Re: [PATCH 6/6] KVM: VMX: Move VERW closer to VMentry for MDS mitigation

From: Sean Christopherson <seanjc@google.com>
To: Pawan Gupta <pawan.kumar.gupta@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>,
	Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
	Dave Hansen <dave.hansen@linux.intel.com>,
	x86@kernel.org, "H. Peter Anvin" <hpa@zytor.com>,
	Peter Zijlstra <peterz@infradead.org>,
	Josh Poimboeuf <jpoimboe@kernel.org>,
	Andy Lutomirski <luto@kernel.org>,
	Jonathan Corbet <corbet@lwn.net>,
	Paolo Bonzini <pbonzini@redhat.com>,
	tony.luck@intel.com, ak@linux.intel.com,
	tim.c.chen@linux.intel.com, linux-kernel@vger.kernel.org,
	linux-doc@vger.kernel.org, kvm@vger.kernel.org,
	Alyssa Milburn <alyssa.milburn@linux.intel.com>,
	Daniel Sneddon <daniel.sneddon@linux.intel.com>,
	antonio.gomez.iglesias@linux.intel.com
Subject: Re: [PATCH  6/6] KVM: VMX: Move VERW closer to VMentry for MDS mitigation
Date: Fri, 20 Oct 2023 15:55:07 -0700	[thread overview]
Message-ID: <ZTMFS8I2s8EroSNe@google.com> (raw)
In-Reply-To: <20231020-delay-verw-v1-6-cff54096326d@linux.intel.com>

On Fri, Oct 20, 2023, Pawan Gupta wrote:
> During VMentry VERW is executed to mitigate MDS. After VERW, any memory
> access like register push onto stack may put host data in MDS affected
> CPU buffers. A guest can then use MDS to sample host data.
> 
> Although likelihood of secrets surviving in registers at current VERW
> callsite is less, but it can't be ruled out. Harden the MDS mitigation
> by moving the VERW mitigation late in VMentry path.
> 
> Note that VERW for MMIO Stale Data mitigation is unchanged because of
> the complexity of per-guest conditional VERW which is not easy to handle
> that late in asm with no GPRs available. If the CPU is also affected by
> MDS, VERW is unconditionally executed late in asm regardless of guest
> having MMIO access.
> 
> Signed-off-by: Pawan Gupta <pawan.kumar.gupta@linux.intel.com>
> ---
>  arch/x86/kvm/vmx/vmenter.S |  9 +++++++++
>  arch/x86/kvm/vmx/vmx.c     | 10 +++++++---
>  2 files changed, 16 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/x86/kvm/vmx/vmenter.S b/arch/x86/kvm/vmx/vmenter.S
> index be275a0410a8..efa716cf4727 100644
> --- a/arch/x86/kvm/vmx/vmenter.S
> +++ b/arch/x86/kvm/vmx/vmenter.S
> @@ -1,6 +1,7 @@
>  /* SPDX-License-Identifier: GPL-2.0 */
>  #include <linux/linkage.h>
>  #include <asm/asm.h>
> +#include <asm/segment.h>
>  #include <asm/bitsperlong.h>
>  #include <asm/kvm_vcpu_regs.h>
>  #include <asm/nospec-branch.h>
> @@ -31,6 +32,8 @@
>  #define VCPU_R15	__VCPU_REGS_R15 * WORD_SIZE
>  #endif
>  
> +#define GUEST_CLEAR_CPU_BUFFERS		USER_CLEAR_CPU_BUFFERS
> +
>  .macro VMX_DO_EVENT_IRQOFF call_insn call_target
>  	/*
>  	 * Unconditionally create a stack frame, getting the correct RSP on the
> @@ -177,10 +180,16 @@ SYM_FUNC_START(__vmx_vcpu_run)
>   * the 'vmx_vmexit' label below.
>   */
>  .Lvmresume:
> +	/* Mitigate CPU data sampling attacks .e.g. MDS */
> +	GUEST_CLEAR_CPU_BUFFERS

I have a very hard time believing that it's worth duplicating the mitigation
for VMRESUME vs. VMLAUNCH just to land it after a Jcc.

 3b1:   48 8b 00                mov    (%rax),%rax
 3b4:   74 18                   je     3ce <__vmx_vcpu_run+0x9e>
 3b6:   eb 0e                   jmp    3c6 <__vmx_vcpu_run+0x96>
 3b8:   0f 00 2d 05 00 00 00    verw   0x5(%rip)        # 3c4 <__vmx_vcpu_run+0x94>
 3bf:   0f 1f 80 00 00 18 00    nopl   0x180000(%rax)
 3c6:   0f 01 c3                vmresume
 3c9:   e9 c9 00 00 00          jmp    497 <vmx_vmexit+0xa7>
 3ce:   eb 0e                   jmp    3de <__vmx_vcpu_run+0xae>
 3d0:   0f 00 2d 05 00 00 00    verw   0x5(%rip)        # 3dc <__vmx_vcpu_run+0xac>
 3d7:   0f 1f 80 00 00 18 00    nopl   0x180000(%rax)
 3de:   0f 01 c2                vmlaunch

Also, would it'd be better to put the NOP first?  Or even better, out of line?
It'd be quite hilarious if the CPU pulled a stupid and speculated on the operand
of the NOP, i.e. if the user/guest controlled RAX allowed for pulling in data
after the VERW.