Re: [BUG] rcu detected stall in shmem_file_write_iter

["David Hildenbrand (Arm)" <david@kernel.org>]

On 2/5/26 13:57, Zw Tang wrote:
> Hi,
> 
> I am reporting a reproducible RCU stall observed on Linux 6.19.0-rc7,
> triggered by a syzkaller C reproducer.
> 
> The stall is reported while a userspace task is executing the tmpfs
> (shmem) write path. The blocked task is a syz-executor process, and the
> RCU report consistently shows it running in the shmem write / folio
> allocation path for an extended period of time.
> 
> The relevant call trace of the stalled task is:
> 
>    shmem_file_write_iter
>      shmem_write_begin
>        shmem_get_folio_gfp
>          __folio_batch_add_and_move
>            folio_batch_move_lru
>              lru_add
>                __mod_zone_page_state
> 

lru_add()->lruvec_add_folio() calls 
update_lru_size()->__update_lru_size()->__mod_zone_page_state().

On SMP systems we end up in mm/vmstat.c:__mod_zone_page_state().

I don't see anything that could block in there.

Livelocking? Or we're actually stuck somewhere else and the 
__mod_zone_page_state is misleading.

Can you try running with lockdep to see if anything pops up?

-- 
Cheers,

David