Badness in as-iosched:1210

Badness in as-iosched:1210

[Zwane Mwaikambo]

I'm not sure wether you want this, it was during error handling from the 
HBA driver (source was disk error).

scsi1: ERROR on channel 0, id 3, lun 0, CDB: Read (10) 00 00 7f de 60 00 00 80 00 
Info fld=0x7fdeb2, Current sdd: sense key Medium Error
Additional sense: Unrecovered read error
end_request: I/O error, dev sdd, sector 8380032
Badness in as_add_request at drivers/block/as-iosched.c:1210
Call Trace:
 [<c0298c30>] as_add_request+0xd0/0x110
 [<c028fee5>] __elv_add_request+0x25/0x40
 [<c02e5a5f>] scsi_queue_next_request+0xaf/0x430
 [<c02938a7>] end_that_request_first+0x17/0x20
 [<c02e5e2f>] scsi_end_request+0x4f/0x130
 [<c02e30eb>] print_sense+0x1b/0x20
 [<c02e6391>] scsi_io_completion+0x2d1/0x460
 [<c0320071>] sd_rw_intr+0x41/0x160
 [<c02e077c>] scsi_finish_command+0x11c/0x160
 [<c030a7c7>] ahc_linux_isr+0x1d7/0x3a0
 [<c02e0606>] scsi_softirq+0xa6/0xd0
 [<c01268da>] do_softirq+0xca/0xd0
 [<c010bffa>] do_IRQ+0x1ba/0x210
 [<c0106f00>] default_idle+0x0/0x40
 [<c010a418>] common_interrupt+0x18/0x20
 [<c0106f00>] default_idle+0x0/0x40
 [<c0106f2e>] default_idle+0x2e/0x40
 [<c0106fba>] cpu_idle+0x3a/0x50
 [<c0105000>] rest_init+0x0/0x80
 [<c055e7dc>] start_kernel+0x16c/0x190

Re: Badness in as-iosched:1210

[Nick Piggin]

Zwane Mwaikambo wrote:

>I'm not sure wether you want this, it was during error handling from the 
>HBA driver (source was disk error).
>
>scsi1: ERROR on channel 0, id 3, lun 0, CDB: Read (10) 00 00 7f de 60 00 00 80 00 
>Info fld=0x7fdeb2, Current sdd: sense key Medium Error
>Additional sense: Unrecovered read error
>end_request: I/O error, dev sdd, sector 8380032
>Badness in as_add_request at drivers/block/as-iosched.c:1210
>
Thanks I'll have a look.

Re: Badness in as-iosched:1210

[Jens Axboe]

On Thu, Apr 24 2003, Nick Piggin wrote:
> Zwane Mwaikambo wrote:
> 
> >I'm not sure wether you want this, it was during error handling from the 
> >HBA driver (source was disk error).
> >
> >scsi1: ERROR on channel 0, id 3, lun 0, CDB: Read (10) 00 00 7f de 60 00 
> >00 80 00 Info fld=0x7fdeb2, Current sdd: sense key Medium Error
> >Additional sense: Unrecovered read error
> >end_request: I/O error, dev sdd, sector 8380032
> >Badness in as_add_request at drivers/block/as-iosched.c:1210
> >
> Thanks I'll have a look.

The debug check looks broken, request could have come from somewhere
else than the block pool.

-- 
Jens Axboe

Re: Badness in as-iosched:1210

[Nick Piggin]

Jens Axboe wrote:

>On Thu, Apr 24 2003, Nick Piggin wrote:
>
>>Zwane Mwaikambo wrote:
>>
>>
>>>I'm not sure wether you want this, it was during error handling from the 
>>>HBA driver (source was disk error).
>>>
>>>scsi1: ERROR on channel 0, id 3, lun 0, CDB: Read (10) 00 00 7f de 60 00 
>>>00 80 00 Info fld=0x7fdeb2, Current sdd: sense key Medium Error
>>>Additional sense: Unrecovered read error
>>>end_request: I/O error, dev sdd, sector 8380032
>>>Badness in as_add_request at drivers/block/as-iosched.c:1210
>>>
>>>
>>Thanks I'll have a look.
>>
>
>The debug check looks broken, request could have come from somewhere
>else than the block pool.
>
Thats right. I thought these requests would all be
!blk_fs_request()s though. It should be only the debug
checks which are wrong.

Re: Badness in as-iosched:1210

[Jens Axboe]

On Fri, Apr 25 2003, Nick Piggin wrote:
> Jens Axboe wrote:
> 
> >On Thu, Apr 24 2003, Nick Piggin wrote:
> >
> >>Zwane Mwaikambo wrote:
> >>
> >>
> >>>I'm not sure wether you want this, it was during error handling from the 
> >>>HBA driver (source was disk error).
> >>>
> >>>scsi1: ERROR on channel 0, id 3, lun 0, CDB: Read (10) 00 00 7f de 60 00 
> >>>00 80 00 Info fld=0x7fdeb2, Current sdd: sense key Medium Error
> >>>Additional sense: Unrecovered read error
> >>>end_request: I/O error, dev sdd, sector 8380032
> >>>Badness in as_add_request at drivers/block/as-iosched.c:1210
> >>>
> >>>
> >>Thanks I'll have a look.
> >>
> >
> >The debug check looks broken, request could have come from somewhere
> >else than the block pool.
> >
> Thats right. I thought these requests would all be
> !blk_fs_request()s though. It should be only the debug
> checks which are wrong.

Exactly, the rest looks ok, the debug trigger is wrong :). The
add_request() strategy is the entry point for all types of requests, not
just blk_fs_request()

-- 
Jens Axboe

Re: Badness in as-iosched:1210

[Nick Piggin]

Jens Axboe wrote:

>On Fri, Apr 25 2003, Nick Piggin wrote:
>
>>Jens Axboe wrote:
>>
>>
>>>On Thu, Apr 24 2003, Nick Piggin wrote:
>>>
>>>
>>>>Zwane Mwaikambo wrote:
>>>>
>>>>
>>>>
>>>>>I'm not sure wether you want this, it was during error handling from the 
>>>>>HBA driver (source was disk error).
>>>>>
>>>>>scsi1: ERROR on channel 0, id 3, lun 0, CDB: Read (10) 00 00 7f de 60 00 
>>>>>00 80 00 Info fld=0x7fdeb2, Current sdd: sense key Medium Error
>>>>>Additional sense: Unrecovered read error
>>>>>end_request: I/O error, dev sdd, sector 8380032
>>>>>Badness in as_add_request at drivers/block/as-iosched.c:1210
>>>>>
>>>>>
>>>>>
>>>>Thanks I'll have a look.
>>>>
>>>>
>>>The debug check looks broken, request could have come from somewhere
>>>else than the block pool.
>>>
>>>
>>Thats right. I thought these requests would all be
>>!blk_fs_request()s though. It should be only the debug
>>checks which are wrong.
>>
>
>Exactly, the rest looks ok, the debug trigger is wrong :). The
>add_request() strategy is the entry point for all types of requests, not
>just blk_fs_request()
>
No but it is as_insert_request which is that entry point. It
should only calls as_add_request for a blk_fs_request.

Re: Badness in as-iosched:1210

[Jens Axboe]

On Fri, Apr 25 2003, Nick Piggin wrote:
> Jens Axboe wrote:
> 
> >On Fri, Apr 25 2003, Nick Piggin wrote:
> >
> >>Jens Axboe wrote:
> >>
> >>
> >>>On Thu, Apr 24 2003, Nick Piggin wrote:
> >>>
> >>>
> >>>>Zwane Mwaikambo wrote:
> >>>>
> >>>>
> >>>>
> >>>>>I'm not sure wether you want this, it was during error handling from 
> >>>>>the HBA driver (source was disk error).
> >>>>>
> >>>>>scsi1: ERROR on channel 0, id 3, lun 0, CDB: Read (10) 00 00 7f de 60 
> >>>>>00 00 80 00 Info fld=0x7fdeb2, Current sdd: sense key Medium Error
> >>>>>Additional sense: Unrecovered read error
> >>>>>end_request: I/O error, dev sdd, sector 8380032
> >>>>>Badness in as_add_request at drivers/block/as-iosched.c:1210
> >>>>>
> >>>>>
> >>>>>
> >>>>Thanks I'll have a look.
> >>>>
> >>>>
> >>>The debug check looks broken, request could have come from somewhere
> >>>else than the block pool.
> >>>
> >>>
> >>Thats right. I thought these requests would all be
> >>!blk_fs_request()s though. It should be only the debug
> >>checks which are wrong.
> >>
> >
> >Exactly, the rest looks ok, the debug trigger is wrong :). The
> >add_request() strategy is the entry point for all types of requests, not
> >just blk_fs_request()
> >
> No but it is as_insert_request which is that entry point. It
> should only calls as_add_request for a blk_fs_request.

Oh I see, yes you are right, I should have looked closer (I just assumed
it was your elevator_add_req_fn, your naming is a bit funny :)

The debug check is still a bit silly, and there's nothing that stops it
from being wrong. So I'd still suggest to kill it.

-- 
Jens Axboe

Re: Badness in as-iosched:1210

[Nick Piggin]

Jens Axboe wrote:

>On Fri, Apr 25 2003, Nick Piggin wrote:
>
>>Jens Axboe wrote:
>>
>>
>>>On Fri, Apr 25 2003, Nick Piggin wrote:
>>>
>>>Exactly, the rest looks ok, the debug trigger is wrong :). The
>>>add_request() strategy is the entry point for all types of requests, not
>>>just blk_fs_request()
>>>
>>>
>>No but it is as_insert_request which is that entry point. It
>>should only calls as_add_request for a blk_fs_request.
>>
>
>Oh I see, yes you are right, I should have looked closer (I just assumed
>it was your elevator_add_req_fn, your naming is a bit funny :)
>
>The debug check is still a bit silly, and there's nothing that stops it
>from being wrong. So I'd still suggest to kill it.
>
Well the debug check is supposed to catch drivers which aren't
behaving nicely or if the reference counting is broken somewhere.

I suppose now that blk_put_request is being used to call
elv_completed_request then it should be pretty safe, right?

So I'll remove the debug stuff then