Re: + writeback-split-inode_wb_list_lock-into-bdi_writebacklist_lock-fix-f ix.patch added to -mm tree

Re: + writeback-split-inode_wb_list_lock-into-bdi_writebacklist_lock-fix-f ix.patch added to -mm tree

[Christoph Hellwig]

On Tue, May 03, 2011 at 01:57:44PM -0700, akpm@linux-foundation.org wrote:
>  	struct backing_dev_info *old = inode->i_data.backing_dev_info;
>  
> -	if (dst == old)
> +	if (dst == old)			/* deadlock avoidance */

That's not an overly useful comment.  It should be a proper block coment
documentation how that we could ever end up with the same bdi as
destination and source.

Which is something I wanted to ask Hugh anyway - do you have traces explaining
how this happens for you?

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: + writeback-split-inode_wb_list_lock-into-bdi_writebacklist_lock-fix-f ix.patch added to -mm tree

[Hugh Dickins]

On Wed, May 4, 2011 at 2:49 AM, Christoph Hellwig <hch@lst.de> wrote:
> On Tue, May 03, 2011 at 01:57:44PM -0700, akpm@linux-foundation.org wrote:
>>       struct backing_dev_info *old = inode->i_data.backing_dev_info;
>>
>> -     if (dst == old)
>> +     if (dst == old)                 /* deadlock avoidance */
>
> That's not an overly useful comment.  It should be a proper block coment
> documentation how that we could ever end up with the same bdi as
> destination and source.

I didn't put in a comment myself because it seemed obvious that we'd
want to avoid calling something named bdi_lock_two(a, b) when a and b
are the same; and I was expecting it to be obvious to you that
actually we could get here with a and b the same.  But apparently not.
 default_backing_dev_info?

>
> Which is something I wanted to ask Hugh anyway - do you have traces explaining
> how this happens for you?

Let's take out the patch and jot down the dmesg, omitting ? stale
lines and function offsets from the backtrace.

Something that may prove relevant: this is openSUSE 11.4, which cannot
boot my kernels unless I have CONFIG_DEVTMPFS=y; and I've set
CONFIG_DEVTMPFS_MOUNT=y too.  No initramfs.

...
VFS: Mounted root (ext2 filesystem) readonly on device 8:1.
devtmpfs: mounted
Freeing unused kernel memory: 352k freed
udev[162]: starting version 166
BUG: spinlock recursion on CPU#0, blkid/299
lock: 78690c30, .magic: dead4ead, .owner: blkid/299, .owner_cpu: 0
Pid: 299, comm: blkid Not tainted 2.6.39-rc5-mm1 #4
Call Trace:
spin_bug
do_raw_spin_lock
_raw_spin_lock_nested
bdi_lock_two
bdev_inode_switch_bdi
__blkdev_get
blkdev_get
blkdev_open
__dentry_open
nameidata_to_filp
do_last
path_openat
do_filp_open
do_sys_open
sys_open
sysenter_do_call

And indeed that "lock: 78690c30" falls inside my
default_backing_dev_info at 78690ae4.

Hugh

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>