Re: [PATCH 02/15] KVM: VMX: Avoid to rdmsrl(MSR_IA32_SYSENTER_ESP)

From: Paolo Bonzini
Date: Thu Nov 18 2021 - 06:21:28 EST


On 11/18/21 12:08, Lai Jiangshan wrote:
From: Lai Jiangshan <laijs@xxxxxxxxxxxxxxxxx>

The value of host MSR_IA32_SYSENTER_ESP is known to be constant for
each CPU: (cpu_entry_stack(cpu) + 1) when 32 bit syscall is enabled or
NULL is 32 bit syscall is not enabled.

So rdmsrl() can be avoided for the first case and both rdmsrl() and
vmcs_writel() can be avoided for the second case.

Signed-off-by: Lai Jiangshan <laijs@xxxxxxxxxxxxxxxxx>

Then it's not constant host state, isn't? The hunk in vmx_set_constant_host_state seems wrong.

Paolo

---
arch/x86/kvm/vmx/vmx.c | 10 +++++++---
1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index 48a34d1a2989..4e4a33226edb 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -1271,7 +1271,6 @@ void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
if (!already_loaded) {
void *gdt = get_current_gdt_ro();
- unsigned long sysenter_esp;
/*
* Flush all EPTP/VPID contexts, the new pCPU may have stale
@@ -1287,8 +1286,11 @@ void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
(unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
- rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
- vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
+ if (IS_ENABLED(CONFIG_IA32_EMULATION) || IS_ENABLED(CONFIG_X86_32)) {
+ /* 22.2.3 */
+ vmcs_writel(HOST_IA32_SYSENTER_ESP,
+ (unsigned long)(cpu_entry_stack(cpu) + 1));
+ }
vmx->loaded_vmcs->cpu = cpu;
}
@@ -4021,6 +4023,8 @@ void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
+ rdmsrl(MSR_IA32_SYSENTER_ESP, tmpl);
+ vmcs_writel(HOST_IA32_SYSENTER_ESP, tmpl);
rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */