On Wed, Jul 18, 2012 at 1:25 PM, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:
> On Tue, 17 Jul 2012, James Bottomley wrote:
>
>> On Tue, 2012-07-17 at 12:51 -0700, Linus Torvalds wrote:
>> > Wrong people cc'd, it looks like.
>> >
>> > Guys, commit a7a20d103994 ("sd: limit the scope of the async probe
>> > domain") is causing boot problems. It's timing-dependent and
>> > apparently sometimes works, which makes sense with that commit.
>> >
>> > However, it *should* have been fixed by commit 43a8d39d0137 ("fix
>> > async probe regression"), but Artem seems to report the problem even
>> > in -rc7.
>> >
>> > Comments?
>>
>> As far as I can tell, the fix should have worked. However, there are a
>> lot of assumptions in the async stuff that end up not being true in the
>> presence of separate async domains. We should be fixing it all here:
>>
>> http://git.kernel.org/?p=linux/kernel/git/jejb/scsi.git;a=shortlog;h=refs/heads/async
>
> Those commits haven't been merged into 3.5-rc7, right? So Artem isn't
> using them.
>
>> I've got to say, I don't understand the bug report. all of those
>> commits were about probing for devices. However, the screen shot
>>
>> https://bugzilla.kernel.org/attachment.cgi?id=75351
>>
>> shows the devices were found, it's the partition tables that weren't.
>> For us to see the message about sda's capacity, we're already in the
>> async code the commits were trying to synchronise with.
>
> That's understandable, given that those commits aren't present in
> Artem's kernel. Are they queued for -stable?
I did not flag them as such given the wider entanglements.
>
>> However, there are some missing messages: there's no partition table
>> print and no final
>>
>> sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
>> sdp->removable ? "removable " : "");
>>
>> So sd_probe_async() got stuck somewhere after the first
>> sd_revalidata_disk().
>
> No, it didn't get stuck. The problem is that the scsi_wait_scan module
> didn't wait for the async scanning to finish.
>
> And the reason for that is one you're already familiar with:
> CONFIG_SCSI_MOD=y.
I had forgotten that additional aspect. but I don't think that is the
root cause. scsi_wait_scan() is a nop in Artem's config given
CONFIG_SCSI_WAIT_SCAN=m. From what I have seen ahci typically ensures
that wait_for_device_probe() is all that is needed to guarantee
scanning is complete which commit a7a20d103994 ("sd: limit the scope
of the async probe domain") breaks, and which commit 43a8d39d0137
("[SCSI] fix async probe regression"), does not resolve because that
requires a call to scsi_complete_async_scans() to close the loop.
> Artem, if you change all your SCSI drivers to be modular rather than
> built-in, that ought to fix the problem. Alternatively, you can simply
> continue to use the "rootwait" option.
I think setting CONFIG_SCSI_WAIT_SCAN=y would also be a workaround, or
going forward with the pending async rework to make sd probe work once
again visible to async_synchronize_domain_full().
--
Dan
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
