Re: 2.6.0-test5-mm3 as-iosched Oops running dbt2 workload

[Mary Edie Meredith <maryedie@osdl.org>]

I tried dbt2 cached run again on the same 8way project machine on
2.6.0-test5-mm4 and got the following similar panic.  Maybe it will give
more clues:


------------[ cut here ]------------
kernel BUG at drivers/block/as-iosched.c:1091!
invalid operand: 0000 [#1]
SMP 
CPU:    3
EIP:    0060:[<c0228b38>]    Not tainted VLI
EFLAGS: 00010046
EIP is at as_move_to_dispatch+0x1b8/0x1d0
eax: 00000000   ebx: 00000000   ecx: 00000000   edx: ffffff41
esi: 00000000   edi: f7a228c0   ebp: f6e865e0   esp: e5923bbc
ds: 007b   es: 007b   ss: 0068
Process kernel (pid: 1640, threadinfo=e5922000 task=d16dc670)
Stack: f27e4b40 00000003 00000000 0020c7f0 00000000 f7a228c0 00000000
00000001 
       f6e865e0 c0228c8b f7a228c0 f6e865e0 00000040 c0223a3f f7818800
f7a228c0 
       00000000 e5922000 f74566e0 c0228e78 f7a228c0 f7818800 f7818800
c0220816 
Call Trace:
 [<c0228c8b>] as_dispatch_request+0x13b/0x2f0
 [<c0223a3f>] generic_make_request+0x16f/0x1f0
 [<c0228e78>] as_next_request+0x38/0x50
 [<c0220816>] elv_next_request+0x16/0x110
 [<c022234a>] generic_unplug_device+0x5a/0x70
 [<c02224b6>] blk_run_queues+0x86/0xa0
 [<c017dc66>] dio_await_one+0x86/0xb0
 [<c017dd9b>] dio_await_completion+0x2b/0x60
 [<c017eb93>] direct_io_worker+0x343/0x480
 [<c017ee4f>] blockdev_direct_IO+0x17f/0x1c7
 [<c0162b30>] blkdev_get_blocks+0x0/0xa0
 [<c0162c3a>] blkdev_direct_IO+0x6a/0x80
 [<c0162b30>] blkdev_get_blocks+0x0/0xa0
 [<c013f6b1>] generic_file_direct_IO+0xe1/0x110
 [<c013ed06>] __generic_file_aio_write_nolock+0x8f6/0xbd0
 [<c0162b30>] blkdev_get_blocks+0x0/0xa0
 [<c0142b2f>] wait_on_page_range+0x2f/0x40
 [<c0142ad0>] page_waiter+0x0/0x30
 [<c0142e28>] sync_page_range_nolock+0xc8/0xce
 [<c013f09c>] generic_file_aio_write_nolock+0xbc/0xd0
 [<c013f212>] generic_file_write_nolock+0xa2/0xc0
 [<c013d632>] generic_file_read+0xb2/0xd0
 [<c0121330>] autoremove_wake_function+0x0/0x50
 [<c0121330>] autoremove_wake_function+0x0/0x50
 [<c0136650>] nanosleep_wake_up+0x0/0x10
 [<c0215067>] raw_file_write+0x37/0x40
 [<c015ac8e>] vfs_write+0xbe/0x130
 [<c015adb2>] sys_write+0x42/0x70
 [<c030d2af>] syscall_call+0x7/0xb
 [<c030007b>] rpc_release_task+0x1db/0x250

Code: f6 e8 dd b7 ff ff 89 b7 c4 00 00 00 e9 ba fe ff ff 8d 45 1c 89 44
24 04 8d 87 c4 00 00 00 89 04 24 e8 3d b9 ff ff e9 ab fe ff ff <0f> 0b
43 04 3d ec 32 c0 e9 57 fe ff f
f 8d 74 26 00 8d bc 27 00 



On Fri, 2003-09-19 at 18:24, Nick Piggin wrote:
> Sigh. Sorry, I'm an idiot...
> 
> If a request is merged with another, it sometimes has to be repositioned
> on the rbtree - you just do a delete then an add. This is a quite
> uncommon case though.
> 
> I changed the way adding works, so collisions must be handled by the
> caller instead of being dumbly fixed by the add routine. Unfortunately
> the uncommon callers weren't handling it properly. Try this please.
> 
> 
> 
> Dave Olien wrote:
> 
> >Andrew,
> >
> >Attached is console output containing a stack trace from an Oops, followed
> >by a Fatal exception, and LOTS of APIC errors.  The machine was hung,
> >printing APIC error messages forever.
> >
> >This looks like another as-iosched problem.  So, I'm copying Nick Piggin
> >on this email.  But the Fatal exception and APIC errors following
> >that are a mystery to me.
> >
> >Mary encountered this running the sapdb dbt2 cached database workload on her
> >project machine.  The project machine was running 2.6.0-test5-mm3.
> >This same test passes on the stp machines.  But Mary's project machine
> >has more processors, and more disks, and a different disk controller type.
> >
> >At this stage, the database has gotten past the database restore phase.
> >That's where it was failing prior to last night's mm3 patch.  Now, the
> >database itself has been running for about 30 minutes.  In the cached
> >case, much of that first 30 minutes is spent loading the cache.
> >
> >This Oops seems to have occurred at about the time the database is
> >transitioning to using its cache.  Most of the I/O after this point
> >is to the log, doing LOTS of sequential writes, with the occasional
> >random read/write.
> >
> >Since this machine has more processors, it's doing transactions
> >more quickly than the same workload on STP machines.  So the log write
> >traffic is probably a lot heavier.
> >
> >
> >------------[ cut here ]------------
> >kernel BUG at drivers/block/as-iosched.c:1230!
> >invalid operand: 0000 [#1]
> >SMP 
> >CPU:    3
> >EIP:    0060:[<c0228146>]    Not tainted VLI
> >EFLAGS: 00010046
> >EIP is at as_dispatch_request+0x236/0x2f0
> >eax: 00000000   ebx: f7a451a0   ecx: 00000000   edx: 00000000
> >esi: 00000000   edi: 00000001   ebp: 00000000   esp: f5f67ef8
> >ds: 007b   es: 007b   ss: 0068
> >Process kernel (pid: 2283, threadinfo=f5f66000 task=f62760a0)
> >Stack: f7a451a0 f5900820 f7a28000 f7a02600 c0235c23 f7a451a0 00000000 f7836000 
> >       f5f67fc4 c0228238 f7a451a0 f7a08420 f7836000 c021fbb6 f7836000 f77e0000 
> >       f7a28000 f7a08420 f7a28000 c023a128 f7836000 f5f02de0 f77e0090 0000000a 
> >Call Trace:
> > [<c0235c23>] DAC960_BA_QueueCommand+0x43/0xb0
> > [<c0228238>] as_next_request+0x38/0x50
> > [<c021fbb6>] elv_next_request+0x16/0x110
> > [<c023a128>] DAC960_ProcessRequest+0x38/0x190
> > [<c023cd40>] DAC960_BA_InterruptHandler+0x90/0xb0
> > [<c010e899>] handle_IRQ_event+0x49/0x80
> > [<c010ec0f>] do_IRQ+0x9f/0x150
> > [<c030cb0c>] common_interrupt+0x18/0x20
> > [<c030007b>] rpcauth_free_credcache+0xbb/0x100
> >
> >Code: 43 50 00 00 00 00 8b 43 54 8b 6b 1c c7 43 5c 00 00 00 00 89 43 58 e9 95 fe ff ff c7 43 48 01 00 00 00 c7 43 4c 00 00 00 00 eb d4 <0f> 0b ce 04 fd da 32 c0 eb c7 8b 43 50 e9 61 ff ff ff 0f 0b b6 
> > <0>Kernel panic: Fatal exception in interrupt
> >In interrupt handler - not syncing
> > <6>APIC error on CPU3: 00(08)
> >APIC error on CPU3: 08(08)
> >APIC error on CPU3: 08(08)
> >APIC error on CPU3: 08(08)
> >APIC error on CPU3: 08(08)
> >APIC error on CPU3: 08(08)
> >APIC error on CPU3: 08(08)
> >APIC error on CPU3: 08(08)
> >APIC error on CPU3: 08(08)
> >APIC error on CPU3: 08(08)
> >APIC error on CPU3: 08(08)
> >APIC error on CPU3: 08(08)
> >APIC error on CPU3: 08(08)
> >APIC error on CPU3: 08(08)
> >-
> >To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> >the body of a message to majordomo@vger.kernel.org
> >More majordomo info at  http://vger.kernel.org/majordomo-info.html
> >Please read the FAQ at  http://www.tux.org/lkml/
> >
> >
> >  
> >
> 
> ______________________________________________________________________
> 
>  linux-2.6-npiggin/drivers/block/as-iosched.c |   21 +++++++++++++++------
>  1 files changed, 15 insertions(+), 6 deletions(-)
> 
> diff -puN drivers/block/as-iosched.c~as-oops-fix drivers/block/as-iosched.c
> --- linux-2.6/drivers/block/as-iosched.c~as-oops-fix	2003-09-20 11:13:26.000000000 +1000
> +++ linux-2.6-npiggin/drivers/block/as-iosched.c	2003-09-20 11:22:55.000000000 +1000
> @@ -1303,7 +1303,7 @@ static struct request *as_next_request(r
>   * Add arq to a list behind alias
>   */
>  static inline void
> -as_add_aliased_request(struct as_rq *arq, struct as_rq *alias)
> +as_add_aliased_request(struct as_data *ad, struct as_rq *arq, struct as_rq *alias)
>  {
>  	/*
>  	 * Another request with the same start sector on the rbtree.
> @@ -1312,6 +1312,11 @@ as_add_aliased_request(struct as_rq *arq
>  	 */
>  	list_add_tail(&arq->request->queuelist,	&alias->request->queuelist);
>  
> +	/*
> +	 * Don't want to have to handle merges.
> +	 */
> +	as_remove_merge_hints(ad->q, arq);
> +
>  }
>  
>  /*
> @@ -1353,7 +1358,7 @@ static void as_add_request(struct as_dat
>  		as_update_arq(ad, arq); /* keep state machine up to date */
>  
>  	} else {
> -		as_add_aliased_request(arq, alias);
> +		as_add_aliased_request(ad, arq, alias);
>  		/*
>  		 * have we been anticipating this request?
>  		 * or does it come from the same process as the one we are
> @@ -1553,8 +1558,10 @@ static void as_merged_request(request_qu
>  		 * currently don't bother. Ditto the next function.
>  		 */
>  		as_del_arq_rb(ad, arq);
> -		if ((alias = as_add_arq_rb(ad, arq)) )
> -			as_add_aliased_request(arq, alias);
> +		if ((alias = as_add_arq_rb(ad, arq)) ) {
> +			list_del_init(&arq->fifo);
> +			as_add_aliased_request(ad, arq, alias);
> +		}
>  		/*
>  		 * Note! At this stage of this and the next function, our next
>  		 * request may not be optimal - eg the request may have "grown"
> @@ -1586,8 +1593,10 @@ as_merged_requests(request_queue_t *q, s
>  	if (rq_rb_key(req) != arq->rb_key) {
>  		struct as_rq *alias;
>  		as_del_arq_rb(ad, arq);
> -		if ((alias = as_add_arq_rb(ad, arq)) )
> -			as_add_aliased_request(arq, alias);
> +		if ((alias = as_add_arq_rb(ad, arq)) ) {
> +			list_del_init(&arq->fifo);
> +			as_add_aliased_request(ad, arq, alias);
> +		}
>  	}
>  
>  	/*
> 
> _
-- 
Mary Edie Meredith <maryedie@osdl.org>
Open Source Development Lab