Skip to contentSkip to conversation

UFS Multi-Circular Queue (MCQ)

RFC PATCH In review Asutosh Das 3 participants 21 messages Sep 2, 2022 – Sep 12, 2022

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

v3 cover UFS Multi-Circular Queue (MCQ)

UFS Multi-Circular Queue (MCQ) has been added in UFSHCI v4.0 to improve storage performance. This patch series is a RFC implementation of this.

This is the initial driver implementation and it has been verified by booting on an emulation platform. During testing, all low power modes were disabled and it was in HS-G1 mode.

Please take a look and let us know your thoughts.

v2 -> v3:
- Split the changes based on functionality
- Addressed queue configuration issues
- Faster SQE tail pointer increments
- Addressed comments from Bart and Manivannan

v1 -> v2:
- Enabled host_tagset
- Added queue num configuration support
- Added one more vops to allow vendor provide the wanted MAC
- Determine nutrs and can_queue by considering both MAC, bqueuedepth and EXT_IID support
- Postponed MCQ initialization and scsi_add_host() to async probe
- Used (EXT_IID, Task Tag) tuple to support up to 4096 tasks (theoretically)

Asutosh Das (2):
ufs: core: prepare ufs for multi circular queue support
ufs: core: mcq: Adds Multi-Circular Queue support

Can Guo (2):
ufs: core: Add Event Specific Interrupt configuration vendor specific
ops
ufs: host: qcom: Add MCQ support

 drivers/ufs/core/Makefile      |   2 +-
 drivers/ufs/core/ufs-mcq.c     | 474 +++++++++++++++++++++++++++++++++++++++++
 drivers/ufs/core/ufshcd-priv.h | 102 +++++++++
 drivers/ufs/core/ufshcd.c      | 370 +++++++++++++++++++++++++-------
 drivers/ufs/host/ufs-qcom.c    | 134 ++++++++++++
 drivers/ufs/host/ufs-qcom.h    |  13 ++
 include/ufs/ufs.h              |   6 +
 include/ufs/ufshcd.h           | 137 ++++++++++++
 include/ufs/ufshci.h           |  79 +++++++
 9 files changed, 1238 insertions(+), 79 deletions(-)
 create mode 100644 drivers/ufs/core/ufs-mcq.c
Signature · 1 line
2.7.4
#1 lore raw

RFC PATCH v3 1/4 ufs: core: prepare ufs for multi circular queue support

Preparatory changes for upcoming multi circular queue.

---
 drivers/ufs/core/ufshcd.c | 99 +++++++++++++++++++++++++----------------------
 1 file changed, 53 insertions(+), 46 deletions(-)
drivers/ufs/core/ufshcd.c +53 -46
index f4f8ded..b119f45 100644
@@ -249,7 +249,7 @@ static int ufshcd_reset_and_restore(struct ufs_hba *hba);
 static int ufshcd_eh_host_reset_handler(struct scsi_cmnd *cmd);
 static int ufshcd_clear_tm_cmd(struct ufs_hba *hba, int tag);
 static void ufshcd_hba_exit(struct ufs_hba *hba);
-static int ufshcd_probe_hba(struct ufs_hba *hba, bool init_dev_params);
+static int ufshcd_probe_hba(struct ufs_hba *hba, bool initial_boot);
 static int ufshcd_setup_clocks(struct ufs_hba *hba, bool on);
 static inline void ufshcd_add_delay_before_dme_cmd(struct ufs_hba *hba);
 static int ufshcd_host_reset_and_restore(struct ufs_hba *hba);
@@ -310,10 +310,11 @@ static void ufshcd_scsi_block_requests(struct ufs_hba *hba)
 		scsi_block_requests(hba->host);
 }
 
-static void ufshcd_add_cmd_upiu_trace(struct ufs_hba *hba, unsigned int tag,
+static void ufshcd_add_cmd_upiu_trace(struct ufs_hba *hba,
+				      struct ufshcd_lrb *lrbp,
 				      enum ufs_trace_str_t str_t)
 {
-	struct utp_upiu_req *rq = hba->lrb[tag].ucd_req_ptr;
+	struct utp_upiu_req *rq = lrbp->ucd_req_ptr;
 	struct utp_upiu_header *header;
 
 	if (!trace_ufshcd_upiu_enabled())
@@ -322,7 +323,7 @@ static void ufshcd_add_cmd_upiu_trace(struct ufs_hba *hba, unsigned int tag,
 	if (str_t == UFS_CMD_SEND)
 		header = &rq->header;
 	else
-		header = &hba->lrb[tag].ucd_rsp_ptr->header;
+		header = &lrbp->ucd_rsp_ptr->header;
 
 	trace_ufshcd_upiu(dev_name(hba->dev), str_t, header, &rq->sc.cdb,
 			  UFS_TSF_CDB);
@@ -379,13 +380,13 @@ static void ufshcd_add_uic_command_trace(struct ufs_hba *hba,
 				 ufshcd_readl(hba, REG_UIC_COMMAND_ARG_3));
 }
 
-static void ufshcd_add_command_trace(struct ufs_hba *hba, unsigned int tag,
+static void ufshcd_add_command_trace(struct ufs_hba *hba,
+				     struct ufshcd_lrb *lrbp,
 				     enum ufs_trace_str_t str_t)
 {
 	u64 lba = 0;
 	u8 opcode = 0, group_id = 0;
 	u32 intr, doorbell;
-	struct ufshcd_lrb *lrbp = &hba->lrb[tag];
 	struct scsi_cmnd *cmd = lrbp->cmd;
 	struct request *rq = scsi_cmd_to_rq(cmd);
 	int transfer_len = -1;
@@ -394,7 +395,7 @@ static void ufshcd_add_command_trace(struct ufs_hba *hba, unsigned int tag,
 		return;
 
 	/* trace UPIU also */
-	ufshcd_add_cmd_upiu_trace(hba, tag, str_t);
+	ufshcd_add_cmd_upiu_trace(hba, lrbp, str_t);
 	if (!trace_ufshcd_command_enabled())
 		return;
 
@@ -419,7 +420,7 @@ static void ufshcd_add_command_trace(struct ufs_hba *hba, unsigned int tag,
 
 	intr = ufshcd_readl(hba, REG_INTERRUPT_STATUS);
 	doorbell = ufshcd_readl(hba, REG_UTP_TRANSFER_REQ_DOOR_BELL);
-	trace_ufshcd_command(dev_name(hba->dev), str_t, tag,
+	trace_ufshcd_command(dev_name(hba->dev), str_t, lrbp->task_tag,
 			doorbell, transfer_len, intr, lba, opcode, group_id);
 }
 
@@ -2134,14 +2135,14 @@ static void ufshcd_update_monitor(struct ufs_hba *hba, const struct ufshcd_lrb *
  * @task_tag: Task tag of the command
  */
 static inline
-void ufshcd_send_command(struct ufs_hba *hba, unsigned int task_tag)
+void ufshcd_send_command(struct ufs_hba *hba, struct ufshcd_lrb *lrbp)
 {
-	struct ufshcd_lrb *lrbp = &hba->lrb[task_tag];
+	int task_tag = lrbp->task_tag;
 	unsigned long flags;
 
 	lrbp->issue_time_stamp = ktime_get();
 	lrbp->compl_time_stamp = ktime_set(0, 0);
-	ufshcd_add_command_trace(hba, task_tag, UFS_CMD_SEND);
+	ufshcd_add_command_trace(hba, lrbp, UFS_CMD_SEND);
 	ufshcd_clk_scaling_start_busy(hba);
 	if (unlikely(ufshcd_should_inform_monitor(hba, lrbp)))
 		ufshcd_start_monitor(hba, lrbp);
@@ -2553,9 +2554,10 @@ void ufshcd_prepare_utp_scsi_cmd_upiu(struct ufshcd_lrb *lrbp, u8 upiu_flags)
 	/* command descriptor fields */
 	ucd_req_ptr->header.dword_0 = UPIU_HEADER_DWORD(
 				UPIU_TRANSACTION_COMMAND, upiu_flags,
-				lrbp->lun, lrbp->task_tag);
+				lrbp->lun, lrbp->task_tag & 0xff);
 	ucd_req_ptr->header.dword_1 = UPIU_HEADER_DWORD(
-				UPIU_COMMAND_SET_TYPE_SCSI, 0, 0, 0);
+				UPIU_COMMAND_SET_TYPE_SCSI, 0, 0,
+				(lrbp->task_tag & 0xf00) << 4);
 
 	/* Total EHS length and Data segment length will be zero */
 	ucd_req_ptr->header.dword_2 = 0;
@@ -2845,7 +2847,7 @@ static int ufshcd_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *cmd)
 		goto out;
 	}
 
-	ufshcd_send_command(hba, tag);
+	ufshcd_send_command(hba, lrbp);
 
 out:
 	rcu_read_unlock();
@@ -2971,7 +2973,7 @@ static int ufshcd_wait_for_dev_cmd(struct ufs_hba *hba,
 
 	if (!time_left) {
 		err = -ETIMEDOUT;
-		dev_dbg(hba->dev, "%s: dev_cmd request timedout, tag %d\n",
+		dev_err(hba->dev, "%s: dev_cmd request timedout, tag %d\n",
 			__func__, lrbp->task_tag);
 		if (!ufshcd_clear_cmds(hba, 1U << lrbp->task_tag))
 			/* successfully cleared the command, retry if needed */
@@ -3021,7 +3023,7 @@ static int ufshcd_exec_dev_cmd(struct ufs_hba *hba,
 
 	ufshcd_add_query_upiu_trace(hba, UFS_QUERY_SEND, lrbp->ucd_req_ptr);
 
-	ufshcd_send_command(hba, tag);
+	ufshcd_send_command(hba, lrbp);
 	err = ufshcd_wait_for_dev_cmd(hba, lrbp, timeout);
 	ufshcd_add_query_upiu_trace(hba, err ? UFS_QUERY_ERR : UFS_QUERY_COMP,
 				    (struct utp_upiu_req *)lrbp->ucd_rsp_ptr);
@@ -4513,6 +4515,7 @@ int ufshcd_make_hba_operational(struct ufs_hba *hba)
 			REG_UTP_TRANSFER_REQ_LIST_BASE_L);
 	ufshcd_writel(hba, upper_32_bits(hba->utrdl_dma_addr),
 			REG_UTP_TRANSFER_REQ_LIST_BASE_H);
+
 	ufshcd_writel(hba, lower_32_bits(hba->utmrdl_dma_addr),
 			REG_UTP_TASK_REQ_LIST_BASE_L);
 	ufshcd_writel(hba, upper_32_bits(hba->utmrdl_dma_addr),
@@ -5320,6 +5323,32 @@ static void ufshcd_release_scsi_cmd(struct ufs_hba *hba,
 	ufshcd_clk_scaling_update_busy(hba);
 }
 
+void ufshcd_compl_one_cqe(struct ufs_hba *hba, int task_tag)
+{
+	struct ufshcd_lrb *lrbp;
+	struct scsi_cmnd *cmd;
+
+	lrbp = &hba->lrb[task_tag];
+	lrbp->compl_time_stamp = ktime_get();
+	cmd = lrbp->cmd;
+	if (cmd) {
+		if (unlikely(ufshcd_should_inform_monitor(hba, lrbp)))
+			ufshcd_update_monitor(hba, lrbp);
+		ufshcd_add_command_trace(hba, lrbp, UFS_CMD_COMP);
+		cmd->result = ufshcd_transfer_rsp_status(hba, lrbp);
+		ufshcd_release_scsi_cmd(hba, lrbp);
+		/* Do not touch lrbp after scsi done */
+		scsi_done(cmd);
+	} else if (lrbp->command_type == UTP_CMD_TYPE_DEV_MANAGE ||
+		   lrbp->command_type == UTP_CMD_TYPE_UFS_STORAGE) {
+		if (hba->dev_cmd.complete) {
+			ufshcd_add_command_trace(hba, lrbp, UFS_DEV_COMP);
+			complete(hba->dev_cmd.complete);
+			ufshcd_clk_scaling_update_busy(hba);
+		}
+	}
+}
+
 /**
  * __ufshcd_transfer_req_compl - handle SCSI and query command completion
  * @hba: per adapter instance
@@ -5328,32 +5357,10 @@ static void ufshcd_release_scsi_cmd(struct ufs_hba *hba,
 static void __ufshcd_transfer_req_compl(struct ufs_hba *hba,
 					unsigned long completed_reqs)
 {
-	struct ufshcd_lrb *lrbp;
-	struct scsi_cmnd *cmd;
-	int index;
-
-	for_each_set_bit(index, &completed_reqs, hba->nutrs) {
-		lrbp = &hba->lrb[index];
-		lrbp->compl_time_stamp = ktime_get();
-		cmd = lrbp->cmd;
-		if (cmd) {
-			if (unlikely(ufshcd_should_inform_monitor(hba, lrbp)))
-				ufshcd_update_monitor(hba, lrbp);
-			ufshcd_add_command_trace(hba, index, UFS_CMD_COMP);
-			cmd->result = ufshcd_transfer_rsp_status(hba, lrbp);
-			ufshcd_release_scsi_cmd(hba, lrbp);
-			/* Do not touch lrbp after scsi done */
-			scsi_done(cmd);
-		} else if (lrbp->command_type == UTP_CMD_TYPE_DEV_MANAGE ||
-			lrbp->command_type == UTP_CMD_TYPE_UFS_STORAGE) {
-			if (hba->dev_cmd.complete) {
-				ufshcd_add_command_trace(hba, index,
-							 UFS_DEV_COMP);
-				complete(hba->dev_cmd.complete);
-				ufshcd_clk_scaling_update_busy(hba);
-			}
-		}
-	}
+	int tag;
+
+	for_each_set_bit(tag, &completed_reqs, hba->nutrs)
+		ufshcd_compl_one_cqe(hba, tag);
 }
 
 /*
@@ -6869,7 +6876,7 @@ static int ufshcd_issue_devman_upiu_cmd(struct ufs_hba *hba,
 
 	ufshcd_add_query_upiu_trace(hba, UFS_QUERY_SEND, lrbp->ucd_req_ptr);
 
-	ufshcd_send_command(hba, tag);
+	ufshcd_send_command(hba, lrbp);
 	/*
 	 * ignore the returning value here - ufshcd_check_query_response is
 	 * bound to fail since dev_cmd.query and dev_cmd.type were left empty.
@@ -8138,11 +8145,11 @@ static int ufshcd_add_lus(struct ufs_hba *hba)
 /**
  * ufshcd_probe_hba - probe hba to detect device and initialize it
  * @hba: per-adapter instance
- * @init_dev_params: whether or not to call ufshcd_device_params_init().
+ * @initial_boot: Whether or not from initial bootup
  *
  * Execute link-startup and verify device initialization
  */
-static int ufshcd_probe_hba(struct ufs_hba *hba, bool init_dev_params)
+static int ufshcd_probe_hba(struct ufs_hba *hba, bool initial_boot)
 {
 	int ret;
 	unsigned long flags;
@@ -8177,7 +8184,7 @@ static int ufshcd_probe_hba(struct ufs_hba *hba, bool init_dev_params)
 	 * Initialize UFS device parameters used by driver, these
 	 * parameters are associated with UFS descriptors.
 	 */
-	if (init_dev_params) {
+	if (initial_boot) {
 		ret = ufshcd_device_params_init(hba);
 		if (ret)
 			goto out;
Signature · 1 line
2.7.4
#2 lore raw

Re: [RFC PATCH v3 1/4] ufs: core: prepare ufs for multi circular queue support

Asutosh,

Quoted 13 lines · Asutosh Das

Preparatory changes for upcoming multi circular queue.

---
 drivers/ufs/core/ufshcd.c | 99 +++++++++++++++++++++++++------------
----------

 1 file changed, 53 insertions(+), 46 deletions(-)

(quoted diff) +0 -0
@@ -2134,14 +2135,14 @@ static void ufshcd_update_monitor(struct
Quoted 3 lines

ufs_hba *hba, const struct ufshcd_lrb *
  * @task_tag: Task tag of the command
  */

You didn't change parameter name in the function description.

Quoted 20 lines

 static inline

(quoted diff) +1 -1
-void ufshcd_send_command(struct ufs_hba *hba, unsigned int task_tag)
+void ufshcd_send_command(struct ufs_hba *hba, struct ufshcd_lrb

*lrbp)
 {
-       struct ufshcd_lrb *lrbp = &hba->lrb[task_tag]; +       int task_tag = lrbp->task_tag;
        unsigned long flags;

        lrbp->issue_time_stamp = ktime_get(); lrbp->compl_time_stamp = ktime_set(0, 0);
-       ufshcd_add_command_trace(hba, task_tag, UFS_CMD_SEND); +       ufshcd_add_command_trace(hba, lrbp, UFS_CMD_SEND); ufshcd_clk_scaling_start_busy(hba);
        if (unlikely(ufshcd_should_inform_monitor(hba, lrbp))) ufshcd_start_monitor(hba, lrbp);

(quoted diff) +0 -0
@@ -2553,9 +2554,10 @@ void ufshcd_prepare_utp_scsi_cmd_upiu(struct
 ufshcd_lrb *lrbp, u8 upiu_flags)
         /* command descriptor fields */

        ucd_req_ptr->header.dword_0 = UPIU_HEADER_DWORD(

Quoted 8 lines

                                UPIU_TRANSACTION_COMMAND, upiu_flags, -                               lrbp->lun, lrbp->task_tag); +                               lrbp->lun, lrbp->task_tag & 0xff); ucd_req_ptr->header.dword_1 = UPIU_HEADER_DWORD(
-                               UPIU_COMMAND_SET_TYPE_SCSI, 0, 0, 0); +                               UPIU_COMMAND_SET_TYPE_SCSI, 0, 0, +                               (lrbp->task_tag & 0xf00) << 4);

Are you sure here "(lrbp->task_tag & 0xf00) << 4" is correct?

this will overwrite other fields, see UPIU_HEADER_DWORD:

#define UPIU_HEADER_DWORD(byte3, byte2, byte1, byte0)\
                        cpu_to_be32((byte3 << 24) | (byte2 << 16) |\
                         (byte1 << 8) | (byte0))
Quoted 22 lines


-       ufshcd_send_command(hba, tag);
+       ufshcd_send_command(hba, lrbp);
        err = ufshcd_wait_for_dev_cmd(hba, lrbp, timeout); ufshcd_add_query_upiu_trace(hba, err ? UFS_QUERY_ERR :
UFS_QUERY_COMP,
                                    (struct utp_upiu_req *)lrbp-

Quoted 1 line

ucd_rsp_ptr);

drivers/ufs/core/ufshcd.c +1 -0
@@ -4513,6 +4515,7 @@ int ufshcd_make_hba_operational(struct ufs_hba
 *hba)
                         REG_UTP_TRANSFER_REQ_LIST_BASE_L);
         ufshcd_writel(hba, upper_32_bits(hba->utrdl_dma_addr),
                         REG_UTP_TRANSFER_REQ_LIST_BASE_H);
+
         ufshcd_writel(hba, lower_32_bits(hba->utmrdl_dma_addr),
                         REG_UTP_TASK_REQ_LIST_BASE_L);

        ufshcd_writel(hba, upper_32_bits(hba->utmrdl_dma_addr),

drivers/ufs/core/ufshcd.c +0 -0
@@ -5320,6 +5323,32 @@ static void ufshcd_release_scsi_cmd(struct
 ufs_hba *hba,
         ufshcd_clk_scaling_update_busy(hba);
  }
Quoted 2 lines


+void ufshcd_compl_one_cqe(struct ufs_hba *hba, int task_tag)

This function does only complete one task. What does cqe stand for?

Quoted 25 lines
drivers/ufs/core/ufshcd.c +9 -0
+{
+       struct ufshcd_lrb *lrbp;
+       struct scsi_cmnd *cmd;
+
+       lrbp = &hba->lrb[task_tag];
+       lrbp->compl_time_stamp = ktime_get();
+       cmd = lrbp->cmd;
+       if (cmd) {
+               if (unlikely(ufshcd_should_inform_monitor(hba,

lrbp)))
+                       ufshcd_update_monitor(hba, lrbp); +               ufshcd_add_command_trace(hba, lrbp, UFS_CMD_COMP); +               cmd->result = ufshcd_transfer_rsp_status(hba, lrbp); +               ufshcd_release_scsi_cmd(hba, lrbp);
+               /* Do not touch lrbp after scsi done */
+               scsi_done(cmd);
+       } else if (lrbp->command_type == UTP_CMD_TYPE_DEV_MANAGE || +                  lrbp->command_type == UTP_CMD_TYPE_UFS_STORAGE) { +               if (hba->dev_cmd.complete) {
+                       ufshcd_add_command_trace(hba, lrbp, UFS_DEV_COMP);
+                       complete(hba->dev_cmd.complete); +                       ufshcd_clk_scaling_update_busy(hba); +               }
+       }

Review

Quoted 2 lines
(quoted diff) +2 -0
+}
+

Kind regards,
Bean

#3 lore raw

Re: [RFC PATCH v3 1/4] ufs: core: prepare ufs for multi circular queue support

Hello Bean
Thanks for your comments. Responses inline.

Quoted 3 lines · Bean Huo

Asutosh,

[...]

Quoted 23 lines
Quoted 23 lines

 static inline

(quoted diff) +1 -1
-void ufshcd_send_command(struct ufs_hba *hba, unsigned int task_tag)
+void ufshcd_send_command(struct ufs_hba *hba, struct ufshcd_lrb

*lrbp)
 {
-       struct ufshcd_lrb *lrbp = &hba->lrb[task_tag]; +       int task_tag = lrbp->task_tag;
        unsigned long flags;

        lrbp->issue_time_stamp = ktime_get(); lrbp->compl_time_stamp = ktime_set(0, 0);
-       ufshcd_add_command_trace(hba, task_tag, UFS_CMD_SEND); +       ufshcd_add_command_trace(hba, lrbp, UFS_CMD_SEND);
        ufshcd_clk_scaling_start_busy(hba);
        if (unlikely(ufshcd_should_inform_monitor(hba, lrbp))) ufshcd_start_monitor(hba, lrbp);

drivers/ufs/core/ufshcd.c +0 -1
@@ -2553,9 +2554,10 @@ void ufshcd_prepare_utp_scsi_cmd_upiu(struct
 ufshcd_lrb *lrbp, u8 upiu_flags)
         /* command descriptor fields */
         ucd_req_ptr->header.dword_0 = UPIU_HEADER_DWORD(
                                 UPIU_TRANSACTION_COMMAND, upiu_flags,
-                               lrbp->lun, lrbp->task_tag);
Quoted 9 lines

drivers/ufs/core/ufshcd.c void ufshcd_prepare_utp_scsi_cmd_upiu(struct ufshcd_lrb *lrbp, u8 upiu_flags) · line 2561

Quoted 6 lines
drivers/ufs/core/ufshcd.c +3 -1
+                               lrbp->lun, lrbp->task_tag & 0xff);
         ucd_req_ptr->header.dword_1 = UPIU_HEADER_DWORD(
-                               UPIU_COMMAND_SET_TYPE_SCSI, 0, 0, 0);
+                               UPIU_COMMAND_SET_TYPE_SCSI, 0, 0,
+                               (lrbp->task_tag & 0xf00) << 4);
 

Are you sure here "(lrbp->task_tag & 0xf00) << 4" is correct?

EXT_IID is the higher nibble in DWORD1. So this looks correct to me.

COMMAND UPIU
0 1 2 3
xx000001b Flags LUN Task Tag
4 5 6 7
IID Command Set Type Reserved Reserved EXT_IID | Reserved

Quoted 32 lines

this will overwrite other fields, see UPIU_HEADER_DWORD:

#define UPIU_HEADER_DWORD(byte3, byte2, byte1, byte0)\
                         cpu_to_be32((byte3 << 24) | (byte2 << 16) |\
                          (byte1 << 8) | (byte0))
Quoted 23 lines
(quoted diff) +1 -1
-       ufshcd_send_command(hba, tag);
+       ufshcd_send_command(hba, lrbp);

        err = ufshcd_wait_for_dev_cmd(hba, lrbp, timeout); ufshcd_add_query_upiu_trace(hba, err ? UFS_QUERY_ERR :
UFS_QUERY_COMP,
                                    (struct utp_upiu_req *)lrbp-

Quoted 1 line

ucd_rsp_ptr);

drivers/ufs/core/ufshcd.c +1 -0
@@ -4513,6 +4515,7 @@ int ufshcd_make_hba_operational(struct ufs_hba
 *hba)
                         REG_UTP_TRANSFER_REQ_LIST_BASE_L);
         ufshcd_writel(hba, upper_32_bits(hba->utrdl_dma_addr),
                         REG_UTP_TRANSFER_REQ_LIST_BASE_H);
+
         ufshcd_writel(hba, lower_32_bits(hba->utmrdl_dma_addr),
                         REG_UTP_TASK_REQ_LIST_BASE_L);

        ufshcd_writel(hba, upper_32_bits(hba->utmrdl_dma_addr),

drivers/ufs/core/ufshcd.c +0 -0
@@ -5320,6 +5323,32 @@ static void ufshcd_release_scsi_cmd(struct
 ufs_hba *hba,
         ufshcd_clk_scaling_update_busy(hba);
  }
 
Quoted 4 lines

drivers/ufs/core/ufshcd.c static void ufshcd_release_scsi_cmd(struct ufs_hba *hba, · line 5326

Quoted 1 line
drivers/ufs/core/ufshcd.c +1 -0
+void ufshcd_compl_one_cqe(struct ufs_hba *hba, int task_tag)

This function does only complete one task. What does cqe stand for?

CQE - Completion Queue Entry. The term 'task' is used for TM commands in scsi, hence there was a comment to change it to CQE.

Quoted 32 lines
Quoted 17 lines
drivers/ufs/core/ufshcd.c +16 -0
+{
+       struct ufshcd_lrb *lrbp;
+       struct scsi_cmnd *cmd;
+
+       lrbp = &hba->lrb[task_tag];
+       lrbp->compl_time_stamp = ktime_get();
+       cmd = lrbp->cmd;
+       if (cmd) {
+               if (unlikely(ufshcd_should_inform_monitor(hba,
 lrbp)))
+                       ufshcd_update_monitor(hba, lrbp);
+               ufshcd_add_command_trace(hba, lrbp, UFS_CMD_COMP);
+               cmd->result = ufshcd_transfer_rsp_status(hba, lrbp);
+               ufshcd_release_scsi_cmd(hba, lrbp);
+               /* Do not touch lrbp after scsi done */
+               scsi_done(cmd);
+       } else if (lrbp->command_type == UTP_CMD_TYPE_DEV_MANAGE ||

drivers/ufs/core/ufshcd.c static void ufshcd_release_scsi_cmd(struct ufs_hba *hba, · line 5351

Quoted 10 lines
drivers/ufs/core/ufshcd.c +9 -0
+                  lrbp->command_type == UTP_CMD_TYPE_UFS_STORAGE) {
+               if (hba->dev_cmd.complete) {
+                       ufshcd_add_command_trace(hba, lrbp,
 UFS_DEV_COMP);
+                       complete(hba->dev_cmd.complete);
+                       ufshcd_clk_scaling_update_busy(hba);
+               }
+       }
+}
+

Kind regards,
Bean

#4 lore raw

Re: [RFC PATCH v3 1/4] ufs: core: prepare ufs for multi circular queue support

Quoted 17 lines · Asutosh Das (asd)
Quoted 15 lines
Quoted 6 lines

ufshcd_lrb *lrbp, u8 upiu_flags)
          /* command descriptor fields */ ucd_req_ptr->header.dword_0 = UPIU_HEADER_DWORD( UPIU_TRANSACTION_COMMAND, upiu_flags,
-                               lrbp->lun, lrbp->task_tag);

Quoted 8 lines

+                               lrbp->lun, lrbp->task_tag & 0xff);
          ucd_req_ptr->header.dword_1 = UPIU_HEADER_DWORD( -                               UPIU_COMMAND_SET_TYPE_SCSI, 0, 0, 0);
+                               UPIU_COMMAND_SET_TYPE_SCSI, 0, 0, +                               (lrbp->task_tag & 0xf00) << 4);

Quoted 2 lines

Are you sure here "(lrbp->task_tag & 0xf00) << 4" is correct?

Quoted 7 lines

EXT_IID is the higher nibble in DWORD1. So this looks correct to me.

COMMAND UPIU
0           1     2     3
xx000001b Flags LUN Task Tag
4                       5       6       7
IID Command Set Type Reserved Reserved EXT_IID | Reserved

Hi Asutosh,

yes, [7:4] of byte 7 in UPIU header for host to device UPIUs EXT_IID, this is correct.

but I think byte7 should be (lrbp->task_tag & 0xf00) >> 4); rather than "<< 4"

or what I missundersood.

Kind regards,
Bean

#6 lore raw

Re: [RFC PATCH v3 1/4] ufs: core: prepare ufs for multi circular queue support

Hello Bart,
Thanks for the comments.

Quoted 7 lines · Bart Van Assche
Quoted 1 line · Asutosh Das

Preparatory changes for upcoming multi circular queue.

One patch per change please and also describe each individual change.
From
https://www.kernel.org/doc/html/latest/process/submitting-patches.html#split-changes: "Separate each logical change into a separate patch".

The intent of this change was to have all the non-mcq related changes to the ufshcd as a separate patch. I would add more details to the commit message of this change.
If there's anything specific in this patch that may need changes, please let me know.

Quoted 3 lines

Thanks,

Bart.

-asd

#7 lore raw

Re: [RFC PATCH v3 1/4] ufs: core: prepare ufs for multi circular queue support

Quoted 4 lines · Asutosh Das (asd)

Hello Bart,
Thanks for the comments.

On 9/8/2022 2:58 PM, Bart Van Assche wrote:

Quoted 8 lines
Quoted 5 lines · Asutosh Das
Quoted 1 line

Preparatory changes for upcoming multi circular queue.

One patch per change please and also describe each individual change.  From https://www.kernel.org/doc/html/latest/process/submitting-patches.html#split-changes: "Separate each logical change into a separate patch".

The intent of this change was to have all the non-mcq related changes to the ufshcd as a separate patch. I would add more details to the commit message of this change.
If there's anything specific in this patch that may need changes, please let me know.

Please follow the "one change per patch" rule. This is a widely followed rule in the Linux kernel community. This rule exists because it is the responsibility of the developer(s) who post a patch series to make it easy for reviewers to review their work.

Thanks,

Bart.

#8 lore raw

Re: [RFC PATCH v3 1/4] ufs: core: prepare ufs for multi circular queue support

Quoted 8 lines · Bart Van Assche
Quoted 7 lines · Asutosh Das (asd)

Hello Bart,
Thanks for the comments.

Quoted 3 lines · Bart Van Assche
Quoted 1 line · Asutosh Das

Preparatory changes for upcoming multi circular queue.

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

One patch per change please and also describe each individual change.  From https://www.kernel.org/doc/html/latest/process/submitting-patches.html#split-changes: "Separate each logical change into a separate patch".

The intent of this change was to have all the non-mcq related changes to the ufshcd as a separate patch. I would add more details to the commit message of this change.
If there's anything specific in this patch that may need changes, please let me know.

Please follow the "one change per patch" rule. This is a widely followed rule in the Linux kernel community. This rule exists because it is the responsibility of the developer(s) who post a patch series to make it easy for reviewers to review their work.

Let me review and restructure the code and push it again. I would like to remove the RFC tag in the next version. Please let me know if that's _not_ OK.

Quoted 3 lines

Thanks,

Bart.

#9 lore raw

RFC PATCH v3 2/4 ufs: core: mcq: Adds Multi-Circular Queue support

In reply to Asutosh Das

In v4.0, UFSHCI specification has added Multi-Circular Queue support to enhance performance. It is backward compatible to the legacy Single Doorbell mode.

This patch adds the support for MCQ to ufshc driver. It uses shared tags across queues. It adds support for vendors to define the number of queues for different queue types. It also adds support ESI which is event specific interrupt mechanism for completion path.

---
 drivers/ufs/core/Makefile      |   2 +-
 drivers/ufs/core/ufs-mcq.c     | 459 +++++++++++++++++++++++++++++++++++++++++
 drivers/ufs/core/ufshcd-priv.h |  96 +++++++++
 drivers/ufs/core/ufshcd.c      | 281 +++++++++++++++++++++----
 include/ufs/ufs.h              |   6 +
 include/ufs/ufshcd.h           | 130 ++++++++++++
 include/ufs/ufshci.h           |  79 +++++++
 7 files changed, 1012 insertions(+), 41 deletions(-)
 create mode 100644 drivers/ufs/core/ufs-mcq.c
drivers/ufs/core/Makefile +1 -1
index 62f38c5..4d02e0f 100644
@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: GPL-2.0
 
 obj-$(CONFIG_SCSI_UFSHCD)		+= ufshcd-core.o
-ufshcd-core-y				+= ufshcd.o ufs-sysfs.o
+ufshcd-core-y				+= ufshcd.o ufs-sysfs.o ufs-mcq.o
 ufshcd-core-$(CONFIG_DEBUG_FS)		+= ufs-debugfs.o
 ufshcd-core-$(CONFIG_SCSI_UFS_BSG)	+= ufs_bsg.o
 ufshcd-core-$(CONFIG_SCSI_UFS_CRYPTO)	+= ufshcd-crypto.o
drivers/ufs/core/ufs-mcq.c new file +459 -0

459 lines not on this page · raw message

drivers/ufs/core/ufshcd-priv.h +96 -0

114 lines not on this page · raw message

drivers/ufs/core/ufshcd.c +241 -40

484 lines not on this page · raw message

include/ufs/ufs.h +6 -0
index 1bba3fe..5112418 100644
@@ -165,6 +165,7 @@ enum attr_idn {
 	QUERY_ATTR_IDN_AVAIL_WB_BUFF_SIZE       = 0x1D,
 	QUERY_ATTR_IDN_WB_BUFF_LIFE_TIME_EST    = 0x1E,
 	QUERY_ATTR_IDN_CURR_WB_BUFF_SIZE        = 0x1F,
+	QUERY_ATTR_IDN_EXT_IID_EN		= 0x2A,
 };
 
 /* Descriptor idn for Query requests */
@@ -352,6 +353,7 @@ enum {
 	UFS_DEV_EXT_TEMP_NOTIF		= BIT(6),
 	UFS_DEV_HPB_SUPPORT		= BIT(7),
 	UFS_DEV_WRITE_BOOSTER_SUP	= BIT(8),
+	UFS_DEV_EXT_IID_SUP		= BIT(16),
 };
 #define UFS_DEV_HPB_SUPPORT_VERSION		0x310
 
@@ -589,6 +591,8 @@ struct ufs_dev_info {
 	u8	*model;
 	u16	wspecversion;
 	u32	clk_gating_wait_us;
+	/* Stores the depth of queue in UFS device */
+	u8	bqueuedepth;
 
 	/* UFS HPB related flag */
 	bool	hpb_enabled;
@@ -601,6 +605,8 @@ struct ufs_dev_info {
 
 	bool	b_rpm_dev_flush_capable;
 	u8	b_presrv_uspc_en;
+	/* UFS EXT_IID Enable */
+	bool	b_ext_iid_en;
 };
 
 /*
include/ufs/ufshcd.h +130 -0

186 lines not on this page · raw message

include/ufs/ufshci.h +79 -0

115 lines not on this page · raw message

Signature · 1 line
2.7.4
#10 lore raw

Re: [RFC PATCH v3 2/4] ufs: core: mcq: Adds Multi-Circular Queue support

Quoted 31 lines · Asutosh Das
drivers/ufs/core/ufs-mcq.c new file +30 -0
+/**
+ * ufshcd_mcq_decide_queue_depth - decide the queue depth
+ * @hba - per adapter instance
+ *
+ * MAC - Max. Active Command of the Host Controller (HC)
+ * HC wouldn't send more than this commands to the device.
+ * The default MAC is 32, but the max. value may vary with
+ * vendor implementation.
+ * Calculates and adjusts the queue depth based on the depth
+ * supported by the HC, ufs device and if ext_iid is supported.
+ */
+u32 ufshcd_mcq_decide_queue_depth(struct ufs_hba *hba)
+{
+       u32 qd, val;
+       int mac;
+
+       mac = ufshcd_mcq_vops_get_hba_mac(hba);
+       if (mac < 0) {
+               val = ufshcd_readl(hba, REG_UFS_MCQ_CFG);
+               mac = (val & MCQ_CFG_MAC_MASK) >> MCQ_CFG_MAC_OFFSET;
+       }
+
+       /*  MAC is a 0 based value. */
+       mac += 1;
+       qd = min_t(u32, mac, hba->dev_info.bqueuedepth);
+       if (!qd)
+               qd = mac;
+
+       /* Adjust queue depth based on ext_iid support. */
+       if (qd > 256 && (!hba->ext_iid_sup || !hba-
Quoted 1 line

dev_info.b_ext_iid_en))

Quoted 5 lines

+               qd = 256;
+       else if (qd > 4096 && hba->ext_iid_sup && hba-

Quoted 1 line

dev_info.b_ext_iid_en)

+               qd = 4096;
+

Astosh,

I don't understand here. How can qd be 4096? MAC in UFSHCI is 9 bits width, the maximum depth will be 512, and hba->dev_info.bqueuedepth is one byte width, maximum will be 256.

Kind regards,
Bean

#11 lore raw

Re: [RFC PATCH v3 2/4] ufs: core: mcq: Adds Multi-Circular Queue support

Hello Bean
Thanks for your comments. Responses inline.

Quoted 35 lines · Bean Huo
Quoted 34 lines · Asutosh Das
drivers/ufs/core/ufs-mcq.c new file +30 -0
+/**
+ * ufshcd_mcq_decide_queue_depth - decide the queue depth
+ * @hba - per adapter instance
+ *
+ * MAC - Max. Active Command of the Host Controller (HC)
+ * HC wouldn't send more than this commands to the device.
+ * The default MAC is 32, but the max. value may vary with
+ * vendor implementation.
+ * Calculates and adjusts the queue depth based on the depth
+ * supported by the HC, ufs device and if ext_iid is supported.
+ */
+u32 ufshcd_mcq_decide_queue_depth(struct ufs_hba *hba)
+{
+       u32 qd, val;
+       int mac;
+
+       mac = ufshcd_mcq_vops_get_hba_mac(hba);
+       if (mac < 0) {
+               val = ufshcd_readl(hba, REG_UFS_MCQ_CFG);
+               mac = (val & MCQ_CFG_MAC_MASK) >> MCQ_CFG_MAC_OFFSET;
+       }
+
+       /*  MAC is a 0 based value. */
+       mac += 1;
+       qd = min_t(u32, mac, hba->dev_info.bqueuedepth);
+       if (!qd)
+               qd = mac;
+
+       /* Adjust queue depth based on ext_iid support. */
+       if (qd > 256 && (!hba->ext_iid_sup || !hba-
Quoted 1 line

dev_info.b_ext_iid_en))

+               qd = 256;
+       else if (qd > 4096 && hba->ext_iid_sup && hba-

Quoted 1 line

dev_info.b_ext_iid_en)

Quoted 9 lines
Quoted 2 lines

+               qd = 4096;
+

Astosh,

I don't understand here. How can qd be 4096? MAC in UFSHCI is 9 bits width, the maximum depth will be 512, and hba->dev_info.bqueuedepth is one byte width, maximum will be 256.

Correct. The spec is fuzzy about it. Because ext_iid provides a provision to increase the Queue depth to 2^12. Even though the parameters that you referred to don't support it now.

Will remove this part in the next version.

Quoted 3 lines

Kind regards,
Bean

#12 lore raw

Re: [RFC PATCH v3 2/4] ufs: core: mcq: Adds Multi-Circular Queue support

In reply to Asutosh Das

drivers/ufs/core/ufs-mcq.c line 128

Quoted 7 lines · Asutosh Das
drivers/ufs/core/ufs-mcq.c new file +7 -0
+struct ufs_hw_queue *ufshcd_mcq_req_to_hwq(struct ufs_hba *hba,
+                                        struct request *req)
+{
+       u32 utag, hwq;
+
+       utag = blk_mq_unique_tag(req);
+       hwq = blk_mq_unique_tag_to_hwq(utag);

I think you could directly get queue number from req->mq_hctx-

Quoted 1 line

queue_num.

#13 lore raw

Re: [RFC PATCH v3 2/4] ufs: core: mcq: Adds Multi-Circular Queue support

In reply to Asutosh Das
Quoted 13 lines · Asutosh Das
drivers/ufs/core/ufs-mcq.c new file +12 -0
+
+static void ufshcd_mcq_config_nr_queues(struct ufs_hba *hba)
+{
+       int i, rem;
+       u32 hbaq_cap, cmp;
+       struct Scsi_Host *host = hba->host;
+
+       hbaq_cap = hba->mcq_capabilities & 0xff;
+
+       rem = hbaq_cap - dev_cmd_queue;
+       /* min() compares variables of same type */
+       hba->nr_queues[HCTX_TYPE_DEFAULT] = min(hbaq_cap -
Quoted 1 line

dev_cmd_queue,

hba->nr_queues[HCTX_TYPE_DEFAULT] = min(rem, num_possible_cpus());

Quoted 11 lines

+                                               num_possible_cpus()); +       rem -= hba->nr_queues[HCTX_TYPE_DEFAULT];
+       if (rem <= 0)
+               goto out;
+       cmp = rem;
+       hba->nr_queues[HCTX_TYPE_POLL] = min(cmp, poll_queues); +       rem -= hba->nr_queues[HCTX_TYPE_POLL];
+       if (rem <= 0)
+               goto out;
+       cmp = rem;
+       hba->nr_queues[HCTX_TYPE_READ] = min(cmp, read_queues);

static read_queues is not initialized.

Quoted 1 line

.....

drivers/ufs/core/ufshcd-priv.h static inline void ufshcd_vops_config_scaling_param(struct ufs_hba *hba, · line 265

Quoted 9 lines
drivers/ufs/core/ufshcd-priv.h +9 -0
+static inline void ufshcd_inc_tp(struct ufs_hw_queue *q)
+{
+       u32 mask = q->max_entries - 1;
+       u32 val;
+
+       q->sq_tp_slot = (q->sq_tp_slot + 1) & mask;
+       val = q->sq_tp_slot * sizeof(struct utp_transfer_req_desc);
+       writel(val, q->mcq_sq_tp);
+}

This function just accesses the submission queue tail pointer. The function name should clearly explain this.

#14 lore raw

Re: [RFC PATCH v3 2/4] ufs: core: mcq: Adds Multi-Circular Queue support

Hello Bean
Thanks for your comments. Responses inline.

Quoted 18 lines · Bean Huo
Quoted 13 lines · Asutosh Das
drivers/ufs/core/ufs-mcq.c new file +12 -0
+
+static void ufshcd_mcq_config_nr_queues(struct ufs_hba *hba)
+{
+       int i, rem;
+       u32 hbaq_cap, cmp;
+       struct Scsi_Host *host = hba->host;
+
+       hbaq_cap = hba->mcq_capabilities & 0xff;
+
+       rem = hbaq_cap - dev_cmd_queue;
+       /* min() compares variables of same type */
+       hba->nr_queues[HCTX_TYPE_DEFAULT] = min(hbaq_cap -
Quoted 1 line

dev_cmd_queue,

  hba->nr_queues[HCTX_TYPE_DEFAULT] = min(rem, num_possible_cpus());

min() compares variables of same type only.

Quoted 12 lines
Quoted 11 lines

+                                               num_possible_cpus()); +       rem -= hba->nr_queues[HCTX_TYPE_DEFAULT];
+       if (rem <= 0)
+               goto out;
+       cmp = rem;
+       hba->nr_queues[HCTX_TYPE_POLL] = min(cmp, poll_queues); +       rem -= hba->nr_queues[HCTX_TYPE_POLL];
+       if (rem <= 0)
+               goto out;
+       cmp = rem;
+       hba->nr_queues[HCTX_TYPE_READ] = min(cmp, read_queues);

Quoted 2 lines

static read_queues is not initialized.

It would be initialized to 0. But I think we should configure it as a module parameter which is missing. Let me add it in the next version.

Quoted 15 lines
Quoted 1 line

.....

drivers/ufs/core/ufshcd-priv.h static inline void ufshcd_vops_config_scaling_param(struct ufs_hba *hba, · line 265

Quoted 9 lines
drivers/ufs/core/ufshcd-priv.h +9 -0
+static inline void ufshcd_inc_tp(struct ufs_hw_queue *q)
+{
+       u32 mask = q->max_entries - 1;
+       u32 val;
+
+       q->sq_tp_slot = (q->sq_tp_slot + 1) & mask;
+       val = q->sq_tp_slot * sizeof(struct utp_transfer_req_desc);
+       writel(val, q->mcq_sq_tp);
+}

This function just accesses the submission queue tail pointer. The function name should clearly explain this.

The reason for this naming was that only SQ's tail pointer can be updated by SW. I can add a _sq_ to this function in the next version. Plmk if you prefer a better name.

Quoted 3 lines
#15 lore raw

Re: [RFC PATCH v3 2/4] ufs: core: mcq: Adds Multi-Circular Queue support

In reply to Bean Huo
Quoted 1 line · Bean Huo

static read_queues is not initialized.

Hi Bean,

The Linux kernel coding style guide does not allow to explicitly initialize static variables to zero. The C standard requires that static variables that are not initialized explicitly are initialized to zero. From https://en.wikipedia.org/wiki/.bss: "In C, statically allocated objects without an explicit initializer are initialized to zero (for arithmetic types) or a null pointer (for pointer types)."

Thanks,

Bart.

#16 lore raw

Re: [RFC PATCH v3 2/4] ufs: core: mcq: Adds Multi-Circular Queue support

In reply to Asutosh Das
Quoted 3 lines · Asutosh Das

In v4.0, UFSHCI specification has added Multi-Circular Queue support to enhance performance. It is backward compatible to the legacy Single Doorbell mode.

The above may be confusing for someone who is not familiar with the UFSHCI 4.0 specification. Consider changing the above text into the following: "The UFSHCI 4.0 register interface defines two modes: (1) a legacy mode that is compatible with UFSHCI 3.0 and (2) an MCQ mode in which multiple submission and completion queues are supported. A capability register indicates whether (1) and/or (2) are supported."

Quoted 12 lines
drivers/ufs/core/ufs-mcq.c new file +12 -0
+static int rw_queue_count_set(const char *val, const struct kernel_param *kp)
+{
+	unsigned int n;
+	int ret;
+
+	ret = kstrtouint(val, 10, &n);
+	if (ret)
+		return ret;
+	if (n > num_possible_cpus())
+		return -EINVAL;
+	return param_set_uint(val, kp);
+}

drivers/ufs/core/ufs-mcq.c line 54

Quoted 10 lines
drivers/ufs/core/ufs-mcq.c new file +10 -0
+
+static const struct kernel_param_ops rw_queue_count_ops = {
+	.set = rw_queue_count_set,
+	.get = param_get_uint,
+};
+
+static unsigned int rw_queues = 8;
+module_param_cb(rw_queues, &rw_queue_count_ops, &rw_queues, 0644);
+MODULE_PARM_DESC(rw_queues,
+		 "Number of interrupt driven I/O queues used for rw. Default value is 8");

rw_queues should be >= 1 and <= 32. However, rw_queue_count_set() allows to set this variable to zero and also to values above 32. I think this needs to be improved.

drivers/ufs/core/ufs-mcq.c line 65

Quoted 10 lines
drivers/ufs/core/ufs-mcq.c new file +10 -0
+/* resources */
+static const struct ufshcd_res_info_t ufshcd_res_info[RES_MAX] = {
+	{"ufs_mem", NULL, NULL},
+	{"mcq", NULL, NULL},
+	{"mcq_sqd", NULL, NULL},
+	{"mcq_sqis", NULL, NULL},
+	{"mcq_cqd", NULL, NULL},
+	{"mcq_cqis", NULL, NULL},
+	{"mcq_vs", NULL, NULL},
+};

Please make the above easier to verify by using designated initializers ([RES_MEM] = { .name = "ufs_mem" }, ...).

drivers/ufs/core/ufs-mcq.c line 72

Quoted 7 lines
drivers/ufs/core/ufs-mcq.c new file +7 -0
+
+/**
+ * ufshcd_mcq_decide_queue_depth - decide the queue depth
+ * @hba - per adapter instance
+ *
+ * MAC - Max. Active Command of the Host Controller (HC)
+ * HC wouldn't send more than this commands to the device.

wouldn't -> won't

drivers/ufs/core/ufs-mcq.c line 93

Quoted 5 lines
drivers/ufs/core/ufs-mcq.c new file +5 -0
+	/*  MAC is a 0 based value. */
+	mac += 1;
+	qd = min_t(u32, mac, hba->dev_info.bqueuedepth);
+	if (!qd)
+		qd = mac;

Please use min_not_zero instead of open-coding it.

drivers/ufs/core/ufs-mcq.c line 106

Quoted 5 lines
drivers/ufs/core/ufs-mcq.c new file +5 -0
+}
+
+/**
+ * ufshcd_mcq_config_mac - Set the #Max Activ Cmds.
+ * @hba - per adpater instance

adpater -> adapter

Quoted 3 lines

+ * @max_active_cmds - maximum # of active commands to the device at any time.
+ *
+ * The controller wouldn't send more than the max_active_cmds to the device at

wouldn't -> won't

Quoted 1 line

+/* ROP - Runtime and operation registers configuration */

The above comment seems misleading to me. In the UFSHCI 4.0 specification the operation and runtime registers start with IS - Interrupt Status while the definitions below are for the MCQ submission and completion queue registers.

drivers/ufs/core/ufs-mcq.c line 167

Quoted 2 lines
drivers/ufs/core/ufs-mcq.c new file +2 -0
+#define MCQ_CFG_n(r, i) \
+	((r) + MCQ_QCFG_SIZE * (i))

Please add a clear comment above this macro that explains its role and also the meaning of its arguments. I assume that 'i' is the queue index?

drivers/ufs/core/ufs-mcq.c line 169

Quoted 2 lines
drivers/ufs/core/ufs-mcq.c new file +2 -0
+#define MCQ_ROP_OFFSET_n(p, i) \
+	(hba->mcq_rop[(p)].offset + hba->mcq_rop[(p)].stride * (i))

Also for this macro, please explain its role. I assume that MCQ_CFG_n() is used to calculate offsets of registers defined in the UFSHCI 4.0 specification and that MCQ_ROP_OFFSET_n() is used to calculate offsets in the data structure with submission and completion queues?

drivers/ufs/core/ufs-mcq.c line 196

Quoted 1 line
drivers/ufs/core/ufs-mcq.c new file +1 -0
+		/* Sbmission Qqueue Doorbell Address Offset */

Sbmission -> Submission
Qqueue -> Queue

drivers/ufs/core/ufs-mcq.c line 216

Quoted 1 line
drivers/ufs/core/ufs-mcq.c new file +1 -0
+		/* Save the base addresses for quicker access */

base -> head and tail?

drivers/ufs/core/ufs-mcq.c line 234

Quoted 6 lines
drivers/ufs/core/ufs-mcq.c new file +6 -0
+		/*
+		 * Submission Qeueue Enable|Size|Completion Queue ID to
+		 * Submission Queue Attribute
+		 */
+		ufsmcq_writel(hba, (1 << 31) | qsize | (i << 16),
+			      MCQ_CFG_n(REG_SQATTR, i));

Qeueue -> Queue

drivers/ufs/core/ufs-mcq.c line 321

Quoted 9 lines
drivers/ufs/core/ufs-mcq.c new file +9 -0
+static void ufshcd_mcq_config_nr_queues(struct ufs_hba *hba)
+{
+	int i, rem;
+	u32 hbaq_cap, cmp;
+	struct Scsi_Host *host = hba->host;
+
+	hbaq_cap = hba->mcq_capabilities & 0xff;
+
+	rem = hbaq_cap - dev_cmd_queue;

Please handle the case dev_cmd_queue > hbaq_cap (rem < 0) explicitly.

drivers/ufs/core/ufs-mcq.c line 322

Quoted 1 line
drivers/ufs/core/ufs-mcq.c new file +1 -0
+	/* min() compares variables of same type */

Please remove this comment. Every kernel developer should know this.

Quoted 8 lines
drivers/ufs/core/ufs-mcq.c new file +8 -0
+	hba->nr_queues[HCTX_TYPE_DEFAULT] = min(hbaq_cap - dev_cmd_queue,
+						num_possible_cpus());
+	rem -= hba->nr_queues[HCTX_TYPE_DEFAULT];
+	if (rem <= 0)
+		goto out;
+	cmp = rem;
+	hba->nr_queues[HCTX_TYPE_POLL] = min(cmp, poll_queues);
+	rem -= hba->nr_queues[HCTX_TYPE_POLL];

drivers/ufs/core/ufs-mcq.c line 340

Quoted 10 lines
drivers/ufs/core/ufs-mcq.c new file +10 -0
+	if (rem <= 0)
+		goto out;
+	cmp = rem;
+	hba->nr_queues[HCTX_TYPE_READ] = min(cmp, read_queues);
+
+out:
+	for (i = 0; i < HCTX_MAX_TYPES; i++)
+		host->nr_hw_queues += hba->nr_queues[i];
+
+	hba->nr_hw_queues = host->nr_hw_queues + dev_cmd_queue;

I think there are code paths in this function that can result in host->nr_hw_queues becoming larger than the number of submission queues supported by the controller. Please fix.

drivers/ufs/core/ufs-mcq.c line 415

Quoted 9 lines
drivers/ufs/core/ufs-mcq.c new file +9 -0
+static int ufshcd_mcq_get_tag(struct ufs_hba *hba,
+				     struct ufs_hw_queue *hwq,
+				     struct cq_entry *cqe)
+{
+	dma_addr_t dma_addr;
+
+	/* Bits 31:7 UCD base address, 6:5 are reserved, 4:0 is SQ ID */
+	dma_addr = ((u64)le32_to_cpu(cqe->command_desc_base_addr_hi) << 32) |
+		   (le32_to_cpu(cqe->command_desc_base_addr_lo) & CQE_UCD_BA);

The above code would become faster if command_desc_base_addr_hi and command_desc_base_addr_lo would be combined into a single __le64 member variable. Has this been considered?

drivers/ufs/core/ufshcd-priv.h static inline void ufshcd_vops_config_scaling_param(struct ufs_hba *hba, · line 265

Quoted 9 lines
drivers/ufs/core/ufshcd-priv.h +9 -0
+static inline void ufshcd_inc_tp(struct ufs_hw_queue *q)
+{
+	u32 mask = q->max_entries - 1;
+	u32 val;
+
+	q->sq_tp_slot = (q->sq_tp_slot + 1) & mask;
+	val = q->sq_tp_slot * sizeof(struct utp_transfer_req_desc);
+	writel(val, q->mcq_sq_tp);
+}

Please consider changing "tp" into "tail" or "tail_pointer" and "hp" into "head" or "head_pointer" to improve code readability.

Quoted 1 line

+EXPORT_SYMBOL_GPL(ufshcd_transfer_req_compl);

Why has this function been exported?

include/ufs/ufshcd.h struct ufs_hba_monitor { · line 746

Quoted 10 lines
include/ufs/ufshcd.h +10 -0
+enum ufshcd_res {
+	RES_MEM,
+	RES_MCQ,
+	RES_MCQ_SQD,
+	RES_MCQ_SQIS,
+	RES_MCQ_CQD,
+	RES_MCQ_CQIS,
+	RES_MCQ_VS,
+	RES_MAX,
+};

Please add a comment that explains the SQD, SQIS etc. acronyms.

include/ufs/ufshcd.h struct ufs_hba_monitor { · line 753

Quoted 6 lines
include/ufs/ufshcd.h +6 -0
+/* MCQ Runtime Operation Pointer info structure */
+struct ufshcd_mcq_rop_info_t {
+	unsigned long offset;
+	unsigned long stride;
+	void __iomem *base;
+};

Please document the role of the members of this data structure.

Thanks,

Bart.

#17 lore raw

RFC PATCH v3 3/4 ufs: core: Add Event Specific Interrupt configuration vendor specific ops

In reply to Asutosh Das

From: Can Guo <[email protected]>

As Event Specific Interrupt message format is not defined in UFSHCI JEDEC specs, and the ESI handling highly depends on how the format is designed, hence add a vendor specific ops such that SoC vendors can configure their own ESI handlers. If ESI vops is not provided or returning error, go with the legacy (central) interrupt way.
Added APIs to enable ESI and config ESI base addresses.

---
 drivers/ufs/core/ufs-mcq.c     | 14 ++++++++++++++
 drivers/ufs/core/ufshcd-priv.h |  8 ++++++++
 drivers/ufs/core/ufshcd.c      |  6 ++++++
 include/ufs/ufshcd.h           |  5 +++++
 4 files changed, 33 insertions(+)
drivers/ufs/core/ufs-mcq.c +14 -0
index e23557b..7f95754 100644
@@ -457,3 +457,17 @@ unsigned long ufshcd_mcq_poll_cqe_lock(struct ufs_hba *hba,
 
 	return completed_reqs;
 }
+
+void ufshcd_mcq_enable_esi(struct ufs_hba *hba)
+{
+	ufshcd_writel(hba, ufshcd_readl(hba, REG_UFS_MEM_CFG) | 0x2,
+		      REG_UFS_MEM_CFG);
+}
+EXPORT_SYMBOL_GPL(ufshcd_mcq_enable_esi);
+
+void ufshcd_mcq_config_esi(struct ufs_hba *hba, struct msi_msg *msg)
+{
+	ufshcd_writel(hba, msg->address_lo, REG_UFS_ESILBA);
+	ufshcd_writel(hba, msg->address_hi, REG_UFS_ESIUBA);
+}
+EXPORT_SYMBOL_GPL(ufshcd_mcq_config_esi);
drivers/ufs/core/ufshcd-priv.h +8 -0
index 4e9e2755..b709695 100644
@@ -254,6 +254,14 @@ static inline int ufshcd_mcq_vops_get_hba_mac(struct ufs_hba *hba)
 	return -EOPNOTSUPP;
 }
 
+static inline int ufshcd_mcq_vops_config_esi(struct ufs_hba *hba)
+{
+	if (hba->vops && hba->vops->config_esi)
+		return hba->vops->config_esi(hba);
+
+	return -EOPNOTSUPP;
+}
+
 static inline void ufshcd_inc_tp(struct ufs_hw_queue *q)
 {
 	u32 mask = q->max_entries - 1;
drivers/ufs/core/ufshcd.c +6 -0
index ec40d46..f5bb0eb 100644
@@ -8306,6 +8306,12 @@ static int ufshcd_config_mcq(struct ufs_hba *hba)
 		goto err;
 	}
 
+	ret = ufshcd_mcq_vops_config_esi(hba);
+	if (ret) {
+		dev_info(hba->dev, "MCQ ESI not available %d\n", ret);
+		ret = 0;
+	}
+
 	ufshcd_enable_intr(hba, UFSHCD_ENABLE_MCQ_INTRS);
 	ufshcd_mcq_make_queues_operational(hba);
 	ufshcd_mcq_config_mac(hba, hba->nutrs);
include/ufs/ufshcd.h +5 -0
index 8c10b09..71bab73 100644
@@ -16,6 +16,7 @@
 #include <linux/blk-crypto-profile.h>
 #include <linux/blk-mq.h>
 #include <linux/devfreq.h>
+#include <linux/msi.h>
 #include <linux/pm_runtime.h>
 #include <scsi/scsi_device.h>
 #include <ufs/unipro.h>
@@ -297,6 +298,7 @@ struct ufs_pwr_mode_info {
  * @get_outstanding_cqs: called to get outstanding completion queues
  * @config_queues: called to config Runtime Operation Pointers
  * @get_hba_mac: called to get vendor specific mac value
+ * @config_esi: called to config Event Specific Interrupt
  */
 struct ufs_hba_variant_ops {
 	const char *name;
@@ -339,6 +341,7 @@ struct ufs_hba_variant_ops {
 				       unsigned long *ocqs);
 	int	(*config_queues)(struct ufs_hba *hba);
 	int	(*get_hba_mac)(struct ufs_hba *hba);
+	int	(*config_esi)(struct ufs_hba *hba);
 };
 
 /* clock gating state  */
@@ -1230,6 +1233,8 @@ extern int ufshcd_dme_get_attr(struct ufs_hba *hba, u32 attr_sel,
 extern int ufshcd_config_pwr_mode(struct ufs_hba *hba,
 			struct ufs_pa_layer_attr *desired_pwr_mode);
 extern int ufshcd_uic_change_pwr_mode(struct ufs_hba *hba, u8 mode);
+extern void ufshcd_mcq_enable_esi(struct ufs_hba *hba);
+extern void ufshcd_mcq_config_esi(struct ufs_hba *hba, struct msi_msg *msg);
 
 /* UIC command interfaces for DME primitives */
 #define DME_LOCAL	0
Signature · 1 line
2.7.4
#18 lore raw

RFC PATCH v3 4/4 ufs: host: qcom: Add MCQ support

In reply to Asutosh Das

From: Can Guo <[email protected]>

Use ESI and define Runtime and Operation definitions for MCQ.

---
 drivers/ufs/core/ufs-mcq.c     |   1 +
 drivers/ufs/core/ufshcd-priv.h |   2 -
 drivers/ufs/host/ufs-qcom.c    | 134 +++++++++++++++++++++++++++++++++++++++++
 drivers/ufs/host/ufs-qcom.h    |  13 ++++
 include/ufs/ufshcd.h           |   2 +
 5 files changed, 150 insertions(+), 2 deletions(-)
drivers/ufs/core/ufs-mcq.c +1 -0
index 7f95754..24370ae 100644
@@ -445,6 +445,7 @@ unsigned long ufshcd_mcq_poll_cqe_nolock(struct ufs_hba *hba,
 
 	return completed_reqs;
 }
+EXPORT_SYMBOL_GPL(ufshcd_mcq_poll_cqe_nolock);
 
 unsigned long ufshcd_mcq_poll_cqe_lock(struct ufs_hba *hba,
 				       struct ufs_hw_queue *hwq)
drivers/ufs/core/ufshcd-priv.h +0 -2
index b709695..a645f8a 100644
@@ -56,8 +56,6 @@ int ufshcd_mcq_memory_alloc(struct ufs_hba *hba);
 void ufshcd_mcq_config_mac(struct ufs_hba *hba, u32 set);
 u32 ufshcd_mcq_read_cqis(struct ufs_hba *hba, int i);
 void ufshcd_mcq_write_cqis(struct ufs_hba *hba, u32 val, int i);
-unsigned long ufshcd_mcq_poll_cqe_nolock(struct ufs_hba *hba,
-					 struct ufs_hw_queue *hwq);
 unsigned long ufshcd_mcq_poll_cqe_lock(struct ufs_hba *hba,
 				       struct ufs_hw_queue *hwq);
 void ufshcd_mcq_make_queues_operational(struct ufs_hba *hba);
drivers/ufs/host/ufs-qcom.c +134 -0

152 lines not on this page · raw message

drivers/ufs/host/ufs-qcom.h +13 -0
index 44466a3..859203a 100644
@@ -53,6 +53,8 @@ enum {
 	 * added in HW Version 3.0.0
 	 */
 	UFS_AH8_CFG				= 0xFC,
+
+	REG_UFS_CFG3				= 0x271C,
 };
 
 /* QCOM UFS host controller vendor specific debug registers */
@@ -72,6 +74,15 @@ enum {
 	UFS_UFS_DBG_RD_EDTL_RAM			= 0x1900,
 };
 
+/* QCOM UFS MCQ vendor specific registers */
+enum {
+	UFS_MEM_MCQ_VS_CFG0				= 0x0,
+	UFS_MEM_SQIS_VS					= 0x4,
+	UFS_MEM_CQIS_VS					= 0x8,
+	UFS_MEM_IAGIS_VS				= 0xC,
+	UFS_MEM_IAGQTYPE_VS				= 0x10,
+};
+
 #define UFS_CNTLR_2_x_x_VEN_REGS_OFFSET(x)	(0x000 + x)
 #define UFS_CNTLR_3_x_x_VEN_REGS_OFFSET(x)	(0x400 + x)
 
@@ -221,6 +232,8 @@ struct ufs_qcom_host {
 	struct reset_controller_dev rcdev;
 
 	struct gpio_desc *device_reset;
+
+	int esi_base;
 };
 
 static inline u32
include/ufs/ufshcd.h +2 -0
index 71bab73..5805128 100644
@@ -1235,6 +1235,8 @@ extern int ufshcd_config_pwr_mode(struct ufs_hba *hba,
 extern int ufshcd_uic_change_pwr_mode(struct ufs_hba *hba, u8 mode);
 extern void ufshcd_mcq_enable_esi(struct ufs_hba *hba);
 extern void ufshcd_mcq_config_esi(struct ufs_hba *hba, struct msi_msg *msg);
+extern unsigned long ufshcd_mcq_poll_cqe_nolock(struct ufs_hba *hba,
+					 struct ufs_hw_queue *hwq);
 
 /* UIC command interfaces for DME primitives */
 #define DME_LOCAL	0
Signature · 1 line
2.7.4
#19 lore raw

Re: [RFC PATCH v3 4/4] ufs: host: qcom: Add MCQ support

drivers/ufs/host/ufs-qcom.c static void ufs_qcom_config_scaling_param(struct ufs_hba *hba, · line 1555

Quoted 4 lines · Asutosh Das
drivers/ufs/host/ufs-qcom.c +4 -0
+static int ufs_qcom_get_hba_mac(struct ufs_hba *hba)
+{
+	return MAX_SUPP_MAC;
+}

The above implementation is not sufficient to convince me that the get_hba_mac vop is useful. Please remove this vop and use MAX_SUPP_MAC directly where ufshcd_mcq_vops_get_hba_mac() is called.

Thanks,

Bart.

#20 lore raw

Re: [RFC PATCH v3 4/4] ufs: host: qcom: Add MCQ support

Hello Bart,
Thanks for taking a look.

Quoted 8 lines · Bart Van Assche

drivers/ufs/host/ufs-qcom.c static void ufs_qcom_config_scaling_param(struct ufs_hba *hba, · line 1555

Quoted 4 lines · Asutosh Das
drivers/ufs/host/ufs-qcom.c +4 -0
+static int ufs_qcom_get_hba_mac(struct ufs_hba *hba)
+{
+    return MAX_SUPP_MAC;
+}

The above implementation is not sufficient to convince me that the get_hba_mac vop is useful. Please remove this vop and use MAX_SUPP_MAC directly where ufshcd_mcq_vops_get_hba_mac() is called.

The default value of MAC is 32 but it can have a max value of 256.
Here it has a value of 64.
So I think this vops would be helpful for other HC vendors as well who may have a different value than the default.

Quoted 3 lines

Thanks,

Bart.

-asd