On 2021-07-21 22:17, Andrew Jones wrote:
On Thu, Jul 15, 2021 at 05:31:53PM +0100, Marc Zyngier wrote:
Document the hypercalls user for the MMIO guard infrastructure.
Signed-off-by: Marc Zyngier <maz@xxxxxxxxxx>
---
Documentation/virt/kvm/arm/index.rst | 1 +
Documentation/virt/kvm/arm/mmio-guard.rst | 73
+++++++++++++++++++++++
2 files changed, 74 insertions(+)
create mode 100644 Documentation/virt/kvm/arm/mmio-guard.rst
diff --git a/Documentation/virt/kvm/arm/index.rst
b/Documentation/virt/kvm/arm/index.rst
index 78a9b670aafe..e77a0ee2e2d4 100644
--- a/Documentation/virt/kvm/arm/index.rst
+++ b/Documentation/virt/kvm/arm/index.rst
@@ -11,3 +11,4 @@ ARM
psci
pvtime
ptp_kvm
+ mmio-guard
diff --git a/Documentation/virt/kvm/arm/mmio-guard.rst
b/Documentation/virt/kvm/arm/mmio-guard.rst
new file mode 100644
index 000000000000..a5563a3e12cc
--- /dev/null
+++ b/Documentation/virt/kvm/arm/mmio-guard.rst
@@ -0,0 +1,73 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+==============
+KVM MMIO guard
+==============
+
+KVM implements device emulation by handling translation faults to any
+IPA range that is not contained a memory slot. Such translation fault
^ in ^ a
+is in most cases passed on to userspace (or in rare cases to the host
+kernel) with the address, size and possibly data of the access for
+emulation.
+
+Should the guest exit with an address that is not one that
corresponds
+to an emulatable device, userspace may take measures that are not the
+most graceful as far as the guest is concerned (such as terminating
it
+or delivering a fatal exception).
+
+There is also an element of trust: by forwarding the request to
+userspace, the kernel asumes that the guest trusts userspace to do
the
assumes
+right thing.
+
+The KVM MMIO guard offers a way to mitigate this last point: a guest
+can request that only certainly regions of the IPA space are valid as
certain
Thanks, all corrections applied.
+MMIO. Only these regions will be handled as an MMIO, and any other
+will result in an exception being delivered to the guest.
+
+This relies on a set of hypercalls defined in the KVM-specific range,
+using the HVC64 calling convention.
+
+* ARM_SMCCC_KVM_FUNC_MMIO_GUARD_INFO
+
+ ============== ======== ================================
+ Function ID: (uint32) 0xC6000002
+ Arguments: none
+ Return Values: (int64) NOT_SUPPORTED(-1) on error, or
+ (uint64) Protection Granule (PG) size in
+ bytes (r0)
+ ============== ======== ================================
+
+* ARM_SMCCC_KVM_FUNC_MMIO_GUARD_ENROLL
+
+ ============== ======== ==============================
+ Function ID: (uint32) 0xC6000003
+ Arguments: none
+ Return Values: (int64) NOT_SUPPORTED(-1) on error, or
+ RET_SUCCESS(0) (r0)
+ ============== ======== ==============================
+
+* ARM_SMCCC_KVM_FUNC_MMIO_GUARD_MAP
+
+ ============== ========
======================================
+ Function ID: (uint32) 0xC6000004
+ Arguments: (uint64) The base of the PG-sized IPA range
+ that is allowed to be accessed as
+ MMIO. Must aligned to the PG size (r1)
align
Hmmm. Ugly mix of tab and spaces. I have no idea what the norm
is here, so I'll just put spaces. I'm sure someone will let me
know if I'm wrong! ;-)
+ (uint64) Index in the MAIR_EL1 register
+ providing the memory attribute that
+ is used by the guest (r2)
+ Return Values: (int64) NOT_SUPPORTED(-1) on error, or
+ RET_SUCCESS(0) (r0)
+ ============== ========
======================================
+
+* ARM_SMCCC_KVM_FUNC_MMIO_GUARD_UNMAP
+
+ ============== ========
======================================
+ Function ID: (uint32) 0xC6000004
copy+paste error, should be 0xC6000005
Gah, well cpotted.
+ Arguments: (uint64) The base of the PG-sized IPA range
+ that is forbidden to be accessed as
is now forbidden
or
was allowed
or just drop that part of the sentence because its covered by the "and
have been previously mapped" part. Something like
PG-sized IPA range aligned to the PG size which has been previously
mapped
(r1)
Picked the latter.
Thanks again,
M.
--
Jazz is not dead. It just smells funny...
