[PATCH v4 0/3] Simplify and optimize the UFS driver

[PATCH v4 0/3] Simplify and optimize the UFS driver

[Bart Van Assche]

Hello Martin,

This patch series that simplifies and optimizes the UFS driver. Please consider
this patch series for kernel v5.5.

Thanks,

Bart.

Changes compared to v3:
- Left out "scsi" from the name of the functions that suspend and resume
  command processing.

Changes compared to v2:
- Use a separate tag set for TMF tags.

Changes compared to v1:
- Use the block layer tag infrastructure for managing TMF tags.

Bart Van Assche (3):
  ufs: Avoid busy-waiting by eliminating tag conflicts
  ufs: Use blk_{get,put}_request() to allocate and free TMFs
  ufs: Simplify the clock scaling mechanism implementation

 drivers/scsi/ufs/ufshcd.c | 384 +++++++++++++++++---------------------
 drivers/scsi/ufs/ufshcd.h |  21 +--
 2 files changed, 173 insertions(+), 232 deletions(-)

[PATCH v4 1/3] ufs: Avoid busy-waiting by eliminating tag conflicts

[Bart Van Assche]

Instead of tracking which tags are in use in the ufs_hba.lrb_in_use
bitmask, rely on the block layer tag allocation mechanism. This patch
removes the following busy-waiting loop if ufshcd_issue_devman_upiu_cmd()
and the block layer accidentally allocate the same tag for a SCSI request:
* ufshcd_queuecommand() returns SCSI_MLQUEUE_HOST_BUSY.
* The SCSI core requeues the SCSI command.

Tested-by: Bean Huo <beanhuo@micron.com>
Reviewed-by: Avri Altman <avri.altman@wdc.com>
Cc: Stanley Chu <stanley.chu@mediatek.com>
Cc: Avri Altman <avri.altman@wdc.com>
Cc: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
---
 drivers/scsi/ufs/ufshcd.c | 121 +++++++++++++++-----------------------
 drivers/scsi/ufs/ufshcd.h |   6 +-
 2 files changed, 50 insertions(+), 77 deletions(-)

diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index 9cbe3b45cf1c..93c920bf77d2 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -497,8 +497,8 @@ static void ufshcd_print_tmrs(struct ufs_hba *hba, unsigned long bitmap)
 static void ufshcd_print_host_state(struct ufs_hba *hba)
 {
 	dev_err(hba->dev, "UFS Host state=%d\n", hba->ufshcd_state);
-	dev_err(hba->dev, "lrb in use=0x%lx, outstanding reqs=0x%lx tasks=0x%lx\n",
-		hba->lrb_in_use, hba->outstanding_reqs, hba->outstanding_tasks);
+	dev_err(hba->dev, "outstanding reqs=0x%lx tasks=0x%lx\n",
+		hba->outstanding_reqs, hba->outstanding_tasks);
 	dev_err(hba->dev, "saved_err=0x%x, saved_uic_err=0x%x\n",
 		hba->saved_err, hba->saved_uic_err);
 	dev_err(hba->dev, "Device power mode=%d, UIC link state=%d\n",
@@ -1596,6 +1596,25 @@ int ufshcd_hold(struct ufs_hba *hba, bool async)
 }
 EXPORT_SYMBOL_GPL(ufshcd_hold);
 
+static bool ufshcd_is_busy(struct request *req, void *priv, bool reserved)
+{
+	int *busy = priv;
+
+	WARN_ON_ONCE(reserved);
+	(*busy)++;
+	return false;
+}
+
+/* Whether or not any tag is in use by a request that is in progress. */
+static bool ufshcd_any_tag_in_use(struct ufs_hba *hba)
+{
+	struct request_queue *q = hba->cmd_queue;
+	int busy = 0;
+
+	blk_mq_tagset_busy_iter(q->tag_set, ufshcd_is_busy, &busy);
+	return busy;
+}
+
 static void ufshcd_gate_work(struct work_struct *work)
 {
 	struct ufs_hba *hba = container_of(work, struct ufs_hba,
@@ -1619,7 +1638,7 @@ static void ufshcd_gate_work(struct work_struct *work)
 
 	if (hba->clk_gating.active_reqs
 		|| hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL
-		|| hba->lrb_in_use || hba->outstanding_tasks
+		|| ufshcd_any_tag_in_use(hba) || hba->outstanding_tasks
 		|| hba->active_uic_cmd || hba->uic_async_done)
 		goto rel_lock;
 
@@ -1673,7 +1692,7 @@ static void __ufshcd_release(struct ufs_hba *hba)
 
 	if (hba->clk_gating.active_reqs || hba->clk_gating.is_suspended
 		|| hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL
-		|| hba->lrb_in_use || hba->outstanding_tasks
+		|| ufshcd_any_tag_in_use(hba) || hba->outstanding_tasks
 		|| hba->active_uic_cmd || hba->uic_async_done
 		|| ufshcd_eh_in_progress(hba))
 		return;
@@ -2443,22 +2462,9 @@ static int ufshcd_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *cmd)
 
 	hba->req_abort_count = 0;
 
-	/* acquire the tag to make sure device cmds don't use it */
-	if (test_and_set_bit_lock(tag, &hba->lrb_in_use)) {
-		/*
-		 * Dev manage command in progress, requeue the command.
-		 * Requeuing the command helps in cases where the request *may*
-		 * find different tag instead of waiting for dev manage command
-		 * completion.
-		 */
-		err = SCSI_MLQUEUE_HOST_BUSY;
-		goto out;
-	}
-
 	err = ufshcd_hold(hba, true);
 	if (err) {
 		err = SCSI_MLQUEUE_HOST_BUSY;
-		clear_bit_unlock(tag, &hba->lrb_in_use);
 		goto out;
 	}
 	WARN_ON(hba->clk_gating.state != CLKS_ON);
@@ -2479,7 +2485,6 @@ static int ufshcd_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *cmd)
 	err = ufshcd_map_sg(hba, lrbp);
 	if (err) {
 		lrbp->cmd = NULL;
-		clear_bit_unlock(tag, &hba->lrb_in_use);
 		goto out;
 	}
 	/* Make sure descriptors are ready before ringing the doorbell */
@@ -2626,44 +2631,6 @@ static int ufshcd_wait_for_dev_cmd(struct ufs_hba *hba,
 	return err;
 }
 
-/**
- * ufshcd_get_dev_cmd_tag - Get device management command tag
- * @hba: per-adapter instance
- * @tag_out: pointer to variable with available slot value
- *
- * Get a free slot and lock it until device management command
- * completes.
- *
- * Returns false if free slot is unavailable for locking, else
- * return true with tag value in @tag.
- */
-static bool ufshcd_get_dev_cmd_tag(struct ufs_hba *hba, int *tag_out)
-{
-	int tag;
-	bool ret = false;
-	unsigned long tmp;
-
-	if (!tag_out)
-		goto out;
-
-	do {
-		tmp = ~hba->lrb_in_use;
-		tag = find_last_bit(&tmp, hba->nutrs);
-		if (tag >= hba->nutrs)
-			goto out;
-	} while (test_and_set_bit_lock(tag, &hba->lrb_in_use));
-
-	*tag_out = tag;
-	ret = true;
-out:
-	return ret;
-}
-
-static inline void ufshcd_put_dev_cmd_tag(struct ufs_hba *hba, int tag)
-{
-	clear_bit_unlock(tag, &hba->lrb_in_use);
-}
-
 /**
  * ufshcd_exec_dev_cmd - API for sending device management requests
  * @hba: UFS hba
@@ -2676,6 +2643,8 @@ static inline void ufshcd_put_dev_cmd_tag(struct ufs_hba *hba, int tag)
 static int ufshcd_exec_dev_cmd(struct ufs_hba *hba,
 		enum dev_cmd_type cmd_type, int timeout)
 {
+	struct request_queue *q = hba->cmd_queue;
+	struct request *req;
 	struct ufshcd_lrb *lrbp;
 	int err;
 	int tag;
@@ -2689,7 +2658,11 @@ static int ufshcd_exec_dev_cmd(struct ufs_hba *hba,
 	 * Even though we use wait_event() which sleeps indefinitely,
 	 * the maximum wait time is bounded by SCSI request timeout.
 	 */
-	wait_event(hba->dev_cmd.tag_wq, ufshcd_get_dev_cmd_tag(hba, &tag));
+	req = blk_get_request(q, REQ_OP_DRV_OUT, 0);
+	if (IS_ERR(req))
+		return PTR_ERR(req);
+	tag = req->tag;
+	WARN_ON_ONCE(!ufshcd_valid_tag(hba, tag));
 
 	init_completion(&wait);
 	lrbp = &hba->lrb[tag];
@@ -2714,8 +2687,7 @@ static int ufshcd_exec_dev_cmd(struct ufs_hba *hba,
 			err ? "query_complete_err" : "query_complete");
 
 out_put_tag:
-	ufshcd_put_dev_cmd_tag(hba, tag);
-	wake_up(&hba->dev_cmd.tag_wq);
+	blk_put_request(req);
 	up_read(&hba->clk_scaling_lock);
 	return err;
 }
@@ -4834,7 +4806,6 @@ static void __ufshcd_transfer_req_compl(struct ufs_hba *hba,
 			cmd->result = result;
 			/* Mark completed command as NULL in LRB */
 			lrbp->cmd = NULL;
-			clear_bit_unlock(index, &hba->lrb_in_use);
 			/* Do not touch lrbp after scsi done */
 			cmd->scsi_done(cmd);
 			__ufshcd_release(hba);
@@ -4856,9 +4827,6 @@ static void __ufshcd_transfer_req_compl(struct ufs_hba *hba,
 	hba->outstanding_reqs ^= completed_reqs;
 
 	ufshcd_clk_scaling_update_busy(hba);
-
-	/* we might have free'd some tags above */
-	wake_up(&hba->dev_cmd.tag_wq);
 }
 
 /**
@@ -5787,6 +5755,8 @@ static int ufshcd_issue_devman_upiu_cmd(struct ufs_hba *hba,
 					enum dev_cmd_type cmd_type,
 					enum query_opcode desc_op)
 {
+	struct request_queue *q = hba->cmd_queue;
+	struct request *req;
 	struct ufshcd_lrb *lrbp;
 	int err = 0;
 	int tag;
@@ -5796,7 +5766,11 @@ static int ufshcd_issue_devman_upiu_cmd(struct ufs_hba *hba,
 
 	down_read(&hba->clk_scaling_lock);
 
-	wait_event(hba->dev_cmd.tag_wq, ufshcd_get_dev_cmd_tag(hba, &tag));
+	req = blk_get_request(q, REQ_OP_DRV_OUT, 0);
+	if (IS_ERR(req))
+		return PTR_ERR(req);
+	tag = req->tag;
+	WARN_ON_ONCE(!ufshcd_valid_tag(hba, tag));
 
 	init_completion(&wait);
 	lrbp = &hba->lrb[tag];
@@ -5870,8 +5844,7 @@ static int ufshcd_issue_devman_upiu_cmd(struct ufs_hba *hba,
 		}
 	}
 
-	ufshcd_put_dev_cmd_tag(hba, tag);
-	wake_up(&hba->dev_cmd.tag_wq);
+	blk_put_request(req);
 	up_read(&hba->clk_scaling_lock);
 	return err;
 }
@@ -6166,9 +6139,6 @@ static int ufshcd_abort(struct scsi_cmnd *cmd)
 	hba->lrb[tag].cmd = NULL;
 	spin_unlock_irqrestore(host->host_lock, flags);
 
-	clear_bit_unlock(tag, &hba->lrb_in_use);
-	wake_up(&hba->dev_cmd.tag_wq);
-
 out:
 	if (!err) {
 		err = SUCCESS;
@@ -8172,6 +8142,7 @@ void ufshcd_remove(struct ufs_hba *hba)
 {
 	ufs_bsg_remove(hba);
 	ufs_sysfs_remove_nodes(hba->dev);
+	blk_cleanup_queue(hba->cmd_queue);
 	scsi_remove_host(hba->host);
 	/* disable interrupts */
 	ufshcd_disable_intr(hba, hba->intr_mask);
@@ -8335,9 +8306,6 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)
 
 	init_rwsem(&hba->clk_scaling_lock);
 
-	/* Initialize device management tag acquire wait queue */
-	init_waitqueue_head(&hba->dev_cmd.tag_wq);
-
 	ufshcd_init_clk_gating(hba);
 
 	ufshcd_init_clk_scaling(hba);
@@ -8371,6 +8339,11 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)
 		goto exit_gating;
 	}
 
+	err = -ENOMEM;
+	hba->cmd_queue = blk_mq_init_queue(&hba->host->tag_set);
+	if (!hba->cmd_queue)
+		goto out_remove_scsi_host;
+
 	/* Reset the attached device */
 	ufshcd_vops_device_reset(hba);
 
@@ -8380,7 +8353,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)
 		dev_err(hba->dev, "Host controller enable failed\n");
 		ufshcd_print_host_regs(hba);
 		ufshcd_print_host_state(hba);
-		goto out_remove_scsi_host;
+		goto free_cmd_queue;
 	}
 
 	/*
@@ -8417,6 +8390,8 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)
 
 	return 0;
 
+free_cmd_queue:
+	blk_cleanup_queue(hba->cmd_queue);
 out_remove_scsi_host:
 	scsi_remove_host(hba->host);
 exit_gating:
diff --git a/drivers/scsi/ufs/ufshcd.h b/drivers/scsi/ufs/ufshcd.h
index e0fe247c719e..6d9521892a84 100644
--- a/drivers/scsi/ufs/ufshcd.h
+++ b/drivers/scsi/ufs/ufshcd.h
@@ -212,13 +212,11 @@ struct ufs_query {
  * @type: device management command type - Query, NOP OUT
  * @lock: lock to allow one command at a time
  * @complete: internal commands completion
- * @tag_wq: wait queue until free command slot is available
  */
 struct ufs_dev_cmd {
 	enum dev_cmd_type type;
 	struct mutex lock;
 	struct completion *complete;
-	wait_queue_head_t tag_wq;
 	struct ufs_query query;
 };
 
@@ -483,7 +481,7 @@ struct ufs_stats {
  * @host: Scsi_Host instance of the driver
  * @dev: device handle
  * @lrb: local reference block
- * @lrb_in_use: lrb in use
+ * @cmd_queue: Used to allocate command tags from hba->host->tag_set.
  * @outstanding_tasks: Bits representing outstanding task requests
  * @outstanding_reqs: Bits representing outstanding transfer requests
  * @capabilities: UFS Controller Capabilities
@@ -541,6 +539,7 @@ struct ufs_hba {
 
 	struct Scsi_Host *host;
 	struct device *dev;
+	struct request_queue *cmd_queue;
 	/*
 	 * This field is to keep a reference to "scsi_device" corresponding to
 	 * "UFS device" W-LU.
@@ -561,7 +560,6 @@ struct ufs_hba {
 	u32 ahit;
 
 	struct ufshcd_lrb *lrb;
-	unsigned long lrb_in_use;
 
 	unsigned long outstanding_tasks;
 	unsigned long outstanding_reqs;
-- 
2.24.0.rc1.363.gb1bccd3e3d-goog

RE: [EXT] [PATCH v4 1/3] ufs: Avoid busy-waiting by eliminating tag conflicts

[Bean Huo (beanhuo)]

Hi, Bart

> +	req = blk_get_request(q, REQ_OP_DRV_OUT, 0);
> +	if (IS_ERR(req))
> +		return PTR_ERR(req);
> +	tag = req->tag;
> +	WARN_ON_ONCE(!ufshcd_valid_tag(hba, tag));
....
> &tag));
> +	req = blk_get_request(q, REQ_OP_DRV_OUT, 0);
> +	if (IS_ERR(req))
> +		return PTR_ERR(req);
> +	tag = req->tag;
> +	WARN_ON_ONCE(!ufshcd_valid_tag(hba, tag));
> 
......
> 
> +	err = -ENOMEM;
> +	hba->cmd_queue = blk_mq_init_queue(&hba->host->tag_set);
> +	if (!hba->cmd_queue)

It is possible  to return ERR_PTR(-ENOMEM)  in  blk_mq_init_queue(),,
Here just checkup if not NULL is not safe enough.

//Bean

Re: [EXT] [PATCH v4 1/3] ufs: Avoid busy-waiting by eliminating tag conflicts

[Bart Van Assche]

On 11/12/19 4:25 AM, Bean Huo (beanhuo) wrote:
>> +	err = -ENOMEM;
>> +	hba->cmd_queue = blk_mq_init_queue(&hba->host->tag_set);
>> +	if (!hba->cmd_queue)
> 
> It is possible  to return ERR_PTR(-ENOMEM)  in  blk_mq_init_queue(),,
> Here just checkup if not NULL is not safe enough.

Agreed, this should be changed into an IS_ERR() check.

Bart.

[PATCH v4 2/3] ufs: Use blk_{get,put}_request() to allocate and free TMFs

[Bart Van Assche]

Manage TMF tags with blk_{get,put}_request() instead of
ufshcd_get_tm_free_slot() / ufshcd_put_tm_slot(). Store a per-request
completion pointer in request.end_io_data instead of using a waitqueue
to report TMF completion.

Tested-by: Bean Huo <beanhuo@micron.com>
Reviewed-by: Avri Altman <avri.altman@wdc.com>
Cc: Stanley Chu <stanley.chu@mediatek.com>
Cc: Avri Altman <avri.altman@wdc.com>
Cc: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
---
 drivers/scsi/ufs/ufshcd.c | 122 +++++++++++++++++++++++---------------
 drivers/scsi/ufs/ufshcd.h |  12 ++--
 2 files changed, 77 insertions(+), 57 deletions(-)

diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index 93c920bf77d2..00c567bbbb64 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -645,40 +645,6 @@ static inline int ufshcd_get_tr_ocs(struct ufshcd_lrb *lrbp)
 	return le32_to_cpu(lrbp->utr_descriptor_ptr->header.dword_2) & MASK_OCS;
 }
 
-/**
- * ufshcd_get_tm_free_slot - get a free slot for task management request
- * @hba: per adapter instance
- * @free_slot: pointer to variable with available slot value
- *
- * Get a free tag and lock it until ufshcd_put_tm_slot() is called.
- * Returns 0 if free slot is not available, else return 1 with tag value
- * in @free_slot.
- */
-static bool ufshcd_get_tm_free_slot(struct ufs_hba *hba, int *free_slot)
-{
-	int tag;
-	bool ret = false;
-
-	if (!free_slot)
-		goto out;
-
-	do {
-		tag = find_first_zero_bit(&hba->tm_slots_in_use, hba->nutmrs);
-		if (tag >= hba->nutmrs)
-			goto out;
-	} while (test_and_set_bit_lock(tag, &hba->tm_slots_in_use));
-
-	*free_slot = tag;
-	ret = true;
-out:
-	return ret;
-}
-
-static inline void ufshcd_put_tm_slot(struct ufs_hba *hba, int slot)
-{
-	clear_bit_unlock(slot, &hba->tm_slots_in_use);
-}
-
 /**
  * ufshcd_utrl_clear - Clear a bit in UTRLCLR register
  * @hba: per adapter instance
@@ -5517,17 +5483,38 @@ static void ufshcd_check_errors(struct ufs_hba *hba)
 	 */
 }
 
+struct ctm_info {
+	struct ufs_hba *hba;
+	unsigned long pending;
+};
+
+static bool ufshcd_compl_tm(struct request *req, void *priv, bool reserved)
+{
+	const struct ctm_info *const ci = priv;
+	struct completion *c;
+
+	WARN_ON_ONCE(reserved);
+	if (test_bit(req->tag, &ci->pending))
+		return true;
+	c = req->end_io_data;
+	if (c)
+		complete(c);
+	return true;
+}
+
 /**
  * ufshcd_tmc_handler - handle task management function completion
  * @hba: per adapter instance
  */
 static void ufshcd_tmc_handler(struct ufs_hba *hba)
 {
-	u32 tm_doorbell;
+	struct request_queue *q = hba->tmf_queue;
+	struct ctm_info ci = {
+		.hba	 = hba,
+		.pending = ufshcd_readl(hba, REG_UTP_TASK_REQ_DOOR_BELL),
+	};
 
-	tm_doorbell = ufshcd_readl(hba, REG_UTP_TASK_REQ_DOOR_BELL);
-	hba->tm_condition = tm_doorbell ^ hba->outstanding_tasks;
-	wake_up(&hba->tm_wq);
+	blk_mq_tagset_busy_iter(q->tag_set, ufshcd_compl_tm, &ci);
 }
 
 /**
@@ -5620,7 +5607,10 @@ static int ufshcd_clear_tm_cmd(struct ufs_hba *hba, int tag)
 static int __ufshcd_issue_tm_cmd(struct ufs_hba *hba,
 		struct utp_task_req_desc *treq, u8 tm_function)
 {
+	struct request_queue *q = hba->tmf_queue;
 	struct Scsi_Host *host = hba->host;
+	DECLARE_COMPLETION_ONSTACK(wait);
+	struct request *req;
 	unsigned long flags;
 	int free_slot, task_tag, err;
 
@@ -5629,7 +5619,10 @@ static int __ufshcd_issue_tm_cmd(struct ufs_hba *hba,
 	 * Even though we use wait_event() which sleeps indefinitely,
 	 * the maximum wait time is bounded by %TM_CMD_TIMEOUT.
 	 */
-	wait_event(hba->tm_tag_wq, ufshcd_get_tm_free_slot(hba, &free_slot));
+	req = blk_get_request(q, REQ_OP_DRV_OUT, BLK_MQ_REQ_RESERVED);
+	req->end_io_data = &wait;
+	free_slot = req->tag;
+	WARN_ON_ONCE(free_slot < 0 || free_slot >= hba->nutmrs);
 	ufshcd_hold(hba, false);
 
 	spin_lock_irqsave(host->host_lock, flags);
@@ -5655,10 +5648,14 @@ static int __ufshcd_issue_tm_cmd(struct ufs_hba *hba,
 	ufshcd_add_tm_upiu_trace(hba, task_tag, "tm_send");
 
 	/* wait until the task management command is completed */
-	err = wait_event_timeout(hba->tm_wq,
-			test_bit(free_slot, &hba->tm_condition),
+	err = wait_for_completion_io_timeout(&wait,
 			msecs_to_jiffies(TM_CMD_TIMEOUT));
 	if (!err) {
+		/*
+		 * Make sure that ufshcd_compl_tm() does not trigger a
+		 * use-after-free.
+		 */
+		req->end_io_data = NULL;
 		ufshcd_add_tm_upiu_trace(hba, task_tag, "tm_complete_err");
 		dev_err(hba->dev, "%s: task management cmd 0x%.2x timed-out\n",
 				__func__, tm_function);
@@ -5677,9 +5674,7 @@ static int __ufshcd_issue_tm_cmd(struct ufs_hba *hba,
 	__clear_bit(free_slot, &hba->outstanding_tasks);
 	spin_unlock_irqrestore(hba->host->host_lock, flags);
 
-	clear_bit(free_slot, &hba->tm_condition);
-	ufshcd_put_tm_slot(hba, free_slot);
-	wake_up(&hba->tm_tag_wq);
+	blk_put_request(req);
 
 	ufshcd_release(hba);
 	return err;
@@ -8142,6 +8137,8 @@ void ufshcd_remove(struct ufs_hba *hba)
 {
 	ufs_bsg_remove(hba);
 	ufs_sysfs_remove_nodes(hba->dev);
+	blk_cleanup_queue(hba->tmf_queue);
+	blk_mq_free_tag_set(&hba->tmf_tag_set);
 	blk_cleanup_queue(hba->cmd_queue);
 	scsi_remove_host(hba->host);
 	/* disable interrupts */
@@ -8221,6 +8218,18 @@ int ufshcd_alloc_host(struct device *dev, struct ufs_hba **hba_handle)
 }
 EXPORT_SYMBOL(ufshcd_alloc_host);
 
+/* This function exists because blk_mq_alloc_tag_set() requires this. */
+static blk_status_t ufshcd_queue_tmf(struct blk_mq_hw_ctx *hctx,
+				     const struct blk_mq_queue_data *qd)
+{
+	WARN_ON_ONCE(true);
+	return BLK_STS_NOTSUPP;
+}
+
+static const struct blk_mq_ops ufshcd_tmf_ops = {
+	.queue_rq = ufshcd_queue_tmf,
+};
+
 /**
  * ufshcd_init - Driver initialization routine
  * @hba: per-adapter instance
@@ -8290,10 +8299,6 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)
 
 	hba->max_pwr_info.is_valid = false;
 
-	/* Initailize wait queue for task management */
-	init_waitqueue_head(&hba->tm_wq);
-	init_waitqueue_head(&hba->tm_tag_wq);
-
 	/* Initialize work queues */
 	INIT_WORK(&hba->eh_work, ufshcd_err_handler);
 	INIT_WORK(&hba->eeh_work, ufshcd_exception_event_handler);
@@ -8344,6 +8349,21 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)
 	if (!hba->cmd_queue)
 		goto out_remove_scsi_host;
 
+	hba->tmf_tag_set = (struct blk_mq_tag_set) {
+		.nr_hw_queues	= 1,
+		.queue_depth	= hba->nutmrs,
+		.ops		= &ufshcd_tmf_ops,
+		.flags		= BLK_MQ_F_NO_SCHED,
+	};
+	err = blk_mq_alloc_tag_set(&hba->tmf_tag_set);
+	if (err < 0)
+		goto free_cmd_queue;
+	hba->tmf_queue = blk_mq_init_queue(&hba->tmf_tag_set);
+	if (IS_ERR(hba->tmf_queue)) {
+		err = PTR_ERR(hba->tmf_queue);
+		goto free_tmf_tag_set;
+	}
+
 	/* Reset the attached device */
 	ufshcd_vops_device_reset(hba);
 
@@ -8353,7 +8373,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)
 		dev_err(hba->dev, "Host controller enable failed\n");
 		ufshcd_print_host_regs(hba);
 		ufshcd_print_host_state(hba);
-		goto free_cmd_queue;
+		goto free_tmf_queue;
 	}
 
 	/*
@@ -8390,6 +8410,10 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)
 
 	return 0;
 
+free_tmf_queue:
+	blk_cleanup_queue(hba->tmf_queue);
+free_tmf_tag_set:
+	blk_mq_free_tag_set(&hba->tmf_tag_set);
 free_cmd_queue:
 	blk_cleanup_queue(hba->cmd_queue);
 out_remove_scsi_host:
diff --git a/drivers/scsi/ufs/ufshcd.h b/drivers/scsi/ufs/ufshcd.h
index 6d9521892a84..5865e16f53a6 100644
--- a/drivers/scsi/ufs/ufshcd.h
+++ b/drivers/scsi/ufs/ufshcd.h
@@ -493,11 +493,9 @@ struct ufs_stats {
  * @irq: Irq number of the controller
  * @active_uic_cmd: handle of active UIC command
  * @uic_cmd_mutex: mutex for uic command
- * @tm_wq: wait queue for task management
- * @tm_tag_wq: wait queue for free task management slots
- * @tm_slots_in_use: bit map of task management request slots in use
+ * @tmf_tag_set: TMF tag set.
+ * @tmf_queue: Used to allocate TMF tags.
  * @pwr_done: completion for power mode change
- * @tm_condition: condition variable for task management
  * @ufshcd_state: UFSHCD states
  * @eh_flags: Error handling flags
  * @intr_mask: Interrupt Mask Bits
@@ -641,10 +639,8 @@ struct ufs_hba {
 	/* Device deviations from standard UFS device spec. */
 	unsigned int dev_quirks;
 
-	wait_queue_head_t tm_wq;
-	wait_queue_head_t tm_tag_wq;
-	unsigned long tm_condition;
-	unsigned long tm_slots_in_use;
+	struct blk_mq_tag_set tmf_tag_set;
+	struct request_queue *tmf_queue;
 
 	struct uic_command *active_uic_cmd;
 	struct mutex uic_cmd_mutex;
-- 
2.24.0.rc1.363.gb1bccd3e3d-goog

[PATCH v4 3/3] ufs: Simplify the clock scaling mechanism implementation

[Bart Van Assche]

Scaling the clock is only safe while no commands are in progress. Use
blk_mq_{un,}freeze_queue() to block submission of new commands and to
wait for ongoing commands to complete. Remove "_scsi" from the name of
the functions that block and unblock requests since these functions
now block both SCSI commands and non-SCSI commands.

Tested-by: Bean Huo <beanhuo@micron.com>
Reviewed-by: Avri Altman <avri.altman@wdc.com>
Cc: Stanley Chu <stanley.chu@mediatek.com>
Cc: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
---
 drivers/scsi/ufs/ufshcd.c | 143 +++++++++++++-------------------------
 drivers/scsi/ufs/ufshcd.h |   3 -
 2 files changed, 47 insertions(+), 99 deletions(-)

diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index 00c567bbbb64..65f05ca8d76f 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -290,16 +290,50 @@ static inline void ufshcd_disable_irq(struct ufs_hba *hba)
 	}
 }
 
-static void ufshcd_scsi_unblock_requests(struct ufs_hba *hba)
+static void ufshcd_unblock_requests(struct ufs_hba *hba)
 {
-	if (atomic_dec_and_test(&hba->scsi_block_reqs_cnt))
-		scsi_unblock_requests(hba->host);
+	struct scsi_device *sdev;
+
+	blk_mq_unfreeze_queue(hba->tmf_queue);
+	blk_mq_unfreeze_queue(hba->cmd_queue);
+	shost_for_each_device(sdev, hba->host)
+		blk_mq_unfreeze_queue(sdev->request_queue);
 }
 
-static void ufshcd_scsi_block_requests(struct ufs_hba *hba)
+static int ufshcd_block_requests(struct ufs_hba *hba, unsigned long timeout)
 {
-	if (atomic_inc_return(&hba->scsi_block_reqs_cnt) == 1)
-		scsi_block_requests(hba->host);
+	struct scsi_device *sdev;
+	unsigned long deadline = jiffies + timeout;
+
+	if (timeout == ULONG_MAX) {
+		shost_for_each_device(sdev, hba->host)
+			blk_mq_freeze_queue(sdev->request_queue);
+		blk_mq_freeze_queue(hba->cmd_queue);
+		blk_mq_freeze_queue(hba->tmf_queue);
+		return 0;
+	}
+
+	shost_for_each_device(sdev, hba->host)
+		blk_freeze_queue_start(sdev->request_queue);
+	blk_freeze_queue_start(hba->cmd_queue);
+	blk_freeze_queue_start(hba->tmf_queue);
+	shost_for_each_device(sdev, hba->host) {
+		if (blk_mq_freeze_queue_wait_timeout(sdev->request_queue,
+				max_t(long, 0, deadline - jiffies)) <= 0) {
+			goto err;
+		}
+	}
+	if (blk_mq_freeze_queue_wait_timeout(hba->cmd_queue,
+				max_t(long, 0, deadline - jiffies)) <= 0)
+		goto err;
+	if (blk_mq_freeze_queue_wait_timeout(hba->tmf_queue,
+				max_t(long, 0, deadline - jiffies)) <= 0)
+		goto err;
+	return 0;
+
+err:
+	ufshcd_unblock_requests(hba);
+	return -ETIMEDOUT;
 }
 
 static void ufshcd_add_cmd_upiu_trace(struct ufs_hba *hba, unsigned int tag,
@@ -971,65 +1005,6 @@ static bool ufshcd_is_devfreq_scaling_required(struct ufs_hba *hba,
 	return false;
 }
 
-static int ufshcd_wait_for_doorbell_clr(struct ufs_hba *hba,
-					u64 wait_timeout_us)
-{
-	unsigned long flags;
-	int ret = 0;
-	u32 tm_doorbell;
-	u32 tr_doorbell;
-	bool timeout = false, do_last_check = false;
-	ktime_t start;
-
-	ufshcd_hold(hba, false);
-	spin_lock_irqsave(hba->host->host_lock, flags);
-	/*
-	 * Wait for all the outstanding tasks/transfer requests.
-	 * Verify by checking the doorbell registers are clear.
-	 */
-	start = ktime_get();
-	do {
-		if (hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL) {
-			ret = -EBUSY;
-			goto out;
-		}
-
-		tm_doorbell = ufshcd_readl(hba, REG_UTP_TASK_REQ_DOOR_BELL);
-		tr_doorbell = ufshcd_readl(hba, REG_UTP_TRANSFER_REQ_DOOR_BELL);
-		if (!tm_doorbell && !tr_doorbell) {
-			timeout = false;
-			break;
-		} else if (do_last_check) {
-			break;
-		}
-
-		spin_unlock_irqrestore(hba->host->host_lock, flags);
-		schedule();
-		if (ktime_to_us(ktime_sub(ktime_get(), start)) >
-		    wait_timeout_us) {
-			timeout = true;
-			/*
-			 * We might have scheduled out for long time so make
-			 * sure to check if doorbells are cleared by this time
-			 * or not.
-			 */
-			do_last_check = true;
-		}
-		spin_lock_irqsave(hba->host->host_lock, flags);
-	} while (tm_doorbell || tr_doorbell);
-
-	if (timeout) {
-		dev_err(hba->dev,
-			"%s: timedout waiting for doorbell to clear (tm=0x%x, tr=0x%x)\n",
-			__func__, tm_doorbell, tr_doorbell);
-		ret = -EBUSY;
-	}
-out:
-	spin_unlock_irqrestore(hba->host->host_lock, flags);
-	ufshcd_release(hba);
-	return ret;
-}
-
 /**
  * ufshcd_scale_gear - scale up/down UFS gear
  * @hba: per adapter instance
@@ -1079,27 +1054,16 @@ static int ufshcd_scale_gear(struct ufs_hba *hba, bool scale_up)
 
 static int ufshcd_clock_scaling_prepare(struct ufs_hba *hba)
 {
-	#define DOORBELL_CLR_TOUT_US		(1000 * 1000) /* 1 sec */
-	int ret = 0;
 	/*
 	 * make sure that there are no outstanding requests when
 	 * clock scaling is in progress
 	 */
-	ufshcd_scsi_block_requests(hba);
-	down_write(&hba->clk_scaling_lock);
-	if (ufshcd_wait_for_doorbell_clr(hba, DOORBELL_CLR_TOUT_US)) {
-		ret = -EBUSY;
-		up_write(&hba->clk_scaling_lock);
-		ufshcd_scsi_unblock_requests(hba);
-	}
-
-	return ret;
+	return ufshcd_block_requests(hba, HZ);
 }
 
 static void ufshcd_clock_scaling_unprepare(struct ufs_hba *hba)
 {
-	up_write(&hba->clk_scaling_lock);
-	ufshcd_scsi_unblock_requests(hba);
+	ufshcd_unblock_requests(hba);
 }
 
 /**
@@ -1471,7 +1435,7 @@ static void ufshcd_ungate_work(struct work_struct *work)
 		hba->clk_gating.is_suspended = false;
 	}
 unblock_reqs:
-	ufshcd_scsi_unblock_requests(hba);
+	ufshcd_unblock_requests(hba);
 }
 
 /**
@@ -1528,7 +1492,7 @@ int ufshcd_hold(struct ufs_hba *hba, bool async)
 		 */
 		/* fallthrough */
 	case CLKS_OFF:
-		ufshcd_scsi_block_requests(hba);
+		ufshcd_block_requests(hba, ULONG_MAX);
 		hba->clk_gating.state = REQ_CLKS_ON;
 		trace_ufshcd_clk_gating(dev_name(hba->dev),
 					hba->clk_gating.state);
@@ -2395,9 +2359,6 @@ static int ufshcd_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *cmd)
 		BUG();
 	}
 
-	if (!down_read_trylock(&hba->clk_scaling_lock))
-		return SCSI_MLQUEUE_HOST_BUSY;
-
 	spin_lock_irqsave(hba->host->host_lock, flags);
 	switch (hba->ufshcd_state) {
 	case UFSHCD_STATE_OPERATIONAL:
@@ -2463,7 +2424,6 @@ static int ufshcd_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *cmd)
 out_unlock:
 	spin_unlock_irqrestore(hba->host->host_lock, flags);
 out:
-	up_read(&hba->clk_scaling_lock);
 	return err;
 }
 
@@ -2617,8 +2577,6 @@ static int ufshcd_exec_dev_cmd(struct ufs_hba *hba,
 	struct completion wait;
 	unsigned long flags;
 
-	down_read(&hba->clk_scaling_lock);
-
 	/*
 	 * Get free slot, sleep if slots are unavailable.
 	 * Even though we use wait_event() which sleeps indefinitely,
@@ -2654,7 +2612,6 @@ static int ufshcd_exec_dev_cmd(struct ufs_hba *hba,
 
 out_put_tag:
 	blk_put_request(req);
-	up_read(&hba->clk_scaling_lock);
 	return err;
 }
 
@@ -5329,7 +5286,7 @@ static void ufshcd_err_handler(struct work_struct *work)
 
 out:
 	spin_unlock_irqrestore(hba->host->host_lock, flags);
-	ufshcd_scsi_unblock_requests(hba);
+	ufshcd_unblock_requests(hba);
 	ufshcd_release(hba);
 	pm_runtime_put_sync(hba->dev);
 }
@@ -5452,8 +5409,8 @@ static void ufshcd_check_errors(struct ufs_hba *hba)
 
 		/* handle fatal errors only when link is functional */
 		if (hba->ufshcd_state == UFSHCD_STATE_OPERATIONAL) {
-			/* block commands from scsi mid-layer */
-			ufshcd_scsi_block_requests(hba);
+			/* block all commands, incl. SCSI commands */
+			ufshcd_block_requests(hba, ULONG_MAX);
 
 			hba->ufshcd_state = UFSHCD_STATE_EH_SCHEDULED;
 
@@ -5759,8 +5716,6 @@ static int ufshcd_issue_devman_upiu_cmd(struct ufs_hba *hba,
 	unsigned long flags;
 	u32 upiu_flags;
 
-	down_read(&hba->clk_scaling_lock);
-
 	req = blk_get_request(q, REQ_OP_DRV_OUT, 0);
 	if (IS_ERR(req))
 		return PTR_ERR(req);
@@ -5840,7 +5795,6 @@ static int ufshcd_issue_devman_upiu_cmd(struct ufs_hba *hba,
 	}
 
 	blk_put_request(req);
-	up_read(&hba->clk_scaling_lock);
 	return err;
 }
 
@@ -8309,8 +8263,6 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)
 	/* Initialize mutex for device management commands */
 	mutex_init(&hba->dev_cmd.lock);
 
-	init_rwsem(&hba->clk_scaling_lock);
-
 	ufshcd_init_clk_gating(hba);
 
 	ufshcd_init_clk_scaling(hba);
@@ -8396,7 +8348,6 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)
 
 	/* Hold auto suspend until async scan completes */
 	pm_runtime_get_sync(dev);
-	atomic_set(&hba->scsi_block_reqs_cnt, 0);
 	/*
 	 * We are assuming that device wasn't put in sleep/power-down
 	 * state exclusively during the boot stage before kernel.
diff --git a/drivers/scsi/ufs/ufshcd.h b/drivers/scsi/ufs/ufshcd.h
index 5865e16f53a6..c44bfcdb26a3 100644
--- a/drivers/scsi/ufs/ufshcd.h
+++ b/drivers/scsi/ufs/ufshcd.h
@@ -520,7 +520,6 @@ struct ufs_stats {
  * @urgent_bkops_lvl: keeps track of urgent bkops level for device
  * @is_urgent_bkops_lvl_checked: keeps track if the urgent bkops level for
  *  device is known or not.
- * @scsi_block_reqs_cnt: reference counting for scsi block requests
  */
 struct ufs_hba {
 	void __iomem *mmio_base;
@@ -724,9 +723,7 @@ struct ufs_hba {
 	enum bkops_status urgent_bkops_lvl;
 	bool is_urgent_bkops_lvl_checked;
 
-	struct rw_semaphore clk_scaling_lock;
 	struct ufs_desc_size desc_size;
-	atomic_t scsi_block_reqs_cnt;
 
 	struct device		bsg_dev;
 	struct request_queue	*bsg_queue;
-- 
2.24.0.rc1.363.gb1bccd3e3d-goog

Re: [PATCH v4 3/3] ufs: Simplify the clock scaling mechanism implementation

[cang]

On 2019-11-12 01:48, Bart Van Assche wrote:
> Scaling the clock is only safe while no commands are in progress. Use
> blk_mq_{un,}freeze_queue() to block submission of new commands and to
> wait for ongoing commands to complete. Remove "_scsi" from the name of
> the functions that block and unblock requests since these functions
> now block both SCSI commands and non-SCSI commands.
> 
> Tested-by: Bean Huo <beanhuo@micron.com>
> Reviewed-by: Avri Altman <avri.altman@wdc.com>
> Cc: Stanley Chu <stanley.chu@mediatek.com>
> Cc: Tomas Winkler <tomas.winkler@intel.com>
> Signed-off-by: Bart Van Assche <bvanassche@acm.org>
> ---
>  drivers/scsi/ufs/ufshcd.c | 143 +++++++++++++-------------------------
>  drivers/scsi/ufs/ufshcd.h |   3 -
>  2 files changed, 47 insertions(+), 99 deletions(-)
> 
> diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
> index 00c567bbbb64..65f05ca8d76f 100644
> --- a/drivers/scsi/ufs/ufshcd.c
> +++ b/drivers/scsi/ufs/ufshcd.c
> @@ -290,16 +290,50 @@ static inline void ufshcd_disable_irq(struct 
> ufs_hba *hba)
>  	}
>  }
> 
> -static void ufshcd_scsi_unblock_requests(struct ufs_hba *hba)
> +static void ufshcd_unblock_requests(struct ufs_hba *hba)
>  {
> -	if (atomic_dec_and_test(&hba->scsi_block_reqs_cnt))
> -		scsi_unblock_requests(hba->host);
> +	struct scsi_device *sdev;
> +
> +	blk_mq_unfreeze_queue(hba->tmf_queue);
> +	blk_mq_unfreeze_queue(hba->cmd_queue);
> +	shost_for_each_device(sdev, hba->host)
> +		blk_mq_unfreeze_queue(sdev->request_queue);
>  }
> 
> -static void ufshcd_scsi_block_requests(struct ufs_hba *hba)
> +static int ufshcd_block_requests(struct ufs_hba *hba, unsigned long 
> timeout)
>  {
> -	if (atomic_inc_return(&hba->scsi_block_reqs_cnt) == 1)
> -		scsi_block_requests(hba->host);
> +	struct scsi_device *sdev;
> +	unsigned long deadline = jiffies + timeout;
> +
> +	if (timeout == ULONG_MAX) {
> +		shost_for_each_device(sdev, hba->host)
> +			blk_mq_freeze_queue(sdev->request_queue);
> +		blk_mq_freeze_queue(hba->cmd_queue);
> +		blk_mq_freeze_queue(hba->tmf_queue);
> +		return 0;
> +	}
> +
> +	shost_for_each_device(sdev, hba->host)
> +		blk_freeze_queue_start(sdev->request_queue);
> +	blk_freeze_queue_start(hba->cmd_queue);
> +	blk_freeze_queue_start(hba->tmf_queue);
> +	shost_for_each_device(sdev, hba->host) {
> +		if (blk_mq_freeze_queue_wait_timeout(sdev->request_queue,
> +				max_t(long, 0, deadline - jiffies)) <= 0) {
> +			goto err;
> +		}
> +	}
> +	if (blk_mq_freeze_queue_wait_timeout(hba->cmd_queue,
> +				max_t(long, 0, deadline - jiffies)) <= 0)
> +		goto err;
> +	if (blk_mq_freeze_queue_wait_timeout(hba->tmf_queue,
> +				max_t(long, 0, deadline - jiffies)) <= 0)
> +		goto err;
> +	return 0;
> +
> +err:
> +	ufshcd_unblock_requests(hba);
> +	return -ETIMEDOUT;
>  }
> 
>  static void ufshcd_add_cmd_upiu_trace(struct ufs_hba *hba, unsigned 
> int tag,
> @@ -971,65 +1005,6 @@ static bool
> ufshcd_is_devfreq_scaling_required(struct ufs_hba *hba,
>  	return false;
>  }
> 
> -static int ufshcd_wait_for_doorbell_clr(struct ufs_hba *hba,
> -					u64 wait_timeout_us)
> -{
> -	unsigned long flags;
> -	int ret = 0;
> -	u32 tm_doorbell;
> -	u32 tr_doorbell;
> -	bool timeout = false, do_last_check = false;
> -	ktime_t start;
> -
> -	ufshcd_hold(hba, false);
> -	spin_lock_irqsave(hba->host->host_lock, flags);
> -	/*
> -	 * Wait for all the outstanding tasks/transfer requests.
> -	 * Verify by checking the doorbell registers are clear.
> -	 */
> -	start = ktime_get();
> -	do {
> -		if (hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL) {
> -			ret = -EBUSY;
> -			goto out;
> -		}
> -
> -		tm_doorbell = ufshcd_readl(hba, REG_UTP_TASK_REQ_DOOR_BELL);
> -		tr_doorbell = ufshcd_readl(hba, REG_UTP_TRANSFER_REQ_DOOR_BELL);
> -		if (!tm_doorbell && !tr_doorbell) {
> -			timeout = false;
> -			break;
> -		} else if (do_last_check) {
> -			break;
> -		}
> -
> -		spin_unlock_irqrestore(hba->host->host_lock, flags);
> -		schedule();
> -		if (ktime_to_us(ktime_sub(ktime_get(), start)) >
> -		    wait_timeout_us) {
> -			timeout = true;
> -			/*
> -			 * We might have scheduled out for long time so make
> -			 * sure to check if doorbells are cleared by this time
> -			 * or not.
> -			 */
> -			do_last_check = true;
> -		}
> -		spin_lock_irqsave(hba->host->host_lock, flags);
> -	} while (tm_doorbell || tr_doorbell);
> -
> -	if (timeout) {
> -		dev_err(hba->dev,
> -			"%s: timedout waiting for doorbell to clear (tm=0x%x, tr=0x%x)\n",
> -			__func__, tm_doorbell, tr_doorbell);
> -		ret = -EBUSY;
> -	}
> -out:
> -	spin_unlock_irqrestore(hba->host->host_lock, flags);
> -	ufshcd_release(hba);
> -	return ret;
> -}
> -
>  /**
>   * ufshcd_scale_gear - scale up/down UFS gear
>   * @hba: per adapter instance
> @@ -1079,27 +1054,16 @@ static int ufshcd_scale_gear(struct ufs_hba
> *hba, bool scale_up)
> 
>  static int ufshcd_clock_scaling_prepare(struct ufs_hba *hba)
>  {
> -	#define DOORBELL_CLR_TOUT_US		(1000 * 1000) /* 1 sec */
> -	int ret = 0;
>  	/*
>  	 * make sure that there are no outstanding requests when
>  	 * clock scaling is in progress
>  	 */
> -	ufshcd_scsi_block_requests(hba);
> -	down_write(&hba->clk_scaling_lock);
> -	if (ufshcd_wait_for_doorbell_clr(hba, DOORBELL_CLR_TOUT_US)) {
> -		ret = -EBUSY;
> -		up_write(&hba->clk_scaling_lock);
> -		ufshcd_scsi_unblock_requests(hba);
> -	}
> -
> -	return ret;
> +	return ufshcd_block_requests(hba, HZ);
>  }
> 
>  static void ufshcd_clock_scaling_unprepare(struct ufs_hba *hba)
>  {
> -	up_write(&hba->clk_scaling_lock);
> -	ufshcd_scsi_unblock_requests(hba);
> +	ufshcd_unblock_requests(hba);
>  }
> 
>  /**
> @@ -1471,7 +1435,7 @@ static void ufshcd_ungate_work(struct work_struct 
> *work)
>  		hba->clk_gating.is_suspended = false;
>  	}
>  unblock_reqs:
> -	ufshcd_scsi_unblock_requests(hba);
> +	ufshcd_unblock_requests(hba);
>  }
> 
>  /**
> @@ -1528,7 +1492,7 @@ int ufshcd_hold(struct ufs_hba *hba, bool async)
>  		 */
>  		/* fallthrough */
>  	case CLKS_OFF:
> -		ufshcd_scsi_block_requests(hba);
> +		ufshcd_block_requests(hba, ULONG_MAX);
>  		hba->clk_gating.state = REQ_CLKS_ON;
>  		trace_ufshcd_clk_gating(dev_name(hba->dev),
>  					hba->clk_gating.state);
> @@ -2395,9 +2359,6 @@ static int ufshcd_queuecommand(struct Scsi_Host
> *host, struct scsi_cmnd *cmd)
>  		BUG();
>  	}
> 
> -	if (!down_read_trylock(&hba->clk_scaling_lock))
> -		return SCSI_MLQUEUE_HOST_BUSY;
> -
>  	spin_lock_irqsave(hba->host->host_lock, flags);
>  	switch (hba->ufshcd_state) {
>  	case UFSHCD_STATE_OPERATIONAL:
> @@ -2463,7 +2424,6 @@ static int ufshcd_queuecommand(struct Scsi_Host
> *host, struct scsi_cmnd *cmd)
>  out_unlock:
>  	spin_unlock_irqrestore(hba->host->host_lock, flags);
>  out:
> -	up_read(&hba->clk_scaling_lock);
>  	return err;
>  }
> 
> @@ -2617,8 +2577,6 @@ static int ufshcd_exec_dev_cmd(struct ufs_hba 
> *hba,
>  	struct completion wait;
>  	unsigned long flags;
> 
> -	down_read(&hba->clk_scaling_lock);
> -
>  	/*
>  	 * Get free slot, sleep if slots are unavailable.
>  	 * Even though we use wait_event() which sleeps indefinitely,
> @@ -2654,7 +2612,6 @@ static int ufshcd_exec_dev_cmd(struct ufs_hba 
> *hba,
> 
>  out_put_tag:
>  	blk_put_request(req);
> -	up_read(&hba->clk_scaling_lock);
>  	return err;
>  }
> 
> @@ -5329,7 +5286,7 @@ static void ufshcd_err_handler(struct work_struct 
> *work)
> 
>  out:
>  	spin_unlock_irqrestore(hba->host->host_lock, flags);
> -	ufshcd_scsi_unblock_requests(hba);
> +	ufshcd_unblock_requests(hba);
>  	ufshcd_release(hba);
>  	pm_runtime_put_sync(hba->dev);
>  }
> @@ -5452,8 +5409,8 @@ static void ufshcd_check_errors(struct ufs_hba 
> *hba)
> 
>  		/* handle fatal errors only when link is functional */
>  		if (hba->ufshcd_state == UFSHCD_STATE_OPERATIONAL) {
> -			/* block commands from scsi mid-layer */
> -			ufshcd_scsi_block_requests(hba);
> +			/* block all commands, incl. SCSI commands */
> +			ufshcd_block_requests(hba, ULONG_MAX);
> 

ufshcd_block_requests() may sleep, but ufshcd_check_errors is called 
from IRQ(atmic) context.

Regards,
Can Guo.

>  			hba->ufshcd_state = UFSHCD_STATE_EH_SCHEDULED;
> 
> @@ -5759,8 +5716,6 @@ static int ufshcd_issue_devman_upiu_cmd(struct
> ufs_hba *hba,
>  	unsigned long flags;
>  	u32 upiu_flags;
> 
> -	down_read(&hba->clk_scaling_lock);
> -
>  	req = blk_get_request(q, REQ_OP_DRV_OUT, 0);
>  	if (IS_ERR(req))
>  		return PTR_ERR(req);
> @@ -5840,7 +5795,6 @@ static int ufshcd_issue_devman_upiu_cmd(struct
> ufs_hba *hba,
>  	}
> 
>  	blk_put_request(req);
> -	up_read(&hba->clk_scaling_lock);
>  	return err;
>  }
> 
> @@ -8309,8 +8263,6 @@ int ufshcd_init(struct ufs_hba *hba, void
> __iomem *mmio_base, unsigned int irq)
>  	/* Initialize mutex for device management commands */
>  	mutex_init(&hba->dev_cmd.lock);
> 
> -	init_rwsem(&hba->clk_scaling_lock);
> -
>  	ufshcd_init_clk_gating(hba);
> 
>  	ufshcd_init_clk_scaling(hba);
> @@ -8396,7 +8348,6 @@ int ufshcd_init(struct ufs_hba *hba, void
> __iomem *mmio_base, unsigned int irq)
> 
>  	/* Hold auto suspend until async scan completes */
>  	pm_runtime_get_sync(dev);
> -	atomic_set(&hba->scsi_block_reqs_cnt, 0);
>  	/*
>  	 * We are assuming that device wasn't put in sleep/power-down
>  	 * state exclusively during the boot stage before kernel.
> diff --git a/drivers/scsi/ufs/ufshcd.h b/drivers/scsi/ufs/ufshcd.h
> index 5865e16f53a6..c44bfcdb26a3 100644
> --- a/drivers/scsi/ufs/ufshcd.h
> +++ b/drivers/scsi/ufs/ufshcd.h
> @@ -520,7 +520,6 @@ struct ufs_stats {
>   * @urgent_bkops_lvl: keeps track of urgent bkops level for device
>   * @is_urgent_bkops_lvl_checked: keeps track if the urgent bkops level 
> for
>   *  device is known or not.
> - * @scsi_block_reqs_cnt: reference counting for scsi block requests
>   */
>  struct ufs_hba {
>  	void __iomem *mmio_base;
> @@ -724,9 +723,7 @@ struct ufs_hba {
>  	enum bkops_status urgent_bkops_lvl;
>  	bool is_urgent_bkops_lvl_checked;
> 
> -	struct rw_semaphore clk_scaling_lock;
>  	struct ufs_desc_size desc_size;
> -	atomic_t scsi_block_reqs_cnt;
> 
>  	struct device		bsg_dev;
>  	struct request_queue	*bsg_queue;

RE: [EXT] Re: [PATCH v4 3/3] ufs: Simplify the clock scaling mechanism implementation

[Bean Huo (beanhuo)]

Hi, Bart

> >  		/* handle fatal errors only when link is functional */
> >  		if (hba->ufshcd_state == UFSHCD_STATE_OPERATIONAL) {
> > -			/* block commands from scsi mid-layer */
> > -			ufshcd_scsi_block_requests(hba);
> > +			/* block all commands, incl. SCSI commands */
> > +			ufshcd_block_requests(hba, ULONG_MAX);
> >
> 
> ufshcd_block_requests() may sleep, but ufshcd_check_errors is called from
> IRQ(atmic) context.
> 
> Regards,
> Can Guo.

[Bean Huo] 
Can Guo's comment is correct, I triggered an error manually, my UFS being killed.
And removing this patch, no problem. Please take this seriously.
Thanks,

//Bean

Re: [EXT] Re: [PATCH v4 3/3] ufs: Simplify the clock scaling mechanism implementation

[Bart Van Assche]

On 11/12/19 7:43 AM, Bean Huo (beanhuo) wrote:
> Can Guo wrote: 
 >> Bart Van Assche wrote:
>>>   		/* handle fatal errors only when link is functional */
>>>   		if (hba->ufshcd_state == UFSHCD_STATE_OPERATIONAL) {
>>> -			/* block commands from scsi mid-layer */
>>> -			ufshcd_scsi_block_requests(hba);
>>> +			/* block all commands, incl. SCSI commands */
>>> +			ufshcd_block_requests(hba, ULONG_MAX);
>>>
>>
>> ufshcd_block_requests() may sleep, but ufshcd_check_errors is called from
>> IRQ(atmic) context.
> 
> [Bean Huo]
> Can Guo's comment is correct, I triggered an error manually, my UFS being killed.
> And removing this patch, no problem. Please take this seriously.

Hi Can and Bean,

Thank you for having reported this. I'll look into keeping 
scsi_block_requests() and scsi_unblock_requests() in this code path.

Thanks,

Bart.

RE: [PATCH v4 0/3] Simplify and optimize the UFS driver

[Avri Altman]

> 
> Hello Martin,
> 
> This patch series that simplifies and optimizes the UFS driver. Please consider
> this patch series for kernel v5.5.
> 
> Thanks,
> 
> Bart.
> 
> Changes compared to v3:
> - Left out "scsi" from the name of the functions that suspend and resume
>   command processing.

Looks good to me.

Thanks,
Avri