[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: 2.6.0-test9-mm3 - AIO test results



On Tue, Nov 18, 2003 at 03:47:53PM -0800, Daniel McNeil wrote:
> Suparna,
> 
> I was unable to reproduce the hang in io_submit() without your patch.
> I ran aiocp with 1k i/o size constantly for 2 hours and it never hung.
> 
> I re-ran with your patch with both as-iosched and deadline and both
> hung in io_submit().  aiocp would run a few times, but I put the
> aiocp in a while loop and it hung on the 1st or 2nd time.  It
> did get most of the way through copying the file before hanging.
> This is on a 2-proc to ide disks running ext3.
> 

Found one race ... not sure if its the one causing the hangs
you see. The attached patch is not a complete fix (there is one
other race to close), but it would be interesting to see if 
this makes any difference for you.

Regards
Suparna

-- 
Suparna Bhattacharya (suparna@in.ibm.com)
Linux Technology Center
IBM Software Labs, India

------------------------------------------------------
Don't access dio fields if its possible that the dio could 
already have been freed asynchronously during i/o completion.
Fixme: This still leaves a window between decrement of
bio_count and accessing dio->waiter during i/o completion 
wherein the dio could get freed by the submission path.


--- pure-mm3/fs/direct-io.c	2003-11-24 13:00:33.000000000 +0530
+++ linux-2.6.0-test9-mm3/fs/direct-io.c	2003-11-24 14:15:30.000000000 +0530
@@ -994,14 +995,17 @@
 	 * reflect the number of to-be-processed BIOs.
 	 */
 	if (dio->is_async) {
-		if (ret == 0)
-			ret = dio->result;
-		if (ret > 0 && dio->result < dio->size && rw == WRITE) {
+		int should_wait = 0;
+
+		if (dio->result < dio->size && rw == WRITE) {
 			dio->waiter = current;
+			should_wait = 1;
 		}
+		if (ret == 0)
+			ret = dio->result;
 		finished_one_bio(dio);		/* This can free the dio */
 		blk_run_queues();
-		if (dio->waiter) {
+		if (should_wait) {
 			/*
 			 * Wait for already issued I/O to drain out and
 			 * release its references to user-space pages
@@ -1013,7 +1017,7 @@
 				set_current_state(TASK_UNINTERRUPTIBLE);
 			}
 			set_current_state(TASK_RUNNING);
-			dio->waiter = NULL;
+			kfree(dio);
 		}
 	} else {
 		finished_one_bio(dio);
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"aart@kvack.org"> aart@kvack.org </a>