Re: [Qemu-devel] Regression with commit 095497ffc66b7f031

[Paolo Bonzini <pbonzini@redhat.com>]

On 15/07/2016 12:41, Juergen Gross wrote:
> On 15/07/16 12:35, Paolo Bonzini wrote:
>>
>>
>> On 15/07/2016 12:12, Gerd Hoffmann wrote:
>>> On Fr, 2016-07-15 at 12:02 +0200, Paolo Bonzini wrote:
>>>>
>>>> On 15/07/2016 10:47, Juergen Gross wrote:
>>>>> Nothing scaring and no real difference between working and not working
>>>>> variant.
>>>>>
>>>>> Meanwhile I've been digging a little bit deeper and found the reason:
>>>>> libxenstore is setting up a reader thread which is waiting for the
>>>>> watch to fire. With above commit the stack size of that thread (16kB)
>>>>> is too small. Setting it to 32kB made qemu work again.
>>>>
>>>> This makes very little sense (not your fault)...  The commit doesn't
>>>> change stack usage at all, TLS should not be on the stack.
>>>>
>>>> Can you capture a backtrace where the 16K stack is exceeded?  Perhaps
>>>> it's only due to inlining decision on the compiler, in which case
>>>> Peter's patch from today is only a bandaid.
>>>
>>> Hmm, I guess I hold off the vnc pull request for now ...
>>
>> Go ahead.  I looked at glibc source code and the patch is okay.
> 
> Paolo, do you know of any interface to obtain the size of the TLS area
> taken from the stack (before calling pthread_create() )? 

https://gcc.gnu.org/ml/gcc-patches/2014-10/msg01643.html has a patch
that _removes_ code to do this from the golang runtime.  The comments
there say that only with glibc before version 2.16 the static TLS size
is taken out of the stack size...

What version of glibc are you using?

Paolo