From: yanjun.zhu@linux.dev
To: mustafa.ismail@intel.com, shiraz.saleem@intel.com,
dledford@redhat.com, jgg@ziepe.ca, linux-rdma@vger.kernel.org,
yanjun.zhu@linux.dev, leonro@nvidia.com
Subject: [PATCH 1/1] RDMA/irdma: optimize rx path by removing unnecessary copy
Date: Fri, 29 Oct 2021 12:23:40 -0400 [thread overview]
Message-ID: <20211029162340.241768-1-yanjun.zhu@linux.dev> (raw)
From: Zhu Yanjun <yanjun.zhu@linux.dev>
In the function irdma_post_recv, the function irdma_copy_sg_list is
not needed since the struct irdma_sge and ib_sge have the similar
member variables. The struct irdma_sge can be replaced with the
struct ib_sge totally.
This can increase the rx performance of irdma.
Signed-off-by: Zhu Yanjun <yanjun.zhu@linux.dev>
---
drivers/infiniband/hw/irdma/uk.c | 38 +++++++++++++-------------
drivers/infiniband/hw/irdma/user.h | 23 ++++++----------
drivers/infiniband/hw/irdma/verbs.c | 41 +++++++++--------------------
3 files changed, 39 insertions(+), 63 deletions(-)
diff --git a/drivers/infiniband/hw/irdma/uk.c b/drivers/infiniband/hw/irdma/uk.c
index 5fb92de1f015..0abbfa2538f1 100644
--- a/drivers/infiniband/hw/irdma/uk.c
+++ b/drivers/infiniband/hw/irdma/uk.c
@@ -13,16 +13,16 @@
* @sge: sge length and stag
* @valid: The wqe valid
*/
-static void irdma_set_fragment(__le64 *wqe, u32 offset, struct irdma_sge *sge,
+static void irdma_set_fragment(__le64 *wqe, u32 offset, struct ib_sge *sge,
u8 valid)
{
if (sge) {
set_64bit_val(wqe, offset,
- FIELD_PREP(IRDMAQPSQ_FRAG_TO, sge->tag_off));
+ FIELD_PREP(IRDMAQPSQ_FRAG_TO, sge->addr));
set_64bit_val(wqe, offset + 8,
FIELD_PREP(IRDMAQPSQ_VALID, valid) |
- FIELD_PREP(IRDMAQPSQ_FRAG_LEN, sge->len) |
- FIELD_PREP(IRDMAQPSQ_FRAG_STAG, sge->stag));
+ FIELD_PREP(IRDMAQPSQ_FRAG_LEN, sge->length) |
+ FIELD_PREP(IRDMAQPSQ_FRAG_STAG, sge->lkey));
} else {
set_64bit_val(wqe, offset, 0);
set_64bit_val(wqe, offset + 8,
@@ -38,14 +38,14 @@ static void irdma_set_fragment(__le64 *wqe, u32 offset, struct irdma_sge *sge,
* @valid: wqe valid flag
*/
static void irdma_set_fragment_gen_1(__le64 *wqe, u32 offset,
- struct irdma_sge *sge, u8 valid)
+ struct ib_sge *sge, u8 valid)
{
if (sge) {
set_64bit_val(wqe, offset,
- FIELD_PREP(IRDMAQPSQ_FRAG_TO, sge->tag_off));
+ FIELD_PREP(IRDMAQPSQ_FRAG_TO, sge->addr));
set_64bit_val(wqe, offset + 8,
- FIELD_PREP(IRDMAQPSQ_GEN1_FRAG_LEN, sge->len) |
- FIELD_PREP(IRDMAQPSQ_GEN1_FRAG_STAG, sge->stag));
+ FIELD_PREP(IRDMAQPSQ_GEN1_FRAG_LEN, sge->length) |
+ FIELD_PREP(IRDMAQPSQ_GEN1_FRAG_STAG, sge->lkey));
} else {
set_64bit_val(wqe, offset, 0);
set_64bit_val(wqe, offset + 8, 0);
@@ -289,7 +289,7 @@ enum irdma_status_code irdma_uk_rdma_write(struct irdma_qp_uk *qp,
return IRDMA_ERR_INVALID_FRAG_COUNT;
for (i = 0; i < op_info->num_lo_sges; i++)
- total_size += op_info->lo_sg_list[i].len;
+ total_size += op_info->lo_sg_list[i].length;
read_fence |= info->read_fence;
@@ -310,7 +310,7 @@ enum irdma_status_code irdma_uk_rdma_write(struct irdma_qp_uk *qp,
irdma_clr_wqes(qp, wqe_idx);
set_64bit_val(wqe, 16,
- FIELD_PREP(IRDMAQPSQ_FRAG_TO, op_info->rem_addr.tag_off));
+ FIELD_PREP(IRDMAQPSQ_FRAG_TO, op_info->rem_addr.addr));
if (info->imm_data_valid) {
set_64bit_val(wqe, 0,
@@ -339,7 +339,7 @@ enum irdma_status_code irdma_uk_rdma_write(struct irdma_qp_uk *qp,
++addl_frag_cnt;
}
- hdr = FIELD_PREP(IRDMAQPSQ_REMSTAG, op_info->rem_addr.stag) |
+ hdr = FIELD_PREP(IRDMAQPSQ_REMSTAG, op_info->rem_addr.lkey) |
FIELD_PREP(IRDMAQPSQ_OPCODE, info->op_type) |
FIELD_PREP(IRDMAQPSQ_IMMDATAFLAG, info->imm_data_valid) |
FIELD_PREP(IRDMAQPSQ_REPORTRTT, info->report_rtt) |
@@ -391,7 +391,7 @@ enum irdma_status_code irdma_uk_rdma_read(struct irdma_qp_uk *qp,
return IRDMA_ERR_INVALID_FRAG_COUNT;
for (i = 0; i < op_info->num_lo_sges; i++)
- total_size += op_info->lo_sg_list[i].len;
+ total_size += op_info->lo_sg_list[i].length;
ret_code = irdma_fragcnt_to_quanta_sq(op_info->num_lo_sges, &quanta);
if (ret_code)
@@ -426,8 +426,8 @@ enum irdma_status_code irdma_uk_rdma_read(struct irdma_qp_uk *qp,
++addl_frag_cnt;
}
set_64bit_val(wqe, 16,
- FIELD_PREP(IRDMAQPSQ_FRAG_TO, op_info->rem_addr.tag_off));
- hdr = FIELD_PREP(IRDMAQPSQ_REMSTAG, op_info->rem_addr.stag) |
+ FIELD_PREP(IRDMAQPSQ_FRAG_TO, op_info->rem_addr.addr));
+ hdr = FIELD_PREP(IRDMAQPSQ_REMSTAG, op_info->rem_addr.lkey) |
FIELD_PREP(IRDMAQPSQ_REPORTRTT, (info->report_rtt ? 1 : 0)) |
FIELD_PREP(IRDMAQPSQ_ADDFRAGCNT, addl_frag_cnt) |
FIELD_PREP(IRDMAQPSQ_OPCODE,
@@ -477,7 +477,7 @@ enum irdma_status_code irdma_uk_send(struct irdma_qp_uk *qp,
return IRDMA_ERR_INVALID_FRAG_COUNT;
for (i = 0; i < op_info->num_sges; i++)
- total_size += op_info->sg_list[i].len;
+ total_size += op_info->sg_list[i].length;
if (info->imm_data_valid)
frag_cnt = op_info->num_sges + 1;
@@ -705,9 +705,9 @@ irdma_uk_inline_rdma_write(struct irdma_qp_uk *qp, struct irdma_post_sq_info *in
read_fence |= info->read_fence;
set_64bit_val(wqe, 16,
- FIELD_PREP(IRDMAQPSQ_FRAG_TO, op_info->rem_addr.tag_off));
+ FIELD_PREP(IRDMAQPSQ_FRAG_TO, op_info->rem_addr.addr));
- hdr = FIELD_PREP(IRDMAQPSQ_REMSTAG, op_info->rem_addr.stag) |
+ hdr = FIELD_PREP(IRDMAQPSQ_REMSTAG, op_info->rem_addr.lkey) |
FIELD_PREP(IRDMAQPSQ_OPCODE, info->op_type) |
FIELD_PREP(IRDMAQPSQ_INLINEDATALEN, op_info->len) |
FIELD_PREP(IRDMAQPSQ_REPORTRTT, info->report_rtt ? 1 : 0) |
@@ -826,7 +826,7 @@ irdma_uk_stag_local_invalidate(struct irdma_qp_uk *qp,
u64 hdr;
u32 wqe_idx;
bool local_fence = false;
- struct irdma_sge sge = {};
+ struct ib_sge sge = {};
info->push_wqe = qp->push_db ? true : false;
op_info = &info->op.inv_local_stag;
@@ -839,7 +839,7 @@ irdma_uk_stag_local_invalidate(struct irdma_qp_uk *qp,
irdma_clr_wqes(qp, wqe_idx);
- sge.stag = op_info->target_stag;
+ sge.lkey = op_info->target_stag;
qp->wqe_ops.iw_set_fragment(wqe, 0, &sge, 0);
set_64bit_val(wqe, 16, 0);
diff --git a/drivers/infiniband/hw/irdma/user.h b/drivers/infiniband/hw/irdma/user.h
index e0e9512ad3d5..2af5986039e2 100644
--- a/drivers/infiniband/hw/irdma/user.h
+++ b/drivers/infiniband/hw/irdma/user.h
@@ -16,7 +16,6 @@
#define irdma_access_privileges u32
#define irdma_physical_fragment u64
#define irdma_address_list u64 *
-#define irdma_sgl struct irdma_sge *
#define IRDMA_MAX_MR_SIZE 0x200000000000ULL
@@ -151,12 +150,6 @@ struct irdma_cq_uk;
struct irdma_qp_uk_init_info;
struct irdma_cq_uk_init_info;
-struct irdma_sge {
- irdma_tagged_offset tag_off;
- u32 len;
- irdma_stag stag;
-};
-
struct irdma_ring {
u32 head;
u32 tail;
@@ -172,7 +165,7 @@ struct irdma_extended_cqe {
};
struct irdma_post_send {
- irdma_sgl sg_list;
+ struct ib_sge *sg_list;
u32 num_sges;
u32 qkey;
u32 dest_qp;
@@ -189,26 +182,26 @@ struct irdma_post_inline_send {
struct irdma_post_rq_info {
u64 wr_id;
- irdma_sgl sg_list;
+ struct ib_sge *sg_list;
u32 num_sges;
};
struct irdma_rdma_write {
- irdma_sgl lo_sg_list;
+ struct ib_sge *lo_sg_list;
u32 num_lo_sges;
- struct irdma_sge rem_addr;
+ struct ib_sge rem_addr;
};
struct irdma_inline_rdma_write {
void *data;
u32 len;
- struct irdma_sge rem_addr;
+ struct ib_sge rem_addr;
};
struct irdma_rdma_read {
- irdma_sgl lo_sg_list;
+ struct ib_sge *lo_sg_list;
u32 num_lo_sges;
- struct irdma_sge rem_addr;
+ struct ib_sge rem_addr;
};
struct irdma_bind_window {
@@ -306,7 +299,7 @@ enum irdma_status_code irdma_uk_stag_local_invalidate(struct irdma_qp_uk *qp,
struct irdma_wqe_uk_ops {
void (*iw_copy_inline_data)(u8 *dest, u8 *src, u32 len, u8 polarity);
u16 (*iw_inline_data_size_to_quanta)(u32 data_size);
- void (*iw_set_fragment)(__le64 *wqe, u32 offset, struct irdma_sge *sge,
+ void (*iw_set_fragment)(__le64 *wqe, u32 offset, struct ib_sge *sge,
u8 valid);
void (*iw_set_mw_bind_wqe)(__le64 *wqe,
struct irdma_bind_window *op_info);
diff --git a/drivers/infiniband/hw/irdma/verbs.c b/drivers/infiniband/hw/irdma/verbs.c
index 02ca1f80968e..7ab9645d6f18 100644
--- a/drivers/infiniband/hw/irdma/verbs.c
+++ b/drivers/infiniband/hw/irdma/verbs.c
@@ -3039,24 +3039,6 @@ static int irdma_dereg_mr(struct ib_mr *ib_mr, struct ib_udata *udata)
return 0;
}
-/**
- * irdma_copy_sg_list - copy sg list for qp
- * @sg_list: copied into sg_list
- * @sgl: copy from sgl
- * @num_sges: count of sg entries
- */
-static void irdma_copy_sg_list(struct irdma_sge *sg_list, struct ib_sge *sgl,
- int num_sges)
-{
- unsigned int i;
-
- for (i = 0; (i < num_sges) && (i < IRDMA_MAX_WQ_FRAGMENT_COUNT); i++) {
- sg_list[i].tag_off = sgl[i].addr;
- sg_list[i].len = sgl[i].length;
- sg_list[i].stag = sgl[i].lkey;
- }
-}
-
/**
* irdma_post_send - kernel application wr
* @ibqp: qp ptr for wr
@@ -3133,7 +3115,7 @@ static int irdma_post_send(struct ib_qp *ibqp,
ret = irdma_uk_inline_send(ukqp, &info, false);
} else {
info.op.send.num_sges = ib_wr->num_sge;
- info.op.send.sg_list = (struct irdma_sge *)
+ info.op.send.sg_list = (struct ib_sge *)
ib_wr->sg_list;
if (iwqp->ibqp.qp_type == IB_QPT_UD ||
iwqp->ibqp.qp_type == IB_QPT_GSI) {
@@ -3169,15 +3151,18 @@ static int irdma_post_send(struct ib_qp *ibqp,
if (ib_wr->send_flags & IB_SEND_INLINE) {
info.op.inline_rdma_write.data = (void *)(uintptr_t)ib_wr->sg_list[0].addr;
- info.op.inline_rdma_write.len = ib_wr->sg_list[0].length;
- info.op.inline_rdma_write.rem_addr.tag_off = rdma_wr(ib_wr)->remote_addr;
- info.op.inline_rdma_write.rem_addr.stag = rdma_wr(ib_wr)->rkey;
+ info.op.inline_rdma_write.len =
+ ib_wr->sg_list[0].length;
+ info.op.inline_rdma_write.rem_addr.addr =
+ rdma_wr(ib_wr)->remote_addr;
+ info.op.inline_rdma_write.rem_addr.lkey =
+ rdma_wr(ib_wr)->rkey;
ret = irdma_uk_inline_rdma_write(ukqp, &info, false);
} else {
info.op.rdma_write.lo_sg_list = (void *)ib_wr->sg_list;
info.op.rdma_write.num_lo_sges = ib_wr->num_sge;
- info.op.rdma_write.rem_addr.tag_off = rdma_wr(ib_wr)->remote_addr;
- info.op.rdma_write.rem_addr.stag = rdma_wr(ib_wr)->rkey;
+ info.op.rdma_write.rem_addr.addr = rdma_wr(ib_wr)->remote_addr;
+ info.op.rdma_write.rem_addr.lkey = rdma_wr(ib_wr)->rkey;
ret = irdma_uk_rdma_write(ukqp, &info, false);
}
@@ -3198,8 +3183,8 @@ static int irdma_post_send(struct ib_qp *ibqp,
break;
}
info.op_type = IRDMA_OP_TYPE_RDMA_READ;
- info.op.rdma_read.rem_addr.tag_off = rdma_wr(ib_wr)->remote_addr;
- info.op.rdma_read.rem_addr.stag = rdma_wr(ib_wr)->rkey;
+ info.op.rdma_read.rem_addr.addr = rdma_wr(ib_wr)->remote_addr;
+ info.op.rdma_read.rem_addr.lkey = rdma_wr(ib_wr)->rkey;
info.op.rdma_read.lo_sg_list = (void *)ib_wr->sg_list;
info.op.rdma_read.num_lo_sges = ib_wr->num_sge;
@@ -3286,7 +3271,6 @@ static int irdma_post_recv(struct ib_qp *ibqp,
struct irdma_qp *iwqp;
struct irdma_qp_uk *ukqp;
struct irdma_post_rq_info post_recv = {};
- struct irdma_sge sg_list[IRDMA_MAX_WQ_FRAGMENT_COUNT];
enum irdma_status_code ret = 0;
unsigned long flags;
int err = 0;
@@ -3301,8 +3285,7 @@ static int irdma_post_recv(struct ib_qp *ibqp,
while (ib_wr) {
post_recv.num_sges = ib_wr->num_sge;
post_recv.wr_id = ib_wr->wr_id;
- irdma_copy_sg_list(sg_list, ib_wr->sg_list, ib_wr->num_sge);
- post_recv.sg_list = sg_list;
+ post_recv.sg_list = ib_wr->sg_list;
ret = irdma_uk_post_receive(ukqp, &post_recv);
if (ret) {
ibdev_dbg(&iwqp->iwdev->ibdev,
--
2.27.0
next reply other threads:[~2021-10-29 8:33 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-10-29 16:23 yanjun.zhu [this message]
2021-10-29 19:53 ` [PATCH 1/1] RDMA/irdma: optimize rx path by removing unnecessary copy Saleem, Shiraz
2021-10-30 2:55 ` yanjun.zhu
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20211029162340.241768-1-yanjun.zhu@linux.dev \
--to=yanjun.zhu@linux.dev \
--cc=dledford@redhat.com \
--cc=jgg@ziepe.ca \
--cc=leonro@nvidia.com \
--cc=linux-rdma@vger.kernel.org \
--cc=mustafa.ismail@intel.com \
--cc=shiraz.saleem@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).