| Message ID | 20201224172010.10701-4-huobean@gmail.com |
|---|---|
| State | New |
| Headers | show |
| Series | Two changes of UFS sysfs | expand |
Hi Rafael Would you please help to review this patch assoociated with PM resume callback implementation? Original code that ufshcd_system/runtime_resume returns '0' in case resume failed due to the UFS device and UFS LINk are powered off(hba- >is_powered == false). In the patch 2/3 Add handling of the return value of pm_runtime_get_sync() is to fix race exists between shudown path and device access through sysfs node. Thanks, Bean On Thu, 2020-12-24 at 18:20 +0100, Bean Huo wrote: > From: Bean Huo <beanhuo@micron.com> > > After ufshcd_shutdown(), both UFS device and UFS LINk are powered > off, > return '0' will mislead the upper PM layer since the device has not > been > successfully resumed yet. This will let pm_runtime_get_sync() caller > mistakenly believe the device/LINK has been resumed, which leads to > request processing timeout that was en-queued later. > > To fix this, let ufshcd_system/runtimie_resume() return -EBUSY in > case of > hba->is_powered == false. > > Signed-off-by: Bean Huo <beanhuo@micron.com> > --- > drivers/scsi/ufs/ufshcd.c | 12 ++++++++---- > 1 file changed, 8 insertions(+), 4 deletions(-) > > diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c > index e221add25a7e..e1bcac51c01f 100644 > --- a/drivers/scsi/ufs/ufshcd.c > +++ b/drivers/scsi/ufs/ufshcd.c > @@ -8950,14 +8950,16 @@ int ufshcd_system_resume(struct ufs_hba *hba) > return -EINVAL; > } > > - if (!hba->is_powered || pm_runtime_suspended(hba->dev)) > + if (!hba->is_powered || pm_runtime_suspended(hba->dev)) { > /* > * Let the runtime resume take care of resuming > * if runtime suspended. > */ > + ret = -EBUSY; > goto out; > - else > + } else { > ret = ufshcd_resume(hba, UFS_SYSTEM_PM); > + } > out: > trace_ufshcd_system_resume(dev_name(hba->dev), ret, > ktime_to_us(ktime_sub(ktime_get(), start)), > @@ -9026,10 +9028,12 @@ int ufshcd_runtime_resume(struct ufs_hba > *hba) > if (!hba) > return -EINVAL; > > - if (!hba->is_powered) > + if (!hba->is_powered) { > + ret = -EBUSY; > goto out; > - else > + } else { > ret = ufshcd_resume(hba, UFS_RUNTIME_PM); > + } > out: > trace_ufshcd_runtime_resume(dev_name(hba->dev), ret, > ktime_to_us(ktime_sub(ktime_get(), start)),
On 2020-12-25 01:20, Bean Huo wrote: > From: Bean Huo <beanhuo@micron.com> > > After ufshcd_shutdown(), both UFS device and UFS LINk are powered off, > return '0' will mislead the upper PM layer since the device has not > been > successfully resumed yet. This will let pm_runtime_get_sync() caller > mistakenly believe the device/LINK has been resumed, which leads to > request processing timeout that was en-queued later. > > To fix this, let ufshcd_system/runtimie_resume() return -EBUSY in case > of > hba->is_powered == false. This change won't work as you expect... During/after shutdown, for UFS's case only, pm_runtime_get_sync(hba->dev) will most likely return 0, because it is already RUNTIME_ACTIVE, pm_runtime_get_sync() will directly return 0... meaning your change won't even be exercised. Check Stanley's change - https://lore.kernel.org/patchwork/patch/1341389/ Can Guo. > > Signed-off-by: Bean Huo <beanhuo@micron.com> > --- > drivers/scsi/ufs/ufshcd.c | 12 ++++++++---- > 1 file changed, 8 insertions(+), 4 deletions(-) > > diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c > index e221add25a7e..e1bcac51c01f 100644 > --- a/drivers/scsi/ufs/ufshcd.c > +++ b/drivers/scsi/ufs/ufshcd.c > @@ -8950,14 +8950,16 @@ int ufshcd_system_resume(struct ufs_hba *hba) > return -EINVAL; > } > > - if (!hba->is_powered || pm_runtime_suspended(hba->dev)) > + if (!hba->is_powered || pm_runtime_suspended(hba->dev)) { > /* > * Let the runtime resume take care of resuming > * if runtime suspended. > */ > + ret = -EBUSY; > goto out; > - else > + } else { > ret = ufshcd_resume(hba, UFS_SYSTEM_PM); > + } > out: > trace_ufshcd_system_resume(dev_name(hba->dev), ret, > ktime_to_us(ktime_sub(ktime_get(), start)), > @@ -9026,10 +9028,12 @@ int ufshcd_runtime_resume(struct ufs_hba *hba) > if (!hba) > return -EINVAL; > > - if (!hba->is_powered) > + if (!hba->is_powered) { > + ret = -EBUSY; > goto out; > - else > + } else { > ret = ufshcd_resume(hba, UFS_RUNTIME_PM); > + } > out: > trace_ufshcd_runtime_resume(dev_name(hba->dev), ret, > ktime_to_us(ktime_sub(ktime_get(), start)),
diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c index e221add25a7e..e1bcac51c01f 100644 --- a/drivers/scsi/ufs/ufshcd.c +++ b/drivers/scsi/ufs/ufshcd.c @@ -8950,14 +8950,16 @@ int ufshcd_system_resume(struct ufs_hba *hba) return -EINVAL; } - if (!hba->is_powered || pm_runtime_suspended(hba->dev)) + if (!hba->is_powered || pm_runtime_suspended(hba->dev)) { /* * Let the runtime resume take care of resuming * if runtime suspended. */ + ret = -EBUSY; goto out; - else + } else { ret = ufshcd_resume(hba, UFS_SYSTEM_PM); + } out: trace_ufshcd_system_resume(dev_name(hba->dev), ret, ktime_to_us(ktime_sub(ktime_get(), start)), @@ -9026,10 +9028,12 @@ int ufshcd_runtime_resume(struct ufs_hba *hba) if (!hba) return -EINVAL; - if (!hba->is_powered) + if (!hba->is_powered) { + ret = -EBUSY; goto out; - else + } else { ret = ufshcd_resume(hba, UFS_RUNTIME_PM); + } out: trace_ufshcd_runtime_resume(dev_name(hba->dev), ret, ktime_to_us(ktime_sub(ktime_get(), start)),