Re: [PATCH][noop-iosched] don't reuse a freed request

From: Jens Axboe
Date: Mon Oct 31 2005 - 02:41:14 EST


On Mon, Oct 31 2005, Arnaldo Carvalho de Melo wrote:
> Hi,
>
> I'm getting the oops below when trying to use qemu with a kernel
> built with just the noop iosched, I'm never had looked at this code before,
> so I did a quick hack that seems enough for my case.
>
> Ah, this is with a fairly recent git tree (today), haven't checked
> if it is present in 2.6.14.
>
> Best Regards,
>
> - Arnaldo
>
> Unable to handle kernel paging request at virtual address c5f20f60
> printing eip:
> c01b0ecd
> *pde = 00017067
> *pte = 05f20000
> Oops: 0000 [#1]
> DEBUG_PAGEALLOC
> Modules linked in:
> CPU: 0
> EIP: 0060:[<c01b0ecd>] Not tainted VLI
> EFLAGS: 00000046 (2.6.14acme)
> EIP is at elv_rq_merge_ok+0x15/0x7b
> eax: 00000014 ebx: c5f20f58 ecx: 000003f8 edx: 00000046
> esi: c12a5a90 edi: c5f20f58 ebp: c11658d0 esp: c11658c4
> ds: 007b es: 007b ss: 0068
> Process swapper (pid: 1, threadinfo=c1165000 task=c1164af0)
> Stack: c0251883 c5ecfe4c c5d688c0 c1165904 c01b0f48 c5f20f58 c12a5a90 00000000
> c5874000 c018c5e1 c5f15f24 0000002b 00000000 c5ecfe4c c5d688c0 c12a5a90
> c1165920 c01b128d c5f20f58 c12a5a90 000a568a 00000000 00000002 c1165960
> Call Trace:
> [<c0102a63>] show_stack+0x78/0x83
> [<c0102b88>] show_registers+0x100/0x167
> [<c0102d35>] die+0xcb/0x140
> [<c0234308>] do_page_fault+0x393/0x53a
> [<c0102777>] error_code+0x4f/0x54
> [<c01b0f48>] elv_try_merge+0x15/0x84
> [<c01b128d>] elv_merge+0x1d/0x4f
> [<c01b41d9>] __make_request+0xb2/0x425
> [<c01b46f9>] generic_make_request+0x125/0x137

Hrmpf, this looks really bad. Tejun, clearly there are still paths where
->last_rq isn't being cleared.

> --- a/drivers/block/ll_rw_blk.c
> +++ b/drivers/block/ll_rw_blk.c
> @@ -1787,6 +1787,9 @@ static inline void blk_free_request(requ
> if (rq->flags & REQ_ELVPRIV)
> elv_put_request(q, rq);
> mempool_free(rq, q->rq.rq_pool);
> +
> + if (rq == q->last_merge)
> + q->last_merge = NULL;
> }
>
> static inline struct request *

It's most likely a bug getting this far in the first place, but does it
fix things for you? I'll get on this asap.

--
Jens Axboe

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/