2009-06-28 11:05:26

by Daniel J Blueman

Subject: [2.6.30-rc1] lockdep false positive in DMA debug API...

With 2.6.30-rc1, I'm seeing the lockdep validator spinning it's wheels
[1] (also already seen [2]) on the DMA debug interface, with the
hashed locks.

A false positive needing annotation, or some changes to the locking are needed?

Daniel

--- [1]

atl1 0000:02:00.0: irq 28 for MSI/MSI-X

atl1 0000:02:00.0: eth0 link is up 1000 Mbps full duplex

# ifconfig eth0 mtu 9000
BUG: MAX_STACK_TRACE_ENTRIES too low!

turning off the locking correctness validator.

Pid: 2383, comm: ifconfig Not tainted 2.6.31-rc1-268c #2

Call Trace:

[<ffffffff8101ad9f>] ? save_stack_trace+0x2f/0x50

[<ffffffff8109c27e>] save_trace+0xbe/0xd0

[<ffffffff8109c2e8>] add_lock_to_list+0x58/0xf0

[<ffffffff8132d631>] ? add_dma_entry+0x41/0x80

[<ffffffff810a0678>] __lock_acquire+0xe88/0x1250

[<ffffffff810a0b5e>] lock_acquire+0x11e/0x170

[<ffffffff8132d631>] ? add_dma_entry+0x41/0x80

[<ffffffff81691230>] _spin_lock_irqsave+0x60/0xa0

[<ffffffff8132d631>] ? add_dma_entry+0x41/0x80

[<ffffffff8132d631>] add_dma_entry+0x41/0x80

[<ffffffff8132ed37>] debug_dma_map_page+0x177/0x1b0

[<ffffffff814b43e5>] atl1_alloc_rx_buffers+0x265/0x330

[<ffffffff814b52c5>] atl1_up+0x85/0x900

[<ffffffff814b5ce0>] atl1_change_mtu+0x80/0xd0

[<ffffffff81599182>] dev_set_mtu+0x52/0x90

[<ffffffff8159b1d2>] dev_ifsioc+0x272/0x2f0

[<ffffffff8159b5b3>] dev_ioctl+0x363/0x610

[<ffffffff81695182>] ? sub_preempt_count+0x142/0x150

[<ffffffff8158379d>] sock_ioctl+0x9d/0x2b0

[<ffffffff8116ddd6>] vfs_ioctl+0x36/0xb0

[<ffffffff8116df7a>] do_vfs_ioctl+0x8a/0x5f0

[<ffffffff8115ed7c>] ? fget_light+0x5c/0x150

[<ffffffff8116e52f>] sys_ioctl+0x4f/0xa0

[<ffffffff8100c4f2>] system_call_fastpath+0x16/0x1b

atl1 0000:02:00.0: irq 28 for MSI/MSI-X

atl1 0000:02:00.0: eth0 link is up 1000 Mbps full duplex

--- [2] http://lkml.org/lkml/2009/6/22/54
--
Daniel J Blueman

2009-06-29 13:15:15

by Joerg Roedel

Subject: Re: [2.6.30-rc1] lockdep false positive in DMA debug API...

On Sun, Jun 28, 2009 at 12:05:17PM +0100, Daniel J Blueman wrote:
> With 2.6.30-rc1, I'm seeing the lockdep validator spinning it's wheels
> [1] (also already seen [2]) on the DMA debug interface, with the
> hashed locks.

If you don't actually mean 2.6.3_1_-rc1 this is a known bug which was
fixed for 2.6.30-final.

Joerg

--
| Advanced Micro Devices GmbH
Operating | Karl-Hammerschmidt-Str. 34, 85609 Dornach bei M?nchen
System |
Research | Gesch?ftsf?hrer: Thomas M. McCoy, Giuliano Meroni
Center | Sitz: Dornach, Gemeinde Aschheim, Landkreis M?nchen
| Registergericht M?nchen, HRB Nr. 43632

2009-07-05 10:09:44

by Daniel J Blueman

Subject: Re: [2.6.30-rc1] lockdep false positive in DMA debug API...

On Mon, Jun 29, 2009 at 2:14 PM, Joerg Roedel<[email protected]> wrote:
> On Sun, Jun 28, 2009 at 12:05:17PM +0100, Daniel J Blueman wrote:
>> With 2.6.30-rc1, I'm seeing the lockdep validator spinning it's wheels
>> [1] (also already seen [2]) on the DMA debug interface, with the
>> hashed locks.
>
> If you don't actually mean 2.6.3_1_-rc1 this is a known bug which was
> fixed for 2.6.30-final.

It's in 2.6.30 proper, but is now fixed in 2.6.31-rc and 2.6.30.1 by
"dma-debug: Put all hash-chain locks into the same lock class":
http://groups.google.com/group/linux.kernel/browse_thread/thread/73ae3a0fc04185c8#

Thanks anyway!
Daniel
--
Daniel J Blueman