On 02/10/20 17:43, Vitaly Kuznetsov wrote:
> KVM was switched to interrupt-based mechanism for 'page ready' event
> delivery in Linux-5.8 (see commit 2635b5c4a0e4 ("KVM: x86: interrupt based
> APF 'page ready' event delivery")) and #PF (ab)use for 'page ready' event
> delivery was removed. Linux guest switched to this new mechanism
> exclusively in 5.9 (see commit b1d405751cd5 ("KVM: x86: Switch KVM guest to
> using interrupts for page ready APF delivery")) so it is not possible to
> get older KVM (APF mechanism won't be enabled). Update the comment in
> exc_page_fault() to reflect the new reality.
>
> Signed-off-by: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>
> ---
> arch/x86/mm/fault.c | 13 ++++++++-----
> 1 file changed, 8 insertions(+), 5 deletions(-)
>
> diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c
> index 6e3e8a124903..3cf77592ac54 100644
> --- a/arch/x86/mm/fault.c
> +++ b/arch/x86/mm/fault.c
> @@ -1446,11 +1446,14 @@ DEFINE_IDTENTRY_RAW_ERRORCODE(exc_page_fault)
> prefetchw(¤t->mm->mmap_lock);
>
> /*
> - * KVM has two types of events that are, logically, interrupts, but
> - * are unfortunately delivered using the #PF vector. These events are
> - * "you just accessed valid memory, but the host doesn't have it right
> - * now, so I'll put you to sleep if you continue" and "that memory
> - * you tried to access earlier is available now."
> + * KVM uses #PF vector to deliver 'page not present' events to guests
> + * (asynchronous page fault mechanism). The event happens when a
> + * userspace task is trying to access some valid (from guest's point of
> + * view) memory which is not currently mapped by the host (e.g. the
> + * memory is swapped out). Note, the corresponding "page ready" event
> + * which is injected when the memory becomes available, is delived via
> + * an interrupt mechanism and not a #PF exception
> + * (see arch/x86/kernel/kvm.c: sysvec_kvm_asyncpf_interrupt()).
> *
> * We are relying on the interrupted context being sane (valid RSP,
> * relevant locks not held, etc.), which is fine as long as the
>
Queued, thanks.
Paolo
