This is a note to let you know that I've just added the patch titled x86,static_call: Use alternative RET encoding to the 5.10-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary The filename of the patch is: x86-static_call-use-alternative-ret-encoding.patch and it can be found in the queue-5.10 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable@xxxxxxxxxxxxxxx> know about it. >From foo@baz Tue Jul 12 05:07:35 PM CEST 2022 From: Peter Zijlstra <peterz@xxxxxxxxxxxxx> Date: Tue, 14 Jun 2022 23:15:39 +0200 Subject: x86,static_call: Use alternative RET encoding From: Peter Zijlstra <peterz@xxxxxxxxxxxxx> commit ee88d363d15617ff50ac24fab0ffec11113b2aeb upstream. In addition to teaching static_call about the new way to spell 'RET', there is an added complication in that static_call() is allowed to rewrite text before it is known which particular spelling is required. In order to deal with this; have a static_call specific fixup in the apply_return() 'alternative' patching routine that will rewrite the static_call trampoline to match the definite sequence. This in turn creates the problem of uniquely identifying static call trampolines. Currently trampolines are 8 bytes, the first 5 being the jmp.d32/ret sequence and the final 3 a byte sequence that spells out 'SCT'. This sequence is used in __static_call_validate() to ensure it is patching a trampoline and not a random other jmp.d32. That is, false-positives shouldn't be plenty, but aren't a big concern. OTOH the new __static_call_fixup() must not have false-positives, and 'SCT' decodes to the somewhat weird but semi plausible sequence: push %rbx rex.XB push %r12 Additionally, there are SLS concerns with immediate jumps. Combined it seems like a good moment to change the signature to a single 3 byte trap instruction that is unique to this usage and will not ever get generated by accident. As such, change the signature to: '0x0f, 0xb9, 0xcc', which decodes to: ud1 %esp, %ecx Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx> Signed-off-by: Borislav Petkov <bp@xxxxxxx> Reviewed-by: Josh Poimboeuf <jpoimboe@xxxxxxxxxx> Signed-off-by: Borislav Petkov <bp@xxxxxxx> [cascardo: skip validation as introduced by 2105a92748e8 ("static_call,x86: Robustify trampoline patching")] Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@xxxxxxxxxxxxx> [bwh: Backported to 5.10: adjust context] Signed-off-by: Ben Hutchings <ben@xxxxxxxxxxxxxxx> Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> --- arch/x86/include/asm/static_call.h | 17 ++++++++++++++++ arch/x86/kernel/alternative.c | 12 +++++++---- arch/x86/kernel/static_call.c | 38 ++++++++++++++++++++++++++++++++++++- 3 files changed, 62 insertions(+), 5 deletions(-) --- a/arch/x86/include/asm/static_call.h +++ b/arch/x86/include/asm/static_call.h @@ -21,6 +21,16 @@ * relative displacement across sections. */ +/* + * The trampoline is 8 bytes and of the general form: + * + * jmp.d32 \func + * ud1 %esp, %ecx + * + * That trailing #UD provides both a speculation stop and serves as a unique + * 3 byte signature identifying static call trampolines. Also see tramp_ud[] + * and __static_call_fixup(). + */ #define __ARCH_DEFINE_STATIC_CALL_TRAMP(name, insns) \ asm(".pushsection .static_call.text, \"ax\" \n" \ ".align 4 \n" \ @@ -34,8 +44,13 @@ #define ARCH_DEFINE_STATIC_CALL_TRAMP(name, func) \ __ARCH_DEFINE_STATIC_CALL_TRAMP(name, ".byte 0xe9; .long " #func " - (. + 4)") +#ifdef CONFIG_RETPOLINE +#define ARCH_DEFINE_STATIC_CALL_NULL_TRAMP(name) \ + __ARCH_DEFINE_STATIC_CALL_TRAMP(name, "jmp __x86_return_thunk") +#else #define ARCH_DEFINE_STATIC_CALL_NULL_TRAMP(name) \ __ARCH_DEFINE_STATIC_CALL_TRAMP(name, "ret; int3; nop; nop; nop") +#endif #define ARCH_ADD_TRAMP_KEY(name) \ @@ -44,4 +59,6 @@ ".long " STATIC_CALL_KEY_STR(name) " - . \n" \ ".popsection \n") +extern bool __static_call_fixup(void *tramp, u8 op, void *dest); + #endif /* _ASM_STATIC_CALL_H */ --- a/arch/x86/kernel/alternative.c +++ b/arch/x86/kernel/alternative.c @@ -693,18 +693,22 @@ void __init_or_module noinline apply_ret s32 *s; for (s = start; s < end; s++) { - void *addr = (void *)s + *s; + void *dest = NULL, *addr = (void *)s + *s; struct insn insn; int len, ret; u8 bytes[16]; - u8 op1; + u8 op; ret = insn_decode_kernel(&insn, addr); if (WARN_ON_ONCE(ret < 0)) continue; - op1 = insn.opcode.bytes[0]; - if (WARN_ON_ONCE(op1 != JMP32_INSN_OPCODE)) + op = insn.opcode.bytes[0]; + if (op == JMP32_INSN_OPCODE) + dest = addr + insn.length + insn.immediate.value; + + if (__static_call_fixup(addr, op, dest) || + WARN_ON_ONCE(dest != &__x86_return_thunk)) continue; DPRINTK("return thunk at: %pS (%px) len: %d to: %pS", --- a/arch/x86/kernel/static_call.c +++ b/arch/x86/kernel/static_call.c @@ -11,6 +11,13 @@ enum insn_type { RET = 3, /* tramp / site cond-tail-call */ }; +/* + * ud1 %esp, %ecx - a 3 byte #UD that is unique to trampolines, chosen such + * that there is no false-positive trampoline identification while also being a + * speculation stop. + */ +static const u8 tramp_ud[] = { 0x0f, 0xb9, 0xcc }; + static const u8 retinsn[] = { RET_INSN_OPCODE, 0xcc, 0xcc, 0xcc, 0xcc }; static void __ref __static_call_transform(void *insn, enum insn_type type, void *func) @@ -32,7 +39,10 @@ static void __ref __static_call_transfor break; case RET: - code = &retinsn; + if (cpu_feature_enabled(X86_FEATURE_RETHUNK)) + code = text_gen_insn(JMP32_INSN_OPCODE, insn, &__x86_return_thunk); + else + code = &retinsn; break; } @@ -97,3 +107,29 @@ void arch_static_call_transform(void *si mutex_unlock(&text_mutex); } EXPORT_SYMBOL_GPL(arch_static_call_transform); + +#ifdef CONFIG_RETPOLINE +/* + * This is called by apply_returns() to fix up static call trampolines, + * specifically ARCH_DEFINE_STATIC_CALL_NULL_TRAMP which is recorded as + * having a return trampoline. + * + * The problem is that static_call() is available before determining + * X86_FEATURE_RETHUNK and, by implication, running alternatives. + * + * This means that __static_call_transform() above can have overwritten the + * return trampoline and we now need to fix things up to be consistent. + */ +bool __static_call_fixup(void *tramp, u8 op, void *dest) +{ + if (memcmp(tramp+5, tramp_ud, 3)) { + /* Not a trampoline site, not our problem. */ + return false; + } + + if (op == RET_INSN_OPCODE || dest == &__x86_return_thunk) + __static_call_transform(tramp, RET, NULL); + + return true; +} +#endif Patches currently in stable-queue which might be from peterz@xxxxxxxxxxxxx are queue-5.10/objtool-cache-instruction-relocs.patch queue-5.10/x86-sev-avoid-using-__x86_return_thunk.patch queue-5.10/objtool-add-elf_create_undef_symbol.patch queue-5.10/x86-ftrace-use-alternative-ret-encoding.patch queue-5.10/objtool-re-add-unwind_hint_-save_restore.patch queue-5.10/x86-bugs-add-retbleed-ibpb.patch queue-5.10/x86-bugs-enable-stibp-for-jmp2ret.patch queue-5.10/x86-retpoline-cleanup-some-ifdefery.patch queue-5.10/objtool-handle-__sanitize_cov-tail-calls.patch queue-5.10/x86-prepare-asm-files-for-straight-line-speculation.patch queue-5.10/kvm-vmx-flatten-__vmx_vcpu_run.patch queue-5.10/x86-kvm-vmx-make-noinstr-clean.patch queue-5.10/objtool-x86-replace-alternatives-with-.retpoline_sites.patch queue-5.10/objtool-skip-magical-retpoline-.altinstr_replacement.patch queue-5.10/x86-retbleed-add-fine-grained-kconfig-knobs.patch queue-5.10/x86-cpu-amd-add-spectral-chicken.patch queue-5.10/objtool-add-straight-line-speculation-validation.patch queue-5.10/kvm-vmx-fix-ibrs-handling-after-vmexit.patch queue-5.10/kvm-vmx-prevent-guest-rsb-poisoning-attacks-with-eibrs.patch queue-5.10/x86-vsyscall_emu-64-don-t-use-ret-in-vsyscall-emulation.patch queue-5.10/tools-arch-update-arch-x86-lib-mem-cpy-set-_64.s-copies-used-in-perf-bench-mem-memcpy.patch queue-5.10/x86-add-straight-line-speculation-mitigation.patch queue-5.10/x86-add-magic-amd-return-thunk.patch queue-5.10/x86-bugs-keep-a-per-cpu-ia32_spec_ctrl-value.patch queue-5.10/x86-alternatives-optimize-optimize_nops.patch queue-5.10/x86-objtool-create-.return_sites.patch queue-5.10/crypto-x86-poly1305-fixup-sls.patch queue-5.10/x86-alternative-handle-jcc-__x86_indirect_thunk_-reg.patch queue-5.10/x86-kvm-fix-setcc-emulation-for-return-thunks.patch queue-5.10/objtool-fix-objtool-regression-on-x32-systems.patch queue-5.10/x86-alternative-relax-text_poke_bp-constraint.patch queue-5.10/x86-retpoline-swizzle-retpoline-thunk.patch queue-5.10/objtool-rework-the-elf_rebuild_reloc_section-logic.patch queue-5.10/x86-speculation-fix-firmware-entry-spec_ctrl-handling.patch queue-5.10/x86-retpoline-remove-unused-replacement-symbols.patch queue-5.10/objtool-fix-symbol-creation.patch queue-5.10/x86-speculation-add-spectre_v2-ibrs-option-to-support-kernel-ibrs.patch queue-5.10/bpf-x86-respect-x86_feature_retpoline.patch queue-5.10/objtool-fix-type-of-reloc-addend.patch queue-5.10/objtool-x86-rewrite-retpoline-thunk-calls.patch queue-5.10/x86-undo-return-thunk-damage.patch queue-5.10/x86-prepare-inline-asm-for-straight-line-speculation.patch queue-5.10/x86-alternative-support-alternative_ternary.patch queue-5.10/kvm-emulate-fix-setcc-emulation-function-offsets-with-sls.patch queue-5.10/objtool-handle-per-arch-retpoline-naming.patch queue-5.10/x86-retpoline-create-a-retpoline-thunk-array.patch queue-5.10/x86-retpoline-simplify-retpolines.patch queue-5.10/x86-asm-fix-register-order.patch queue-5.10/x86-speculation-fill-rsb-on-vmexit-for-ibrs.patch queue-5.10/objtool-add-entry-unret-validation.patch queue-5.10/objtool-keep-track-of-retpoline-call-sites.patch queue-5.10/kvm-vmx-convert-launched-argument-to-flags.patch queue-5.10/objtool-add-elf_create_reloc-helper.patch queue-5.10/objtool-make-.altinstructions-section-entry-size-consistent.patch queue-5.10/x86-bpf-use-alternative-ret-encoding.patch queue-5.10/x86-common-stamp-out-the-stepping-madness.patch queue-5.10/x86-bugs-split-spectre_v2_select_mitigation-and-spectre_v2_user_select_mitigation.patch queue-5.10/x86-bugs-report-intel-retbleed-vulnerability.patch queue-5.10/bpf-x86-simplify-computing-label-offsets.patch queue-5.10/x86-cpufeatures-move-retpoline-flags-to-word-11.patch queue-5.10/x86-speculation-fix-spec_ctrl-write-on-smt-state-change.patch queue-5.10/x86-retpoline-use-mfunction-return.patch queue-5.10/x86-xen-rename-sys-entry-points.patch queue-5.10/objtool-only-rewrite-unconditional-retpoline-thunk-calls.patch queue-5.10/x86-bugs-optimize-spec_ctrl-msr-writes.patch queue-5.10/x86-alternative-optimize-single-byte-nops-at-an-arbitrary-position.patch queue-5.10/objtool-fix-code-relocs-vs-weak-symbols.patch queue-5.10/x86-bugs-report-amd-retbleed-vulnerability.patch queue-5.10/x86-static_call-use-alternative-ret-encoding.patch queue-5.10/x86-speculation-fix-rsb-filling-with-config_retpoline-n.patch queue-5.10/x86-asm-fixup-odd-gen-for-each-reg.h-usage.patch queue-5.10/x86-alternative-add-debug-prints-to-apply_retpolines.patch queue-5.10/objtool-extract-elf_symbol_add.patch queue-5.10/x86-use-return-thunk-in-asm-code.patch queue-5.10/objtool-remove-reloc-symbol-type-checks-in-get_alt_entry.patch queue-5.10/objtool-classify-symbols.patch queue-5.10/intel_idle-disable-ibrs-during-long-idle.patch queue-5.10/objtool-correctly-handle-retpoline-thunk-calls.patch queue-5.10/objtool-fix-.symtab_shndx-handling-for-elf_create_undef_symbol.patch queue-5.10/x86-retpoline-move-the-retpoline-thunk-declarations-to-nospec-branch.h.patch queue-5.10/objtool-support-asm-jump-tables.patch queue-5.10/x86-alternative-implement-.retpoline_sites-support.patch queue-5.10/objtool-x86-ignore-__x86_indirect_alt_-symbols.patch queue-5.10/objtool-fix-sls-validation-for-kcov-tail-call-replacement.patch queue-5.10/x86-alternative-try-inline-spectre_v2-retpoline-amd.patch queue-5.10/x86-entry-remove-skip_r11rcx.patch queue-5.10/objtool-explicitly-avoid-self-modifying-code-in-.altinstr_replacement.patch queue-5.10/x86-speculation-use-cached-host-spec_ctrl-value-for-guest-entry-exit.patch queue-5.10/x86-bugs-add-amd-retbleed-boot-parameter.patch queue-5.10/objtool-create-reloc-sections-implicitly.patch queue-5.10/x86-entry-add-kernel-ibrs-implementation.patch queue-5.10/objtool-treat-.text.__x86.-as-noinstr.patch queue-5.10/x86-lib-atomic64_386_32-rename-things.patch queue-5.10/objtool-introduce-cfi-hash.patch queue-5.10/objtool-default-ignore-int3-for-unreachable.patch queue-5.10/objtool-extract-elf_strtab_concat.patch queue-5.10/objtool-teach-get_alt_entry-about-more-relocation-types.patch queue-5.10/objtool-update-retpoline-validation.patch