Re: [PATCH v2] x86/kvm: Disable KVM_ASYNC_PF_SEND_ALWAYS

[Thomas Gleixner]

Vivek Goyal <vgoyal@xxxxxxxxxx> writes:
> On Mon, Mar 09, 2020 at 09:22:15PM +0100, Peter Zijlstra wrote:
>> On Mon, Mar 09, 2020 at 08:05:18PM +0100, Thomas Gleixner wrote:
>> > Andy Lutomirski <luto@xxxxxxxxxx> writes:
>> 
>> > > I'm okay with the save/restore dance, I guess.  It's just yet more
>> > > entry crud to deal with architecture nastiness, except that this
>> > > nastiness is 100% software and isn't Intel/AMD's fault.
>> > 
>> > And we can do it in C and don't have to fiddle with it in the ASM
>> > maze.
>> 
>> Right; I'd still love to kill KVM_ASYNC_PF_SEND_ALWAYS though, even if
>> we do the save/restore in do_nmi(). That is some wild brain melt. Also,
>> AFAIK none of the distros are actually shipping a PREEMPT=y kernel
>> anyway, so killing it shouldn't matter much.
>
> It will be nice if we can retain KVM_ASYNC_PF_SEND_ALWAYS. I have another
> use case outside CONFIG_PREEMPT.
>
> I am trying to extend async pf interface to also report page fault errors
> to the guest.
>
> https://lore.kernel.org/kvm/20200331194011.24834-1-vgoyal@xxxxxxxxxx/
>
> Right now async page fault interface assumes that host will always be
> able to successfully resolve the page fault and sooner or later PAGE_READY
> event will be sent to guest. And there is no mechnaism to report the
> errors back to guest.
>
> I am trying to add enhance virtiofs to directly map host page cache in guest.
>
> https://lore.kernel.org/linux-fsdevel/20200304165845.3081-1-vgoyal@xxxxxxxxxx/
>
> There it is possible that a file page on host is mapped in guest and file
> got truncated and page is not there anymore. Guest tries to access it,
> and it generates async page fault. On host we will get -EFAULT and I 
> need to propagate it back to guest so that guest can either send SIGBUS
> to process which caused this. Or if kernel was trying to do memcpy(),
> then be able to use execpetion table error handling and be able to
> return with error.  (memcpy_mcflush()).
>
> For the second case to work, I will need async pf events to come in
> even if guest is in kernel and CONFIG_PREEMPT=n.

What?

> So it would be nice if we can keep KVM_ASYNC_PF_SEND_ALWAYS around.

No. If you want this stuff to be actually useful and correct, then
please redesign it from scratch w/o abusing #PF. It want's to be a
separate vector and then the pagefault resulting from your example above
becomes a real #PF without any bells and whistels.

Thanks,

        tglx