[ltt-dev] Force buffer sizes to be at least page size

[Chase Douglas]

On Jul 15, 2009, at 2:08 PM, Mathieu Desnoyers wrote:
> * Mathieu Desnoyers (compudj at krystal.dyndns.org) wrote:
>> * Mathieu Desnoyers (compudj at krystal.dyndns.org) wrote:
>>> * Chase Douglas (cndougla at linux.vnet.ibm.com) wrote:
>>>> During some testing across various architectures we found that  
>>>> our ppc64
>>>> environment with 64K page sizes was causing lttctl to hang when
>>>> destroying a trace if the buffer size was less than 64K
>>>> (channel.all.bufsize). We had been testing with 16K buffer sizes,  
>>>> and
>>>> every other architecture ran fine. This led us to believe that  
>>>> ltt is
>>>> not handling buffer sizes less than a page size correctly.
>>>>
>>>> We did not determine the root cause of the failure, instead  
>>>> opting for a
>>>> quick patch to force the buffer sizes to be at least page size. I  
>>>> have
>>>> attached the patch. Please take this under advisement either to
>>>> determine the root cause or to force proper buffer sizes as  
>>>> necessary.
>>>>
>>>
>>> Actually, the subbuffer size (not only the buffer size) should be at
>>> least a page. I'll do a patch for LTTng shortly.
>>>
>>> Thanks for the report !
>>>
>>
>> Can you give this patch a try to see if it fixes your problem ?  
>> (without
>> your lttctl patch, the problem must be dealt at the kernel-level)
>>
>
> The patch seems to work OK on my system with 4k pages when trying
> forcing small subbuffer size values with lttctl.
>
> It's now in lttng 0.148.

I tested the patch and it did fix the issue.

Thanks!