3.5.0-rc5: BUG: soft lockup - CPU#0 stuck for 22s

[Christian Kujau]

On Mon, 2 Jul 2012 at 14:50, Benjamin Herrenschmidt wrote:
> > while trying to upgrade from 3.4.0 to 3.5.0-rc5 on this Powerbook G4 
> > (powerpc 32 bit), this happens during booting:
> > 
> > --------------
> > usb 2-1: new full-speed USB device number 4 using ohci_hcd
> > SCSI subsystem initialized
> > scsi0 : SBP-2 IEEE-1394
> > scsi1 : SBP-2 IEEE-1394
> > firewire_sbp2 fw1.0: logged in to LUN 0000 (0 retries)
> > scsi 0:0:0:0: Direct-Access     Ext Hard  Disk            PQ: 0 ANSI: 4
> 
> Interesting... I observed something roughly similar on a dual G4
> the other day associated with a 30s to 1mn pause during boot. RCU
> was complaining loudly.
> 
> In my case, it did continue booting normally, is that the case for you ?
> 
> Also if I compile the kernel without CONFIG_SMP, it did go away as well,
> do you observe that too ?
> 
> I don't have a spare cycle to investigate this problem this week I'm
> afraid, it might help if you could try a bisection though.

OK, after a git-bisect, the following commit has been identified as bad:

---------------------------
a7a20d103994fd760766e6c9d494daa569cbfe06 is the first bad commit
commit a7a20d103994fd760766e6c9d494daa569cbfe06
Author: Dan Williams <dan.j.williams at intel.com>
Date:   Thu Mar 22 17:05:11 2012 -0700

    [SCSI] sd: limit the scope of the async probe domain
---------------------------

Unfortunately it's not possible to "git revert" just this single commit:

------------
$ git revert a7a20d103994fd760766e6c9d494daa569cbfe06
error: could not revert a7a20d1... [SCSI] sd: limit the scope of the async probe domain
hint: after resolving the conflicts, mark the corrected paths  
hint: with 'git add <paths>' or 'git rm <paths>'
hint: and commit the result with 'git commit'
------------

I _guess_ this is probably due to:

------------
commit ea80dadec7a06889562b478cf0b87afbe62b7ac8
Author: James Bottomley <JBottomley at Parallels.com>
Date:   Wed Jun 6 14:54:13 2012 +0900

    [SCSI] Fix sd_probe_domain config problem
------------

The whole git-bisect log, dmesg, .config and so on is here:

  http://nerdbynature.de/bits/3.5.0-rc5/soft_lockup/

Anyone got an idea how to go from here?

Thanks,
Christian.
-- 
BOFH excuse #380:

Operators killed when huge stack of backup tapes fell over.