Re: [PATCH] KVM: x86: Take PIC lock on KVM_GET_IRQCHIP path

["Carlos López" <clopez@suse.de>]

On 5/29/26 3:24 PM, Sean Christopherson wrote:
> On Fri, May 29, 2026, Carlos López wrote:
>> When userspace issues the KVM_SET_IRQCHIP ioctl to set the state of
>> the PIC, kvm_vm_ioctl_set_irqchip() grabs @kvm->arch.vpic->lock before
>> updating the state. However, the KVM_GET_IRQCHIP ioctl to retrieve the
>> same PIC state does not grab such lock, potentially causing torn reads
>> for userspace.
> 
> Meh, if userspace hasn't fully paused the VM, save/restore is going to fail
> anyways.  Heck, torn reads is probably _better_ than the alternative, because
> at least that might cause visible failure during the restore.  If there are
> concurrent modifications in-flight, then KVM_GET_IRQCHIP is going to return
> stale data (assuming userspace doesn't redo KVM_GET_IRQCHIP), i.e. save/restore
> will effectively corrupt the guest.

Right, do you want a v2 to at least prevent userspace from reading a
torn state? It seems wrong to have this asymmetry with KVM_SET_IRQCHIP
and other save/restore ioctls (e.g. KVM_{G,S}ET_PIT).

>> Fix this by grabbing the lock on the read path.
>>
>> This issue goes all the way back. The bug was introduced with the
>> addition of PIC ioctl code itself in 6ceb9d791eee ("KVM: Add get/
>> set irqchip ioctls for in-kernel PIC live migration support"). Later,
>> 894a9c5543ab ("KVM: x86: missing locking in PIT/IRQCHIP/SET_BSP_CPU
>> ioctl paths") added the locking for kvm_vm_ioctl_set_irqchip(), but
>> missed kvm_vm_ioctl_get_irqchip().
>>
>> Fixes: 6ceb9d791eee ("KVM: Add get/set irqchip ioctls for in-kernel PIC live migration support")
>> Fixes: 894a9c5543ab ("KVM: x86: missing locking in PIT/IRQCHIP/SET_BSP_CPU ioctl paths")
>> Cc: stable@vger.kernel.org
> 
> This isn't stable material.  There's basically zero chance this actively
> problematic for any VMM.
> 
> Honestly, it's tempting to I'm tempted to do the opposite, and yank out the
> locking for the KVM_SET_IRQCHIP path, because userspace really can't be relying
> on kernel locking for correctness across save/restore.  I don't _actually_ think
> we should do that, but it certainly is tempting.
> 
> Ah, actually, maybe SET has locking because it's also used to reset PIC state,
> i.e. isn't limited to just save/restore?  Doesn't really matter.
> 
>> Reported-by: Claude Code:claude-opus-4.6
>> Signed-off-by: Carlos López <clopez@suse.de>
>> ---
>>  arch/x86/kvm/irq.c | 8 ++++++--
>>  1 file changed, 6 insertions(+), 2 deletions(-)
>>
>> diff --git a/arch/x86/kvm/irq.c b/arch/x86/kvm/irq.c
>> index 9519fec09ee6..251df563427b 100644
>> --- a/arch/x86/kvm/irq.c
>> +++ b/arch/x86/kvm/irq.c
>> @@ -584,14 +584,18 @@ int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
>>  
>>  	r = 0;
>>  	switch (chip->chip_id) {
>> -	case KVM_IRQCHIP_PIC_MASTER:
>> +	case KVM_IRQCHIP_PIC_MASTER: {
>> +		guard(spinlock)(&pic->lock);
> 
> I'd much rather use "manual" spin_(un)lock() instead of guard().  Or scoped_guard()
> to avoid the curly braces, but even then, I find this:
> 
> 		scoped_guard(spinlock, &pic->lock)
> 			memcpy(&chip->chip.pic, &pic->pics[0],
> 			       sizeof(struct kvm_pic_state));
> 
> to be much harder to read than:
> 
> 		spin_lock(&pic->lock);
> 		memcpy(&chip->chip.pic, &pic->pics[0],
> 			sizeof(struct kvm_pic_state));
> 		spin_unlock(&pic->lock);
> 
> And no one can reasonably argue that guard() or scoped_guard() makes the this
> particular code more robust.

Oh well, the guard seemed more readable to me but I don't have a strong
opinion either way, I can add this to v2 (if you want it).