[PATCH] tools/perf probe: Print a hint if adding a probe fails

[PATCH] tools/perf probe: Print a hint if adding a probe fails

[Naveen N. Rao]

Adding a probe can fail in a few scenarios. perf already checks for the
address in the kprobe blacklist. However, the address could still be a
jump label, or have a BUG_ON(). In such cases, it isn't always evident
why adding the probe failed. Add a hint so that the user knows how to
proceed.

Reported-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
---
 tools/perf/builtin-probe.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/tools/perf/builtin-probe.c b/tools/perf/builtin-probe.c
index 6b150756677014..ff9f3fdce600dd 100644
--- a/tools/perf/builtin-probe.c
+++ b/tools/perf/builtin-probe.c
@@ -352,8 +352,11 @@ static int perf_add_probe_events(struct perf_probe_event *pevs, int npevs)
 	}
 
 	ret = apply_perf_probe_events(pevs, npevs);
-	if (ret < 0)
+	if (ret < 0) {
+		pr_info("Hint: Check dmesg to understand reason for probe failure.\n"
+			"      Consider probing at the next/previous instruction.\n");
 		goto out_cleanup;
+	}
 
 	for (i = k = 0; i < npevs; i++)
 		k += pevs[i].ntevs;

base-commit: 0808b3d5b7514dc856178dbc509929329bbf301d
-- 
2.31.1

Re: [PATCH] tools/perf probe: Print a hint if adding a probe fails

[Masami Hiramatsu]

Hi Naveen,

On Thu, 10 Jun 2021 15:14:42 +0530
"Naveen N. Rao" <naveen.n.rao@linux.vnet.ibm.com> wrote:

> Adding a probe can fail in a few scenarios. perf already checks for the
> address in the kprobe blacklist. However, the address could still be a
> jump label, or have a BUG_ON(). In such cases, it isn't always evident
> why adding the probe failed. Add a hint so that the user knows how to
> proceed.
> 

Thanks for the report.

Since now there is <tracefs>/error_log, if you see any errors in registering
probe-events, perf probe should dump the error_log for the hint message.
Also, kprobes should return the correct different error code for each
errors.

Thank you,

> Reported-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
> Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
> ---
>  tools/perf/builtin-probe.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/tools/perf/builtin-probe.c b/tools/perf/builtin-probe.c
> index 6b150756677014..ff9f3fdce600dd 100644
> --- a/tools/perf/builtin-probe.c
> +++ b/tools/perf/builtin-probe.c
> @@ -352,8 +352,11 @@ static int perf_add_probe_events(struct perf_probe_event *pevs, int npevs)
>  	}
>  
>  	ret = apply_perf_probe_events(pevs, npevs);
> -	if (ret < 0)
> +	if (ret < 0) {
> +		pr_info("Hint: Check dmesg to understand reason for probe failure.\n"
> +			"      Consider probing at the next/previous instruction.\n");
>  		goto out_cleanup;
> +	}
>  
>  	for (i = k = 0; i < npevs; i++)
>  		k += pevs[i].ntevs;
> 
> base-commit: 0808b3d5b7514dc856178dbc509929329bbf301d
> -- 
> 2.31.1
> 


-- 
Masami Hiramatsu <mhiramat@kernel.org>

Re: [PATCH] tools/perf probe: Print a hint if adding a probe fails

[Arnaldo Carvalho de Melo]

Em Thu, Jun 10, 2021 at 07:29:26PM +0900, Masami Hiramatsu escreveu:
> Hi Naveen,
> 
> On Thu, 10 Jun 2021 15:14:42 +0530
> "Naveen N. Rao" <naveen.n.rao@linux.vnet.ibm.com> wrote:
> 
> > Adding a probe can fail in a few scenarios. perf already checks for the
> > address in the kprobe blacklist. However, the address could still be a
> > jump label, or have a BUG_ON(). In such cases, it isn't always evident
> > why adding the probe failed. Add a hint so that the user knows how to
> > proceed.
> > 
> 
> Thanks for the report.
> 
> Since now there is <tracefs>/error_log, if you see any errors in registering
> probe-events, perf probe should dump the error_log for the hint message.
> Also, kprobes should return the correct different error code for each
> errors.

Was there any followup on this? I think we should do as Masami suggests,
Naveen, could you do it?

Thanks,

- Arnaldo
 
> Thank you,
> 
> > Reported-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
> > Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
> > ---
> >  tools/perf/builtin-probe.c | 5 ++++-
> >  1 file changed, 4 insertions(+), 1 deletion(-)
> > 
> > diff --git a/tools/perf/builtin-probe.c b/tools/perf/builtin-probe.c
> > index 6b150756677014..ff9f3fdce600dd 100644
> > --- a/tools/perf/builtin-probe.c
> > +++ b/tools/perf/builtin-probe.c
> > @@ -352,8 +352,11 @@ static int perf_add_probe_events(struct perf_probe_event *pevs, int npevs)
> >  	}
> >  
> >  	ret = apply_perf_probe_events(pevs, npevs);
> > -	if (ret < 0)
> > +	if (ret < 0) {
> > +		pr_info("Hint: Check dmesg to understand reason for probe failure.\n"
> > +			"      Consider probing at the next/previous instruction.\n");
> >  		goto out_cleanup;
> > +	}
> >  
> >  	for (i = k = 0; i < npevs; i++)
> >  		k += pevs[i].ntevs;
> > 
> > base-commit: 0808b3d5b7514dc856178dbc509929329bbf301d
> > -- 
> > 2.31.1
> > 
> 
> 
> -- 
> Masami Hiramatsu <mhiramat@kernel.org>

-- 

- Arnaldo

Re: [PATCH] tools/perf probe: Print a hint if adding a probe fails

[Naveen N. Rao]

Hi Arnaldo,

Arnaldo Carvalho de Melo wrote:
> Em Thu, Jun 10, 2021 at 07:29:26PM +0900, Masami Hiramatsu escreveu:
>> Hi Naveen,
>> 
>> On Thu, 10 Jun 2021 15:14:42 +0530
>> "Naveen N. Rao" <naveen.n.rao@linux.vnet.ibm.com> wrote:
>> 
>> > Adding a probe can fail in a few scenarios. perf already checks for the
>> > address in the kprobe blacklist. However, the address could still be a
>> > jump label, or have a BUG_ON(). In such cases, it isn't always evident
>> > why adding the probe failed. Add a hint so that the user knows how to
>> > proceed.
>> > 
>> 
>> Thanks for the report.
>> 
>> Since now there is <tracefs>/error_log, if you see any errors in registering
>> probe-events, perf probe should dump the error_log for the hint message.
>> Also, kprobes should return the correct different error code for each
>> errors.
> 
> Was there any followup on this? I think we should do as Masami suggests,
> Naveen, could you do it?

Thanks for checking. We have been discussing this on the kernel side of 
things here (sorry, missed copying you on that):
http://lkml.kernel.org/r/20210610085617.1590138-1-naveen.n.rao@linux.vnet.ibm.com

I will work on the related changes to perf after the kernel changes.


Thanks,
Naveen