[patch] elv_former_request reversion

[patch] elv_former_request reversion

[Andrew Morton]

This morning's fix for elv_former_request() is causing oopses all over the
place in the IO scheduler.

Jens, remember that I did try that fix a while ago, and the same happened.

I believe it has exposed a new problem at the __make_request/attempt_front_merge
level: if attempt_front_merge() actually succeeds, the wrong request gets freed
up in elv_merged_request().

It may be best to back this change out until it can be fixed up for real.


diff -puN drivers/block/elevator.c~deadline-hack drivers/block/elevator.c
--- 25/drivers/block/elevator.c~deadline-hack	2003-02-15 15:56:56.000000000 -0800
+++ 25-akpm/drivers/block/elevator.c	2003-02-15 15:57:09.000000000 -0800
@@ -399,7 +399,7 @@ struct request *elv_former_request(reque
 	elevator_t *e = &q->elevator;
 
 	if (e->elevator_former_req_fn)
-		return e->elevator_former_req_fn(q, rq);
+		return e->elevator_latter_req_fn(q, rq);
 
 	prev = rq->queuelist.prev;
 	if (prev != &q->queue_head && prev != &rq->queuelist)

_

Re: [patch] elv_former_request reversion

[Jens Axboe]

On Sat, Feb 15 2003, Andrew Morton wrote:
> 
> This morning's fix for elv_former_request() is causing oopses all over the
> place in the IO scheduler.
> 
> Jens, remember that I did try that fix a while ago, and the same happened.
> 
> I believe it has exposed a new problem at the
> __make_request/attempt_front_merge level: if attempt_front_merge()
> actually succeeds, the wrong request gets freed up in
> elv_merged_request().
> 
> It may be best to back this change out until it can be fixed up for
> real.

Yes agreed, I had forgotten about that point... Will fix.

-- 
Jens Axboe

Re: [patch] elv_former_request reversion

[Jens Axboe]

On Sun, Feb 16 2003, Jens Axboe wrote:
> On Sat, Feb 15 2003, Andrew Morton wrote:
> > 
> > This morning's fix for elv_former_request() is causing oopses all over the
> > place in the IO scheduler.
> > 
> > Jens, remember that I did try that fix a while ago, and the same happened.
> > 
> > I believe it has exposed a new problem at the
> > __make_request/attempt_front_merge level: if attempt_front_merge()
> > actually succeeds, the wrong request gets freed up in
> > elv_merged_request().
> > 
> > It may be best to back this change out until it can be fixed up for
> > real.
> 
> Yes agreed, I had forgotten about that point... Will fix.

Andrew, does this work for you?

===== drivers/block/deadline-iosched.c 1.14 vs edited =====
--- 1.14/drivers/block/deadline-iosched.c	Fri Feb 14 13:57:15 2003
+++ edited/drivers/block/deadline-iosched.c	Sun Feb 16 12:57:35 2003
@@ -297,6 +297,9 @@
 		deadline_del_drq_rb(dd, drq);
 	}
 
+	if (q->last_merge == &rq->queuelist)
+		q->last_merge = NULL;
+
 	list_del_init(&rq->queuelist);
 }
 
@@ -424,12 +427,7 @@
 {
 	request_queue_t *q = drq->request->q;
 
-	if (q->last_merge == &drq->request->queuelist)
-		q->last_merge = NULL;
-
-	deadline_del_drq_hash(drq);
-	deadline_del_drq_rb(dd, drq);
-	list_del_init(&drq->fifo);
+	deadline_remove_request(q, drq->request);
 	list_add_tail(&drq->request->queuelist, dd->dispatch);
 }
 
===== drivers/block/elevator.c 1.39 vs edited =====
--- 1.39/drivers/block/elevator.c	Sun Feb 16 00:57:09 2003
+++ edited/drivers/block/elevator.c	Sun Feb 16 11:32:35 2003
@@ -399,7 +399,7 @@
 	elevator_t *e = &q->elevator;
 
 	if (e->elevator_former_req_fn)
-		return e->elevator_latter_req_fn(q, rq);
+		return e->elevator_former_req_fn(q, rq);
 
 	prev = rq->queuelist.prev;
 	if (prev != &q->queue_head && prev != &rq->queuelist)

-- 
Jens Axboe

Re: [patch] elv_former_request reversion

[Andrew Morton]

Jens Axboe <axboe@suse.de> wrote:
>
> Andrew, does this work for you?

Yes.  I threw a bunch of nastiness at that on IDE and SCSI, 2-way and 4-way. 
No problems.  Thanks.

Re: [patch] elv_former_request reversion

[Anton Blanchard]

> Andrew, does this work for you?

Looks good on my ppc64 box too, it was hitting the BUG() when running
SDET before.

Anton