Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <9d9bfd6f-087b-88dc-1b7e-ada26a9318fa@intel.com>
Date:   Mon, 19 Dec 2022 15:19:52 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Firefox/102.0 Thunderbird/102.6.0
Subject: Re: [PATCH] perf tool: Fix output unexpected messages in quiet mode
Content-Language: en-US
To:     Yang Jihong <yangjihong1@huawei.com>, peterz@infradead.org,
        mingo@redhat.com, acme@kernel.org, mark.rutland@arm.com,
        alexander.shishkin@linux.intel.com, jolsa@kernel.org,
        namhyung@kernel.org, irogers@google.com, kan.liang@linux.intel.com,
        kim.phillips@amd.com, german.gomez@arm.com, ravi.bangoria@amd.com,
        linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org
References: <20221219024911.62741-1-yangjihong1@huawei.com>
 <194c676b-3a73-7f34-1f29-a157b5e4731f@intel.com>
 <7aecae79-5b6d-09ea-dfd5-4d4810875f3d@huawei.com>
 <d2189ccc-26aa-250a-ad43-59d53ef01d5e@intel.com>
 <0e06416c-fac3-1149-2660-9a1981e3460d@huawei.com>
From:   Adrian Hunter <adrian.hunter@intel.com>
Organization: Intel Finland Oy, Registered Address: PL 281, 00181 Helsinki,
 Business Identity Code: 0357606 - 4, Domiciled in Helsinki
In-Reply-To: <0e06416c-fac3-1149-2660-9a1981e3460d@huawei.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 19/12/22 15:14, Yang Jihong wrote:
> Hello,
> 
> On 2022/12/19 18:28, Adrian Hunter wrote:
>> On 19/12/22 11:28, Yang Jihong wrote:
>>> Hello,
>>>
>>> On 2022/12/19 14:59, Adrian Hunter wrote:
>>>> On 19/12/22 04:49, Yang Jihong wrote:
>>>>> When perf uses quiet mode, perf_quiet_option sets debug_peo_args to -1,
>>>>
>>>> Seems like redirect_to_stderr has similar issue?
>>> The redirect_to_stderr is used only in the veprintf function:
>>>
>>>    int veprintf(int level, int var, const char *fmt, va_list args)
>>>    {
>>>            int ret = 0;
>>>
>>>            if (var >= level) {
>>>                    if (use_browser >= 1 && redirect_to_stderr <= 0) {
>>>                            ui_helpline__vshow(fmt, args);
>>>                    } else {
>>>                            ret = fprintf_time(debug_file);
>>>                            ret += vfprintf(debug_file, fmt, args);
>>>                    }
>>>            }
>>>
>>>            return ret;
>>>    }
>>>
>>> If use quiet mode, verbose sets to -1. (also assigned in the perf_quiet_option function)
>>> Because "var >= level" is false, veprintf function returns directly, which avoids this problem.
>>>
>>> However, there are cases where:
>>>    # perf --debug stderr=-1 report -vvv 2>/tmp/debug
>>>
>>> If stderr is -1, should we redirect pr_debug in this case?
>>
>> tools/perf/Documentation/perf.txt says debug variables are in value
>> range (0, 10), so -1 is invalid anyway.
>>
> Okay, I see.
>>>
>>> Because I'm not sure if this is a problem,
>>> if redirect_to_stderr needs to be fixed as well,
>>> let me know and I'll submit a patch to fix it.
>>
>> perf_quiet_option() sets redirect_to_stderr = -1 with the intention
>> that it turns it off, but it doesn't, although it gets turned of by
>> verbose = -1. Perhaps set redirect_to_stderr = 0 in perf_quiet_option() ?
>>
> I think it's better to fix redirect_to_stderr and debug_peo_args in the same way (because they're similar):
> 
> Solution A:
>   diff --git a/tools/perf/util/debug.c b/tools/perf/util/debug.c
>   index 65e6c22f38e4..908b26e579e5 100644
>   --- a/tools/perf/util/debug.c
>   +++ b/tools/perf/util/debug.c
>   @@ -68,7 +68,7 @@ int veprintf(int level, int var, const char *fmt, va_list args)
>           int ret = 0;
> 
>           if (var >= level) {
>   -               if (use_browser >= 1 && !redirect_to_stderr) {
>   +               if (use_browser >= 1 && redirect_to_stderr <= 0) {
>                           ui_helpline__vshow(fmt, args);
>                   } else {
>                           ret = fprintf_time(debug_file);
>   diff --git a/tools/perf/util/debug.h b/tools/perf/util/debug.h
>   index f99468a7f681..f6ab84c93ec0 100644
>   --- a/tools/perf/util/debug.h
>   +++ b/tools/perf/util/debug.h
>   @@ -41,7 +41,7 @@ extern int debug_data_convert;
> 
>    /* Special macro to print perf_event_open arguments/return value. */
>    #define pr_debug2_peo(fmt, ...) {                              \
>   -       if (debug_peo_args)                                             \
>   +       if (debug_peo_args > 0)                                         \
>                   pr_debugN(0, pr_fmt(fmt), ##__VA_ARGS__);       \
>           else                                                    \
>                   pr_debugN(2, pr_fmt(fmt), ##__VA_ARGS__);       \
>   diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
>   index 999dd1700502..45cf144c5d5d 100644
>   --- a/tools/perf/util/evsel.c
>   +++ b/tools/perf/util/evsel.c
>   @@ -1775,7 +1775,7 @@ static int __open_attr__fprintf(FILE *fp, const char *name, const char *val,
> 
>    static void display_attr(struct perf_event_attr *attr)
>    {
>   -       if (verbose >= 2 || debug_peo_args) {
>   +       if (verbose >= 2 || debug_peo_args > 0) {
>                   fprintf(stderr, "%.60s\n", graph_dotted_line);
>                   fprintf(stderr, "perf_event_attr:\n");
>                   perf_event_attr__fprintf(stderr, attr, __open_attr__fprintf, NULL);
> 
> or
> Solution B:
> 
>   diff --git a/tools/perf/util/debug.c b/tools/perf/util/debug.c
>   index 908b26e579e5..e3acc213edd1 100644
>   --- a/tools/perf/util/debug.c
>   +++ b/tools/perf/util/debug.c
>   @@ -241,6 +241,9 @@ int perf_quiet_option(void)
>                   opt++;
>           }
> 
>   +       redirect_to_stderr = 0;
>   +       debug_peo_args = 0;
>   +

That seems good to me.

> 
>> I see we have another problem as well: places that check "if (verbose)"
>> instead of "if (verbose > 0)"
> Yes,  places that "if (verbose)" also have problems, I'll submit a patch to change them to "if (verbose > 0)"

Thank you!

> 
> In addition, I found another problem, perf lock/stat/probe does not call the perf_quiet_option function when quiet is true.
> 
> Thanks,
> Yang