> I don’t see the err_data getting freed in ofi_cq_readerr(). Also, just adding 
> a free
> may not be sufficient, shouldn’t it have something like the saved_err_data 
> mechanism
> that is in util_eq.c to keep the err_data in scope for compatibility reasons?

I agree that the EQ and CQ implementations should be closer aligned.

The util_cq code allocates and frees the err_entry, but not the err_data.  I 
don't see that any providers actually use this field for CQ entries, so it 
likely isn't an issue in practice.  It seems cleaner that if util_cq frees 
err_data, it should also allocate it.

But... util_eq frees err_data without allocating it...

To 'fix' the CQ, I would add a new call, ofi_cq_write_error_data() that can 
take the err_data and size as input parameters.  That would allow util_cq to 
include the allocation as part of the event data.

- Sean
_______________________________________________
ofiwg mailing list
[email protected]
https://lists.openfabrics.org/mailman/listinfo/ofiwg

Reply via email to