> On Jul 10, 2015, at 4:43 PM, Anna Schumaker <[email protected]> wrote:
>
> Hi Chuck,
>
> On 07/09/2015 04:43 PM, Chuck Lever wrote:
>> Only the RPC/RDMA header is sent when making an RDMA_NOMSG call.
>> That header resides in the first element of the iovec array
>> passed to rpcrdma_ep_post().
>>
>> Instead of special casing the iovec element with the pad, just
>> sync all the elements in the send iovec. Syncing the zero pad is
>> not strictly necessary, but the pad is rarely if ever used these
>> days, and the extra cost in that case is small.
>>
>> Signed-off-by: Chuck Lever <[email protected]>
>> diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c
>> index cdf5220..9199436 100644
>> --- a/net/sunrpc/xprtrdma/verbs.c
>> +++ b/net/sunrpc/xprtrdma/verbs.c
>> @@ -1316,22 +1318,15 @@ rpcrdma_ep_post(struct rpcrdma_ia *ia,
>>
>> send_wr.next = NULL;
>> send_wr.wr_id = RPCRDMA_IGNORE_COMPLETION;
>> - send_wr.sg_list = req->rl_send_iov;
>> + send_wr.sg_list = iov;
>> send_wr.num_sge = req->rl_niovs;
>> send_wr.opcode = IB_WR_SEND;
>> - if (send_wr.num_sge == 4) /* no need to sync any pad (constant) */
>> - ib_dma_sync_single_for_device(ia->ri_device,
>> - req->rl_send_iov[3].addr,
>> - req->rl_send_iov[3].length,
>> - DMA_TO_DEVICE);
>> - ib_dma_sync_single_for_device(ia->ri_device,
>> - req->rl_send_iov[1].addr,
>> - req->rl_send_iov[1].length,
>> - DMA_TO_DEVICE);
>> - ib_dma_sync_single_for_device(ia->ri_device,
>> - req->rl_send_iov[0].addr,
>> - req->rl_send_iov[0].length,
>> - DMA_TO_DEVICE);
>> +
>> + for (i = 0; i < send_wr.num_sge; i++)
>> + ib_dma_sync_single_for_device(device, iov[i].addr,
>> + iov[i].length, DMA_TO_DEVICE);
>
> Two questions here:
> 1) Is syncing order important? The original code went 3 - 1 - 0, but now
> we're going 0 - 1 - 2 - 3.
No, the syncing order isn’t important. What’s important is that
the sge’s be synced _before_ the SEND Work Request is posted so
that the HCA can see the freshly marshaled RPC.
> 2) Is iov[2] the zero pad you mentioned in your commit message?
Yes, it is.
> If not, do you know why it wasn't already syncing?
The pad always contains zeroes. Syncing it is optional, AFAICT.
v2 of this series will have some changes in this area, and the
zero pad will be gone entirely.
> Thanks,
> Anna
>
>> + dprintk("RPC: %s: posting %d s/g entries\n",
>> + __func__, send_wr.num_sge);
>>
>> if (DECR_CQCOUNT(ep) > 0)
>> send_wr.send_flags = 0;
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
--
Chuck Lever
[email protected]
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
