[ofw] CM ref counting issues...

Sean Hefty sean.hefty at intel.com
Tue Dec 8 21:39:42 PST 2009


>>   4.00005c  fffffa80036c2720 fffe752f Blocked    nt!KiSwapContext+0x7f
>
>Why is this thread blocked trying to disconnect the QP?  Could this be holding
>the ref on the CM ID?

I'm looking into this thread in more detail, since I think there may be a
separate issue here.  I just don't see a clear relation between it and the IB CM
hang.  It should not be holding a reference on the CM ID.  When the timeout
occurs, we're not in an IB CM callback - the cm id signal flag is not set.

This is winverbs processing a work item asynchronously after a DREQ or DREP to
transition the QP into the error state.  It doesn't do anything with the
winverbs endpoint or the cm id, just the QP. 

>
>>                                         nt!KiSwapThread+0x2fa
>>                                         nt!KeWaitForSingleObject+0x2da
>> winverbs!WvProviderDisableRemove+0x4a
>>                                         winverbs!WvQpAcquire+0x2a
>>                                         winverbs!WvEpDisconnectQp+0x31
>> winverbs!WvEpDisconnectHandler+0xcd
>>                                         winverbs!WorkQueueHandler+0x4b
>>                                         nt!IopProcessWorkItem+0x27
>>                                         nt!ExpWorkerThread+0x11a
>>                                         nt!PspSystemThreadStartup+0x57
>>                                         nt!KxStartSystemThread+0x16






More information about the ofw mailing list