On Wed, Sep 13, 2023, isaku.yamahata@xxxxxxxxx wrote:
> From: Isaku Yamahata <isaku.yamahata@xxxxxxxxx>
>
> When resolving kvm page fault and hwpoisoned page is given, KVM exit
> with HWPOISONED flag so that user space VMM, e.g. qemu, handle it.
>
> - Add a new flag POISON to KVM_EXIT_MEMORY_FAULT to indicate the page is
> poisoned.
> - Make kvm_gmem_get_pfn() return hwpoison state by -EHWPOISON when the
> folio is hw-poisoned.
> - When page is hw-poisoned on faulting in private gmem, return
> KVM_EXIT_MEMORY_FAULT with HWPOISONED flag.
>
> Signed-off-by: Isaku Yamahata <isaku.yamahata@xxxxxxxxx>
> ---
...
> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> index eb900344a054..48329cb44415 100644
> --- a/include/uapi/linux/kvm.h
> +++ b/include/uapi/linux/kvm.h
> @@ -527,7 +527,8 @@ struct kvm_run {
> } notify;
> /* KVM_EXIT_MEMORY_FAULT */
> struct {
> -#define KVM_MEMORY_EXIT_FLAG_PRIVATE (1ULL << 3)
> +#define KVM_MEMORY_EXIT_FLAG_PRIVATE BIT_ULL(3)
> +#define KVM_MEMORY_EXIT_FLAG_HWPOISON BIT_ULL(4)
Rather than add a flag, I think we should double down on returning -1 + errno
when exiting with vcpu->run->exit_reason == KVM_EXIT_MEMORY_FAULT, as is being
proposed in Anish's series for accelerating UFFD-like behavior in KVM[*].
Then KVM can simply return -EFAULT or -EHWPOISON to communicate why KVM is
existing at a higher level, and let the kvm_run structure provide the finer
details about the access itself. E.g. kvm_faultin_pfn_private() can simply
propagate the return value from kvm_gmem_get_pfn() without having to identify
*why* kvm_gmem_get_pfn() failed.
static int kvm_faultin_pfn_private(struct kvm_vcpu *vcpu,
struct kvm_page_fault *fault)
{
int max_order, r;
if (!kvm_slot_can_be_private(fault->slot)) {
kvm_mmu_prepare_memory_fault_exit(vcpu, fault);
return -EFAULT;
}
r = kvm_gmem_get_pfn(vcpu->kvm, fault->slot, fault->gfn, &fault->pfn,
&max_order);
if (r) {
kvm_mmu_prepare_memory_fault_exit(vcpu, fault);
return r;
}
...
}
[*] https://lore.kernel.org/all/20230908222905.1321305-5-amoorthy@xxxxxxxxxx
