[openib-general] Re: [PATCH] mthca: correct max_rd_atomic handling

[Michael S. Tsirkin]

Quoting r. Roland Dreier <rdreier at cisco.com>:
> Subject: Re: [PATCH] mthca: correct max_rd_atomic handling
> 
> Hmm, I'm not sure about this change any more now.  Why is setting swe
> being tied to sra_max?

Note its tied to IB_QP_MAX_QP_RD_ATOMIC - the attribute bit, not the value.

Well, as I see it SWE isnt really used in IB spec, but hardware needs it.

Our hardware requires/allows you to set this bit in exactly same transitions
where sre and sae are required/optional, and this unsurprisingly is when
IB_QP_MAX_QP_RD_ATOMIC is required/optional.

In other transitions its supposed to be 0 since thats the value reserved
bits should have. VAPI and current mthca code seems to set this bit always: this
seems to violate what the documentation says, but seems to work nevertheless.

> We should be able to post RDMA writes even if
> reads/atomics are disabled, right?
> 
>  - R.
> 

Right. Thats why we set it inconditionally when IB_QP_MAX_QP_RD_ATOMIC attribute
bit is set, and IB_QP_MAX_QP_RD_ATOMIC is required to go to RTS.

-- 
MST