x86,kvm,vmx: Preserve CR4 across VM entry
CR4 isn't constant; at least the TSD and PCE bits can vary.
TBH, treating CR0 and CR3 as constant scares me a bit, too, but it looks
like it's correct.
This adds a branch and a read from cr4 to each vm entry. Because it is
extremely likely that consecutive entries into the same vcpu will have
the same host cr4 value, this fixes up the vmcs instead of restoring cr4
after the fact. A subsequent patch will add a kernel-wide cr4 shadow,
reducing the overhead in the common case to just two memory reads and a
branch.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: stable@vger.kernel.org
Cc: Petr Matousek <pmatouse@redhat.com>
Cc: Gleb Natapov <gleb@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index d9dcfa2..0acac81 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -472,6 +472,7 @@
int gs_ldt_reload_needed;
int fs_reload_needed;
u64 msr_host_bndcfgs;
+ unsigned long vmcs_host_cr4; /* May not match real cr4 */
} host_state;
struct {
int vm86_active;
@@ -4267,11 +4268,16 @@
u32 low32, high32;
unsigned long tmpl;
struct desc_ptr dt;
+ unsigned long cr4;
vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
- vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
+ /* Save the most likely value for this task's CR4 in the VMCS. */
+ cr4 = read_cr4();
+ vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
+ vmx->host_state.vmcs_host_cr4 = cr4;
+
vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
#ifdef CONFIG_X86_64
/*
@@ -7514,7 +7520,7 @@
static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
{
struct vcpu_vmx *vmx = to_vmx(vcpu);
- unsigned long debugctlmsr;
+ unsigned long debugctlmsr, cr4;
/* Record the guest's net vcpu time for enforced NMI injections. */
if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
@@ -7540,6 +7546,12 @@
if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
+ cr4 = read_cr4();
+ if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
+ vmcs_writel(HOST_CR4, cr4);
+ vmx->host_state.vmcs_host_cr4 = cr4;
+ }
+
/* When single-stepping over STI and MOV SS, we must clear the
* corresponding interruptibility bits in the guest state. Otherwise
* vmentry fails as it then expects bit 14 (BS) in pending debug