Re: [PATCH] mm/swap: annotate data races for lru_rotate_pvecs

[Marco Elver]

On Fri, 28 Feb 2020 at 05:40, Qian Cai <cai@xxxxxx> wrote:
>
> Read to lru_add_pvec->nr could be interrupted and then write to the same
> variable. The write has local interrupt disabled, but the plain reads
> result in data races. However, it is unlikely the compilers could
> do much damage here given that lru_add_pvec->nr is a "unsigned char" and
> there is an existing compiler barrier. Thus, annotate the reads using the
> data_race() macro. The data races were reported by KCSAN,

Note that, the fact that the writer has local interrupts disabled for
the write is irrelevant because it's the interrupt that triggered
while the read was happening that led to the concurrent write.

I assume you ran this with CONFIG_KCSAN_INTERRUPT_WATCHER=y?  The
option is disabled by default (see its help-text). I don't know if we
want to deal with data races due to interrupts right now, especially
those that just result in 'data_race' annotations. Thoughts?

Thanks,
-- Marco

>  BUG: KCSAN: data-race in lru_add_drain_cpu / rotate_reclaimable_page
>
>  write to 0xffff9291ebcb8a40 of 1 bytes by interrupt on cpu 23:
>   rotate_reclaimable_page+0x2df/0x490
>   pagevec_add at include/linux/pagevec.h:81
>   (inlined by) rotate_reclaimable_page at mm/swap.c:259
>   end_page_writeback+0x1b5/0x2b0
>   end_swap_bio_write+0x1d0/0x280
>   bio_endio+0x297/0x560
>   dec_pending+0x218/0x430 [dm_mod]
>   clone_endio+0xe4/0x2c0 [dm_mod]
>   bio_endio+0x297/0x560
>   blk_update_request+0x201/0x920
>   scsi_end_request+0x6b/0x4a0
>   scsi_io_completion+0xb7/0x7e0
>   scsi_finish_command+0x1ed/0x2a0
>   scsi_softirq_done+0x1c9/0x1d0
>   blk_done_softirq+0x181/0x1d0
>   __do_softirq+0xd9/0x57c
>   irq_exit+0xa2/0xc0
>   do_IRQ+0x8b/0x190
>   ret_from_intr+0x0/0x42
>   delay_tsc+0x46/0x80
>   __const_udelay+0x3c/0x40
>   __udelay+0x10/0x20
>   kcsan_setup_watchpoint+0x202/0x3a0
>   __tsan_read1+0xc2/0x100
>   lru_add_drain_cpu+0xb8/0x3f0
>   lru_add_drain+0x25/0x40
>   shrink_active_list+0xe1/0xc80
>   shrink_lruvec+0x766/0xb70
>   shrink_node+0x2d6/0xca0
>   do_try_to_free_pages+0x1f7/0x9a0
>   try_to_free_pages+0x252/0x5b0
>   __alloc_pages_slowpath+0x458/0x1290
>   __alloc_pages_nodemask+0x3bb/0x450
>   alloc_pages_vma+0x8a/0x2c0
>   do_anonymous_page+0x16e/0x6f0
>   __handle_mm_fault+0xcd5/0xd40
>   handle_mm_fault+0xfc/0x2f0
>   do_page_fault+0x263/0x6f9
>   page_fault+0x34/0x40
>
>  read to 0xffff9291ebcb8a40 of 1 bytes by task 37761 on cpu 23:
>   lru_add_drain_cpu+0xb8/0x3f0
>   lru_add_drain_cpu at mm/swap.c:602
>   lru_add_drain+0x25/0x40
>   shrink_active_list+0xe1/0xc80
>   shrink_lruvec+0x766/0xb70
>   shrink_node+0x2d6/0xca0
>   do_try_to_free_pages+0x1f7/0x9a0
>   try_to_free_pages+0x252/0x5b0
>   __alloc_pages_slowpath+0x458/0x1290
>   __alloc_pages_nodemask+0x3bb/0x450
>   alloc_pages_vma+0x8a/0x2c0
>   do_anonymous_page+0x16e/0x6f0
>   __handle_mm_fault+0xcd5/0xd40
>   handle_mm_fault+0xfc/0x2f0
>   do_page_fault+0x263/0x6f9
>   page_fault+0x34/0x40
>
>  2 locks held by oom02/37761:
>   #0: ffff9281e5928808 (&mm->mmap_sem#2){++++}, at: do_page_fault
>   #1: ffffffffb3ade380 (fs_reclaim){+.+.}, at: fs_reclaim_acquire.part
>  irq event stamp: 1949217
>  trace_hardirqs_on_thunk+0x1a/0x1c
>  __do_softirq+0x2e7/0x57c
>  __do_softirq+0x34c/0x57c
>  irq_exit+0xa2/0xc0
>
>  Reported by Kernel Concurrency Sanitizer on:
>  CPU: 23 PID: 37761 Comm: oom02 Not tainted 5.6.0-rc3-next-20200226+ #6
>  Hardware name: HP ProLiant BL660c Gen9, BIOS I38 10/17/2018
>
> Signed-off-by: Qian Cai <cai@xxxxxx>
> ---
>
> BTW, while at it, I had also looked at other pagevec there, but could
> not tell for  sure if they could be interrupted resulting in data races,
> so I leave them out for now.
>
>  mm/swap.c | 5 +++--
>  1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/mm/swap.c b/mm/swap.c
> index cf39d24ada2a..c922f99dab85 100644
> --- a/mm/swap.c
> +++ b/mm/swap.c
> @@ -599,7 +599,8 @@ void lru_add_drain_cpu(int cpu)
>                 __pagevec_lru_add(pvec);
>
>         pvec = &per_cpu(lru_rotate_pvecs, cpu);
> -       if (pagevec_count(pvec)) {
> +       /* Disabling interrupts below acts as a compiler barrier. */
> +       if (data_race(pagevec_count(pvec))) {
>                 unsigned long flags;
>
>                 /* No harm done if a racing interrupt already did this */
> @@ -744,7 +745,7 @@ void lru_add_drain_all(void)
>                 struct work_struct *work = &per_cpu(lru_add_drain_work, cpu);
>
>                 if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) ||
> -                   pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) ||
> +                   data_race(pagevec_count(&per_cpu(lru_rotate_pvecs, cpu))) ||
>                     pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) ||
>                     pagevec_count(&per_cpu(lru_deactivate_pvecs, cpu)) ||
>                     pagevec_count(&per_cpu(lru_lazyfree_pvecs, cpu)) ||
> --
> 2.21.0 (Apple Git-122.2)
>