Re: [PATCH 0/2 v2] Syscalls tracing

[Frederic Weisbecker <fweisbec@gmail.com>]

On Fri, Mar 13, 2009 at 05:17:49PM +0100, Ingo Molnar wrote:
> 
> * Ingo Molnar <mingo@elte.hu> wrote:
> 
> > 
> > * Frederic Weisbecker <fweisbec@gmail.com> wrote:
> > 
> > > On Fri, Mar 13, 2009 at 03:42:10PM +0100, Frederic Weisbecker wrote:
> > > > tracing/syscalls: core infrastructure to trace syscalls
> > > > 
> > > > This new iteration addresses a good part of the previous reviews.
> > > 
> > > 
> > > Ah I just discovered that you applied the previous version 
> > > today. But the v2 is not a delta :-s
> > > 
> > > I can rebase them but not until Sunday.
> > 
> > No problem, i'll deltify them and will have a look.
> 
> Ok, i did the deltas, tidied them up and put them into 
> tip:tracing/syscalls.
> 
> Nice stuff! Here's some sample output:
> 
> aldebaran:/debug/tracing> head trace
> # tracer: syscall
> #
> #           TASK-PID    CPU#    TIMESTAMP  FUNCTION
> #              | |       |          |         |
>            <...>-4405  [003]   188.452934: sys_dup2(oldfd: a, newfd: 1) 
>            <...>-4405  [003]   188.452939: sys_dup2 -> 0x1
>            <...>-4405  [003]   188.452940: sys_fcntl(fd: a, cmd: 1, arg: 0) 
>            <...>-4405  [003]   188.452941: sys_fcntl -> 0x1
>            <...>-4405  [003]   188.452942: sys_close(fd: a) 
>            <...>-4405  [003]   188.452943: sys_close -> 0x0
> 
> A suggestion:
> 
>  - Would be nice for all the registered syscalls to show up
>    under /debug/events/syscalls/, one directory per syscall,
>    with an 'enable' and a 'format' file as well.


Nice idea.
It is somehow planned since I want to let the user export the traces
through their binary values (so they will need the format).

 
> And a bugreport:
> 
>  - when using function_graph (after having used the syscall 
>    tracer) i dont see graph traces anymore - only the syscall 
>    trace entries:


Ah exact, it seems I forgot to reset the buffer while switching the tracer.


> # tracer: function_graph
> #
> # CPU  DURATION                  FUNCTION CALLS
> # |     |   |                     |   |   |   |
> ##### CPU 9 buffer started ####
>  at-spi-registry-3063  [009]   322.915058: sys_read -> 0x5a8
>  at-spi-registry-3063  [009]   322.915059: sys_write(fd: 6, buf: 632840, count: 5a8) 
>  at-spi-registry-3063  [009]   322.915062: sys_write -> 0x5a8
>  at-spi-registry-3063  [009]   322.915062: sys_read(fd: 4, buf: 632840, count: 40000) 
> 
> That's not intended, right?


Indeed it's not. I'm cooking the fix.

Thanks.


> 	Ingo