diff mbox series

scsi: ufs: Allow async suspend/resume callbacks

Message ID 20210728012743.1063928-1-paillon@google.com
State New
Headers show
Series scsi: ufs: Allow async suspend/resume callbacks | expand

Commit Message

Vincent Palomares July 28, 2021, 1:27 a.m. UTC
Allow UFS suspend/resume callbacks to run in parallel with other
suspend/resume callbacks. This can recoup dozens of milliseconds on the
resume path if UFS hardware needs to be powered back on.

Suspending and resuming asynchronously is safe to do so long as the driver
callbacks only depend on resources made available by either a) parent
devices or b) devices explicitly marked as suppliers with device_link_add.

Cc: Bjorn Helgaas <bhelgaas@google.com>
Cc: Jaegeuk Kim <jaegeuk@kernel.org>
Cc: Bart Van Assche <bvanassche@acm.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Stanley Chu <stanley.chu@mediatek.com>
Cc: Can Guo <cang@codeaurora.org>
Cc: Asutosh Das <asutoshd@codeaurora.org>
Cc: Avri Altman <avri.altman@wdc.com>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Vincent Palomares <paillon@google.com>
---

Are there any suspend/resume dependencies for UFS drivers not tracked by
the device parent relationship?

drivers/scsi/ufs/ufshcd.c | 1 +
 1 file changed, 1 insertion(+)

Comments

Bart Van Assche July 28, 2021, 5:57 p.m. UTC | #1
On 7/27/21 6:27 PM, Vincent Palomares wrote:
> Are there any suspend/resume dependencies for UFS drivers not tracked by
> the device parent relationship?

Not that I know of. Since I like this patch:

Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Avri Altman July 29, 2021, 6:48 a.m. UTC | #2
> Allow UFS suspend/resume callbacks to run in parallel with other

> suspend/resume callbacks. This can recoup dozens of milliseconds on the

> resume path if UFS hardware needs to be powered back on.

> 

> Suspending and resuming asynchronously is safe to do so long as the driver

> callbacks only depend on resources made available by either a) parent

> devices or b) devices explicitly marked as suppliers with device_link_add.

> 

> Cc: Bjorn Helgaas <bhelgaas@google.com>

> Cc: Jaegeuk Kim <jaegeuk@kernel.org>

> Cc: Bart Van Assche <bvanassche@acm.org>

> Cc: Adrian Hunter <adrian.hunter@intel.com>

> Cc: Stanley Chu <stanley.chu@mediatek.com>

> Cc: Can Guo <cang@codeaurora.org>

> Cc: Asutosh Das <asutoshd@codeaurora.org>

> Cc: Avri Altman <avri.altman@wdc.com>

> Cc: Martin K. Petersen <martin.petersen@oracle.com>

> Signed-off-by: Vincent Palomares <paillon@google.com>

> ---

> 

> Are there any suspend/resume dependencies for UFS drivers not tracked by

> the device parent relationship?

> 

> drivers/scsi/ufs/ufshcd.c | 1 +

>  1 file changed, 1 insertion(+)

> 

> diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c

> index b87ff68aa9aa..9ec5c308a0ea 100644

> --- a/drivers/scsi/ufs/ufshcd.c

> +++ b/drivers/scsi/ufs/ufshcd.c

> @@ -9625,6 +9625,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem

> *mmio_base, unsigned int irq)

>         async_schedule(ufshcd_async_scan, hba);

>         ufs_sysfs_add_nodes(hba->dev);

> 

> +       device_enable_async_suspend(dev);

>         return 0;

Isn't device_enable_async_suspend is being called for each lun in scsi_sysfs_add_sdev Anyway?

Thanks,
Avri
> 

>  free_tmf_queue:

> --

> 2.32.0.432.gabb21c7263-goog
Bart Van Assche July 30, 2021, 4:05 p.m. UTC | #3
On 7/28/21 11:48 PM, Avri Altman wrote:
> Vincent wrote:

>> diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c

>> index b87ff68aa9aa..9ec5c308a0ea 100644

>> --- a/drivers/scsi/ufs/ufshcd.c

>> +++ b/drivers/scsi/ufs/ufshcd.c

>> @@ -9625,6 +9625,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem

>> *mmio_base, unsigned int irq)

>>          async_schedule(ufshcd_async_scan, hba);

>>          ufs_sysfs_add_nodes(hba->dev);

>>

>> +       device_enable_async_suspend(dev);

>>          return 0;

> Isn't device_enable_async_suspend is being called for each lun in scsi_sysfs_add_sdev Anyway?


Hi Avri,

Our measurements have shown that resume takes longer than it should with 
encryption enabled. While suspending we change the power mode of the UFS 
device to a mode in which it loses crypto keys. Restoring crypto keys 
during resume (blk_ksm_reprogram_all_keys()) takes about 31 ms. This is 
the long pole and takes much more time than resuming LUNs. This patch 
makes UFS resume happen concurrently with resuming other devices in the 
system instead of serializing it. Measurements have shown that this 
patch significantly improves the time needed to resume an Android device.

Bart.
Avri Altman July 30, 2021, 6:17 p.m. UTC | #4
> On 7/28/21 11:48 PM, Avri Altman wrote:

> > Vincent wrote:

> >> diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c

> >> index b87ff68aa9aa..9ec5c308a0ea 100644

> >> --- a/drivers/scsi/ufs/ufshcd.c

> >> +++ b/drivers/scsi/ufs/ufshcd.c

> >> @@ -9625,6 +9625,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem

> >> *mmio_base, unsigned int irq)

> >>          async_schedule(ufshcd_async_scan, hba);

> >>          ufs_sysfs_add_nodes(hba->dev);

> >>

> >> +       device_enable_async_suspend(dev);

> >>          return 0;

> > Isn't device_enable_async_suspend is being called for each lun in

> scsi_sysfs_add_sdev Anyway?

> 

> Hi Avri,

> 

> Our measurements have shown that resume takes longer than it should with

> encryption enabled. While suspending we change the power mode of the UFS

> device to a mode in which it loses crypto keys. Restoring crypto keys

> during resume (blk_ksm_reprogram_all_keys()) takes about 31 ms. This is

> the long pole and takes much more time than resuming LUNs. This patch

> makes UFS resume happen concurrently with resuming other devices in the

> system instead of serializing it. Measurements have shown that this

> patch significantly improves the time needed to resume an Android device.

OK.
Thanks for the extra info.

Thanks,
Avri

> 

> Bart.
Martin K. Petersen July 31, 2021, 3:48 a.m. UTC | #5
Vincent,

> Allow UFS suspend/resume callbacks to run in parallel with other

> suspend/resume callbacks. This can recoup dozens of milliseconds on the

> resume path if UFS hardware needs to be powered back on.


Applied to 5.15/scsi-staging, thanks!

-- 
Martin K. Petersen	Oracle Linux Engineering
diff mbox series

Patch

diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index b87ff68aa9aa..9ec5c308a0ea 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -9625,6 +9625,7 @@  int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)
 	async_schedule(ufshcd_async_scan, hba);
 	ufs_sysfs_add_nodes(hba->dev);
 
+	device_enable_async_suspend(dev);
 	return 0;
 
 free_tmf_queue: