From 76b043848fd22dbf7f8bf3a1452f8c70d557b860 Mon Sep 17 00:00:00 2001 From: David Woodhouse Date: Thu, 11 Jan 2018 21:46:25 +0000 Subject: x86/retpoline: Add initial retpoline support Enable the use of -mindirect-branch=thunk-extern in newer GCC, and provide the corresponding thunks. Provide assembler macros for invoking the thunks in the same way that GCC does, from native and inline assembler. This adds X86_FEATURE_RETPOLINE and sets it by default on all CPUs. In some circumstances, IBRS microcode features may be used instead, and the retpoline can be disabled. On AMD CPUs if lfence is serialising, the retpoline can be dramatically simplified to a simple "lfence; jmp *\reg". A future patch, after it has been verified that lfence really is serialising in all circumstances, can enable this by setting the X86_FEATURE_RETPOLINE_AMD feature bit in addition to X86_FEATURE_RETPOLINE. Do not align the retpoline in the altinstr section, because there is no guarantee that it stays aligned when it's copied over the oldinstr during alternative patching. [ Andi Kleen: Rename the macros, add CONFIG_RETPOLINE option, export thunks] [ tglx: Put actual function CALL/JMP in front of the macros, convert to symbolic labels ] [ dwmw2: Convert back to numeric labels, merge objtool fixes ] Signed-off-by: David Woodhouse Signed-off-by: Thomas Gleixner Acked-by: Arjan van de Ven Acked-by: Ingo Molnar Cc: gnomes@lxorguk.ukuu.org.uk Cc: Rik van Riel Cc: Andi Kleen Cc: Josh Poimboeuf Cc: thomas.lendacky@amd.com Cc: Peter Zijlstra Cc: Linus Torvalds Cc: Jiri Kosina Cc: Andy Lutomirski Cc: Dave Hansen Cc: Kees Cook Cc: Tim Chen Cc: Greg Kroah-Hartman Cc: Paul Turner Link: https://lkml.kernel.org/r/1515707194-20531-4-git-send-email-dwmw@amazon.co.uk --- arch/x86/include/asm/nospec-branch.h | 128 +++++++++++++++++++++++++++++++++++ 1 file changed, 128 insertions(+) create mode 100644 arch/x86/include/asm/nospec-branch.h (limited to 'arch/x86/include/asm/nospec-branch.h') diff --git a/arch/x86/include/asm/nospec-branch.h b/arch/x86/include/asm/nospec-branch.h new file mode 100644 index 000000000000..e20e92ef2ca8 --- /dev/null +++ b/arch/x86/include/asm/nospec-branch.h @@ -0,0 +1,128 @@ +/* SPDX-License-Identifier: GPL-2.0 */ + +#ifndef __NOSPEC_BRANCH_H__ +#define __NOSPEC_BRANCH_H__ + +#include +#include +#include + +#ifdef __ASSEMBLY__ + +/* + * This should be used immediately before a retpoline alternative. It tells + * objtool where the retpolines are so that it can make sense of the control + * flow by just reading the original instruction(s) and ignoring the + * alternatives. + */ +.macro ANNOTATE_NOSPEC_ALTERNATIVE + .Lannotate_\@: + .pushsection .discard.nospec + .long .Lannotate_\@ - . + .popsection +.endm + +/* + * These are the bare retpoline primitives for indirect jmp and call. + * Do not use these directly; they only exist to make the ALTERNATIVE + * invocation below less ugly. + */ +.macro RETPOLINE_JMP reg:req + call .Ldo_rop_\@ +.Lspec_trap_\@: + pause + jmp .Lspec_trap_\@ +.Ldo_rop_\@: + mov \reg, (%_ASM_SP) + ret +.endm + +/* + * This is a wrapper around RETPOLINE_JMP so the called function in reg + * returns to the instruction after the macro. + */ +.macro RETPOLINE_CALL reg:req + jmp .Ldo_call_\@ +.Ldo_retpoline_jmp_\@: + RETPOLINE_JMP \reg +.Ldo_call_\@: + call .Ldo_retpoline_jmp_\@ +.endm + +/* + * JMP_NOSPEC and CALL_NOSPEC macros can be used instead of a simple + * indirect jmp/call which may be susceptible to the Spectre variant 2 + * attack. + */ +.macro JMP_NOSPEC reg:req +#ifdef CONFIG_RETPOLINE + ANNOTATE_NOSPEC_ALTERNATIVE + ALTERNATIVE_2 __stringify(jmp *\reg), \ + __stringify(RETPOLINE_JMP \reg), X86_FEATURE_RETPOLINE, \ + __stringify(lfence; jmp *\reg), X86_FEATURE_RETPOLINE_AMD +#else + jmp *\reg +#endif +.endm + +.macro CALL_NOSPEC reg:req +#ifdef CONFIG_RETPOLINE + ANNOTATE_NOSPEC_ALTERNATIVE + ALTERNATIVE_2 __stringify(call *\reg), \ + __stringify(RETPOLINE_CALL \reg), X86_FEATURE_RETPOLINE,\ + __stringify(lfence; call *\reg), X86_FEATURE_RETPOLINE_AMD +#else + call *\reg +#endif +.endm + +#else /* __ASSEMBLY__ */ + +#define ANNOTATE_NOSPEC_ALTERNATIVE \ + "999:\n\t" \ + ".pushsection .discard.nospec\n\t" \ + ".long 999b - .\n\t" \ + ".popsection\n\t" + +#if defined(CONFIG_X86_64) && defined(RETPOLINE) + +/* + * Since the inline asm uses the %V modifier which is only in newer GCC, + * the 64-bit one is dependent on RETPOLINE not CONFIG_RETPOLINE. + */ +# define CALL_NOSPEC \ + ANNOTATE_NOSPEC_ALTERNATIVE \ + ALTERNATIVE( \ + "call *%[thunk_target]\n", \ + "call __x86_indirect_thunk_%V[thunk_target]\n", \ + X86_FEATURE_RETPOLINE) +# define THUNK_TARGET(addr) [thunk_target] "r" (addr) + +#elif defined(CONFIG_X86_32) && defined(CONFIG_RETPOLINE) +/* + * For i386 we use the original ret-equivalent retpoline, because + * otherwise we'll run out of registers. We don't care about CET + * here, anyway. + */ +# define CALL_NOSPEC ALTERNATIVE("call *%[thunk_target]\n", \ + " jmp 904f;\n" \ + " .align 16\n" \ + "901: call 903f;\n" \ + "902: pause;\n" \ + " jmp 902b;\n" \ + " .align 16\n" \ + "903: addl $4, %%esp;\n" \ + " pushl %[thunk_target];\n" \ + " ret;\n" \ + " .align 16\n" \ + "904: call 901b;\n", \ + X86_FEATURE_RETPOLINE) + +# define THUNK_TARGET(addr) [thunk_target] "rm" (addr) +#else /* No retpoline */ +# define CALL_NOSPEC "call *%[thunk_target]\n" +# define THUNK_TARGET(addr) [thunk_target] "rm" (addr) +#endif + +#endif /* __ASSEMBLY__ */ +#endif /* __NOSPEC_BRANCH_H__ */ -- cgit v1.2.1 From da285121560e769cc31797bba6422eea71d473e0 Mon Sep 17 00:00:00 2001 From: David Woodhouse Date: Thu, 11 Jan 2018 21:46:26 +0000 Subject: x86/spectre: Add boot time option to select Spectre v2 mitigation Add a spectre_v2= option to select the mitigation used for the indirect branch speculation vulnerability. Currently, the only option available is retpoline, in its various forms. This will be expanded to cover the new IBRS/IBPB microcode features. The RETPOLINE_AMD feature relies on a serializing LFENCE for speculation control. For AMD hardware, only set RETPOLINE_AMD if LFENCE is a serializing instruction, which is indicated by the LFENCE_RDTSC feature. [ tglx: Folded back the LFENCE/AMD fixes and reworked it so IBRS integration becomes simple ] Signed-off-by: David Woodhouse Signed-off-by: Thomas Gleixner Cc: gnomes@lxorguk.ukuu.org.uk Cc: Rik van Riel Cc: Andi Kleen Cc: Josh Poimboeuf Cc: thomas.lendacky@amd.com Cc: Peter Zijlstra Cc: Linus Torvalds Cc: Jiri Kosina Cc: Andy Lutomirski Cc: Dave Hansen Cc: Kees Cook Cc: Tim Chen Cc: Greg Kroah-Hartman Cc: Paul Turner Link: https://lkml.kernel.org/r/1515707194-20531-5-git-send-email-dwmw@amazon.co.uk --- arch/x86/include/asm/nospec-branch.h | 10 ++++++++++ 1 file changed, 10 insertions(+) (limited to 'arch/x86/include/asm/nospec-branch.h') diff --git a/arch/x86/include/asm/nospec-branch.h b/arch/x86/include/asm/nospec-branch.h index e20e92ef2ca8..ea034fa6e261 100644 --- a/arch/x86/include/asm/nospec-branch.h +++ b/arch/x86/include/asm/nospec-branch.h @@ -124,5 +124,15 @@ # define THUNK_TARGET(addr) [thunk_target] "rm" (addr) #endif +/* The Spectre V2 mitigation variants */ +enum spectre_v2_mitigation { + SPECTRE_V2_NONE, + SPECTRE_V2_RETPOLINE_MINIMAL, + SPECTRE_V2_RETPOLINE_MINIMAL_AMD, + SPECTRE_V2_RETPOLINE_GENERIC, + SPECTRE_V2_RETPOLINE_AMD, + SPECTRE_V2_IBRS, +}; + #endif /* __ASSEMBLY__ */ #endif /* __NOSPEC_BRANCH_H__ */ -- cgit v1.2.1 From 117cc7a908c83697b0b737d15ae1eb5943afe35b Mon Sep 17 00:00:00 2001 From: David Woodhouse Date: Fri, 12 Jan 2018 11:11:27 +0000 Subject: 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. [ak: numbers again for the RSB stuffing labels] Signed-off-by: David Woodhouse Signed-off-by: Thomas Gleixner Tested-by: Peter Zijlstra (Intel) Cc: gnomes@lxorguk.ukuu.org.uk Cc: Rik van Riel Cc: Andi Kleen Cc: Josh Poimboeuf Cc: thomas.lendacky@amd.com Cc: Linus Torvalds Cc: Jiri Kosina Cc: Andy Lutomirski Cc: Dave Hansen Cc: Kees Cook Cc: Tim Chen Cc: Greg Kroah-Hartman Cc: Paul Turner Link: https://lkml.kernel.org/r/1515755487-8524-1-git-send-email-dwmw@amazon.co.uk --- arch/x86/include/asm/nospec-branch.h | 78 +++++++++++++++++++++++++++++++++++- 1 file changed, 77 insertions(+), 1 deletion(-) (limited to 'arch/x86/include/asm/nospec-branch.h') diff --git a/arch/x86/include/asm/nospec-branch.h b/arch/x86/include/asm/nospec-branch.h index ea034fa6e261..402a11c803c3 100644 --- a/arch/x86/include/asm/nospec-branch.h +++ b/arch/x86/include/asm/nospec-branch.h @@ -7,6 +7,48 @@ #include #include +/* + * 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 ") but let's not go there. + */ + +#define RSB_CLEAR_LOOPS 32 /* To forcibly overwrite all entries */ +#define RSB_FILL_LOOPS 16 /* To avoid underflow */ + +/* + * Google experimented with loop-unrolling and this turned out to be + * the optimal version — two calls, each with their own speculation + * trap should their return address end up getting used, in a loop. + */ +#define __FILL_RETURN_BUFFER(reg, nr, sp) \ + mov $(nr/2), reg; \ +771: \ + call 772f; \ +773: /* speculation trap */ \ + pause; \ + jmp 773b; \ +772: \ + call 774f; \ +775: /* speculation trap */ \ + pause; \ + jmp 775b; \ +774: \ + dec reg; \ + jnz 771b; \ + add $(BITS_PER_LONG/8) * nr, sp; + #ifdef __ASSEMBLY__ /* @@ -74,6 +116,20 @@ #else call *\reg #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 + ANNOTATE_NOSPEC_ALTERNATIVE + ALTERNATIVE "jmp .Lskip_rsb_\@", \ + __stringify(__FILL_RETURN_BUFFER(\reg,\nr,%_ASM_SP)) \ + \ftr +.Lskip_rsb_\@: +#endif .endm #else /* __ASSEMBLY__ */ @@ -119,7 +175,7 @@ X86_FEATURE_RETPOLINE) # define THUNK_TARGET(addr) [thunk_target] "rm" (addr) -#else /* No retpoline */ +#else /* No retpoline for C / inline asm */ # define CALL_NOSPEC "call *%[thunk_target]\n" # define THUNK_TARGET(addr) [thunk_target] "rm" (addr) #endif @@ -134,5 +190,25 @@ 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 (ANNOTATE_NOSPEC_ALTERNATIVE + ALTERNATIVE("jmp 910f", + __stringify(__FILL_RETURN_BUFFER(%0, RSB_CLEAR_LOOPS, %1)), + X86_FEATURE_RETPOLINE) + "910:" + : "=&r" (loops), ASM_CALL_CONSTRAINT + : "r" (loops) : "memory" ); +#endif +} #endif /* __ASSEMBLY__ */ #endif /* __NOSPEC_BRANCH_H__ */ -- cgit v1.2.1