Split Device Model¶
We usually emulate devices in the Device Model. However, in some cases, we need to emulate devices in the ACRN Hypervisor. For example, the post-launched RTVM needs to emulate passthrough PCI(e) devices in the ACRN Hypervisor so that it can continue to run even if the Device Model is no longer working. In spite of this, the Device Model still owns the overall resource management such as memory/MMIO space, interrupt pin, etc.
One communication method provided by the ACRN Hypervisor aligns the resource information for the Device Model with the ACRN Hypervisor emulated device.
Let’s take the passthrough PCI(e) device as an example. Before we split passthrough PCI(e) devices from the Device Model to the ACRN Hypervisor, the whole picture looks like this:
After we split passthrough PCI(e) devices from the Device Model to the ACRN Hypervisor, the whole picture looks like this:
Interfaces Design¶
In order to achieve this, we have added a new pair of hypercalls to align the PCI(e) BAR and INTx information.
-
int32_t
hcall_assign_pcidev
(struct acrn_vcpu *vcpu, struct acrn_vm *target_vm, uint64_t param1, uint64_t param2) Assign one PCI dev to VM.
- Preconditions
is_sos_vm(vcpu->vm)
- Return
0 on success, non-zero on error.
- Parameters
vcpu
: Pointer to vCPU that initiates the hypercalltarget_vm
: Pointer to target VM data structureparam1
: not usedparam2
: guest physical address. This gpa points to data structure of acrn_assign_pcidev including assign PCI device info
-
int32_t
hcall_deassign_pcidev
(struct acrn_vcpu *vcpu, struct acrn_vm *target_vm, uint64_t param1, uint64_t param2) Deassign one PCI dev to VM.
- Preconditions
is_sos_vm(vcpu->vm)
- Return
0 on success, non-zero on error.
- Parameters
vcpu
: Pointer to vCPU that initiates the hypercalltarget_vm
: Pointer to target VM data structureparam1
: not usedparam2
: guest physical address. This gpa points to data structure of acrn_assign_pcidev including deassign PCI device info