19. Intel Trust Domain Extensions (TDX)¶
Intel’s Trust Domain Extensions (TDX) protect confidential guest VMs from the host and physical attacks by isolating the guest register state and by encrypting the guest memory. In TDX, a special module running in a special mode sits between the host and the guest and manages the guest/host separation.
Since the host cannot directly access guest registers or memory, much normal functionality of a hypervisor must be moved into the guest. This is implemented using a Virtualization Exception (#VE) that is handled by the guest kernel. A #VE is handled entirely inside the guest kernel, but some require the hypervisor to be consulted.
TDX includes new hypercall-like mechanisms for communicating from the guest to the hypervisor or the TDX module.
19.1. New TDX Exceptions¶
TDX guests behave differently from bare-metal and traditional VMX guests. In TDX guests, otherwise normal instructions or memory accesses can cause #VE or #GP exceptions.
Instructions marked with an ‘*’ conditionally cause exceptions. The details for these instructions are discussed below.
19.1.1. Instruction-based #VE¶
Port I/O (INS, OUTS, IN, OUT)
19.1.2. Instruction-based #GP¶
All VMX instructions: INVEPT, INVVPID, VMCLEAR, VMFUNC, VMLAUNCH, VMPTRLD, VMPTRST, VMREAD, VMRESUME, VMWRITE, VMXOFF, VMXON
19.1.3. RDMSR/WRMSR Behavior¶
MSR access behavior falls into three categories:
In general, the #GP MSRs should not be used in guests. Their use likely indicates a bug in the guest. The guest may try to handle the #GP with a hypercall but it is unlikely to succeed.
The #VE MSRs are typically able to be handled by the hypervisor. Guests can make a hypercall to the hypervisor to handle the #VE.
The “just works” MSRs do not need any special guest handling. They might be implemented by directly passing through the MSR to the hardware or by trapping and handling in the TDX module. Other than possibly being slow, these MSRs appear to function just as they would on bare metal.
19.1.4. CPUID Behavior¶
For some CPUID leaves and sub-leaves, the virtualized bit fields of CPUID return values (in guest EAX/EBX/ECX/EDX) are configurable by the hypervisor. For such cases, the Intel TDX module architecture defines two virtualization types:
Bit fields for which the hypervisor controls the value seen by the guest TD.
Bit fields for which the hypervisor configures the value such that the guest TD either sees their native value or a value of 0. For these bit fields, the hypervisor can mask off the native values, but it can not turn on values.
A #VE is generated for CPUID leaves and sub-leaves that the TDX module does not know how to handle. The guest kernel may ask the hypervisor for the value with a hypercall.
19.2. #VE on Memory Accesses¶
There are essentially two classes of TDX memory: private and shared. Private memory receives full TDX protections. Its content is protected against access from the hypervisor. Shared memory is expected to be shared between guest and hypervisor and does not receive full TDX protections.
A TD guest is in control of whether its memory accesses are treated as private or shared. It selects the behavior with a bit in its page table entries. This helps ensure that a guest does not place sensitive information in shared memory, exposing it to the untrusted hypervisor.
19.2.2. #VE on Private Pages¶
An access to private mappings can also cause a #VE. Since all kernel memory is also private memory, the kernel might theoretically need to handle a #VE on arbitrary kernel memory accesses. This is not feasible, so TDX guests ensure that all guest memory has been “accepted” before memory is used by the kernel.
A modest amount of memory (typically 512M) is pre-accepted by the firmware before the kernel runs to ensure that the kernel can start up without being subjected to a #VE.
The hypervisor is permitted to unilaterally move accepted pages to a “blocked” state. However, if it does this, page access will not generate a #VE. It will, instead, cause a “TD Exit” where the hypervisor is required to handle the exception.
19.3. Linux #VE handler¶
Just like page faults or #GP’s, #VE exceptions can be either handled or be fatal. Typically, an unhandled userspace #VE results in a SIGSEGV. An unhandled kernel #VE results in an oops.
Handling nested exceptions on x86 is typically nasty business. A #VE could be interrupted by an NMI which triggers another #VE and hilarity ensues. The TDX #VE architecture anticipated this scenario and includes a feature to make it slightly less nasty.
During #VE handling, the TDX module ensures that all interrupts (including NMIs) are blocked. The block remains in place until the guest makes a TDG.VP.VEINFO.GET TDCALL. This allows the guest to control when interrupts or a new #VE can be delivered.
However, the guest kernel must still be careful to avoid potential #VE-triggering actions (discussed above) while this block is in place. While the block is in place, any #VE is elevated to a double fault (#DF) which is not recoverable.
19.4. MMIO handling¶
In non-TDX VMs, MMIO is usually implemented by giving a guest access to a mapping which will cause a VMEXIT on access, and then the hypervisor emulates the access. That is not possible in TDX guests because VMEXIT will expose the register state to the host. TDX guests don’t trust the host and can’t have their state exposed to the host.
In TDX, MMIO regions typically trigger a #VE exception in the guest. The guest #VE handler then emulates the MMIO instruction inside the guest and converts it into a controlled TDCALL to the host, rather than exposing guest state to the host.
MMIO addresses on x86 are just special physical addresses. They can theoretically be accessed with any instruction that accesses memory. However, the kernel instruction decoding method is limited. It is only designed to decode instructions like those generated by io.h macros.
MMIO access via other means (like structure overlays) may result in an oops.
TDX reference material is collected here: