Re: [PATCH] fs: fix schedule while atomic caused by gfp of erofs_allocpage

[Zhaoyang Huang]

On Tue, Jul 16, 2024 at 2:20 PM Gao Xiang <hsiangkao@xxxxxxxxxxxxxxxxx> wrote:
>
>
>
> On 2024/7/16 14:14, Zhaoyang Huang wrote:
> > On Tue, Jul 16, 2024 at 1:50 PM Gao Xiang <hsiangkao@xxxxxxxxxxxxxxxxx> wrote:
> >>
> >>
> >>
> >> On 2024/7/16 13:44, zhaoyang.huang wrote:
> >>> From: Zhaoyang Huang <zhaoyang.huang@xxxxxxxxxx>
> >>>
> >>> scheduling while atomic was reported as below where the schedule_timeout
> >>> comes from too_many_isolated when doing direct_reclaim. Fix this by
> >>> masking GFP_DIRECT_RECLAIM from gfp.
> >>>
> >>> [  175.610416][  T618] BUG: scheduling while atomic: kworker/u16:6/618/0x00000000
> >>> [  175.643480][  T618] CPU: 2 PID: 618 Comm: kworker/u16:6 Tainted: G
> >>> [  175.645791][  T618] Workqueue: loop20 loop_workfn
> >>> [  175.646394][  T618] Call trace:
> >>> [  175.646785][  T618]  dump_backtrace+0xf4/0x140
> >>> [  175.647345][  T618]  show_stack+0x20/0x2c
> >>> [  175.647846][  T618]  dump_stack_lvl+0x60/0x84
> >>> [  175.648394][  T618]  dump_stack+0x18/0x24
> >>> [  175.648895][  T618]  __schedule_bug+0x64/0x90
> >>> [  175.649445][  T618]  __schedule+0x680/0x9b8
> >>> [  175.649970][  T618]  schedule+0x130/0x1b0
> >>> [  175.650470][  T618]  schedule_timeout+0xac/0x1d0
> >>> [  175.651050][  T618]  schedule_timeout_uninterruptible+0x24/0x34
> >>> [  175.651789][  T618]  __alloc_pages_slowpath+0x8dc/0x121c
> >>> [  175.652455][  T618]  __alloc_pages+0x294/0x2fc
> >>> [  175.653011][  T618]  erofs_allocpage+0x48/0x58
> >>> [  175.653572][  T618]  z_erofs_runqueue+0x314/0x8a4
> >>> [  175.654161][  T618]  z_erofs_readahead+0x258/0x318
> >>> [  175.654761][  T618]  read_pages+0x88/0x394
> >>> [  175.655275][  T618]  page_cache_ra_unbounded+0x1cc/0x23c
> >>> [  175.655939][  T618]  page_cache_ra_order+0x27c/0x33c
> >>> [  175.656559][  T618]  ondemand_readahead+0x224/0x334
> >>> [  175.657169][  T618]  page_cache_async_ra+0x60/0x9c
> >>> [  175.657767][  T618]  filemap_get_pages+0x19c/0x7cc
> >>> [  175.658367][  T618]  filemap_read+0xf0/0x484
> >>> [  175.658901][  T618]  generic_file_read_iter+0x4c/0x15c
> >>> [  175.659543][  T618]  do_iter_read+0x224/0x348
> >>> [  175.660100][  T618]  vfs_iter_read+0x24/0x38
> >>> [  175.660635][  T618]  loop_process_work+0x408/0xa68
> >>> [  175.661236][  T618]  loop_workfn+0x28/0x34
> >>> [  175.661751][  T618]  process_scheduled_works+0x254/0x4e8
> >>> [  175.662417][  T618]  worker_thread+0x24c/0x33c
> >>> [  175.662974][  T618]  kthread+0x110/0x1b8
> >>> [  175.663465][  T618]  ret_from_fork+0x10/0x20
> >>>
> >>> Signed-off-by: Zhaoyang Huang <zhaoyang.huang@xxxxxxxxxx>
> >>
> >> I don't see why it's an atomic context,
> >> so this patch is incorrect.
> > Sorry, I should provide more details. page_cache_ra_unbounded() will
> > call filemap_invalidate_lock_shared(mapping) to ensure the integrity
> > of page cache during readahead, which will disable preempt.
>
> Why a rwsem sleepable lock disable preemption?
emm, that's the original design of down_read()

> context should be always non-atomic context, which is applied
> to all kernel filesystems.
 AFAICT, filemap_fault/read have added the folios of readahead to page
cache which means the aops->readahead basically just need to map the
block to this folios and then launch the bio. The erofs is a little
bit different to others as it has to alloc_pages for decompression
when doing this.
>
> Thanks,
> Gao Xiang
>
> >>
> >> Thanks,
> >> Gao Xiang