ufs: core: fix lockdep warning of clk_scaling_lock

PATCH Superseded [email protected] 2 participants 3 messages Jul 25, 2022 – Jul 26, 2022

Subsystems (3)

SCSI Arm MediaTek UFS host controller

lore

Share & tools

lore

[email protected]

author

2022-07-25 04:30 UTC #0 lore raw

PATCH v2 1/1 ufs: core: fix lockdep warning of clk_scaling_lock

From: Peter Wang <[email protected]>

There have a lockdep warning like below in current flow. kworker/u16:0: Possible unsafe locking scenario:

kworker/u16:0:        CPU0                    CPU1
kworker/u16:0:        ----                    ----

kworker/u16:0: lock(&hba->clk_scaling_lock);
kworker/u16:0: lock(&hba->dev_cmd.lock);
kworker/u16:0: lock(&hba->clk_scaling_lock);
kworker/u16:0: lock(&hba->dev_cmd.lock);
kworker/u16:0:

Before this patch clk_scaling_lock was held in reader mode during the ufshcd_wb_toggle() call. With this patch applied clk_scaling_lock is not held while ufshcd_wb_toggle() is called.

This is safe because ufshcd_wb_toggle will held clk_scaling_lock in reader mode "again" in flow ufshcd_wb_toggle -> __ufshcd_wb_toggle -> ufshcd_query_flag_retry -> ufshcd_query_flag -> ufshcd_exec_dev_cmd -> down_read(&hba->clk_scaling_lock);
The protect should enough and make sure clock is not change while send command.

Signed-off-by Peter Wang <[email protected]>

---
 drivers/ufs/core/ufshcd.c | 22 +++++++++++-----------
 1 file changed, 11 insertions(+), 11 deletions(-)

drivers/ufs/core/ufshcd.c +11 ~~-11~~

index c7b337480e3e..209089bd8085 100644
@@ -1249,12 +1249,10 @@ static int ufshcd_clock_scaling_prepare(struct ufs_hba *hba)
 	return ret;
 }
 
-static void ufshcd_clock_scaling_unprepare(struct ufs_hba *hba, bool writelock)
+static void ufshcd_clock_scaling_unprepare(struct ufs_hba *hba)
 {
-	if (writelock)
-		up_write(&hba->clk_scaling_lock);
-	else
-		up_read(&hba->clk_scaling_lock);
+	up_write(&hba->clk_scaling_lock);
+
 	ufshcd_scsi_unblock_requests(hba);
 	ufshcd_release(hba);
 }

@@ -1271,7 +1269,7 @@ static void ufshcd_clock_scaling_unprepare(struct ufs_hba *hba, bool writelock)
 static int ufshcd_devfreq_scale(struct ufs_hba *hba, bool scale_up)
 {
 	int ret = 0;
-	bool is_writelock = true;
+	bool wb_toggle = false;
 
 	ret = ufshcd_clock_scaling_prepare(hba);
 	if (ret)

@@ -1300,13 +1298,15 @@ static int ufshcd_devfreq_scale(struct ufs_hba *hba, bool scale_up)
 		}
 	}
 
-	/* Enable Write Booster if we have scaled up else disable it */
-	downgrade_write(&hba->clk_scaling_lock);
-	is_writelock = false;
-	ufshcd_wb_toggle(hba, scale_up);
+	wb_toggle = true;
 
 out_unprepare:
-	ufshcd_clock_scaling_unprepare(hba, is_writelock);
+	ufshcd_clock_scaling_unprepare(hba);
+
+	/* Enable Write Booster if we have scaled up else disable it */
+	if (wb_toggle)
+		ufshcd_wb_toggle(hba, scale_up);
+
 	return ret;
 }

Signature · 1 line

2.18.0

Bart Van Assche

reviewer

2022-07-25 16:56 UTC #1 lore raw

Quoted 13 lines · On 7/24/22 21:30, [email protected] wrote:

From: Peter Wang <[email protected]>
There have a lockdep warning like below in current flow. kworker/u16:0: Possible unsafe locking scenario:
kworker/u16:0:        CPU0                    CPU1
kworker/u16:0:        ----                    ----
kworker/u16:0: lock(&hba->clk_scaling_lock);
kworker/u16:0: lock(&hba->dev_cmd.lock);
kworker/u16:0: lock(&hba->clk_scaling_lock);
kworker/u16:0: lock(&hba->dev_cmd.lock);
kworker/u16:0:

Quoted 7 lines

Before this patch clk_scaling_lock was held in reader mode during the ufshcd_wb_toggle() call. With this patch applied clk_scaling_lock is not held while ufshcd_wb_toggle() is called.
This is safe because ufshcd_wb_toggle will held clk_scaling_lock in reader mode "again" in flow ufshcd_wb_toggle -> __ufshcd_wb_toggle -> ufshcd_query_flag_retry -> ufshcd_query_flag -> ufshcd_exec_dev_cmd -> down_read(&hba->clk_scaling_lock);
The protect should enough and make sure clock is not change while send command.

Since this is a bug fix, please add a Fixes: tag.

drivers/ufs/core/ufshcd.c static int ufshcd_devfreq_scale(struct ufs_hba *hba, bool scale_up) · line 1311

Quoted 10 lines

out_unprepare:

drivers/ufs/core/ufshcd.c +6 -1

-	ufshcd_clock_scaling_unprepare(hba, is_writelock);
+	ufshcd_clock_scaling_unprepare(hba);
+
+	/* Enable Write Booster if we have scaled up else disable it */
+	if (wb_toggle)
+		ufshcd_wb_toggle(hba, scale_up);
+
  	return ret;
  }

Can the above patch can have the following unwanted effect?
* ufshcd_devfreq_scale() calls ufshcd_clock_scaling_unprepare().
* Clock scaling to a lower frequency happens.
* ufshcd_wb_toggle() enables the write booster.

Shouldn't the above ufshcd_wb_toggle() call be surrounded by down_read() and up_read() calls in addition to a check whether the WriteBooster really should be enabled instead of using 'scale_up'?

Thanks,

Bart.

Peter Wang

author

2022-07-26 08:42 UTC #2 lore raw

Hi Bart,

Quoted 17 lines · Bart Van Assche

Quoted 16 lines · On 7/24/22 21:30, [email protected] wrote:
From: Peter Wang <[email protected]>
There have a lockdep warning like below in current flow. kworker/u16:0: Possible unsafe locking scenario:
kworker/u16:0: CPU0 CPU1 kworker/u16:0: ---- ---- kworker/u16:0: lock(&hba->clk_scaling_lock); kworker/u16:0: lock(&hba->dev_cmd.lock); kworker/u16:0: lock(&hba->clk_scaling_lock); kworker/u16:0: lock(&hba->dev_cmd.lock); kworker/u16:0:
Before this patch clk_scaling_lock was held in reader mode during the ufshcd_wb_toggle() call. With this patch applied clk_scaling_lock is not held while ufshcd_wb_toggle() is called.

Quoted 6 lines

Quoted 4 lines
This is safe because ufshcd_wb_toggle will held clk_scaling_lock in reader mode "again" in flow ufshcd_wb_toggle -> __ufshcd_wb_toggle -> ufshcd_query_flag_retry -> ufshcd_query_flag -> ufshcd_exec_dev_cmd -> down_read(&hba->clk_scaling_lock);
The protect should enough and make sure clock is not change while send command.
Since this is a bug fix, please add a Fixes: tag.

Will add Fixes: tag in next version.

Quoted 17 lines

Quoted 11 lines
out_unprepare:
- ufshcd_clock_scaling_unprepare(hba, is_writelock); + ufshcd_clock_scaling_unprepare(hba);
drivers/ufs/core/ufshcd.c +5 -0
+
+    /* Enable Write Booster if we have scaled up else disable it */
+    if (wb_toggle)
+        ufshcd_wb_toggle(hba, scale_up);
+
return ret;
}
Can the above patch can have the following unwanted effect?
* ufshcd_devfreq_scale() calls ufshcd_clock_scaling_unprepare().
* Clock scaling to a lower frequency happens.
* ufshcd_wb_toggle() enables the write booster.

Quoted 7 lines

Shouldn't the above ufshcd_wb_toggle() call be surrounded by down_read() and up_read() calls in addition to a check whether the WriteBooster really should be enabled instead of using 'scale_up'?
Thanks,
Bart.

You means ufshcd_devfreq_scale may have racing in two thread, right?
Then yes, it may have this unwanted effect in this condition.
But ufshcd_wb_toggle should not hold clk_scaling_lock, or the deadlock may happen.
I will change this patch to protect ufshcd_devfreq_scale racing and ufshcd_wb_toggle in next version.

Thanks.
Peter