Skip to contentSkip to conversation

ufs: Fix handling of active power mode in ufshcd_suspend()

PATCH In review Bart Van Assche 2 participants 3 messages May 12, 2021 – May 14, 2021 SCSI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ufs: Fix handling of active power mode in ufshcd_suspend()

If the rpm_lvl and spm_lvl sysfs attributes indicate that ufshcd_suspend() should keep the link active, re-enable clock gating instead of disabling clocks.

Cc Can Guo <[email protected]>
Cc Alim Akhtar <[email protected]>
Cc Avri Altman <[email protected]>
Cc Stanley Chu <[email protected]>
Cc Bean Huo <[email protected]>
Cc Adrian Hunter <[email protected]>
---
 drivers/scsi/ufs/ufshcd.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
drivers/scsi/ufs/ufshcd.c +2 -4
index f540b0cc253f..c96e36aab989 100644
@@ -8690,9 +8690,8 @@ static int ufshcd_suspend(struct ufs_hba *hba, enum ufs_pm_op pm_op)
 	ufshcd_clk_scaling_suspend(hba, true);
 
 	if (req_dev_pwr_mode == UFS_ACTIVE_PWR_MODE &&
-			req_link_state == UIC_LINK_ACTIVE_STATE) {
-		goto disable_clks;
-	}
+			req_link_state == UIC_LINK_ACTIVE_STATE)
+		goto enable_gating;
 
 	if ((req_dev_pwr_mode == hba->curr_dev_pwr_mode) &&
 	    (req_link_state == hba->uic_link_state))
@@ -8754,7 +8753,6 @@ static int ufshcd_suspend(struct ufs_hba *hba, enum ufs_pm_op pm_op)
 	if (ret)
 		goto set_dev_active;
 
-disable_clks:
 	/*
 	 * Call vendor specific suspend callback. As these callbacks may access
 	 * vendor specific host controller register space call them before the
#1 lore raw
Quoted 41 lines · Bart Van Assche

If the rpm_lvl and spm_lvl sysfs attributes indicate that ufshcd_suspend() should keep the link active, re-enable clock gating instead of disabling clocks.

Cc Can Guo <[email protected]>
Cc Alim Akhtar <[email protected]>
Cc Avri Altman <[email protected]>
Cc Stanley Chu <[email protected]>
Cc Bean Huo <[email protected]>
Cc Adrian Hunter <[email protected]>
---
 drivers/scsi/ufs/ufshcd.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
drivers/scsi/ufs/ufshcd.c +2 -3
index f540b0cc253f..c96e36aab989 100644
@@ -8690,9 +8690,8 @@ static int ufshcd_suspend(struct ufs_hba *hba,
 enum ufs_pm_op pm_op)
 	ufshcd_clk_scaling_suspend(hba, true);
 
 	if (req_dev_pwr_mode == UFS_ACTIVE_PWR_MODE &&
-			req_link_state == UIC_LINK_ACTIVE_STATE) {
-		goto disable_clks;
-	}
+			req_link_state == UIC_LINK_ACTIVE_STATE)
+		goto enable_gating;
 
 	if ((req_dev_pwr_mode == hba->curr_dev_pwr_mode) &&
drivers/scsi/ufs/ufshcd.c +0 -1
 	    (req_link_state == hba->uic_link_state))
@@ -8754,7 +8753,6 @@ static int ufshcd_suspend(struct ufs_hba *hba,
 enum ufs_pm_op pm_op)
 	if (ret)
 		goto set_dev_active;
 
-disable_clks:
 	/*
 	 * Call vendor specific suspend callback. As these callbacks may 
Quoted 3 lines

access
* vendor specific host controller register space call them before
the

Hi Bart,

The change is unnecessary, ufshcd_suspend() is indeed keeping link alive even if we are disabling clocks. In __ufshcd_setup_clocks(), we have checks on clock sources so that we leave certain clock sources ON if the link is alive. And we have many of our customers tested the case rpm/spm_lvl == 0, it is working well. Please check below changes:

https://lore.kernel.org/patchwork/patch/1345336/
https://lore.kernel.org/patchwork/patch/1345337/

With this change, after suspend (rpm/spm_lvl == 0), leaving clock gating running is risky:
1. clock gating may run into concurrency with resume.
2. In ufshcd_resume(), we also have a ufshcd_release(), it will cause unbalanced usage of clock gating.

And it seems quite opposite from what you want - you want to keep link alive but you are leaving clock gating enabled, then when clock gating kicks start, it will put the link into hibern8, but not keep link alive.

Thanks,
Can Guo.

#2 lore raw
Quoted 19 lines · Can Guo

The change is unnecessary, ufshcd_suspend() is indeed keeping link alive even if
we are disabling clocks. In __ufshcd_setup_clocks(), we have checks on clock sources
so that we leave certain clock sources ON if the link is alive. And we have many
of our customers tested the case rpm/spm_lvl == 0, it is working well. Please check
below changes:

https://lore.kernel.org/patchwork/patch/1345336/
https://lore.kernel.org/patchwork/patch/1345337/

With this change, after suspend (rpm/spm_lvl == 0), leaving clock gating running is risky:
1. clock gating may run into concurrency with resume.
2. In ufshcd_resume(), we also have a ufshcd_release(), it will cause unbalanced usage of clock gating.

Quoted 3 lines

And it seems quite opposite from what you want - you want to keep link alive but you are leaving clock gating enabled, then when clock gating kicks start, it will put the link into hibern8, but not keep link alive.

Hi Can,

Thank you for the detailed feedback. I will drop this patch.

Bart.