x86/tdx: Allow extended topology CPUID leafs to be emulated by hypercall
The current TDX module does not handle extended topology leaves
explicitly and will generate a #VE but the current #VE handler
implementation blindly returns 0 for those CPUID leaves.
This currently causes TDX guests to see 0 values when querying the numa
topology leading to incorrect numa configurations.
This patch fixes this behavior by emulating the extended topology leaves
using the CPUID hypercall.
Change-Id: I427807e3ac8d9e3be50a6fac40ebd3f54b445b0c
Signed-off-by: Sagi Shahar <sagis@google.com>
diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c
index ccae238..629bf25 100644
--- a/arch/x86/coco/tdx/tdx.c
+++ b/arch/x86/coco/tdx/tdx.c
@@ -381,13 +381,16 @@ static int handle_cpuid(struct pt_regs *regs, struct ve_info *ve)
};
/*
- * Only allow VMM to control range reserved for hypervisor
- * communication.
+ * Only emulate CPUID in 2 cases:
+ * - CPUID is in the range reserved for hypervisor communication.
+ * - CPUID is an extended topology leaf which is not emulated natively
+ * by the TDX module.
*
- * Return all-zeros for any CPUID outside the range. It matches CPU
- * behaviour for non-supported leaf.
+ * Return all-zeros for any other CPUID. It matches CPU behaviour for
+ * non-supported leaf.
*/
- if (regs->ax < 0x40000000 || regs->ax > 0x4FFFFFFF) {
+ if ((regs->ax < 0x40000000 || regs->ax > 0x4FFFFFFF) &&
+ regs->ax != 0x0b && regs->ax != 0x1f) {
regs->ax = regs->bx = regs->cx = regs->dx = 0;
return ve_instr_len(ve);
}
