Re: 5.14.9 aarch64 OOPS Workqueue: btrfs-delalloc btrfs_work_helper

[Nikolay Borisov <nborisov@suse.com>]

On 22.10.21 г. 13:44, Nikolay Borisov wrote:
> 
> 
> On 22.10.21 г. 5:36, Chris Murphy wrote:
>> OK I have a vmcore file:
>> https://dustymabe.fedorapeople.org/bz2011928-vmcore/
>>
>> lib/modules/5.14.10-300.fc35.aarch64/vmlinuz
>> https://drive.google.com/file/d/1xXM8XGRi_Wzyupbm4MSNteF0rwUzO4GE/view?usp=sharing
>>
> 
> So the problem is we have a null inode:
> 
> 
> crash> struct async_chunk ffff00012a78eb08
> struct async_chunk {
>   inode = 0x0,
>   locked_page = 0xfffffc000508c240,
>   start = 0,
>   end = 4095,
>   write_flags = 0,
>   extents = {
>     next = 0xffff00012a78eb30,
>     prev = 0xffff00012a78eb30
>   },
>   blkcg_css = 0x0,
>   work = {
>     func = 0xffffd7c4c03c05c0 <async_cow_start>,
>     ordered_func = 0xffffd7c4c03c1bf0 <async_cow_submit>,
>     ordered_free = 0xffffd7c4c03be2e0 <async_cow_free>,
>     normal_work = {
>       data = {
>         counter = 256
>       },
>       entry = {
>         next = 0xffff00012a78eb68,
>         prev = 0xffff00012a78eb68
>       },
>       func = 0xffffd7c4c03f9e84 <btrfs_work_helper>
>     },
>     ordered_list = {
>       next = 0xffff00012a78ee80,
>       prev = 0xffff0000c6d83510
>     },
>     wq = 0xffff0000c6d83500,
>     flags = 3
>   },
>   pending = 0xffff00012a78eb00
> }
> 
> 
> But this makes no sense since before submit_compressed_extents is called
> we have an explicit check for async_hunk->inode presence but AFAICS this
> is not done in a concurrent context. So this either leaves some hw issue
> or some race which manifests due to ARM's weak mm.

I also looked at the assembly generated in async_cow_submit to see if
anything funny happens while the async_chunk->inode check is performed -
everything looks fine. Also given that the extents list is empty and the
inode is NULL I'd assume that the "write" side is also correct i.e the
code in async_cow_start. This pretty much excludes a codegen problem.

Chris can you add the following line in submit_compressed_extents right
before the BTRFS_I() function is called:

 WARN_ON(!async_chunk->inode);

And re-run the workload again?


> 
>>
>> --
>> Chris Murphy
>>
>