Re: debugctl msr

[stephane eranian]

Markus,

I found a couple more bugs with ds.c:

- in ds_put_context(), you need to mark task->thread.ds_ctx = NULL or
this_system_context = NULL when you are
  done, otherwise a subsequent session on the same task or CPU will
think there is already a context allocated but
  the pointer will be stale.

- in ptrace.c:ptrace_disable(), you systematically invoke ds_release()
without checking if TIF_BTS_TRACE_TS
  is set. That causes extraneous calls to ds_release() which messes up
the reference counting if PEBS is in use.


I ran into those issues trying to make perfmon work. So far, I have
gotten per-thread PEBS to work. Per-cpu is still
crashing the machine.


On Wed, Nov 19, 2008 at 11:26 PM, Markus Metzger
<markus.t.metzger@xxxxxxxxxxxxxx> wrote:
> On Wed, 2008-11-19 at 20:20 +0100, stephane eranian wrote:
>
>> Yes, I have narrowed this down to the following lines:
>>         current->mm->total_vm  -= context->pages[qual];
>>         current->mm->locked_vm -= context->pages[qual];
>>
>> I think this is again related to the problem of which thread call
>> ds_release(). In my test
>> case, this is the monitored thread as it exits. By the time it gets
>> there current->mm is NULL.
>
> Yes, this is again the ptrace-ness of the approach. The entire code
> assumes that there is one tracer task that controls another traced task.
>
> You're right, though, that I should only do the memory accounting
> if the buffer had been allocated by ds.c.
> That's a plain bug. Perfmon2 is the first user that uses its own
> buffers.
>
>
>> > The point I was trying to make is that buffer overflows should not be
>> > handled on higher levels (i.e. users of ds.c). That's why I am so
>> > reluctant to expose the interrupt threshold in the ds.c interface.
>> >
>> But the threshold is a characteristic of the buffer, not the interrupt handler.
>> Depending on the tool, it may be interesting to set the threshold earlier than
>> at the end of the buffer.
>
> Good point.
>
> Would you want to change the threshold or would it be OK if this became
> another parameter to ds_request()?
>
> regards,
> markus.
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/