[PATCH] perf tool: report user-friendly error from timechart

[PATCH] perf tool: report user-friendly error from timechart

[Ramkumar Ramachandra]

Earlier, when 'timechart record' was invoked by a user without
permissions to debugfs:

$ perf timechart record
invalid or unsupported event: 'power:cpu_frequency'
Run 'perf list' for a list of valid events

 usage: perf record [<options>] [<command>]
    or: perf record [<options>] -- <command> [<options>]
[... detailed usage information ...]

This is highly user unfriendly, as the real problem is not reported to
the user correctly. Now, the user gets a much more friendly:

$ perf timechart record
Error:  No permissions to read $debugfs/tracing/events/power/cpu_frequency
Hint:   Change the permissions of debugfs: /sys/kernel/debug
        The directory will be present if your kernel was compiled with debugfs support.

Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com>
---
 tools/perf/builtin-timechart.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/tools/perf/builtin-timechart.c b/tools/perf/builtin-timechart.c
index c2e0231..ceaf9b8 100644
--- a/tools/perf/builtin-timechart.c
+++ b/tools/perf/builtin-timechart.c
@@ -1056,6 +1056,14 @@ static int __cmd_record(int argc, const char **argv)
 	}
 #endif
 
+	/* Perform a quick sanity check */
+	if (!is_valid_tracepoint("power:cpu_frequency")) {
+		fprintf(stderr, "Error:\tNo permissions to read $debugfs/tracing/events/power/cpu_frequency\n");
+		fprintf(stderr, "Hint:\tChange the permissions of debugfs: /sys/kernel/debug\n");
+		fprintf(stderr, "\tThe directory will be present if your kernel was compiled with debugfs support.\n");
+		return 1;
+	}
+
 	rec_argc = record_elems + argc - 1;
 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
 
-- 
1.8.4.477.g5d89aa9

Re: [PATCH] perf tool: report user-friendly error from timechart

[Ingo Molnar]

* Ramkumar Ramachandra <artagnon@gmail.com> wrote:

> +	/* Perform a quick sanity check */
> +	if (!is_valid_tracepoint("power:cpu_frequency")) {
> +		fprintf(stderr, "Error:\tNo permissions to read $debugfs/tracing/events/power/cpu_frequency\n");
> +		fprintf(stderr, "Hint:\tChange the permissions of debugfs: /sys/kernel/debug\n");
> +		fprintf(stderr, "\tThe directory will be present if your kernel was compiled with debugfs support.\n");

Is missing permissions the only way how is_valid_tracepoint() can fail?

What if debugfs has the right permissions but CONFIG_TRACEPOINTS is 
disabled in the kernel?

Thanks,

	Ingo

Re: [PATCH] perf tool: report user-friendly error from timechart

[Ramkumar Ramachandra]

Ingo Molnar wrote:
>> +     /* Perform a quick sanity check */
>> +     if (!is_valid_tracepoint("power:cpu_frequency")) {
>> +             fprintf(stderr, "Error:\tNo permissions to read $debugfs/tracing/events/power/cpu_frequency\n");
>> +             fprintf(stderr, "Hint:\tChange the permissions of debugfs: /sys/kernel/debug\n");
>> +             fprintf(stderr, "\tThe directory will be present if your kernel was compiled with debugfs support.\n");
>
> Is missing permissions the only way how is_valid_tracepoint() can fail?
>
> What if debugfs has the right permissions but CONFIG_TRACEPOINTS is
> disabled in the kernel?

I'm thinking about all this in terms of files present in debugfs. Will
the cpu_frequecy file be present if tracepoints is disabled? (should I
quickly check using User-Mode Linux?). If it won't be present, then we
can just change the last line to "compiled with debugfs and
tracepoints support".

Re: [PATCH] perf tool: report user-friendly error from timechart

[Ingo Molnar]

* Ramkumar Ramachandra <artagnon@gmail.com> wrote:

> Ingo Molnar wrote:
> >> +     /* Perform a quick sanity check */
> >> +     if (!is_valid_tracepoint("power:cpu_frequency")) {
> >> +             fprintf(stderr, "Error:\tNo permissions to read $debugfs/tracing/events/power/cpu_frequency\n");
> >> +             fprintf(stderr, "Hint:\tChange the permissions of debugfs: /sys/kernel/debug\n");
> >> +             fprintf(stderr, "\tThe directory will be present if your kernel was compiled with debugfs support.\n");
> >
> > Is missing permissions the only way how is_valid_tracepoint() can fail?
> >
> > What if debugfs has the right permissions but CONFIG_TRACEPOINTS is
> > disabled in the kernel?
> 
> I'm thinking about all this in terms of files present in debugfs. Will 
> the cpu_frequecy file be present if tracepoints is disabled? (should I 
> quickly check using User-Mode Linux?). If it won't be present, then we 
> can just change the last line to "compiled with debugfs and tracepoints 
> support".

The debug/tracing/events directory won't be present if tracing is 
disabled.

Thanks,

	Ingo

Re: [PATCH] perf tool: report user-friendly error from timechart

[David Ahern]

On 10/3/13 6:38 AM, Ingo Molnar wrote:
>
> * Ramkumar Ramachandra <artagnon@gmail.com> wrote:
>
>> +	/* Perform a quick sanity check */
>> +	if (!is_valid_tracepoint("power:cpu_frequency")) {
>> +		fprintf(stderr, "Error:\tNo permissions to read $debugfs/tracing/events/power/cpu_frequency\n");
>> +		fprintf(stderr, "Hint:\tChange the permissions of debugfs: /sys/kernel/debug\n");
>> +		fprintf(stderr, "\tThe directory will be present if your kernel was compiled with debugfs support.\n");
>
> Is missing permissions the only way how is_valid_tracepoint() can fail?
>
> What if debugfs has the right permissions but CONFIG_TRACEPOINTS is
> disabled in the kernel?

There are a number of reasons that function can fail. The complete 
solution is to plumb various error numbers and on failure request a 
string for that failure. Take a look at util/target.[ch] as an example.

The comment applies to the perf-trace patch as well, but it gets more 
complicated to handle the error paths from perf_evsel__newtp when they 
dip into the tracepoint code

David

Re: [PATCH] perf tool: report user-friendly error from timechart

[Arnaldo Carvalho de Melo]

Em Thu, Oct 03, 2013 at 07:35:18AM -0600, David Ahern escreveu:
> On 10/3/13 6:38 AM, Ingo Molnar wrote:
> >
> >* Ramkumar Ramachandra <artagnon@gmail.com> wrote:
> >
> >>+	/* Perform a quick sanity check */
> >>+	if (!is_valid_tracepoint("power:cpu_frequency")) {
> >>+		fprintf(stderr, "Error:\tNo permissions to read $debugfs/tracing/events/power/cpu_frequency\n");
> >>+		fprintf(stderr, "Hint:\tChange the permissions of debugfs: /sys/kernel/debug\n");
> >>+		fprintf(stderr, "\tThe directory will be present if your kernel was compiled with debugfs support.\n");
> >
> >Is missing permissions the only way how is_valid_tracepoint() can fail?
> >
> >What if debugfs has the right permissions but CONFIG_TRACEPOINTS is
> >disabled in the kernel?
> 
> There are a number of reasons that function can fail. The complete
> solution is to plumb various error numbers and on failure request a
> string for that failure. Take a look at util/target.[ch] as an
> example.
> 
> The comment applies to the perf-trace patch as well, but it gets
> more complicated to handle the error paths from perf_evsel__newtp
> when they dip into the tracepoint code

See the patch I posted, in that case we can use the old errno way, i.e.
do nothing and just look at it in the perf_evsel__newtp/
perf_evlist__add_newtp callers.

And is_valid_tracepoint() is a too big hammer, it traverses the whole
directory looking for a match instead of plain build the path and do an
access, its one of those things I need to ditch at some point. So far I
just try to do a perf_evlist__add_newtp and if it fails, look at errno.

- Arnaldo

Re: [PATCH] perf tool: report user-friendly error from timechart

[David Ahern]

On 10/3/13 11:08 AM, Arnaldo Carvalho de Melo wrote:

> And is_valid_tracepoint() is a too big hammer, it traverses the whole
> directory looking for a match instead of plain build the path and do an
> access, its one of those things I need to ditch at some point. So far I
> just try to do a perf_evlist__add_newtp and if it fails, look at errno.

And a number of tracepoints depend on CONFIG settings so it is more than 
just looking at whether debugfs is mounted and accessible/readable. 
e.g., look at the message I put into perf-lock

David