linux-kernel - Re: perf probe + uprobes missing events

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: <20200407213943.3a92e040d4ce30dc55e9aa1f@kernel.org>
Date:   Tue, 7 Apr 2020 21:39:43 +0900
From:   Masami Hiramatsu <mhiramat@...nel.org>
To:     Arnaldo Carvalho de Melo <acme@...nel.org>
Cc:     Jiri Olsa <jolsa@...nel.org>, Kajol Jain <kjain@...ux.ibm.com>,
        Andi Kleen <andi@...stfloor.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        linux-perf-users@...r.kernel.org
Subject: Re: perf probe + uprobes missing events

Hi Arnaldo,

On Mon, 6 Apr 2020 11:53:56 -0300
Arnaldo Carvalho de Melo <acme@...nel.org> wrote:

> Hi Masami,
> 
> 	I'm trying to use 'perf probe' to debug 'perf test', but I'm not
> getting repeatable results, take a look:
> 
> I'm trying to figure out which expresssions are _really_ being tested
> byu the 'perf test expr' testcase, so I added a probe to the
> expr__parse() routine, asking for the expr string to be printed:
> 
> [root@...e ~]# perf probe -x ~/bin/perf -L expr__parse
> <expr__parse@...me/acme/git/perf/tools/perf/util/expr.c:0>
>       0  int expr__parse(double *final_val, struct expr_parse_ctx *ctx, const char *expr, int runtime)
>       1  {
>       2         return __expr__parse(final_val, ctx, expr, EXPR_PARSE, runtime) ? -1 : 0;
>       3  }
> 
>          static bool
>          already_seen(const char *val, const char *one, const char **other,
> 
> [root@...e ~]#
> 
> [root@...e ~]# perf probe -x ~/bin/perf expr__parse expr:string
> Target program is compiled without optimization. Skipping prologue.
> Probe on address 0x5cb11b to force probing at the function entry.
> 
> Added new event:
>   probe_perf:expr__parse (on expr__parse in /home/acme/bin/perf with expr:string)
> 
> You can now use it in all perf tools, such as:
> 
> 	perf record -e probe_perf:expr__parse -aR sleep 1
> 
> [root@...e ~]#

Hmm, These operation looks good to me. However,

> 
> [root@...e ~]# perf trace -e probe_perf:expr_* perf test -F expr
>  7: Simple expression parser                              : Ok
>      0.000 :2994175/2994175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1+1")
>      0.015 :2994175/2994175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "FOO+BAR")
>      0.018 :2994175/2994175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "(BAR/2)%2")
>      0.020 :2994175/2994175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1 - -4")
>      0.023 :2994175/2994175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "(FOO-1)*2 + (BAR/2)%2 - -4")
>      0.026 :2994175/2994175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1-1 | 1")
>      0.029 :2994175/2994175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1-1 & 1")
>      0.031 :2994175/2994175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "min(1,2) + 1")
>      0.034 :2994175/2994175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "max(1,2) + 1")
>      0.036 :2994175/2994175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1+1 if 3*4 else 0")
>      0.039 :2994175/2994175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "FOO/0")
>      0.041 :2994175/2994175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "BAR/")
> [root@...e ~]# perf trace -e probe_perf:expr_* perf test -F expr
>  7: Simple expression parser                              : Ok
>      0.000 perf/2996042 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1+1")
>      0.021 perf/2996042 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "FOO+BAR")
>      0.025 perf/2996042 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "(BAR/2)%2")
>      0.029 perf/2996042 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1 - -4")
>      0.033 perf/2996042 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "(FOO-1)*2 + (BAR/2)%2 - -4")
>      0.041 perf/2996042 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1-1 | 1")
>      0.044 perf/2996042 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1-1 & 1")
> [root@...e ~]# perf trace -e probe_perf:expr_* perf test -F expr
>  7: Simple expression parser                              : Ok
>      0.000 perf/2996416 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1+1")
>      0.020 perf/2996416 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "FOO+BAR")
>      0.025 perf/2996416 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "(BAR/2)%2")
>      0.029 perf/2996416 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1 - -4")
>      0.032 perf/2996416 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "(FOO-1)*2 + (BAR/2)%2 - -4")
>      0.037 perf/2996416 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1-1 | 1")
>      0.040 perf/2996416 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1-1 & 1")
>      0.043 perf/2996416 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "min(1,2) + 1")
>      0.046 perf/2996416 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "max(1,2) + 1")
>      0.049 perf/2996416 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1+1 if 3*4 else 0")
>      0.053 perf/2996416 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "FOO/0")
>      0.056 perf/2996416 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "BAR/")
> [root@...e ~]# perf trace -e probe_perf:expr_* perf test -F expr
>  7: Simple expression parser                              : Ok
> [root@...e ~]# perf trace -e probe_perf:expr_* perf test -F expr
>  7: Simple expression parser                              : Ok
> [root@...e ~]# perf trace -e probe_perf:expr_* perf test -F expr
>  7: Simple expression parser                              : Ok
> [root@...e ~]# perf trace -e probe_perf:expr_* perf test -F expr
>  7: Simple expression parser                              : Ok
> [root@...e ~]# perf trace -e probe_perf:expr_* perf test -F expr
>  7: Simple expression parser                              : Ok
>      0.000 :2998175/2998175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1+1")
>      0.014 :2998175/2998175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "FOO+BAR")
>      0.017 :2998175/2998175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "(BAR/2)%2")
>      0.021 :2998175/2998175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1 - -4")
>      0.024 :2998175/2998175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "(FOO-1)*2 + (BAR/2)%2 - -4")
>      0.030 :2998175/2998175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1-1 | 1")
>      0.032 :2998175/2998175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1-1 & 1")
>      0.035 :2998175/2998175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "min(1,2) + 1")
>      0.038 :2998175/2998175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "max(1,2) + 1")
>      0.040 :2998175/2998175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "1+1 if 3*4 else 0")
>      0.044 :2998175/2998175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "FOO/0")
>      0.046 :2998175/2998175 probe_perf:expr__parse(__probe_ip: 6074674, expr_string: "BAR/")
> [root@...e ~]#

Strange. This seems bug in uprobes. Did you try to enable ftrace event
and dump ftrace trace buffer for each time?
Also, it may help if you dump the /sys/kernel/debug/tracing/uprobe_profile.
It may indicate how many times the probe is hit and missed.

OK, I'll also try to reproduce it.

Thank you,

> 
> 
> Do you have any idea why that happens?
> 
> - Arnaldo


-- 
Masami Hiramatsu <mhiramat@...nel.org>