[Openib-windows] RE: [PATCH] query-cq

[Fab Tillier]

> From: Yossi Leybovich [mailto:sleybo at mellanox.co.il]
> Sent: Tuesday, September 06, 2005 7:16 AM
> 
> Hello Fab,
>
> Attached is a patch for query cq, the patch also includes few checks to the
> alts test.
> Our verification team found that the return size in query_cq is different from
> the size that was return in create_cq VERB.
> The reason is that when creating cq  the THH layer allocate some spare cqes
> (for DB coalsing) and return the size that was requested by the user
> But the query_cq IOCTL use the kernel tables for the size which use the actual
> size (usually a bigger size).

Isn't the bigger size the correct one to return?  This seems to be what current
verb implementations do, and what the IB spec suggests.  If it isn't, don't
kernel clients end up getting the wrong value when they query their CQ size?

The value should be the same as what was returned in the call to create_cq,
which is (hw\mt23108\vapi\Hca\hcahal\tavor\thhul_cqm\thhul_cqm.c at 890):

*num_o_cqes_p= (1U << new_cq->cur_buf.log2_num_o_cqes) - 1 -
	new_cq->cur_buf.spare_cqes;

If the kernel clients get fixed, then the user-mode clients automatically get
the right value and we only need to update the ALTS test.

Does that make sense?

Thanks,

- Fab