Hi Jens,

I'm wondering about what is needed in ps3disk_remove(), to clean
up the request_queue (priv->queue) the driver uses.
ps3disk_remove() is called from the main device_shutdown()
routine via the ps3_system_bus remove routine.

The current code follows, and as can be seen, there is no check
or handling of the queue before blk_cleanup_queue() is called.
Can it be assumed that the queue is empty at this point,
or is something else needed here?

-Geoff

static int ps3disk_remove(struct ps3_system_bus_device *_dev)
{
	struct ps3_storage_device *dev = to_ps3_storage_device(&_dev->core);
	struct ps3disk_private *priv = dev->sbd.core.driver_data;

	mutex_lock(&ps3disk_mask_mutex);
	__clear_bit(priv->gendisk->first_minor / PS3DISK_MINORS,
		    &ps3disk_mask);
	mutex_unlock(&ps3disk_mask_mutex);

	del_gendisk(priv->gendisk);
	blk_cleanup_queue(priv->queue);
	put_disk(priv->gendisk);
	dev_notice(&dev->sbd.core, "Synchronizing disk cache\n");
	ps3disk_sync_cache(dev);
	ps3stor_teardown(dev);
	kfree(dev->bounce_buf);
	dev->bounce_buf = NULL;
	kfree(priv);
	dev->sbd.core.driver_data = NULL;
	return 0;
}







--

[ message continues ]

The queue should be empty, at least there should be no one else
submitting IO for the device given the contet in which it is called. But
I suppose we could have pending IO (given the kexec context you
explained to me in person), so the below patch could be needed. It may
actually be a good idea to put this into blk_cleanup_queue(), since we
should have process context there anyway.

diff --git a/drivers/block/ps3disk.c b/drivers/block/ps3disk.c
index d797e20..dce64d0 100644
--- a/drivers/block/ps3disk.c
+++ b/drivers/block/ps3disk.c
@@ -537,6 +537,8 @@ static int ps3disk_remove(struct ps3_system_bus_device *_dev)
 	struct ps3_storage_device *dev = to_ps3_storage_device(&_dev->core);
 	struct ps3disk_private *priv = dev->sbd.core.driver_data;
 
+	blk_sync_queue(priv->queue);
+
 	mutex_lock(&ps3disk_mask_mutex);
 	__clear_bit(priv->gendisk->first_minor / PS3DISK_MINORS,
 		    &ps3disk_mask);

-- 
Jens Axboe

--

[ message continues ]