ed62e6ed00
driver inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I8A08Z Compared with normal doorbell, using record doorbell can shorten the process of ringing the doorbell and reduce the latency. Signed-off-by: Yangyang Li <liyangyang20@huawei.com> Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com> Signed-off-by: Ran Zhou <zhouran10@h-partners.com> (cherry picked from commit 23f6e3ca5e9fa66b41e3d33a9c4a88429cfc61ab)
151 lines
5.0 KiB
Diff
151 lines
5.0 KiB
Diff
From 2880d64c8d73375978d2767c5dd7803b444f9016 Mon Sep 17 00:00:00 2001
|
|
From: Yangyang Li <liyangyang20@huawei.com>
|
|
Date: Mon, 16 Oct 2023 16:10:06 +0800
|
|
Subject: [PATCH] libhns: Support SRQ record doorbell
|
|
|
|
Compared with normal doorbell, using record doorbell can shorten the
|
|
process of ringing the doorbell and reduce the latency.
|
|
|
|
During SRQ creation, the kernel driver will allocate doorbell buffer
|
|
and notify userspace whether the SRQ record doorbell is enabled with
|
|
the flag HNS_ROCE_RSP_SRQ_CAP_RECORD_DB. The userspace driver will decide
|
|
whether to use record doorbell or normal doorbell based on this flag
|
|
in post SRQ recv process.
|
|
|
|
This patch relies on the corresponding kernel patch:
|
|
RDMA/hns: Support SRQ record doorbell
|
|
|
|
Signed-off-by: Yangyang Li <liyangyang20@huawei.com>
|
|
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
|
|
---
|
|
providers/hns/hns_roce_u.h | 4 +++-
|
|
providers/hns/hns_roce_u_db.c | 1 +
|
|
providers/hns/hns_roce_u_hw_v2.c | 14 +++++++++-----
|
|
providers/hns/hns_roce_u_verbs.c | 12 +++++++-----
|
|
4 files changed, 20 insertions(+), 11 deletions(-)
|
|
|
|
diff --git a/providers/hns/hns_roce_u.h b/providers/hns/hns_roce_u.h
|
|
index ae9ae51..197bde9 100644
|
|
--- a/providers/hns/hns_roce_u.h
|
|
+++ b/providers/hns/hns_roce_u.h
|
|
@@ -187,6 +187,7 @@ struct hns_roce_buf {
|
|
enum hns_roce_db_type {
|
|
HNS_ROCE_QP_TYPE_DB,
|
|
HNS_ROCE_CQ_TYPE_DB,
|
|
+ HNS_ROCE_SRQ_TYPE_DB,
|
|
HNS_ROCE_DB_TYPE_NUM
|
|
};
|
|
|
|
@@ -351,7 +352,8 @@ struct hns_roce_srq {
|
|
unsigned int max_gs;
|
|
unsigned int rsv_sge;
|
|
unsigned int wqe_shift;
|
|
- unsigned int *db;
|
|
+ unsigned int *rdb;
|
|
+ unsigned int cap_flags;
|
|
unsigned short counter;
|
|
struct list_node xrc_srcq_node;
|
|
};
|
|
diff --git a/providers/hns/hns_roce_u_db.c b/providers/hns/hns_roce_u_db.c
|
|
index 73a71de..bbef988 100644
|
|
--- a/providers/hns/hns_roce_u_db.c
|
|
+++ b/providers/hns/hns_roce_u_db.c
|
|
@@ -41,6 +41,7 @@
|
|
static const unsigned int db_size[] = {
|
|
[HNS_ROCE_QP_TYPE_DB] = 4,
|
|
[HNS_ROCE_CQ_TYPE_DB] = 4,
|
|
+ [HNS_ROCE_SRQ_TYPE_DB] = 4,
|
|
};
|
|
|
|
static struct hns_roce_db_page *hns_roce_add_db_page(
|
|
diff --git a/providers/hns/hns_roce_u_hw_v2.c b/providers/hns/hns_roce_u_hw_v2.c
|
|
index ac40d5d..714a34e 100644
|
|
--- a/providers/hns/hns_roce_u_hw_v2.c
|
|
+++ b/providers/hns/hns_roce_u_hw_v2.c
|
|
@@ -2109,11 +2109,15 @@ static void fill_wqe_idx(struct hns_roce_srq *srq, unsigned int wqe_idx)
|
|
idx_que->head++;
|
|
}
|
|
|
|
-static void update_srq_db(struct hns_roce_db *db, struct hns_roce_srq *srq)
|
|
+static void update_srq_db(struct hns_roce_context *ctx, struct hns_roce_db *db,
|
|
+ struct hns_roce_srq *srq)
|
|
{
|
|
hr_reg_write(db, DB_TAG, srq->srqn);
|
|
hr_reg_write(db, DB_CMD, HNS_ROCE_V2_SRQ_DB);
|
|
hr_reg_write(db, DB_PI, srq->idx_que.head);
|
|
+
|
|
+ hns_roce_write64(ctx, ctx->uar + ROCEE_VF_DB_CFG0_OFFSET,
|
|
+ (__le32 *)db);
|
|
}
|
|
|
|
static int check_srq_recv(struct hns_roce_context *ctx)
|
|
@@ -2176,10 +2180,10 @@ static int hns_roce_u_v2_post_srq_recv(struct ibv_srq *ib_srq,
|
|
*/
|
|
udma_to_device_barrier();
|
|
|
|
- update_srq_db(&srq_db, srq);
|
|
-
|
|
- hns_roce_write64(ctx, ctx->uar + ROCEE_VF_DB_CFG0_OFFSET,
|
|
- (__le32 *)&srq_db);
|
|
+ if (srq->cap_flags & HNS_ROCE_RSP_SRQ_CAP_RECORD_DB)
|
|
+ *srq->rdb = srq->idx_que.head & 0xffff;
|
|
+ else
|
|
+ update_srq_db(ctx, &srq_db, srq);
|
|
}
|
|
|
|
hns_roce_spin_unlock(&srq->hr_lock);
|
|
diff --git a/providers/hns/hns_roce_u_verbs.c b/providers/hns/hns_roce_u_verbs.c
|
|
index f76341c..1c2d94d 100644
|
|
--- a/providers/hns/hns_roce_u_verbs.c
|
|
+++ b/providers/hns/hns_roce_u_verbs.c
|
|
@@ -866,7 +866,8 @@ static int exec_srq_create_cmd(struct ibv_context *context,
|
|
|
|
cmd_ex.buf_addr = (uintptr_t)srq->wqe_buf.buf;
|
|
cmd_ex.que_addr = (uintptr_t)srq->idx_que.buf.buf;
|
|
- cmd_ex.db_addr = (uintptr_t)srq->db;
|
|
+ cmd_ex.db_addr = (uintptr_t)srq->rdb;
|
|
+ cmd_ex.req_cap_flags |= HNS_ROCE_SRQ_CAP_RECORD_DB;
|
|
|
|
ret = ibv_cmd_create_srq_ex(context, &srq->verbs_srq, init_attr,
|
|
&cmd_ex.ibv_cmd, sizeof(cmd_ex),
|
|
@@ -875,6 +876,7 @@ static int exec_srq_create_cmd(struct ibv_context *context,
|
|
return ret;
|
|
|
|
srq->srqn = resp_ex.srqn;
|
|
+ srq->cap_flags = resp_ex.cap_flags;
|
|
|
|
return 0;
|
|
}
|
|
@@ -932,8 +934,8 @@ static struct ibv_srq *create_srq(struct ibv_context *context,
|
|
if (alloc_srq_buf(srq))
|
|
goto err_free_srq;
|
|
|
|
- srq->db = hns_roce_alloc_db(hr_ctx, HNS_ROCE_QP_TYPE_DB);
|
|
- if (!srq->db)
|
|
+ srq->rdb = hns_roce_alloc_db(hr_ctx, HNS_ROCE_SRQ_TYPE_DB);
|
|
+ if (!srq->rdb)
|
|
goto err_srq_buf;
|
|
|
|
ret = exec_srq_create_cmd(context, srq, init_attr);
|
|
@@ -956,7 +958,7 @@ err_destroy_srq:
|
|
ibv_cmd_destroy_srq(&srq->verbs_srq.srq);
|
|
|
|
err_srq_db:
|
|
- hns_roce_free_db(hr_ctx, srq->db, HNS_ROCE_QP_TYPE_DB);
|
|
+ hns_roce_free_db(hr_ctx, srq->rdb, HNS_ROCE_SRQ_TYPE_DB);
|
|
|
|
err_srq_buf:
|
|
free_srq_buf(srq);
|
|
@@ -1048,7 +1050,7 @@ int hns_roce_u_destroy_srq(struct ibv_srq *ibv_srq)
|
|
|
|
hns_roce_clear_srq(ctx, srq->srqn);
|
|
|
|
- hns_roce_free_db(ctx, srq->db, HNS_ROCE_QP_TYPE_DB);
|
|
+ hns_roce_free_db(ctx, srq->rdb, HNS_ROCE_SRQ_TYPE_DB);
|
|
free_srq_buf(srq);
|
|
free(srq);
|
|
|
|
--
|
|
2.25.1
|
|
|