lkml.org 
[lkml]   [2018]   [Jan]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[PATCH v2.1] x86/retpoline: Fill return stack buffer on vmexit
Date
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@amazon.co.uk>
---
v2: Reduce the size of the ALTERNATIVE insns, fix .align (again)!
Sent in private email for testing, hence this second public post is
v2.1: Add CONFIG_RETPOLINE around RSB stuffing

Sorry, Josh, objtool broke again! :)

arch/x86/include/asm/nospec-branch.h | 76 ++++++++++++++++++++++++++++++++++++
arch/x86/kvm/svm.c | 4 ++
arch/x86/kvm/vmx.c | 4 ++
3 files changed, 84 insertions(+)

diff --git a/arch/x86/include/asm/nospec-branch.h b/arch/x86/include/asm/nospec-branch.h
index 7d70ea9..b55ff79 100644
--- a/arch/x86/include/asm/nospec-branch.h
+++ b/arch/x86/include/asm/nospec-branch.h
@@ -7,6 +7,50 @@
#include <asm/alternative-asm.h>
#include <asm/cpufeatures.h>

+/*
+ * 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.
+ */
+
+#define RSB_CLEAR_LOOPS 32 /* To forcibly overwrite all entries */
+#define RSB_FILL_LOOPS 16 /* To avoid underflow */
+
+#define __FILL_RETURN_BUFFER_PREP(reg, nr) \
+ mov $(nr/2), reg
+
+#define __FILL_RETURN_BUFFER(reg, nr, sp, uniq) \
+ .align 16; \
+.Ldo_call1_ ## uniq: \
+ call .Ldo_call2_ ## uniq; \
+.Ltrap1_ ## uniq: \
+ pause; \
+ jmp .Ltrap1_ ## uniq; \
+ .align 16; \
+.Ldo_call2_ ## uniq: \
+ call .Ldo_loop_ ## uniq; \
+.Ltrap2_ ## uniq: \
+ pause; \
+ jmp .Ltrap2_ ## uniq; \
+ .align 16; \
+.Ldo_loop_ ## uniq: \
+ dec reg; \
+ jnz .Ldo_call1_ ## uniq; \
+ add $(BITS_PER_LONG/8) * nr, sp; \
+.Lskip_rsb_ ## uniq:;
+
+
#ifdef __ASSEMBLY__

/*
@@ -61,6 +105,19 @@
#endif
.endm

+ /*
+ * A simpler FILL_RETURN_BUFFER macro. Don't make people use the CPP
+ * monstrosity above, manually.
+ */
+.macro FILL_RETURN_BUFFER reg:req nr:req ftr:req
+#ifdef CONFIG_RETPOLINE
+ ALTERNATIVE "jmp .Lskip_rsb_\@", \
+ __stringify(__FILL_RETURN_BUFFER_PREP(\reg,\nr)), \
+ \ftr
+ __FILL_RETURN_BUFFER(\reg,\nr,%_ASM_SP,\@)
+#endif
+.endm
+
#else /* __ASSEMBLY__ */

#if defined(CONFIG_X86_64) && defined(RETPOLINE)
@@ -115,5 +172,24 @@ enum spectre_v2_mitigation {
SPECTRE_V2_IBRS,
};

+/*
+ * On VMEXIT we must ensure that no RSB predictions learned in the guest
+ * can be followed in the host, by overwriting the RSB completely. Both
+ * retpoline and IBRS mitigations for Spectre v2 need this; only on future
+ * CPUs with IBRS_ATT *might* it be avoided.
+ */
+static inline void vmexit_fill_RSB(void)
+{
+#ifdef CONFIG_RETPOLINE
+ unsigned long loops = RSB_CLEAR_LOOPS / 2;
+
+ asm volatile (ALTERNATIVE("jmp .Lskip_rsb__%=",
+ __stringify(__FILL_RETURN_BUFFER_PREP(%0, RSB_CLEAR_LOOPS)),
+ X86_FEATURE_RETPOLINE)
+ __stringify(__FILL_RETURN_BUFFER(%0, RSB_CLEAR_LOOPS, %1, _%=))
+ : "=&r" (loops), ASM_CALL_CONSTRAINT
+ : "r" (loops) : "memory" );
+#endif
+}
#endif /* __ASSEMBLY__ */
#endif /* __NOSPEC_BRANCH_H__ */
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index 0e68f0b..2744b973 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -45,6 +45,7 @@
#include <asm/debugreg.h>
#include <asm/kvm_para.h>
#include <asm/irq_remapping.h>
+#include <asm/nospec-branch.h>

#include <asm/virtext.h>
#include "trace.h"
@@ -4985,6 +4986,9 @@ static void svm_vcpu_run(struct kvm_vcpu *vcpu)
#endif
);

+ /* Eliminate branch target predictions from guest mode */
+ vmexit_fill_RSB();
+
#ifdef CONFIG_X86_64
wrmsrl(MSR_GS_BASE, svm->host.gs_base);
#else
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 62ee436..d1e25db 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -50,6 +50,7 @@
#include <asm/apic.h>
#include <asm/irq_remapping.h>
#include <asm/mmu_context.h>
+#include <asm/nospec-branch.h>

#include "trace.h"
#include "pmu.h"
@@ -9403,6 +9404,9 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
#endif
);

+ /* Eliminate branch target predictions from guest mode */
+ vmexit_fill_RSB();
+
/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
if (debugctlmsr)
update_debugctlmsr(debugctlmsr);
--
2.7.4
\
 
 \ /
  Last update: 2018-01-14 23:23    [W:0.099 / U:5.980 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site