Re: [PATCH] KVM: x86: allow RSM from 64-bit mode

[Laszlo Ersek <lersek@redhat.com>]

On 11/03/15 14:46, Paolo Bonzini wrote:
> 
> 
> On 03/11/2015 14:40, Laszlo Ersek wrote:
>> On 11/03/15 14:29, Paolo Bonzini wrote:
>>> The SDM says that exiting system management mode from 64-bit mode
>>> is invalid, but that would be too good to be true.  But actually,
>>> most of the code is already there to support exiting from compat
>>> mode (EFER.LME=1, EFER.LMA=0).  Getting all the way from 64-bit
>>> mode to real mode only requires clearing CS.L and CR4.PCIDE.
>>>
>>> Cc: stable@vger.kernel.org
>>> Fixes: 660a5d517aaab9187f93854425c4c63f4a09195c
>>> Cc: Laszlo Ersek <lersek@redhat.com>
>>> Cc: Radim Krčmář <rkrcmar@redhat.com>
>>> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
>>> ---
>>>  arch/x86/kvm/emulate.c | 30 +++++++++++++++++++++++++-----
>>>  1 file changed, 25 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
>>> index b60fed56671b..1505587d06e9 100644
>>> --- a/arch/x86/kvm/emulate.c
>>> +++ b/arch/x86/kvm/emulate.c
>>> @@ -2484,16 +2484,36 @@ static int em_rsm(struct x86_emulate_ctxt *ctxt)
>>>  
>>>  	/*
>>>  	 * Get back to real mode, to prepare a safe state in which to load
>>> -	 * CR0/CR3/CR4/EFER.
>>> -	 *
>>> -	 * CR4.PCIDE must be zero, because it is a 64-bit mode only feature.
>>> +	 * CR0/CR3/CR4/EFER.  It's all a bit more complicated if the vCPU
>>> +	 * supports long mode.
>>>  	 */
>>> +	cr4 = ctxt->ops->get_cr(ctxt, 4);
>>> +	if (emulator_has_longmode(ctxt)) {
>>> +		struct desc_struct cs_desc;
>>> +
>>> +		/* Zero CR4.PCIDE before CR0.PG.  */
>>> +		if (cr4 & X86_CR4_PCIDE) {
>>> +			ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PCIDE);
>>> +			cr4 &= ~X86_CR4_PCIDE;
>>> +		}
>>> +
>>> +		/* A 32-bit code segment is required to clear EFER.LMA.  */
>>> +		memset(&cs_desc, 0, sizeof(cs_desc));
>>> +		cs_desc.type = 0xb;
>>> +		cs_desc.s = cs_desc.g = cs_desc.p = 1;
>>> +		ctxt->ops->set_segment(ctxt, 0, &cs_desc, 0, VCPU_SREG_CS);
>>> +	}
>>> +
>>> +	/* For the 64-bit case, this will clear EFER.LMA.  */
>>>  	cr0 = ctxt->ops->get_cr(ctxt, 0);
>>>  	if (cr0 & X86_CR0_PE)
>>>  		ctxt->ops->set_cr(ctxt, 0, cr0 & ~(X86_CR0_PG | X86_CR0_PE));
>>> -	cr4 = ctxt->ops->get_cr(ctxt, 4);
>>> +
>>> +	/* Now clear CR4.PAE (which must be done before clearing EFER.LME).  */
>>>  	if (cr4 & X86_CR4_PAE)
>>>  		ctxt->ops->set_cr(ctxt, 4, cr4 & ~X86_CR4_PAE);
>>> +
>>> +	/* And finally go back to 32-bit mode.  */
>>>  	efer = 0;
>>>  	ctxt->ops->set_msr(ctxt, MSR_EFER, efer);
>>>  
>>> @@ -4454,7 +4474,7 @@ static const struct opcode twobyte_table[256] = {
>>>  	F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
>>>  	/* 0xA8 - 0xAF */
>>>  	I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
>>> -	II(No64 | EmulateOnUD | ImplicitOps, em_rsm, rsm),
>>> +	II(EmulateOnUD | ImplicitOps, em_rsm, rsm),
>>>  	F(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
>>>  	F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
>>>  	F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
>>>
>>
>> What branch should I test this on top of?
> 
> Just use whatever you were using before, and revert commit c9db607
> ("UefiCpuPkg: PiSmmCpuDxeSmm: do not execute RSM from 64-bit mode",
> 2015-10-14) from your OVMF branch.

Right, I planned to do that OVMF-side revert; I just wasn't sure if e.g.
kvm/queue had some prerequisite patches for this.

> This is how I tested it, in fact.

I'll try to report back soon.

Thanks!
Laszlo