[ofa-general] Re: IPOIB CM (NOSRQ)[PATCH V5] patch

[Pradeep Satyanarayana]

Michael S. Tsirkin wrote:
>> Quoting Pradeep Satyanarayana <pradeeps at linux.vnet.ibm.com>:
>> Subject: IPOIB CM (NOSRQ)[PATCH V5] patch
>>
>> Here is a fifth version of the IPOIB_CM_NOSRQ patch. This patch will
>> benefit adapters that do not support shared receive queues.
>>
>> This patch incorporates the following review comments and subsequent
>> discussions on this mailing list from v4:
>>
>> 1. Reduce the number of if(srq) tests in the packet receive path
> 
> I could still count at least 2 of these, and I don't see why there can't be just 1,
> or even 0 if the QP pool is hidden under the SRQ interface.

Yes, there are 2 of these now. Previously, only ipoib_poll() needed
to be altered to incorporate this. Now I would have to add 
ipoib_drain_cq() as well.

As previously mentioned we do need to keep in mind the maintainability
aspects and the way it is, all the changes are well contained.

Isn't it time that we should stop quibbling about one extra if(srq)? If
you are so inclined you can submit a patch on top of this one. We can
then debate the merits of that patch and make an appropriate decision.

> 
>> +int current_rc_qp = 0; /* Active RC QPs for NOSRQ */
>>  #define IPOIB_CM_IETF_ID 0x1000000000000000ULL
>>
>>  #define IPOIB_CM_RX_UPDATE_TIME (256 * HZ)
> 
> I don't see any locking for current_rc_qp, which looks wrong.

Yes, I will correct that.

Pradeep