41 lines
1.7 KiB
ReStructuredText
41 lines
1.7 KiB
ReStructuredText
.. _mmio-device-passthrough:
|
|
|
|
MMIO Device Passthrough
|
|
########################
|
|
|
|
The ACRN Hypervisor supports both PCI and MMIO device passthrough.
|
|
However there are some constraints on and hypervisor assumptions about
|
|
MMIO devices: there can be no DMA access to the MMIO device and the MMIO
|
|
device may not use IRQs.
|
|
|
|
Here is how ACRN supports MMIO device passthrough:
|
|
|
|
* For a pre-launched VM, the VM configuration tells the ACRN hypervisor
|
|
the addresses of the physical MMIO device's regions and where they are
|
|
mapped to in the pre-launched VM. The hypervisor then removes these
|
|
MMIO regions from the Service VM and fills the vACPI table for this MMIO
|
|
device based on the device's physical ACPI table.
|
|
|
|
* For a post-launched VM, the same actions are done as in a
|
|
pre-launched VM, plus we use the command line to tell which MMIO
|
|
device we want to pass through to the post-launched VM.
|
|
|
|
If the MMIO device has ACPI Tables, use ``--acpidev_pt HID[,UID]`` and
|
|
if not, use ``--mmiodev_pt MMIO_regions``.
|
|
|
|
.. note::
|
|
The vTPM and PT TPM in the ACRN-DM have the same HID so we
|
|
can't support them both at the same time. The VM will fail to boot if
|
|
both are used.
|
|
|
|
These issues remain to be implemented:
|
|
|
|
* Save the MMIO regions in a field of the VM structure in order to
|
|
release the resources when the post-launched VM shuts down abnormally.
|
|
* Allocate the guest MMIO regions for the MMIO device in a guest-reserved
|
|
MMIO region instead of being hard-coded. With this, we could add more
|
|
passthrough MMIO devices.
|
|
* De-assign the MMIO device from the Service VM first before passing
|
|
through it to the post-launched VM and not only removing the MMIO
|
|
regions from the Service VM.
|