On Tue, Aug 16, 2022, Peter Xu wrote:
> Since at it, renaming kvm_handle_bad_page to kvm_handle_error_pfn assuming
Please put parantheses after function names, e.g. kvm_handle_bad_page().
> that'll match better with what it does, e.g. KVM_PFN_ERR_SIGPENDING is not
> accurately a bad page but just one kind of errors.
...
> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> index 3e1317325e1f..23dc46da2f18 100644
> --- a/arch/x86/kvm/mmu/mmu.c
> +++ b/arch/x86/kvm/mmu/mmu.c
> @@ -3134,8 +3134,13 @@ static void kvm_send_hwpoison_signal(unsigned long address, struct task_struct *
> send_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, PAGE_SHIFT, tsk);
> }
>
> -static int kvm_handle_bad_page(struct kvm_vcpu *vcpu, gfn_t gfn, kvm_pfn_t pfn)
> +static int kvm_handle_error_pfn(struct kvm_vcpu *vcpu, gfn_t gfn, kvm_pfn_t pfn)
> {
> + if (is_sigpending_pfn(pfn)) {
> + kvm_handle_signal_exit(vcpu);
> + return -EINTR;
> + }
...
> @@ -2648,9 +2651,12 @@ kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
> if (atomic)
> return KVM_PFN_ERR_FAULT;
>
> - npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
> + npages = hva_to_pfn_slow(addr, async, write_fault, interruptible,
> + writable, &pfn);
> if (npages == 1)
> return pfn;
> + if (npages == -EINTR)
> + return KVM_PFN_ERR_SIGPENDING;
This patch should be split into 3 parts:
1. Add KVM_PFN_ERR_SIGPENDING and the above code
2. Add the interruptible flag
3. Add handling in x86 and rename kvm_handle_bad_page()
With #3 merged with patch 3.
That was if there's oddball arch code that reacts poorly to KVM_PFN_ERR_SIGPENDING,
those errors will bisect to #1.
And if there's a typo in the plumbing, that bisects to #2.
And if something goes sideways in x86, those bugs bisect to #3 (patch 3), and it's
easy to revert just the x86 changes (though I can't imagine that's likely).
