Re: scsi_scan: WARNING: at include/linux/blkdev.h:431 blk_queue_init_tags+0x107/0x120()

["Jesper Juhl" <jesper.juhl@gmail.com>]

Hi,

2008/5/12 James Bottomley <James.Bottomley@hansenpartnership.com>:
> On Thu, 2008-05-08 at 12:55 +0900, FUJITA Tomonori wrote:
>> On Wed, 07 May 2008 17:40:26 -0500
>> James Bottomley <James.Bottomley@HansenPartnership.com> wrote:
>> > > The problem is that the commit 75ad23b expects that we hold the queue
>> > > lock for __blk_queue_free_tags, blk_queue_free_tags and
>> > > blk_queue_init_tags but we haven't.
>> > >
>> > > The simple fix is using queue_flag_set/clear_unlocked for them, then
>> > > it should work as before. However, it would be better to hold the
>> > > queue lock for blk_queue_free_tags and blk_queue_init_tags (we can
>> > > hold the queue lock in scsi_activate_tcq and scsi_deactivate_tcq).
>> >
>> > So is this the fix that everyone agrees on?  And if so, whose tree is it
>> > going through (I tend to think block, since the original breakage came
>> > from the block tree).
>>
>> The patch is fine from the perspective of the SCSI mid-layer. But it
>> would be not from the perspective of the block layer. For example,
>> blk_queue_init_tags is expected to be called with holding the queue
>> lock (since it could call blk_queue_resize_tags internally) though
>> only the SCSI mid-layer uses those APIs for now.
>
> There are two cases for blk_queue_init_tags; one is for a shared tag map
> and the other is for individual tag maps.  If you look at the code and
> its use, blk_queue_resize_tags is *only* called for the individual tag
> map case (or for the first caller in a shared tag map case).
>
> The blk_queue_resize_tags() also returns -EBUSY on shared tag maps,
> except the first one; which really exposes a bug in SCSI: we want to set
> the shared tag map to some host specific depth and then carve off pieces
> of it in the devices, so we don't want to adjust the queue down to
> whatever the first seen device wants, and we don't want all other
> devices to fail to adjust the scsi_queue depth the mid-layer uses.
>
> This should fix up scsi.
>
> James
>
> ---
> diff --git a/drivers/scsi/scsi.c b/drivers/scsi/scsi.c
> index 110e776..cc0af0f 100644
> --- a/drivers/scsi/scsi.c
> +++ b/drivers/scsi/scsi.c
> @@ -893,9 +893,11 @@ void scsi_adjust_queue_depth(struct scsi_device *sdev, int tagged, int tags)
>
>        spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
>
> -       /* Check to see if the queue is managed by the block layer.
> -        * If it is, and we fail to adjust the depth, exit. */
> +       /* Check to see if the queue is managed by the block layer,
> +        * and that we don't have a shared tag map.  If so, and we
> +        * fail to adjust the depth, exit. */
>        if (blk_queue_tagged(sdev->request_queue) &&
> +           !sdev->host->bqt &&
>            blk_queue_resize_tags(sdev->request_queue, tags) != 0)
>                goto out;
>

I just tested this patch by applying it to the tip of Linus' git tree,
recompiled, rebooted and everything seems to be fine.
>From where I'm sitting this looks good.  Thanks James.

Tested-by: Jesper Juhl <jesper.juhl@gmail.com>

-- 
Jesper Juhl <jesper.juhl@gmail.com>
Don't top-post http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please http://www.expita.com/nomime.html