linux-kernel - Re: [PATCH v5 5/5] perf-stat: introduce bpf_counter

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <7DBDECAE-D100-44C0-B5D3-DE48631430B5@fb.com>
Date:   Thu, 29 Apr 2021 22:40:01 +0000
From:   Song Liu <songliubraving@...com>
To:     Jiri Olsa <jolsa@...hat.com>
CC:     Song Liu <song@...nel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        Kernel Team <Kernel-team@...com>,
        "acme@...nel.org" <acme@...nel.org>,
        "acme@...hat.com" <acme@...hat.com>,
        "namhyung@...nel.org" <namhyung@...nel.org>,
        "jolsa@...nel.org" <jolsa@...nel.org>
Subject: Re: [PATCH v5 5/5] perf-stat: introduce bpf_counter_ops->disable()



> On Apr 27, 2021, at 12:30 PM, Song Liu <songliubraving@...com> wrote:
> 
> 
> 
>> On Apr 27, 2021, at 5:33 AM, Jiri Olsa <jolsa@...hat.com> wrote:
>> 
>> On Mon, Apr 26, 2021 at 10:18:57PM +0000, Song Liu wrote:
>>> 
>>> 
>>>> On Apr 26, 2021, at 2:27 PM, Jiri Olsa <jolsa@...hat.com> wrote:
>>>> 
>>>> On Sun, Apr 25, 2021 at 02:43:33PM -0700, Song Liu wrote:
>>>> 
>>>> SNIP
>>>> 
>>>>> +static inline int bpf_counter__disable(struct evsel *evsel __maybe_unused)
>>>>> +{
>>>>> +	return 0;
>>>>> +}
>>>>> +
>>>>> static inline int bpf_counter__read(struct evsel *evsel __maybe_unused)
>>>>> {
>>>>> 	return -EAGAIN;
>>>>> diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
>>>>> index d29a8a118973c..e71041c890102 100644
>>>>> --- a/tools/perf/util/evlist.c
>>>>> +++ b/tools/perf/util/evlist.c
>>>>> @@ -17,6 +17,7 @@
>>>>> #include "evsel.h"
>>>>> #include "debug.h"
>>>>> #include "units.h"
>>>>> +#include "bpf_counter.h"
>>>>> #include <internal/lib.h> // page_size
>>>>> #include "affinity.h"
>>>>> #include "../perf.h"
>>>>> @@ -421,6 +422,9 @@ static void __evlist__disable(struct evlist *evlist, char *evsel_name)
>>>>> 	if (affinity__setup(&affinity) < 0)
>>>>> 		return;
>>>>> 
>>>>> +	evlist__for_each_entry(evlist, pos)
>>>>> +		bpf_counter__disable(pos);
>>>> 
>>>> I was wondering why you don't check evsel__is_bpf like
>>>> for the enable case.. and realized that we don't skip
>>>> bpf evsels in __evlist__enable and __evlist__disable
>>>> like we do in read_affinity_counters
>>>> 
>>>> so I guess there's extra affinity setup and bunch of
>>>> wrong ioctls being called?
>>> 
>>> We actually didn't do wrong ioctls because the following check:
>>> 
>>>      if (... || !pos->core.fd)
>>>               continue;
>>> 
>>> in __evlist__enable and __evlist__disable. That we don't allocate 
>>> core.fd for is_bpf events. 
>>> 
>>> It is probably good to be more safe with an extra check of 
>>> evsel__is_bpf(). But it is not required with current code. 
>> 
>> hum, but it will do all the affinity setup no? for no reason,
>> if there's no non-bpb event
> 
> Yes, it will do the affinity setup. Let me see how to get something
> like all_counters_use_bpf here (or within builtin-stat.c).
> 

Would something like the following work? It is not clean (skipping some 
useful logic in __evlist__[enable|disable]). But it seems to work in the
tests.

Thanks,
Song



>From ecb75a1fa747ca5521bcda972840df1e97c09b11 Mon Sep 17 00:00:00 2001
From: Song Liu <song@...nel.org>
Date: Wed, 28 Apr 2021 17:41:28 -0700
Subject: [PATCH] perf-stat: skip evlist__[enable|disable] when all events uses
 BPF

When all events of a perf-stat session use BPF, it is not necessary to
call evlist__enable() and evlist__disable(). Skip them when
all_counters_use_bpf is true.

Signed-off-by: Song Liu <song@...nel.org>
---
 tools/perf/builtin-stat.c | 12 +++++++++---
 tools/perf/util/evlist.c  |  3 ---
 2 files changed, 9 insertions(+), 6 deletions(-)

diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 5a830ae09418e..44459e0352fda 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -572,7 +572,8 @@ static int enable_counters(void)
         * - we have initial delay configured
         */
        if (!target__none(&target) || stat_config.initial_delay) {
-               evlist__enable(evsel_list);
+               if (!all_counters_use_bpf)
+                       evlist__enable(evsel_list);
                if (stat_config.initial_delay > 0)
                        pr_info(EVLIST_ENABLED_MSG);
        }
@@ -581,13 +582,18 @@ static int enable_counters(void)

 static void disable_counters(void)
 {
+       struct evsel *counter;
        /*
         * If we don't have tracee (attaching to task or cpu), counters may
         * still be running. To get accurate group ratios, we must stop groups
         * from counting before reading their constituent counters.
         */
-       if (!target__none(&target))
-               evlist__disable(evsel_list);
+       if (!target__none(&target)) {
+               evlist__for_each_entry(evsel_list, counter)
+                       bpf_counter__disable(counter);
+               if (!all_counters_use_bpf)
+                       evlist__disable(evsel_list);
+       }
 }

 static volatile int workload_exec_errno;
diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 6e5c41528c7d0..6ea3e677dc1e7 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -425,9 +425,6 @@ static void __evlist__disable(struct evlist *evlist, char *evsel_name)
        if (affinity__setup(&affinity) < 0)
                return;

-       evlist__for_each_entry(evlist, pos)
-               bpf_counter__disable(pos);
-
        /* Disable 'immediate' events last */
        for (imm = 0; imm <= 1; imm++) {
                evlist__for_each_cpu(evlist, i, cpu) {
--
2.30.2