xiaomi-sm8450/android_kernel_xiaomi_sm8450
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

RDMA/mlx5: Use PA mapping for PI handover

Browse Source 
If possibe, avoid doing a UMR operation to register data and protection
buffers (via MTT/KLM mkeys). Instead, use the local DMA key and map the
SG lists using PA access. This is safe, since the internal key for data
and protection never exposed to the remote server (only signature key
might be exposed). If PA mappings are not possible, perform mapping
using MTT/KLM descriptors.

The setup of the tested benchmark (using iSER ULP):
 - 2 servers with 24 cores (1 initiator and 1 target)
 - ConnectX-4/ConnectX-5 adapters
 - 24 target sessions with 1 LUN each
 - ramdisk backstore
 - PI active

Performance results running fio (24 jobs, 128 iodepth) using
write_generate=1 and read_verify=1 (w/w.o patch):

bs      IOPS(read)        IOPS(write)
----    ----------        ----------
512   1266.4K/1262.4K    1720.1K/1732.1K
4k    793139/570902      1129.6K/773982
32k   72660/72086        97229/96164

Using write_generate=0 and read_verify=0 (w/w.o patch):
bs      IOPS(read)        IOPS(write)
----    ----------        ----------
512   1590.2K/1600.1K    1828.2K/1830.3K
4k    1078.1K/937272     1142.1K/815304
32k   77012/77369        98125/97435

Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Suggested-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
This commit is contained in:
Max Gurtovoy
2019-06-11 18:52:56 +03:00
committed by Jason Gunthorpe
parent de0ae958de
commit 2563e2f30a
3 changed files with 112 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

76
drivers/infiniband/hw/mlx5/qp.c
View File

@@ -4562,7 +4562,7 @@ static int set_sig_data_segment(const struct ib_send_wr *send_wr,
data_len = pi_mr->data_length;
data_key = pi_mr->ibmr.lkey;
data_va = pi_mr->ibmr.iova;
data_va = pi_mr->data_iova;
if (pi_mr->meta_ndescs) {
prot_len = pi_mr->meta_length;
prot_key = pi_mr->ibmr.lkey;
@@ -4912,6 +4912,7 @@ static int _mlx5_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
struct mlx5_ib_qp *qp;
struct mlx5_ib_mr *mr;
struct mlx5_ib_mr *pi_mr;
struct mlx5_ib_mr pa_pi_mr;
struct ib_sig_attrs *sig_attrs;
struct mlx5_wqe_xrc_seg *xrc;
struct mlx5_bf *bf;
@@ -5026,35 +5027,62 @@ static int _mlx5_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
break;
case IB_WR_REG_MR_INTEGRITY:
memset(&reg_pi_wr, 0, sizeof(struct ib_reg_wr));
qp->sq.wr_data[idx] = IB_WR_REG_MR_INTEGRITY;
mr = to_mmr(reg_wr(wr)->mr);
pi_mr = mr->pi_mr;
reg_pi_wr.mr = &pi_mr->ibmr;
reg_pi_wr.access = reg_wr(wr)->access;
reg_pi_wr.key = pi_mr->ibmr.rkey;
if (pi_mr) {
memset(&reg_pi_wr, 0,
sizeof(struct ib_reg_wr));
qp->sq.wr_data[idx] = IB_WR_REG_MR_INTEGRITY;
ctrl->imm = cpu_to_be32(reg_pi_wr.key);
/* UMR for data + protection registration */
err = set_reg_wr(qp, &reg_pi_wr, &seg, &size,
&cur_edge, false);
if (err) {
*bad_wr = wr;
goto out;
}
finish_wqe(qp, ctrl, seg, size, cur_edge, idx,
wr->wr_id, nreq, fence,
MLX5_OPCODE_UMR);
reg_pi_wr.mr = &pi_mr->ibmr;
reg_pi_wr.access = reg_wr(wr)->access;
reg_pi_wr.key = pi_mr->ibmr.rkey;
err = begin_wqe(qp, &seg, &ctrl, wr, &idx,
&size, &cur_edge, nreq);
if (err) {
mlx5_ib_warn(dev, "\n");
err = -ENOMEM;
*bad_wr = wr;
goto out;
ctrl->imm = cpu_to_be32(reg_pi_wr.key);
/* UMR for data + prot registration */
err = set_reg_wr(qp, &reg_pi_wr, &seg,
&size, &cur_edge,
false);
if (err) {
*bad_wr = wr;
goto out;
}
finish_wqe(qp, ctrl, seg, size,
cur_edge, idx, wr->wr_id,
nreq, fence,
MLX5_OPCODE_UMR);
err = begin_wqe(qp, &seg, &ctrl, wr,
&idx, &size, &cur_edge,
nreq);
if (err) {
mlx5_ib_warn(dev, "\n");
err = -ENOMEM;
*bad_wr = wr;
goto out;
}
} else {
memset(&pa_pi_mr, 0,
sizeof(struct mlx5_ib_mr));
/* No UMR, use local_dma_lkey */
pa_pi_mr.ibmr.lkey =
mr->ibmr.pd->local_dma_lkey;
pa_pi_mr.ndescs = mr->ndescs;
pa_pi_mr.data_length = mr->data_length;
pa_pi_mr.data_iova = mr->data_iova;
if (mr->meta_ndescs) {
pa_pi_mr.meta_ndescs =
mr->meta_ndescs;
pa_pi_mr.meta_length =
mr->meta_length;
pa_pi_mr.pi_iova = mr->pi_iova;
}
pa_pi_mr.ibmr.length = mr->ibmr.length;
mr->pi_mr = &pa_pi_mr;
}
ctrl->imm = cpu_to_be32(mr->ibmr.rkey);
/* UMR for sig MR */