linux-kernel - Re: Mysterious lockdep warning from the SCSI workqueue code

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-Id: <1198365406.30889.14.camel@lappy>
Date:	Sun, 23 Dec 2007 00:16:45 +0100
From:	Peter Zijlstra <peterz@...radead.org>
To:	"Rafael J. Wysocki" <rjw@...k.pl>
Cc:	James Bottomley <James.Bottomley@...eleye.com>,
	linux-kernel <linux-kernel@...r.kernel.org>,
	Arjan van de Ven <arjan@...radead.org>,
	Ingo Molnar <mingo@...e.hu>
Subject: Re: Mysterious lockdep warning from the SCSI workqueue code


On Sat, 2007-12-22 at 23:56 +0100, Rafael J. Wysocki wrote:
> On Saturday, 22 of December 2007, James Bottomley wrote:
> > I've no idea why this is occurring:
> > 
> > WARNING: at kernel/lockdep.c:700 look_up_lock_class()
> > Pid: 2068, comm: scsi_wq_3 Not tainted 2.6.24-rc6 #38
> >  [<c010535a>] show_trace_log_lvl+0x1a/0x30
> >  [<c0105ce2>] show_trace+0x12/0x20
> >  [<c010601c>] dump_stack+0x6c/0x80
> >  [<c014696d>] __lock_acquire+0x46d/0x10b0
> >  [<c0147628>] lock_acquire+0x78/0xa0
> >  [<c0136d18>] run_workqueue+0x128/0x1d0
> >  [<c013795c>] worker_thread+0x8c/0xf0
> >  [<c013ab32>] kthread+0x42/0x70
> >  [<c0104f47>] kernel_thread_helper+0x7/0x10
> >  =======================
> > 
> > it looks to be happening on the initial workqueue.c:run_workqueue()
> > 
> > 	spin_lock_irq(&cwq->lock);
> > 
> > It's caused by lockdep.c:look_up_lock_class()
> > 
> > 	WARN_ON_ONCE(class->name != lock->name);
> > 
> > but I'm not entirely sure what this actually means, not fully
> > understanding lockdep classes and how we acquire them in scsi.
> 
> I guess Ingo and/or Peter should have a look at this.

This makes me think of an invalid use of lockdep_set_class(). The
typical way to trigger this is:

struct lock_class_key my_keys[10];

struct my_obj *create_my_obj1()
{
	...
	spin_lock_init(&my_obj.lock);
	lockdep_set_class(&my_obj.lock, my_keys + foo);
	...
}

struct my_obj *create_my_obj2()
{
	...
	spin_lock_init(&my_obj.lock);
	lockdep_set_class(&my_obj.lock, my_keys + bar);
	...
}

This initializes &my_obj.lock to the same class (provides foo and bar
evaluate to the same), but give it a different name.

#define lockdep_set_class(lock, key) \
                lockdep_init_map(&(lock)->dep_map, #key, key, 0)

The name is #key, so: "my_keys + foo" vs "my_keys + bar"

The possible fixes are:
  - make key identical by changing the expression
  - use lockdep_set_class_and_name() and specify a
    consistent name by hand.



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