[PATCH] scsi: ufs: move shutting_down back to ufshcd_shutdown

[PATCH] scsi: ufs: move shutting_down back to ufshcd_shutdown

[Jaegeuk Kim]

The commit b294ff3e3449 ("scsi: ufs: core: Enable power management for wlun")
moved hba->shutting_down from ufshcd_shutdown to ufshcd_wl_shutdown, which
introduced regression as belows.

ufshcd_err_handler started; HBA state eh_non_fatal; powered 1; shutting down 1; saved_err = 4; saved_uic_err = 64; force_reset = 0
...
task:init            state:D stack:    0 pid:    1 ppid:     0 flags:0x04000008
Call trace:
 __switch_to+0x25c/0x5e0
 __schedule+0x68c/0xaa8
 schedule+0x12c/0x24c
 schedule_timeout+0x98/0x138
 wait_for_common_io+0x13c/0x30c
 blk_execute_rq+0xb0/0x10c
 __scsi_execute+0x100/0x27c
 ufshcd_set_dev_pwr_mode+0x1c8/0x408
 __ufshcd_wl_suspend+0x564/0x688
 ufshcd_wl_shutdown+0xa8/0xc0
 device_shutdown+0x234/0x578
 kernel_restart+0x4c/0x140
 __arm64_sys_reboot+0x3a0/0x414
 el0_svc_common+0xd0/0x1e4
 el0_svc+0x28/0x88
 el0_sync_handler+0x8c/0xf0
 el0_sync+0x1c0/0x200

The init for reboot was stuck, since ufshcd_err_hanlder was skipped when
shutting down WLUN. This patch allows to run the error handler and let
disable it during final ufshcd_shutdown only.

Cc: stable@vger.kernel.org
Fixes: b294ff3e3449 ("scsi: ufs: core: Enable power management for wlun")
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
---
 drivers/scsi/ufs/ufshcd.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index 460d2b440d2e..a37813b474d0 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -9178,10 +9178,6 @@ static void ufshcd_wl_shutdown(struct device *dev)
 
 	hba = shost_priv(sdev->host);
 
-	down(&hba->host_sem);
-	hba->shutting_down = true;
-	up(&hba->host_sem);
-
 	/* Turn on everything while shutting down */
 	ufshcd_rpm_get_sync(hba);
 	scsi_device_quiesce(sdev);
@@ -9387,6 +9383,10 @@ EXPORT_SYMBOL(ufshcd_runtime_resume);
  */
 int ufshcd_shutdown(struct ufs_hba *hba)
 {
+	down(&hba->host_sem);
+	hba->shutting_down = true;
+	up(&hba->host_sem);
+
 	if (ufshcd_is_ufs_dev_poweroff(hba) && ufshcd_is_link_off(hba))
 		goto out;
 
-- 
2.35.1.574.g5d30c73bfb-goog

Re: [PATCH] scsi: ufs: move shutting_down back to ufshcd_shutdown

[Adrian Hunter]

On 25/02/2022 01:56, Jaegeuk Kim wrote:
> The commit b294ff3e3449 ("scsi: ufs: core: Enable power management for wlun")
> moved hba->shutting_down from ufshcd_shutdown to ufshcd_wl_shutdown, which
> introduced regression as belows.
> 
> ufshcd_err_handler started; HBA state eh_non_fatal; powered 1; shutting down 1; saved_err = 4; saved_uic_err = 64; force_reset = 0
> ...
> task:init            state:D stack:    0 pid:    1 ppid:     0 flags:0x04000008
> Call trace:
>  __switch_to+0x25c/0x5e0
>  __schedule+0x68c/0xaa8
>  schedule+0x12c/0x24c
>  schedule_timeout+0x98/0x138
>  wait_for_common_io+0x13c/0x30c
>  blk_execute_rq+0xb0/0x10c
>  __scsi_execute+0x100/0x27c
>  ufshcd_set_dev_pwr_mode+0x1c8/0x408
>  __ufshcd_wl_suspend+0x564/0x688
>  ufshcd_wl_shutdown+0xa8/0xc0
>  device_shutdown+0x234/0x578
>  kernel_restart+0x4c/0x140
>  __arm64_sys_reboot+0x3a0/0x414
>  el0_svc_common+0xd0/0x1e4
>  el0_svc+0x28/0x88
>  el0_sync_handler+0x8c/0xf0
>  el0_sync+0x1c0/0x200
> 
> The init for reboot was stuck, since ufshcd_err_hanlder was skipped when
> shutting down WLUN. This patch allows to run the error handler and let
> disable it during final ufshcd_shutdown only.

I do not understand why it is stuck?  If there was a non-fatal error, the
request should complete anyway, or at least timeout.

> 
> Cc: stable@vger.kernel.org
> Fixes: b294ff3e3449 ("scsi: ufs: core: Enable power management for wlun")
> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
> ---
>  drivers/scsi/ufs/ufshcd.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
> index 460d2b440d2e..a37813b474d0 100644
> --- a/drivers/scsi/ufs/ufshcd.c
> +++ b/drivers/scsi/ufs/ufshcd.c
> @@ -9178,10 +9178,6 @@ static void ufshcd_wl_shutdown(struct device *dev)
>  
>  	hba = shost_priv(sdev->host);
>  
> -	down(&hba->host_sem);
> -	hba->shutting_down = true;
> -	up(&hba->host_sem);
> -
>  	/* Turn on everything while shutting down */
>  	ufshcd_rpm_get_sync(hba);
>  	scsi_device_quiesce(sdev);
> @@ -9387,6 +9383,10 @@ EXPORT_SYMBOL(ufshcd_runtime_resume);
>   */
>  int ufshcd_shutdown(struct ufs_hba *hba)
>  {
> +	down(&hba->host_sem);
> +	hba->shutting_down = true;
> +	up(&hba->host_sem);

This does not seem right because the device is not accessible after
__ufshcd_wl_suspend(hba, UFS_SHUTDOWN_PM), so shutting_down needs to
be set before.


> +
>  	if (ufshcd_is_ufs_dev_poweroff(hba) && ufshcd_is_link_off(hba))
>  		goto out;
>