Re: [PATCH] x86/retpoline: Fill return stack buffer on vmexit

[Woodhouse, David]

On Thu, 2018-01-11 at 09:49 +0100, Boris Petkov wrote:
> On January 11, 2018 9:42:38 AM GMT+01:00, Peter Zijlstra  wrote:
> >Or we teach the alternative thing to patch in a jmp to end instead of
> >NOP padding the entire thing as soon as the jmp (3 bytes) fits ?
> 
> Or, even better: use alternative_call() to call functions instead of patching gazillion bytes.

For this one I kind of wanted to keep it as a macro so we can select
which register it uses. I've taken the bulk of it out of the
ALTERNATIVE, and just switch between the first 'mov' instruction and a
jmp over the whole lot.

Looks like this now...

From 302622182f56825b7cf2c39ce88ea8c462d587fe Mon Sep 17 00:00:00 2001
From: David Woodhouse <dwmw@xxxxxxxxxxxx>
Date: Wed, 10 Jan 2018 22:32:24 +0000
Subject: [PATCH] x86/retpoline: Fill return stack buffer on vmexit

In accordance with the Intel and AMD documentation, we need to overwrite
all entries in the RSB on exiting a guest, to prevent malicious branch
target predictions from affecting the host kernel. This is needed both
for retpoline and for IBRS.

Signed-off-by: David Woodhouse <dwmw@xxxxxxxxxxxx>
---
Âarch/x86/include/asm/nospec-branch.h | 72 ++++++++++++++++++++++++++++++++++++
Âarch/x86/kvm/svm.cÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ|ÂÂ4 ++
Âarch/x86/kvm/vmx.cÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ|ÂÂ4 ++
Â3 files changed, 80 insertions(+)

diff --git a/arch/x86/include/asm/nospec-branch.h b/arch/x86/include/asm/nospec-branch.h
index 7d70ea9..8fbc8b9 100644
--- a/arch/x86/include/asm/nospec-branch.h
+++ b/arch/x86/include/asm/nospec-branch.h
@@ -7,6 +7,50 @@
Â#include <asm/alternative-asm.h>
Â#include <asm/cpufeatures.h>
Â
+/*
+ * Fill the CPU return stack buffer.
+ *
+ * Each entry in the RSB, if used for a speculative 'ret', contains an
+ * infinite 'pause; jmp' loop to capture speculative execution.
+ *
+ * This is required in various cases for retpoline and IBRS-based
+ * mitigations for the Spectre variant 2 vulnerability. Sometimes to
+ * eliminate potentially bogus entries from the RSB, and sometimes
+ * purely to ensure that it doesn't get empty, which on some CPUs would
+ * allow predictions from other (unwanted!) sources to be used.
+ *
+ * We define a CPP macro such that it can be used from both .S files and
+ * inline assembly. It's possible to do a .macro and then include that
+ * from C via asm(".include <asm/nospec-branch.h>") but let's not go there