On 11/12/19 4:11 PM, cang@xxxxxxxxxxxxxx wrote:
On 2019-11-13 01:37, Bart Van Assche wrote:
@@ -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);
ufshcd_hold(async == true) is used in ufshcd_queuecommand() path because
ufshcd_queuecommand() can be entered under atomic contexts.
Thus ufshcd_block_requests() here has the same risk causing scheduling
while atomic.
FYI, it is not easy to hit above scenario in small scale of test.
Hi Bean,
How about replacing patch 4/4 with the attached patch?
Thanks,
Bart.
>From 112fd52ef68927ab9b19fd84765ea31aacd2d0de Mon Sep 17 00:00:00 2001
From: Bart Van Assche <bvanassche@xxxxxxx>
Date: Thu, 10 Oct 2019 15:56:35 -0700
Subject: [PATCH] ufs: Simplify the clock scaling mechanism implementation
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. This patch removes a semaphore
down and up operation pair from the hot path.
Cc: Bean Huo <beanhuo@xxxxxxxxxx>
Cc: Avri Altman <avri.altman@xxxxxxx>
Cc: Stanley Chu <stanley.chu@xxxxxxxxxxxx>
Cc: Tomas Winkler <tomas.winkler@xxxxxxxxx>
Signed-off-by: Bart Van Assche <bvanassche@xxxxxxx>
---
drivers/scsi/ufs/ufshcd.c | 132 ++++++++++++++------------------------
drivers/scsi/ufs/ufshcd.h | 1 -
2 files changed, 48 insertions(+), 85 deletions(-)
diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index 99ce1d651f03..f00d665715d1 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -302,6 +302,52 @@ static void ufshcd_scsi_block_requests(struct ufs_hba *hba)
scsi_block_requests(hba->host);
}
+static void ufshcd_unblock_requests(struct ufs_hba *hba)
+{
+ 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 int ufshcd_block_requests(struct ufs_hba *hba, unsigned long timeout)
+{
+ 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,
const char *str)
{
@@ -971,65 +1017,6 @@ 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 +1066,16 @@ 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);
}
/**
@@ -2394,9 +2370,6 @@ 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:
@@ -2462,7 +2435,6 @@ 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;
}
@@ -2616,8 +2588,6 @@ 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,
@@ -2653,7 +2623,6 @@ 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;
}
@@ -5771,8 +5740,6 @@ 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);
@@ -5852,7 +5819,6 @@ static int ufshcd_issue_devman_upiu_cmd(struct ufs_hba *hba,
}
blk_put_request(req);
- up_read(&hba->clk_scaling_lock);
return err;
}
@@ -8321,8 +8287,6 @@ 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);
diff --git a/drivers/scsi/ufs/ufshcd.h b/drivers/scsi/ufs/ufshcd.h
index 5865e16f53a6..5ebb920ae874 100644
--- a/drivers/scsi/ufs/ufshcd.h
+++ b/drivers/scsi/ufs/ufshcd.h
@@ -724,7 +724,6 @@ 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;
