linux-kernel - Re: [PATCH 1/2] KVM: x86: Return emulator error if RDMSR/WRMSR emulation failed

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite for Android: free password hash cracker in your pocket
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <Yw5aeFp9rTs4tkDb@google.com>
Date:   Tue, 30 Aug 2022 18:44:08 +0000
From:   Sean Christopherson <seanjc@...gle.com>
To:     Hou Wenlong <houwenlong.hwl@...group.com>
Cc:     kvm@...r.kernel.org, Paolo Bonzini <pbonzini@...hat.com>,
        Vitaly Kuznetsov <vkuznets@...hat.com>,
        Wanpeng Li <wanpengli@...cent.com>,
        Jim Mattson <jmattson@...gle.com>,
        Joerg Roedel <joro@...tes.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        Ingo Molnar <mingo@...hat.com>, Borislav Petkov <bp@...en8.de>,
        Dave Hansen <dave.hansen@...ux.intel.com>, x86@...nel.org,
        "H. Peter Anvin" <hpa@...or.com>, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 1/2] KVM: x86: Return emulator error if RDMSR/WRMSR
 emulation failed

On Thu, Jul 28, 2022, Hou Wenlong wrote:
> The return value of emulator_{get|set}_mst_with_filter()
> is confused, since msr access error and emulator error
> are mixed. Although, KVM_MSR_RET_* doesn't conflict with
> X86EMUL_IO_NEEDED at present, it is better to convert
> msr access error to emulator error if error value is
> needed.
> 
> Signed-off-by: Hou Wenlong <houwenlong.hwl@...group.com>
> ---
>  arch/x86/kvm/x86.c | 22 ++++++++++++----------
>  1 file changed, 12 insertions(+), 10 deletions(-)
> 
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 5366f884e9a7..8df89b9c212f 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -7908,11 +7908,12 @@ static int emulator_get_msr_with_filter(struct x86_emulate_ctxt *ctxt,
>  	int r;
>  
>  	r = kvm_get_msr_with_filter(vcpu, msr_index, pdata);
> -
> -	if (r && kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_RDMSR, 0,
> -				    complete_emulated_rdmsr, r)) {
> -		/* Bounce to user space */
> -		return X86EMUL_IO_NEEDED;
> +	if (r) {
> +		if (kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_RDMSR, 0,
> +				       complete_emulated_rdmsr, r))
> +			r = X86EMUL_IO_NEEDED;
> +		else
> +			r = X86EMUL_UNHANDLEABLE;

This should be X86EMUL_PROPAGATE_FAULT, X86EMUL_UNHANDLEABLE is used to indicate
that KVM needs to bail all the way to userspace.

I definitely like the idea of converting to X86EMUL_* here instead of spreading
it across these helpers and the emulator, but in that case should convert _all_
types.

And I think it makes sense to opportunistically handle "r < 0" in the get helper.
KVM may not return -errno today, but assuming that will always hold true is
unnecessarily risking.

E.g. what about:


static int emulator_get_msr_with_filter(struct x86_emulate_ctxt *ctxt,
					u32 msr_index, u64 *pdata)
{
	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
	int r;

	r = kvm_get_msr_with_filter(vcpu, msr_index, pdata);
	if (r < 0)
		return X86EMUL_UNHANDLEABLE;

	if (r) {
		if (kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_RDMSR, 0,
				       complete_emulated_rdmsr, r))
			return X86EMUL_IO_NEEDED;
		else
			return X86EMUL_PROPAGATE_FAULT;
	}

	return X86EMUL_CONTINUE;
}

static int emulator_set_msr_with_filter(struct x86_emulate_ctxt *ctxt,
					u32 msr_index, u64 data)
{
	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
	int r;

	r = kvm_set_msr_with_filter(vcpu, msr_index, data);
	if (r < 0)
		return X86EMUL_UNHANDLEABLE;

	if (r) {
		if (kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_WRMSR, data,
				       complete_emulated_msr_access, r))
			return X86EMUL_IO_NEEDED;
		else
			return X86EMUL_PROPAGATE_FAULT;
	}

	return X86EMUL_CONTINUE;
}


Or maybe even add a helper to do the translation?  Can't tell if this is a net
positive or not.  It's a bit gratuitous, but it does ensure consistent behavior
for RDMSR vs. WRMSR.

static int emulator_handle_msr_return(struct kvm_vcpu *vcpu *, int r,
				      u32 msr, u64 data, u32 exit_reason,
				      int (*comp)(struct kvm_vcpu *vcpu))
{
	if (r < 0)
		return X86EMUL_UNHANDLEABLE;

	if (r) {
		if (kvm_msr_user_space(vcpu, msr, exit_reason, data, comp, r))
			return X86EMUL_IO_NEEDED;
		else
			return X86EMUL_UNHANDLEABLE;
	}

	return X86EMUL_CONTINUE;
}

static int emulator_get_msr_with_filter(struct x86_emulate_ctxt *ctxt,
					u32 msr_index, u64 *pdata)
{
	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
	int r;

	r = kvm_get_msr_with_filter(vcpu, msr_index, pdata);
	return emulator_handle_msr_return(vcpu, r, msr_index, 0,
					  KVM_EXIT_X86_RDMSR,
					  complete_emulated_rdmsr);
}

static int emulator_set_msr_with_filter(struct x86_emulate_ctxt *ctxt,
					u32 msr_index, u64 data)
{
	struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
	int r;

	r = kvm_set_msr_with_filter(vcpu, msr_index, data);
	return emulator_handle_msr_return(vcpu, r, msr_index, data,
					  KVM_EXIT_X86_WRMSR,
					  complete_emulated_msr_access);
}


And then the emulator side of things can be:

static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
{
	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
	u64 msr_data;
	int r;

	msr_data = (u32)reg_read(ctxt, VCPU_REGS_RAX)
		| ((u64)reg_read(ctxt, VCPU_REGS_RDX) << 32);
	r = ctxt->ops->set_msr_with_filter(ctxt, msr_index, msr_data);

	if (r == X86EMUL_PROPAGATE_FAULT)
		return emulate_gp(ctxt, 0);

	return r;
}

static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
{
	u64 msr_index = reg_read(ctxt, VCPU_REGS_RCX);
	u64 msr_data;
	int r;

	r = ctxt->ops->get_msr_with_filter(ctxt, msr_index, &msr_data);

	if (r == X86EMUL_PROPAGATE_FAULT)
		return emulate_gp(ctxt, 0);

	if (r == X86EMUL_CONTINUE) {
		*reg_write(ctxt, VCPU_REGS_RAX) = (u32)msr_data;
		*reg_write(ctxt, VCPU_REGS_RDX) = msr_data >> 32;
	}
	return r;
}