[PATCH v1] scsi: ufs: fix rpm racing issue

[PATCH v1] scsi: ufs: fix rpm racing issue

[peter.wang]

From: Peter Wang <peter.wang@mediatek.com>

Device WLun may enter RPM_SUSPENDED and the consumer is RPM_ACTIVE.

The error state is
Device Wlun RPM: status:2, usage_count:0
Consumer: 0:0:0:49476 Link: rpm_active:1, RPM: status:2, usage_count:0
Consumer: 0:0:0:49456 Link: rpm_active:1, RPM: status:2, usage_count:0
Consumer: 0:0:0:0 Link: rpm_active:1, RPM: status:2, usage_count:0
Consumer: 0:0:0:1 Link: rpm_active:1, RPM: status:2, usage_count:0
Consumer: 0:0:0:2 Link: rpm_active:1, RPM: status:0, usage_count:0

Because rpm enable before rpm delay set, scsi_autopm_put_device
invoke by scsi_sysfs_add_sdev may cause consumer enter suspend immediately.
Thas will always set rpm_active to 1.
If driver_probe_device invoke pm_runtime_get_suppliers just befor
previous scsi_autopm_put_device, the rpm_active will change to 1
after pm_runtime_put_suppliers.
Set this delay to avoid scsi_autopm_put_device enter suspend immediately.

Signed-off-by: Peter Wang <peter.wang@mediatek.com>
---
 drivers/scsi/ufs/ufshcd.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index 3f9caafa91bf..1250792d16be 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -5024,10 +5024,14 @@ static int ufshcd_slave_configure(struct scsi_device *sdev)
 	 * Block runtime-pm until all consumers are added.
 	 * Refer ufshcd_setup_links().
 	 */
-	if (is_device_wlun(sdev))
+	if (is_device_wlun(sdev)) {
 		pm_runtime_get_noresume(&sdev->sdev_gendev);
-	else if (ufshcd_is_rpm_autosuspend_allowed(hba))
+	} else if (ufshcd_is_rpm_autosuspend_allowed(hba)) {
 		sdev->rpm_autosuspend = 1;
+		pm_runtime_set_autosuspend_delay(&sdev->sdev_gendev,
+			RPM_AUTOSUSPEND_DELAY_MS);
+	}
+
 	/*
 	 * Do not print messages during runtime PM to avoid never-ending cycles
 	 * of messages written back to storage by user space causing runtime
-- 
2.18.0

RE: [PATCH v1] scsi: ufs: fix rpm racing issue

[Avri Altman]

 
> From: Peter Wang <peter.wang@mediatek.com>
> 
> Device WLun may enter RPM_SUSPENDED and the consumer is RPM_ACTIVE.
> 
> The error state is
> Device Wlun RPM: status:2, usage_count:0
> Consumer: 0:0:0:49476 Link: rpm_active:1, RPM: status:2, usage_count:0
> Consumer: 0:0:0:49456 Link: rpm_active:1, RPM: status:2, usage_count:0
> Consumer: 0:0:0:0 Link: rpm_active:1, RPM: status:2, usage_count:0
> Consumer: 0:0:0:1 Link: rpm_active:1, RPM: status:2, usage_count:0
> Consumer: 0:0:0:2 Link: rpm_active:1, RPM: status:0, usage_count:0
> 
> Because rpm enable before rpm delay set, scsi_autopm_put_device
> invoke by scsi_sysfs_add_sdev may cause consumer enter suspend
> immediately.
> Thas will always set rpm_active to 1.
> If driver_probe_device invoke pm_runtime_get_suppliers just befor
> previous scsi_autopm_put_device, the rpm_active will change to 1
> after pm_runtime_put_suppliers.
> Set this delay to avoid scsi_autopm_put_device enter suspend immediately.
> 
> Signed-off-by: Peter Wang <peter.wang@mediatek.com>
> ---
>  drivers/scsi/ufs/ufshcd.c | 8 ++++++--
>  1 file changed, 6 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
> index 3f9caafa91bf..1250792d16be 100644
> --- a/drivers/scsi/ufs/ufshcd.c
> +++ b/drivers/scsi/ufs/ufshcd.c
> @@ -5024,10 +5024,14 @@ static int ufshcd_slave_configure(struct
> scsi_device *sdev)
>          * Block runtime-pm until all consumers are added.
>          * Refer ufshcd_setup_links().
>          */
> -       if (is_device_wlun(sdev))
> +       if (is_device_wlun(sdev)) {
>                 pm_runtime_get_noresume(&sdev->sdev_gendev);
> -       else if (ufshcd_is_rpm_autosuspend_allowed(hba))
> +       } else if (ufshcd_is_rpm_autosuspend_allowed(hba)) {
>                 sdev->rpm_autosuspend = 1;
> +               pm_runtime_set_autosuspend_delay(&sdev->sdev_gendev,
> +                       RPM_AUTOSUSPEND_DELAY_MS);
> +       }
Maybe remove ufshcd_blk_pm_runtime_init from ufshcd_scsi_add_wlus,
And call here ufshcd_blk_pm_runtime_init() ?

Thanks,
Avri

> +
>         /*
>          * Do not print messages during runtime PM to avoid never-ending cycles
>          * of messages written back to storage by user space causing runtime
> --
> 2.18.0

Re: [PATCH v1] scsi: ufs: fix rpm racing issue

[Peter Wang]

On 4/25/22 4:35 PM, Avri Altman wrote:
>   
>> From: Peter Wang <peter.wang@mediatek.com>
>>
>> Device WLun may enter RPM_SUSPENDED and the consumer is RPM_ACTIVE.
>>
>> The error state is
>> Device Wlun RPM: status:2, usage_count:0
>> Consumer: 0:0:0:49476 Link: rpm_active:1, RPM: status:2, usage_count:0
>> Consumer: 0:0:0:49456 Link: rpm_active:1, RPM: status:2, usage_count:0
>> Consumer: 0:0:0:0 Link: rpm_active:1, RPM: status:2, usage_count:0
>> Consumer: 0:0:0:1 Link: rpm_active:1, RPM: status:2, usage_count:0
>> Consumer: 0:0:0:2 Link: rpm_active:1, RPM: status:0, usage_count:0
>>
>> Because rpm enable before rpm delay set, scsi_autopm_put_device
>> invoke by scsi_sysfs_add_sdev may cause consumer enter suspend
>> immediately.
>> Thas will always set rpm_active to 1.
>> If driver_probe_device invoke pm_runtime_get_suppliers just befor
>> previous scsi_autopm_put_device, the rpm_active will change to 1
>> after pm_runtime_put_suppliers.
>> Set this delay to avoid scsi_autopm_put_device enter suspend immediately.
>>
>> Signed-off-by: Peter Wang <peter.wang@mediatek.com>
>> ---
>>   drivers/scsi/ufs/ufshcd.c | 8 ++++++--
>>   1 file changed, 6 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
>> index 3f9caafa91bf..1250792d16be 100644
>> --- a/drivers/scsi/ufs/ufshcd.c
>> +++ b/drivers/scsi/ufs/ufshcd.c
>> @@ -5024,10 +5024,14 @@ static int ufshcd_slave_configure(struct
>> scsi_device *sdev)
>>           * Block runtime-pm until all consumers are added.
>>           * Refer ufshcd_setup_links().
>>           */
>> -       if (is_device_wlun(sdev))
>> +       if (is_device_wlun(sdev)) {
>>                  pm_runtime_get_noresume(&sdev->sdev_gendev);
>> -       else if (ufshcd_is_rpm_autosuspend_allowed(hba))
>> +       } else if (ufshcd_is_rpm_autosuspend_allowed(hba)) {
>>                  sdev->rpm_autosuspend = 1;
>> +               pm_runtime_set_autosuspend_delay(&sdev->sdev_gendev,
>> +                       RPM_AUTOSUSPEND_DELAY_MS);
>> +       }
> Maybe remove ufshcd_blk_pm_runtime_init from ufshcd_scsi_add_wlus,
> And call here ufshcd_blk_pm_runtime_init() ?
>
> Thanks,
> Avri

Hi Arvi,


Thanks review, but I need discard this patch because it is not correct.

Sorry for inconvenient.


Thanks

Peter

>
>> +
>>          /*
>>           * Do not print messages during runtime PM to avoid never-ending cycles
>>           * of messages written back to storage by user space causing runtime
>> --
>> 2.18.0