[PATCH] KVM: nVMX: Reset RFLAGS on VM-exit

[PATCH] KVM: nVMX: Reset RFLAGS on VM-exit

[Jan Kiszka]

From: Jan Kiszka <jan.kiszka@siemens.com>

Ouch, how could this work so well that far? We need to clear RFLAGS to
the reset value as specified by the SDM. Particularly, IF must be off
after VM-exit!

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
---
 arch/x86/kvm/vmx.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 486078d..252f9c4 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -7454,6 +7454,7 @@ static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
 
 	kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
 	kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
+	vmx_set_rflags(vcpu, 0x02);
 	/*
 	 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
 	 * actually changed, because it depends on the current state of
-- 
1.7.3.4

Re: [PATCH] KVM: nVMX: Reset RFLAGS on VM-exit

[Nadav Har'El]

On Sun, Mar 03, 2013, Jan Kiszka wrote about "[PATCH] KVM: nVMX: Reset RFLAGS on VM-exit":
> From: Jan Kiszka <jan.kiszka@siemens.com>
> 
> Ouch, how could this work so well that far? We need to clear RFLAGS to
> the reset value as specified by the SDM. Particularly, IF must be off
> after VM-exit!

nested_vmx_succeed() or nested_vmx_fail*() were already clearing some of
the fields that I understood was necessary to clear. But they did not clear
the IF - I never realised (and didn't verify now) that this is part of the
spec. And since L1 KVM anyways enters L2 with interrupts disabled, nested KVM
would not see a difference.

> +	vmx_set_rflags(vcpu, 0x02);

There's a macro X86_EFLAGS_BIT1 which you can use for this 0x02.


-- 
Nadav Har'El                        |         Sunday, Mar 3 2013, 22 Adar 5773
nyh@math.technion.ac.il             |-----------------------------------------
Phone +972-523-790466, ICQ 13349191 |"I don't use drugs, my dreams are
http://nadav.harel.org.il           |frightening enough." -- M. C. Escher

Re: [PATCH] KVM: nVMX: Reset RFLAGS on VM-exit

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 1115 bytes --]

On 2013-03-03 20:15, Nadav Har'El wrote:
> On Sun, Mar 03, 2013, Jan Kiszka wrote about "[PATCH] KVM: nVMX: Reset RFLAGS on VM-exit":
>> From: Jan Kiszka <jan.kiszka@siemens.com>
>>
>> Ouch, how could this work so well that far? We need to clear RFLAGS to
>> the reset value as specified by the SDM. Particularly, IF must be off
>> after VM-exit!
> 
> nested_vmx_succeed() or nested_vmx_fail*() were already clearing some of
> the fields that I understood was necessary to clear. But they did not clear
> the IF - I never realised (and didn't verify now) that this is part of the
> spec. And since L1 KVM anyways enters L2 with interrupts disabled, nested KVM
> would not see a difference.

But we didn't restore rflags on vmexit either, we reused the guest value
for L1. It was pure luck (and due to KVM's short IRQ-off phase after
vmexit) when things worked. But I bet this should fix some spurious issues.

> 
>> +	vmx_set_rflags(vcpu, 0x02);
> 
> There's a macro X86_EFLAGS_BIT1 which you can use for this 0x02.

Ah, good to know. Checked vmx_vcpu_reset and found this value.

Jan



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 263 bytes --]