WARNING: SOFTIRQ-safe -> SOFTIRQ-unsafe (bio_associate_blkg_from_css+0x3e5/0x650 pool_map+0x23/0x70)

WARNING: SOFTIRQ-safe -> SOFTIRQ-unsafe (bio_associate_blkg_from_css+0x3e5/0x650 pool_map+0x23/0x70)

[Bruno Goncalves]

Hello,

We've hit the issue below twice during our tests on kernel 5.14.0-rc3.
Unfortunately, we don't have a reliable reproducer.

[  496.176739] =====================================================
[  496.182844] WARNING: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected
[  496.189471] 5.14.0-rc3 #1 Not tainted
[  496.193152] -----------------------------------------------------
[  496.199252] systemd-udevd/12979 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire:
[  496.206399] ffff995b358dc5a0 (&q->queue_lock){....}-{2:2}, at:
bio_associate_blkg_from_css+0x3e5/0x650
[  496.215726]
               and this task is already holding:
[  496.221563] ffff995b081d0ad8 (&pool->lock#3){..-.}-{2:2}, at:
pool_map+0x23/0x70 [dm_thin_pool]
[  496.230282] which would create a new lock dependency:
[  496.235344]  (&pool->lock#3){..-.}-{2:2} -> (&q->queue_lock){....}-{2:2}
[  496.242059]
               but this new dependency connects a SOFTIRQ-irq-safe lock:
[  496.249988]  (&pool->lock#3){..-.}-{2:2}
[  496.249993]
               ... which became SOFTIRQ-irq-safe at:
[  496.260111]   lock_acquire+0xb5/0x2b0
[  496.263785]   _raw_spin_lock_irqsave+0x48/0x60
[  496.268240]   overwrite_endio+0x46/0x70 [dm_thin_pool]
[  496.273393]   clone_endio+0xb9/0x1e0
[  496.276979]   clone_endio+0xb9/0x1e0
[  496.280565]   blk_update_request+0x25b/0x420
[  496.284846]   blk_mq_end_request+0x1c/0x130
[  496.289041]   blk_complete_reqs+0x37/0x40
[  496.293071]   __do_softirq+0xde/0x485
[  496.296744]   run_ksoftirqd+0x3a/0x70
[  496.300427]   smpboot_thread_fn+0xf2/0x1c0
[  496.304534]   kthread+0x143/0x160
[  496.307863]   ret_from_fork+0x22/0x30
[  496.311538]
               to a SOFTIRQ-irq-unsafe lock:
[  496.317037]  (&blkcg->lock){+.+.}-{2:2}
[  496.317041]
               ... which became SOFTIRQ-irq-unsafe at:
[  496.327243] ...
[  496.327244]   lock_acquire+0xb5/0x2b0
[  496.332680]   _raw_spin_lock+0x2c/0x40
[  496.336443]   ioc_weight_write+0x153/0x260
[  496.340551]   kernfs_fop_write_iter+0x134/0x1d0
[  496.345095]   new_sync_write+0x10b/0x180
[  496.349030]   vfs_write+0x26a/0x380
[  496.352530]   ksys_write+0x58/0xd0
[  496.355944]   do_syscall_64+0x5c/0x80
[  496.359616]   entry_SYSCALL_64_after_hwframe+0x44/0xae
[  496.364776]

More logs are available on
https://arr-cki-prod-datawarehouse-public.s3.amazonaws.com/index.html?prefix=datawarehouse-public/2021/07/29/345042572/build_x86_64_redhat%3A1463074774/tests/lvm_snapper_test/

Thank you,
Bruno

Re: WARNING: SOFTIRQ-safe -> SOFTIRQ-unsafe (bio_associate_blkg_from_css+0x3e5/0x650 pool_map+0x23/0x70)

[Ming Lei]

On Mon, Aug 02, 2021 at 05:11:08PM +0200, Bruno Goncalves wrote:
> Hello,
> 
> We've hit the issue below twice during our tests on kernel 5.14.0-rc3.
> Unfortunately, we don't have a reliable reproducer.
> 
> [  496.176739] =====================================================
> [  496.182844] WARNING: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected
> [  496.189471] 5.14.0-rc3 #1 Not tainted
> [  496.193152] -----------------------------------------------------
> [  496.199252] systemd-udevd/12979 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire:
> [  496.206399] ffff995b358dc5a0 (&q->queue_lock){....}-{2:2}, at:
> bio_associate_blkg_from_css+0x3e5/0x650
> [  496.215726]
>                and this task is already holding:
> [  496.221563] ffff995b081d0ad8 (&pool->lock#3){..-.}-{2:2}, at:
> pool_map+0x23/0x70 [dm_thin_pool]
> [  496.230282] which would create a new lock dependency:
> [  496.235344]  (&pool->lock#3){..-.}-{2:2} -> (&q->queue_lock){....}-{2:2}
> [  496.242059]
>                but this new dependency connects a SOFTIRQ-irq-safe lock:
> [  496.249988]  (&pool->lock#3){..-.}-{2:2}
> [  496.249993]
>                ... which became SOFTIRQ-irq-safe at:
> [  496.260111]   lock_acquire+0xb5/0x2b0
> [  496.263785]   _raw_spin_lock_irqsave+0x48/0x60
> [  496.268240]   overwrite_endio+0x46/0x70 [dm_thin_pool]
> [  496.273393]   clone_endio+0xb9/0x1e0
> [  496.276979]   clone_endio+0xb9/0x1e0
> [  496.280565]   blk_update_request+0x25b/0x420
> [  496.284846]   blk_mq_end_request+0x1c/0x130
> [  496.289041]   blk_complete_reqs+0x37/0x40
> [  496.293071]   __do_softirq+0xde/0x485
> [  496.296744]   run_ksoftirqd+0x3a/0x70
> [  496.300427]   smpboot_thread_fn+0xf2/0x1c0
> [  496.304534]   kthread+0x143/0x160
> [  496.307863]   ret_from_fork+0x22/0x30
> [  496.311538]
>                to a SOFTIRQ-irq-unsafe lock:
> [  496.317037]  (&blkcg->lock){+.+.}-{2:2}
> [  496.317041]
>                ... which became SOFTIRQ-irq-unsafe at:
> [  496.327243] ...
> [  496.327244]   lock_acquire+0xb5/0x2b0
> [  496.332680]   _raw_spin_lock+0x2c/0x40
> [  496.336443]   ioc_weight_write+0x153/0x260
> [  496.340551]   kernfs_fop_write_iter+0x134/0x1d0
> [  496.345095]   new_sync_write+0x10b/0x180
> [  496.349030]   vfs_write+0x26a/0x380
> [  496.352530]   ksys_write+0x58/0xd0
> [  496.355944]   do_syscall_64+0x5c/0x80
> [  496.359616]   entry_SYSCALL_64_after_hwframe+0x44/0xae
> [  496.364776]
> 
> More logs are available on
> https://arr-cki-prod-datawarehouse-public.s3.amazonaws.com/index.html?prefix=datawarehouse-public/2021/07/29/345042572/build_x86_64_redhat%3A1463074774/tests/lvm_snapper_test/

The following patch should fix the warning:

https://lore.kernel.org/linux-block/20210803070608.1766400-1-ming.lei@redhat.com/T/#u


Thanks,
Ming