[PATCH] kvm/eventfd: fix the misleading comment in kvm_irqfd_assign

[PATCH] kvm/eventfd: fix the misleading comment in kvm_irqfd_assign

[Longpeng(Mike)]

From: Longpeng <longpeng2@huawei.com>

The comment above the invocation of vfs_poll() is misleading, move
it to the right place.

Fixes: 684a0b719ddb ("KVM: eventfd: Fix lock order inversion")
Signed-off-by: Longpeng <longpeng2@huawei.com>
---
 virt/kvm/eventfd.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/virt/kvm/eventfd.c b/virt/kvm/eventfd.c
index 2ad013b..cd01814 100644
--- a/virt/kvm/eventfd.c
+++ b/virt/kvm/eventfd.c
@@ -406,12 +406,12 @@ bool __attribute__((weak)) kvm_arch_irqfd_route_changed(
 
 	spin_unlock_irq(&kvm->irqfds.lock);
 
+	events = vfs_poll(f.file, &irqfd->pt);
+
 	/*
 	 * Check if there was an event already pending on the eventfd
 	 * before we registered, and trigger it as if we didn't miss it.
 	 */
-	events = vfs_poll(f.file, &irqfd->pt);
-
 	if (events & EPOLLIN)
 		schedule_work(&irqfd->inject);
 
-- 
1.8.3.1

Re: [PATCH] kvm/eventfd: fix the misleading comment in kvm_irqfd_assign

[Christian Borntraeger]

Am 29.11.21 um 04:43 schrieb Longpeng(Mike):
> From: Longpeng <longpeng2@huawei.com>
> 
> The comment above the invocation of vfs_poll() is misleading, move
> it to the right place.
> 
I think that the current variant is better.
events is only used in that function to check for EPOLLIN, so the
assignment and the if belong together from a "what am I doing here" perspective.

> Fixes: 684a0b719ddb ("KVM: eventfd: Fix lock order inversion")
> Signed-off-by: Longpeng <longpeng2@huawei.com>
> ---
>   virt/kvm/eventfd.c | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/virt/kvm/eventfd.c b/virt/kvm/eventfd.c
> index 2ad013b..cd01814 100644
> --- a/virt/kvm/eventfd.c
> +++ b/virt/kvm/eventfd.c
> @@ -406,12 +406,12 @@ bool __attribute__((weak)) kvm_arch_irqfd_route_changed(
> 
>   	spin_unlock_irq(&kvm->irqfds.lock);
> 
> +	events = vfs_poll(f.file, &irqfd->pt);
> +
>   	/*
>   	 * Check if there was an event already pending on the eventfd
>   	 * before we registered, and trigger it as if we didn't miss it.
>   	 */
> -	events = vfs_poll(f.file, &irqfd->pt);
> -
>   	if (events & EPOLLIN)
>   		schedule_work(&irqfd->inject);
> 

RE: [PATCH] kvm/eventfd: fix the misleading comment in kvm_irqfd_assign

[Longpeng (Mike, Cloud Infrastructure Service Product Dept.)]

> -----Original Message-----
> From: Christian Borntraeger [mailto:borntraeger@linux.ibm.com]
> Sent: Tuesday, November 30, 2021 1:14 AM
> To: Longpeng (Mike, Cloud Infrastructure Service Product Dept.)
> <longpeng2@huawei.com>; pbonzini@redhat.com
> Cc: cornelia.huck@de.ibm.com; kvm@vger.kernel.org;
> linux-kernel@vger.kernel.org; Gonglei (Arei) <arei.gonglei@huawei.com>
> Subject: Re: [PATCH] kvm/eventfd: fix the misleading comment in kvm_irqfd_assign
> 
> 
> 
> Am 29.11.21 um 04:43 schrieb Longpeng(Mike):
> > From: Longpeng <longpeng2@huawei.com>
> >
> > The comment above the invocation of vfs_poll() is misleading, move
> > it to the right place.
> >
> I think that the current variant is better.
> events is only used in that function to check for EPOLLIN, so the
> assignment and the if belong together from a "what am I doing here" perspective.
> 

Hi Christian,

I think that add the irqfd->wait to the file's wait queue is much more
important, the current variant may lead to ignoring it.

Both of these two variants are supported in the current kernel:

[1] https://elixir.bootlin.com/linux/latest/source/drivers/vfio/virqfd.c#L169
```
	events = vfs_poll(irqfd.file, &virqfd->pt);

	/*
	 * Check if there was an event already pending on the eventfd
	 * before we registered and trigger it as if we didn't miss it.
	 */
	if (events & EPOLLIN) {
		if ((!handler || handler(opaque, data)) && thread)
			schedule_work(&virqfd->inject);
	}
```

[2] https://elixir.bootlin.com/linux/latest/source/drivers/virt/acrn/irqfd.c#L161
```
	/* Check the pending event in this stage */
	events = vfs_poll(f.file, &irqfd->pt);

	if (events & EPOLLIN)
		acrn_irqfd_inject(irqfd);
```

Since there's no any code changes, I agree to drop this unmeaning change.

Thanks.

> > Fixes: 684a0b719ddb ("KVM: eventfd: Fix lock order inversion")
> > Signed-off-by: Longpeng <longpeng2@huawei.com>
> > ---
> >   virt/kvm/eventfd.c | 4 ++--
> >   1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/virt/kvm/eventfd.c b/virt/kvm/eventfd.c
> > index 2ad013b..cd01814 100644
> > --- a/virt/kvm/eventfd.c
> > +++ b/virt/kvm/eventfd.c
> > @@ -406,12 +406,12 @@ bool __attribute__((weak))
> kvm_arch_irqfd_route_changed(
> >
> >   	spin_unlock_irq(&kvm->irqfds.lock);
> >
> > +	events = vfs_poll(f.file, &irqfd->pt);
> > +
> >   	/*
> >   	 * Check if there was an event already pending on the eventfd
> >   	 * before we registered, and trigger it as if we didn't miss it.
> >   	 */
> > -	events = vfs_poll(f.file, &irqfd->pt);
> > -
> >   	if (events & EPOLLIN)
> >   		schedule_work(&irqfd->inject);
> >