On 09/27/2012 10:42 PM, Alan Stern wrote:

On Thu, 27 Sep 2012, Aaron Lu wrote:

quoted

quoted

Moreover, I'd like to migrate SCSI drivers to the PM handling based on struct
dev_pm_ops eventually and your change is kind of going in the opposite
direction.  I don't know how much effort the migration is going to take,
though, so perhaps we can just make this change first.

Does the following change look OK?

diff --git a/drivers/scsi/scsi_pm.c b/drivers/scsi/scsi_pm.c
index dc0ad85..1fb7ccc 100644
--- a/drivers/scsi/scsi_pm.c
+++ b/drivers/scsi/scsi_pm.c

@@ -143,7 +143,15 @@ static int scsi_runtime_suspend(struct device *dev)
 
 	dev_dbg(dev, "scsi_runtime_suspend\n");
 	if (scsi_is_sdev_device(dev)) {
-		err = scsi_dev_type_suspend(dev, PMSG_AUTO_SUSPEND);
+		err = scsi_device_quiesce(to_scsi_device(dev));
+		if (err)
+			goto out;
+
+		err = pm_generic_runtime_suspend(dev);
+		if (!err)
+			goto out;
+
+		scsi_device_resume(to_scsi_device(dev));
 		if (err == -EAGAIN)
 			pm_schedule_suspend(dev, jiffies_to_msecs(
 				round_jiffies_up_relative(HZ/10)));

Maybe in the end it will be better, but for now this looks like 
unnecessary code duplication.  Basically you are copying 
scsi_dev_type_suspend() into scsi_runtime_suspend(), except that you 
omitted the debugging statement.

And I've used pm_generic_runtime_suspend :-)
That would allow me to write runtime callbacks of dev_pm_ops for
indivisual scsi drivers, like sr.

Thanks,
Aaron