On Fri, May 13, 2022 at 03:15:50PM +0800, Guangguan Wang wrote:
> As cdc msg's length is 44B, cdc msgs can be sent inline in
> most rdma devices, which can help reducing sending latency.
>
> In my test environment, which are 2 VMs running on the same
> physical host and whose NICs(ConnectX-4Lx) are working on
> SR-IOV mode, qperf shows 0.4us-0.7us improvement in latency.
>
> Test command:
> server: smc_run taskset -c 1 qperf
> client: smc_run taskset -c 1 qperf <server ip> -oo \
> msg_size:1:2K:*2 -t 30 -vu tcp_lat
>
> The results shown below:
> msgsize before after
> 1B 11.9 us 11.2 us (-0.7 us)
> 2B 11.7 us 11.2 us (-0.5 us)
> 4B 11.7 us 11.3 us (-0.4 us)
> 8B 11.6 us 11.2 us (-0.4 us)
> 16B 11.7 us 11.3 us (-0.4 us)
> 32B 11.7 us 11.3 us (-0.4 us)
> 64B 11.7 us 11.2 us (-0.5 us)
> 128B 11.6 us 11.2 us (-0.4 us)
> 256B 11.8 us 11.2 us (-0.6 us)
> 512B 11.8 us 11.4 us (-0.4 us)
> 1KB 11.9 us 11.4 us (-0.5 us)
> 2KB 12.1 us 11.5 us (-0.6 us)
>
> Signed-off-by: Guangguan Wang <guangguan.wang@xxxxxxxxxxxxxxxxx>
> ---
> net/smc/smc_ib.c | 1 +
> net/smc/smc_wr.c | 5 ++++-
> 2 files changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/net/smc/smc_ib.c b/net/smc/smc_ib.c
> index a3e2d3b89568..1dcce9e4f4ca 100644
> --- a/net/smc/smc_ib.c
> +++ b/net/smc/smc_ib.c
> @@ -671,6 +671,7 @@ int smc_ib_create_queue_pair(struct smc_link *lnk)
> .max_recv_wr = SMC_WR_BUF_CNT * 3,
> .max_send_sge = SMC_IB_MAX_SEND_SGE,
> .max_recv_sge = sges_per_buf,
> + .max_inline_data = SMC_WR_TX_SIZE,
> },
> .sq_sig_type = IB_SIGNAL_REQ_WR,
> .qp_type = IB_QPT_RC,
> diff --git a/net/smc/smc_wr.c b/net/smc/smc_wr.c
> index 24be1d03fef9..8a2f9a561197 100644
> --- a/net/smc/smc_wr.c
> +++ b/net/smc/smc_wr.c
> @@ -554,10 +554,11 @@ void smc_wr_remember_qp_attr(struct smc_link *lnk)
> static void smc_wr_init_sge(struct smc_link *lnk)
> {
> int sges_per_buf = (lnk->lgr->smc_version == SMC_V2) ? 2 : 1;
> + bool send_inline = (lnk->qp_attr.cap.max_inline_data >= SMC_WR_TX_SIZE);
When will it be false? You are creating QPs with max_inline_data == SMC_WR_TX_SIZE?
> u32 i;
>
> for (i = 0; i < lnk->wr_tx_cnt; i++) {
> - lnk->wr_tx_sges[i].addr =
> + lnk->wr_tx_sges[i].addr = send_inline ? (u64)(&lnk->wr_tx_bufs[i]) :
> lnk->wr_tx_dma_addr + i * SMC_WR_BUF_SIZE;
> lnk->wr_tx_sges[i].length = SMC_WR_TX_SIZE;
> lnk->wr_tx_sges[i].lkey = lnk->roce_pd->local_dma_lkey;
> @@ -575,6 +576,8 @@ static void smc_wr_init_sge(struct smc_link *lnk)
> lnk->wr_tx_ibs[i].opcode = IB_WR_SEND;
> lnk->wr_tx_ibs[i].send_flags =
> IB_SEND_SIGNALED | IB_SEND_SOLICITED;
> + if (send_inline)
> + lnk->wr_tx_ibs[i].send_flags |= IB_SEND_INLINE;
If you try to transfer data == SMC_WR_TX_SIZE, you will get -ENOMEM error.
IB drivers check that length < qp->max_inline_data.
Thanks
> lnk->wr_tx_rdmas[i].wr_tx_rdma[0].wr.opcode = IB_WR_RDMA_WRITE;
> lnk->wr_tx_rdmas[i].wr_tx_rdma[1].wr.opcode = IB_WR_RDMA_WRITE;
> lnk->wr_tx_rdmas[i].wr_tx_rdma[0].wr.sg_list =
> --
> 2.24.3 (Apple Git-128)
>
