* [PATCH for-next 1/5] RDMA/hns: Rename wqe buffer related functions
2020-03-02 12:11 [PATCH for-next 0/5] RDMA/hns: Refactor wqe related codes Weihang Li
@ 2020-03-02 12:11 ` Weihang Li
2020-03-05 6:27 ` Leon Romanovsky
2020-03-02 12:11 ` [PATCH for-next 2/5] RDMA/hns: Optimize wqe buffer filling process for post send Weihang Li
` (3 subsequent siblings)
4 siblings, 1 reply; 18+ messages in thread
From: Weihang Li @ 2020-03-02 12:11 UTC (permalink / raw)
To: dledford, jgg; +Cc: leon, linux-rdma, linuxarm
From: Xi Wang <wangxi11@huawei.com>
There are serval global functions related to wqe buffer in the hns driver
and are called in different files. These symbols cannot directly represent
the namespace they belong to. So add prefix 'hns_roce_' to 3 wqe buffer
related global functions: get_recv_wqe(), get_send_wqe(), and
get_send_extend_sge().
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
---
drivers/infiniband/hw/hns/hns_roce_device.h | 6 +++---
drivers/infiniband/hw/hns/hns_roce_hw_v1.c | 9 +++++----
drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 10 +++++-----
drivers/infiniband/hw/hns/hns_roce_qp.c | 6 +++---
4 files changed, 16 insertions(+), 15 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index d7dcf6e..b6ae12d 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -1238,9 +1238,9 @@ struct ib_qp *hns_roce_create_qp(struct ib_pd *ib_pd,
int hns_roce_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
int attr_mask, struct ib_udata *udata);
void init_flush_work(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp);
-void *get_recv_wqe(struct hns_roce_qp *hr_qp, int n);
-void *get_send_wqe(struct hns_roce_qp *hr_qp, int n);
-void *get_send_extend_sge(struct hns_roce_qp *hr_qp, int n);
+void *hns_roce_get_recv_wqe(struct hns_roce_qp *hr_qp, int n);
+void *hns_roce_get_send_wqe(struct hns_roce_qp *hr_qp, int n);
+void *hns_roce_get_extend_sge(struct hns_roce_qp *hr_qp, int n);
bool hns_roce_wq_overflow(struct hns_roce_wq *hr_wq, int nreq,
struct ib_cq *ib_cq);
enum hns_roce_qp_state to_hns_roce_state(enum ib_qp_state state);
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
index c05a905..2e53045 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v1.c
@@ -106,7 +106,7 @@ static int hns_roce_v1_post_send(struct ib_qp *ibqp,
goto out;
}
- wqe = get_send_wqe(qp, wqe_idx);
+ wqe = hns_roce_get_send_wqe(qp, wqe_idx);
qp->sq.wrid[wqe_idx] = wr->wr_id;
/* Corresponding to the RC and RD type wqe process separately */
@@ -378,7 +378,7 @@ static int hns_roce_v1_post_recv(struct ib_qp *ibqp,
goto out;
}
- ctrl = get_recv_wqe(hr_qp, wqe_idx);
+ ctrl = hns_roce_get_recv_wqe(hr_qp, wqe_idx);
roce_set_field(ctrl->rwqe_byte_12,
RQ_WQE_CTRL_RWQE_BYTE_12_RWQE_SGE_NUM_M,
@@ -2284,9 +2284,10 @@ static int hns_roce_v1_poll_one(struct hns_roce_cq *hr_cq,
if (is_send) {
/* SQ conrespond to CQE */
- sq_wqe = get_send_wqe(*cur_qp, roce_get_field(cqe->cqe_byte_4,
+ sq_wqe = hns_roce_get_send_wqe(*cur_qp,
+ roce_get_field(cqe->cqe_byte_4,
CQE_BYTE_4_WQE_INDEX_M,
- CQE_BYTE_4_WQE_INDEX_S)&
+ CQE_BYTE_4_WQE_INDEX_S) &
((*cur_qp)->sq.wqe_cnt-1));
switch (le32_to_cpu(sq_wqe->flag) & HNS_ROCE_WQE_OPCODE_MASK) {
case HNS_ROCE_WQE_OPCODE_SEND:
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 82021fa..88d671a 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -127,7 +127,7 @@ static void set_extend_sge(struct hns_roce_qp *qp, const struct ib_send_wr *wr,
* should calculate how many sges in the first page and the second
* page.
*/
- dseg = get_send_extend_sge(qp, (*sge_ind) & (qp->sge.sge_cnt - 1));
+ dseg = hns_roce_get_extend_sge(qp, (*sge_ind) & (qp->sge.sge_cnt - 1));
fi_sge_num = (round_up((uintptr_t)dseg, 1 << shift) -
(uintptr_t)dseg) /
sizeof(struct hns_roce_v2_wqe_data_seg);
@@ -137,7 +137,7 @@ static void set_extend_sge(struct hns_roce_qp *qp, const struct ib_send_wr *wr,
set_data_seg_v2(dseg++, sg + i);
(*sge_ind)++;
}
- dseg = get_send_extend_sge(qp,
+ dseg = hns_roce_get_extend_sge(qp,
(*sge_ind) & (qp->sge.sge_cnt - 1));
for (i = 0; i < se_sge_num; i++) {
set_data_seg_v2(dseg++, sg + fi_sge_num + i);
@@ -329,7 +329,7 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
goto out;
}
- wqe = get_send_wqe(qp, wqe_idx);
+ wqe = hns_roce_get_send_wqe(qp, wqe_idx);
qp->sq.wrid[wqe_idx] = wr->wr_id;
owner_bit =
~(((qp->sq.head + nreq) >> ilog2(qp->sq.wqe_cnt)) & 0x1);
@@ -676,7 +676,7 @@ static int hns_roce_v2_post_recv(struct ib_qp *ibqp,
goto out;
}
- wqe = get_recv_wqe(hr_qp, wqe_idx);
+ wqe = hns_roce_get_recv_wqe(hr_qp, wqe_idx);
dseg = (struct hns_roce_v2_wqe_data_seg *)wqe;
for (i = 0; i < wr->num_sge; i++) {
if (!wr->sg_list[i].length)
@@ -2935,7 +2935,7 @@ static int hns_roce_handle_recv_inl_wqe(struct hns_roce_v2_cqe *cqe,
sge_list = (*cur_qp)->rq_inl_buf.wqe_list[wr_cnt].sg_list;
sge_num = (*cur_qp)->rq_inl_buf.wqe_list[wr_cnt].sge_cnt;
- wqe_buf = get_recv_wqe(*cur_qp, wr_cnt);
+ wqe_buf = hns_roce_get_recv_wqe(*cur_qp, wr_cnt);
data_len = wc->byte_len;
for (sge_cnt = 0; (sge_cnt < sge_num) && (data_len); sge_cnt++) {
diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c
index 10c4354..cdc8b19 100644
--- a/drivers/infiniband/hw/hns/hns_roce_qp.c
+++ b/drivers/infiniband/hw/hns/hns_roce_qp.c
@@ -1469,17 +1469,17 @@ static void *get_wqe(struct hns_roce_qp *hr_qp, int offset)
return hns_roce_buf_offset(&hr_qp->hr_buf, offset);
}
-void *get_recv_wqe(struct hns_roce_qp *hr_qp, int n)
+void *hns_roce_get_recv_wqe(struct hns_roce_qp *hr_qp, int n)
{
return get_wqe(hr_qp, hr_qp->rq.offset + (n << hr_qp->rq.wqe_shift));
}
-void *get_send_wqe(struct hns_roce_qp *hr_qp, int n)
+void *hns_roce_get_send_wqe(struct hns_roce_qp *hr_qp, int n)
{
return get_wqe(hr_qp, hr_qp->sq.offset + (n << hr_qp->sq.wqe_shift));
}
-void *get_send_extend_sge(struct hns_roce_qp *hr_qp, int n)
+void *hns_roce_get_extend_sge(struct hns_roce_qp *hr_qp, int n)
{
return hns_roce_buf_offset(&hr_qp->hr_buf, hr_qp->sge.offset +
(n << hr_qp->sge.sge_shift));
--
2.8.1
^ permalink raw reply related [flat|nested] 18+ messages in thread* Re: [PATCH for-next 1/5] RDMA/hns: Rename wqe buffer related functions
2020-03-02 12:11 ` [PATCH for-next 1/5] RDMA/hns: Rename wqe buffer related functions Weihang Li
@ 2020-03-05 6:27 ` Leon Romanovsky
0 siblings, 0 replies; 18+ messages in thread
From: Leon Romanovsky @ 2020-03-05 6:27 UTC (permalink / raw)
To: Weihang Li; +Cc: dledford, jgg, linux-rdma, linuxarm
On Mon, Mar 02, 2020 at 08:11:29PM +0800, Weihang Li wrote:
> From: Xi Wang <wangxi11@huawei.com>
>
> There are serval global functions related to wqe buffer in the hns driver
> and are called in different files. These symbols cannot directly represent
> the namespace they belong to. So add prefix 'hns_roce_' to 3 wqe buffer
> related global functions: get_recv_wqe(), get_send_wqe(), and
> get_send_extend_sge().
>
> Signed-off-by: Xi Wang <wangxi11@huawei.com>
> Signed-off-by: Weihang Li <liweihang@huawei.com>
> ---
> drivers/infiniband/hw/hns/hns_roce_device.h | 6 +++---
> drivers/infiniband/hw/hns/hns_roce_hw_v1.c | 9 +++++----
> drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 10 +++++-----
> drivers/infiniband/hw/hns/hns_roce_qp.c | 6 +++---
> 4 files changed, 16 insertions(+), 15 deletions(-)
>
Thanks,
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
^ permalink raw reply [flat|nested] 18+ messages in thread
* [PATCH for-next 2/5] RDMA/hns: Optimize wqe buffer filling process for post send
2020-03-02 12:11 [PATCH for-next 0/5] RDMA/hns: Refactor wqe related codes Weihang Li
2020-03-02 12:11 ` [PATCH for-next 1/5] RDMA/hns: Rename wqe buffer related functions Weihang Li
@ 2020-03-02 12:11 ` Weihang Li
2020-03-05 6:32 ` Leon Romanovsky
2020-03-02 12:11 ` [PATCH for-next 3/5] RDMA/hns: Optimize the wr opcode conversion from ib to hns Weihang Li
` (2 subsequent siblings)
4 siblings, 1 reply; 18+ messages in thread
From: Weihang Li @ 2020-03-02 12:11 UTC (permalink / raw)
To: dledford, jgg; +Cc: leon, linux-rdma, linuxarm
From: Xi Wang <wangxi11@huawei.com>
Encapsulates the wqe buffer process details for datagram seg, fast mr seg
and atomic seg.
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
---
drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 63 +++++++++++++++---------------
1 file changed, 32 insertions(+), 31 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 88d671a..c8c345f 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -57,10 +57,10 @@ static void set_data_seg_v2(struct hns_roce_v2_wqe_data_seg *dseg,
}
static void set_frmr_seg(struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
- struct hns_roce_wqe_frmr_seg *fseg,
- const struct ib_reg_wr *wr)
+ void *wqe, const struct ib_reg_wr *wr)
{
struct hns_roce_mr *mr = to_hr_mr(wr->mr);
+ struct hns_roce_wqe_frmr_seg *fseg = wqe;
/* use ib_access_flags */
roce_set_bit(rc_sq_wqe->byte_4, V2_RC_FRMR_WQE_BYTE_4_BIND_EN_S,
@@ -92,16 +92,26 @@ static void set_frmr_seg(struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
V2_RC_FRMR_WQE_BYTE_40_BLK_MODE_S, 0);
}
-static void set_atomic_seg(struct hns_roce_wqe_atomic_seg *aseg,
- const struct ib_atomic_wr *wr)
+static void set_atomic_seg(const struct ib_send_wr *wr, void *wqe,
+ struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
+ int valid_num_sge)
{
- if (wr->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
- aseg->fetchadd_swap_data = cpu_to_le64(wr->swap);
- aseg->cmp_data = cpu_to_le64(wr->compare_add);
+ struct hns_roce_wqe_atomic_seg *aseg;
+
+ set_data_seg_v2(wqe, wr->sg_list);
+ aseg = wqe + sizeof(struct hns_roce_v2_wqe_data_seg);
+
+ if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
+ aseg->fetchadd_swap_data = cpu_to_le64(atomic_wr(wr)->swap);
+ aseg->cmp_data = cpu_to_le64(atomic_wr(wr)->compare_add);
} else {
- aseg->fetchadd_swap_data = cpu_to_le64(wr->compare_add);
+ aseg->fetchadd_swap_data =
+ cpu_to_le64(atomic_wr(wr)->compare_add);
aseg->cmp_data = 0;
}
+
+ roce_set_field(rc_sq_wqe->byte_16, V2_RC_SEND_WQE_BYTE_16_SGE_NUM_M,
+ V2_RC_SEND_WQE_BYTE_16_SGE_NUM_S, valid_num_sge);
}
static void set_extend_sge(struct hns_roce_qp *qp, const struct ib_send_wr *wr,
@@ -154,11 +164,11 @@ static void set_extend_sge(struct hns_roce_qp *qp, const struct ib_send_wr *wr,
static int set_rwqe_data_seg(struct ib_qp *ibqp, const struct ib_send_wr *wr,
struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
void *wqe, unsigned int *sge_ind,
- int valid_num_sge,
- const struct ib_send_wr **bad_wr)
+ int valid_num_sge)
{
struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
struct hns_roce_v2_wqe_data_seg *dseg = wqe;
+ struct ib_device *ibdev = &hr_dev->ib_dev;
struct hns_roce_qp *qp = to_hr_qp(ibqp);
int j = 0;
int i;
@@ -166,15 +176,14 @@ static int set_rwqe_data_seg(struct ib_qp *ibqp, const struct ib_send_wr *wr,
if (wr->send_flags & IB_SEND_INLINE && valid_num_sge) {
if (le32_to_cpu(rc_sq_wqe->msg_len) >
hr_dev->caps.max_sq_inline) {
- *bad_wr = wr;
- dev_err(hr_dev->dev, "inline len(1-%d)=%d, illegal",
- rc_sq_wqe->msg_len, hr_dev->caps.max_sq_inline);
+ ibdev_err(ibdev, "inline len(1-%d)=%d, illegal",
+ rc_sq_wqe->msg_len,
+ hr_dev->caps.max_sq_inline);
return -EINVAL;
}
if (wr->opcode == IB_WR_RDMA_READ) {
- *bad_wr = wr;
- dev_err(hr_dev->dev, "Not support inline data!\n");
+ ibdev_err(ibdev, "Not support inline data!\n");
return -EINVAL;
}
@@ -285,7 +294,6 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
struct hns_roce_v2_ud_send_wqe *ud_sq_wqe;
struct hns_roce_v2_rc_send_wqe *rc_sq_wqe;
struct hns_roce_qp *qp = to_hr_qp(ibqp);
- struct hns_roce_wqe_frmr_seg *fseg;
struct device *dev = hr_dev->dev;
unsigned int owner_bit;
unsigned int sge_idx;
@@ -547,8 +555,7 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
break;
case IB_WR_REG_MR:
hr_op = HNS_ROCE_V2_WQE_OP_FAST_REG_PMR;
- fseg = wqe;
- set_frmr_seg(rc_sq_wqe, fseg, reg_wr(wr));
+ set_frmr_seg(rc_sq_wqe, wqe, reg_wr(wr));
break;
case IB_WR_ATOMIC_CMP_AND_SWP:
hr_op = HNS_ROCE_V2_WQE_OP_ATOM_CMP_AND_SWAP;
@@ -582,23 +589,17 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
V2_RC_SEND_WQE_BYTE_4_OPCODE_S, hr_op);
if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
- wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
- struct hns_roce_v2_wqe_data_seg *dseg;
-
- dseg = wqe;
- set_data_seg_v2(dseg, wr->sg_list);
- wqe += sizeof(struct hns_roce_v2_wqe_data_seg);
- set_atomic_seg(wqe, atomic_wr(wr));
- roce_set_field(rc_sq_wqe->byte_16,
- V2_RC_SEND_WQE_BYTE_16_SGE_NUM_M,
- V2_RC_SEND_WQE_BYTE_16_SGE_NUM_S,
+ wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
+ set_atomic_seg(wr, wqe, rc_sq_wqe,
valid_num_sge);
- } else if (wr->opcode != IB_WR_REG_MR) {
+ else if (wr->opcode != IB_WR_REG_MR) {
ret = set_rwqe_data_seg(ibqp, wr, rc_sq_wqe,
wqe, &sge_idx,
- valid_num_sge, bad_wr);
- if (ret)
+ valid_num_sge);
+ if (ret) {
+ *bad_wr = wr;
goto out;
+ }
}
} else {
dev_err(dev, "Illegal qp_type(0x%x)\n", ibqp->qp_type);
--
2.8.1
^ permalink raw reply related [flat|nested] 18+ messages in thread* Re: [PATCH for-next 2/5] RDMA/hns: Optimize wqe buffer filling process for post send
2020-03-02 12:11 ` [PATCH for-next 2/5] RDMA/hns: Optimize wqe buffer filling process for post send Weihang Li
@ 2020-03-05 6:32 ` Leon Romanovsky
0 siblings, 0 replies; 18+ messages in thread
From: Leon Romanovsky @ 2020-03-05 6:32 UTC (permalink / raw)
To: Weihang Li; +Cc: dledford, jgg, linux-rdma, linuxarm
On Mon, Mar 02, 2020 at 08:11:30PM +0800, Weihang Li wrote:
> From: Xi Wang <wangxi11@huawei.com>
>
> Encapsulates the wqe buffer process details for datagram seg, fast mr seg
> and atomic seg.
>
> Signed-off-by: Xi Wang <wangxi11@huawei.com>
> Signed-off-by: Weihang Li <liweihang@huawei.com>
> ---
> drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 63 +++++++++++++++---------------
> 1 file changed, 32 insertions(+), 31 deletions(-)
>
Thanks,
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
^ permalink raw reply [flat|nested] 18+ messages in thread
* [PATCH for-next 3/5] RDMA/hns: Optimize the wr opcode conversion from ib to hns
2020-03-02 12:11 [PATCH for-next 0/5] RDMA/hns: Refactor wqe related codes Weihang Li
2020-03-02 12:11 ` [PATCH for-next 1/5] RDMA/hns: Rename wqe buffer related functions Weihang Li
2020-03-02 12:11 ` [PATCH for-next 2/5] RDMA/hns: Optimize wqe buffer filling process for post send Weihang Li
@ 2020-03-02 12:11 ` Weihang Li
2020-03-05 6:18 ` Leon Romanovsky
2020-03-02 12:11 ` [PATCH for-next 4/5] RDMA/hns: Optimize base address table config flow for qp buffer Weihang Li
2020-03-02 12:11 ` [PATCH for-next 5/5] RDMA/hns: Optimize wqe buffer set flow for post send Weihang Li
4 siblings, 1 reply; 18+ messages in thread
From: Weihang Li @ 2020-03-02 12:11 UTC (permalink / raw)
To: dledford, jgg; +Cc: leon, linux-rdma, linuxarm
From: Xi Wang <wangxi11@huawei.com>
Simplify the wr opcode conversion from ib to hns by using a map table
instead of the switch-case statement.
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
---
drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 70 ++++++++++++++++++------------
1 file changed, 43 insertions(+), 27 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index c8c345f..ea61ccb 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -56,6 +56,47 @@ static void set_data_seg_v2(struct hns_roce_v2_wqe_data_seg *dseg,
dseg->len = cpu_to_le32(sg->length);
}
+/*
+ * mapped-value = 1 + real-value
+ * The hns wr opcode real value is start from 0, In order to distinguish between
+ * initialized and uninitialized map values, we plus 1 to the actual value when
+ * defining the mapping, so that the validity can be identified by checking the
+ * mapped value is greater than 0.
+ */
+#define HR_OPC_MAP(ib_key, hr_key) \
+ [IB_WR_ ## ib_key] = 1 + HNS_ROCE_V2_WQE_OP_ ## hr_key
+
+static const u32 hns_roce_op_code[] = {
+ HR_OPC_MAP(RDMA_WRITE, RDMA_WRITE),
+ HR_OPC_MAP(RDMA_WRITE_WITH_IMM, RDMA_WRITE_WITH_IMM),
+ HR_OPC_MAP(SEND, SEND),
+ HR_OPC_MAP(SEND_WITH_IMM, SEND_WITH_IMM),
+ HR_OPC_MAP(RDMA_READ, RDMA_READ),
+ HR_OPC_MAP(ATOMIC_CMP_AND_SWP, ATOM_CMP_AND_SWAP),
+ HR_OPC_MAP(ATOMIC_FETCH_AND_ADD, ATOM_FETCH_AND_ADD),
+ HR_OPC_MAP(SEND_WITH_INV, SEND_WITH_INV),
+ HR_OPC_MAP(LOCAL_INV, LOCAL_INV),
+ HR_OPC_MAP(MASKED_ATOMIC_CMP_AND_SWP, ATOM_MSK_CMP_AND_SWAP),
+ HR_OPC_MAP(MASKED_ATOMIC_FETCH_AND_ADD, ATOM_MSK_FETCH_AND_ADD),
+ HR_OPC_MAP(REG_MR, FAST_REG_PMR),
+ [IB_WR_RESERVED1] = 0,
+};
+
+static inline u32 to_hr_opcode(u32 ib_opcode)
+{
+ u32 hr_opcode = 0;
+
+ if (ib_opcode < IB_WR_RESERVED1)
+ hr_opcode = hns_roce_op_code[ib_opcode];
+
+ /* exist a valid mapping definition for ib code */
+ if (hr_opcode > 0)
+ return hr_opcode - 1;
+
+ /* default hns roce wr opcode */
+ return HNS_ROCE_V2_WQE_OP_MASK;
+}
+
static void set_frmr_seg(struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
void *wqe, const struct ib_reg_wr *wr)
{
@@ -303,7 +344,6 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
void *wqe = NULL;
bool loopback;
u32 tmp_len;
- u32 hr_op;
u8 *smac;
int nreq;
int ret;
@@ -517,76 +557,52 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
wqe += sizeof(struct hns_roce_v2_rc_send_wqe);
switch (wr->opcode) {
case IB_WR_RDMA_READ:
- hr_op = HNS_ROCE_V2_WQE_OP_RDMA_READ;
rc_sq_wqe->rkey =
cpu_to_le32(rdma_wr(wr)->rkey);
rc_sq_wqe->va =
cpu_to_le64(rdma_wr(wr)->remote_addr);
break;
case IB_WR_RDMA_WRITE:
- hr_op = HNS_ROCE_V2_WQE_OP_RDMA_WRITE;
rc_sq_wqe->rkey =
cpu_to_le32(rdma_wr(wr)->rkey);
rc_sq_wqe->va =
cpu_to_le64(rdma_wr(wr)->remote_addr);
break;
case IB_WR_RDMA_WRITE_WITH_IMM:
- hr_op = HNS_ROCE_V2_WQE_OP_RDMA_WRITE_WITH_IMM;
rc_sq_wqe->rkey =
cpu_to_le32(rdma_wr(wr)->rkey);
rc_sq_wqe->va =
cpu_to_le64(rdma_wr(wr)->remote_addr);
break;
- case IB_WR_SEND:
- hr_op = HNS_ROCE_V2_WQE_OP_SEND;
- break;
- case IB_WR_SEND_WITH_INV:
- hr_op = HNS_ROCE_V2_WQE_OP_SEND_WITH_INV;
- break;
- case IB_WR_SEND_WITH_IMM:
- hr_op = HNS_ROCE_V2_WQE_OP_SEND_WITH_IMM;
- break;
case IB_WR_LOCAL_INV:
- hr_op = HNS_ROCE_V2_WQE_OP_LOCAL_INV;
roce_set_bit(rc_sq_wqe->byte_4,
V2_RC_SEND_WQE_BYTE_4_SO_S, 1);
rc_sq_wqe->inv_key =
cpu_to_le32(wr->ex.invalidate_rkey);
break;
case IB_WR_REG_MR:
- hr_op = HNS_ROCE_V2_WQE_OP_FAST_REG_PMR;
set_frmr_seg(rc_sq_wqe, wqe, reg_wr(wr));
break;
case IB_WR_ATOMIC_CMP_AND_SWP:
- hr_op = HNS_ROCE_V2_WQE_OP_ATOM_CMP_AND_SWAP;
rc_sq_wqe->rkey =
cpu_to_le32(atomic_wr(wr)->rkey);
rc_sq_wqe->va =
cpu_to_le64(atomic_wr(wr)->remote_addr);
break;
case IB_WR_ATOMIC_FETCH_AND_ADD:
- hr_op = HNS_ROCE_V2_WQE_OP_ATOM_FETCH_AND_ADD;
rc_sq_wqe->rkey =
cpu_to_le32(atomic_wr(wr)->rkey);
rc_sq_wqe->va =
cpu_to_le64(atomic_wr(wr)->remote_addr);
break;
- case IB_WR_MASKED_ATOMIC_CMP_AND_SWP:
- hr_op =
- HNS_ROCE_V2_WQE_OP_ATOM_MSK_CMP_AND_SWAP;
- break;
- case IB_WR_MASKED_ATOMIC_FETCH_AND_ADD:
- hr_op =
- HNS_ROCE_V2_WQE_OP_ATOM_MSK_FETCH_AND_ADD;
- break;
default:
- hr_op = HNS_ROCE_V2_WQE_OP_MASK;
break;
}
roce_set_field(rc_sq_wqe->byte_4,
V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
- V2_RC_SEND_WQE_BYTE_4_OPCODE_S, hr_op);
+ V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
+ to_hr_opcode(wr->opcode));
if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
--
2.8.1
^ permalink raw reply related [flat|nested] 18+ messages in thread* Re: [PATCH for-next 3/5] RDMA/hns: Optimize the wr opcode conversion from ib to hns
2020-03-02 12:11 ` [PATCH for-next 3/5] RDMA/hns: Optimize the wr opcode conversion from ib to hns Weihang Li
@ 2020-03-05 6:18 ` Leon Romanovsky
2020-03-05 11:22 ` liweihang
2020-03-05 13:24 ` Jason Gunthorpe
0 siblings, 2 replies; 18+ messages in thread
From: Leon Romanovsky @ 2020-03-05 6:18 UTC (permalink / raw)
To: Weihang Li; +Cc: dledford, jgg, linux-rdma, linuxarm
On Mon, Mar 02, 2020 at 08:11:31PM +0800, Weihang Li wrote:
> From: Xi Wang <wangxi11@huawei.com>
>
> Simplify the wr opcode conversion from ib to hns by using a map table
> instead of the switch-case statement.
>
> Signed-off-by: Xi Wang <wangxi11@huawei.com>
> Signed-off-by: Weihang Li <liweihang@huawei.com>
> ---
> drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 70 ++++++++++++++++++------------
> 1 file changed, 43 insertions(+), 27 deletions(-)
>
> diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
> index c8c345f..ea61ccb 100644
> --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
> +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
> @@ -56,6 +56,47 @@ static void set_data_seg_v2(struct hns_roce_v2_wqe_data_seg *dseg,
> dseg->len = cpu_to_le32(sg->length);
> }
>
> +/*
> + * mapped-value = 1 + real-value
> + * The hns wr opcode real value is start from 0, In order to distinguish between
> + * initialized and uninitialized map values, we plus 1 to the actual value when
> + * defining the mapping, so that the validity can be identified by checking the
> + * mapped value is greater than 0.
> + */
> +#define HR_OPC_MAP(ib_key, hr_key) \
> + [IB_WR_ ## ib_key] = 1 + HNS_ROCE_V2_WQE_OP_ ## hr_key
> +
> +static const u32 hns_roce_op_code[] = {
> + HR_OPC_MAP(RDMA_WRITE, RDMA_WRITE),
> + HR_OPC_MAP(RDMA_WRITE_WITH_IMM, RDMA_WRITE_WITH_IMM),
> + HR_OPC_MAP(SEND, SEND),
> + HR_OPC_MAP(SEND_WITH_IMM, SEND_WITH_IMM),
> + HR_OPC_MAP(RDMA_READ, RDMA_READ),
> + HR_OPC_MAP(ATOMIC_CMP_AND_SWP, ATOM_CMP_AND_SWAP),
> + HR_OPC_MAP(ATOMIC_FETCH_AND_ADD, ATOM_FETCH_AND_ADD),
> + HR_OPC_MAP(SEND_WITH_INV, SEND_WITH_INV),
> + HR_OPC_MAP(LOCAL_INV, LOCAL_INV),
> + HR_OPC_MAP(MASKED_ATOMIC_CMP_AND_SWP, ATOM_MSK_CMP_AND_SWAP),
> + HR_OPC_MAP(MASKED_ATOMIC_FETCH_AND_ADD, ATOM_MSK_FETCH_AND_ADD),
> + HR_OPC_MAP(REG_MR, FAST_REG_PMR),
> + [IB_WR_RESERVED1] = 0,
hns_roce_op_code[] is declared as static, everything is initialized to
0, there is no need to set 0 again.
> +};
> +
> +static inline u32 to_hr_opcode(u32 ib_opcode)
No inline functions in *.c, please.
> +{
> + u32 hr_opcode = 0;
> +
> + if (ib_opcode < IB_WR_RESERVED1)
if (ib_opcode > ARRAY_SIZE(hns_roce_op_code) - 1)
return HNS_ROCE_V2_WQE_OP_MASK;
return hns_roce_op_code[ib_opcode];
> + hr_opcode = hns_roce_op_code[ib_opcode];
> +
> + /* exist a valid mapping definition for ib code */
> + if (hr_opcode > 0)
> + return hr_opcode - 1;
> +
> + /* default hns roce wr opcode */
> + return HNS_ROCE_V2_WQE_OP_MASK;
> +}
> +
> static void set_frmr_seg(struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
> void *wqe, const struct ib_reg_wr *wr)
> {
> @@ -303,7 +344,6 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
> void *wqe = NULL;
> bool loopback;
> u32 tmp_len;
> - u32 hr_op;
> u8 *smac;
> int nreq;
> int ret;
> @@ -517,76 +557,52 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
> wqe += sizeof(struct hns_roce_v2_rc_send_wqe);
> switch (wr->opcode) {
> case IB_WR_RDMA_READ:
> - hr_op = HNS_ROCE_V2_WQE_OP_RDMA_READ;
> rc_sq_wqe->rkey =
> cpu_to_le32(rdma_wr(wr)->rkey);
> rc_sq_wqe->va =
> cpu_to_le64(rdma_wr(wr)->remote_addr);
> break;
> case IB_WR_RDMA_WRITE:
> - hr_op = HNS_ROCE_V2_WQE_OP_RDMA_WRITE;
> rc_sq_wqe->rkey =
> cpu_to_le32(rdma_wr(wr)->rkey);
> rc_sq_wqe->va =
> cpu_to_le64(rdma_wr(wr)->remote_addr);
> break;
> case IB_WR_RDMA_WRITE_WITH_IMM:
> - hr_op = HNS_ROCE_V2_WQE_OP_RDMA_WRITE_WITH_IMM;
> rc_sq_wqe->rkey =
> cpu_to_le32(rdma_wr(wr)->rkey);
> rc_sq_wqe->va =
> cpu_to_le64(rdma_wr(wr)->remote_addr);
> break;
> - case IB_WR_SEND:
> - hr_op = HNS_ROCE_V2_WQE_OP_SEND;
> - break;
> - case IB_WR_SEND_WITH_INV:
> - hr_op = HNS_ROCE_V2_WQE_OP_SEND_WITH_INV;
> - break;
> - case IB_WR_SEND_WITH_IMM:
> - hr_op = HNS_ROCE_V2_WQE_OP_SEND_WITH_IMM;
> - break;
> case IB_WR_LOCAL_INV:
> - hr_op = HNS_ROCE_V2_WQE_OP_LOCAL_INV;
> roce_set_bit(rc_sq_wqe->byte_4,
> V2_RC_SEND_WQE_BYTE_4_SO_S, 1);
> rc_sq_wqe->inv_key =
> cpu_to_le32(wr->ex.invalidate_rkey);
> break;
> case IB_WR_REG_MR:
> - hr_op = HNS_ROCE_V2_WQE_OP_FAST_REG_PMR;
> set_frmr_seg(rc_sq_wqe, wqe, reg_wr(wr));
> break;
> case IB_WR_ATOMIC_CMP_AND_SWP:
> - hr_op = HNS_ROCE_V2_WQE_OP_ATOM_CMP_AND_SWAP;
> rc_sq_wqe->rkey =
> cpu_to_le32(atomic_wr(wr)->rkey);
> rc_sq_wqe->va =
> cpu_to_le64(atomic_wr(wr)->remote_addr);
> break;
> case IB_WR_ATOMIC_FETCH_AND_ADD:
> - hr_op = HNS_ROCE_V2_WQE_OP_ATOM_FETCH_AND_ADD;
> rc_sq_wqe->rkey =
> cpu_to_le32(atomic_wr(wr)->rkey);
> rc_sq_wqe->va =
> cpu_to_le64(atomic_wr(wr)->remote_addr);
> break;
> - case IB_WR_MASKED_ATOMIC_CMP_AND_SWP:
> - hr_op =
> - HNS_ROCE_V2_WQE_OP_ATOM_MSK_CMP_AND_SWAP;
> - break;
> - case IB_WR_MASKED_ATOMIC_FETCH_AND_ADD:
> - hr_op =
> - HNS_ROCE_V2_WQE_OP_ATOM_MSK_FETCH_AND_ADD;
> - break;
> default:
> - hr_op = HNS_ROCE_V2_WQE_OP_MASK;
> break;
> }
>
> roce_set_field(rc_sq_wqe->byte_4,
> V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
> - V2_RC_SEND_WQE_BYTE_4_OPCODE_S, hr_op);
> + V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
> + to_hr_opcode(wr->opcode));
>
> if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
> wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
> --
> 2.8.1
>
^ permalink raw reply [flat|nested] 18+ messages in thread* Re: [PATCH for-next 3/5] RDMA/hns: Optimize the wr opcode conversion from ib to hns
2020-03-05 6:18 ` Leon Romanovsky
@ 2020-03-05 11:22 ` liweihang
2020-03-05 12:09 ` Leon Romanovsky
2020-03-05 13:24 ` Jason Gunthorpe
1 sibling, 1 reply; 18+ messages in thread
From: liweihang @ 2020-03-05 11:22 UTC (permalink / raw)
To: Leon Romanovsky
Cc: dledford@redhat.com, jgg@ziepe.ca, linux-rdma@vger.kernel.org,
Linuxarm
On 2020/3/5 14:18, Leon Romanovsky wrote:
> On Mon, Mar 02, 2020 at 08:11:31PM +0800, Weihang Li wrote:
>> From: Xi Wang <wangxi11@huawei.com>
>>
>> Simplify the wr opcode conversion from ib to hns by using a map table
>> instead of the switch-case statement.
>>
>> Signed-off-by: Xi Wang <wangxi11@huawei.com>
>> Signed-off-by: Weihang Li <liweihang@huawei.com>
>> ---
>> drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 70 ++++++++++++++++++------------
>> 1 file changed, 43 insertions(+), 27 deletions(-)
>>
>> diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>> index c8c345f..ea61ccb 100644
>> --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>> +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>> @@ -56,6 +56,47 @@ static void set_data_seg_v2(struct hns_roce_v2_wqe_data_seg *dseg,
>> dseg->len = cpu_to_le32(sg->length);
>> }
>>
>> +/*
>> + * mapped-value = 1 + real-value
>> + * The hns wr opcode real value is start from 0, In order to distinguish between
>> + * initialized and uninitialized map values, we plus 1 to the actual value when
>> + * defining the mapping, so that the validity can be identified by checking the
>> + * mapped value is greater than 0.
>> + */
>> +#define HR_OPC_MAP(ib_key, hr_key) \
>> + [IB_WR_ ## ib_key] = 1 + HNS_ROCE_V2_WQE_OP_ ## hr_key
>> +
>> +static const u32 hns_roce_op_code[] = {
>> + HR_OPC_MAP(RDMA_WRITE, RDMA_WRITE),
>> + HR_OPC_MAP(RDMA_WRITE_WITH_IMM, RDMA_WRITE_WITH_IMM),
>> + HR_OPC_MAP(SEND, SEND),
>> + HR_OPC_MAP(SEND_WITH_IMM, SEND_WITH_IMM),
>> + HR_OPC_MAP(RDMA_READ, RDMA_READ),
>> + HR_OPC_MAP(ATOMIC_CMP_AND_SWP, ATOM_CMP_AND_SWAP),
>> + HR_OPC_MAP(ATOMIC_FETCH_AND_ADD, ATOM_FETCH_AND_ADD),
>> + HR_OPC_MAP(SEND_WITH_INV, SEND_WITH_INV),
>> + HR_OPC_MAP(LOCAL_INV, LOCAL_INV),
>> + HR_OPC_MAP(MASKED_ATOMIC_CMP_AND_SWP, ATOM_MSK_CMP_AND_SWAP),
>> + HR_OPC_MAP(MASKED_ATOMIC_FETCH_AND_ADD, ATOM_MSK_FETCH_AND_ADD),
>> + HR_OPC_MAP(REG_MR, FAST_REG_PMR),
>> + [IB_WR_RESERVED1] = 0,
>
> hns_roce_op_code[] is declared as static, everything is initialized to
> 0, there is no need to set 0 again.
OK, thank you.
>
>> +};
>> +
>> +static inline u32 to_hr_opcode(u32 ib_opcode)
>
> No inline functions in *.c, please.
Hi Leon,
Thanks for your comments.
But I'm confused about when we should use static inline and when we should
use macros if a function is only used in a *.c. A few days ago, Jason
suggested me to use static inline functions, you can check the link below:
https://patchwork.kernel.org/patch/11372851/
Are there any rules about that in kernel or in our rdma subsystem? Should
I use a macro, just remove the keyword "inline" from this definition or
move this definition to .h?
>
>> +{
>> + u32 hr_opcode = 0;
>> +
>> + if (ib_opcode < IB_WR_RESERVED1)
>
> if (ib_opcode > ARRAY_SIZE(hns_roce_op_code) - 1)
> return HNS_ROCE_V2_WQE_OP_MASK;
>
> return hns_roce_op_code[ib_opcode];
>
The index of ib_key in hns_roce_op_code[] is not continuous, so there
are some invalid ib_wr_opcode for hns between the valid index.
For hardware of HIP08, HNS_ROCE_V2_WQE_OP_MASK means invalid opcode but
not zero. So we have to check if the ib_wr_opcode has a mapping value in
hns_roce_op_code[], and if the mapping result is zero, we have to return
HNS_ROCE_V2_WQE_OP_MASK. Is it ok like this?
{
u32 hr_opcode = 0;
if (ib_opcode > ARRAY_SIZE(hns_roce_op_code) - 1)
return HNS_ROCE_V2_WQE_OP_MASK;
hr_opcode = hns_roce_op_code[ib_opcode];
/* exist a valid mapping definition for ib code */
if (hr_opcode > 0)
return hr_opcode - 1;
else
return HNS_ROCE_V2_WQE_OP_MASK;
}
Thanks,
Weihang
>
>> + hr_opcode = hns_roce_op_code[ib_opcode];
>> +
>> + /* exist a valid mapping definition for ib code */
>> + if (hr_opcode > 0)
>> + return hr_opcode - 1;
>> +
>> + /* default hns roce wr opcode */
>> + return HNS_ROCE_V2_WQE_OP_MASK;
>> +}
>> +
>> static void set_frmr_seg(struct hns_roce_v2_rc_send_wqe *rc_sq_wqe,
>> void *wqe, const struct ib_reg_wr *wr)
>> {
>> @@ -303,7 +344,6 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
>> void *wqe = NULL;
>> bool loopback;
>> u32 tmp_len;
>> - u32 hr_op;
>> u8 *smac;
>> int nreq;
>> int ret;
>> @@ -517,76 +557,52 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
>> wqe += sizeof(struct hns_roce_v2_rc_send_wqe);
>> switch (wr->opcode) {
>> case IB_WR_RDMA_READ:
>> - hr_op = HNS_ROCE_V2_WQE_OP_RDMA_READ;
>> rc_sq_wqe->rkey =
>> cpu_to_le32(rdma_wr(wr)->rkey);
>> rc_sq_wqe->va =
>> cpu_to_le64(rdma_wr(wr)->remote_addr);
>> break;
>> case IB_WR_RDMA_WRITE:
>> - hr_op = HNS_ROCE_V2_WQE_OP_RDMA_WRITE;
>> rc_sq_wqe->rkey =
>> cpu_to_le32(rdma_wr(wr)->rkey);
>> rc_sq_wqe->va =
>> cpu_to_le64(rdma_wr(wr)->remote_addr);
>> break;
>> case IB_WR_RDMA_WRITE_WITH_IMM:
>> - hr_op = HNS_ROCE_V2_WQE_OP_RDMA_WRITE_WITH_IMM;
>> rc_sq_wqe->rkey =
>> cpu_to_le32(rdma_wr(wr)->rkey);
>> rc_sq_wqe->va =
>> cpu_to_le64(rdma_wr(wr)->remote_addr);
>> break;
>> - case IB_WR_SEND:
>> - hr_op = HNS_ROCE_V2_WQE_OP_SEND;
>> - break;
>> - case IB_WR_SEND_WITH_INV:
>> - hr_op = HNS_ROCE_V2_WQE_OP_SEND_WITH_INV;
>> - break;
>> - case IB_WR_SEND_WITH_IMM:
>> - hr_op = HNS_ROCE_V2_WQE_OP_SEND_WITH_IMM;
>> - break;
>> case IB_WR_LOCAL_INV:
>> - hr_op = HNS_ROCE_V2_WQE_OP_LOCAL_INV;
>> roce_set_bit(rc_sq_wqe->byte_4,
>> V2_RC_SEND_WQE_BYTE_4_SO_S, 1);
>> rc_sq_wqe->inv_key =
>> cpu_to_le32(wr->ex.invalidate_rkey);
>> break;
>> case IB_WR_REG_MR:
>> - hr_op = HNS_ROCE_V2_WQE_OP_FAST_REG_PMR;
>> set_frmr_seg(rc_sq_wqe, wqe, reg_wr(wr));
>> break;
>> case IB_WR_ATOMIC_CMP_AND_SWP:
>> - hr_op = HNS_ROCE_V2_WQE_OP_ATOM_CMP_AND_SWAP;
>> rc_sq_wqe->rkey =
>> cpu_to_le32(atomic_wr(wr)->rkey);
>> rc_sq_wqe->va =
>> cpu_to_le64(atomic_wr(wr)->remote_addr);
>> break;
>> case IB_WR_ATOMIC_FETCH_AND_ADD:
>> - hr_op = HNS_ROCE_V2_WQE_OP_ATOM_FETCH_AND_ADD;
>> rc_sq_wqe->rkey =
>> cpu_to_le32(atomic_wr(wr)->rkey);
>> rc_sq_wqe->va =
>> cpu_to_le64(atomic_wr(wr)->remote_addr);
>> break;
>> - case IB_WR_MASKED_ATOMIC_CMP_AND_SWP:
>> - hr_op =
>> - HNS_ROCE_V2_WQE_OP_ATOM_MSK_CMP_AND_SWAP;
>> - break;
>> - case IB_WR_MASKED_ATOMIC_FETCH_AND_ADD:
>> - hr_op =
>> - HNS_ROCE_V2_WQE_OP_ATOM_MSK_FETCH_AND_ADD;
>> - break;
>> default:
>> - hr_op = HNS_ROCE_V2_WQE_OP_MASK;
>> break;
>> }
>>
>> roce_set_field(rc_sq_wqe->byte_4,
>> V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
>> - V2_RC_SEND_WQE_BYTE_4_OPCODE_S, hr_op);
>> + V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
>> + to_hr_opcode(wr->opcode));
>>
>> if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
>> wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
>> --
>> 2.8.1
>>
>
^ permalink raw reply [flat|nested] 18+ messages in thread* Re: [PATCH for-next 3/5] RDMA/hns: Optimize the wr opcode conversion from ib to hns
2020-03-05 11:22 ` liweihang
@ 2020-03-05 12:09 ` Leon Romanovsky
2020-03-05 13:12 ` liweihang
2020-03-05 13:28 ` liweihang
0 siblings, 2 replies; 18+ messages in thread
From: Leon Romanovsky @ 2020-03-05 12:09 UTC (permalink / raw)
To: liweihang
Cc: dledford@redhat.com, jgg@ziepe.ca, linux-rdma@vger.kernel.org,
Linuxarm
On Thu, Mar 05, 2020 at 11:22:18AM +0000, liweihang wrote:
> On 2020/3/5 14:18, Leon Romanovsky wrote:
> > On Mon, Mar 02, 2020 at 08:11:31PM +0800, Weihang Li wrote:
> >> From: Xi Wang <wangxi11@huawei.com>
> >>
> >> Simplify the wr opcode conversion from ib to hns by using a map table
> >> instead of the switch-case statement.
> >>
> >> Signed-off-by: Xi Wang <wangxi11@huawei.com>
> >> Signed-off-by: Weihang Li <liweihang@huawei.com>
> >> ---
> >> drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 70 ++++++++++++++++++------------
> >> 1 file changed, 43 insertions(+), 27 deletions(-)
> >>
> >> diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
> >> index c8c345f..ea61ccb 100644
> >> --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
> >> +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
> >> @@ -56,6 +56,47 @@ static void set_data_seg_v2(struct hns_roce_v2_wqe_data_seg *dseg,
> >> dseg->len = cpu_to_le32(sg->length);
> >> }
> >>
> >> +/*
> >> + * mapped-value = 1 + real-value
> >> + * The hns wr opcode real value is start from 0, In order to distinguish between
> >> + * initialized and uninitialized map values, we plus 1 to the actual value when
> >> + * defining the mapping, so that the validity can be identified by checking the
> >> + * mapped value is greater than 0.
> >> + */
> >> +#define HR_OPC_MAP(ib_key, hr_key) \
> >> + [IB_WR_ ## ib_key] = 1 + HNS_ROCE_V2_WQE_OP_ ## hr_key
> >> +
> >> +static const u32 hns_roce_op_code[] = {
> >> + HR_OPC_MAP(RDMA_WRITE, RDMA_WRITE),
> >> + HR_OPC_MAP(RDMA_WRITE_WITH_IMM, RDMA_WRITE_WITH_IMM),
> >> + HR_OPC_MAP(SEND, SEND),
> >> + HR_OPC_MAP(SEND_WITH_IMM, SEND_WITH_IMM),
> >> + HR_OPC_MAP(RDMA_READ, RDMA_READ),
> >> + HR_OPC_MAP(ATOMIC_CMP_AND_SWP, ATOM_CMP_AND_SWAP),
> >> + HR_OPC_MAP(ATOMIC_FETCH_AND_ADD, ATOM_FETCH_AND_ADD),
> >> + HR_OPC_MAP(SEND_WITH_INV, SEND_WITH_INV),
> >> + HR_OPC_MAP(LOCAL_INV, LOCAL_INV),
> >> + HR_OPC_MAP(MASKED_ATOMIC_CMP_AND_SWP, ATOM_MSK_CMP_AND_SWAP),
> >> + HR_OPC_MAP(MASKED_ATOMIC_FETCH_AND_ADD, ATOM_MSK_FETCH_AND_ADD),
> >> + HR_OPC_MAP(REG_MR, FAST_REG_PMR),
> >> + [IB_WR_RESERVED1] = 0,
> >
> > hns_roce_op_code[] is declared as static, everything is initialized to
> > 0, there is no need to set 0 again.
>
> OK, thank you.
>
> >
> >> +};
> >> +
> >> +static inline u32 to_hr_opcode(u32 ib_opcode)
> >
> > No inline functions in *.c, please.
>
> Hi Leon,
>
> Thanks for your comments.
>
> But I'm confused about when we should use static inline and when we should
> use macros if a function is only used in a *.c. A few days ago, Jason
> suggested me to use static inline functions, you can check the link below:
>
> https://patchwork.kernel.org/patch/11372851/
>
> Are there any rules about that in kernel or in our rdma subsystem? Should
> I use a macro, just remove the keyword "inline" from this definition or
> move this definition to .h?
Just drop "inline" word from the declaration.
https://elixir.bootlin.com/linux/latest/source/Documentation/process/coding-style.rst#L882
>
> >
> >> +{
> >> + u32 hr_opcode = 0;
> >> +
> >> + if (ib_opcode < IB_WR_RESERVED1)
> >
> > if (ib_opcode > ARRAY_SIZE(hns_roce_op_code) - 1)
> > return HNS_ROCE_V2_WQE_OP_MASK;
> >
> > return hns_roce_op_code[ib_opcode];
> >
>
> The index of ib_key in hns_roce_op_code[] is not continuous, so there
> are some invalid ib_wr_opcode for hns between the valid index.
>
> For hardware of HIP08, HNS_ROCE_V2_WQE_OP_MASK means invalid opcode but
> not zero. So we have to check if the ib_wr_opcode has a mapping value in
> hns_roce_op_code[], and if the mapping result is zero, we have to return
> HNS_ROCE_V2_WQE_OP_MASK. Is it ok like this?
I didn't mean that you will use my code as is, what about this?
if (ib_opcode > ARRAY_SIZE(hns_roce_op_code) - 1)
return HNS_ROCE_V2_WQE_OP_MASK;
return hns_roce_op_code[ib_opcode] ?: HNS_ROCE_V2_WQE_OP_MASK;
Thanks
^ permalink raw reply [flat|nested] 18+ messages in thread* Re: [PATCH for-next 3/5] RDMA/hns: Optimize the wr opcode conversion from ib to hns
2020-03-05 12:09 ` Leon Romanovsky
@ 2020-03-05 13:12 ` liweihang
2020-03-05 13:28 ` liweihang
1 sibling, 0 replies; 18+ messages in thread
From: liweihang @ 2020-03-05 13:12 UTC (permalink / raw)
To: Leon Romanovsky
Cc: dledford@redhat.com, jgg@ziepe.ca, linux-rdma@vger.kernel.org,
Linuxarm
On 2020/3/5 20:09, Leon Romanovsky wrote:
> On Thu, Mar 05, 2020 at 11:22:18AM +0000, liweihang wrote:
>> On 2020/3/5 14:18, Leon Romanovsky wrote:
>>> On Mon, Mar 02, 2020 at 08:11:31PM +0800, Weihang Li wrote:
>>>> From: Xi Wang <wangxi11@huawei.com>
>>>>
>>>> Simplify the wr opcode conversion from ib to hns by using a map table
>>>> instead of the switch-case statement.
>>>>
>>>> Signed-off-by: Xi Wang <wangxi11@huawei.com>
>>>> Signed-off-by: Weihang Li <liweihang@huawei.com>
>>>> ---
>>>> drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 70 ++++++++++++++++++------------
>>>> 1 file changed, 43 insertions(+), 27 deletions(-)
>>>>
>>>> diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>>>> index c8c345f..ea61ccb 100644
>>>> --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>>>> +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>>>> @@ -56,6 +56,47 @@ static void set_data_seg_v2(struct hns_roce_v2_wqe_data_seg *dseg,
>>>> dseg->len = cpu_to_le32(sg->length);
>>>> }
>>>>
>>>> +/*
>>>> + * mapped-value = 1 + real-value
>>>> + * The hns wr opcode real value is start from 0, In order to distinguish between
>>>> + * initialized and uninitialized map values, we plus 1 to the actual value when
>>>> + * defining the mapping, so that the validity can be identified by checking the
>>>> + * mapped value is greater than 0.
>>>> + */
>>>> +#define HR_OPC_MAP(ib_key, hr_key) \
>>>> + [IB_WR_ ## ib_key] = 1 + HNS_ROCE_V2_WQE_OP_ ## hr_key
>>>> +
>>>> +static const u32 hns_roce_op_code[] = {
>>>> + HR_OPC_MAP(RDMA_WRITE, RDMA_WRITE),
>>>> + HR_OPC_MAP(RDMA_WRITE_WITH_IMM, RDMA_WRITE_WITH_IMM),
>>>> + HR_OPC_MAP(SEND, SEND),
>>>> + HR_OPC_MAP(SEND_WITH_IMM, SEND_WITH_IMM),
>>>> + HR_OPC_MAP(RDMA_READ, RDMA_READ),
>>>> + HR_OPC_MAP(ATOMIC_CMP_AND_SWP, ATOM_CMP_AND_SWAP),
>>>> + HR_OPC_MAP(ATOMIC_FETCH_AND_ADD, ATOM_FETCH_AND_ADD),
>>>> + HR_OPC_MAP(SEND_WITH_INV, SEND_WITH_INV),
>>>> + HR_OPC_MAP(LOCAL_INV, LOCAL_INV),
>>>> + HR_OPC_MAP(MASKED_ATOMIC_CMP_AND_SWP, ATOM_MSK_CMP_AND_SWAP),
>>>> + HR_OPC_MAP(MASKED_ATOMIC_FETCH_AND_ADD, ATOM_MSK_FETCH_AND_ADD),
>>>> + HR_OPC_MAP(REG_MR, FAST_REG_PMR),
>>>> + [IB_WR_RESERVED1] = 0,
>>>
>>> hns_roce_op_code[] is declared as static, everything is initialized to
>>> 0, there is no need to set 0 again.
>>
>> OK, thank you.
>>
>>>
>>>> +};
>>>> +
>>>> +static inline u32 to_hr_opcode(u32 ib_opcode)
>>>
>>> No inline functions in *.c, please.
>>
>> Hi Leon,
>>
>> Thanks for your comments.
>>
>> But I'm confused about when we should use static inline and when we should
>> use macros if a function is only used in a *.c. A few days ago, Jason
>> suggested me to use static inline functions, you can check the link below:
>>
>> https://patchwork.kernel.org/patch/11372851/
>>
>> Are there any rules about that in kernel or in our rdma subsystem? Should
>> I use a macro, just remove the keyword "inline" from this definition or
>> move this definition to .h?
>
> Just drop "inline" word from the declaration.
> https://elixir.bootlin.com/linux/latest/source/Documentation/process/coding-style.rst#L882
>
I got it, thank you :)
>>
>>>
>>>> +{
>>>> + u32 hr_opcode = 0;
>>>> +
>>>> + if (ib_opcode < IB_WR_RESERVED1)
>>>
>>> if (ib_opcode > ARRAY_SIZE(hns_roce_op_code) - 1)
>>> return HNS_ROCE_V2_WQE_OP_MASK;
>>>
>>> return hns_roce_op_code[ib_opcode];
>>>
>>
>> The index of ib_key in hns_roce_op_code[] is not continuous, so there
>> are some invalid ib_wr_opcode for hns between the valid index.
>>
>> For hardware of HIP08, HNS_ROCE_V2_WQE_OP_MASK means invalid opcode but
>> not zero. So we have to check if the ib_wr_opcode has a mapping value in
>> hns_roce_op_code[], and if the mapping result is zero, we have to return
>> HNS_ROCE_V2_WQE_OP_MASK. Is it ok like this?
>
> I didn't mean that you will use my code as is, what about this?
>
> if (ib_opcode > ARRAY_SIZE(hns_roce_op_code) - 1)
> return HNS_ROCE_V2_WQE_OP_MASK;
>
> return hns_roce_op_code[ib_opcode] ?: HNS_ROCE_V2_WQE_OP_MASK;
>
> Thanks
>
OK, thanks for your suggestions.
^ permalink raw reply [flat|nested] 18+ messages in thread* Re: [PATCH for-next 3/5] RDMA/hns: Optimize the wr opcode conversion from ib to hns
2020-03-05 12:09 ` Leon Romanovsky
2020-03-05 13:12 ` liweihang
@ 2020-03-05 13:28 ` liweihang
2020-03-05 14:30 ` Leon Romanovsky
1 sibling, 1 reply; 18+ messages in thread
From: liweihang @ 2020-03-05 13:28 UTC (permalink / raw)
To: Leon Romanovsky
Cc: dledford@redhat.com, jgg@ziepe.ca, linux-rdma@vger.kernel.org,
Linuxarm
On 2020/3/5 20:09, Leon Romanovsky wrote:
> On Thu, Mar 05, 2020 at 11:22:18AM +0000, liweihang wrote:
>> On 2020/3/5 14:18, Leon Romanovsky wrote:
>>> On Mon, Mar 02, 2020 at 08:11:31PM +0800, Weihang Li wrote:
>>>> From: Xi Wang <wangxi11@huawei.com>
>>>>
>>>> Simplify the wr opcode conversion from ib to hns by using a map table
>>>> instead of the switch-case statement.
>>>>
>>>> Signed-off-by: Xi Wang <wangxi11@huawei.com>
>>>> Signed-off-by: Weihang Li <liweihang@huawei.com>
>>>> ---
>>>> drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 70 ++++++++++++++++++------------
>>>> 1 file changed, 43 insertions(+), 27 deletions(-)
>>>>
>>>> diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>>>> index c8c345f..ea61ccb 100644
>>>> --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>>>> +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>>>> @@ -56,6 +56,47 @@ static void set_data_seg_v2(struct hns_roce_v2_wqe_data_seg *dseg,
>>>> dseg->len = cpu_to_le32(sg->length);
>>>> }
>>>>
>>>> +/*
>>>> + * mapped-value = 1 + real-value
>>>> + * The hns wr opcode real value is start from 0, In order to distinguish between
>>>> + * initialized and uninitialized map values, we plus 1 to the actual value when
>>>> + * defining the mapping, so that the validity can be identified by checking the
>>>> + * mapped value is greater than 0.
>>>> + */
>>>> +#define HR_OPC_MAP(ib_key, hr_key) \
>>>> + [IB_WR_ ## ib_key] = 1 + HNS_ROCE_V2_WQE_OP_ ## hr_key
>>>> +
>>>> +static const u32 hns_roce_op_code[] = {
>>>> + HR_OPC_MAP(RDMA_WRITE, RDMA_WRITE),
>>>> + HR_OPC_MAP(RDMA_WRITE_WITH_IMM, RDMA_WRITE_WITH_IMM),
>>>> + HR_OPC_MAP(SEND, SEND),
>>>> + HR_OPC_MAP(SEND_WITH_IMM, SEND_WITH_IMM),
>>>> + HR_OPC_MAP(RDMA_READ, RDMA_READ),
>>>> + HR_OPC_MAP(ATOMIC_CMP_AND_SWP, ATOM_CMP_AND_SWAP),
>>>> + HR_OPC_MAP(ATOMIC_FETCH_AND_ADD, ATOM_FETCH_AND_ADD),
>>>> + HR_OPC_MAP(SEND_WITH_INV, SEND_WITH_INV),
>>>> + HR_OPC_MAP(LOCAL_INV, LOCAL_INV),
>>>> + HR_OPC_MAP(MASKED_ATOMIC_CMP_AND_SWP, ATOM_MSK_CMP_AND_SWAP),
>>>> + HR_OPC_MAP(MASKED_ATOMIC_FETCH_AND_ADD, ATOM_MSK_FETCH_AND_ADD),
>>>> + HR_OPC_MAP(REG_MR, FAST_REG_PMR),
>>>> + [IB_WR_RESERVED1] = 0,
>>>
>>> hns_roce_op_code[] is declared as static, everything is initialized to
>>> 0, there is no need to set 0 again.
>>
>> OK, thank you.
>>
>>>
>>>> +};
>>>> +
>>>> +static inline u32 to_hr_opcode(u32 ib_opcode)
>>>
>>> No inline functions in *.c, please.
>>
>> Hi Leon,
>>
>> Thanks for your comments.
>>
>> But I'm confused about when we should use static inline and when we should
>> use macros if a function is only used in a *.c. A few days ago, Jason
>> suggested me to use static inline functions, you can check the link below:
>>
>> https://patchwork.kernel.org/patch/11372851/
>>
>> Are there any rules about that in kernel or in our rdma subsystem? Should
>> I use a macro, just remove the keyword "inline" from this definition or
>> move this definition to .h?
>
> Just drop "inline" word from the declaration.
> https://elixir.bootlin.com/linux/latest/source/Documentation/process/coding-style.rst#L882
>
>>
>>>
>>>> +{
>>>> + u32 hr_opcode = 0;
>>>> +
>>>> + if (ib_opcode < IB_WR_RESERVED1)
>>>
>>> if (ib_opcode > ARRAY_SIZE(hns_roce_op_code) - 1)
>>> return HNS_ROCE_V2_WQE_OP_MASK;
>>>
>>> return hns_roce_op_code[ib_opcode];
>>>
>>
>> The index of ib_key in hns_roce_op_code[] is not continuous, so there
>> are some invalid ib_wr_opcode for hns between the valid index.
>>
>> For hardware of HIP08, HNS_ROCE_V2_WQE_OP_MASK means invalid opcode but
>> not zero. So we have to check if the ib_wr_opcode has a mapping value in
>> hns_roce_op_code[], and if the mapping result is zero, we have to return
>> HNS_ROCE_V2_WQE_OP_MASK. Is it ok like this?
>
> I didn't mean that you will use my code as is, what about this?
>
> if (ib_opcode > ARRAY_SIZE(hns_roce_op_code) - 1)
> return HNS_ROCE_V2_WQE_OP_MASK;
>
> return hns_roce_op_code[ib_opcode] ?: HNS_ROCE_V2_WQE_OP_MASK;
>
> Thanks
>
One more question, should I add a Reviewed-by tag for anyone who has comments
on my patch, or I should only do this when the reviewer asked me to do it?
For example, should I add a reviewed-by tag for you in this patch? Thank you :)
^ permalink raw reply [flat|nested] 18+ messages in thread* Re: [PATCH for-next 3/5] RDMA/hns: Optimize the wr opcode conversion from ib to hns
2020-03-05 13:28 ` liweihang
@ 2020-03-05 14:30 ` Leon Romanovsky
0 siblings, 0 replies; 18+ messages in thread
From: Leon Romanovsky @ 2020-03-05 14:30 UTC (permalink / raw)
To: liweihang
Cc: dledford@redhat.com, jgg@ziepe.ca, linux-rdma@vger.kernel.org,
Linuxarm
On Thu, Mar 05, 2020 at 01:28:28PM +0000, liweihang wrote:
> On 2020/3/5 20:09, Leon Romanovsky wrote:
> > On Thu, Mar 05, 2020 at 11:22:18AM +0000, liweihang wrote:
> >> On 2020/3/5 14:18, Leon Romanovsky wrote:
> >>> On Mon, Mar 02, 2020 at 08:11:31PM +0800, Weihang Li wrote:
> >>>> From: Xi Wang <wangxi11@huawei.com>
> >>>>
> >>>> Simplify the wr opcode conversion from ib to hns by using a map table
> >>>> instead of the switch-case statement.
> >>>>
> >>>> Signed-off-by: Xi Wang <wangxi11@huawei.com>
> >>>> Signed-off-by: Weihang Li <liweihang@huawei.com>
> >>>> ---
> >>>> drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 70 ++++++++++++++++++------------
> >>>> 1 file changed, 43 insertions(+), 27 deletions(-)
> >>>>
> >>>> diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
> >>>> index c8c345f..ea61ccb 100644
> >>>> --- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
> >>>> +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
> >>>> @@ -56,6 +56,47 @@ static void set_data_seg_v2(struct hns_roce_v2_wqe_data_seg *dseg,
> >>>> dseg->len = cpu_to_le32(sg->length);
> >>>> }
> >>>>
> >>>> +/*
> >>>> + * mapped-value = 1 + real-value
> >>>> + * The hns wr opcode real value is start from 0, In order to distinguish between
> >>>> + * initialized and uninitialized map values, we plus 1 to the actual value when
> >>>> + * defining the mapping, so that the validity can be identified by checking the
> >>>> + * mapped value is greater than 0.
> >>>> + */
> >>>> +#define HR_OPC_MAP(ib_key, hr_key) \
> >>>> + [IB_WR_ ## ib_key] = 1 + HNS_ROCE_V2_WQE_OP_ ## hr_key
> >>>> +
> >>>> +static const u32 hns_roce_op_code[] = {
> >>>> + HR_OPC_MAP(RDMA_WRITE, RDMA_WRITE),
> >>>> + HR_OPC_MAP(RDMA_WRITE_WITH_IMM, RDMA_WRITE_WITH_IMM),
> >>>> + HR_OPC_MAP(SEND, SEND),
> >>>> + HR_OPC_MAP(SEND_WITH_IMM, SEND_WITH_IMM),
> >>>> + HR_OPC_MAP(RDMA_READ, RDMA_READ),
> >>>> + HR_OPC_MAP(ATOMIC_CMP_AND_SWP, ATOM_CMP_AND_SWAP),
> >>>> + HR_OPC_MAP(ATOMIC_FETCH_AND_ADD, ATOM_FETCH_AND_ADD),
> >>>> + HR_OPC_MAP(SEND_WITH_INV, SEND_WITH_INV),
> >>>> + HR_OPC_MAP(LOCAL_INV, LOCAL_INV),
> >>>> + HR_OPC_MAP(MASKED_ATOMIC_CMP_AND_SWP, ATOM_MSK_CMP_AND_SWAP),
> >>>> + HR_OPC_MAP(MASKED_ATOMIC_FETCH_AND_ADD, ATOM_MSK_FETCH_AND_ADD),
> >>>> + HR_OPC_MAP(REG_MR, FAST_REG_PMR),
> >>>> + [IB_WR_RESERVED1] = 0,
> >>>
> >>> hns_roce_op_code[] is declared as static, everything is initialized to
> >>> 0, there is no need to set 0 again.
> >>
> >> OK, thank you.
> >>
> >>>
> >>>> +};
> >>>> +
> >>>> +static inline u32 to_hr_opcode(u32 ib_opcode)
> >>>
> >>> No inline functions in *.c, please.
> >>
> >> Hi Leon,
> >>
> >> Thanks for your comments.
> >>
> >> But I'm confused about when we should use static inline and when we should
> >> use macros if a function is only used in a *.c. A few days ago, Jason
> >> suggested me to use static inline functions, you can check the link below:
> >>
> >> https://patchwork.kernel.org/patch/11372851/
> >>
> >> Are there any rules about that in kernel or in our rdma subsystem? Should
> >> I use a macro, just remove the keyword "inline" from this definition or
> >> move this definition to .h?
> >
> > Just drop "inline" word from the declaration.
> > https://elixir.bootlin.com/linux/latest/source/Documentation/process/coding-style.rst#L882
> >
> >>
> >>>
> >>>> +{
> >>>> + u32 hr_opcode = 0;
> >>>> +
> >>>> + if (ib_opcode < IB_WR_RESERVED1)
> >>>
> >>> if (ib_opcode > ARRAY_SIZE(hns_roce_op_code) - 1)
> >>> return HNS_ROCE_V2_WQE_OP_MASK;
> >>>
> >>> return hns_roce_op_code[ib_opcode];
> >>>
> >>
> >> The index of ib_key in hns_roce_op_code[] is not continuous, so there
> >> are some invalid ib_wr_opcode for hns between the valid index.
> >>
> >> For hardware of HIP08, HNS_ROCE_V2_WQE_OP_MASK means invalid opcode but
> >> not zero. So we have to check if the ib_wr_opcode has a mapping value in
> >> hns_roce_op_code[], and if the mapping result is zero, we have to return
> >> HNS_ROCE_V2_WQE_OP_MASK. Is it ok like this?
> >
> > I didn't mean that you will use my code as is, what about this?
> >
> > if (ib_opcode > ARRAY_SIZE(hns_roce_op_code) - 1)
> > return HNS_ROCE_V2_WQE_OP_MASK;
> >
> > return hns_roce_op_code[ib_opcode] ?: HNS_ROCE_V2_WQE_OP_MASK;
> >
> > Thanks
> >
>
> One more question, should I add a Reviewed-by tag for anyone who has comments
> on my patch, or I should only do this when the reviewer asked me to do it?
>
> For example, should I add a reviewed-by tag for you in this patch? Thank you :)
Yes, please.
The words "Reviewed .../ Acked ..." are the actual request to add.
Thanks
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [PATCH for-next 3/5] RDMA/hns: Optimize the wr opcode conversion from ib to hns
2020-03-05 6:18 ` Leon Romanovsky
2020-03-05 11:22 ` liweihang
@ 2020-03-05 13:24 ` Jason Gunthorpe
2020-03-05 14:33 ` Leon Romanovsky
1 sibling, 1 reply; 18+ messages in thread
From: Jason Gunthorpe @ 2020-03-05 13:24 UTC (permalink / raw)
To: Leon Romanovsky; +Cc: Weihang Li, dledford, linux-rdma, linuxarm
On Thu, Mar 05, 2020 at 08:18:39AM +0200, Leon Romanovsky wrote:
> On Mon, Mar 02, 2020 at 08:11:31PM +0800, Weihang Li wrote:
> > From: Xi Wang <wangxi11@huawei.com>
> >
> > Simplify the wr opcode conversion from ib to hns by using a map table
> > instead of the switch-case statement.
> >
> > Signed-off-by: Xi Wang <wangxi11@huawei.com>
> > Signed-off-by: Weihang Li <liweihang@huawei.com>
> > drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 70 ++++++++++++++++++------------
> > 1 file changed, 43 insertions(+), 27 deletions(-)
> >
> > diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
> > index c8c345f..ea61ccb 100644
> > +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
> > @@ -56,6 +56,47 @@ static void set_data_seg_v2(struct hns_roce_v2_wqe_data_seg *dseg,
> > dseg->len = cpu_to_le32(sg->length);
> > }
> >
> > +/*
> > + * mapped-value = 1 + real-value
> > + * The hns wr opcode real value is start from 0, In order to distinguish between
> > + * initialized and uninitialized map values, we plus 1 to the actual value when
> > + * defining the mapping, so that the validity can be identified by checking the
> > + * mapped value is greater than 0.
> > + */
> > +#define HR_OPC_MAP(ib_key, hr_key) \
> > + [IB_WR_ ## ib_key] = 1 + HNS_ROCE_V2_WQE_OP_ ## hr_key
> > +
> > +static const u32 hns_roce_op_code[] = {
> > + HR_OPC_MAP(RDMA_WRITE, RDMA_WRITE),
> > + HR_OPC_MAP(RDMA_WRITE_WITH_IMM, RDMA_WRITE_WITH_IMM),
> > + HR_OPC_MAP(SEND, SEND),
> > + HR_OPC_MAP(SEND_WITH_IMM, SEND_WITH_IMM),
> > + HR_OPC_MAP(RDMA_READ, RDMA_READ),
> > + HR_OPC_MAP(ATOMIC_CMP_AND_SWP, ATOM_CMP_AND_SWAP),
> > + HR_OPC_MAP(ATOMIC_FETCH_AND_ADD, ATOM_FETCH_AND_ADD),
> > + HR_OPC_MAP(SEND_WITH_INV, SEND_WITH_INV),
> > + HR_OPC_MAP(LOCAL_INV, LOCAL_INV),
> > + HR_OPC_MAP(MASKED_ATOMIC_CMP_AND_SWP, ATOM_MSK_CMP_AND_SWAP),
> > + HR_OPC_MAP(MASKED_ATOMIC_FETCH_AND_ADD, ATOM_MSK_FETCH_AND_ADD),
> > + HR_OPC_MAP(REG_MR, FAST_REG_PMR),
> > + [IB_WR_RESERVED1] = 0,
>
> hns_roce_op_code[] is declared as static, everything is initialized to
> 0, there is no need to set 0 again.
It is needed to guarentee the size of the array.
> > +};
> > +
> > +static inline u32 to_hr_opcode(u32 ib_opcode)
>
> No inline functions in *.c, please.
Did you see CH say this is not such a strong rule apparently?
Jason
^ permalink raw reply [flat|nested] 18+ messages in thread* Re: [PATCH for-next 3/5] RDMA/hns: Optimize the wr opcode conversion from ib to hns
2020-03-05 13:24 ` Jason Gunthorpe
@ 2020-03-05 14:33 ` Leon Romanovsky
0 siblings, 0 replies; 18+ messages in thread
From: Leon Romanovsky @ 2020-03-05 14:33 UTC (permalink / raw)
To: Jason Gunthorpe; +Cc: Weihang Li, dledford, linux-rdma, linuxarm
On Thu, Mar 05, 2020 at 09:24:53AM -0400, Jason Gunthorpe wrote:
> On Thu, Mar 05, 2020 at 08:18:39AM +0200, Leon Romanovsky wrote:
> > On Mon, Mar 02, 2020 at 08:11:31PM +0800, Weihang Li wrote:
> > > From: Xi Wang <wangxi11@huawei.com>
> > >
> > > Simplify the wr opcode conversion from ib to hns by using a map table
> > > instead of the switch-case statement.
> > >
> > > Signed-off-by: Xi Wang <wangxi11@huawei.com>
> > > Signed-off-by: Weihang Li <liweihang@huawei.com>
> > > drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 70 ++++++++++++++++++------------
> > > 1 file changed, 43 insertions(+), 27 deletions(-)
> > >
> > > diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
> > > index c8c345f..ea61ccb 100644
> > > +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
> > > @@ -56,6 +56,47 @@ static void set_data_seg_v2(struct hns_roce_v2_wqe_data_seg *dseg,
> > > dseg->len = cpu_to_le32(sg->length);
> > > }
> > >
> > > +/*
> > > + * mapped-value = 1 + real-value
> > > + * The hns wr opcode real value is start from 0, In order to distinguish between
> > > + * initialized and uninitialized map values, we plus 1 to the actual value when
> > > + * defining the mapping, so that the validity can be identified by checking the
> > > + * mapped value is greater than 0.
> > > + */
> > > +#define HR_OPC_MAP(ib_key, hr_key) \
> > > + [IB_WR_ ## ib_key] = 1 + HNS_ROCE_V2_WQE_OP_ ## hr_key
> > > +
> > > +static const u32 hns_roce_op_code[] = {
> > > + HR_OPC_MAP(RDMA_WRITE, RDMA_WRITE),
> > > + HR_OPC_MAP(RDMA_WRITE_WITH_IMM, RDMA_WRITE_WITH_IMM),
> > > + HR_OPC_MAP(SEND, SEND),
> > > + HR_OPC_MAP(SEND_WITH_IMM, SEND_WITH_IMM),
> > > + HR_OPC_MAP(RDMA_READ, RDMA_READ),
> > > + HR_OPC_MAP(ATOMIC_CMP_AND_SWP, ATOM_CMP_AND_SWAP),
> > > + HR_OPC_MAP(ATOMIC_FETCH_AND_ADD, ATOM_FETCH_AND_ADD),
> > > + HR_OPC_MAP(SEND_WITH_INV, SEND_WITH_INV),
> > > + HR_OPC_MAP(LOCAL_INV, LOCAL_INV),
> > > + HR_OPC_MAP(MASKED_ATOMIC_CMP_AND_SWP, ATOM_MSK_CMP_AND_SWAP),
> > > + HR_OPC_MAP(MASKED_ATOMIC_FETCH_AND_ADD, ATOM_MSK_FETCH_AND_ADD),
> > > + HR_OPC_MAP(REG_MR, FAST_REG_PMR),
> > > + [IB_WR_RESERVED1] = 0,
> >
> > hns_roce_op_code[] is declared as static, everything is initialized to
> > 0, there is no need to set 0 again.
>
> It is needed to guarentee the size of the array.
Not really, it depends on their implementation of to_hr_opcode().
>
> > > +};
> > > +
> > > +static inline u32 to_hr_opcode(u32 ib_opcode)
> >
> > No inline functions in *.c, please.
>
> Did you see CH say this is not such a strong rule apparently?
Yes, he talked about container_of which makes sense.
>
> Jason
^ permalink raw reply [flat|nested] 18+ messages in thread
* [PATCH for-next 4/5] RDMA/hns: Optimize base address table config flow for qp buffer
2020-03-02 12:11 [PATCH for-next 0/5] RDMA/hns: Refactor wqe related codes Weihang Li
` (2 preceding siblings ...)
2020-03-02 12:11 ` [PATCH for-next 3/5] RDMA/hns: Optimize the wr opcode conversion from ib to hns Weihang Li
@ 2020-03-02 12:11 ` Weihang Li
2020-03-04 19:18 ` Jason Gunthorpe
2020-03-02 12:11 ` [PATCH for-next 5/5] RDMA/hns: Optimize wqe buffer set flow for post send Weihang Li
4 siblings, 1 reply; 18+ messages in thread
From: Weihang Li @ 2020-03-02 12:11 UTC (permalink / raw)
To: dledford, jgg; +Cc: leon, linux-rdma, linuxarm
From: Xi Wang <wangxi11@huawei.com>
Currently, before the qp is created, a page size needs to be calculated for
the base address table to store all base addresses in the mtr. As a result,
the parameter configuration of the mtr is complex. So integrate the process
of calculating the base table page size into the hem related interface to
simplify the process of using mtr.
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
---
drivers/infiniband/hw/hns/hns_roce_device.h | 4 ---
drivers/infiniband/hw/hns/hns_roce_hem.c | 16 +++++++----
drivers/infiniband/hw/hns/hns_roce_qp.c | 42 +++++++----------------------
3 files changed, 21 insertions(+), 41 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
index b6ae12d..f6b3cf6 100644
--- a/drivers/infiniband/hw/hns/hns_roce_device.h
+++ b/drivers/infiniband/hw/hns/hns_roce_device.h
@@ -669,10 +669,6 @@ struct hns_roce_qp {
struct ib_umem *umem;
struct hns_roce_mtt mtt;
struct hns_roce_mtr mtr;
-
- /* this define must less than HNS_ROCE_MAX_BT_REGION */
-#define HNS_ROCE_WQE_REGION_MAX 3
- struct hns_roce_buf_region regions[HNS_ROCE_WQE_REGION_MAX];
int wqe_bt_pg_shift;
u32 buff_size;
diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c
index e822157..8380d71 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hem.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hem.c
@@ -1383,6 +1383,7 @@ static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev,
void *cpu_base;
u64 phy_base;
int ret = 0;
+ int ba_num;
int offset;
int total;
int step;
@@ -1393,12 +1394,16 @@ static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev,
if (root_hem)
return 0;
+ ba_num = hns_roce_hem_list_calc_root_ba(regions, region_cnt, unit);
+ if (ba_num < 1)
+ return -ENOMEM;
+
INIT_LIST_HEAD(&temp_root);
- total = r->offset;
+ offset = r->offset;
/* indicate to last region */
r = ®ions[region_cnt - 1];
- root_hem = hem_list_alloc_item(hr_dev, total, r->offset + r->count - 1,
- unit, true, 0);
+ root_hem = hem_list_alloc_item(hr_dev, offset, r->offset + r->count - 1,
+ ba_num, true, 0);
if (!root_hem)
return -ENOMEM;
list_add(&root_hem->list, &temp_root);
@@ -1410,7 +1415,7 @@ static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev,
INIT_LIST_HEAD(&temp_list[i]);
total = 0;
- for (i = 0; i < region_cnt && total < unit; i++) {
+ for (i = 0; i < region_cnt && total < ba_num; i++) {
r = ®ions[i];
if (!r->count)
continue;
@@ -1443,7 +1448,8 @@ static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev,
/* if exist mid bt, link L1 to L0 */
list_for_each_entry_safe(hem, temp_hem,
&hem_list->mid_bt[i][1], list) {
- offset = hem->start / step * BA_BYTE_LEN;
+ offset = (hem->start - r->offset) / step *
+ BA_BYTE_LEN;
hem_list_link_bt(hr_dev, cpu_base + offset,
hem->dma_addr);
total++;
diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c
index cdc8b19..37e5760 100644
--- a/drivers/infiniband/hw/hns/hns_roce_qp.c
+++ b/drivers/infiniband/hw/hns/hns_roce_qp.c
@@ -579,30 +579,6 @@ static int split_wqe_buf_region(struct hns_roce_dev *hr_dev,
return region_cnt;
}
-static int calc_wqe_bt_page_shift(struct hns_roce_dev *hr_dev,
- struct hns_roce_buf_region *regions,
- int region_cnt)
-{
- int bt_pg_shift;
- int ba_num;
- int ret;
-
- bt_pg_shift = PAGE_SHIFT + hr_dev->caps.mtt_ba_pg_sz;
-
- /* all root ba entries must in one bt page */
- do {
- ba_num = (1 << bt_pg_shift) / BA_BYTE_LEN;
- ret = hns_roce_hem_list_calc_root_ba(regions, region_cnt,
- ba_num);
- if (ret <= ba_num)
- break;
-
- bt_pg_shift++;
- } while (ret > ba_num);
-
- return bt_pg_shift - PAGE_SHIFT;
-}
-
static int set_extend_sge_param(struct hns_roce_dev *hr_dev,
struct hns_roce_qp *hr_qp)
{
@@ -768,7 +744,10 @@ static void free_rq_inline_buf(struct hns_roce_qp *hr_qp)
static int map_wqe_buf(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp,
u32 page_shift, bool is_user)
{
- dma_addr_t *buf_list[ARRAY_SIZE(hr_qp->regions)] = { NULL };
+/* WQE buffer include 3 parts: SQ, extend SGE and RQ. */
+#define HNS_ROCE_WQE_REGION_MAX 3
+ struct hns_roce_buf_region regions[HNS_ROCE_WQE_REGION_MAX] = {};
+ dma_addr_t *buf_list[HNS_ROCE_WQE_REGION_MAX] = { NULL };
struct ib_device *ibdev = &hr_dev->ib_dev;
struct hns_roce_buf_region *r;
int region_count;
@@ -776,18 +755,18 @@ static int map_wqe_buf(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp,
int ret;
int i;
- region_count = split_wqe_buf_region(hr_dev, hr_qp, hr_qp->regions,
- ARRAY_SIZE(hr_qp->regions), page_shift);
+ region_count = split_wqe_buf_region(hr_dev, hr_qp, regions,
+ ARRAY_SIZE(regions), page_shift);
/* alloc a tmp list to store WQE buffers address */
- ret = hns_roce_alloc_buf_list(hr_qp->regions, buf_list, region_count);
+ ret = hns_roce_alloc_buf_list(regions, buf_list, region_count);
if (ret) {
ibdev_err(ibdev, "Failed to alloc WQE buffer list\n");
return ret;
}
for (i = 0; i < region_count; i++) {
- r = &hr_qp->regions[i];
+ r = ®ions[i];
if (is_user)
buf_count = hns_roce_get_umem_bufs(hr_dev, buf_list[i],
r->count, r->offset, hr_qp->umem,
@@ -805,11 +784,10 @@ static int map_wqe_buf(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp,
}
}
- hr_qp->wqe_bt_pg_shift = calc_wqe_bt_page_shift(hr_dev, hr_qp->regions,
- region_count);
+ hr_qp->wqe_bt_pg_shift = hr_dev->caps.mtt_ba_pg_sz;
hns_roce_mtr_init(&hr_qp->mtr, PAGE_SHIFT + hr_qp->wqe_bt_pg_shift,
page_shift);
- ret = hns_roce_mtr_attach(hr_dev, &hr_qp->mtr, buf_list, hr_qp->regions,
+ ret = hns_roce_mtr_attach(hr_dev, &hr_qp->mtr, buf_list, regions,
region_count);
if (ret)
ibdev_err(ibdev, "Failed to attatch WQE's mtr\n");
--
2.8.1
^ permalink raw reply related [flat|nested] 18+ messages in thread* Re: [PATCH for-next 4/5] RDMA/hns: Optimize base address table config flow for qp buffer
2020-03-02 12:11 ` [PATCH for-next 4/5] RDMA/hns: Optimize base address table config flow for qp buffer Weihang Li
@ 2020-03-04 19:18 ` Jason Gunthorpe
2020-03-05 6:08 ` liweihang
0 siblings, 1 reply; 18+ messages in thread
From: Jason Gunthorpe @ 2020-03-04 19:18 UTC (permalink / raw)
To: Weihang Li; +Cc: dledford, leon, linux-rdma, linuxarm
On Mon, Mar 02, 2020 at 08:11:32PM +0800, Weihang Li wrote:
> From: Xi Wang <wangxi11@huawei.com>
>
> Currently, before the qp is created, a page size needs to be calculated for
> the base address table to store all base addresses in the mtr. As a result,
> the parameter configuration of the mtr is complex. So integrate the process
> of calculating the base table page size into the hem related interface to
> simplify the process of using mtr.
This patch doesn't apply, and somehow it didn't get linked to the
series in patchworks too. Looks like this patch was replaced somehow
as it references the wrong In-Reply-To
Please be more careful that things go into patchworks properly and
resend this series so it applies
Jason
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: [PATCH for-next 4/5] RDMA/hns: Optimize base address table config flow for qp buffer
2020-03-04 19:18 ` Jason Gunthorpe
@ 2020-03-05 6:08 ` liweihang
0 siblings, 0 replies; 18+ messages in thread
From: liweihang @ 2020-03-05 6:08 UTC (permalink / raw)
To: Jason Gunthorpe
Cc: dledford@redhat.com, leon@kernel.org, linux-rdma@vger.kernel.org,
Linuxarm
On 2020/3/5 3:18, Jason Gunthorpe wrote:
> On Mon, Mar 02, 2020 at 08:11:32PM +0800, Weihang Li wrote:
>> From: Xi Wang <wangxi11@huawei.com>
>>
>> Currently, before the qp is created, a page size needs to be calculated for
>> the base address table to store all base addresses in the mtr. As a result,
>> the parameter configuration of the mtr is complex. So integrate the process
>> of calculating the base table page size into the hem related interface to
>> simplify the process of using mtr.
>
> This patch doesn't apply, and somehow it didn't get linked to the
> series in patchworks too. Looks like this patch was replaced somehow
> as it references the wrong In-Reply-To
>
> Please be more careful that things go into patchworks properly and
> resend this series so it applies
>
> Jason
>
Sorry for that, I will resend a new version.
But I couldn't find the reason for this, maybe there were some network issues
with the machine I sent this series from.
Thanks,
Weihang
^ permalink raw reply [flat|nested] 18+ messages in thread
* [PATCH for-next 5/5] RDMA/hns: Optimize wqe buffer set flow for post send
2020-03-02 12:11 [PATCH for-next 0/5] RDMA/hns: Refactor wqe related codes Weihang Li
` (3 preceding siblings ...)
2020-03-02 12:11 ` [PATCH for-next 4/5] RDMA/hns: Optimize base address table config flow for qp buffer Weihang Li
@ 2020-03-02 12:11 ` Weihang Li
4 siblings, 0 replies; 18+ messages in thread
From: Weihang Li @ 2020-03-02 12:11 UTC (permalink / raw)
To: dledford, jgg; +Cc: leon, linux-rdma, linuxarm
From: Xi Wang <wangxi11@huawei.com>
Splits hns_roce_v2_post_send() into three sub-functions: set_rc_wqe(),
set_ud_wqe() and update_sq_db() to simplify the code.
Signed-off-by: Xi Wang <wangxi11@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
---
drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 473 ++++++++++++++---------------
1 file changed, 225 insertions(+), 248 deletions(-)
diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index ea61ccb..f6bd0cd 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -294,6 +294,214 @@ static int check_send_valid(struct hns_roce_dev *hr_dev,
return 0;
}
+static inline int calc_wr_sge_num(const struct ib_send_wr *wr, u32 *sge_len)
+{
+ u32 len = 0;
+ int num = 0;
+ int i;
+
+ for (i = 0; i < wr->num_sge; i++) {
+ if (likely(wr->sg_list[i].length)) {
+ len += wr->sg_list[i].length;
+ num++;
+ }
+ }
+
+ *sge_len = len;
+ return num;
+}
+
+static inline int set_ud_wqe(struct hns_roce_qp *qp,
+ const struct ib_send_wr *wr,
+ void *wqe, unsigned int *sge_idx,
+ unsigned int owner_bit)
+{
+ struct hns_roce_dev *hr_dev = to_hr_dev(qp->ibqp.device);
+ struct hns_roce_ah *ah = to_hr_ah(ud_wr(wr)->ah);
+ struct hns_roce_v2_ud_send_wqe *ud_sq_wqe = wqe;
+ unsigned int curr_idx = *sge_idx;
+ int valid_num_sge;
+ u32 msg_len = 0;
+ bool loopback;
+ u8 *smac;
+
+ valid_num_sge = calc_wr_sge_num(wr, &msg_len);
+ memset(ud_sq_wqe, 0, sizeof(*ud_sq_wqe));
+
+ roce_set_field(ud_sq_wqe->dmac, V2_UD_SEND_WQE_DMAC_0_M,
+ V2_UD_SEND_WQE_DMAC_0_S, ah->av.mac[0]);
+ roce_set_field(ud_sq_wqe->dmac, V2_UD_SEND_WQE_DMAC_1_M,
+ V2_UD_SEND_WQE_DMAC_1_S, ah->av.mac[1]);
+ roce_set_field(ud_sq_wqe->dmac, V2_UD_SEND_WQE_DMAC_2_M,
+ V2_UD_SEND_WQE_DMAC_2_S, ah->av.mac[2]);
+ roce_set_field(ud_sq_wqe->dmac, V2_UD_SEND_WQE_DMAC_3_M,
+ V2_UD_SEND_WQE_DMAC_3_S, ah->av.mac[3]);
+ roce_set_field(ud_sq_wqe->byte_48, V2_UD_SEND_WQE_BYTE_48_DMAC_4_M,
+ V2_UD_SEND_WQE_BYTE_48_DMAC_4_S, ah->av.mac[4]);
+ roce_set_field(ud_sq_wqe->byte_48, V2_UD_SEND_WQE_BYTE_48_DMAC_5_M,
+ V2_UD_SEND_WQE_BYTE_48_DMAC_5_S, ah->av.mac[5]);
+
+ /* MAC loopback */
+ smac = (u8 *)hr_dev->dev_addr[qp->port];
+ loopback = ether_addr_equal_unaligned(ah->av.mac, smac) ? 1 : 0;
+
+ roce_set_bit(ud_sq_wqe->byte_40,
+ V2_UD_SEND_WQE_BYTE_40_LBI_S, loopback);
+
+ roce_set_field(ud_sq_wqe->byte_4,
+ V2_UD_SEND_WQE_BYTE_4_OPCODE_M,
+ V2_UD_SEND_WQE_BYTE_4_OPCODE_S,
+ HNS_ROCE_V2_WQE_OP_SEND);
+
+ ud_sq_wqe->msg_len = cpu_to_le32(msg_len);
+
+ switch (wr->opcode) {
+ case IB_WR_SEND_WITH_IMM:
+ case IB_WR_RDMA_WRITE_WITH_IMM:
+ ud_sq_wqe->immtdata = cpu_to_le32(be32_to_cpu(wr->ex.imm_data));
+ break;
+ default:
+ ud_sq_wqe->immtdata = 0;
+ break;
+ }
+
+ /* Set sig attr */
+ roce_set_bit(ud_sq_wqe->byte_4, V2_UD_SEND_WQE_BYTE_4_CQE_S,
+ (wr->send_flags & IB_SEND_SIGNALED) ? 1 : 0);
+
+ /* Set se attr */
+ roce_set_bit(ud_sq_wqe->byte_4, V2_UD_SEND_WQE_BYTE_4_SE_S,
+ (wr->send_flags & IB_SEND_SOLICITED) ? 1 : 0);
+
+ roce_set_bit(ud_sq_wqe->byte_4, V2_UD_SEND_WQE_BYTE_4_OWNER_S,
+ owner_bit);
+
+ roce_set_field(ud_sq_wqe->byte_16, V2_UD_SEND_WQE_BYTE_16_PD_M,
+ V2_UD_SEND_WQE_BYTE_16_PD_S, to_hr_pd(qp->ibqp.pd)->pdn);
+
+ roce_set_field(ud_sq_wqe->byte_16, V2_UD_SEND_WQE_BYTE_16_SGE_NUM_M,
+ V2_UD_SEND_WQE_BYTE_16_SGE_NUM_S, valid_num_sge);
+
+ roce_set_field(ud_sq_wqe->byte_20,
+ V2_UD_SEND_WQE_BYTE_20_MSG_START_SGE_IDX_M,
+ V2_UD_SEND_WQE_BYTE_20_MSG_START_SGE_IDX_S,
+ curr_idx & (qp->sge.sge_cnt - 1));
+
+ roce_set_field(ud_sq_wqe->byte_24, V2_UD_SEND_WQE_BYTE_24_UDPSPN_M,
+ V2_UD_SEND_WQE_BYTE_24_UDPSPN_S, 0);
+ ud_sq_wqe->qkey = cpu_to_le32(ud_wr(wr)->remote_qkey & 0x80000000 ?
+ qp->qkey : ud_wr(wr)->remote_qkey);
+ roce_set_field(ud_sq_wqe->byte_32, V2_UD_SEND_WQE_BYTE_32_DQPN_M,
+ V2_UD_SEND_WQE_BYTE_32_DQPN_S, ud_wr(wr)->remote_qpn);
+
+ roce_set_field(ud_sq_wqe->byte_36, V2_UD_SEND_WQE_BYTE_36_VLAN_M,
+ V2_UD_SEND_WQE_BYTE_36_VLAN_S, ah->av.vlan_id);
+ roce_set_field(ud_sq_wqe->byte_36, V2_UD_SEND_WQE_BYTE_36_HOPLIMIT_M,
+ V2_UD_SEND_WQE_BYTE_36_HOPLIMIT_S, ah->av.hop_limit);
+ roce_set_field(ud_sq_wqe->byte_36, V2_UD_SEND_WQE_BYTE_36_TCLASS_M,
+ V2_UD_SEND_WQE_BYTE_36_TCLASS_S, ah->av.tclass);
+ roce_set_field(ud_sq_wqe->byte_40, V2_UD_SEND_WQE_BYTE_40_FLOW_LABEL_M,
+ V2_UD_SEND_WQE_BYTE_40_FLOW_LABEL_S, ah->av.flowlabel);
+ roce_set_field(ud_sq_wqe->byte_40, V2_UD_SEND_WQE_BYTE_40_SL_M,
+ V2_UD_SEND_WQE_BYTE_40_SL_S, ah->av.sl);
+ roce_set_field(ud_sq_wqe->byte_40, V2_UD_SEND_WQE_BYTE_40_PORTN_M,
+ V2_UD_SEND_WQE_BYTE_40_PORTN_S, qp->port);
+
+ roce_set_bit(ud_sq_wqe->byte_40, V2_UD_SEND_WQE_BYTE_40_UD_VLAN_EN_S,
+ ah->av.vlan_en ? 1 : 0);
+ roce_set_field(ud_sq_wqe->byte_48, V2_UD_SEND_WQE_BYTE_48_SGID_INDX_M,
+ V2_UD_SEND_WQE_BYTE_48_SGID_INDX_S, ah->av.gid_index);
+
+ memcpy(&ud_sq_wqe->dgid[0], &ah->av.dgid[0], GID_LEN_V2);
+
+ set_extend_sge(qp, wr, &curr_idx, valid_num_sge);
+
+ *sge_idx = curr_idx;
+
+ return 0;
+}
+
+static inline int set_rc_wqe(struct hns_roce_qp *qp,
+ const struct ib_send_wr *wr,
+ void *wqe, unsigned int *sge_idx,
+ unsigned int owner_bit)
+{
+ struct hns_roce_v2_rc_send_wqe *rc_sq_wqe = wqe;
+ unsigned int curr_idx = *sge_idx;
+ int valid_num_sge;
+ u32 msg_len = 0;
+ int ret = 0;
+
+ valid_num_sge = calc_wr_sge_num(wr, &msg_len);
+ memset(rc_sq_wqe, 0, sizeof(*rc_sq_wqe));
+
+ rc_sq_wqe->msg_len = cpu_to_le32(msg_len);
+
+ switch (wr->opcode) {
+ case IB_WR_SEND_WITH_IMM:
+ case IB_WR_RDMA_WRITE_WITH_IMM:
+ rc_sq_wqe->immtdata = cpu_to_le32(be32_to_cpu(wr->ex.imm_data));
+ break;
+ case IB_WR_SEND_WITH_INV:
+ rc_sq_wqe->inv_key = cpu_to_le32(wr->ex.invalidate_rkey);
+ break;
+ default:
+ rc_sq_wqe->immtdata = 0;
+ break;
+ }
+
+ roce_set_bit(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_FENCE_S,
+ (wr->send_flags & IB_SEND_FENCE) ? 1 : 0);
+
+ roce_set_bit(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_SE_S,
+ (wr->send_flags & IB_SEND_SOLICITED) ? 1 : 0);
+
+ roce_set_bit(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_CQE_S,
+ (wr->send_flags & IB_SEND_SIGNALED) ? 1 : 0);
+
+ roce_set_bit(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_OWNER_S,
+ owner_bit);
+
+ wqe += sizeof(struct hns_roce_v2_rc_send_wqe);
+ switch (wr->opcode) {
+ case IB_WR_RDMA_READ:
+ case IB_WR_RDMA_WRITE:
+ case IB_WR_RDMA_WRITE_WITH_IMM:
+ rc_sq_wqe->rkey = cpu_to_le32(rdma_wr(wr)->rkey);
+ rc_sq_wqe->va = cpu_to_le64(rdma_wr(wr)->remote_addr);
+ break;
+ case IB_WR_LOCAL_INV:
+ roce_set_bit(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_SO_S, 1);
+ rc_sq_wqe->inv_key = cpu_to_le32(wr->ex.invalidate_rkey);
+ break;
+ case IB_WR_REG_MR:
+ set_frmr_seg(rc_sq_wqe, wqe, reg_wr(wr));
+ break;
+ case IB_WR_ATOMIC_CMP_AND_SWP:
+ case IB_WR_ATOMIC_FETCH_AND_ADD:
+ rc_sq_wqe->rkey = cpu_to_le32(atomic_wr(wr)->rkey);
+ rc_sq_wqe->va = cpu_to_le64(atomic_wr(wr)->remote_addr);
+ break;
+ default:
+ break;
+ }
+
+ roce_set_field(rc_sq_wqe->byte_4, V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
+ V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
+ to_hr_opcode(wr->opcode));
+
+ if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
+ wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
+ set_atomic_seg(wr, wqe, rc_sq_wqe, valid_num_sge);
+ else if (wr->opcode != IB_WR_REG_MR)
+ ret = set_rwqe_data_seg(&qp->ibqp, wr, rc_sq_wqe,
+ wqe, &curr_idx, valid_num_sge);
+
+ *sge_idx = curr_idx;
+
+ return ret;
+}
+
static inline void update_sq_db(struct hns_roce_dev *hr_dev,
struct hns_roce_qp *qp)
{
@@ -331,23 +539,15 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
const struct ib_send_wr **bad_wr)
{
struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
- struct hns_roce_ah *ah = to_hr_ah(ud_wr(wr)->ah);
- struct hns_roce_v2_ud_send_wqe *ud_sq_wqe;
- struct hns_roce_v2_rc_send_wqe *rc_sq_wqe;
+ struct ib_device *ibdev = &hr_dev->ib_dev;
struct hns_roce_qp *qp = to_hr_qp(ibqp);
- struct device *dev = hr_dev->dev;
+ unsigned long flags = 0;
unsigned int owner_bit;
unsigned int sge_idx;
unsigned int wqe_idx;
- unsigned long flags;
- int valid_num_sge;
void *wqe = NULL;
- bool loopback;
- u32 tmp_len;
- u8 *smac;
int nreq;
int ret;
- int i;
spin_lock_irqsave(&qp->sq.lock, flags);
@@ -370,8 +570,8 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
wqe_idx = (qp->sq.head + nreq) & (qp->sq.wqe_cnt - 1);
if (unlikely(wr->num_sge > qp->sq.max_gs)) {
- dev_err(dev, "num_sge=%d > qp->sq.max_gs=%d\n",
- wr->num_sge, qp->sq.max_gs);
+ ibdev_err(ibdev, "num_sge=%d > qp->sq.max_gs=%d\n",
+ wr->num_sge, qp->sq.max_gs);
ret = -EINVAL;
*bad_wr = wr;
goto out;
@@ -381,248 +581,25 @@ static int hns_roce_v2_post_send(struct ib_qp *ibqp,
qp->sq.wrid[wqe_idx] = wr->wr_id;
owner_bit =
~(((qp->sq.head + nreq) >> ilog2(qp->sq.wqe_cnt)) & 0x1);
- valid_num_sge = 0;
- tmp_len = 0;
-
- for (i = 0; i < wr->num_sge; i++) {
- if (likely(wr->sg_list[i].length)) {
- tmp_len += wr->sg_list[i].length;
- valid_num_sge++;
- }
- }
/* Corresponding to the QP type, wqe process separately */
- if (ibqp->qp_type == IB_QPT_GSI) {
- ud_sq_wqe = wqe;
- memset(ud_sq_wqe, 0, sizeof(*ud_sq_wqe));
-
- roce_set_field(ud_sq_wqe->dmac, V2_UD_SEND_WQE_DMAC_0_M,
- V2_UD_SEND_WQE_DMAC_0_S, ah->av.mac[0]);
- roce_set_field(ud_sq_wqe->dmac, V2_UD_SEND_WQE_DMAC_1_M,
- V2_UD_SEND_WQE_DMAC_1_S, ah->av.mac[1]);
- roce_set_field(ud_sq_wqe->dmac, V2_UD_SEND_WQE_DMAC_2_M,
- V2_UD_SEND_WQE_DMAC_2_S, ah->av.mac[2]);
- roce_set_field(ud_sq_wqe->dmac, V2_UD_SEND_WQE_DMAC_3_M,
- V2_UD_SEND_WQE_DMAC_3_S, ah->av.mac[3]);
- roce_set_field(ud_sq_wqe->byte_48,
- V2_UD_SEND_WQE_BYTE_48_DMAC_4_M,
- V2_UD_SEND_WQE_BYTE_48_DMAC_4_S,
- ah->av.mac[4]);
- roce_set_field(ud_sq_wqe->byte_48,
- V2_UD_SEND_WQE_BYTE_48_DMAC_5_M,
- V2_UD_SEND_WQE_BYTE_48_DMAC_5_S,
- ah->av.mac[5]);
-
- /* MAC loopback */
- smac = (u8 *)hr_dev->dev_addr[qp->port];
- loopback = ether_addr_equal_unaligned(ah->av.mac,
- smac) ? 1 : 0;
-
- roce_set_bit(ud_sq_wqe->byte_40,
- V2_UD_SEND_WQE_BYTE_40_LBI_S, loopback);
-
- roce_set_field(ud_sq_wqe->byte_4,
- V2_UD_SEND_WQE_BYTE_4_OPCODE_M,
- V2_UD_SEND_WQE_BYTE_4_OPCODE_S,
- HNS_ROCE_V2_WQE_OP_SEND);
-
- ud_sq_wqe->msg_len =
- cpu_to_le32(le32_to_cpu(ud_sq_wqe->msg_len) + tmp_len);
-
- switch (wr->opcode) {
- case IB_WR_SEND_WITH_IMM:
- case IB_WR_RDMA_WRITE_WITH_IMM:
- ud_sq_wqe->immtdata =
- cpu_to_le32(be32_to_cpu(wr->ex.imm_data));
- break;
- default:
- ud_sq_wqe->immtdata = 0;
- break;
- }
-
- /* Set sig attr */
- roce_set_bit(ud_sq_wqe->byte_4,
- V2_UD_SEND_WQE_BYTE_4_CQE_S,
- (wr->send_flags & IB_SEND_SIGNALED) ? 1 : 0);
-
- /* Set se attr */
- roce_set_bit(ud_sq_wqe->byte_4,
- V2_UD_SEND_WQE_BYTE_4_SE_S,
- (wr->send_flags & IB_SEND_SOLICITED) ? 1 : 0);
-
- roce_set_bit(ud_sq_wqe->byte_4,
- V2_UD_SEND_WQE_BYTE_4_OWNER_S, owner_bit);
-
- roce_set_field(ud_sq_wqe->byte_16,
- V2_UD_SEND_WQE_BYTE_16_PD_M,
- V2_UD_SEND_WQE_BYTE_16_PD_S,
- to_hr_pd(ibqp->pd)->pdn);
-
- roce_set_field(ud_sq_wqe->byte_16,
- V2_UD_SEND_WQE_BYTE_16_SGE_NUM_M,
- V2_UD_SEND_WQE_BYTE_16_SGE_NUM_S,
- valid_num_sge);
-
- roce_set_field(ud_sq_wqe->byte_20,
- V2_UD_SEND_WQE_BYTE_20_MSG_START_SGE_IDX_M,
- V2_UD_SEND_WQE_BYTE_20_MSG_START_SGE_IDX_S,
- sge_idx & (qp->sge.sge_cnt - 1));
-
- roce_set_field(ud_sq_wqe->byte_24,
- V2_UD_SEND_WQE_BYTE_24_UDPSPN_M,
- V2_UD_SEND_WQE_BYTE_24_UDPSPN_S, 0);
- ud_sq_wqe->qkey =
- cpu_to_le32(ud_wr(wr)->remote_qkey & 0x80000000 ?
- qp->qkey : ud_wr(wr)->remote_qkey);
- roce_set_field(ud_sq_wqe->byte_32,
- V2_UD_SEND_WQE_BYTE_32_DQPN_M,
- V2_UD_SEND_WQE_BYTE_32_DQPN_S,
- ud_wr(wr)->remote_qpn);
-
- roce_set_field(ud_sq_wqe->byte_36,
- V2_UD_SEND_WQE_BYTE_36_VLAN_M,
- V2_UD_SEND_WQE_BYTE_36_VLAN_S,
- ah->av.vlan_id);
- roce_set_field(ud_sq_wqe->byte_36,
- V2_UD_SEND_WQE_BYTE_36_HOPLIMIT_M,
- V2_UD_SEND_WQE_BYTE_36_HOPLIMIT_S,
- ah->av.hop_limit);
- roce_set_field(ud_sq_wqe->byte_36,
- V2_UD_SEND_WQE_BYTE_36_TCLASS_M,
- V2_UD_SEND_WQE_BYTE_36_TCLASS_S,
- ah->av.tclass);
- roce_set_field(ud_sq_wqe->byte_40,
- V2_UD_SEND_WQE_BYTE_40_FLOW_LABEL_M,
- V2_UD_SEND_WQE_BYTE_40_FLOW_LABEL_S,
- ah->av.flowlabel);
- roce_set_field(ud_sq_wqe->byte_40,
- V2_UD_SEND_WQE_BYTE_40_SL_M,
- V2_UD_SEND_WQE_BYTE_40_SL_S,
- ah->av.sl);
- roce_set_field(ud_sq_wqe->byte_40,
- V2_UD_SEND_WQE_BYTE_40_PORTN_M,
- V2_UD_SEND_WQE_BYTE_40_PORTN_S,
- qp->port);
-
- roce_set_bit(ud_sq_wqe->byte_40,
- V2_UD_SEND_WQE_BYTE_40_UD_VLAN_EN_S,
- ah->av.vlan_en ? 1 : 0);
- roce_set_field(ud_sq_wqe->byte_48,
- V2_UD_SEND_WQE_BYTE_48_SGID_INDX_M,
- V2_UD_SEND_WQE_BYTE_48_SGID_INDX_S,
- hns_get_gid_index(hr_dev, qp->phy_port,
- ah->av.gid_index));
-
- memcpy(&ud_sq_wqe->dgid[0], &ah->av.dgid[0],
- GID_LEN_V2);
-
- set_extend_sge(qp, wr, &sge_idx, valid_num_sge);
- } else if (ibqp->qp_type == IB_QPT_RC) {
- rc_sq_wqe = wqe;
- memset(rc_sq_wqe, 0, sizeof(*rc_sq_wqe));
-
- rc_sq_wqe->msg_len =
- cpu_to_le32(le32_to_cpu(rc_sq_wqe->msg_len) + tmp_len);
-
- switch (wr->opcode) {
- case IB_WR_SEND_WITH_IMM:
- case IB_WR_RDMA_WRITE_WITH_IMM:
- rc_sq_wqe->immtdata =
- cpu_to_le32(be32_to_cpu(wr->ex.imm_data));
- break;
- case IB_WR_SEND_WITH_INV:
- rc_sq_wqe->inv_key =
- cpu_to_le32(wr->ex.invalidate_rkey);
- break;
- default:
- rc_sq_wqe->immtdata = 0;
- break;
- }
-
- roce_set_bit(rc_sq_wqe->byte_4,
- V2_RC_SEND_WQE_BYTE_4_FENCE_S,
- (wr->send_flags & IB_SEND_FENCE) ? 1 : 0);
-
- roce_set_bit(rc_sq_wqe->byte_4,
- V2_RC_SEND_WQE_BYTE_4_SE_S,
- (wr->send_flags & IB_SEND_SOLICITED) ? 1 : 0);
-
- roce_set_bit(rc_sq_wqe->byte_4,
- V2_RC_SEND_WQE_BYTE_4_CQE_S,
- (wr->send_flags & IB_SEND_SIGNALED) ? 1 : 0);
-
- roce_set_bit(rc_sq_wqe->byte_4,
- V2_RC_SEND_WQE_BYTE_4_OWNER_S, owner_bit);
-
- wqe += sizeof(struct hns_roce_v2_rc_send_wqe);
- switch (wr->opcode) {
- case IB_WR_RDMA_READ:
- rc_sq_wqe->rkey =
- cpu_to_le32(rdma_wr(wr)->rkey);
- rc_sq_wqe->va =
- cpu_to_le64(rdma_wr(wr)->remote_addr);
- break;
- case IB_WR_RDMA_WRITE:
- rc_sq_wqe->rkey =
- cpu_to_le32(rdma_wr(wr)->rkey);
- rc_sq_wqe->va =
- cpu_to_le64(rdma_wr(wr)->remote_addr);
- break;
- case IB_WR_RDMA_WRITE_WITH_IMM:
- rc_sq_wqe->rkey =
- cpu_to_le32(rdma_wr(wr)->rkey);
- rc_sq_wqe->va =
- cpu_to_le64(rdma_wr(wr)->remote_addr);
- break;
- case IB_WR_LOCAL_INV:
- roce_set_bit(rc_sq_wqe->byte_4,
- V2_RC_SEND_WQE_BYTE_4_SO_S, 1);
- rc_sq_wqe->inv_key =
- cpu_to_le32(wr->ex.invalidate_rkey);
- break;
- case IB_WR_REG_MR:
- set_frmr_seg(rc_sq_wqe, wqe, reg_wr(wr));
- break;
- case IB_WR_ATOMIC_CMP_AND_SWP:
- rc_sq_wqe->rkey =
- cpu_to_le32(atomic_wr(wr)->rkey);
- rc_sq_wqe->va =
- cpu_to_le64(atomic_wr(wr)->remote_addr);
- break;
- case IB_WR_ATOMIC_FETCH_AND_ADD:
- rc_sq_wqe->rkey =
- cpu_to_le32(atomic_wr(wr)->rkey);
- rc_sq_wqe->va =
- cpu_to_le64(atomic_wr(wr)->remote_addr);
- break;
- default:
- break;
- }
-
- roce_set_field(rc_sq_wqe->byte_4,
- V2_RC_SEND_WQE_BYTE_4_OPCODE_M,
- V2_RC_SEND_WQE_BYTE_4_OPCODE_S,
- to_hr_opcode(wr->opcode));
-
- if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
- wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
- set_atomic_seg(wr, wqe, rc_sq_wqe,
- valid_num_sge);
- else if (wr->opcode != IB_WR_REG_MR) {
- ret = set_rwqe_data_seg(ibqp, wr, rc_sq_wqe,
- wqe, &sge_idx,
- valid_num_sge);
- if (ret) {
- *bad_wr = wr;
- goto out;
- }
- }
- } else {
- dev_err(dev, "Illegal qp_type(0x%x)\n", ibqp->qp_type);
+ if (ibqp->qp_type == IB_QPT_GSI)
+ ret = set_ud_wqe(qp, wr, wqe, &sge_idx, owner_bit);
+ else if (ibqp->qp_type == IB_QPT_RC)
+ ret = set_rc_wqe(qp, wr, wqe, &sge_idx, owner_bit);
+
+ else {
+ ibdev_err(ibdev, "Illegal qp_type(0x%x)\n",
+ ibqp->qp_type);
spin_unlock_irqrestore(&qp->sq.lock, flags);
*bad_wr = wr;
return -EOPNOTSUPP;
}
+
+ if (ret) {
+ *bad_wr = wr;
+ goto out;
+ }
}
out:
--
2.8.1
^ permalink raw reply related [flat|nested] 18+ messages in thread