Re: [PATCH v1 02/14] svcrdma: Add svc_rdma_map_reply_hdr()

[Chuck Lever <chuck.lever@oracle.com>]

> On Mar 21, 2017, at 1:54 PM, Sagi Grimberg <sagi@grimberg.me> wrote:
> 
> 
>> 	svc_rdma_build_send_wr(ctxt, 1);
>> 	ret = svc_rdma_send(rdma, &ctxt->send_wr);
>> 	if (ret) {
>> +		svc_rdma_unmap_dma(ctxt);
>> +		svc_rdma_put_context(ctxt, 1);
>> 		ret = -EIO;
>> -		goto out_unmap;
>> 	}
> 
> Any specific reason to not go with the goto scheme?

Only one "goto out_unmap" call site is left, and this
isn't a performance critical path.


> Can't this function grow more error paths in the future?

I can't think of any. The only time this code changes is
when overhauls like this happens.

I don't mind putting the out_unmap label back.


> btw, I'm assuming svc_rdma_unmap_dma() is the opposite of
> svc_rdma_map_reply_hdr() ?

svc_rdma_map_reply_hdr() DMA-maps the transport header buffer.

svc_rdma_unmap_dma() DMA-unmaps everything associated with
the ctxt.


--
Chuck Lever