On 6/2/23 01:49, Kai-Heng Feng wrote:
diff --git a/drivers/scsi/scsi_scan.c b/drivers/scsi/scsi_scan.c
index d217be323cc6..092f37464101 100644
--- a/drivers/scsi/scsi_scan.c
+++ b/drivers/scsi/scsi_scan.c
@@ -1621,6 +1621,11 @@ void scsi_rescan_device(struct device *dev)
{
struct scsi_device *sdev = to_scsi_device(dev);
+#ifdef CONFIG_PM_SLEEP
+ if (dev->power.is_suspended)
+ wait_for_completion(&dev->power.completion);
+#endif
+
device_lock(dev);
scsi_attach_vpd(sdev);
Directly accessing dev->power.completion from the SCSI core seems like a
layering violation to me. Isn't this an object that should only be
accessed directly by the device driver power management core?
Additionally, what guarantees that the desired power state has been
reached after wait_for_completion(&dev->power.completion) has finished?
I think we need another solution. The device_lock() and device_unlock()
calls have been introduced by commit e27829dc92e5 ("scsi: serialize
->rescan against ->remove"). I think there are other ways to serialize
scsi_rescan_device() against scsi_remove_device(), e.g. via
host->scan_mutex. Is this something that has been considered?
Thanks,
Bart.
