Re: [PATCH v5] locking/rwsem: Make handoff bit handling more consistent

[john . p . donnelly]

On 4/11/22 1:40 PM, Waiman Long wrote:
> On 4/11/22 14:26, john.p.donnelly@xxxxxxxxxx wrote:
> > On 2/14/22 9:47 AM, chenguanyou wrote:
> > > Hi Waiman, Greg,
> > > This patch has been merged in branch linux-5.16.y.
> > > Can we take it to the linux-5.10.y LTS version?
> > >
> > > thanks,
> >
> > Hi,
> >
> > As a FYI:
> >
> > We have observed that following lockup with this commit added to 
> > 5.15.LTS:
> >
> > d257cc8cb8d5 - locking/rwsem: Make handoff bit handling more 
> > consistent (4 months ago) <Waiman Long>
> >
> > The "fio" test suit fails with LVM devices composed of four NVME 
> > devices with these observed lockup, panic.
> >
> >
> >
> > ext4:
> >
> > PID: 3682   TASK: ffff8f489ae34bc0  CPU: 2   COMMAND: "dio/dm-0"
> >  #0 [fffffe0000083e50] crash_nmi_callback at ffffffff828772b3
> >  #1 [fffffe0000083e58] nmi_handle at ffffffff82840778
> >  #2 [fffffe0000083ea0] default_do_nmi at ffffffff8337a1e2
> >  #3 [fffffe0000083ec8] exc_nmi at ffffffff8337a48d
> >  #4 [fffffe0000083ef0] end_repeat_nmi at ffffffff8340153b
> >     [exception RIP: _raw_spin_lock_irq+23]
> >     RIP: ffffffff8338b2e7  RSP: ffff9c4409b47c78  RFLAGS: 00000046
> >     RAX: 0000000000000000  RBX: ffff8f489ae34bc0  RCX: 0000000000000000
> >     RDX: 0000000000000001  RSI: 0000000000000000  RDI: ffff8f47f7b90104
> >     RBP: ffff9c4409b47d20   R8: 0000000000000000   R9: 0000000000000000
> >     R10: 0000000000000000  R11: 0000000000000000  R12: ffff8f47f7b90104
> >     R13: ffff9c4409b47cb0  R14: ffff8f47f7b900f0  R15: 0000000000000000
> >     ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
> >     <NMI exception stack> ---
> >  #5 [ffff9c4409b47c78] _raw_spin_lock_irq at ffffffff8338b2e7
> >  #6 [ffff9c4409b47c78] rwsem_down_write_slowpath at ffffffff82925be9
> >  #7 [ffff9c4409b47d28] ext4_map_blocks at ffffffffc11c26dc [ext4]
> >  #8 [ffff9c4409b47d98] ext4_convert_unwritten_extents at
> > ffffffffc11ad9e0 [ext4]
> >  #9 [ffff9c4409b47df0] ext4_dio_write_end_io at ffffffffc11b22aa [ext4]
> >
> > xfs:
> >
> > PID: 3719   TASK: ffff9f81d2d74bc0  CPU: 37  COMMAND: "dio/dm-0"
> >  #0 [fffffe0000894e50] crash_nmi_callback at ffffffffad6772b3
> >  #1 [fffffe0000894e58] nmi_handle at ffffffffad640778
> >  #2 [fffffe0000894ea0] default_do_nmi at ffffffffae17a1e2
> >  #3 [fffffe0000894ec8] exc_nmi at ffffffffae17a48d
> >  #4 [fffffe0000894ef0] end_repeat_nmi at ffffffffae20153b
> >     [exception RIP: _raw_spin_lock_irq+23]
> >     RIP: ffffffffae18b2e7  RSP: ffffbb7ec9637c48  RFLAGS: 00000046
> >     RAX: 0000000000000000  RBX: ffff9f81d2d74bc0  RCX: 0000000000000000
> >     RDX: 0000000000000001  RSI: 0000000000000000  RDI: ffff9f81c04a918c
> >     RBP: ffffbb7ec9637ce8   R8: 0000000000000000   R9: 0000000000000000
> >     R10: 0000000000000000  R11: 0000000000000000  R12: ffff9f81c04a918c
> >     R13: ffffbb7ec9637c80  R14: ffff9f81c04a9178  R15: 0000000000000000
> >     ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
> >     <NMI exception stack> ---
> >  #5 [ffffbb7ec9637c48] _raw_spin_lock_irq at ffffffffae18b2e7
> >  #6 [ffffbb7ec9637c48] rwsem_down_write_slowpath at ffffffffad725be9
> >  #7 [ffffbb7ec9637cf0] xfs_trans_alloc_inode at ffffffffc074f2bd [xfs]
> >  #8 [ffffbb7ec9637d50] xfs_iomap_write_unwritten at ffffffffc073ad15
> > [xfs]
> >  #9 [ffffbb7ec9637dd0] xfs_dio_write_end_io at ffffffffc072db62 [xfs]
> >
> >
> > I have reached out to Waiman and he suggested this for our next test 
> > pass:
> >
> >
> > 1ee326196c6658 locking/rwsem: Always try to wake waiters in out_nolock 
> > path
>
> Does this commit help to avoid the lockup problem?
>
> Commit 1ee326196c6658 fixes a potential missed wakeup problem when a 
> reader first in the wait queue is interrupted out without acquiring the 
> lock. It is actually not a fix for commit d257cc8cb8d5. However, this 
> commit changes the out_nolock path behavior of writers by leaving the 
> handoff bit set when the wait queue isn't empty. That likely makes the 
> missed wakeup problem easier to reproduce.
>
> Cheers,
> Longman

Hi,


We are testing now

ETA for fio soak test completion is  ~15hr from now.

I wanted to share the stack traces for future reference + occurrences.


Cheers.

JD



...