x86/speculation: Regression fixes
cover x86/speculation: Regression fixes
Fix two regressions introudced by recent speculation mitigations on the 4.14 branch:
- Crash on older 32-bit processors
- Build warning from objtool
Ben.
Ben Hutchings (1):
Revert "x86/speculation: Change FILL_RETURN_BUFFER to work with
objtool"
Peter Zijlstra (1):
x86/nospec: Fix i386 RSB stuffing
arch/x86/include/asm/nospec-branch.h | 24 +++++++++++++++++++----- 1 file changed, 19 insertions(+), 5 deletions(-)
PATCH 1/2 x86/nospec: Fix i386 RSB stuffing
commit 332924973725e8cdcc783c175f68cf7e162cb9e5 upstream.
Turns out that i386 doesn't unconditionally have LFENCE, as such the loop in __FILL_RETURN_BUFFER isn't actually speculation safe on such chips.
[bwh: Backported to 4.14:
- __FILL_RETURN_BUFFER takes an sp parameter
- Open-code __FILL_RETURN_SLOT]
--- arch/x86/include/asm/nospec-branch.h | 14 ++++++++++++++ 1 file changed, 14 insertions(+)
arch/x86/include/asm/nospec-branch.h +14 -0
@@ -38,6 +38,7 @@* the optimal version — two calls, each with their own speculation * trap should their return address end up getting used, in a loop. */ +#ifdef CONFIG_X86_64 #define __FILL_RETURN_BUFFER (reg, nr, sp) \ mov $(nr/2), reg; \ 771: \
@@ -58,6 +59,19 @@ jnz 771b; \/* barrier for jnz misprediction */ \ lfence; +#else +/* +* i386 doesn't unconditionally have LFENCE, as such it can't +* do a loop. +*/ +#define __FILL_RETURN_BUFFER (reg, nr, sp) \ + .rept nr; \ + call 772f; \ + int3; \ +772:; \ + .endr; \ + add $(BITS_PER_LONG/8) * nr, sp; +#endif #define ISSUE_UNBALANCED_RET_GUARD (sp) \ call 992f; \
PATCH 2/2 Revert "x86/speculation: Change FILL_RETURN_BUFFER to work with objtool"
In reply to Ben HutchingsThis reverts commit c95afe5bcad40e1f0292bfc0a625c4aa080cc971, which was commit 089dd8e53126ebaf506e2dc0bf89d652c36bfc12 upstream.
The necessary changes to objtool have not been backported to 4.14. Backporting this commit alone only added build warnings.
--- arch/x86/include/asm/nospec-branch.h | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
arch/x86/include/asm/nospec-branch.h +5 -5
@@ -4,13 +4,11 @@#define _ASM_X86_NOSPEC_BRANCH_H_#include <linux/static_key.h> -#include <linux/frame.h> #include <asm/alternative.h> #include <asm/alternative-asm.h> #include <asm/cpufeatures.h> #include <asm/msr-index.h> -#include <asm/unwind_hints.h> #include <asm/percpu.h> /*
@@ -54,9 +52,9 @@lfence ; \ jmp 775b; \ 774: \ - add $(BITS_PER_LONG/8) * 2, sp; \dec reg; \ jnz 771b; \ + add $(BITS_PER_LONG/8) * nr, sp; \/* barrier for jnz misprediction */ \ lfence;#else
@@ -167,8 +165,10 @@* monstrosity above, manually. */ .macro FILL_RETURN_BUFFER reg:req nr:req ftr:req - ALTERNATIVE "jmp .Lskip_rsb_\@", "", \ftr - __FILL_RETURN_BUFFER(\reg,\nr,%_ASM_SP) + ANNOTATE_NOSPEC_ALTERNATIVE +ALTERNATIVE "jmp .Lskip_rsb_\@" , \ + __stringify(__FILL_RETURN_BUFFER(\reg,\nr,%_ASM_SP)) \ + \ftr .Lskip_rsb_\@: .endm
Greg KH
maintainerQuoted 12 lines · Ben Hutchings
Fix two regressions introudced by recent speculation mitigations on the 4.14 branch:
- Crash on older 32-bit processors
- Build warning from objtoolBen.
Ben Hutchings (1):
Revert "x86/speculation: ChangeFILL_RETURN_BUFFERto work with
objtool"
Quoted 6 lines
Peter Zijlstra (1):
x86/nospec: Fix i386 RSB stuffingarch/x86/include/asm/nospec-branch.h | 24 +++++++++++++++++++----- 1 file changed, 19 insertions(+), 5 deletions(-)
Now queued up, thanks.
greg k-h