Commit 03cadd36 authored by Michael Holzheu's avatar Michael Holzheu Committed by Martin Schwidefsky

[S390] tape390: Fix request queue handling in block driver

When setting a channel attached tape online under Linux 2.6.31, the

"vol_id" process from udev hangs in sync_page():
 2 sync_page+144 [0x1dfaac]
 3 __wait_on_bit_lock+194 [0x58c23e]
 4 __lock_page+116 [0x1df9dc]
 5 truncate_inode_pages_range+728 [0x1ed7cc]
 6 __blkdev_put+244 [0x25f738]
 7 __fput+300 [0x229c4c]
 8 filp_close+122 [0x225a3a]

The reason for that is an error in the request queue handling. It can
happen that we fetch a request, but do not process it further because
the number of queued requests exceeds TAPEBLOCK_MIN_REQUEUE.
To fix this, we should call blk_peek_request() instead of
blk_fetch_request() in the while condition and fetch the request in
the loop body afterwards.

This bug was introduced with the patch "block: implement and enforce
request peek/start/fetch" (9934c8c0)
Signed-off-by: default avatarMichael Holzheu <holzheu@linux.vnet.ibm.com>
Signed-off-by: default avatarMartin Schwidefsky <schwidefsky@de.ibm.com>
parent 7874b1b6
...@@ -162,9 +162,10 @@ tapeblock_requeue(struct work_struct *work) { ...@@ -162,9 +162,10 @@ tapeblock_requeue(struct work_struct *work) {
spin_lock_irq(&device->blk_data.request_queue_lock); spin_lock_irq(&device->blk_data.request_queue_lock);
while ( while (
!blk_queue_plugged(queue) && !blk_queue_plugged(queue) &&
(req = blk_fetch_request(queue)) && blk_peek_request(queue) &&
nr_queued < TAPEBLOCK_MIN_REQUEUE nr_queued < TAPEBLOCK_MIN_REQUEUE
) { ) {
req = blk_fetch_request(queue);
if (rq_data_dir(req) == WRITE) { if (rq_data_dir(req) == WRITE) {
DBF_EVENT(1, "TBLOCK: Rejecting write request\n"); DBF_EVENT(1, "TBLOCK: Rejecting write request\n");
spin_unlock_irq(&device->blk_data.request_queue_lock); spin_unlock_irq(&device->blk_data.request_queue_lock);
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment