cai@lca.pw>, Joe Lawrence <joe.lawrence@redhat.com>,
	Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>,
	Sri Krishna chowdary <schowdary@nvidia.com>,
	"Uladzislau Rezki (Sony)" <urezki@gmail.com>,
	Andy Shevchenko <andriy.shevchenko@linux.intel.com>,
	Changbin Du <changbin.du@intel.com>,
	Ard Biesheuvel <ardb@kernel.org>,
	"David S. Miller" <davem@davemloft.net>,
	Kent Overstreet <kent.overstreet@gmail.com>,
	Gary Hook <Gary.Hook@amd.com>, Arnd Bergmann <arnd@arndb.de>,
	linux-kernel@vger.kernel.org,
	Stephane Eranian <eranian@google.com>,
	Andi Kleen <ak@linux.intel.com>
Subject: Re: [PATCH v3 10/10] perf/cgroup: Do not switch system-wide events in cgroup switch
Date: Thu, 14 Nov 2019 15:49:57 -0500	[thread overview]
Message-ID: <94c8c876-f236-7052-24ef-536f6870a8d5@linux.intel.com> (raw)
In-Reply-To: <4bc51bf9-1d47-063a-e811-d05fb42c8838@linux.intel.com>



On 11/14/2019 10:24 AM, Liang, Kan wrote:
> 
> 
> On 11/14/2019 10:16 AM, Liang, Kan wrote:
>>
>>
>> On 11/14/2019 8:57 AM, Peter Zijlstra wrote:
>>> On Thu, Nov 14, 2019 at 08:46:51AM -0500, Liang, Kan wrote:
>>>>
>>>>
>>>> On 11/14/2019 5:43 AM, Peter Zijlstra wrote:
>>>>> On Wed, Nov 13, 2019 at 04:30:42PM -0800, Ian Rogers wrote:
>>>>>> From: Kan Liang <kan.liang@linux.intel.com>
>>>>>>
>>>>>> When counting system-wide events and cgroup events simultaneously, 
>>>>>> the
>>>>>> system-wide events are always scheduled out then back in during 
>>>>>> cgroup
>>>>>> switches, bringing extra overhead and possibly missing events. 
>>>>>> Switching
>>>>>> out system wide flexible events may be necessary if the scheduled in
>>>>>> task's cgroups have pinned events that need to be scheduled in at 
>>>>>> a higher
>>>>>> priority than the system wide flexible events.
>>>>>
>>>>> I'm thinking this patch is actively broken. groups->index 'group' wide
>>>>> and therefore across cpu/cgroup boundaries.
>>>>>
>>>>> There is no !cgroup to cgroup hierarchy as this patch seems to assume,
>>>>> specifically look at how the merge sort in visit_groups_merge() allows
>>>>> cgroup events to be picked before !cgroup events.
>>>>
>>>>
>>>> No, the patch intends to avoid switch !cgroup during cgroup context 
>>>> switch.
>>>
>>> Which is wrong.
>>>
>> Why we want to switch !cgroup system-wide event in context switch?
>>
>> How should current perf handle this case?
> 

It seems hard to find a simple case to explain why we should not switch 
!cgroup during cgroup context switch.

Let me try to explain it using ftrace.

Case 1:
User A do system-wide monitoring for 1 second. No other users.
      #perf stat -e branches -a -- sleep 1

The counter counts between 765531.617703 and 765532.620184.
Everything is collected.

            <...>-59160 [027] d.h. 765531.617697: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 0
            <...>-59160 [027] d.h. 765531.617701: write_msr: 
MSR_IA32_PMC0(4c1), value 800000000001
            <...>-59160 [027] d.h. 765531.617702: write_msr: 
MSR_P6_EVNTSEL0(186), value 5300c4
            <...>-59160 [027] d.h. 765531.617703: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 70000000f
           <idle>-0     [027] d.h. 765532.620184: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 0
           <idle>-0     [027] d.h. 765532.620185: write_msr: 
MSR_P6_EVNTSEL0(186), value 1300c4
           <idle>-0     [027] d.h. 765532.620186: rdpmc: 0, value 
80000b3e87a4
           <idle>-0     [027] d.h. 765532.620187: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 70000000f


Case 2:
User A do system-wide monitoring for 1 second.
      #perf stat -e branches -a -- sleep 1
At the meantime, User B do cgroup monitoring.
      #perf stat -e cycles -G cgroup

The User A expects to collect everything from 765580.196521 to 
765581.198150. But it doesn't.

Because of cgroup context switch, the system-wide event for user A stops 
counting at [765580.213882, 765580.213884],
[765580.213913, 765580.213915], ..., [765580.774304, 765580.774307].

I think it breaks the usage of User A.

Furthermore, switching !cgroup system-wide event also brings extra 
overhead, which is unnecessary.

            <...>-121292 [027] d.h. 765580.196514: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 0
            <...>-121292 [027] d.h. 765580.196519: write_msr: 
MSR_IA32_PMC0(4c1), value 800000000001
            <...>-121292 [027] d.h. 765580.196520: write_msr: 
MSR_P6_EVNTSEL0(186), value 5300c4
            <...>-121292 [027] d.h. 765580.196521: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 70000000f
           <idle>-0     [027] d... 765580.213878: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 0
           <idle>-0     [027] d... 765580.213880: write_msr: 
MSR_P6_EVNTSEL0(186), value 1300c4
           <idle>-0     [027] d... 765580.213880: rdpmc: 0, value 
800000357bc1
           <idle>-0     [027] d... 765580.213882: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 70000000f
      simics-poll-25601 [027] d... 765580.213884: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 0
      simics-poll-25601 [027] d... 765580.213888: write_msr: 
MSR_CORE_PERF_FIXED_CTR1(30a), value 800015820cbe
      simics-poll-25601 [027] d... 765580.213889: read_msr: 
MSR_CORE_PERF_FIXED_CTR_CTRL(38d), value 0
      simics-poll-25601 [027] d... 765580.213890: write_msr: 
MSR_CORE_PERF_FIXED_CTR_CTRL(38d), value b0
      simics-poll-25601 [027] d... 765580.213890: write_msr: 
MSR_IA32_PMC0(4c1), value 800000357bc1
      simics-poll-25601 [027] d... 765580.213891: write_msr: 
MSR_P6_EVNTSEL0(186), value 5300c4
      simics-poll-25601 [027] d... 765580.213892: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 70000000f
      simics-poll-25601 [027] d... 765580.213910: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 0
      simics-poll-25601 [027] d... 765580.213911: read_msr: 
MSR_CORE_PERF_FIXED_CTR_CTRL(38d), value b0
      simics-poll-25601 [027] d... 765580.213911: write_msr: 
MSR_CORE_PERF_FIXED_CTR_CTRL(38d), value 0
      simics-poll-25601 [027] d... 765580.213911: rdpmc: 40000001, value 
80001582b676
      simics-poll-25601 [027] d... 765580.213912: write_msr: 
MSR_P6_EVNTSEL0(186), value 1300c4
      simics-poll-25601 [027] d... 765580.213913: rdpmc: 0, value 
800000358491
      simics-poll-25601 [027] d... 765580.213913: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 70000000f
           <idle>-0     [027] d... 765580.213915: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 0
           <idle>-0     [027] d... 765580.213916: write_msr: 
MSR_IA32_PMC0(4c1), value 800000358491
           <idle>-0     [027] d... 765580.213916: write_msr: 
MSR_P6_EVNTSEL0(186), value 5300c4
           <idle>-0     [027] d... 765580.213917: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 70000000f

... ...

      simics-poll-25601 [027] d... 765580.774301: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 0
      simics-poll-25601 [027] d... 765580.774302: read_msr: 
MSR_CORE_PERF_FIXED_CTR_CTRL(38d), value b0
      simics-poll-25601 [027] d... 765580.774302: write_msr: 
MSR_CORE_PERF_FIXED_CTR_CTRL(38d), value 0
      simics-poll-25601 [027] d... 765580.774302: rdpmc: 40000001, value 
8000165e927b
      simics-poll-25601 [027] d... 765580.774303: write_msr: 
MSR_P6_EVNTSEL0(186), value 1300c4
      simics-poll-25601 [027] d... 765580.774303: rdpmc: 0, value 
8000059298ce
      simics-poll-25601 [027] d... 765580.774304: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 70000000f
            <...>-135379 [027] d... 765580.774307: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 0
            <...>-135379 [027] d... 765580.774308: write_msr: 
MSR_IA32_PMC0(4c1), value 8000059298ce
            <...>-135379 [027] d... 765580.774309: write_msr: 
MSR_P6_EVNTSEL0(186), value 5300c4
            <...>-135379 [027] d... 765580.774309: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 70000000f
            <...>-147127 [027] d.h. 765581.198150: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 0
            <...>-147127 [027] d.h. 765581.198153: write_msr: 
MSR_P6_EVNTSEL0(186), value 1300c4
            <...>-147127 [027] d.h. 765581.198153: rdpmc: 0, value 
80000a573368
            <...>-147127 [027] d.h. 765581.198155: write_msr: 
MSR_CORE_PERF_GLOBAL_CTRL(38f), value 70000000f


Thanks,
Kan






  reply	other threads:[~2019-11-14 20:50 UTC|newest]

Thread overview: 80+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-11-14  0:30 [PATCH v3 00/10] Optimize cgroup context switch Ian Rogers
2019-11-14  0:30 ` [PATCH v3 01/10] perf/cgroup: Reorder perf_cgroup_connect() Ian Rogers
2019-11-14  8:50   ` Peter Zijlstra
2019-11-14  0:30 ` [PATCH v3 02/10] lib: introduce generic min max heap Ian Rogers
2019-11-14  9:32   ` Peter Zijlstra
2019-11-14  9:35   ` Peter Zijlstra
2019-11-17 18:28   ` Joe Perches
2019-11-18  8:40     ` Peter Zijlstra
2019-11-18 11:50       ` Joe Perches
2019-11-18 12:21         ` Peter Zijlstra
2019-11-14  0:30 ` [PATCH v3 03/10] perf: Use min_max_heap in visit_groups_merge Ian Rogers
2019-11-14  9:39   ` Peter Zijlstra
2019-11-14  0:30 ` [PATCH v3 04/10] perf: Add per perf_cpu_context min_heap storage Ian Rogers
2019-11-14  9:51   ` Peter Zijlstra
2019-11-16  1:19     ` Ian Rogers
2019-11-14  0:30 ` [PATCH v3 05/10] perf/cgroup: Grow per perf_cpu_context heap storage Ian Rogers
2019-11-14  9:54   ` Peter Zijlstra
2019-11-14  0:30 ` [PATCH v3 06/10] perf/cgroup: Order events in RB tree by cgroup id Ian Rogers
2019-11-14  0:30 ` [PATCH v3 07/10] perf: simplify and rename visit_groups_merge Ian Rogers
2019-11-14 10:03   ` Peter Zijlstra
2019-11-16  1:20     ` Ian Rogers
2019-11-14  0:30 ` [PATCH v3 08/10] perf: cache perf_event_groups_first for cgroups Ian Rogers
2019-11-14 10:25   ` Peter Zijlstra
2019-11-16  1:20     ` Ian Rogers
2019-11-18  8:37       ` Peter Zijlstra
2019-11-14  0:30 ` [PATCH v3 09/10] perf: optimize event_filter_match during sched_in Ian Rogers
2019-11-14  0:30 ` [PATCH v3 10/10] perf/cgroup: Do not switch system-wide events in cgroup switch Ian Rogers
2019-11-14 10:43   ` Peter Zijlstra
2019-11-14 13:46     ` Liang, Kan
2019-11-14 13:57       ` Peter Zijlstra
2019-11-14 15:16         ` Liang, Kan
2019-11-14 15:24           ` Liang, Kan
2019-11-14 20:49             ` Liang, Kan [this message]
2019-11-14  0:42 ` [PATCH v3 00/10] Optimize cgroup context switch Ian Rogers
2019-11-14 10:45 ` Peter Zijlstra
2019-11-14 18:17   ` Ian Rogers
2019-12-06 23:16     ` Ian Rogers
2019-11-16  1:18 ` [PATCH v4 " Ian Rogers
2019-11-16  1:18   ` [PATCH v4 01/10] perf/cgroup: Reorder perf_cgroup_connect() Ian Rogers
2019-11-16  1:18   ` [PATCH v4 02/10] lib: introduce generic min max heap Ian Rogers
2019-11-21 11:11     ` Joe Perches
2019-11-16  1:18   ` [PATCH v4 03/10] perf: Use min_max_heap in visit_groups_merge Ian Rogers
2019-11-16  1:18   ` [PATCH v4 04/10] perf: Add per perf_cpu_context min_heap storage Ian Rogers
2019-11-16  1:18   ` [PATCH v4 05/10] perf/cgroup: Grow per perf_cpu_context heap storage Ian Rogers
2019-11-16  1:18   ` [PATCH v4 06/10] perf/cgroup: Order events in RB tree by cgroup id Ian Rogers
2019-11-16  1:18   ` [PATCH v4 07/10] perf: simplify and rename visit_groups_merge Ian Rogers
2019-11-16  1:18   ` [PATCH v4 08/10] perf: cache perf_event_groups_first for cgroups Ian Rogers
2019-11-16  1:18   ` [PATCH v4 09/10] perf: optimize event_filter_match during sched_in Ian Rogers
2019-11-16  1:18   ` [PATCH v4 10/10] perf/cgroup: Do not switch system-wide events in cgroup switch Ian Rogers
2019-12-06 23:15   ` [PATCH v5 00/10] Optimize cgroup context switch Ian Rogers
2019-12-06 23:15     ` [PATCH v5 01/10] perf/cgroup: Reorder perf_cgroup_connect() Ian Rogers
2019-12-06 23:15     ` [PATCH v5 02/10] lib: introduce generic min-heap Ian Rogers
2019-12-06 23:15     ` [PATCH v5 03/10] perf: Use min_max_heap in visit_groups_merge Ian Rogers
2019-12-08  7:10       ` kbuild test robot
2019-12-06 23:15     ` [PATCH v5 04/10] perf: Add per perf_cpu_context min_heap storage Ian Rogers
2019-12-06 23:15     ` [PATCH v5 05/10] perf/cgroup: Grow per perf_cpu_context heap storage Ian Rogers
2019-12-06 23:15     ` [PATCH v5 06/10] perf/cgroup: Order events in RB tree by cgroup id Ian Rogers
2019-12-06 23:15     ` [PATCH v5 07/10] perf: simplify and rename visit_groups_merge Ian Rogers
2019-12-06 23:15     ` [PATCH v5 08/10] perf: cache perf_event_groups_first for cgroups Ian Rogers
2019-12-06 23:15     ` [PATCH v5 09/10] perf: optimize event_filter_match during sched_in Ian Rogers
2019-12-06 23:15     ` [PATCH v5 10/10] perf/cgroup: Do not switch system-wide events in cgroup switch Ian Rogers
2020-02-14  7:51     ` [PATCH v6 0/6] Optimize cgroup context switch Ian Rogers
2020-02-14  7:51       ` [PATCH v6 1/6] perf/cgroup: Reorder perf_cgroup_connect() Ian Rogers
2020-02-14 16:11         ` Shuah Khan
2020-02-14 17:37           ` Peter Zijlstra
2020-03-06 14:42         ` [tip: perf/core] " tip-bot2 for Peter Zijlstra
2020-02-14  7:51       ` [PATCH v6 2/6] lib: introduce generic min-heap Ian Rogers
2020-02-14 22:06         ` Randy Dunlap
2020-02-17 16:29         ` Peter Zijlstra
2020-03-06 14:42         ` [tip: perf/core] lib: Introduce " tip-bot2 for Ian Rogers
2020-02-14  7:51       ` [PATCH v6 3/6] perf: Use min_heap in visit_groups_merge Ian Rogers
2020-02-17 17:23         ` Peter Zijlstra
2020-03-06 14:42         ` [tip: perf/core] perf/core: Use min_heap in visit_groups_merge() tip-bot2 for Ian Rogers
2020-02-14  7:51       ` [PATCH v6 4/6] perf: Add per perf_cpu_context min_heap storage Ian Rogers
2020-03-06 14:42         ` [tip: perf/core] perf/core: " tip-bot2 for Ian Rogers
2020-02-14  7:51       ` [PATCH v6 5/6] perf/cgroup: Grow per perf_cpu_context heap storage Ian Rogers
2020-03-06 14:42         ` [tip: perf/core] " tip-bot2 for Ian Rogers
2020-02-14  7:51       ` [PATCH v6 6/6] perf/cgroup: Order events in RB tree by cgroup id Ian Rogers
2020-02-14 19:32       ` [PATCH v6 0/6] Optimize cgroup context switch Ian Rogers
2020-02-17 16:18       ` Peter Zijlstra

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=94c8c876-f236-7052-24ef-536f6870a8d5@linux.intel.com \
    --to=kan.liang@linux.intel.com \
    --cc=Gary.Hook@amd.com \
    --cc=acme@kernel.org \
    --cc=ak@linux.intel.com \
    --cc=akpm@linux-foundation.org \
    --cc=alexander.shishkin@linux.intel.com \
    --cc=andriy.shevchenko@linux.intel.com \
    --cc=ardb@kernel.org \
    --cc=arnd@arndb.de \
    --cc=cai@lca.pw \
    --cc=catalin.marinas@arm.com \
    --cc=changbin.du@intel.com \
    --cc=davem@davemloft.net \
    --cc=eranian@google.com \
    --cc=irogers@google.com \
    --cc=joe.lawrence@redhat.com \
    --cc=jolsa@redhat.com \
    --cc=keescook@chromium.org \
    --cc=kent.overstreet@gmail.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mark.rutland@arm.com \
    --cc=mchehab+samsung@kernel.org \
    --cc=mingo@redhat.com \
    --cc=namhyung@kernel.org \
    --cc=penguin-kernel@i-love.sakura.ne.jp \
    --cc=peterz@infradead.org \
    --cc=pmladek@suse.com \
    --cc=schowdary@nvidia.com \
    --cc=urezki@gmail.com \
    --cc=yamada.masahiro@socionext.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body. 
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).