[PROBLEM] Badness in cfq_account_completion at drivers/block/cfq-iosched.c:916

[PROBLEM] Badness in cfq_account_completion at drivers/block/cfq-iosched.c:916

[Srihari Vijayaraghavan]

I see zillion of these error messages in vanilla
2.6.10:
Jan 11 16:10:13 linux kernel:  [<c0220ee0>]
scsi_end_request+0xa9/0xd6
Jan 11 16:10:13 linux kernel:  [<c0221181>]
scsi_io_completion+0xee/0x48e
Jan 11 16:10:13 linux kernel:  [<c028145f>]
_spin_unlock_irqrestore+0x5/0x6
Jan 11 16:10:13 linux kernel:  [<c01233d0>]
__mod_timer+0x122/0x160
Jan 11 16:10:13 linux kernel:  [<c01f187e>]
i8042_interrupt+0x4b/0x19f
Jan 11 16:10:13 linux kernel:  [<f881ca9c>]
sd_rw_intr+0x52/0x2a3 [sd_mod]
Jan 11 16:10:14 linux kernel:  [<c021d0e2>]
scsi_finish_command+0x17/0xb2
Jan 11 16:10:15 linux kernel:  [<c0123cc8>]
run_timer_softirq+0x175/0x17d
Jan 11 16:10:15 linux kernel:  [<c021d05e>]
scsi_softirq+0x9c/0xd3
Jan 11 16:10:15 linux kernel:  [<c011fde2>]
__do_softirq+0x62/0xce
Jan 11 16:10:15 linux kernel:  [<c0105112>]
do_softirq+0x42/0x49
Jan 11 16:10:15 linux kernel:  =======================
Jan 11 16:10:15 linux kernel:  [<c0105022>]
do_IRQ+0x42/0x54
Jan 11 16:10:16 linux kernel:  [<c010376e>]
common_interrupt+0x1a/0x20
Jan 11 16:10:16 linux kernel:  [<c010101e>]
default_idle+0x0/0x33
Jan 11 16:10:16 linux kernel:  [<c0101047>]
default_idle+0x29/0x33
Jan 11 16:10:16 linux kernel:  [<c01010b2>]
cpu_idle+0x2e/0x3e
Jan 11 16:10:16 linux kernel:  [<c03388ff>]
start_kernel+0x18d/0x1c9
Jan 11 16:10:16 linux kernel:  [<c0338345>]
unknown_bootoption+0x0/0x1bc
Jan 11 16:10:16 linux kernel: Badness in
cfq_account_completion at
drivers/block/cfq-iosched.c:916
Jan 11 16:10:16 linux kernel:  [<c020566a>]
cfq_completed_request+0xf6/0xfe
Jan 11 16:10:16 linux kernel:  [<c01ffbff>]
__blk_put_request+0x4d/0x90
Jan 11 16:10:16 linux kernel:  [<c0200e7e>]
end_that_request_last+0xb5/0xe3
Jan 11 16:10:16 linux kernel:  [<c0220ee0>]
scsi_end_request+0xa9/0xd6
Jan 11 16:10:16 linux kernel:  [<c0221181>]
scsi_io_completion+0xee/0x48e
Jan 11 16:10:16 linux kernel:  [<c028145f>]
_spin_unlock_irqrestore+0x5/0x6
Jan 11 16:10:16 linux kernel:  [<c01233d0>]
__mod_timer+0x122/0x160
Jan 11 16:10:16 linux kernel:  [<c01f187e>]
i8042_interrupt+0x4b/0x19f
Jan 11 16:10:16 linux kernel:  [<f881ca9c>]
sd_rw_intr+0x52/0x2a3 [sd_mod]
Jan 11 16:10:16 linux kernel:  [<c021d0e2>]
scsi_finish_command+0x17/0xb2
Jan 11 16:10:16 linux kernel:  [<c0123cc8>]
run_timer_softirq+0x175/0x17d
Jan 11 16:10:16 linux kernel:  [<c021d05e>]
scsi_softirq+0x9c/0xd3
Jan 11 16:10:16 linux kernel:  [<c011fde2>]
__do_softirq+0x62/0xce
Jan 11 16:10:16 linux kernel:  [<c0105112>]
do_softirq+0x42/0x49

It is an IBM x360, 2 Xeon (HT), 4 GB, Hardware RAID
70+ GB on IBM ServeRaid etc.

Thanks
Hari

PS: Please cc me in replies, as I am not subscribed to
LKML. I will see if I see same error messages in
-bk<latest>.


Find local movie times and trailers on Yahoo! Movies.
http://au.movies.yahoo.com

Re: [PROBLEM] Badness in cfq_account_completion at drivers/block/cfq-iosched.c:916

[Jens Axboe]

On Tue, Jan 11 2005, Srihari Vijayaraghavan wrote:
> I see zillion of these error messages in vanilla
> 2.6.10:
> Jan 11 16:10:13 linux kernel:  [<c0220ee0>]
> scsi_end_request+0xa9/0xd6

[snip]

Does this fix it?

===== drivers/block/cfq-iosched.c 1.17 vs edited =====
--- 1.17/drivers/block/cfq-iosched.c	2004-12-24 09:12:58 +01:00
+++ edited/drivers/block/cfq-iosched.c	2005-01-11 10:03:17 +01:00
@@ -622,8 +622,10 @@
 			cfq_sort_rr_list(cfqq, 0);
 		}
 
-		crq->accounted = 0;
-		cfqq->cfqd->rq_in_driver--;
+		if (crq->accounted) {
+			crq->accounted = 0;
+			cfqq->cfqd->rq_in_driver--;
+		}
 	}
 	list_add(&rq->queuelist, &q->queue_head);
 }

-- 
Jens Axboe

Re: [PROBLEM] Badness in cfq_account_completion at drivers/block/cfq-iosched.c:916

[Srihari Vijayaraghavan]

 --- Jens Axboe <axboe@suse.de> wrote: 
> ... 
> Does this fix it?
> 
> ===== drivers/block/cfq-iosched.c 1.17 vs edited
> =====
> --- 1.17/drivers/block/cfq-iosched.c	2004-12-24
> 09:12:58 +01:00
> +++ edited/drivers/block/cfq-iosched.c	2005-01-11
> 10:03:17 +01:00
> @@ -622,8 +622,10 @@
>  			cfq_sort_rr_list(cfqq, 0);
>  		}
>  
> -		crq->accounted = 0;
> -		cfqq->cfqd->rq_in_driver--;
> +		if (crq->accounted) {
> +			crq->accounted = 0;
> +			cfqq->cfqd->rq_in_driver--;
> +		}
>  	}
>  	list_add(&rq->queuelist, &q->queue_head);
>  }

Yes, it does fix the problem with cfq, and the system
works fine. No more "Badness" error messages. Thanks
Jens.

While you are at it, is this acceptable?:
--- test/drivers/block/elevator.c.orig	2005-01-11
15:47:07.000000000 +1100
+++ test/drivers/block/elevator.c	2005-01-12
12:16:19.365813400 +1100
@@ -170,8 +170,6 @@
 #else
 #error "You must build at least 1 IO scheduler into
the kernel"
 #endif
-	printk(KERN_INFO "elevator: using %s as default io
scheduler\n",
-							chosen_elevator);
 }
 
 static int __init elevator_setup(char *str)
@@ -516,6 +514,9 @@
 	spin_unlock_irq(&elv_list_lock);
 
 	printk(KERN_INFO "io scheduler %s registered\n",
e->elevator_name);
+	if (!strcmp(e->elevator_name, chosen_elevator))
+		printk(KERN_INFO "elevator: using %s as default io
scheduler\n",
+                                                     
  e->elevator_name);
 	return 0;
 }
 EXPORT_SYMBOL_GPL(elv_register);

It has an advantage of working even when one uses
"elevator=" kernel boot parameter. If it is wrong
completely, I am sorry about it.

Thank you.
Hari

PS: I am using web email interface, if things appear
funny, sorry about that.


Find local movie times and trailers on Yahoo! Movies.
http://au.movies.yahoo.com

Re: [PROBLEM] Badness in cfq_account_completion at drivers/block/cfq-iosched.c:916

[Jens Axboe]

On Wed, Jan 12 2005, Srihari Vijayaraghavan wrote:
>  --- Jens Axboe <axboe@suse.de> wrote: 
> > ... 
> > Does this fix it?
> > 
> > ===== drivers/block/cfq-iosched.c 1.17 vs edited
> > =====
> > --- 1.17/drivers/block/cfq-iosched.c	2004-12-24
> > 09:12:58 +01:00
> > +++ edited/drivers/block/cfq-iosched.c	2005-01-11
> > 10:03:17 +01:00
> > @@ -622,8 +622,10 @@
> >  			cfq_sort_rr_list(cfqq, 0);
> >  		}
> >  
> > -		crq->accounted = 0;
> > -		cfqq->cfqd->rq_in_driver--;
> > +		if (crq->accounted) {
> > +			crq->accounted = 0;
> > +			cfqq->cfqd->rq_in_driver--;
> > +		}
> >  	}
> >  	list_add(&rq->queuelist, &q->queue_head);
> >  }
> 
> Yes, it does fix the problem with cfq, and the system
> works fine. No more "Badness" error messages. Thanks
> Jens.

Super, thanks.

> While you are at it, is this acceptable?:
> --- test/drivers/block/elevator.c.orig	2005-01-11
> 15:47:07.000000000 +1100
> +++ test/drivers/block/elevator.c	2005-01-12
> 12:16:19.365813400 +1100
> @@ -170,8 +170,6 @@
>  #else
>  #error "You must build at least 1 IO scheduler into
> the kernel"
>  #endif
> -	printk(KERN_INFO "elevator: using %s as default io
> scheduler\n",
> -							chosen_elevator);
>  }
>  
>  static int __init elevator_setup(char *str)
> @@ -516,6 +514,9 @@
>  	spin_unlock_irq(&elv_list_lock);
>  
>  	printk(KERN_INFO "io scheduler %s registered\n",
> e->elevator_name);
> +	if (!strcmp(e->elevator_name, chosen_elevator))
> +		printk(KERN_INFO "elevator: using %s as default io
> scheduler\n",
> +                                                     
>   e->elevator_name);
>  	return 0;
>  }
>  EXPORT_SYMBOL_GPL(elv_register);
> 
> It has an advantage of working even when one uses
> "elevator=" kernel boot parameter. If it is wrong
> completely, I am sorry about it.

Yes that's a good idea, perhaps just adding "(default)" at the end of
the default io scheduler is better so we save that extra line.

-- 
Jens Axboe