Re: pci_stub and kvm

Re: pci_stub and kvm

[Avi Kivity]

On 07/01/2009 07:18 AM, Yinghai Lu wrote:
> [ 1966.343286]
> [ 1966.343288] =======================================================
> [ 1966.356756] [ INFO: possible circular locking dependency detected ]
> [ 1966.356759] 2.6.31-rc1-tip-00978-g99123e5-dirty #438
> [ 1966.356761] -------------------------------------------------------
> [ 1966.356764] events/0/387 is trying to acquire lock:
> [ 1966.356766]  (&kvm->lock){+.+.+.}, at: [<ffffffff8100af27>]
> kvm_assigned_dev_interrupt_work_handler+0x42/0x13a
> [ 1966.356786]
> [ 1966.356787] but task is already holding lock:
> [ 1966.356789]  (&match->interrupt_work){+.+...}, at:
> [<ffffffff810986e9>] worker_thread+0x175/0x2f6
> [ 1966.356797]
> [ 1966.356798] which lock already depends on the new lock.
> [ 1966.356799]
> [ 1966.356800]
> [ 1966.356801] the existing dependency chain (in reverse order) is:
> [ 1966.356803]
> [ 1966.356803] ->  #1 (&match->interrupt_work){+.+...}:
> [ 1966.356809]        [<ffffffff810b3bf6>] __lock_acquire+0x1396/0x1710
> [ 1966.356817]        [<ffffffff810b403c>] lock_acquire+0xcc/0x104
> [ 1966.356821]        [<ffffffff810994a8>] __cancel_work_timer+0x121/0x247
> [ 1966.356825]        [<ffffffff8109962c>] cancel_work_sync+0x23/0x39
> [ 1966.356828]        [<ffffffff8100b280>] kvm_deassign_irq+0xf1/0x183
> [ 1966.356832]        [<ffffffff8100db6c>] kvm_vm_ioctl+0x8c8/0xc1a
> [ 1966.356837]        [<ffffffff81156e56>] vfs_ioctl+0x3e/0xa3
> [ 1966.356846]        [<ffffffff8115741c>] do_vfs_ioctl+0x4be/0x511
> [ 1966.356850]        [<ffffffff811574c5>] sys_ioctl+0x56/0x8d
> [ 1966.356854]        [<ffffffff81034fdb>] system_call_fastpath+0x16/0x1b
> [ 1966.356860]        [<ffffffffffffffff>] 0xffffffffffffffff
> [ 1966.356869]
> [ 1966.356870] ->  #0 (&kvm->lock){+.+.+.}:
> [ 1966.356872]        [<ffffffff810b392b>] __lock_acquire+0x10cb/0x1710
> [ 1966.356875]        [<ffffffff810b403c>] lock_acquire+0xcc/0x104
> [ 1966.356878]        [<ffffffff81cde487>] mutex_lock_nested+0x75/0x2fa
> [ 1966.356886]        [<ffffffff8100af27>]
> kvm_assigned_dev_interrupt_work_handler+0x42/0x13a
> [ 1966.356890]        [<ffffffff81098743>] worker_thread+0x1cf/0x2f6
> [ 1966.356892]        [<ffffffff8109e335>] kthread+0xa8/0xb0
> [ 1966.356899]        [<ffffffff8103609a>] child_rip+0xa/0x20
> [ 1966.356906]        [<ffffffffffffffff>] 0xffffffffffffffff
>    

This is already fixed in kvm.git.  I'm not sure about merging it to 
2.6.30 since the race is very rare and involves device assignment (which 
is not very mainstream), while the fix touches the core kvm parts.

-- 
error compiling committee.c: too many arguments to function

Re: pci_stub and kvm

[Yinghai Lu]

On Wed, Jul 1, 2009 at 12:49 AM, Avi Kivity<avi@redhat.com> wrote:
> On 07/01/2009 07:18 AM, Yinghai Lu wrote:
>>
>> [ 1966.343286]
>> [ 1966.343288] =======================================================
>> [ 1966.356756] [ INFO: possible circular locking dependency detected ]
>> [ 1966.356759] 2.6.31-rc1-tip-00978-g99123e5-dirty #438
>> [ 1966.356761] -------------------------------------------------------
>> [ 1966.356764] events/0/387 is trying to acquire lock:
>> [ 1966.356766]  (&kvm->lock){+.+.+.}, at: [<ffffffff8100af27>]
>> kvm_assigned_dev_interrupt_work_handler+0x42/0x13a
>> [ 1966.356786]
>> [ 1966.356787] but task is already holding lock:
>> [ 1966.356789]  (&match->interrupt_work){+.+...}, at:
>> [<ffffffff810986e9>] worker_thread+0x175/0x2f6
>> [ 1966.356797]
>> [ 1966.356798] which lock already depends on the new lock.
>> [ 1966.356799]
>> [ 1966.356800]
>> [ 1966.356801] the existing dependency chain (in reverse order) is:
>> [ 1966.356803]
>> [ 1966.356803] ->  #1 (&match->interrupt_work){+.+...}:
>> [ 1966.356809]        [<ffffffff810b3bf6>] __lock_acquire+0x1396/0x1710
>> [ 1966.356817]        [<ffffffff810b403c>] lock_acquire+0xcc/0x104
>> [ 1966.356821]        [<ffffffff810994a8>] __cancel_work_timer+0x121/0x247
>> [ 1966.356825]        [<ffffffff8109962c>] cancel_work_sync+0x23/0x39
>> [ 1966.356828]        [<ffffffff8100b280>] kvm_deassign_irq+0xf1/0x183
>> [ 1966.356832]        [<ffffffff8100db6c>] kvm_vm_ioctl+0x8c8/0xc1a
>> [ 1966.356837]        [<ffffffff81156e56>] vfs_ioctl+0x3e/0xa3
>> [ 1966.356846]        [<ffffffff8115741c>] do_vfs_ioctl+0x4be/0x511
>> [ 1966.356850]        [<ffffffff811574c5>] sys_ioctl+0x56/0x8d
>> [ 1966.356854]        [<ffffffff81034fdb>] system_call_fastpath+0x16/0x1b
>> [ 1966.356860]        [<ffffffffffffffff>] 0xffffffffffffffff
>> [ 1966.356869]
>> [ 1966.356870] ->  #0 (&kvm->lock){+.+.+.}:
>> [ 1966.356872]        [<ffffffff810b392b>] __lock_acquire+0x10cb/0x1710
>> [ 1966.356875]        [<ffffffff810b403c>] lock_acquire+0xcc/0x104
>> [ 1966.356878]        [<ffffffff81cde487>] mutex_lock_nested+0x75/0x2fa
>> [ 1966.356886]        [<ffffffff8100af27>]
>> kvm_assigned_dev_interrupt_work_handler+0x42/0x13a
>> [ 1966.356890]        [<ffffffff81098743>] worker_thread+0x1cf/0x2f6
>> [ 1966.356892]        [<ffffffff8109e335>] kthread+0xa8/0xb0
>> [ 1966.356899]        [<ffffffff8103609a>] child_rip+0xa/0x20
>> [ 1966.356906]        [<ffffffffffffffff>] 0xffffffffffffffff
>>
>
> This is already fixed in kvm.git.  I'm not sure about merging it to 2.6.30
> since the race is very rare and involves device assignment (which is not
> very mainstream), while the fix touches the core kvm parts.

tip/master merged with kvm/master doesn't have that warning.

YH