[PATCH v2 20/21] docs: add uefi variable service documentation

[Gerd Hoffmann]

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
---
 docs/devel/index-internals.rst |  1 +
 docs/devel/uefi-vars.rst       | 66 ++++++++++++++++++++++++++++++++++
 hw/uefi/LIMITATIONS.md         |  7 ++++
 3 files changed, 74 insertions(+)
 create mode 100644 docs/devel/uefi-vars.rst
 create mode 100644 hw/uefi/LIMITATIONS.md

diff --git a/docs/devel/index-internals.rst b/docs/devel/index-internals.rst
index ab9fbc448263..56aa589f663c 100644
--- a/docs/devel/index-internals.rst
+++ b/docs/devel/index-internals.rst
@@ -18,6 +18,7 @@ Details about QEMU's various subsystems including how to add 
features to them.
    s390-cpu-topology
    s390-dasd-ipl
    tracing
+   uefi-vars
    vfio-iommufd
    writing-monitor-commands
    virtio-backends
diff --git a/docs/devel/uefi-vars.rst b/docs/devel/uefi-vars.rst
new file mode 100644
index 000000000000..3e7bd98b5208
--- /dev/null
+++ b/docs/devel/uefi-vars.rst
@@ -0,0 +1,66 @@
+==============
+UEFI variables
+==============
+
+Guest UEFI variable management
+==============================
+
+The traditional approach for UEFI Variable storage in qemu guests is
+to work as close as possible to physical hardware.  That means
+providing pflash as storage and leaving the management of variables
+and flash to the guest.
+
+Secure boot support comes with the requirement that the UEFI variable
+storage must be protected against direct access by the OS.  All update
+requests must pass the sanity checks.  (Parts of) the firmware must
+run with a higher privilege level than the OS so this can be enforced
+by the firmware.  On x86 this has been implemented using System
+Management Mode (SMM) in qemu and kvm, which again is the same
+approach taken by physical hardware.  Only privileged code running in
+SMM mode is allowed to access flash storage.
+
+Communication with the firmware code running in SMM mode works by
+serializing the requests to a shared buffer, then trapping into SMM
+mode via SMI.  The SMM code processes the request, stores the reply in
+the same buffer and returns.
+
+Host UEFI variable service
+==========================
+
+Instead of running the privileged code inside the guest we can run it
+on the host.  The serialization protocol can be reused.  The
+communication with the host uses a virtual device, which essentially
+configures the shared buffer location and size, and traps to the host
+to process the requests.
+
+The ``uefi-vars`` device implements the UEFI virtual device.  It comes
+in ``uefi-vars-isa`` and ``uefi-vars-sysbus`` flavours.  The device
+reimplements the handlers needed, specifically
+``EfiSmmVariableProtocol`` and ``VarCheckPolicyLibMmiHandler``.  It
+also consumes events (``EfiEndOfDxeEventGroup``,
+``EfiEventReadyToBoot`` and ``EfiEventExitBootServices``).
+
+The advantage of the approach is that we do not need a special
+privilege level for the firmware to protect itself, i.e. it does not
+depend on SMM emulation on x64, which allows the removal of a bunch of
+complex code for SMM emulation from the linux kernel
+(CONFIG_KVM_SMM=n).  It also allows support for secure boot on arm
+without implementing secure world (el3) emulation in kvm.
+
+Of course there are also downsides.  The added device increases the
+attack surface of the host, and we are adding some code duplication
+because we have to reimplement some edk2 functionality in qemu.
+
+usage on x86_64 (isa)
+---------------------
+
+.. code::
+
+   qemu-system-x86_64 -device uefi-vars-isa,jsonfile=/path/to/vars.json
+
+usage on aarch64 (sysbus)
+-------------------------
+
+.. code::
+
+   qemu-system-aarch64 -M virt,x-uefi-vars=on
diff --git a/hw/uefi/LIMITATIONS.md b/hw/uefi/LIMITATIONS.md
new file mode 100644
index 000000000000..29308bd587aa
--- /dev/null
+++ b/hw/uefi/LIMITATIONS.md
@@ -0,0 +1,7 @@
+known issues and limitations
+----------------------------
+
+* works only on little endian hosts
+  - accessing structs in guest ram is done without endian conversion.
+* works only for 64-bit guests
+  - UINTN is mapped to uint64_t, for 32-bit guests that would be uint32_t
-- 
2.47.1