Skip to contentSkip to conversation

RDMA/hns: Fix and refactor CMDQ related code

PATCH Superseded Weihang Li 2 participants 9 messages Feb 4, 2021 – Feb 7, 2021 1 message not in the archive

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Weihang Li

author
#0 lore raw

cover RDMA/hns: Fix and refactor CMDQ related code

Remove some dead code in process of CMDQ transmission, and fix an issue about missing error code.

Lang Cheng (6):
RDMA/hns: Remove unused member and variable of CMDQ RDMA/hns: Remove unsupported CMDQ mode
RDMA/hns: Fixes missing error code of CMDQ RDMA/hns: Remove redundant operations on CMDQ RDMA/hns: Adjust fields and variables about CMDQ tail/head RDMA/hns: Refactor process of posting CMDQ

 drivers/infiniband/hw/hns/hns_roce_common.h |   4 +-
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 142 ++++++++--------------------
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  |   7 --
 3 files changed, 42 insertions(+), 111 deletions(-)
Signature · 1 line
2.8.1
author
#1 lore raw

PATCH 1/6 RDMA/hns: Remove unused member and variable of CMDQ

From: Lang Cheng <[email protected]>

last_status of structure hns_roce_v2_cmq has never been used, and the variable named 'complete' in __hns_roce_cmq_send() is meaningless.

Fixes a04ff739f2a9 ("RDMA/hns: Add command queue support for hip08 RoCE driver")
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 9 +++------
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h | 1 -
 2 files changed, 3 insertions(+), 7 deletions(-)
drivers/infiniband/hw/hns/hns_roce_hw_v2.c +3 -6
index a5bbfb1..7a5a41d 100644
@@ -1243,7 +1243,6 @@ static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 	struct hns_roce_v2_priv *priv = hr_dev->priv;
 	struct hns_roce_v2_cmq_ring *csq = &priv->cmq.csq;
 	struct hns_roce_cmq_desc *desc_to_use;
-	bool complete = false;
 	u32 timeout = 0;
 	int handle = 0;
 	u16 desc_ret;
@@ -1290,7 +1289,6 @@ static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 	}
 
 	if (hns_roce_cmq_csq_done(hr_dev)) {
-		complete = true;
 		handle = 0;
 		while (handle < num) {
 			/* get the result of hardware write back */
@@ -1302,16 +1300,15 @@ static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 				ret = 0;
 			else
 				ret = -EIO;
-			priv->cmq.last_status = desc_ret;
+
 			ntc++;
 			handle++;
 			if (ntc == csq->desc_num)
 				ntc = 0;
 		}
-	}
-
-	if (!complete)
+	} else {
 		ret = -EAGAIN;
+	}
 
 	/* clean the command send queue */
 	handle = hns_roce_cmq_csq_clean(hr_dev);
drivers/infiniband/hw/hns/hns_roce_hw_v2.h +0 -1
index 69bc072..9f97e32 100644
@@ -1839,7 +1839,6 @@ struct hns_roce_v2_cmq {
 	struct hns_roce_v2_cmq_ring csq;
 	struct hns_roce_v2_cmq_ring crq;
 	u16 tx_timeout;
-	u16 last_status;
 };
 
 enum hns_roce_link_table_type {
Signature · 1 line
2.8.1
author
#2 lore raw

PATCH 2/6 RDMA/hns: Remove unsupported CMDQ mode

In reply to Weihang Li

From: Lang Cheng <[email protected]>

HIP08/09 only supports CMDQ in non-interrupt mode, and the firmware always ignores the flag to indicate the mode. Therefore, remove the dead code.

Fixes a04ff739f2a9 ("RDMA/hns: Add command queue support for hip08 RoCE driver")
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 24 ++++++++----------------
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h |  2 --
 2 files changed, 8 insertions(+), 18 deletions(-)
drivers/infiniband/hw/hns/hns_roce_hw_v2.c +8 -16
index 7a5a41d..260c17c 100644
@@ -1197,8 +1197,7 @@ static void hns_roce_cmq_setup_basic_desc(struct hns_roce_cmq_desc *desc,
 {
 	memset((void *)desc, 0, sizeof(struct hns_roce_cmq_desc));
 	desc->opcode = cpu_to_le16(opcode);
-	desc->flag =
-		cpu_to_le16(HNS_ROCE_CMD_FLAG_NO_INTR | HNS_ROCE_CMD_FLAG_IN);
+	desc->flag = cpu_to_le16(HNS_ROCE_CMD_FLAG_IN);
 	if (is_read)
 		desc->flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_WR);
 	else
@@ -1275,18 +1274,12 @@ static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 	/* Write to hardware */
 	roce_write(hr_dev, ROCEE_TX_CMQ_TAIL_REG, csq->next_to_use);
 
-	/*
-	 * If the command is sync, wait for the firmware to write back,
-	 * if multi descriptors to be sent, use the first one to check
-	 */
-	if (le16_to_cpu(desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
-		do {
-			if (hns_roce_cmq_csq_done(hr_dev))
-				break;
-			udelay(1);
-			timeout++;
-		} while (timeout < priv->cmq.tx_timeout);
-	}
+	do {
+		if (hns_roce_cmq_csq_done(hr_dev))
+			break;
+		udelay(1);
+		timeout++;
+	} while (timeout < priv->cmq.tx_timeout);
 
 	if (hns_roce_cmq_csq_done(hr_dev)) {
 		handle = 0;
@@ -1626,8 +1619,7 @@ static int hns_roce_set_vf_switch_param(struct hns_roce_dev *hr_dev, int vf_id)
 	if (ret)
 		return ret;
 
-	desc.flag =
-		cpu_to_le16(HNS_ROCE_CMD_FLAG_NO_INTR | HNS_ROCE_CMD_FLAG_IN);
+	desc.flag = cpu_to_le16(HNS_ROCE_CMD_FLAG_IN);
 	desc.flag &= cpu_to_le16(~HNS_ROCE_CMD_FLAG_WR);
 	roce_set_bit(swt->cfg, VF_SWITCH_DATA_CFG_ALW_LPBK_S, 1);
 	roce_set_bit(swt->cfg, VF_SWITCH_DATA_CFG_ALW_LCL_LPBK_S, 0);
drivers/infiniband/hw/hns/hns_roce_hw_v2.h +0 -2
index 9f97e32..986a287 100644
@@ -128,14 +128,12 @@
 #define HNS_ROCE_CMD_FLAG_OUT_VALID_SHIFT	1
 #define HNS_ROCE_CMD_FLAG_NEXT_SHIFT		2
 #define HNS_ROCE_CMD_FLAG_WR_OR_RD_SHIFT	3
-#define HNS_ROCE_CMD_FLAG_NO_INTR_SHIFT		4
 #define HNS_ROCE_CMD_FLAG_ERR_INTR_SHIFT	5
 
 #define HNS_ROCE_CMD_FLAG_IN		BIT(HNS_ROCE_CMD_FLAG_IN_VALID_SHIFT)
 #define HNS_ROCE_CMD_FLAG_OUT		BIT(HNS_ROCE_CMD_FLAG_OUT_VALID_SHIFT)
 #define HNS_ROCE_CMD_FLAG_NEXT		BIT(HNS_ROCE_CMD_FLAG_NEXT_SHIFT)
 #define HNS_ROCE_CMD_FLAG_WR		BIT(HNS_ROCE_CMD_FLAG_WR_OR_RD_SHIFT)
-#define HNS_ROCE_CMD_FLAG_NO_INTR	BIT(HNS_ROCE_CMD_FLAG_NO_INTR_SHIFT)
 #define HNS_ROCE_CMD_FLAG_ERR_INTR	BIT(HNS_ROCE_CMD_FLAG_ERR_INTR_SHIFT)
 
 #define HNS_ROCE_CMQ_DESC_NUM_S		3
Signature · 1 line
2.8.1

A message that is not in the archive (sent off-list or not yet received): [email protected]

author
#4 lore raw

Re: [PATCH for-next 2/6] RDMA/hns: Remove unsupported CMDQ mode

Quoted 58 lines · chenglang
Quoted 56 lines · Weihang Li

From: Lang Cheng <[email protected]>

HIP08/09 only supports CMDQ in non-interrupt mode, and the firmware always ignores the flag to indicate the mode. Therefore, remove the dead code.

Fixes a04ff739f2a9 ("RDMA/hns: Add command queue support for hip08 RoCE driver")
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 24 ++++++++----------------
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h |  2 --
 2 files changed, 8 insertions(+), 18 deletions(-)
drivers/infiniband/hw/hns/hns_roce_hw_v2.c +7 -14
index 7a5a41d..260c17c 100644
@@ -1197,8 +1197,7 @@ static void hns_roce_cmq_setup_basic_desc(struct hns_roce_cmq_desc *desc,
 {
 	memset((void *)desc, 0, sizeof(struct hns_roce_cmq_desc));
 	desc->opcode = cpu_to_le16(opcode);
-	desc->flag =
-		cpu_to_le16(HNS_ROCE_CMD_FLAG_NO_INTR | HNS_ROCE_CMD_FLAG_IN);
+	desc->flag = cpu_to_le16(HNS_ROCE_CMD_FLAG_IN);
 	if (is_read)
 		desc->flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_WR);
 	else
@@ -1275,18 +1274,12 @@ static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 	/* Write to hardware */
 	roce_write(hr_dev, ROCEE_TX_CMQ_TAIL_REG, csq->next_to_use);
 
-	/*
-	 * If the command is sync, wait for the firmware to write back,
-	 * if multi descriptors to be sent, use the first one to check
-	 */
-	if (le16_to_cpu(desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
-		do {
-			if (hns_roce_cmq_csq_done(hr_dev))
-				break;
-			udelay(1);
-			timeout++;
-		} while (timeout < priv->cmq.tx_timeout);
-	}
+	do {
+		if (hns_roce_cmq_csq_done(hr_dev))
+			break;
+		udelay(1);
+		timeout++;
+	} while (timeout < priv->cmq.tx_timeout);
 
 	if (hns_roce_cmq_csq_done(hr_dev)) {
 		handle = 0;
@@ -1626,8 +1619,7 @@ static int hns_roce_set_vf_switch_param(struct hns_roce_dev *hr_dev, int vf_id)
 	if (ret)
 		return ret;
 
Quoted 7 lines

drivers/infiniband/hw/hns/hns_roce_hw_v2.c static int hns_roce_set_vf_switch_param(struct hns_roce_dev *hr_dev, int vf_id) · line 1630

Quoted 2 lines
drivers/infiniband/hw/hns/hns_roce_hw_v2.c +0 -2
-	desc.flag =
-		cpu_to_le16(HNS_ROCE_CMD_FLAG_NO_INTR | HNS_ROCE_CMD_FLAG_IN);

The old firmware needs this redundant flag, it is best cleaned up after the firmware version is released.

Thanks.

Got it, I will drop this one from the series.

Thanks
Weihang

Quoted 24 lines
Quoted 24 lines
drivers/infiniband/hw/hns/hns_roce_hw_v2.c +1 -0
+	desc.flag = cpu_to_le16(HNS_ROCE_CMD_FLAG_IN);
 	desc.flag &= cpu_to_le16(~HNS_ROCE_CMD_FLAG_WR);
 	roce_set_bit(swt->cfg, VF_SWITCH_DATA_CFG_ALW_LPBK_S, 1);
 	roce_set_bit(swt->cfg, VF_SWITCH_DATA_CFG_ALW_LCL_LPBK_S, 0);
drivers/infiniband/hw/hns/hns_roce_hw_v2.h +0 -2
index 9f97e32..986a287 100644
@@ -128,14 +128,12 @@
 #define HNS_ROCE_CMD_FLAG_OUT_VALID_SHIFT	1
 #define HNS_ROCE_CMD_FLAG_NEXT_SHIFT		2
 #define HNS_ROCE_CMD_FLAG_WR_OR_RD_SHIFT	3
-#define HNS_ROCE_CMD_FLAG_NO_INTR_SHIFT		4
 #define HNS_ROCE_CMD_FLAG_ERR_INTR_SHIFT	5
 
 #define HNS_ROCE_CMD_FLAG_IN		BIT(HNS_ROCE_CMD_FLAG_IN_VALID_SHIFT)
 #define HNS_ROCE_CMD_FLAG_OUT		BIT(HNS_ROCE_CMD_FLAG_OUT_VALID_SHIFT)
 #define HNS_ROCE_CMD_FLAG_NEXT		BIT(HNS_ROCE_CMD_FLAG_NEXT_SHIFT)
 #define HNS_ROCE_CMD_FLAG_WR		BIT(HNS_ROCE_CMD_FLAG_WR_OR_RD_SHIFT)
-#define HNS_ROCE_CMD_FLAG_NO_INTR	BIT(HNS_ROCE_CMD_FLAG_NO_INTR_SHIFT)
 #define HNS_ROCE_CMD_FLAG_ERR_INTR	BIT(HNS_ROCE_CMD_FLAG_ERR_INTR_SHIFT)
 
 #define HNS_ROCE_CMQ_DESC_NUM_S		3
#5 lore raw

Re: [PATCH for-next 2/6] RDMA/hns: Remove unsupported CMDQ mode

Quoted 65 lines · liweihang
Quoted 64 lines · chenglang
Quoted 53 lines · Weihang Li

From: Lang Cheng <[email protected]>

HIP08/09 only supports CMDQ in non-interrupt mode, and the firmware always ignores the flag to indicate the mode. Therefore, remove the dead code.

Fixes a04ff739f2a9 ("RDMA/hns: Add command queue support for hip08 RoCE driver")
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 24 ++++++++----------------
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h |  2 --
 2 files changed, 8 insertions(+), 18 deletions(-)
drivers/infiniband/hw/hns/hns_roce_hw_v2.c +7 -14
index 7a5a41d..260c17c 100644
@@ -1197,8 +1197,7 @@ static void hns_roce_cmq_setup_basic_desc(struct hns_roce_cmq_desc *desc,
 {
 	memset((void *)desc, 0, sizeof(struct hns_roce_cmq_desc));
 	desc->opcode = cpu_to_le16(opcode);
-	desc->flag =
-		cpu_to_le16(HNS_ROCE_CMD_FLAG_NO_INTR | HNS_ROCE_CMD_FLAG_IN);
+	desc->flag = cpu_to_le16(HNS_ROCE_CMD_FLAG_IN);
 	if (is_read)
 		desc->flag |= cpu_to_le16(HNS_ROCE_CMD_FLAG_WR);
 	else
@@ -1275,18 +1274,12 @@ static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 	/* Write to hardware */
 	roce_write(hr_dev, ROCEE_TX_CMQ_TAIL_REG, csq->next_to_use);
 
-	/*
-	 * If the command is sync, wait for the firmware to write back,
-	 * if multi descriptors to be sent, use the first one to check
-	 */
-	if (le16_to_cpu(desc->flag) & HNS_ROCE_CMD_FLAG_NO_INTR) {
-		do {
-			if (hns_roce_cmq_csq_done(hr_dev))
-				break;
-			udelay(1);
-			timeout++;
-		} while (timeout < priv->cmq.tx_timeout);
-	}
+	do {
+		if (hns_roce_cmq_csq_done(hr_dev))
+			break;
+		udelay(1);
+		timeout++;
+	} while (timeout < priv->cmq.tx_timeout);
 
 	if (hns_roce_cmq_csq_done(hr_dev)) {
 		handle = 0;

drivers/infiniband/hw/hns/hns_roce_hw_v2.c static int hns_roce_set_vf_switch_param(struct hns_roce_dev *hr_dev, int vf_id) · line 1630

Quoted 5 lines
drivers/infiniband/hw/hns/hns_roce_hw_v2.c +0 -2
@@ -1626,8 +1619,7 @@ static int hns_roce_set_vf_switch_param(struct hns_roce_dev *hr_dev, int vf_id)
 	if (ret)
 		return ret;
 
-	desc.flag =
-		cpu_to_le16(HNS_ROCE_CMD_FLAG_NO_INTR | HNS_ROCE_CMD_FLAG_IN);

The old firmware needs this redundant flag, it is best cleaned up after the firmware version is released.

Quoted 3 lines
Quoted 1 line

Thanks.

Got it, I will drop this one from the series.

And how will it work new kernel with old FW? Or isn't it possible and all users must upgrade their kernel/FW at the same time?

Thanks

Quoted 29 lines

Thanks
Weihang

Quoted 24 lines
Quoted 24 lines
drivers/infiniband/hw/hns/hns_roce_hw_v2.c +1 -0
+	desc.flag = cpu_to_le16(HNS_ROCE_CMD_FLAG_IN);
 	desc.flag &= cpu_to_le16(~HNS_ROCE_CMD_FLAG_WR);
 	roce_set_bit(swt->cfg, VF_SWITCH_DATA_CFG_ALW_LPBK_S, 1);
 	roce_set_bit(swt->cfg, VF_SWITCH_DATA_CFG_ALW_LCL_LPBK_S, 0);
drivers/infiniband/hw/hns/hns_roce_hw_v2.h +0 -2
index 9f97e32..986a287 100644
@@ -128,14 +128,12 @@
 #define HNS_ROCE_CMD_FLAG_OUT_VALID_SHIFT	1
 #define HNS_ROCE_CMD_FLAG_NEXT_SHIFT		2
 #define HNS_ROCE_CMD_FLAG_WR_OR_RD_SHIFT	3
-#define HNS_ROCE_CMD_FLAG_NO_INTR_SHIFT		4
 #define HNS_ROCE_CMD_FLAG_ERR_INTR_SHIFT	5
 
 #define HNS_ROCE_CMD_FLAG_IN		BIT(HNS_ROCE_CMD_FLAG_IN_VALID_SHIFT)
 #define HNS_ROCE_CMD_FLAG_OUT		BIT(HNS_ROCE_CMD_FLAG_OUT_VALID_SHIFT)
 #define HNS_ROCE_CMD_FLAG_NEXT		BIT(HNS_ROCE_CMD_FLAG_NEXT_SHIFT)
 #define HNS_ROCE_CMD_FLAG_WR		BIT(HNS_ROCE_CMD_FLAG_WR_OR_RD_SHIFT)
-#define HNS_ROCE_CMD_FLAG_NO_INTR	BIT(HNS_ROCE_CMD_FLAG_NO_INTR_SHIFT)
 #define HNS_ROCE_CMD_FLAG_ERR_INTR	BIT(HNS_ROCE_CMD_FLAG_ERR_INTR_SHIFT)
 
 #define HNS_ROCE_CMQ_DESC_NUM_S		3
author
#6 lore raw

PATCH 3/6 RDMA/hns: Fixes missing error code of CMDQ

In reply to Weihang Li

From: Lang Cheng <[email protected]>

When posting a multi-descriptors command, the error code of previous failed descriptors may be rewrote to 0 by a later successful descriptor.

Fixes a04ff739f2a9 ("RDMA/hns: Add command queue support for hip08 RoCE driver")
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
drivers/infiniband/hw/hns/hns_roce_hw_v2.c +3 -4
index 260c17c..13f7897 100644
@@ -1245,7 +1245,7 @@ static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 	u32 timeout = 0;
 	int handle = 0;
 	u16 desc_ret;
-	int ret = 0;
+	int ret;
 	int ntc;
 
 	spin_lock_bh(&csq->lock);
@@ -1283,15 +1283,14 @@ static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 
 	if (hns_roce_cmq_csq_done(hr_dev)) {
 		handle = 0;
+		ret = 0;
 		while (handle < num) {
 			/* get the result of hardware write back */
 			desc_to_use = &csq->desc[ntc];
 			desc[handle] = *desc_to_use;
 			dev_dbg(hr_dev->dev, "Get cmq desc:\n");
 			desc_ret = le16_to_cpu(desc[handle].retval);
-			if (desc_ret == CMD_EXEC_SUCCESS)
-				ret = 0;
-			else
+			if (unlikely(desc_ret != CMD_EXEC_SUCCESS))
 				ret = -EIO;
 
 			ntc++;
Signature · 1 line
2.8.1
author
#7 lore raw

PATCH 4/6 RDMA/hns: Remove redundant operations on CMDQ

In reply to Weihang Li

From: Lang Cheng <[email protected]>

CMDQ works serially, after each successful transmission, the head and tail pointers will be equal, so there is no need to check whether the queue is full. At the same time, since the descriptor of each transmission is new, there is no need to perform a cleanup operation. Then, the field named next_to_clean in structure hns_roce_v2_cmq_ring is redundant.

Fixes a04ff739f2a9 ("RDMA/hns: Add command queue support for hip08 RoCE driver")
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 54 +++---------------------------
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h |  1 -
 2 files changed, 5 insertions(+), 50 deletions(-)
drivers/infiniband/hw/hns/hns_roce_hw_v2.c +5 -49
index 13f7897..04ff0ec 100644
@@ -1059,15 +1059,6 @@ static int hns_roce_v2_rst_process_cmd(struct hns_roce_dev *hr_dev)
 	return 0;
 }
 
-static int hns_roce_cmq_space(struct hns_roce_v2_cmq_ring *ring)
-{
-	int ntu = ring->next_to_use;
-	int ntc = ring->next_to_clean;
-	int used = (ntu - ntc + ring->desc_num) % ring->desc_num;
-
-	return ring->desc_num - used - 1;
-}
-
 static int hns_roce_alloc_cmq_desc(struct hns_roce_dev *hr_dev,
 				   struct hns_roce_v2_cmq_ring *ring)
 {
@@ -1107,7 +1098,6 @@ static int hns_roce_init_cmq_ring(struct hns_roce_dev *hr_dev, bool ring_type)
 					    &priv->cmq.csq : &priv->cmq.crq;
 
 	ring->flag = ring_type;
-	ring->next_to_clean = 0;
 	ring->next_to_use = 0;
 
 	return hns_roce_alloc_cmq_desc(hr_dev, ring);
@@ -1212,30 +1202,6 @@ static int hns_roce_cmq_csq_done(struct hns_roce_dev *hr_dev)
 	return head == priv->cmq.csq.next_to_use;
 }
 
-static int hns_roce_cmq_csq_clean(struct hns_roce_dev *hr_dev)
-{
-	struct hns_roce_v2_priv *priv = hr_dev->priv;
-	struct hns_roce_v2_cmq_ring *csq = &priv->cmq.csq;
-	struct hns_roce_cmq_desc *desc;
-	u16 ntc = csq->next_to_clean;
-	u32 head;
-	int clean = 0;
-
-	desc = &csq->desc[ntc];
-	head = roce_read(hr_dev, ROCEE_TX_CMQ_HEAD_REG);
-	while (head != ntc) {
-		memset(desc, 0, sizeof(*desc));
-		ntc++;
-		if (ntc == csq->desc_num)
-			ntc = 0;
-		desc = &csq->desc[ntc];
-		clean++;
-	}
-	csq->next_to_clean = ntc;
-
-	return clean;
-}
-
 static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 			       struct hns_roce_cmq_desc *desc, int num)
 {
@@ -1250,15 +1216,6 @@ static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 
 	spin_lock_bh(&csq->lock);
 
-	if (num > hns_roce_cmq_space(csq)) {
-		spin_unlock_bh(&csq->lock);
-		return -EBUSY;
-	}
-
-	/*
-	 * Record the location of desc in the cmq for this time
-	 * which will be use for hardware to write back
-	 */
 	ntc = csq->next_to_use;
 
 	while (handle < num) {
@@ -1299,15 +1256,14 @@ static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 				ntc = 0;
 		}
 	} else {
+		/* FW/HW reset or incorrect number of desc */
+		ntc = roce_read(hr_dev, ROCEE_TX_CMQ_HEAD_REG);
+		dev_warn(hr_dev->dev, "CMDQ move head from %d to %d\n",
+			 csq->next_to_use, ntc);
+		csq->next_to_use = ntc;
 		ret = -EAGAIN;
 	}
 
-	/* clean the command send queue */
-	handle = hns_roce_cmq_csq_clean(hr_dev);
-	if (handle != num)
-		dev_warn(hr_dev->dev, "Cleaned %d, need to clean %d\n",
-			 handle, num);
-
 	spin_unlock_bh(&csq->lock);
 
 	return ret;
drivers/infiniband/hw/hns/hns_roce_hw_v2.h +0 -1
index 986a287..db77d2c 100644
@@ -1828,7 +1828,6 @@ struct hns_roce_v2_cmq_ring {
 	u16 buf_size;
 	u16 desc_num;
 	int next_to_use;
-	int next_to_clean;
 	u8 flag;
 	spinlock_t lock; /* command queue lock */
 };
Signature · 1 line
2.8.1
author
#8 lore raw

PATCH 5/6 RDMA/hns: Adjust fields and variables about CMDQ tail/head

In reply to Weihang Li

From: Lang Cheng <[email protected]>

The register 0x07014 is actually the head pointer of CMDQ, and 0x07010 means tail pointer. Current definitions are confusing, so rename them and related variables.

The next_to_use of structure hns_roce_v2_cmq_ring has the same semantics as head, merge them into one member. The next_to_clean of structure hns_roce_v2_cmq_ring has the same semantics as tail. After deleting next_to_clean, tail should also be deleted.

---
 drivers/infiniband/hw/hns/hns_roce_common.h |  4 ++--
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 37 +++++++++++++++--------------
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  |  3 ---
 3 files changed, 21 insertions(+), 23 deletions(-)
drivers/infiniband/hw/hns/hns_roce_common.h +2 -2
index 3ca6e88..23c438c 100644
@@ -364,8 +364,8 @@
 #define ROCEE_TX_CMQ_BASEADDR_L_REG		0x07000
 #define ROCEE_TX_CMQ_BASEADDR_H_REG		0x07004
 #define ROCEE_TX_CMQ_DEPTH_REG			0x07008
-#define ROCEE_TX_CMQ_TAIL_REG			0x07010
-#define ROCEE_TX_CMQ_HEAD_REG			0x07014
+#define ROCEE_TX_CMQ_HEAD_REG			0x07010
+#define ROCEE_TX_CMQ_TAIL_REG			0x07014
 
 #define ROCEE_RX_CMQ_BASEADDR_L_REG		0x07018
 #define ROCEE_RX_CMQ_BASEADDR_H_REG		0x0701c
drivers/infiniband/hw/hns/hns_roce_hw_v2.c +19 -18
index 04ff0ec..23a69cf 100644
@@ -1098,7 +1098,7 @@ static int hns_roce_init_cmq_ring(struct hns_roce_dev *hr_dev, bool ring_type)
 					    &priv->cmq.csq : &priv->cmq.crq;
 
 	ring->flag = ring_type;
-	ring->next_to_use = 0;
+	ring->head = 0;
 
 	return hns_roce_alloc_cmq_desc(hr_dev, ring);
 }
@@ -1196,10 +1196,10 @@ static void hns_roce_cmq_setup_basic_desc(struct hns_roce_cmq_desc *desc,
 
 static int hns_roce_cmq_csq_done(struct hns_roce_dev *hr_dev)
 {
-	u32 head = roce_read(hr_dev, ROCEE_TX_CMQ_HEAD_REG);
+	u32 tail = roce_read(hr_dev, ROCEE_TX_CMQ_TAIL_REG);
 	struct hns_roce_v2_priv *priv = hr_dev->priv;
 
-	return head == priv->cmq.csq.next_to_use;
+	return tail == priv->cmq.csq.head;
 }
 
 static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
@@ -1211,25 +1211,25 @@ static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 	u32 timeout = 0;
 	int handle = 0;
 	u16 desc_ret;
+	u32 tail;
 	int ret;
-	int ntc;
 
 	spin_lock_bh(&csq->lock);
 
-	ntc = csq->next_to_use;
+	tail = csq->head;
 
 	while (handle < num) {
-		desc_to_use = &csq->desc[csq->next_to_use];
+		desc_to_use = &csq->desc[csq->head];
 		*desc_to_use = desc[handle];
 		dev_dbg(hr_dev->dev, "set cmq desc:\n");
-		csq->next_to_use++;
-		if (csq->next_to_use == csq->desc_num)
-			csq->next_to_use = 0;
+		csq->head++;
+		if (csq->head == csq->desc_num)
+			csq->head = 0;
 		handle++;
 	}
 
 	/* Write to hardware */
-	roce_write(hr_dev, ROCEE_TX_CMQ_TAIL_REG, csq->next_to_use);
+	roce_write(hr_dev, ROCEE_TX_CMQ_HEAD_REG, csq->head);
 
 	do {
 		if (hns_roce_cmq_csq_done(hr_dev))
@@ -1243,24 +1243,25 @@ static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 		ret = 0;
 		while (handle < num) {
 			/* get the result of hardware write back */
-			desc_to_use = &csq->desc[ntc];
+			desc_to_use = &csq->desc[tail];
 			desc[handle] = *desc_to_use;
 			dev_dbg(hr_dev->dev, "Get cmq desc:\n");
 			desc_ret = le16_to_cpu(desc[handle].retval);
 			if (unlikely(desc_ret != CMD_EXEC_SUCCESS))
 				ret = -EIO;
 
-			ntc++;
+			tail++;
 			handle++;
-			if (ntc == csq->desc_num)
-				ntc = 0;
+			if (tail == csq->desc_num)
+				tail = 0;
 		}
 	} else {
 		/* FW/HW reset or incorrect number of desc */
-		ntc = roce_read(hr_dev, ROCEE_TX_CMQ_HEAD_REG);
-		dev_warn(hr_dev->dev, "CMDQ move head from %d to %d\n",
-			 csq->next_to_use, ntc);
-		csq->next_to_use = ntc;
+		tail = roce_read(hr_dev, ROCEE_TX_CMQ_TAIL_REG);
+		dev_warn(hr_dev->dev, "CMDQ move tail from %d to %d\n",
+			 csq->head, tail);
+		csq->head = tail;
+
 		ret = -EAGAIN;
 	}
 
drivers/infiniband/hw/hns/hns_roce_hw_v2.h +0 -3
index db77d2c..691b757 100644
@@ -1823,11 +1823,8 @@ struct hns_roce_v2_cmq_ring {
 	dma_addr_t desc_dma_addr;
 	struct hns_roce_cmq_desc *desc;
 	u32 head;
-	u32 tail;
-
 	u16 buf_size;
 	u16 desc_num;
-	int next_to_use;
 	u8 flag;
 	spinlock_t lock; /* command queue lock */
 };
Signature · 1 line
2.8.1
author
#9 lore raw

PATCH 6/6 RDMA/hns: Refactor process of posting CMDQ

In reply to Weihang Li

From: Lang Cheng <[email protected]>

Simplify __hns_roce_cmq_send() then remove the redundant variables.

---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 39 ++++++++++++------------------
 1 file changed, 16 insertions(+), 23 deletions(-)
drivers/infiniband/hw/hns/hns_roce_hw_v2.c +16 -23
index 23a69cf..69b210a 100644
@@ -1207,25 +1207,20 @@ static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 {
 	struct hns_roce_v2_priv *priv = hr_dev->priv;
 	struct hns_roce_v2_cmq_ring *csq = &priv->cmq.csq;
-	struct hns_roce_cmq_desc *desc_to_use;
 	u32 timeout = 0;
-	int handle = 0;
 	u16 desc_ret;
 	u32 tail;
 	int ret;
+	int i;
 
 	spin_lock_bh(&csq->lock);
 
 	tail = csq->head;
 
-	while (handle < num) {
-		desc_to_use = &csq->desc[csq->head];
-		*desc_to_use = desc[handle];
-		dev_dbg(hr_dev->dev, "set cmq desc:\n");
-		csq->head++;
+	for (i = 0; i < num; i++) {
+		csq->desc[csq->head++] = desc[i];
 		if (csq->head == csq->desc_num)
 			csq->head = 0;
-		handle++;
 	}
 
 	/* Write to hardware */
@@ -1235,25 +1230,23 @@ static int __hns_roce_cmq_send(struct hns_roce_dev *hr_dev,
 		if (hns_roce_cmq_csq_done(hr_dev))
 			break;
 		udelay(1);
-		timeout++;
-	} while (timeout < priv->cmq.tx_timeout);
+	} while (++timeout < priv->cmq.tx_timeout);
 
 	if (hns_roce_cmq_csq_done(hr_dev)) {
-		handle = 0;
-		ret = 0;
-		while (handle < num) {
-			/* get the result of hardware write back */
-			desc_to_use = &csq->desc[tail];
-			desc[handle] = *desc_to_use;
-			dev_dbg(hr_dev->dev, "Get cmq desc:\n");
-			desc_ret = le16_to_cpu(desc[handle].retval);
-			if (unlikely(desc_ret != CMD_EXEC_SUCCESS))
-				ret = -EIO;
-
-			tail++;
-			handle++;
+		for (ret = 0, i = 0; i < num; i++) {
+			/* check the result of hardware write back */
+			desc[i] = csq->desc[tail++];
 			if (tail == csq->desc_num)
 				tail = 0;
+
+			desc_ret = le16_to_cpu(desc[i].retval);
+			if (likely(desc_ret == CMD_EXEC_SUCCESS))
+				continue;
+
+			dev_err_ratelimited(hr_dev->dev,
+					    "Cmdq IO error, opcode = %x, return = %x\n",
+					    desc->opcode, desc_ret);
+			ret = -EIO;
 		}
 	} else {
 		/* FW/HW reset or incorrect number of desc */
Signature · 1 line
2.8.1