There might be a racing issue between UFS shutdown and runtime resume
flow described as below,
Thread #1: In UFS shutdown flow with ufshcd_shutdown() is running.
Thread #2: In UFS runtime-resume flow which invokes
ufshcd_runtime_resume() because UFS was in runtime-suspended
state while an I/O request was issued.
In this scenario, racing may happen and possibly lead to system hang
if Thread #2 accesses UFS host's register map after host's resource,
like power or clocks, are disabled by Thread #1.
To avoid this racing, use PM public function pm_runtime_get_sync() in
shutdown flow instead of internal function ufshcd_runtime_resume() for
consolidated control of RPM status.
One concern is that pm_runtime_get_sync() may be better paired with
pm_runtime_put_sync(), however shutdown could be one-way path thus the
pairing is not required.
Signed-off-by: Stanley Chu <stanley....@mediatek.com>
Signed-off-by: Peter Wang <peter.w...@mediatek.com>
---
drivers/scsi/ufs/ufshcd.c | 7 ++-----
1 file changed, 2 insertions(+), 5 deletions(-)
diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index a208589426b1..cce7303f8653 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -8095,11 +8095,8 @@ int ufshcd_shutdown(struct ufs_hba *hba)
if (ufshcd_is_ufs_dev_poweroff(hba) && ufshcd_is_link_off(hba))
goto out;
- if (pm_runtime_suspended(hba->dev)) {
- ret = ufshcd_runtime_resume(hba);
- if (ret)
- goto out;
- }
+ if (pm_runtime_get_sync(hba->dev) < 0)
+ goto out;
ret = ufshcd_suspend(hba, UFS_SHUTDOWN_PM);
out:
--
2.18.0
