[bug] oops in linux-3.8-rc2

[bug] oops in linux-3.8-rc2

[Tom Gundersen]

Hi guys,

I ran into a bug today with 3.8-rc2 [0], following an "make
modules_install && make install" of the kernel.

I have been using the same kernel without problems for some days, the
only change immediately before the oops was that I switched from plain
partitions to using multi-device and subvolumes.

Is there any more info I could give?

Cheers,

Tom

[0]: <https://plus.google.com/114015603831160344127/posts/fzEYnF64NkZ>
(I'm sorry about the quality of the picture, hopefully it is still
useful).

Re: [bug] oops in linux-3.8-rc2

[David Sterba]

Hi,

On Mon, Jan 07, 2013 at 01:51:21AM +0100, Tom Gundersen wrote:
> I ran into a bug today with 3.8-rc2 [0], following an "make
> modules_install && make install" of the kernel.
> 
> I have been using the same kernel without problems for some days, the
> only change immediately before the oops was that I switched from plain
> partitions to using multi-device and subvolumes.
> 
> [0]: <https://plus.google.com/114015603831160344127/posts/fzEYnF64NkZ>

BUG at mm/page-writeback.c:2164

2160 int clear_page_dirty_for_io(struct page *page)
2161 {
2162         struct address_space *mapping = page_mapping(page);
2163
2164         BUG_ON(!PageLocked(page));

btrfs on the stack is only __btrfs_buffered_write +
btrfs_file_aio_write, ie. just a buffered write, originating from
pwrite64.

The bug_on condition is clear, the question is why the page is not
locked.


david

Re: [bug] oops in linux-3.8-rc2

[Chris Mason]

On Mon, Jan 07, 2013 at 04:10:37AM -0700, David Sterba wrote:
> Hi,
> 
> On Mon, Jan 07, 2013 at 01:51:21AM +0100, Tom Gundersen wrote:
> > I ran into a bug today with 3.8-rc2 [0], following an "make
> > modules_install && make install" of the kernel.
> > 
> > I have been using the same kernel without problems for some days, the
> > only change immediately before the oops was that I switched from plain
> > partitions to using multi-device and subvolumes.
> > 
> > [0]: <https://plus.google.com/114015603831160344127/posts/fzEYnF64NkZ>
> 
> BUG at mm/page-writeback.c:2164
> 
> 2160 int clear_page_dirty_for_io(struct page *page)
> 2161 {
> 2162         struct address_space *mapping = page_mapping(page);
> 2163
> 2164         BUG_ON(!PageLocked(page));
> 
> btrfs on the stack is only __btrfs_buffered_write +
> btrfs_file_aio_write, ie. just a buffered write, originating from
> pwrite64.
> 
> The bug_on condition is clear, the question is why the page is not
> locked.

The big question is, can you make it happen again?

-chris

Re: [bug] oops in linux-3.8-rc2

[Tom Gundersen]

On Mon, Jan 7, 2013 at 2:00 PM, Chris Mason <chris.mason@fusionio.com> wrote:
> The big question is, can you make it happen again?

I wish I could, but I don't want to ;-) At least not on this machine
(it is, very foolishly, my work machine).

I'll try to reproduce it in a VM and let you know if/when I manage.

Cheers,

Tom