[ofw] [Patch][Core] Fix possible BSOD during shutdown

[Alex Naslednikov]

With all my wish for the first option to be true, currently the second one looks more realistic :)
 

-----Original Message-----
From: Fab Tillier [mailto:ftillier at microsoft.com] 
Sent: Wednesday, May 12, 2010 6:52 PM
To: Alex Naslednikov; Smith, Stan; ofw at lists.openfabrics.org
Subject: RE: [Patch][Core] Fix possible BSOD during shutdown

Alex Naslednikov wrote on Wed, 12 May 2010 at 02:15:53

> The only reason to leave there ASSERT is for debug purposes.
> This assert previously helped me to get into the problem easily.
> 
> But taking into account that we would like to debug the original race, 
> this assert can help us a lot.

So are you working on a true fix, or will this just remain forever a fix for the symptom but not the cause?

-Fab

> 
> -----Original Message-----
> From: Smith, Stan [mailto:stan.smith at intel.com]
> Sent: Tuesday, May 11, 2010 7:39 PM
> To: Alex Naslednikov; ofw at lists.openfabrics.org
> Subject: RE: [Patch][Core] Fix possible BSOD during shutdown
> 
> Alex Naslednikov wrote:
>> Avoid the BSOD that can happen as a race consequence.
>> Signed-off by: Alexander Naslednikov (xalex at mellanox.co.il)
>> Index: D:/windows/MLNX_VPI_trunk/core/al/kernel/al_smi.c
>> ===================================================================
>> --- D:/windows/MLNX_VPI_trunk/core/al/kernel/al_smi.c (revision 5854)
>> +++ D:/windows/MLNX_VPI_trunk/core/al/kernel/al_smi.c (revision 5855)
>> @@ -3299,7 +3299,14 @@
>> 
>>   CL_ASSERT( mad_svc_context );
>>   CL_ASSERT( p_mad_response );
>> - CL_ASSERT( p_mad_response->send_context1 );
>> +
>> +
>> + if ( !p_mad_response->send_context1 ) {  CL_ASSERT(
>> + p_mad_response->send_context1 );  ib_put_mad( p_mad_response ); 
>> + AL_EXIT( AL_DBG_SMI );  return; }
>> 
>>   /* Initialize pointers. */
>>   p_spl_qp_svc = mad_svc_context;
> 
> Hello,
>   Your patch does indeed cure the intermittent shutdown crash. As 
> noted in previous email, the fix does not address the 'real' race 
> condition; although preventing any crash is goodness.
> Point being, why retain the 'CL_ASSERT( p_mad_response->send_context1 
> );' statement?
> 
> stan.
> _______________________________________________
> ofw mailing list
> ofw at lists.openfabrics.org
> http://lists.openfabrics.org/cgi-bin/mailman/listinfo/ofw