| Message ID | 20210715163159.1480168-6-maz@kernel.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | KVM: arm64: MMIO guard PV services | expand |
On Thu, Jul 15, 2021 at 05:31:48PM +0100, Marc Zyngier wrote: > Plumb the MMIO checking code into the MMIO fault handling code. > Nothing allows a region to be registered yet, so there should be > no funtional change either. Typo: functional > Signed-off-by: Marc Zyngier <maz@kernel.org> > --- > arch/arm64/kvm/mmio.c | 10 ++++++++++ > 1 file changed, 10 insertions(+) > > diff --git a/arch/arm64/kvm/mmio.c b/arch/arm64/kvm/mmio.c > index 3dd38a151d2a..fd5747279d27 100644 > --- a/arch/arm64/kvm/mmio.c > +++ b/arch/arm64/kvm/mmio.c > @@ -6,6 +6,7 @@ > > #include <linux/kvm_host.h> > #include <asm/kvm_emulate.h> > +#include <asm/kvm_mmu.h> > #include <trace/events/kvm.h> > > #include "trace.h" > @@ -130,6 +131,10 @@ int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa) > int len; > u8 data_buf[8]; > > + /* Check failed? Return to the guest for debriefing... */ > + if (!kvm_check_ioguard_page(vcpu, fault_ipa)) > + return 1; > + > /* > * No valid syndrome? Ask userspace for help if it has > * volunteered to do so, and bail out otherwise. > @@ -156,6 +161,11 @@ int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa) > len = kvm_vcpu_dabt_get_as(vcpu); > rt = kvm_vcpu_dabt_get_rd(vcpu); > > + /* If we cross a page boundary, check that too... */ > + if (((fault_ipa + len - 1) & PAGE_MASK) != (fault_ipa & PAGE_MASK) && > + !kvm_check_ioguard_page(vcpu, fault_ipa + len - 1)) > + return 1; > + I find this a little odd as the checks straddle the invalid syndrome check, meaning that the relative priorities of KVM_ARCH_FLAG_MMIO_GUARD and KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER are unclear. Will
On Tue, 27 Jul 2021 19:11:21 +0100, Will Deacon <will@kernel.org> wrote: > > On Thu, Jul 15, 2021 at 05:31:48PM +0100, Marc Zyngier wrote: > > Plumb the MMIO checking code into the MMIO fault handling code. > > Nothing allows a region to be registered yet, so there should be > > no funtional change either. > > Typo: functional > > > Signed-off-by: Marc Zyngier <maz@kernel.org> > > --- > > arch/arm64/kvm/mmio.c | 10 ++++++++++ > > 1 file changed, 10 insertions(+) > > > > diff --git a/arch/arm64/kvm/mmio.c b/arch/arm64/kvm/mmio.c > > index 3dd38a151d2a..fd5747279d27 100644 > > --- a/arch/arm64/kvm/mmio.c > > +++ b/arch/arm64/kvm/mmio.c > > @@ -6,6 +6,7 @@ > > > > #include <linux/kvm_host.h> > > #include <asm/kvm_emulate.h> > > +#include <asm/kvm_mmu.h> > > #include <trace/events/kvm.h> > > > > #include "trace.h" > > @@ -130,6 +131,10 @@ int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa) > > int len; > > u8 data_buf[8]; > > > > + /* Check failed? Return to the guest for debriefing... */ > > + if (!kvm_check_ioguard_page(vcpu, fault_ipa)) > > + return 1; > > + > > /* > > * No valid syndrome? Ask userspace for help if it has > > * volunteered to do so, and bail out otherwise. > > @@ -156,6 +161,11 @@ int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa) > > len = kvm_vcpu_dabt_get_as(vcpu); > > rt = kvm_vcpu_dabt_get_rd(vcpu); > > > > + /* If we cross a page boundary, check that too... */ > > + if (((fault_ipa + len - 1) & PAGE_MASK) != (fault_ipa & PAGE_MASK) && > > + !kvm_check_ioguard_page(vcpu, fault_ipa + len - 1)) > > + return 1; > > + > > I find this a little odd as the checks straddle the invalid syndrome check, > meaning that the relative priorities of KVM_ARCH_FLAG_MMIO_GUARD and > KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER are unclear. Good point. And the combination of both flags on its own is odd. Maybe KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER should be ignored or deemed incompatible with the MMIO guard feature. The lack of syndrome information means that we cannot really test for the boundaries of the access (len is invalid), so I'd be tempted to inject an abort in this case. Thoughts? M.
On Wed, Jul 28, 2021 at 11:21:52AM +0100, Marc Zyngier wrote: > On Tue, 27 Jul 2021 19:11:21 +0100, > Will Deacon <will@kernel.org> wrote: > > > > On Thu, Jul 15, 2021 at 05:31:48PM +0100, Marc Zyngier wrote: > > > Plumb the MMIO checking code into the MMIO fault handling code. > > > Nothing allows a region to be registered yet, so there should be > > > no funtional change either. > > > > Typo: functional > > > > > Signed-off-by: Marc Zyngier <maz@kernel.org> > > > --- > > > arch/arm64/kvm/mmio.c | 10 ++++++++++ > > > 1 file changed, 10 insertions(+) > > > > > > diff --git a/arch/arm64/kvm/mmio.c b/arch/arm64/kvm/mmio.c > > > index 3dd38a151d2a..fd5747279d27 100644 > > > --- a/arch/arm64/kvm/mmio.c > > > +++ b/arch/arm64/kvm/mmio.c > > > @@ -6,6 +6,7 @@ > > > > > > #include <linux/kvm_host.h> > > > #include <asm/kvm_emulate.h> > > > +#include <asm/kvm_mmu.h> > > > #include <trace/events/kvm.h> > > > > > > #include "trace.h" > > > @@ -130,6 +131,10 @@ int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa) > > > int len; > > > u8 data_buf[8]; > > > > > > + /* Check failed? Return to the guest for debriefing... */ > > > + if (!kvm_check_ioguard_page(vcpu, fault_ipa)) > > > + return 1; > > > + > > > /* > > > * No valid syndrome? Ask userspace for help if it has > > > * volunteered to do so, and bail out otherwise. > > > @@ -156,6 +161,11 @@ int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa) > > > len = kvm_vcpu_dabt_get_as(vcpu); > > > rt = kvm_vcpu_dabt_get_rd(vcpu); > > > > > > + /* If we cross a page boundary, check that too... */ > > > + if (((fault_ipa + len - 1) & PAGE_MASK) != (fault_ipa & PAGE_MASK) && > > > + !kvm_check_ioguard_page(vcpu, fault_ipa + len - 1)) > > > + return 1; > > > + > > > > I find this a little odd as the checks straddle the invalid syndrome check, > > meaning that the relative priorities of KVM_ARCH_FLAG_MMIO_GUARD and > > KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER are unclear. > > Good point. And the combination of both flags on its own is odd. Maybe > KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER should be ignored or deemed > incompatible with the MMIO guard feature. > > The lack of syndrome information means that we cannot really test for > the boundaries of the access (len is invalid), so I'd be tempted to > inject an abort in this case. > > Thoughts? I agree. Probably worth rejecting both flags anyway so the VMM knows what it's getting, but injecting an abort into the guest if we don't have sufficient syndrom information to triage it safely feels like the right thing to do. Will
diff --git a/arch/arm64/kvm/mmio.c b/arch/arm64/kvm/mmio.c index 3dd38a151d2a..fd5747279d27 100644 --- a/arch/arm64/kvm/mmio.c +++ b/arch/arm64/kvm/mmio.c @@ -6,6 +6,7 @@ #include <linux/kvm_host.h> #include <asm/kvm_emulate.h> +#include <asm/kvm_mmu.h> #include <trace/events/kvm.h> #include "trace.h" @@ -130,6 +131,10 @@ int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa) int len; u8 data_buf[8]; + /* Check failed? Return to the guest for debriefing... */ + if (!kvm_check_ioguard_page(vcpu, fault_ipa)) + return 1; + /* * No valid syndrome? Ask userspace for help if it has * volunteered to do so, and bail out otherwise. @@ -156,6 +161,11 @@ int io_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa) len = kvm_vcpu_dabt_get_as(vcpu); rt = kvm_vcpu_dabt_get_rd(vcpu); + /* If we cross a page boundary, check that too... */ + if (((fault_ipa + len - 1) & PAGE_MASK) != (fault_ipa & PAGE_MASK) && + !kvm_check_ioguard_page(vcpu, fault_ipa + len - 1)) + return 1; + if (is_write) { data = vcpu_data_guest_to_host(vcpu, vcpu_get_reg(vcpu, rt), len);
Plumb the MMIO checking code into the MMIO fault handling code. Nothing allows a region to be registered yet, so there should be no funtional change either. Signed-off-by: Marc Zyngier <maz@kernel.org> --- arch/arm64/kvm/mmio.c | 10 ++++++++++ 1 file changed, 10 insertions(+)