Re: [Probably regression] Possible circular locking dependency from ATA in current mainline

[Tejun Heo <htejun@gmail.com>]

Hello, Rafael.

On Sun, Dec 26, 2010 at 11:16:26PM +0100, Rafael J. Wysocki wrote:
> [  319.281892] =======================================================
> [  319.281893] [ INFO: possible circular locking dependency detected ]
> [  319.281894] 2.6.37-rc7+ #16
> [  319.281895] -------------------------------------------------------
> [  319.281896] kworker/1:0/13333 is trying to acquire lock:
> [  319.281897]  (&bdev->bd_mutex){+.+.+.}, at: [<ffffffff811271a6>] revalidate_disk+0x56/0x90
> [  319.281902] 
> [  319.281903] but task is already holding lock:
> [  319.281903]  (&ap->scsi_scan_mutex){+.+...}, at: [<ffffffff81360420>] ata_scsi_dev_rescan+0x30/0x120

This dependency is true.

> [  319.281910] -> #3 (&ap->scsi_scan_mutex){+.+...}:
> [  319.281927] -> #2 ((&ap->scsi_rescan_task)){+.+...}:
> [  319.281937] -> #1 (events){+.+.+.}:
> [  319.281970] -> #0 (&bdev->bd_mutex){+.+.+.}:

But the above is bogus.  It's only linked through the system_wq.  The
current lockdep notation assumes single execution resource and
triggers spuriously in certain cases.  I'll think about how to relax
it.

Thank you.

-- 
tejun