linux-kernel - Re: [PATCH v2 2/2] perf stat: Stop repeating when ref_perf

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: <4181eab7-9d86-4043-bff8-1be898946b12@linaro.org>
Date: Fri, 13 Sep 2024 09:45:46 +0100
From: James Clark <james.clark@...aro.org>
To: Levi Yun <yeoreum.yun@....com>
Cc: nd@....com, linux-perf-users@...r.kernel.org,
 linux-kernel@...r.kernel.org, peterz@...radead.org, mingo@...hat.com,
 acme@...nel.org, namhyung@...nel.org, mark.rutland@....com,
 alexander.shishkin@...ux.intel.com, jolsa@...nel.org, irogers@...gle.com,
 adrian.hunter@...el.com, asmadeus@...ewreck.org
Subject: Re: [PATCH v2 2/2] perf stat: Stop repeating when ref_perf_stat()
 returns -1



On 13/09/2024 09:36, James Clark wrote:
> 
> 
> On 13/09/2024 03:02, Levi Yun wrote:
>> Exit when run_perf_stat() returns an error to avoid continuously
>> repeating the same error message. It's not expected that COUNTER_FATAL
>> or internal errors are recoverable so there's no point in retrying.
>>
>> This fixes the following flood of error messages for permission issues,
>> for example when perf_event_paranoid==3:
>>    perf stat -r 1044 -- false
>>
>>    Error:
>>    Access to performance monitoring and observability operations is 
>> limited.
>>    ...
>>    Error:
>>    Access to performance monitoring and observability operations is 
>> limited.
>>    ...
>>    (repeating for 1044 times).
>>
>> Signed-off-by: Levi Yun <yeoreum.yun@....com>
>> ---
>> Changes in v2:
>>    - Add some comments.
>> ---
>>   tools/perf/builtin-stat.c | 10 +++++++++-
>>   1 file changed, 9 insertions(+), 1 deletion(-)
>>
>> diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
>> index 954eb37ce7b8..0153925f2382 100644
>> --- a/tools/perf/builtin-stat.c
>> +++ b/tools/perf/builtin-stat.c
>> @@ -2875,7 +2875,15 @@ int cmd_stat(int argc, const char **argv)
>>               evlist__reset_prev_raw_counts(evsel_list);
>>
>>           status = run_perf_stat(argc, argv, run_idx);
>> -        if (forever && status != -1 && !interval) {
>> +        /*
>> +         * * Meet COUNTER_FATAL situation (i.e) can't open event 
>> counter.
>> +         * * In this case, there is a high chance of failure in the 
>> next attempt
>> +         * * as well with the same reason. so, stop it.
>> +         * */
> 
> There's something wrong with the formatting here.
> 
> But I don't think the comment answers my question about the other return 
> codes. It just states what the code does.
> 
> There are many more return -1's than just for COUNTER_FATAL, so it's not 
> just that situation anyway. And in addition to that, there is -ENOMEM 
> and others that aren't -1 which aren't explained that they are 
> deliberately explicit retry or ignores.
> 

If I'm understanding what I think it means, what about something like this:

/*
  * Returns -1 for fatal errors which signifies to not continue
  * when in repeat mode.
  *
  * Returns < -1 error codes when stat record is used. These
  * result in the stat information being displayed, but writing
  * to the file fails and is non fatal.
  */
static int __run_perf_stat(int argc, const char **argv, int run_idx)
{