Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20221117161449.114086-1-pbonzini@redhat.com> <CALzav=cxtgaVV2tORqDo93AuUW+5BSLdjsah=YASQdPMwnf2iA@mail.gmail.com>
 <11f5e652ca2c2a4507316c3426b25d0d5cd66120.camel@linux.intel.com>
In-Reply-To: <11f5e652ca2c2a4507316c3426b25d0d5cd66120.camel@linux.intel.com>
From:   David Matlack <dmatlack@google.com>
Date:   Thu, 17 Nov 2022 19:49:11 -0800
Message-ID: <CALzav=eeVJAOwUbg1QRnkWhhzn5WT8jgjWo2b-tu9uBJJKdG8A@mail.gmail.com>
Subject: Re: [PATCH] KVM: x86/mmu: simplify kvm_tdp_mmu_map flow when guest
 has to retry
To:     Robert Hoo <robert.hu@linux.intel.com>
Cc:     Paolo Bonzini <pbonzini@redhat.com>, linux-kernel@vger.kernel.org,
        kvm@vger.kernel.org, seanjc@google.com
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Nov 17, 2022 at 6:01 PM Robert Hoo <robert.hu@linux.intel.com> wrote:
>
> On Thu, 2022-11-17 at 10:43 -0800, David Matlack wrote:
> > On Thu, Nov 17, 2022 at 8:14 AM Paolo Bonzini <pbonzini@redhat.com>
> > wrote:
> > >                 if (is_shadow_present_pte(iter.old_spte))
> > > -                       ret = tdp_mmu_split_huge_page(kvm, &iter,
> > > sp, true);
> > > +                       r = tdp_mmu_split_huge_page(kvm, &iter, sp,
> > > true);
> > >                 else
> > > -                       ret = tdp_mmu_link_sp(kvm, &iter, sp,
> > > true);
> > > +                       r = tdp_mmu_link_sp(kvm, &iter, sp, true);
> >
> > Can this fix be squashed into [1]? It seems like a serious enough
> > bug.
> > If 2 threads race to update the same PTE, KVM will return -EBUSY out
> > to userspace from KVM_RUN, I think. I'm not sure about QEMU, but that
> > would be fatal for the VM in Vanadium.
> >
> > [1]
> > https://lore.kernel.org/kvm/20221109185905.486172-3-dmatlack@google.com/
> >
> I think in you patch it's all right, since then before
> kvm_tdp_mmu_map() returns, it must go through
> tdp_mmu_map_handle_target_level(), it returns RET_PF_* enum.

Ah that's right. kvm_tdp_mmu_map() won't actually return 0/-EBUSY,
because it either returns RET_PF_RETRY or goes through
tdp_mmu_map_handle_target_level().