2020-11-10 09:07:25

by YangYang

Subject: [PATCH] scsi: ufs: Fix a bug in ufshcd_system_resume()

During system resume, ufshcd_system_resume() won't resume UFS host if
runtime suspended. After that, scsi_bus_resume() try to set SCSI host's
RPM status to RPM_ACTIVE, this will fail because UFS host's RPM status
is still RPM_SUSPENDED. So fix it.

scsi host0: scsi_runtime_suspend()
ufshcd_runtime_suspend()
scsi host0: scsi_bus_suspend()
ufshcd_system_suspend()
----------------------------------
ufshcd_pltfrm_resume()
scsi host0: scsi_bus_resume()
scsi host0: scsi_bus_resume_common()
scsi host0: pm_runtime_set_active(dev)

scsi host0: runtime PM trying to activate child device host0 but parent
(8800000.ufshc) is not active

Fixes: 57d104c153d3 ("ufs: add UFS power management support")
Signed-off-by: Yang Yang <[email protected]>
---
drivers/scsi/ufs/ufshcd.c | 15 +++++++++------
1 file changed, 9 insertions(+), 6 deletions(-)

diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index b8f573a02713..9e666e1ad58c 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -8767,11 +8767,7 @@ int ufshcd_system_resume(struct ufs_hba *hba)
if (!hba)
return -EINVAL;

- if (!hba->is_powered || pm_runtime_suspended(hba->dev))
- /*
- * Let the runtime resume take care of resuming
- * if runtime suspended.
- */
+ if (!hba->is_powered)
goto out;
else
ret = ufshcd_resume(hba, UFS_SYSTEM_PM);
@@ -8779,8 +8775,15 @@ int ufshcd_system_resume(struct ufs_hba *hba)
trace_ufshcd_system_resume(dev_name(hba->dev), ret,
ktime_to_us(ktime_sub(ktime_get(), start)),
hba->curr_dev_pwr_mode, hba->uic_link_state);
- if (!ret)
+ if (!ret) {
hba->is_sys_suspended = false;
+
+ if (pm_runtime_suspended(hba->dev)) {
+ pm_runtime_disable(hba->dev);
+ pm_runtime_set_active(hba->dev);
+ pm_runtime_enable(hba->dev);
+ }
+ }
return ret;
}
EXPORT_SYMBOL(ufshcd_system_resume);
--
2.17.1

2020-11-13 05:09:14

by Can Guo

Subject: Re: [PATCH] scsi: ufs: Fix a bug in ufshcd_system_resume()

On 2020-11-10 16:55, Yang Yang wrote:
> During system resume, ufshcd_system_resume() won't resume UFS host if
> runtime suspended. After that, scsi_bus_resume() try to set SCSI host's
> RPM status to RPM_ACTIVE, this will fail because UFS host's RPM status
> is still RPM_SUSPENDED. So fix it.
>
> scsi host0: scsi_runtime_suspend()
> ufshcd_runtime_suspend()
> scsi host0: scsi_bus_suspend()
> ufshcd_system_suspend()
> ----------------------------------
> ufshcd_pltfrm_resume()
> scsi host0: scsi_bus_resume()
> scsi host0: scsi_bus_resume_common()
> scsi host0: pm_runtime_set_active(dev)
>
> scsi host0: runtime PM trying to activate child device host0 but
> parent
> (8800000.ufshc) is not active
>
> Fixes: 57d104c153d3 ("ufs: add UFS power management support")
> Signed-off-by: Yang Yang <[email protected]>
> ---
> drivers/scsi/ufs/ufshcd.c | 15 +++++++++------
> 1 file changed, 9 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
> index b8f573a02713..9e666e1ad58c 100644
> --- a/drivers/scsi/ufs/ufshcd.c
> +++ b/drivers/scsi/ufs/ufshcd.c
> @@ -8767,11 +8767,7 @@ int ufshcd_system_resume(struct ufs_hba *hba)
> if (!hba)
> return -EINVAL;
>
> - if (!hba->is_powered || pm_runtime_suspended(hba->dev))
> - /*
> - * Let the runtime resume take care of resuming
> - * if runtime suspended.
> - */
> + if (!hba->is_powered)
> goto out;
> else
> ret = ufshcd_resume(hba, UFS_SYSTEM_PM);
> @@ -8779,8 +8775,15 @@ int ufshcd_system_resume(struct ufs_hba *hba)
> trace_ufshcd_system_resume(dev_name(hba->dev), ret,
> ktime_to_us(ktime_sub(ktime_get(), start)),
> hba->curr_dev_pwr_mode, hba->uic_link_state);
> - if (!ret)
> + if (!ret) {
> hba->is_sys_suspended = false;
> +
> + if (pm_runtime_suspended(hba->dev)) {
> + pm_runtime_disable(hba->dev);
> + pm_runtime_set_active(hba->dev);
> + pm_runtime_enable(hba->dev);
> + }
> + }
> return ret;
> }
> EXPORT_SYMBOL(ufshcd_system_resume);

It is designed like this - if hba is runtime suspend, do not resume it
during system resume, which is the so called deferred resume feature.
This is to leave the runtime PM management to block layer PM, which
is more power efficiency.

> scsi host0: runtime PM trying to activate child device host0 but parent
> (8800000.ufshc) is not active
The log is not harmful or fatal, maybe just annoying. Do you see real
problem with it? If yes, please share it here.

Thanks,

Can Guo.