Hi,
I was running a couple of tests with perf mem from tip.git and
found that perf mem -t load rep was failing only in TUI mode.
With --stdio, it works fine. I did a bit of investigation and found
out that the TUI mode is failing because some symbol (kernel)
is showing up with a huge (negative) size. I don't understand
the connection between the TUI vs. tty browser and symbols.
To reproduce:
$ perf mem -t load rec my_test_program
$ perf mem -t load rep
$ perf mem -t load rep --stdio
Debugging with TUI is a pain, so I stopped.
I have a patch to improve perf mem a bit but I'd like to get this problem
fixed first.
Maybe you have a idea of what's going on.
Thanks.
On 12/9/14 9:15 PM, Stephane Eranian wrote:
> Hi,
>
> I was running a couple of tests with perf mem from tip.git and
> found that perf mem -t load rep was failing only in TUI mode.
> With --stdio, it works fine. I did a bit of investigation and found
> out that the TUI mode is failing because some symbol (kernel)
> is showing up with a huge (negative) size. I don't understand
> the connection between the TUI vs. tty browser and symbols.
>
> To reproduce:
> $ perf mem -t load rec my_test_program
> $ perf mem -t load rep
> $ perf mem -t load rep --stdio
>
> Debugging with TUI is a pain, so I stopped.
>
> I have a patch to improve perf mem a bit but I'd like to get this problem
> fixed first.
>
> Maybe you have a idea of what's going on.
I usually avoid the TUI by not installing slang-devel.
But, trying out Linus' current tip (a0e4467726c - post 3.18) this seems
to work fine:
perf mem -t load rec ls
perf mem -t load rep
But kernel version for the box is 3.16.6-203.fc20.x86_64.
David
On Wed, Dec 10, 2014 at 12:18 AM, David Ahern <[email protected]> wrote:
> On 12/9/14 9:15 PM, Stephane Eranian wrote:
>>
>> Hi,
>>
>> I was running a couple of tests with perf mem from tip.git and
>> found that perf mem -t load rep was failing only in TUI mode.
>> With --stdio, it works fine. I did a bit of investigation and found
>> out that the TUI mode is failing because some symbol (kernel)
>> is showing up with a huge (negative) size. I don't understand
>> the connection between the TUI vs. tty browser and symbols.
>>
>> To reproduce:
>> $ perf mem -t load rec my_test_program
>> $ perf mem -t load rep
>> $ perf mem -t load rep --stdio
>>
>> Debugging with TUI is a pain, so I stopped.
>>
>> I have a patch to improve perf mem a bit but I'd like to get this problem
>> fixed first.
>>
>> Maybe you have a idea of what's going on.
>
>
> I usually avoid the TUI by not installing slang-devel.
>
> But, trying out Linus' current tip (a0e4467726c - post 3.18) this seems to
> work fine:
> perf mem -t load rec ls
> perf mem -t load rep
>
What about tip.git?
> But kernel version for the box is 3.16.6-203.fc20.x86_64.
>
> David
On Tue, Dec 09, 2014 at 11:15:13PM -0500, Stephane Eranian wrote:
> Hi,
>
> I was running a couple of tests with perf mem from tip.git and
> found that perf mem -t load rep was failing only in TUI mode.
> With --stdio, it works fine. I did a bit of investigation and found
> out that the TUI mode is failing because some symbol (kernel)
> is showing up with a huge (negative) size. I don't understand
> the connection between the TUI vs. tty browser and symbols.
>
> To reproduce:
> $ perf mem -t load rec my_test_program
> $ perf mem -t load rep
> $ perf mem -t load rep --stdio
hi,
I couldn't reproduce any failure on the acme's perf/core
since you indicated it's data dependent, could you share
your perf.data?
thanks,
jirka
>
> Debugging with TUI is a pain, so I stopped.
>
> I have a patch to improve perf mem a bit but I'd like to get this problem
> fixed first.
>
> Maybe you have a idea of what's going on.
>
> Thanks.
On Wed, Dec 10, 2014 at 10:30 AM, Jiri Olsa <[email protected]> wrote:
> On Tue, Dec 09, 2014 at 11:15:13PM -0500, Stephane Eranian wrote:
>> Hi,
>>
>> I was running a couple of tests with perf mem from tip.git and
>> found that perf mem -t load rep was failing only in TUI mode.
>> With --stdio, it works fine. I did a bit of investigation and found
>> out that the TUI mode is failing because some symbol (kernel)
>> is showing up with a huge (negative) size. I don't understand
>> the connection between the TUI vs. tty browser and symbols.
>>
>> To reproduce:
>> $ perf mem -t load rec my_test_program
>> $ perf mem -t load rep
>> $ perf mem -t load rep --stdio
>
> hi,
> I couldn't reproduce any failure on the acme's perf/core
>
What about tip.git?
> since you indicated it's data dependent, could you share
> your perf.data?
>
I can send it to you directly.
> thanks,
> jirka
>
>>
>> Debugging with TUI is a pain, so I stopped.
>>
>> I have a patch to improve perf mem a bit but I'd like to get this problem
>> fixed first.
>>
>> Maybe you have a idea of what's going on.
>>
>> Thanks.
On Wed, Dec 10, 2014 at 10:46:59AM -0500, Stephane Eranian wrote:
> On Wed, Dec 10, 2014 at 10:30 AM, Jiri Olsa <[email protected]> wrote:
> > On Tue, Dec 09, 2014 at 11:15:13PM -0500, Stephane Eranian wrote:
> >> Hi,
> >>
> >> I was running a couple of tests with perf mem from tip.git and
> >> found that perf mem -t load rep was failing only in TUI mode.
> >> With --stdio, it works fine. I did a bit of investigation and found
> >> out that the TUI mode is failing because some symbol (kernel)
> >> is showing up with a huge (negative) size. I don't understand
> >> the connection between the TUI vs. tty browser and symbols.
> >>
> >> To reproduce:
> >> $ perf mem -t load rec my_test_program
> >> $ perf mem -t load rep
> >> $ perf mem -t load rep --stdio
> >
> > hi,
> > I couldn't reproduce any failure on the acme's perf/core
> >
> What about tip.git?
works for me
>
> > since you indicated it's data dependent, could you share
> > your perf.data?
> >
> I can send it to you directly.
ok, I'll check
jirka