Merge tag 'objtool-core-2020-06-01' of git://git.kernel.org/pub/scm/linux/kernel...
authorLinus Torvalds <torvalds@linux-foundation.org>
Mon, 1 Jun 2020 20:13:00 +0000 (13:13 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Mon, 1 Jun 2020 20:13:00 +0000 (13:13 -0700)
Pull objtool updates from Ingo Molnar:
 "There are a lot of objtool changes in this cycle, all across the map:

   - Speed up objtool significantly, especially when there are large
     number of sections

   - Improve objtool's understanding of special instructions such as
     IRET, to reduce the number of annotations required

   - Implement 'noinstr' validation

   - Do baby steps for non-x86 objtool use

   - Simplify/fix retpoline decoding

   - Add vmlinux validation

   - Improve documentation

   - Fix various bugs and apply smaller cleanups"

* tag 'objtool-core-2020-06-01' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (54 commits)
  objtool: Enable compilation of objtool for all architectures
  objtool: Move struct objtool_file into arch-independent header
  objtool: Exit successfully when requesting help
  objtool: Add check_kcov_mode() to the uaccess safelist
  samples/ftrace: Fix asm function ELF annotations
  objtool: optimize add_dead_ends for split sections
  objtool: use gelf_getsymshndx to handle >64k sections
  objtool: Allow no-op CFI ops in alternatives
  x86/retpoline: Fix retpoline unwind
  x86: Change {JMP,CALL}_NOSPEC argument
  x86: Simplify retpoline declaration
  x86/speculation: Change FILL_RETURN_BUFFER to work with objtool
  objtool: Add support for intra-function calls
  objtool: Move the IRET hack into the arch decoder
  objtool: Remove INSN_STACK
  objtool: Make handle_insn_ops() unconditional
  objtool: Rework allocating stack_ops on decode
  objtool: UNWIND_HINT_RET_OFFSET should not check registers
  objtool: is_fentry_call() crashes if call has no destination
  x86,smap: Fix smap_{save,restore}() alternatives
  ...

46 files changed:
arch/x86/crypto/aesni-intel_asm.S
arch/x86/crypto/camellia-aesni-avx-asm_64.S
arch/x86/crypto/camellia-aesni-avx2-asm_64.S
arch/x86/crypto/crc32c-pcl-intel-asm_64.S
arch/x86/entry/entry_32.S
arch/x86/entry/entry_64.S
arch/x86/include/asm/GEN-for-each-reg.h [new file with mode: 0644]
arch/x86/include/asm/asm-prototypes.h
arch/x86/include/asm/nospec-branch.h
arch/x86/include/asm/orc_types.h
arch/x86/include/asm/processor.h
arch/x86/include/asm/smap.h
arch/x86/include/asm/unwind_hints.h
arch/x86/kernel/ftrace.c
arch/x86/kernel/ftrace_32.S
arch/x86/kernel/ftrace_64.S
arch/x86/lib/checksum_32.S
arch/x86/lib/retpoline.S
arch/x86/platform/efi/efi_stub_64.S
include/linux/frame.h
lib/Kconfig.debug
samples/ftrace/ftrace-direct-modify.c
samples/ftrace/ftrace-direct-too.c
samples/ftrace/ftrace-direct.c
scripts/link-vmlinux.sh
tools/arch/x86/include/asm/orc_types.h
tools/objtool/Build
tools/objtool/Documentation/stack-validation.txt
tools/objtool/Makefile
tools/objtool/arch.h
tools/objtool/arch/x86/decode.c
tools/objtool/arch/x86/include/cfi_regs.h [new file with mode: 0644]
tools/objtool/builtin-check.c
tools/objtool/builtin-orc.c
tools/objtool/builtin.h
tools/objtool/cfi.h
tools/objtool/check.c
tools/objtool/check.h
tools/objtool/elf.c
tools/objtool/elf.h
tools/objtool/objtool.c
tools/objtool/objtool.h [new file with mode: 0644]
tools/objtool/orc.h [deleted file]
tools/objtool/orc_dump.c
tools/objtool/orc_gen.c
tools/objtool/weak.c [new file with mode: 0644]

index cad6e1b..54e7d15 100644 (file)
@@ -2758,7 +2758,7 @@ SYM_FUNC_START(aesni_xts_crypt8)
        pxor INC, STATE4
        movdqu IV, 0x30(OUTP)
 
-       CALL_NOSPEC %r11
+       CALL_NOSPEC r11
 
        movdqu 0x00(OUTP), INC
        pxor INC, STATE1
@@ -2803,7 +2803,7 @@ SYM_FUNC_START(aesni_xts_crypt8)
        _aesni_gf128mul_x_ble()
        movups IV, (IVP)
 
-       CALL_NOSPEC %r11
+       CALL_NOSPEC r11
 
        movdqu 0x40(OUTP), INC
        pxor INC, STATE1
index d01ddd7..ecc0a9a 100644 (file)
@@ -1228,7 +1228,7 @@ SYM_FUNC_START_LOCAL(camellia_xts_crypt_16way)
        vpxor 14 * 16(%rax), %xmm15, %xmm14;
        vpxor 15 * 16(%rax), %xmm15, %xmm15;
 
-       CALL_NOSPEC %r9;
+       CALL_NOSPEC r9;
 
        addq $(16 * 16), %rsp;
 
index 563ef6e..0907243 100644 (file)
@@ -1339,7 +1339,7 @@ SYM_FUNC_START_LOCAL(camellia_xts_crypt_32way)
        vpxor 14 * 32(%rax), %ymm15, %ymm14;
        vpxor 15 * 32(%rax), %ymm15, %ymm15;
 
-       CALL_NOSPEC %r9;
+       CALL_NOSPEC r9;
 
        addq $(16 * 32), %rsp;
 
index 0e6690e..8501ec4 100644 (file)
@@ -75,7 +75,7 @@
 
 .text
 SYM_FUNC_START(crc_pcl)
-#define    bufp                %rdi
+#define    bufp                rdi
 #define    bufp_dw     %edi
 #define    bufp_w      %di
 #define    bufp_b      %dil
@@ -105,9 +105,9 @@ SYM_FUNC_START(crc_pcl)
        ## 1) ALIGN:
        ################################################################
 
-       mov     bufp, bufptmp           # rdi = *buf
-       neg     bufp
-       and     $7, bufp                # calculate the unalignment amount of
+       mov     %bufp, bufptmp          # rdi = *buf
+       neg     %bufp
+       and     $7, %bufp               # calculate the unalignment amount of
                                        # the address
        je      proc_block              # Skip if aligned
 
@@ -123,13 +123,13 @@ SYM_FUNC_START(crc_pcl)
 do_align:
        #### Calculate CRC of unaligned bytes of the buffer (if any)
        movq    (bufptmp), tmp          # load a quadward from the buffer
-       add     bufp, bufptmp           # align buffer pointer for quadword
+       add     %bufp, bufptmp          # align buffer pointer for quadword
                                        # processing
-       sub     bufp, len               # update buffer length
+       sub     %bufp, len              # update buffer length
 align_loop:
        crc32b  %bl, crc_init_dw        # compute crc32 of 1-byte
        shr     $8, tmp                 # get next byte
-       dec     bufp
+       dec     %bufp
        jne     align_loop
 
 proc_block:
@@ -169,10 +169,10 @@ continue_block:
        xor     crc2, crc2
 
        ## branch into array
-       lea     jump_table(%rip), bufp
-       movzxw  (bufp, %rax, 2), len
-       lea     crc_array(%rip), bufp
-       lea     (bufp, len, 1), bufp
+       lea     jump_table(%rip), %bufp
+       movzxw  (%bufp, %rax, 2), len
+       lea     crc_array(%rip), %bufp
+       lea     (%bufp, len, 1), %bufp
        JMP_NOSPEC bufp
 
        ################################################################
@@ -218,9 +218,9 @@ LABEL crc_ %i
        ## 4) Combine three results:
        ################################################################
 
-       lea     (K_table-8)(%rip), bufp         # first entry is for idx 1
+       lea     (K_table-8)(%rip), %bufp                # first entry is for idx 1
        shlq    $3, %rax                        # rax *= 8
-       pmovzxdq (bufp,%rax), %xmm0             # 2 consts: K1:K2
+       pmovzxdq (%bufp,%rax), %xmm0            # 2 consts: K1:K2
        leal    (%eax,%eax,2), %eax             # rax *= 3 (total *24)
        subq    %rax, tmp                       # tmp -= rax*24
 
index b67bae7..7e7ffb7 100644 (file)
@@ -816,7 +816,7 @@ SYM_CODE_START(ret_from_fork)
 
        /* kernel thread */
 1:     movl    %edi, %eax
-       CALL_NOSPEC %ebx
+       CALL_NOSPEC ebx
        /*
         * A kernel thread is allowed to return here after successfully
         * calling do_execve().  Exit to userspace to complete the execve()
@@ -1501,7 +1501,7 @@ SYM_CODE_START_LOCAL_NOALIGN(common_exception_read_cr2)
 
        TRACE_IRQS_OFF
        movl    %esp, %eax                      # pt_regs pointer
-       CALL_NOSPEC %edi
+       CALL_NOSPEC edi
        jmp     ret_from_exception
 SYM_CODE_END(common_exception_read_cr2)
 
@@ -1522,7 +1522,7 @@ SYM_CODE_START_LOCAL_NOALIGN(common_exception)
 
        TRACE_IRQS_OFF
        movl    %esp, %eax                      # pt_regs pointer
-       CALL_NOSPEC %edi
+       CALL_NOSPEC edi
        jmp     ret_from_exception
 SYM_CODE_END(common_exception)
 
index 3063aa9..64fe3d8 100644 (file)
@@ -348,7 +348,7 @@ SYM_CODE_START(ret_from_fork)
        /* kernel thread */
        UNWIND_HINT_EMPTY
        movq    %r12, %rdi
-       CALL_NOSPEC %rbx
+       CALL_NOSPEC rbx
        /*
         * A kernel thread is allowed to return here after successfully
         * calling do_execve().  Exit to userspace to complete the execve()
diff --git a/arch/x86/include/asm/GEN-for-each-reg.h b/arch/x86/include/asm/GEN-for-each-reg.h
new file mode 100644 (file)
index 0000000..1b07fb1
--- /dev/null
@@ -0,0 +1,25 @@
+#ifdef CONFIG_64BIT
+GEN(rax)
+GEN(rbx)
+GEN(rcx)
+GEN(rdx)
+GEN(rsi)
+GEN(rdi)
+GEN(rbp)
+GEN(r8)
+GEN(r9)
+GEN(r10)
+GEN(r11)
+GEN(r12)
+GEN(r13)
+GEN(r14)
+GEN(r15)
+#else
+GEN(eax)
+GEN(ebx)
+GEN(ecx)
+GEN(edx)
+GEN(esi)
+GEN(edi)
+GEN(ebp)
+#endif
index ce92c4a..9bf2620 100644 (file)
@@ -17,24 +17,19 @@ extern void cmpxchg8b_emu(void);
 #endif
 
 #ifdef CONFIG_RETPOLINE
-#ifdef CONFIG_X86_32
-#define INDIRECT_THUNK(reg) extern asmlinkage void __x86_indirect_thunk_e ## reg(void);
-#else
-#define INDIRECT_THUNK(reg) extern asmlinkage void __x86_indirect_thunk_r ## reg(void);
-INDIRECT_THUNK(8)
-INDIRECT_THUNK(9)
-INDIRECT_THUNK(10)
-INDIRECT_THUNK(11)
-INDIRECT_THUNK(12)
-INDIRECT_THUNK(13)
-INDIRECT_THUNK(14)
-INDIRECT_THUNK(15)
-#endif
-INDIRECT_THUNK(ax)
-INDIRECT_THUNK(bx)
-INDIRECT_THUNK(cx)
-INDIRECT_THUNK(dx)
-INDIRECT_THUNK(si)
-INDIRECT_THUNK(di)
-INDIRECT_THUNK(bp)
+
+#define DECL_INDIRECT_THUNK(reg) \
+       extern asmlinkage void __x86_indirect_thunk_ ## reg (void);
+
+#define DECL_RETPOLINE(reg) \
+       extern asmlinkage void __x86_retpoline_ ## reg (void);
+
+#undef GEN
+#define GEN(reg) DECL_INDIRECT_THUNK(reg)
+#include <asm/GEN-for-each-reg.h>
+
+#undef GEN
+#define GEN(reg) DECL_RETPOLINE(reg)
+#include <asm/GEN-for-each-reg.h>
+
 #endif /* CONFIG_RETPOLINE */
index 7e9a281..d52d1aa 100644 (file)
@@ -4,20 +4,13 @@
 #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>
-
-/*
- * 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.
- */
-#define ANNOTATE_NOSPEC_ALTERNATIVE \
-       ANNOTATE_IGNORE_ALTERNATIVE
+#include <asm/unwind_hints.h>
 
 /*
  * Fill the CPU return stack buffer.
 #define __FILL_RETURN_BUFFER(reg, nr, sp)      \
        mov     $(nr/2), reg;                   \
 771:                                           \
+       ANNOTATE_INTRA_FUNCTION_CALL;           \
        call    772f;                           \
 773:   /* speculation trap */                  \
+       UNWIND_HINT_EMPTY;                      \
        pause;                                  \
        lfence;                                 \
        jmp     773b;                           \
 772:                                           \
+       ANNOTATE_INTRA_FUNCTION_CALL;           \
        call    774f;                           \
 775:   /* speculation trap */                  \
+       UNWIND_HINT_EMPTY;                      \
        pause;                                  \
        lfence;                                 \
        jmp     775b;                           \
 774:                                           \
+       add     $(BITS_PER_LONG/8) * 2, sp;     \
        dec     reg;                            \
-       jnz     771b;                           \
-       add     $(BITS_PER_LONG/8) * nr, sp;
+       jnz     771b;
 
 #ifdef __ASSEMBLY__
 
        .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
-       lfence
-       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
  */
 .macro JMP_NOSPEC reg:req
 #ifdef CONFIG_RETPOLINE
-       ANNOTATE_NOSPEC_ALTERNATIVE
-       ALTERNATIVE_2 __stringify(ANNOTATE_RETPOLINE_SAFE; jmp *\reg),  \
-               __stringify(RETPOLINE_JMP \reg), X86_FEATURE_RETPOLINE, \
-               __stringify(lfence; ANNOTATE_RETPOLINE_SAFE; jmp *\reg), X86_FEATURE_RETPOLINE_AMD
+       ALTERNATIVE_2 __stringify(ANNOTATE_RETPOLINE_SAFE; jmp *%\reg), \
+                     __stringify(jmp __x86_retpoline_\reg), X86_FEATURE_RETPOLINE, \
+                     __stringify(lfence; ANNOTATE_RETPOLINE_SAFE; jmp *%\reg), X86_FEATURE_RETPOLINE_AMD
 #else
-       jmp     *\reg
+       jmp     *%\reg
 #endif
 .endm
 
 .macro CALL_NOSPEC reg:req
 #ifdef CONFIG_RETPOLINE
-       ANNOTATE_NOSPEC_ALTERNATIVE
-       ALTERNATIVE_2 __stringify(ANNOTATE_RETPOLINE_SAFE; call *\reg), \
-               __stringify(RETPOLINE_CALL \reg), X86_FEATURE_RETPOLINE,\
-               __stringify(lfence; ANNOTATE_RETPOLINE_SAFE; call *\reg), X86_FEATURE_RETPOLINE_AMD
+       ALTERNATIVE_2 __stringify(ANNOTATE_RETPOLINE_SAFE; call *%\reg), \
+                     __stringify(call __x86_retpoline_\reg), X86_FEATURE_RETPOLINE, \
+                     __stringify(lfence; ANNOTATE_RETPOLINE_SAFE; call *%\reg), X86_FEATURE_RETPOLINE_AMD
 #else
-       call    *\reg
+       call    *%\reg
 #endif
 .endm
 
   */
 .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
+       ALTERNATIVE "jmp .Lskip_rsb_\@", "", \ftr
+       __FILL_RETURN_BUFFER(\reg,\nr,%_ASM_SP)
 .Lskip_rsb_\@:
 #endif
 .endm
  * which is ensured when CONFIG_RETPOLINE is defined.
  */
 # define CALL_NOSPEC                                           \
-       ANNOTATE_NOSPEC_ALTERNATIVE                             \
        ALTERNATIVE_2(                                          \
        ANNOTATE_RETPOLINE_SAFE                                 \
        "call *%[thunk_target]\n",                              \
-       "call __x86_indirect_thunk_%V[thunk_target]\n",         \
+       "call __x86_retpoline_%V[thunk_target]\n",              \
        X86_FEATURE_RETPOLINE,                                  \
        "lfence;\n"                                             \
        ANNOTATE_RETPOLINE_SAFE                                 \
        "call *%[thunk_target]\n",                              \
        X86_FEATURE_RETPOLINE_AMD)
+
 # define THUNK_TARGET(addr) [thunk_target] "r" (addr)
 
 #else /* CONFIG_X86_32 */
  * here, anyway.
  */
 # define CALL_NOSPEC                                           \
-       ANNOTATE_NOSPEC_ALTERNATIVE                             \
        ALTERNATIVE_2(                                          \
        ANNOTATE_RETPOLINE_SAFE                                 \
        "call *%[thunk_target]\n",                              \
index 6e06090..d255349 100644 (file)
@@ -58,8 +58,7 @@
 #define ORC_TYPE_CALL                  0
 #define ORC_TYPE_REGS                  1
 #define ORC_TYPE_REGS_IRET             2
-#define UNWIND_HINT_TYPE_SAVE          3
-#define UNWIND_HINT_TYPE_RESTORE       4
+#define UNWIND_HINT_TYPE_RET_OFFSET    3
 
 #ifndef __ASSEMBLY__
 /*
index c4e8fd7..29ee0c0 100644 (file)
@@ -728,7 +728,6 @@ static inline void sync_core(void)
        unsigned int tmp;
 
        asm volatile (
-               UNWIND_HINT_SAVE
                "mov %%ss, %0\n\t"
                "pushq %q0\n\t"
                "pushq %%rsp\n\t"
@@ -738,7 +737,6 @@ static inline void sync_core(void)
                "pushq %q0\n\t"
                "pushq $1f\n\t"
                "iretq\n\t"
-               UNWIND_HINT_RESTORE
                "1:"
                : "=&r" (tmp), ASM_CALL_CONSTRAINT : : "cc", "memory");
 #endif
index 27c47d1..8b58d69 100644 (file)
@@ -57,8 +57,10 @@ static __always_inline unsigned long smap_save(void)
 {
        unsigned long flags;
 
-       asm volatile (ALTERNATIVE("", "pushf; pop %0; " __ASM_CLAC,
-                                 X86_FEATURE_SMAP)
+       asm volatile ("# smap_save\n\t"
+                     ALTERNATIVE("jmp 1f", "", X86_FEATURE_SMAP)
+                     "pushf; pop %0; " __ASM_CLAC "\n\t"
+                     "1:"
                      : "=rm" (flags) : : "memory", "cc");
 
        return flags;
@@ -66,7 +68,10 @@ static __always_inline unsigned long smap_save(void)
 
 static __always_inline void smap_restore(unsigned long flags)
 {
-       asm volatile (ALTERNATIVE("", "push %0; popf", X86_FEATURE_SMAP)
+       asm volatile ("# smap_restore\n\t"
+                     ALTERNATIVE("jmp 1f", "", X86_FEATURE_SMAP)
+                     "push %0; popf\n\t"
+                     "1:"
                      : : "g" (flags) : "memory", "cc");
 }
 
index f5e2eb1..7d903fd 100644 (file)
        UNWIND_HINT sp_offset=\sp_offset
 .endm
 
-.macro UNWIND_HINT_SAVE
-       UNWIND_HINT type=UNWIND_HINT_TYPE_SAVE
-.endm
-
-.macro UNWIND_HINT_RESTORE
-       UNWIND_HINT type=UNWIND_HINT_TYPE_RESTORE
+/*
+ * RET_OFFSET: Used on instructions that terminate a function; mostly RETURN
+ * and sibling calls. On these, sp_offset denotes the expected offset from
+ * initial_func_cfi.
+ */
+.macro UNWIND_HINT_RET_OFFSET sp_offset=8
+       UNWIND_HINT type=UNWIND_HINT_TYPE_RET_OFFSET sp_offset=\sp_offset
 .endm
 
-#else /* !__ASSEMBLY__ */
-
-#define UNWIND_HINT(sp_reg, sp_offset, type, end)              \
-       "987: \n\t"                                             \
-       ".pushsection .discard.unwind_hints\n\t"                \
-       /* struct unwind_hint */                                \
-       ".long 987b - .\n\t"                                    \
-       ".short " __stringify(sp_offset) "\n\t"                 \
-       ".byte " __stringify(sp_reg) "\n\t"                     \
-       ".byte " __stringify(type) "\n\t"                       \
-       ".byte " __stringify(end) "\n\t"                        \
-       ".balign 4 \n\t"                                        \
-       ".popsection\n\t"
-
-#define UNWIND_HINT_SAVE UNWIND_HINT(0, 0, UNWIND_HINT_TYPE_SAVE, 0)
-
-#define UNWIND_HINT_RESTORE UNWIND_HINT(0, 0, UNWIND_HINT_TYPE_RESTORE, 0)
-
 #endif /* __ASSEMBLY__ */
 
 #endif /* _ASM_X86_UNWIND_HINTS_H */
index b0e6417..c84d28e 100644 (file)
@@ -282,7 +282,8 @@ static inline void tramp_free(void *tramp) { }
 
 /* Defined as markers to the end of the ftrace default trampolines */
 extern void ftrace_regs_caller_end(void);
-extern void ftrace_epilogue(void);
+extern void ftrace_regs_caller_ret(void);
+extern void ftrace_caller_end(void);
 extern void ftrace_caller_op_ptr(void);
 extern void ftrace_regs_caller_op_ptr(void);
 
@@ -334,7 +335,7 @@ create_trampoline(struct ftrace_ops *ops, unsigned int *tramp_size)
                call_offset = (unsigned long)ftrace_regs_call;
        } else {
                start_offset = (unsigned long)ftrace_caller;
-               end_offset = (unsigned long)ftrace_epilogue;
+               end_offset = (unsigned long)ftrace_caller_end;
                op_offset = (unsigned long)ftrace_caller_op_ptr;
                call_offset = (unsigned long)ftrace_call;
        }
@@ -366,6 +367,13 @@ create_trampoline(struct ftrace_ops *ops, unsigned int *tramp_size)
        if (WARN_ON(ret < 0))
                goto fail;
 
+       if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
+               ip = trampoline + (ftrace_regs_caller_ret - ftrace_regs_caller);
+               ret = probe_kernel_read(ip, (void *)retq, RET_SIZE);
+               if (WARN_ON(ret < 0))
+                       goto fail;
+       }
+
        /*
         * The address of the ftrace_ops that is used for this trampoline
         * is stored at the end of the trampoline. This will be used to
@@ -433,7 +441,7 @@ void set_ftrace_ops_ro(void)
                        end_offset = (unsigned long)ftrace_regs_caller_end;
                } else {
                        start_offset = (unsigned long)ftrace_caller;
-                       end_offset = (unsigned long)ftrace_epilogue;
+                       end_offset = (unsigned long)ftrace_caller_end;
                }
                size = end_offset - start_offset;
                size = size + RET_SIZE + sizeof(void *);
index e8a9f83..e405fe1 100644 (file)
@@ -189,5 +189,5 @@ return_to_handler:
        movl    %eax, %ecx
        popl    %edx
        popl    %eax
-       JMP_NOSPEC %ecx
+       JMP_NOSPEC ecx
 #endif
index 369e61f..aa5d28a 100644 (file)
@@ -23,7 +23,7 @@
 #endif /* CONFIG_FRAME_POINTER */
 
 /* Size of stack used to save mcount regs in save_mcount_regs */
-#define MCOUNT_REG_SIZE                (SS+8 + MCOUNT_FRAME_SIZE)
+#define MCOUNT_REG_SIZE                (FRAME_SIZE + MCOUNT_FRAME_SIZE)
 
 /*
  * gcc -pg option adds a call to 'mcount' in most functions.
@@ -77,7 +77,7 @@
        /*
         * We add enough stack to save all regs.
         */
-       subq $(MCOUNT_REG_SIZE - MCOUNT_FRAME_SIZE), %rsp
+       subq $(FRAME_SIZE), %rsp
        movq %rax, RAX(%rsp)
        movq %rcx, RCX(%rsp)
        movq %rdx, RDX(%rsp)
@@ -157,8 +157,12 @@ SYM_INNER_LABEL(ftrace_call, SYM_L_GLOBAL)
         * think twice before adding any new code or changing the
         * layout here.
         */
-SYM_INNER_LABEL(ftrace_epilogue, SYM_L_GLOBAL)
+SYM_INNER_LABEL(ftrace_caller_end, SYM_L_GLOBAL)
 
+       jmp ftrace_epilogue
+SYM_FUNC_END(ftrace_caller);
+
+SYM_FUNC_START(ftrace_epilogue)
 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 SYM_INNER_LABEL(ftrace_graph_call, SYM_L_GLOBAL)
        jmp ftrace_stub
@@ -170,14 +174,12 @@ SYM_INNER_LABEL(ftrace_graph_call, SYM_L_GLOBAL)
  */
 SYM_INNER_LABEL_ALIGN(ftrace_stub, SYM_L_WEAK)
        retq
-SYM_FUNC_END(ftrace_caller)
+SYM_FUNC_END(ftrace_epilogue)
 
 SYM_FUNC_START(ftrace_regs_caller)
        /* Save the current flags before any operations that can change them */
        pushfq
 
-       UNWIND_HINT_SAVE
-
        /* added 8 bytes to save flags */
        save_mcount_regs 8
        /* save_mcount_regs fills in first two parameters */
@@ -233,10 +235,13 @@ SYM_INNER_LABEL(ftrace_regs_call, SYM_L_GLOBAL)
        movq ORIG_RAX(%rsp), %rax
        movq %rax, MCOUNT_REG_SIZE-8(%rsp)
 
-       /* If ORIG_RAX is anything but zero, make this a call to that */
+       /*
+        * If ORIG_RAX is anything but zero, make this a call to that.
+        * See arch_ftrace_set_direct_caller().
+        */
        movq ORIG_RAX(%rsp), %rax
-       cmpq    $0, %rax
-       je      1f
+       testq   %rax, %rax
+       jz      1f
 
        /* Swap the flags with orig_rax */
        movq MCOUNT_REG_SIZE(%rsp), %rdi
@@ -244,20 +249,14 @@ SYM_INNER_LABEL(ftrace_regs_call, SYM_L_GLOBAL)
        movq %rax, MCOUNT_REG_SIZE(%rsp)
 
        restore_mcount_regs 8
+       /* Restore flags */
+       popfq
 
-       jmp     2f
+SYM_INNER_LABEL(ftrace_regs_caller_ret, SYM_L_GLOBAL);
+       UNWIND_HINT_RET_OFFSET
+       jmp     ftrace_epilogue
 
 1:     restore_mcount_regs
-
-
-2:
-       /*
-        * The stack layout is nondetermistic here, depending on which path was
-        * taken.  This confuses objtool and ORC, rightfully so.  For now,
-        * pretend the stack always looks like the non-direct case.
-        */
-       UNWIND_HINT_RESTORE
-
        /* Restore flags */
        popfq
 
@@ -268,7 +267,6 @@ SYM_INNER_LABEL(ftrace_regs_call, SYM_L_GLOBAL)
         * to the return.
         */
 SYM_INNER_LABEL(ftrace_regs_caller_end, SYM_L_GLOBAL)
-
        jmp ftrace_epilogue
 
 SYM_FUNC_END(ftrace_regs_caller)
@@ -303,7 +301,7 @@ trace:
         * function tracing is enabled.
         */
        movq ftrace_trace_function, %r8
-       CALL_NOSPEC %r8
+       CALL_NOSPEC r8
        restore_mcount_regs
 
        jmp fgraph_trace
@@ -340,6 +338,6 @@ SYM_CODE_START(return_to_handler)
        movq 8(%rsp), %rdx
        movq (%rsp), %rax
        addq $24, %rsp
-       JMP_NOSPEC %rdi
+       JMP_NOSPEC rdi
 SYM_CODE_END(return_to_handler)
 #endif
index 4742e8f..d1d7689 100644 (file)
@@ -153,7 +153,7 @@ SYM_FUNC_START(csum_partial)
        negl %ebx
        lea 45f(%ebx,%ebx,2), %ebx
        testl %esi, %esi
-       JMP_NOSPEC %ebx
+       JMP_NOSPEC ebx
 
        # Handle 2-byte-aligned regions
 20:    addw (%esi), %ax
@@ -436,7 +436,7 @@ SYM_FUNC_START(csum_partial_copy_generic)
        andl $-32,%edx
        lea 3f(%ebx,%ebx), %ebx
        testl %esi, %esi 
-       JMP_NOSPEC %ebx
+       JMP_NOSPEC ebx
 1:     addl $64,%esi
        addl $64,%edi 
        SRC(movb -32(%edx),%bl) ; SRC(movb (%edx),%bl)
index 363ec13..b4c43a9 100644 (file)
@@ -7,15 +7,31 @@
 #include <asm/alternative-asm.h>
 #include <asm/export.h>
 #include <asm/nospec-branch.h>
+#include <asm/unwind_hints.h>
+#include <asm/frame.h>
 
 .macro THUNK reg
        .section .text.__x86.indirect_thunk
 
+       .align 32
 SYM_FUNC_START(__x86_indirect_thunk_\reg)
-       CFI_STARTPROC
-       JMP_NOSPEC %\reg
-       CFI_ENDPROC
+       JMP_NOSPEC \reg
 SYM_FUNC_END(__x86_indirect_thunk_\reg)
+
+SYM_FUNC_START_NOALIGN(__x86_retpoline_\reg)
+       ANNOTATE_INTRA_FUNCTION_CALL
+       call    .Ldo_rop_\@
+.Lspec_trap_\@:
+       UNWIND_HINT_EMPTY
+       pause
+       lfence
+       jmp     .Lspec_trap_\@
+.Ldo_rop_\@:
+       mov     %\reg, (%_ASM_SP)
+       UNWIND_HINT_RET_OFFSET
+       ret
+SYM_FUNC_END(__x86_retpoline_\reg)
+
 .endm
 
 /*
@@ -24,25 +40,24 @@ SYM_FUNC_END(__x86_indirect_thunk_\reg)
  * only see one instance of "__x86_indirect_thunk_\reg" rather
  * than one per register with the correct names. So we do it
  * the simple and nasty way...
+ *
+ * Worse, you can only have a single EXPORT_SYMBOL per line,
+ * and CPP can't insert newlines, so we have to repeat everything
+ * at least twice.
  */
-#define __EXPORT_THUNK(sym) _ASM_NOKPROBE(sym); EXPORT_SYMBOL(sym)
-#define EXPORT_THUNK(reg) __EXPORT_THUNK(__x86_indirect_thunk_ ## reg)
-#define GENERATE_THUNK(reg) THUNK reg ; EXPORT_THUNK(reg)
-
-GENERATE_THUNK(_ASM_AX)
-GENERATE_THUNK(_ASM_BX)
-GENERATE_THUNK(_ASM_CX)
-GENERATE_THUNK(_ASM_DX)
-GENERATE_THUNK(_ASM_SI)
-GENERATE_THUNK(_ASM_DI)
-GENERATE_THUNK(_ASM_BP)
-#ifdef CONFIG_64BIT
-GENERATE_THUNK(r8)
-GENERATE_THUNK(r9)
-GENERATE_THUNK(r10)
-GENERATE_THUNK(r11)
-GENERATE_THUNK(r12)
-GENERATE_THUNK(r13)
-GENERATE_THUNK(r14)
-GENERATE_THUNK(r15)
-#endif
+
+#define __EXPORT_THUNK(sym)    _ASM_NOKPROBE(sym); EXPORT_SYMBOL(sym)
+#define EXPORT_THUNK(reg)      __EXPORT_THUNK(__x86_indirect_thunk_ ## reg)
+#define EXPORT_RETPOLINE(reg)  __EXPORT_THUNK(__x86_retpoline_ ## reg)
+
+#undef GEN
+#define GEN(reg) THUNK reg
+#include <asm/GEN-for-each-reg.h>
+
+#undef GEN
+#define GEN(reg) EXPORT_THUNK(reg)
+#include <asm/GEN-for-each-reg.h>
+
+#undef GEN
+#define GEN(reg) EXPORT_RETPOLINE(reg)
+#include <asm/GEN-for-each-reg.h>
index 15da118..90380a1 100644 (file)
@@ -21,7 +21,7 @@ SYM_FUNC_START(__efi_call)
        mov %r8, %r9
        mov %rcx, %r8
        mov %rsi, %rcx
-       CALL_NOSPEC %rdi
+       CALL_NOSPEC rdi
        leave
        ret
 SYM_FUNC_END(__efi_call)
index 02d3ca2..303cda6 100644 (file)
        static void __used __section(.discard.func_stack_frame_non_standard) \
                *__func_stack_frame_non_standard_##func = func
 
+/*
+ * This macro indicates that the following intra-function call is valid.
+ * Any non-annotated intra-function call will cause objtool to issue a warning.
+ */
+#define ANNOTATE_INTRA_FUNCTION_CALL                           \
+       999:                                                    \
+       .pushsection .discard.intra_function_calls;             \
+       .long 999b;                                             \
+       .popsection;
+
 #else /* !CONFIG_STACK_VALIDATION */
 
 #define STACK_FRAME_NON_STANDARD(func)
+#define ANNOTATE_INTRA_FUNCTION_CALL
 
 #endif /* CONFIG_STACK_VALIDATION */
 
index f3322a6..613ee4f 100644 (file)
@@ -369,6 +369,11 @@ config STACK_VALIDATION
          For more information, see
          tools/objtool/Documentation/stack-validation.txt.
 
+config VMLINUX_VALIDATION
+       bool
+       depends on STACK_VALIDATION && DEBUG_ENTRY && !PARAVIRT
+       default y
+
 config DEBUG_FORCE_WEAK_PER_CPU
        bool "Force weak per-cpu definitions"
        depends on DEBUG_KERNEL
index e04229d..c13a5bc 100644 (file)
@@ -20,18 +20,22 @@ static unsigned long my_ip = (unsigned long)schedule;
 
 asm (
 "      .pushsection    .text, \"ax\", @progbits\n"
+"      .type           my_tramp1, @function\n"
 "   my_tramp1:"
 "      pushq %rbp\n"
 "      movq %rsp, %rbp\n"
 "      call my_direct_func1\n"
 "      leave\n"
+"      .size           my_tramp1, .-my_tramp1\n"
 "      ret\n"
+"      .type           my_tramp2, @function\n"
 "   my_tramp2:"
 "      pushq %rbp\n"
 "      movq %rsp, %rbp\n"
 "      call my_direct_func2\n"
 "      leave\n"
 "      ret\n"
+"      .size           my_tramp2, .-my_tramp2\n"
 "      .popsection\n"
 );
 
index 27efa5f..d5c5022 100644 (file)
@@ -15,6 +15,7 @@ extern void my_tramp(void *);
 
 asm (
 "      .pushsection    .text, \"ax\", @progbits\n"
+"      .type           my_tramp, @function\n"
 "   my_tramp:"
 "      pushq %rbp\n"
 "      movq %rsp, %rbp\n"
@@ -27,6 +28,7 @@ asm (
 "      popq %rdi\n"
 "      leave\n"
 "      ret\n"
+"      .size           my_tramp, .-my_tramp\n"
 "      .popsection\n"
 );
 
index a2e3063..63ca06d 100644 (file)
@@ -13,6 +13,7 @@ extern void my_tramp(void *);
 
 asm (
 "      .pushsection    .text, \"ax\", @progbits\n"
+"      .type           my_tramp, @function\n"
 "   my_tramp:"
 "      pushq %rbp\n"
 "      movq %rsp, %rbp\n"
@@ -21,6 +22,7 @@ asm (
 "      popq %rdi\n"
 "      leave\n"
 "      ret\n"
+"      .size           my_tramp, .-my_tramp\n"
 "      .popsection\n"
 );
 
index d09ab4a..3adef49 100755 (executable)
@@ -55,6 +55,29 @@ modpost_link()
        ${LD} ${KBUILD_LDFLAGS} -r -o ${1} ${objects}
 }
 
+objtool_link()
+{
+       local objtoolopt;
+
+       if [ -n "${CONFIG_VMLINUX_VALIDATION}" ]; then
+               objtoolopt="check"
+               if [ -z "${CONFIG_FRAME_POINTER}" ]; then
+                       objtoolopt="${objtoolopt} --no-fp"
+               fi
+               if [ -n "${CONFIG_GCOV_KERNEL}" ]; then
+                       objtoolopt="${objtoolopt} --no-unreachable"
+               fi
+               if [ -n "${CONFIG_RETPOLINE}" ]; then
+                       objtoolopt="${objtoolopt} --retpoline"
+               fi
+               if [ -n "${CONFIG_X86_SMAP}" ]; then
+                       objtoolopt="${objtoolopt} --uaccess"
+               fi
+               info OBJTOOL ${1}
+               tools/objtool/objtool ${objtoolopt} ${1}
+       fi
+}
+
 # Link of vmlinux
 # ${1} - output file
 # ${2}, ${3}, ... - optional extra .o files
@@ -251,6 +274,7 @@ ${MAKE} -f "${srctree}/scripts/Makefile.build" obj=init need-builtin=1
 #link vmlinux.o
 info LD vmlinux.o
 modpost_link vmlinux.o
+objtool_link vmlinux.o
 
 # modpost vmlinux.o to check for section mismatches
 ${MAKE} -f "${srctree}/scripts/Makefile.modpost" MODPOST_VMLINUX=1
index 6e06090..d255349 100644 (file)
@@ -58,8 +58,7 @@
 #define ORC_TYPE_CALL                  0
 #define ORC_TYPE_REGS                  1
 #define ORC_TYPE_REGS_IRET             2
-#define UNWIND_HINT_TYPE_SAVE          3
-#define UNWIND_HINT_TYPE_RESTORE       4
+#define UNWIND_HINT_TYPE_RET_OFFSET    3
 
 #ifndef __ASSEMBLY__
 /*
index 66f44f5..b7222d5 100644 (file)
@@ -1,11 +1,16 @@
 objtool-y += arch/$(SRCARCH)/
+
+objtool-y += weak.o
+
+objtool-$(SUBCMD_CHECK) += check.o
+objtool-$(SUBCMD_CHECK) += special.o
+objtool-$(SUBCMD_ORC) += check.o
+objtool-$(SUBCMD_ORC) += orc_gen.o
+objtool-$(SUBCMD_ORC) += orc_dump.o
+
 objtool-y += builtin-check.o
 objtool-y += builtin-orc.o
-objtool-y += check.o
-objtool-y += orc_gen.o
-objtool-y += orc_dump.o
 objtool-y += elf.o
-objtool-y += special.o
 objtool-y += objtool.o
 
 objtool-y += libstring.o
index de09467..0542e46 100644 (file)
@@ -289,6 +289,47 @@ they mean, and suggestions for how to fix them.
       might be corrupt due to a gcc bug.  For more details, see:
       https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70646
 
+9. file.o: warning: objtool: funcA() call to funcB() with UACCESS enabled
+
+   This means that an unexpected call to a non-whitelisted function exists
+   outside of arch-specific guards.
+   X86: SMAP (stac/clac): __uaccess_begin()/__uaccess_end()
+   ARM: PAN: uaccess_enable()/uaccess_disable()
+
+   These functions should be called to denote a minimal critical section around
+   access to __user variables. See also: https://lwn.net/Articles/517475/
+
+   The intention of the warning is to prevent calls to funcB() from eventually
+   calling schedule(), potentially leaking the AC flags state, and not
+   restoring them correctly.
+
+   It also helps verify that there are no unexpected calls to funcB() which may
+   access user space pages with protections against doing so disabled.
+
+   To fix, either:
+   1) remove explicit calls to funcB() from funcA().
+   2) add the correct guards before and after calls to low level functions like
+      __get_user_size()/__put_user_size().
+   3) add funcB to uaccess_safe_builtin whitelist in tools/objtool/check.c, if
+      funcB obviously does not call schedule(), and is marked notrace (since
+      function tracing inserts additional calls, which is not obvious from the
+      sources).
+
+10. file.o: warning: func()+0x5c: alternative modifies stack
+
+    This means that an alternative includes instructions that modify the
+    stack. The problem is that there is only one ORC unwind table, this means
+    that the ORC unwind entries must be valid for each of the alternatives.
+    The easiest way to enforce this is to ensure alternatives do not contain
+    any ORC entries, which in turn implies the above constraint.
+
+11. file.o: warning: unannotated intra-function call
+
+   This warning means that a direct call is done to a destination which
+   is not at the beginning of a function. If this is a legit call, you
+   can remove this warning by putting the ANNOTATE_INTRA_FUNCTION_CALL
+   directive right before the call.
+
 
 If the error doesn't seem to make sense, it could be a bug in objtool.
 Feel free to ask the objtool maintainer for help.
index f591c4d..7770edc 100644 (file)
@@ -35,7 +35,8 @@ all: $(OBJTOOL)
 
 INCLUDES := -I$(srctree)/tools/include \
            -I$(srctree)/tools/arch/$(HOSTARCH)/include/uapi \
-           -I$(srctree)/tools/arch/$(SRCARCH)/include
+           -I$(srctree)/tools/arch/$(SRCARCH)/include  \
+           -I$(srctree)/tools/objtool/arch/$(SRCARCH)/include
 WARNINGS := $(EXTRA_WARNINGS) -Wno-switch-default -Wno-switch-enum -Wno-packed
 CFLAGS   := -Werror $(WARNINGS) $(KBUILD_HOSTCFLAGS) -g $(INCLUDES) $(LIBELF_FLAGS)
 LDFLAGS  += $(LIBELF_LIBS) $(LIBSUBCMD) $(KBUILD_HOSTLDFLAGS)
@@ -45,14 +46,24 @@ elfshdr := $(shell echo '$(pound)include <libelf.h>' | $(CC) $(CFLAGS) -x c -E -
 CFLAGS += $(if $(elfshdr),,-DLIBELF_USE_DEPRECATED)
 
 AWK = awk
+
+SUBCMD_CHECK := n
+SUBCMD_ORC := n
+
+ifeq ($(SRCARCH),x86)
+       SUBCMD_CHECK := y
+       SUBCMD_ORC := y
+endif
+
+export SUBCMD_CHECK SUBCMD_ORC
 export srctree OUTPUT CFLAGS SRCARCH AWK
 include $(srctree)/tools/build/Makefile.include
 
 $(OBJTOOL_IN): fixdep FORCE
+       @$(CONFIG_SHELL) ./sync-check.sh
        @$(MAKE) $(build)=objtool
 
 $(OBJTOOL): $(LIBSUBCMD) $(OBJTOOL_IN)
-       @$(CONFIG_SHELL) ./sync-check.sh
        $(QUIET_LINK)$(CC) $(OBJTOOL_IN) $(LDFLAGS) -o $@
 
 
index ced3765..eda15a5 100644 (file)
@@ -8,9 +8,11 @@
 
 #include <stdbool.h>
 #include <linux/list.h>
-#include "elf.h"
+#include "objtool.h"
 #include "cfi.h"
 
+#include <asm/orc_types.h>
+
 enum insn_type {
        INSN_JUMP_CONDITIONAL,
        INSN_JUMP_UNCONDITIONAL,
@@ -20,7 +22,6 @@ enum insn_type {
        INSN_CALL_DYNAMIC,
        INSN_RETURN,
        INSN_CONTEXT_SWITCH,
-       INSN_STACK,
        INSN_BUG,
        INSN_NOP,
        INSN_STAC,
@@ -64,15 +65,23 @@ struct op_src {
 struct stack_op {
        struct op_dest dest;
        struct op_src src;
+       struct list_head list;
 };
 
-void arch_initial_func_cfi_state(struct cfi_state *state);
+struct instruction;
+
+void arch_initial_func_cfi_state(struct cfi_init_state *state);
 
-int arch_decode_instruction(struct elf *elf, struct section *sec,
+int arch_decode_instruction(const struct elf *elf, const struct section *sec,
                            unsigned long offset, unsigned int maxlen,
                            unsigned int *len, enum insn_type *type,
-                           unsigned long *immediate, struct stack_op *op);
+                           unsigned long *immediate,
+                           struct list_head *ops_list);
 
 bool arch_callee_saved_reg(unsigned char reg);
 
+unsigned long arch_jump_destination(struct instruction *insn);
+
+unsigned long arch_dest_rela_offset(int addend);
+
 #endif /* _ARCH_H */
index a62e032..4b504fc 100644 (file)
@@ -11,6 +11,7 @@
 #include "../../../arch/x86/lib/inat.c"
 #include "../../../arch/x86/lib/insn.c"
 
+#include "../../check.h"
 #include "../../elf.h"
 #include "../../arch.h"
 #include "../../warn.h"
@@ -26,7 +27,7 @@ static unsigned char op_to_cfi_reg[][2] = {
        {CFI_DI, CFI_R15},
 };
 
-static int is_x86_64(struct elf *elf)
+static int is_x86_64(const struct elf *elf)
 {
        switch (elf->ehdr.e_machine) {
        case EM_X86_64:
@@ -66,16 +67,34 @@ bool arch_callee_saved_reg(unsigned char reg)
        }
 }
 
-int arch_decode_instruction(struct elf *elf, struct section *sec,
+unsigned long arch_dest_rela_offset(int addend)
+{
+       return addend + 4;
+}
+
+unsigned long arch_jump_destination(struct instruction *insn)
+{
+       return insn->offset + insn->len + insn->immediate;
+}
+
+#define ADD_OP(op) \
+       if (!(op = calloc(1, sizeof(*op)))) \
+               return -1; \
+       else for (list_add_tail(&op->list, ops_list); op; op = NULL)
+
+int arch_decode_instruction(const struct elf *elf, const struct section *sec,
                            unsigned long offset, unsigned int maxlen,
                            unsigned int *len, enum insn_type *type,
-                           unsigned long *immediate, struct stack_op *op)
+                           unsigned long *immediate,
+                           struct list_head *ops_list)
 {
        struct insn insn;
        int x86_64, sign;
        unsigned char op1, op2, rex = 0, rex_b = 0, rex_r = 0, rex_w = 0,
                      rex_x = 0, modrm = 0, modrm_mod = 0, modrm_rm = 0,
                      modrm_reg = 0, sib = 0;
+       struct stack_op *op = NULL;
+       struct symbol *sym;
 
        x86_64 = is_x86_64(elf);
        if (x86_64 == -1)
@@ -85,7 +104,7 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
        insn_get_length(&insn);
 
        if (!insn_complete(&insn)) {
-               WARN_FUNC("can't decode instruction", sec, offset);
+               WARN("can't decode instruction at %s:0x%lx", sec->name, offset);
                return -1;
        }
 
@@ -123,40 +142,44 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
                if (rex_w && !rex_b && modrm_mod == 3 && modrm_rm == 4) {
 
                        /* add/sub reg, %rsp */
-                       *type = INSN_STACK;
-                       op->src.type = OP_SRC_ADD;
-                       op->src.reg = op_to_cfi_reg[modrm_reg][rex_r];
-                       op->dest.type = OP_DEST_REG;
-                       op->dest.reg = CFI_SP;
+                       ADD_OP(op) {
+                               op->src.type = OP_SRC_ADD;
+                               op->src.reg = op_to_cfi_reg[modrm_reg][rex_r];
+                               op->dest.type = OP_DEST_REG;
+                               op->dest.reg = CFI_SP;
+                       }
                }
                break;
 
        case 0x50 ... 0x57:
 
                /* push reg */
-               *type = INSN_STACK;
-               op->src.type = OP_SRC_REG;
-               op->src.reg = op_to_cfi_reg[op1 & 0x7][rex_b];
-               op->dest.type = OP_DEST_PUSH;
+               ADD_OP(op) {
+                       op->src.type = OP_SRC_REG;
+                       op->src.reg = op_to_cfi_reg[op1 & 0x7][rex_b];
+                       op->dest.type = OP_DEST_PUSH;
+               }
 
                break;
 
        case 0x58 ... 0x5f:
 
                /* pop reg */
-               *type = INSN_STACK;
-               op->src.type = OP_SRC_POP;
-               op->dest.type = OP_DEST_REG;
-               op->dest.reg = op_to_cfi_reg[op1 & 0x7][rex_b];
+               ADD_OP(op) {
+                       op->src.type = OP_SRC_POP;
+                       op->dest.type = OP_DEST_REG;
+                       op->dest.reg = op_to_cfi_reg[op1 & 0x7][rex_b];
+               }
 
                break;
 
        case 0x68:
        case 0x6a:
                /* push immediate */
-               *type = INSN_STACK;
-               op->src.type = OP_SRC_CONST;
-               op->dest.type = OP_DEST_PUSH;
+               ADD_OP(op) {
+                       op->src.type = OP_SRC_CONST;
+                       op->dest.type = OP_DEST_PUSH;
+               }
                break;
 
        case 0x70 ... 0x7f:
@@ -170,12 +193,13 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
 
                if (modrm == 0xe4) {
                        /* and imm, %rsp */
-                       *type = INSN_STACK;
-                       op->src.type = OP_SRC_AND;
-                       op->src.reg = CFI_SP;
-                       op->src.offset = insn.immediate.value;
-                       op->dest.type = OP_DEST_REG;
-                       op->dest.reg = CFI_SP;
+                       ADD_OP(op) {
+                               op->src.type = OP_SRC_AND;
+                               op->src.reg = CFI_SP;
+                               op->src.offset = insn.immediate.value;
+                               op->dest.type = OP_DEST_REG;
+                               op->dest.reg = CFI_SP;
+                       }
                        break;
                }
 
@@ -187,34 +211,37 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
                        break;
 
                /* add/sub imm, %rsp */
-               *type = INSN_STACK;
-               op->src.type = OP_SRC_ADD;
-               op->src.reg = CFI_SP;
-               op->src.offset = insn.immediate.value * sign;
-               op->dest.type = OP_DEST_REG;
-               op->dest.reg = CFI_SP;
+               ADD_OP(op) {
+                       op->src.type = OP_SRC_ADD;
+                       op->src.reg = CFI_SP;
+                       op->src.offset = insn.immediate.value * sign;
+                       op->dest.type = OP_DEST_REG;
+                       op->dest.reg = CFI_SP;
+               }
                break;
 
        case 0x89:
                if (rex_w && !rex_r && modrm_mod == 3 && modrm_reg == 4) {
 
                        /* mov %rsp, reg */
-                       *type = INSN_STACK;
-                       op->src.type = OP_SRC_REG;
-                       op->src.reg = CFI_SP;
-                       op->dest.type = OP_DEST_REG;
-                       op->dest.reg = op_to_cfi_reg[modrm_rm][rex_b];
+                       ADD_OP(op) {
+                               op->src.type = OP_SRC_REG;
+                               op->src.reg = CFI_SP;
+                               op->dest.type = OP_DEST_REG;
+                               op->dest.reg = op_to_cfi_reg[modrm_rm][rex_b];
+                       }
                        break;
                }
 
                if (rex_w && !rex_b && modrm_mod == 3 && modrm_rm == 4) {
 
                        /* mov reg, %rsp */
-                       *type = INSN_STACK;
-                       op->src.type = OP_SRC_REG;
-                       op->src.reg = op_to_cfi_reg[modrm_reg][rex_r];
-                       op->dest.type = OP_DEST_REG;
-                       op->dest.reg = CFI_SP;
+                       ADD_OP(op) {
+                               op->src.type = OP_SRC_REG;
+                               op->src.reg = op_to_cfi_reg[modrm_reg][rex_r];
+                               op->dest.type = OP_DEST_REG;
+                               op->dest.reg = CFI_SP;
+                       }
                        break;
                }
 
@@ -224,22 +251,24 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
                    (modrm_mod == 1 || modrm_mod == 2) && modrm_rm == 5) {
 
                        /* mov reg, disp(%rbp) */
-                       *type = INSN_STACK;
-                       op->src.type = OP_SRC_REG;
-                       op->src.reg = op_to_cfi_reg[modrm_reg][rex_r];
-                       op->dest.type = OP_DEST_REG_INDIRECT;
-                       op->dest.reg = CFI_BP;
-                       op->dest.offset = insn.displacement.value;
+                       ADD_OP(op) {
+                               op->src.type = OP_SRC_REG;
+                               op->src.reg = op_to_cfi_reg[modrm_reg][rex_r];
+                               op->dest.type = OP_DEST_REG_INDIRECT;
+                               op->dest.reg = CFI_BP;
+                               op->dest.offset = insn.displacement.value;
+                       }
 
                } else if (rex_w && !rex_b && modrm_rm == 4 && sib == 0x24) {
 
                        /* mov reg, disp(%rsp) */
-                       *type = INSN_STACK;
-                       op->src.type = OP_SRC_REG;
-                       op->src.reg = op_to_cfi_reg[modrm_reg][rex_r];
-                       op->dest.type = OP_DEST_REG_INDIRECT;
-                       op->dest.reg = CFI_SP;
-                       op->dest.offset = insn.displacement.value;
+                       ADD_OP(op) {
+                               op->src.type = OP_SRC_REG;
+                               op->src.reg = op_to_cfi_reg[modrm_reg][rex_r];
+                               op->dest.type = OP_DEST_REG_INDIRECT;
+                               op->dest.reg = CFI_SP;
+                               op->dest.offset = insn.displacement.value;
+                       }
                }
 
                break;
@@ -248,23 +277,25 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
                if (rex_w && !rex_b && modrm_mod == 1 && modrm_rm == 5) {
 
                        /* mov disp(%rbp), reg */
-                       *type = INSN_STACK;
-                       op->src.type = OP_SRC_REG_INDIRECT;
-                       op->src.reg = CFI_BP;
-                       op->src.offset = insn.displacement.value;
-                       op->dest.type = OP_DEST_REG;
-                       op->dest.reg = op_to_cfi_reg[modrm_reg][rex_r];
+                       ADD_OP(op) {
+                               op->src.type = OP_SRC_REG_INDIRECT;
+                               op->src.reg = CFI_BP;
+                               op->src.offset = insn.displacement.value;
+                               op->dest.type = OP_DEST_REG;
+                               op->dest.reg = op_to_cfi_reg[modrm_reg][rex_r];
+                       }
 
                } else if (rex_w && !rex_b && sib == 0x24 &&
                           modrm_mod != 3 && modrm_rm == 4) {
 
                        /* mov disp(%rsp), reg */
-                       *type = INSN_STACK;
-                       op->src.type = OP_SRC_REG_INDIRECT;
-                       op->src.reg = CFI_SP;
-                       op->src.offset = insn.displacement.value;
-                       op->dest.type = OP_DEST_REG;
-                       op->dest.reg = op_to_cfi_reg[modrm_reg][rex_r];
+                       ADD_OP(op) {
+                               op->src.type = OP_SRC_REG_INDIRECT;
+                               op->src.reg = CFI_SP;
+                               op->src.offset = insn.displacement.value;
+                               op->dest.type = OP_DEST_REG;
+                               op->dest.reg = op_to_cfi_reg[modrm_reg][rex_r];
+                       }
                }
 
                break;
@@ -272,28 +303,30 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
        case 0x8d:
                if (sib == 0x24 && rex_w && !rex_b && !rex_x) {
 
-                       *type = INSN_STACK;
-                       if (!insn.displacement.value) {
-                               /* lea (%rsp), reg */
-                               op->src.type = OP_SRC_REG;
-                       } else {
-                               /* lea disp(%rsp), reg */
-                               op->src.type = OP_SRC_ADD;
-                               op->src.offset = insn.displacement.value;
+                       ADD_OP(op) {
+                               if (!insn.displacement.value) {
+                                       /* lea (%rsp), reg */
+                                       op->src.type = OP_SRC_REG;
+                               } else {
+                                       /* lea disp(%rsp), reg */
+                                       op->src.type = OP_SRC_ADD;
+                                       op->src.offset = insn.displacement.value;
+                               }
+                               op->src.reg = CFI_SP;
+                               op->dest.type = OP_DEST_REG;
+                               op->dest.reg = op_to_cfi_reg[modrm_reg][rex_r];
                        }
-                       op->src.reg = CFI_SP;
-                       op->dest.type = OP_DEST_REG;
-                       op->dest.reg = op_to_cfi_reg[modrm_reg][rex_r];
 
                } else if (rex == 0x48 && modrm == 0x65) {
 
                        /* lea disp(%rbp), %rsp */
-                       *type = INSN_STACK;
-                       op->src.type = OP_SRC_ADD;
-                       op->src.reg = CFI_BP;
-                       op->src.offset = insn.displacement.value;
-                       op->dest.type = OP_DEST_REG;
-                       op->dest.reg = CFI_SP;
+                       ADD_OP(op) {
+                               op->src.type = OP_SRC_ADD;
+                               op->src.reg = CFI_BP;
+                               op->src.offset = insn.displacement.value;
+                               op->dest.type = OP_DEST_REG;
+                               op->dest.reg = CFI_SP;
+                       }
 
                } else if (rex == 0x49 && modrm == 0x62 &&
                           insn.displacement.value == -8) {
@@ -304,12 +337,13 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
                         * Restoring rsp back to its original value after a
                         * stack realignment.
                         */
-                       *type = INSN_STACK;
-                       op->src.type = OP_SRC_ADD;
-                       op->src.reg = CFI_R10;
-                       op->src.offset = -8;
-                       op->dest.type = OP_DEST_REG;
-                       op->dest.reg = CFI_SP;
+                       ADD_OP(op) {
+                               op->src.type = OP_SRC_ADD;
+                               op->src.reg = CFI_R10;
+                               op->src.offset = -8;
+                               op->dest.type = OP_DEST_REG;
+                               op->dest.reg = CFI_SP;
+                       }
 
                } else if (rex == 0x49 && modrm == 0x65 &&
                           insn.displacement.value == -16) {
@@ -320,21 +354,23 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
                         * Restoring rsp back to its original value after a
                         * stack realignment.
                         */
-                       *type = INSN_STACK;
-                       op->src.type = OP_SRC_ADD;
-                       op->src.reg = CFI_R13;
-                       op->src.offset = -16;
-                       op->dest.type = OP_DEST_REG;
-                       op->dest.reg = CFI_SP;
+                       ADD_OP(op) {
+                               op->src.type = OP_SRC_ADD;
+                               op->src.reg = CFI_R13;
+                               op->src.offset = -16;
+                               op->dest.type = OP_DEST_REG;
+                               op->dest.reg = CFI_SP;
+                       }
                }
 
                break;
 
        case 0x8f:
                /* pop to mem */
-               *type = INSN_STACK;
-               op->src.type = OP_SRC_POP;
-               op->dest.type = OP_DEST_MEM;
+               ADD_OP(op) {
+                       op->src.type = OP_SRC_POP;
+                       op->dest.type = OP_DEST_MEM;
+               }
                break;
 
        case 0x90:
@@ -343,16 +379,18 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
 
        case 0x9c:
                /* pushf */
-               *type = INSN_STACK;
-               op->src.type = OP_SRC_CONST;
-               op->dest.type = OP_DEST_PUSHF;
+               ADD_OP(op) {
+                       op->src.type = OP_SRC_CONST;
+                       op->dest.type = OP_DEST_PUSHF;
+               }
                break;
 
        case 0x9d:
                /* popf */
-               *type = INSN_STACK;
-               op->src.type = OP_SRC_POPF;
-               op->dest.type = OP_DEST_MEM;
+               ADD_OP(op) {
+                       op->src.type = OP_SRC_POPF;
+                       op->dest.type = OP_DEST_MEM;
+               }
                break;
 
        case 0x0f:
@@ -387,16 +425,18 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
                } else if (op2 == 0xa0 || op2 == 0xa8) {
 
                        /* push fs/gs */
-                       *type = INSN_STACK;
-                       op->src.type = OP_SRC_CONST;
-                       op->dest.type = OP_DEST_PUSH;
+                       ADD_OP(op) {
+                               op->src.type = OP_SRC_CONST;
+                               op->dest.type = OP_DEST_PUSH;
+                       }
 
                } else if (op2 == 0xa1 || op2 == 0xa9) {
 
                        /* pop fs/gs */
-                       *type = INSN_STACK;
-                       op->src.type = OP_SRC_POP;
-                       op->dest.type = OP_DEST_MEM;
+                       ADD_OP(op) {
+                               op->src.type = OP_SRC_POP;
+                               op->dest.type = OP_DEST_MEM;
+                       }
                }
 
                break;
@@ -409,8 +449,8 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
                 * mov bp, sp
                 * pop bp
                 */
-               *type = INSN_STACK;
-               op->dest.type = OP_DEST_LEAVE;
+               ADD_OP(op)
+                       op->dest.type = OP_DEST_LEAVE;
 
                break;
 
@@ -429,14 +469,41 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
                *type = INSN_RETURN;
                break;
 
+       case 0xcf: /* iret */
+               /*
+                * Handle sync_core(), which has an IRET to self.
+                * All other IRET are in STT_NONE entry code.
+                */
+               sym = find_symbol_containing(sec, offset);
+               if (sym && sym->type == STT_FUNC) {
+                       ADD_OP(op) {
+                               /* add $40, %rsp */
+                               op->src.type = OP_SRC_ADD;
+                               op->src.reg = CFI_SP;
+                               op->src.offset = 5*8;
+                               op->dest.type = OP_DEST_REG;
+                               op->dest.reg = CFI_SP;
+                       }
+                       break;
+               }
+
+               /* fallthrough */
+
        case 0xca: /* retf */
        case 0xcb: /* retf */
-       case 0xcf: /* iret */
                *type = INSN_CONTEXT_SWITCH;
                break;
 
        case 0xe8:
                *type = INSN_CALL;
+               /*
+                * For the impact on the stack, a CALL behaves like
+                * a PUSH of an immediate value (the return address).
+                */
+               ADD_OP(op) {
+                       op->src.type = OP_SRC_CONST;
+                       op->dest.type = OP_DEST_PUSH;
+               }
                break;
 
        case 0xfc:
@@ -464,9 +531,10 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
                else if (modrm_reg == 6) {
 
                        /* push from mem */
-                       *type = INSN_STACK;
-                       op->src.type = OP_SRC_CONST;
-                       op->dest.type = OP_DEST_PUSH;
+                       ADD_OP(op) {
+                               op->src.type = OP_SRC_CONST;
+                               op->dest.type = OP_DEST_PUSH;
+                       }
                }
 
                break;
@@ -480,7 +548,7 @@ int arch_decode_instruction(struct elf *elf, struct section *sec,
        return 0;
 }
 
-void arch_initial_func_cfi_state(struct cfi_state *state)
+void arch_initial_func_cfi_state(struct cfi_init_state *state)
 {
        int i;
 
diff --git a/tools/objtool/arch/x86/include/cfi_regs.h b/tools/objtool/arch/x86/include/cfi_regs.h
new file mode 100644 (file)
index 0000000..79bc517
--- /dev/null
@@ -0,0 +1,25 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+
+#ifndef _OBJTOOL_CFI_REGS_H
+#define _OBJTOOL_CFI_REGS_H
+
+#define CFI_AX                 0
+#define CFI_DX                 1
+#define CFI_CX                 2
+#define CFI_BX                 3
+#define CFI_SI                 4
+#define CFI_DI                 5
+#define CFI_BP                 6
+#define CFI_SP                 7
+#define CFI_R8                 8
+#define CFI_R9                 9
+#define CFI_R10                        10
+#define CFI_R11                        11
+#define CFI_R12                        12
+#define CFI_R13                        13
+#define CFI_R14                        14
+#define CFI_R15                        15
+#define CFI_RA                 16
+#define CFI_NUM_REGS           17
+
+#endif /* _OBJTOOL_CFI_REGS_H */
index 10fbe75..7a44174 100644 (file)
  */
 
 #include <subcmd/parse-options.h>
+#include <string.h>
 #include "builtin.h"
-#include "check.h"
+#include "objtool.h"
 
-bool no_fp, no_unreachable, retpoline, module, backtrace, uaccess, stats;
+bool no_fp, no_unreachable, retpoline, module, backtrace, uaccess, stats, validate_dup, vmlinux;
 
 static const char * const check_usage[] = {
        "objtool check [<options>] file.o",
@@ -32,12 +33,14 @@ const struct option check_options[] = {
        OPT_BOOLEAN('b', "backtrace", &backtrace, "unwind on error"),
        OPT_BOOLEAN('a', "uaccess", &uaccess, "enable uaccess checking"),
        OPT_BOOLEAN('s', "stats", &stats, "print statistics"),
+       OPT_BOOLEAN('d', "duplicate", &validate_dup, "duplicate validation for vmlinux.o"),
+       OPT_BOOLEAN('l', "vmlinux", &vmlinux, "vmlinux.o validation"),
        OPT_END(),
 };
 
 int cmd_check(int argc, const char **argv)
 {
-       const char *objname;
+       const char *objname, *s;
 
        argc = parse_options(argc, argv, check_options, check_usage, 0);
 
@@ -46,5 +49,9 @@ int cmd_check(int argc, const char **argv)
 
        objname = argv[0];
 
+       s = strstr(objname, "vmlinux.o");
+       if (s && !s[9])
+               vmlinux = true;
+
        return check(objname, false);
 }
index 5f7cc61..b1dfe20 100644 (file)
@@ -14,8 +14,7 @@
 
 #include <string.h>
 #include "builtin.h"
-#include "check.h"
-
+#include "objtool.h"
 
 static const char *orc_usage[] = {
        "objtool orc generate [<options>] file.o",
index 0b90790..85c979c 100644 (file)
@@ -8,7 +8,7 @@
 #include <subcmd/parse-options.h>
 
 extern const struct option check_options[];
-extern bool no_fp, no_unreachable, retpoline, module, backtrace, uaccess, stats;
+extern bool no_fp, no_unreachable, retpoline, module, backtrace, uaccess, stats, validate_dup, vmlinux;
 
 extern int cmd_check(int argc, const char **argv);
 extern int cmd_orc(int argc, const char **argv);
index 4427bf8..c7c59c6 100644 (file)
@@ -6,38 +6,33 @@
 #ifndef _OBJTOOL_CFI_H
 #define _OBJTOOL_CFI_H
 
+#include "cfi_regs.h"
+
 #define CFI_UNDEFINED          -1
 #define CFI_CFA                        -2
 #define CFI_SP_INDIRECT                -3
 #define CFI_BP_INDIRECT                -4
 
-#define CFI_AX                 0
-#define CFI_DX                 1
-#define CFI_CX                 2
-#define CFI_BX                 3
-#define CFI_SI                 4
-#define CFI_DI                 5
-#define CFI_BP                 6
-#define CFI_SP                 7
-#define CFI_R8                 8
-#define CFI_R9                 9
-#define CFI_R10                        10
-#define CFI_R11                        11
-#define CFI_R12                        12
-#define CFI_R13                        13
-#define CFI_R14                        14
-#define CFI_R15                        15
-#define CFI_RA                 16
-#define CFI_NUM_REGS           17
-
 struct cfi_reg {
        int base;
        int offset;
 };
 
-struct cfi_state {
+struct cfi_init_state {
+       struct cfi_reg regs[CFI_NUM_REGS];
        struct cfi_reg cfa;
+};
+
+struct cfi_state {
        struct cfi_reg regs[CFI_NUM_REGS];
+       struct cfi_reg vals[CFI_NUM_REGS];
+       struct cfi_reg cfa;
+       int stack_size;
+       int drap_reg, drap_offset;
+       unsigned char type;
+       bool bp_scratch;
+       bool drap;
+       bool end;
 };
 
 #endif /* _OBJTOOL_CFI_H */
index 3c6da70..63d65a7 100644 (file)
@@ -7,10 +7,10 @@
 #include <stdlib.h>
 
 #include "builtin.h"
+#include "cfi.h"
+#include "arch.h"
 #include "check.h"
-#include "elf.h"
 #include "special.h"
-#include "arch.h"
 #include "warn.h"
 
 #include <linux/hashtable.h>
@@ -27,16 +27,17 @@ struct alternative {
 };
 
 const char *objname;
-struct cfi_state initial_func_cfi;
+struct cfi_init_state initial_func_cfi;
 
 struct instruction *find_insn(struct objtool_file *file,
                              struct section *sec, unsigned long offset)
 {
        struct instruction *insn;
 
-       hash_for_each_possible(file->insn_hash, insn, hash, offset)
+       hash_for_each_possible(file->insn_hash, insn, hash, sec_offset_hash(sec, offset)) {
                if (insn->sec == sec && insn->offset == offset)
                        return insn;
+       }
 
        return NULL;
 }
@@ -226,18 +227,31 @@ static bool dead_end_function(struct objtool_file *file, struct symbol *func)
        return __dead_end_function(file, func, 0);
 }
 
-static void clear_insn_state(struct insn_state *state)
+static void init_cfi_state(struct cfi_state *cfi)
 {
        int i;
 
-       memset(state, 0, sizeof(*state));
-       state->cfa.base = CFI_UNDEFINED;
        for (i = 0; i < CFI_NUM_REGS; i++) {
-               state->regs[i].base = CFI_UNDEFINED;
-               state->vals[i].base = CFI_UNDEFINED;
+               cfi->regs[i].base = CFI_UNDEFINED;
+               cfi->vals[i].base = CFI_UNDEFINED;
        }
-       state->drap_reg = CFI_UNDEFINED;
-       state->drap_offset = -1;
+       cfi->cfa.base = CFI_UNDEFINED;
+       cfi->drap_reg = CFI_UNDEFINED;
+       cfi->drap_offset = -1;
+}
+
+static void init_insn_state(struct insn_state *state, struct section *sec)
+{
+       memset(state, 0, sizeof(*state));
+       init_cfi_state(&state->cfi);
+
+       /*
+        * We need the full vmlinux for noinstr validation, otherwise we can
+        * not correctly determine insn->call_dest->sec (external symbols do
+        * not have a section).
+        */
+       if (vmlinux && sec)
+               state->noinstr = sec->noinstr;
 }
 
 /*
@@ -263,6 +277,10 @@ static int decode_instructions(struct objtool_file *file)
                    strncmp(sec->name, ".discard.", 9))
                        sec->text = true;
 
+               if (!strcmp(sec->name, ".noinstr.text") ||
+                   !strcmp(sec->name, ".entry.text"))
+                       sec->noinstr = true;
+
                for (offset = 0; offset < sec->len; offset += insn->len) {
                        insn = malloc(sizeof(*insn));
                        if (!insn) {
@@ -271,7 +289,8 @@ static int decode_instructions(struct objtool_file *file)
                        }
                        memset(insn, 0, sizeof(*insn));
                        INIT_LIST_HEAD(&insn->alts);
-                       clear_insn_state(&insn->state);
+                       INIT_LIST_HEAD(&insn->stack_ops);
+                       init_cfi_state(&insn->cfi);
 
                        insn->sec = sec;
                        insn->offset = offset;
@@ -280,11 +299,11 @@ static int decode_instructions(struct objtool_file *file)
                                                      sec->len - offset,
                                                      &insn->len, &insn->type,
                                                      &insn->immediate,
-                                                     &insn->stack_op);
+                                                     &insn->stack_ops);
                        if (ret)
                                goto err;
 
-                       hash_add(file->insn_hash, &insn->hash, insn->offset);
+                       hash_add(file->insn_hash, &insn->hash, sec_offset_hash(sec, insn->offset));
                        list_add_tail(&insn->list, &file->insn_list);
                        nr_insns++;
                }
@@ -314,6 +333,19 @@ err:
        return ret;
 }
 
+static struct instruction *find_last_insn(struct objtool_file *file,
+                                         struct section *sec)
+{
+       struct instruction *insn = NULL;
+       unsigned int offset;
+       unsigned int end = (sec->len > 10) ? sec->len - 10 : 0;
+
+       for (offset = sec->len - 1; offset >= end && !insn; offset--)
+               insn = find_insn(file, sec, offset);
+
+       return insn;
+}
+
 /*
  * Mark "ud2" instructions and manually annotated dead ends.
  */
@@ -322,7 +354,6 @@ static int add_dead_ends(struct objtool_file *file)
        struct section *sec;
        struct rela *rela;
        struct instruction *insn;
-       bool found;
 
        /*
         * By default, "ud2" is a dead end unless otherwise annotated, because
@@ -348,15 +379,8 @@ static int add_dead_ends(struct objtool_file *file)
                if (insn)
                        insn = list_prev_entry(insn, list);
                else if (rela->addend == rela->sym->sec->len) {
-                       found = false;
-                       list_for_each_entry_reverse(insn, &file->insn_list, list) {
-                               if (insn->sec == rela->sym->sec) {
-                                       found = true;
-                                       break;
-                               }
-                       }
-
-                       if (!found) {
+                       insn = find_last_insn(file, rela->sym->sec);
+                       if (!insn) {
                                WARN("can't find unreachable insn at %s+0x%x",
                                     rela->sym->sec->name, rela->addend);
                                return -1;
@@ -390,15 +414,8 @@ reachable:
                if (insn)
                        insn = list_prev_entry(insn, list);
                else if (rela->addend == rela->sym->sec->len) {
-                       found = false;
-                       list_for_each_entry_reverse(insn, &file->insn_list, list) {
-                               if (insn->sec == rela->sym->sec) {
-                                       found = true;
-                                       break;
-                               }
-                       }
-
-                       if (!found) {
+                       insn = find_last_insn(file, rela->sym->sec);
+                       if (!insn) {
                                WARN("can't find reachable insn at %s+0x%x",
                                     rela->sym->sec->name, rela->addend);
                                return -1;
@@ -490,6 +507,7 @@ static const char *uaccess_safe_builtin[] = {
        "__asan_report_store16_noabort",
        /* KCOV */
        "write_comp_data",
+       "check_kcov_mode",
        "__sanitizer_cov_trace_pc",
        "__sanitizer_cov_trace_const_cmp1",
        "__sanitizer_cov_trace_const_cmp2",
@@ -585,13 +603,14 @@ static int add_jump_destinations(struct objtool_file *file)
                                               insn->offset, insn->len);
                if (!rela) {
                        dest_sec = insn->sec;
-                       dest_off = insn->offset + insn->len + insn->immediate;
+                       dest_off = arch_jump_destination(insn);
                } else if (rela->sym->type == STT_SECTION) {
                        dest_sec = rela->sym->sec;
-                       dest_off = rela->addend + 4;
+                       dest_off = arch_dest_rela_offset(rela->addend);
                } else if (rela->sym->sec->idx) {
                        dest_sec = rela->sym->sec;
-                       dest_off = rela->sym->sym.st_value + rela->addend + 4;
+                       dest_off = rela->sym->sym.st_value +
+                                  arch_dest_rela_offset(rela->addend);
                } else if (strstr(rela->sym->name, "_indirect_thunk_")) {
                        /*
                         * Retpoline jumps are really dynamic jumps in
@@ -665,6 +684,16 @@ static int add_jump_destinations(struct objtool_file *file)
        return 0;
 }
 
+static void remove_insn_ops(struct instruction *insn)
+{
+       struct stack_op *op, *tmp;
+
+       list_for_each_entry_safe(op, tmp, &insn->stack_ops, list) {
+               list_del(&op->list);
+               free(op);
+       }
+}
+
 /*
  * Find the destination instructions for all calls.
  */
@@ -681,7 +710,7 @@ static int add_call_destinations(struct objtool_file *file)
                rela = find_rela_by_dest_range(file->elf, insn->sec,
                                               insn->offset, insn->len);
                if (!rela) {
-                       dest_off = insn->offset + insn->len + insn->immediate;
+                       dest_off = arch_jump_destination(insn);
                        insn->call_dest = find_func_by_offset(insn->sec, dest_off);
                        if (!insn->call_dest)
                                insn->call_dest = find_symbol_by_offset(insn->sec, dest_off);
@@ -690,10 +719,7 @@ static int add_call_destinations(struct objtool_file *file)
                                continue;
 
                        if (!insn->call_dest) {
-                               WARN_FUNC("unsupported intra-function call",
-                                         insn->sec, insn->offset);
-                               if (retpoline)
-                                       WARN("If this is a retpoline, please patch it in with alternatives and annotate it with ANNOTATE_NOSPEC_ALTERNATIVE.");
+                               WARN_FUNC("unannotated intra-function call", insn->sec, insn->offset);
                                return -1;
                        }
 
@@ -704,17 +730,27 @@ static int add_call_destinations(struct objtool_file *file)
                        }
 
                } else if (rela->sym->type == STT_SECTION) {
+                       dest_off = arch_dest_rela_offset(rela->addend);
                        insn->call_dest = find_func_by_offset(rela->sym->sec,
-                                                             rela->addend+4);
+                                                             dest_off);
                        if (!insn->call_dest) {
-                               WARN_FUNC("can't find call dest symbol at %s+0x%x",
+                               WARN_FUNC("can't find call dest symbol at %s+0x%lx",
                                          insn->sec, insn->offset,
                                          rela->sym->sec->name,
-                                         rela->addend + 4);
+                                         dest_off);
                                return -1;
                        }
                } else
                        insn->call_dest = rela->sym;
+
+               /*
+                * Whatever stack impact regular CALLs have, should be undone
+                * by the RETURN of the called function.
+                *
+                * Annotated intra-function calls retain the stack_ops but
+                * are converted to JUMP, see read_intra_function_calls().
+                */
+               remove_insn_ops(insn);
        }
 
        return 0;
@@ -742,7 +778,9 @@ static int handle_group_alt(struct objtool_file *file,
                            struct instruction *orig_insn,
                            struct instruction **new_insn)
 {
+       static unsigned int alt_group_next_index = 1;
        struct instruction *last_orig_insn, *last_new_insn, *insn, *fake_jump = NULL;
+       unsigned int alt_group = alt_group_next_index++;
        unsigned long dest_off;
 
        last_orig_insn = NULL;
@@ -751,7 +789,7 @@ static int handle_group_alt(struct objtool_file *file,
                if (insn->offset >= special_alt->orig_off + special_alt->orig_len)
                        break;
 
-               insn->alt_group = true;
+               insn->alt_group = alt_group;
                last_orig_insn = insn;
        }
 
@@ -763,7 +801,8 @@ static int handle_group_alt(struct objtool_file *file,
                }
                memset(fake_jump, 0, sizeof(*fake_jump));
                INIT_LIST_HEAD(&fake_jump->alts);
-               clear_insn_state(&fake_jump->state);
+               INIT_LIST_HEAD(&fake_jump->stack_ops);
+               init_cfi_state(&fake_jump->cfi);
 
                fake_jump->sec = special_alt->new_sec;
                fake_jump->offset = FAKE_JUMP_OFFSET;
@@ -784,6 +823,7 @@ static int handle_group_alt(struct objtool_file *file,
        }
 
        last_new_insn = NULL;
+       alt_group = alt_group_next_index++;
        insn = *new_insn;
        sec_for_each_insn_from(file, insn) {
                if (insn->offset >= special_alt->new_off + special_alt->new_len)
@@ -793,6 +833,7 @@ static int handle_group_alt(struct objtool_file *file,
 
                insn->ignore = orig_insn->ignore_alts;
                insn->func = orig_insn->func;
+               insn->alt_group = alt_group;
 
                /*
                 * Since alternative replacement code is copy/pasted by the
@@ -821,7 +862,7 @@ static int handle_group_alt(struct objtool_file *file,
                if (!insn->immediate)
                        continue;
 
-               dest_off = insn->offset + insn->len + insn->immediate;
+               dest_off = arch_jump_destination(insn);
                if (dest_off == special_alt->new_off + special_alt->new_len) {
                        if (!fake_jump) {
                                WARN("%s: alternative jump to end of section",
@@ -916,6 +957,12 @@ static int add_special_section_alts(struct objtool_file *file)
                }
 
                if (special_alt->group) {
+                       if (!special_alt->orig_len) {
+                               WARN_FUNC("empty alternative entry",
+                                         orig_insn->sec, orig_insn->offset);
+                               continue;
+                       }
+
                        ret = handle_group_alt(file, special_alt, orig_insn,
                                               &new_insn);
                        if (ret)
@@ -1253,15 +1300,10 @@ static int read_unwind_hints(struct objtool_file *file)
                        return -1;
                }
 
-               cfa = &insn->state.cfa;
-
-               if (hint->type == UNWIND_HINT_TYPE_SAVE) {
-                       insn->save = true;
-                       continue;
+               cfa = &insn->cfi.cfa;
 
-               } else if (hint->type == UNWIND_HINT_TYPE_RESTORE) {
-                       insn->restore = true;
-                       insn->hint = true;
+               if (hint->type == UNWIND_HINT_TYPE_RET_OFFSET) {
+                       insn->ret_offset = hint->sp_offset;
                        continue;
                }
 
@@ -1299,8 +1341,8 @@ static int read_unwind_hints(struct objtool_file *file)
                }
 
                cfa->offset = hint->sp_offset;
-               insn->state.type = hint->type;
-               insn->state.end = hint->end;
+               insn->cfi.type = hint->type;
+               insn->cfi.end = hint->end;
        }
 
        return 0;
@@ -1341,6 +1383,104 @@ static int read_retpoline_hints(struct objtool_file *file)
        return 0;
 }
 
+static int read_instr_hints(struct objtool_file *file)
+{
+       struct section *sec;
+       struct instruction *insn;
+       struct rela *rela;
+
+       sec = find_section_by_name(file->elf, ".rela.discard.instr_end");
+       if (!sec)
+               return 0;
+
+       list_for_each_entry(rela, &sec->rela_list, list) {
+               if (rela->sym->type != STT_SECTION) {
+                       WARN("unexpected relocation symbol type in %s", sec->name);
+                       return -1;
+               }
+
+               insn = find_insn(file, rela->sym->sec, rela->addend);
+               if (!insn) {
+                       WARN("bad .discard.instr_end entry");
+                       return -1;
+               }
+
+               insn->instr--;
+       }
+
+       sec = find_section_by_name(file->elf, ".rela.discard.instr_begin");
+       if (!sec)
+               return 0;
+
+       list_for_each_entry(rela, &sec->rela_list, list) {
+               if (rela->sym->type != STT_SECTION) {
+                       WARN("unexpected relocation symbol type in %s", sec->name);
+                       return -1;
+               }
+
+               insn = find_insn(file, rela->sym->sec, rela->addend);
+               if (!insn) {
+                       WARN("bad .discard.instr_begin entry");
+                       return -1;
+               }
+
+               insn->instr++;
+       }
+
+       return 0;
+}
+
+static int read_intra_function_calls(struct objtool_file *file)
+{
+       struct instruction *insn;
+       struct section *sec;
+       struct rela *rela;
+
+       sec = find_section_by_name(file->elf, ".rela.discard.intra_function_calls");
+       if (!sec)
+               return 0;
+
+       list_for_each_entry(rela, &sec->rela_list, list) {
+               unsigned long dest_off;
+
+               if (rela->sym->type != STT_SECTION) {
+                       WARN("unexpected relocation symbol type in %s",
+                            sec->name);
+                       return -1;
+               }
+
+               insn = find_insn(file, rela->sym->sec, rela->addend);
+               if (!insn) {
+                       WARN("bad .discard.intra_function_call entry");
+                       return -1;
+               }
+
+               if (insn->type != INSN_CALL) {
+                       WARN_FUNC("intra_function_call not a direct call",
+                                 insn->sec, insn->offset);
+                       return -1;
+               }
+
+               /*
+                * Treat intra-function CALLs as JMPs, but with a stack_op.
+                * See add_call_destinations(), which strips stack_ops from
+                * normal CALLs.
+                */
+               insn->type = INSN_JUMP_UNCONDITIONAL;
+
+               dest_off = insn->offset + insn->len + insn->immediate;
+               insn->jump_dest = find_insn(file, insn->sec, dest_off);
+               if (!insn->jump_dest) {
+                       WARN_FUNC("can't find call dest at %s+0x%lx",
+                                 insn->sec, insn->offset,
+                                 insn->sec->name, dest_off);
+                       return -1;
+               }
+       }
+
+       return 0;
+}
+
 static void mark_rodata(struct objtool_file *file)
 {
        struct section *sec;
@@ -1357,8 +1497,8 @@ static void mark_rodata(struct objtool_file *file)
         * .rodata.str1.* sections are ignored; they don't contain jump tables.
         */
        for_each_sec(file, sec) {
-               if ((!strncmp(sec->name, ".rodata", 7) && !strstr(sec->name, ".str1.")) ||
-                   !strcmp(sec->name, C_JUMP_TABLE_SECTION)) {
+               if (!strncmp(sec->name, ".rodata", 7) &&
+                   !strstr(sec->name, ".str1.")) {
                        sec->rodata = true;
                        found = true;
                }
@@ -1396,6 +1536,10 @@ static int decode_sections(struct objtool_file *file)
        if (ret)
                return ret;
 
+       ret = read_intra_function_calls(file);
+       if (ret)
+               return ret;
+
        ret = add_call_destinations(file);
        if (ret)
                return ret;
@@ -1412,12 +1556,16 @@ static int decode_sections(struct objtool_file *file)
        if (ret)
                return ret;
 
+       ret = read_instr_hints(file);
+       if (ret)
+               return ret;
+
        return 0;
 }
 
 static bool is_fentry_call(struct instruction *insn)
 {
-       if (insn->type == INSN_CALL &&
+       if (insn->type == INSN_CALL && insn->call_dest &&
            insn->call_dest->type == STT_NOTYPE &&
            !strcmp(insn->call_dest->name, "__fentry__"))
                return true;
@@ -1425,40 +1573,57 @@ static bool is_fentry_call(struct instruction *insn)
        return false;
 }
 
-static bool has_modified_stack_frame(struct insn_state *state)
+static bool has_modified_stack_frame(struct instruction *insn, struct insn_state *state)
 {
+       u8 ret_offset = insn->ret_offset;
+       struct cfi_state *cfi = &state->cfi;
        int i;
 
-       if (state->cfa.base != initial_func_cfi.cfa.base ||
-           state->cfa.offset != initial_func_cfi.cfa.offset ||
-           state->stack_size != initial_func_cfi.cfa.offset ||
-           state->drap)
+       if (cfi->cfa.base != initial_func_cfi.cfa.base || cfi->drap)
+               return true;
+
+       if (cfi->cfa.offset != initial_func_cfi.cfa.offset + ret_offset)
+               return true;
+
+       if (cfi->stack_size != initial_func_cfi.cfa.offset + ret_offset)
                return true;
 
-       for (i = 0; i < CFI_NUM_REGS; i++)
-               if (state->regs[i].base != initial_func_cfi.regs[i].base ||
-                   state->regs[i].offset != initial_func_cfi.regs[i].offset)
+       /*
+        * If there is a ret offset hint then don't check registers
+        * because a callee-saved register might have been pushed on
+        * the stack.
+        */
+       if (ret_offset)
+               return false;
+
+       for (i = 0; i < CFI_NUM_REGS; i++) {
+               if (cfi->regs[i].base != initial_func_cfi.regs[i].base ||
+                   cfi->regs[i].offset != initial_func_cfi.regs[i].offset)
                        return true;
+       }
 
        return false;
 }
 
 static bool has_valid_stack_frame(struct insn_state *state)
 {
-       if (state->cfa.base == CFI_BP && state->regs[CFI_BP].base == CFI_CFA &&
-           state->regs[CFI_BP].offset == -16)
+       struct cfi_state *cfi = &state->cfi;
+
+       if (cfi->cfa.base == CFI_BP && cfi->regs[CFI_BP].base == CFI_CFA &&
+           cfi->regs[CFI_BP].offset == -16)
                return true;
 
-       if (state->drap && state->regs[CFI_BP].base == CFI_BP)
+       if (cfi->drap && cfi->regs[CFI_BP].base == CFI_BP)
                return true;
 
        return false;
 }
 
-static int update_insn_state_regs(struct instruction *insn, struct insn_state *state)
+static int update_cfi_state_regs(struct instruction *insn,
+                                 struct cfi_state *cfi,
+                                 struct stack_op *op)
 {
-       struct cfi_reg *cfa = &state->cfa;
-       struct stack_op *op = &insn->stack_op;
+       struct cfi_reg *cfa = &cfi->cfa;
 
        if (cfa->base != CFI_SP && cfa->base != CFI_SP_INDIRECT)
                return 0;
@@ -1479,20 +1644,19 @@ static int update_insn_state_regs(struct instruction *insn, struct insn_state *s
        return 0;
 }
 
-static void save_reg(struct insn_state *state, unsigned char reg, int base,
-                    int offset)
+static void save_reg(struct cfi_state *cfi, unsigned char reg, int base, int offset)
 {
        if (arch_callee_saved_reg(reg) &&
-           state->regs[reg].base == CFI_UNDEFINED) {
-               state->regs[reg].base = base;
-               state->regs[reg].offset = offset;
+           cfi->regs[reg].base == CFI_UNDEFINED) {
+               cfi->regs[reg].base = base;
+               cfi->regs[reg].offset = offset;
        }
 }
 
-static void restore_reg(struct insn_state *state, unsigned char reg)
+static void restore_reg(struct cfi_state *cfi, unsigned char reg)
 {
-       state->regs[reg].base = CFI_UNDEFINED;
-       state->regs[reg].offset = 0;
+       cfi->regs[reg].base = initial_func_cfi.regs[reg].base;
+       cfi->regs[reg].offset = initial_func_cfi.regs[reg].offset;
 }
 
 /*
@@ -1548,11 +1712,11 @@ static void restore_reg(struct insn_state *state, unsigned char reg)
  *   41 5d                     pop    %r13
  *   c3                                retq
  */
-static int update_insn_state(struct instruction *insn, struct insn_state *state)
+static int update_cfi_state(struct instruction *insn, struct cfi_state *cfi,
+                            struct stack_op *op)
 {
-       struct stack_op *op = &insn->stack_op;
-       struct cfi_reg *cfa = &state->cfa;
-       struct cfi_reg *regs = state->regs;
+       struct cfi_reg *cfa = &cfi->cfa;
+       struct cfi_reg *regs = cfi->regs;
 
        /* stack operations don't make sense with an undefined CFA */
        if (cfa->base == CFI_UNDEFINED) {
@@ -1563,8 +1727,8 @@ static int update_insn_state(struct instruction *insn, struct insn_state *state)
                return 0;
        }
 
-       if (state->type == ORC_TYPE_REGS || state->type == ORC_TYPE_REGS_IRET)
-               return update_insn_state_regs(insn, state);
+       if (cfi->type == ORC_TYPE_REGS || cfi->type == ORC_TYPE_REGS_IRET)
+               return update_cfi_state_regs(insn, cfi, op);
 
        switch (op->dest.type) {
 
@@ -1579,16 +1743,16 @@ static int update_insn_state(struct instruction *insn, struct insn_state *state)
 
                                /* mov %rsp, %rbp */
                                cfa->base = op->dest.reg;
-                               state->bp_scratch = false;
+                               cfi->bp_scratch = false;
                        }
 
                        else if (op->src.reg == CFI_SP &&
-                                op->dest.reg == CFI_BP && state->drap) {
+                                op->dest.reg == CFI_BP && cfi->drap) {
 
                                /* drap: mov %rsp, %rbp */
                                regs[CFI_BP].base = CFI_BP;
-                               regs[CFI_BP].offset = -state->stack_size;
-                               state->bp_scratch = false;
+                               regs[CFI_BP].offset = -cfi->stack_size;
+                               cfi->bp_scratch = false;
                        }
 
                        else if (op->src.reg == CFI_SP && cfa->base == CFI_SP) {
@@ -1603,8 +1767,8 @@ static int update_insn_state(struct instruction *insn, struct insn_state *state)
                                 *   ...
                                 *   mov    %rax, %rsp
                                 */
-                               state->vals[op->dest.reg].base = CFI_CFA;
-                               state->vals[op->dest.reg].offset = -state->stack_size;
+                               cfi->vals[op->dest.reg].base = CFI_CFA;
+                               cfi->vals[op->dest.reg].offset = -cfi->stack_size;
                        }
 
                        else if (op->src.reg == CFI_BP && op->dest.reg == CFI_SP &&
@@ -1615,14 +1779,14 @@ static int update_insn_state(struct instruction *insn, struct insn_state *state)
                                 *
                                 * Restore the original stack pointer (Clang).
                                 */
-                               state->stack_size = -state->regs[CFI_BP].offset;
+                               cfi->stack_size = -cfi->regs[CFI_BP].offset;
                        }
 
                        else if (op->dest.reg == cfa->base) {
 
                                /* mov %reg, %rsp */
                                if (cfa->base == CFI_SP &&
-                                   state->vals[op->src.reg].base == CFI_CFA) {
+                                   cfi->vals[op->src.reg].base == CFI_CFA) {
 
                                        /*
                                         * This is needed for the rare case
@@ -1632,8 +1796,8 @@ static int update_insn_state(struct instruction *insn, struct insn_state *state)
                                         *   ...
                                         *   mov    %rcx, %rsp
                                         */
-                                       cfa->offset = -state->vals[op->src.reg].offset;
-                                       state->stack_size = cfa->offset;
+                                       cfa->offset = -cfi->vals[op->src.reg].offset;
+                                       cfi->stack_size = cfa->offset;
 
                                } else {
                                        cfa->base = CFI_UNDEFINED;
@@ -1647,7 +1811,7 @@ static int update_insn_state(struct instruction *insn, struct insn_state *state)
                        if (op->dest.reg == CFI_SP && op->src.reg == CFI_SP) {
 
                                /* add imm, %rsp */
-                               state->stack_size -= op->src.offset;
+                               cfi->stack_size -= op->src.offset;
                                if (cfa->base == CFI_SP)
                                        cfa->offset -= op->src.offset;
                                break;
@@ -1656,14 +1820,14 @@ static int update_insn_state(struct instruction *insn, struct insn_state *state)
                        if (op->dest.reg == CFI_SP && op->src.reg == CFI_BP) {
 
                                /* lea disp(%rbp), %rsp */
-                               state->stack_size = -(op->src.offset + regs[CFI_BP].offset);
+                               cfi->stack_size = -(op->src.offset + regs[CFI_BP].offset);
                                break;
                        }
 
                        if (op->src.reg == CFI_SP && cfa->base == CFI_SP) {
 
                                /* drap: lea disp(%rsp), %drap */
-                               state->drap_reg = op->dest.reg;
+                               cfi->drap_reg = op->dest.reg;
 
                                /*
                                 * lea disp(%rsp), %reg
@@ -1675,25 +1839,25 @@ static int update_insn_state(struct instruction *insn, struct insn_state *state)
                                 *   ...
                                 *   mov    %rcx, %rsp
                                 */
-                               state->vals[op->dest.reg].base = CFI_CFA;
-                               state->vals[op->dest.reg].offset = \
-                                       -state->stack_size + op->src.offset;
+                               cfi->vals[op->dest.reg].base = CFI_CFA;
+                               cfi->vals[op->dest.reg].offset = \
+                                       -cfi->stack_size + op->src.offset;
 
                                break;
                        }
 
-                       if (state->drap && op->dest.reg == CFI_SP &&
-                           op->src.reg == state->drap_reg) {
+                       if (cfi->drap && op->dest.reg == CFI_SP &&
+                           op->src.reg == cfi->drap_reg) {
 
                                 /* drap: lea disp(%drap), %rsp */
                                cfa->base = CFI_SP;
-                               cfa->offset = state->stack_size = -op->src.offset;
-                               state->drap_reg = CFI_UNDEFINED;
-                               state->drap = false;
+                               cfa->offset = cfi->stack_size = -op->src.offset;
+                               cfi->drap_reg = CFI_UNDEFINED;
+                               cfi->drap = false;
                                break;
                        }
 
-                       if (op->dest.reg == state->cfa.base) {
+                       if (op->dest.reg == cfi->cfa.base) {
                                WARN_FUNC("unsupported stack register modification",
                                          insn->sec, insn->offset);
                                return -1;
@@ -1703,18 +1867,18 @@ static int update_insn_state(struct instruction *insn, struct insn_state *state)
 
                case OP_SRC_AND:
                        if (op->dest.reg != CFI_SP ||
-                           (state->drap_reg != CFI_UNDEFINED && cfa->base != CFI_SP) ||
-                           (state->drap_reg == CFI_UNDEFINED && cfa->base != CFI_BP)) {
+                           (cfi->drap_reg != CFI_UNDEFINED && cfa->base != CFI_SP) ||
+                           (cfi->drap_reg == CFI_UNDEFINED && cfa->base != CFI_BP)) {
                                WARN_FUNC("unsupported stack pointer realignment",
                                          insn->sec, insn->offset);
                                return -1;
                        }
 
-                       if (state->drap_reg != CFI_UNDEFINED) {
+                       if (cfi->drap_reg != CFI_UNDEFINED) {
                                /* drap: and imm, %rsp */
-                               cfa->base = state->drap_reg;
-                               cfa->offset = state->stack_size = 0;
-                               state->drap = true;
+                               cfa->base = cfi->drap_reg;
+                               cfa->offset = cfi->stack_size = 0;
+                               cfi->drap = true;
                        }
 
                        /*
@@ -1726,57 +1890,55 @@ static int update_insn_state(struct instruction *insn, struct insn_state *state)
 
                case OP_SRC_POP:
                case OP_SRC_POPF:
-                       if (!state->drap && op->dest.type == OP_DEST_REG &&
-                           op->dest.reg == cfa->base) {
+                       if (!cfi->drap && op->dest.reg == cfa->base) {
 
                                /* pop %rbp */
                                cfa->base = CFI_SP;
                        }
 
-                       if (state->drap && cfa->base == CFI_BP_INDIRECT &&
-                           op->dest.type == OP_DEST_REG &&
-                           op->dest.reg == state->drap_reg &&
-                           state->drap_offset == -state->stack_size) {
+                       if (cfi->drap && cfa->base == CFI_BP_INDIRECT &&
+                           op->dest.reg == cfi->drap_reg &&
+                           cfi->drap_offset == -cfi->stack_size) {
 
                                /* drap: pop %drap */
-                               cfa->base = state->drap_reg;
+                               cfa->base = cfi->drap_reg;
                                cfa->offset = 0;
-                               state->drap_offset = -1;
+                               cfi->drap_offset = -1;
 
-                       } else if (regs[op->dest.reg].offset == -state->stack_size) {
+                       } else if (regs[op->dest.reg].offset == -cfi->stack_size) {
 
                                /* pop %reg */
-                               restore_reg(state, op->dest.reg);
+                               restore_reg(cfi, op->dest.reg);
                        }
 
-                       state->stack_size -= 8;
+                       cfi->stack_size -= 8;
                        if (cfa->base == CFI_SP)
                                cfa->offset -= 8;
 
                        break;
 
                case OP_SRC_REG_INDIRECT:
-                       if (state->drap && op->src.reg == CFI_BP &&
-                           op->src.offset == state->drap_offset) {
+                       if (cfi->drap && op->src.reg == CFI_BP &&
+                           op->src.offset == cfi->drap_offset) {
 
                                /* drap: mov disp(%rbp), %drap */
-                               cfa->base = state->drap_reg;
+                               cfa->base = cfi->drap_reg;
                                cfa->offset = 0;
-                               state->drap_offset = -1;
+                               cfi->drap_offset = -1;
                        }
 
-                       if (state->drap && op->src.reg == CFI_BP &&
+                       if (cfi->drap && op->src.reg == CFI_BP &&
                            op->src.offset == regs[op->dest.reg].offset) {
 
                                /* drap: mov disp(%rbp), %reg */
-                               restore_reg(state, op->dest.reg);
+                               restore_reg(cfi, op->dest.reg);
 
                        } else if (op->src.reg == cfa->base &&
                            op->src.offset == regs[op->dest.reg].offset + cfa->offset) {
 
                                /* mov disp(%rbp), %reg */
                                /* mov disp(%rsp), %reg */
-                               restore_reg(state, op->dest.reg);
+                               restore_reg(cfi, op->dest.reg);
                        }
 
                        break;
@@ -1791,78 +1953,78 @@ static int update_insn_state(struct instruction *insn, struct insn_state *state)
 
        case OP_DEST_PUSH:
        case OP_DEST_PUSHF:
-               state->stack_size += 8;
+               cfi->stack_size += 8;
                if (cfa->base == CFI_SP)
                        cfa->offset += 8;
 
                if (op->src.type != OP_SRC_REG)
                        break;
 
-               if (state->drap) {
-                       if (op->src.reg == cfa->base && op->src.reg == state->drap_reg) {
+               if (cfi->drap) {
+                       if (op->src.reg == cfa->base && op->src.reg == cfi->drap_reg) {
 
                                /* drap: push %drap */
                                cfa->base = CFI_BP_INDIRECT;
-                               cfa->offset = -state->stack_size;
+                               cfa->offset = -cfi->stack_size;
 
                                /* save drap so we know when to restore it */
-                               state->drap_offset = -state->stack_size;
+                               cfi->drap_offset = -cfi->stack_size;
 
-                       } else if (op->src.reg == CFI_BP && cfa->base == state->drap_reg) {
+                       } else if (op->src.reg == CFI_BP && cfa->base == cfi->drap_reg) {
 
                                /* drap: push %rbp */
-                               state->stack_size = 0;
+                               cfi->stack_size = 0;
 
                        } else if (regs[op->src.reg].base == CFI_UNDEFINED) {
 
                                /* drap: push %reg */
-                               save_reg(state, op->src.reg, CFI_BP, -state->stack_size);
+                               save_reg(cfi, op->src.reg, CFI_BP, -cfi->stack_size);
                        }
 
                } else {
 
                        /* push %reg */
-                       save_reg(state, op->src.reg, CFI_CFA, -state->stack_size);
+                       save_reg(cfi, op->src.reg, CFI_CFA, -cfi->stack_size);
                }
 
                /* detect when asm code uses rbp as a scratch register */
                if (!no_fp && insn->func && op->src.reg == CFI_BP &&
                    cfa->base != CFI_BP)
-                       state->bp_scratch = true;
+                       cfi->bp_scratch = true;
                break;
 
        case OP_DEST_REG_INDIRECT:
 
-               if (state->drap) {
-                       if (op->src.reg == cfa->base && op->src.reg == state->drap_reg) {
+               if (cfi->drap) {
+                       if (op->src.reg == cfa->base && op->src.reg == cfi->drap_reg) {
 
                                /* drap: mov %drap, disp(%rbp) */
                                cfa->base = CFI_BP_INDIRECT;
                                cfa->offset = op->dest.offset;
 
                                /* save drap offset so we know when to restore it */
-                               state->drap_offset = op->dest.offset;
+                               cfi->drap_offset = op->dest.offset;
                        }
 
                        else if (regs[op->src.reg].base == CFI_UNDEFINED) {
 
                                /* drap: mov reg, disp(%rbp) */
-                               save_reg(state, op->src.reg, CFI_BP, op->dest.offset);
+                               save_reg(cfi, op->src.reg, CFI_BP, op->dest.offset);
                        }
 
                } else if (op->dest.reg == cfa->base) {
 
                        /* mov reg, disp(%rbp) */
                        /* mov reg, disp(%rsp) */
-                       save_reg(state, op->src.reg, CFI_CFA,
-                                op->dest.offset - state->cfa.offset);
+                       save_reg(cfi, op->src.reg, CFI_CFA,
+                                op->dest.offset - cfi->cfa.offset);
                }
 
                break;
 
        case OP_DEST_LEAVE:
-               if ((!state->drap && cfa->base != CFI_BP) ||
-                   (state->drap && cfa->base != state->drap_reg)) {
+               if ((!cfi->drap && cfa->base != CFI_BP) ||
+                   (cfi->drap && cfa->base != cfi->drap_reg)) {
                        WARN_FUNC("leave instruction with modified stack frame",
                                  insn->sec, insn->offset);
                        return -1;
@@ -1870,10 +2032,10 @@ static int update_insn_state(struct instruction *insn, struct insn_state *state)
 
                /* leave (mov %rbp, %rsp; pop %rbp) */
 
-               state->stack_size = -state->regs[CFI_BP].offset - 8;
-               restore_reg(state, CFI_BP);
+               cfi->stack_size = -cfi->regs[CFI_BP].offset - 8;
+               restore_reg(cfi, CFI_BP);
 
-               if (!state->drap) {
+               if (!cfi->drap) {
                        cfa->base = CFI_SP;
                        cfa->offset -= 8;
                }
@@ -1888,7 +2050,7 @@ static int update_insn_state(struct instruction *insn, struct insn_state *state)
                }
 
                /* pop mem */
-               state->stack_size -= 8;
+               cfi->stack_size -= 8;
                if (cfa->base == CFI_SP)
                        cfa->offset -= 8;
 
@@ -1903,41 +2065,86 @@ static int update_insn_state(struct instruction *insn, struct insn_state *state)
        return 0;
 }
 
-static bool insn_state_match(struct instruction *insn, struct insn_state *state)
+static int handle_insn_ops(struct instruction *insn, struct insn_state *state)
+{
+       struct stack_op *op;
+
+       list_for_each_entry(op, &insn->stack_ops, list) {
+               struct cfi_state old_cfi = state->cfi;
+               int res;
+
+               res = update_cfi_state(insn, &state->cfi, op);
+               if (res)
+                       return res;
+
+               if (insn->alt_group && memcmp(&state->cfi, &old_cfi, sizeof(struct cfi_state))) {
+                       WARN_FUNC("alternative modifies stack", insn->sec, insn->offset);
+                       return -1;
+               }
+
+               if (op->dest.type == OP_DEST_PUSHF) {
+                       if (!state->uaccess_stack) {
+                               state->uaccess_stack = 1;
+                       } else if (state->uaccess_stack >> 31) {
+                               WARN_FUNC("PUSHF stack exhausted",
+                                         insn->sec, insn->offset);
+                               return 1;
+                       }
+                       state->uaccess_stack <<= 1;
+                       state->uaccess_stack  |= state->uaccess;
+               }
+
+               if (op->src.type == OP_SRC_POPF) {
+                       if (state->uaccess_stack) {
+                               state->uaccess = state->uaccess_stack & 1;
+                               state->uaccess_stack >>= 1;
+                               if (state->uaccess_stack == 1)
+                                       state->uaccess_stack = 0;
+                       }
+               }
+       }
+
+       return 0;
+}
+
+static bool insn_cfi_match(struct instruction *insn, struct cfi_state *cfi2)
 {
-       struct insn_state *state1 = &insn->state, *state2 = state;
+       struct cfi_state *cfi1 = &insn->cfi;
        int i;
 
-       if (memcmp(&state1->cfa, &state2->cfa, sizeof(state1->cfa))) {
+       if (memcmp(&cfi1->cfa, &cfi2->cfa, sizeof(cfi1->cfa))) {
+
                WARN_FUNC("stack state mismatch: cfa1=%d%+d cfa2=%d%+d",
                          insn->sec, insn->offset,
-                         state1->cfa.base, state1->cfa.offset,
-                         state2->cfa.base, state2->cfa.offset);
+                         cfi1->cfa.base, cfi1->cfa.offset,
+                         cfi2->cfa.base, cfi2->cfa.offset);
 
-       } else if (memcmp(&state1->regs, &state2->regs, sizeof(state1->regs))) {
+       } else if (memcmp(&cfi1->regs, &cfi2->regs, sizeof(cfi1->regs))) {
                for (i = 0; i < CFI_NUM_REGS; i++) {
-                       if (!memcmp(&state1->regs[i], &state2->regs[i],
+                       if (!memcmp(&cfi1->regs[i], &cfi2->regs[i],
                                    sizeof(struct cfi_reg)))
                                continue;
 
                        WARN_FUNC("stack state mismatch: reg1[%d]=%d%+d reg2[%d]=%d%+d",
                                  insn->sec, insn->offset,
-                                 i, state1->regs[i].base, state1->regs[i].offset,
-                                 i, state2->regs[i].base, state2->regs[i].offset);
+                                 i, cfi1->regs[i].base, cfi1->regs[i].offset,
+                                 i, cfi2->regs[i].base, cfi2->regs[i].offset);
                        break;
                }
 
-       } else if (state1->type != state2->type) {
+       } else if (cfi1->type != cfi2->type) {
+
                WARN_FUNC("stack state mismatch: type1=%d type2=%d",
-                         insn->sec, insn->offset, state1->type, state2->type);
+                         insn->sec, insn->offset, cfi1->type, cfi2->type);
+
+       } else if (cfi1->drap != cfi2->drap ||
+                  (cfi1->drap && cfi1->drap_reg != cfi2->drap_reg) ||
+                  (cfi1->drap && cfi1->drap_offset != cfi2->drap_offset)) {
 
-       } else if (state1->drap != state2->drap ||
-                (state1->drap && state1->drap_reg != state2->drap_reg) ||
-                (state1->drap && state1->drap_offset != state2->drap_offset)) {
                WARN_FUNC("stack state mismatch: drap1=%d(%d,%d) drap2=%d(%d,%d)",
                          insn->sec, insn->offset,
-                         state1->drap, state1->drap_reg, state1->drap_offset,
-                         state2->drap, state2->drap_reg, state2->drap_offset);
+                         cfi1->drap, cfi1->drap_reg, cfi1->drap_offset,
+                         cfi2->drap, cfi2->drap_reg, cfi2->drap_offset);
 
        } else
                return true;
@@ -1963,6 +2170,13 @@ static inline const char *call_dest_name(struct instruction *insn)
 
 static int validate_call(struct instruction *insn, struct insn_state *state)
 {
+       if (state->noinstr && state->instr <= 0 &&
+           (!insn->call_dest || !insn->call_dest->sec->noinstr)) {
+               WARN_FUNC("call to %s() leaves .noinstr.text section",
+                               insn->sec, insn->offset, call_dest_name(insn));
+               return 1;
+       }
+
        if (state->uaccess && !func_uaccess_safe(insn->call_dest)) {
                WARN_FUNC("call to %s() with UACCESS enabled",
                                insn->sec, insn->offset, call_dest_name(insn));
@@ -1980,7 +2194,7 @@ static int validate_call(struct instruction *insn, struct insn_state *state)
 
 static int validate_sibling_call(struct instruction *insn, struct insn_state *state)
 {
-       if (has_modified_stack_frame(state)) {
+       if (has_modified_stack_frame(insn, state)) {
                WARN_FUNC("sibling call from callable instruction with modified stack frame",
                                insn->sec, insn->offset);
                return 1;
@@ -1991,6 +2205,12 @@ static int validate_sibling_call(struct instruction *insn, struct insn_state *st
 
 static int validate_return(struct symbol *func, struct instruction *insn, struct insn_state *state)
 {
+       if (state->noinstr && state->instr > 0) {
+               WARN_FUNC("return with instrumentation enabled",
+                         insn->sec, insn->offset);
+               return 1;
+       }
+
        if (state->uaccess && !func_uaccess_safe(func)) {
                WARN_FUNC("return with UACCESS enabled",
                          insn->sec, insn->offset);
@@ -2009,13 +2229,13 @@ static int validate_return(struct symbol *func, struct instruction *insn, struct
                return 1;
        }
 
-       if (func && has_modified_stack_frame(state)) {
+       if (func && has_modified_stack_frame(insn, state)) {
                WARN_FUNC("return with modified stack frame",
                          insn->sec, insn->offset);
                return 1;
        }
 
-       if (state->bp_scratch) {
+       if (state->cfi.bp_scratch) {
                WARN_FUNC("BP used as a scratch register",
                          insn->sec, insn->offset);
                return 1;
@@ -2024,6 +2244,30 @@ static int validate_return(struct symbol *func, struct instruction *insn, struct
        return 0;
 }
 
+/*
+ * Alternatives should not contain any ORC entries, this in turn means they
+ * should not contain any CFI ops, which implies all instructions should have
+ * the same same CFI state.
+ *
+ * It is possible to constuct alternatives that have unreachable holes that go
+ * unreported (because they're NOPs), such holes would result in CFI_UNDEFINED
+ * states which then results in ORC entries, which we just said we didn't want.
+ *
+ * Avoid them by copying the CFI entry of the first instruction into the whole
+ * alternative.
+ */
+static void fill_alternative_cfi(struct objtool_file *file, struct instruction *insn)
+{
+       struct instruction *first_insn = insn;
+       int alt_group = insn->alt_group;
+
+       sec_for_each_insn_continue(file, insn) {
+               if (insn->alt_group != alt_group)
+                       break;
+               insn->cfi = first_insn->cfi;
+       }
+}
+
 /*
  * Follow the branch starting at the given instruction, and recursively follow
  * any other branches (jumps).  Meanwhile, track the frame pointer state at
@@ -2031,23 +2275,16 @@ static int validate_return(struct symbol *func, struct instruction *insn, struct
  * tools/objtool/Documentation/stack-validation.txt.
  */
 static int validate_branch(struct objtool_file *file, struct symbol *func,
-                          struct instruction *first, struct insn_state state)
+                          struct instruction *insn, struct insn_state state)
 {
        struct alternative *alt;
-       struct instruction *insn, *next_insn;
+       struct instruction *next_insn;
        struct section *sec;
        u8 visited;
        int ret;
 
-       insn = first;
        sec = insn->sec;
 
-       if (insn->alt_group && list_empty(&insn->alts)) {
-               WARN_FUNC("don't know how to handle branch to middle of alternative instruction group",
-                         sec, insn->offset);
-               return 1;
-       }
-
        while (1) {
                next_insn = next_insn_same_sec(file, insn);
 
@@ -2065,59 +2302,24 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,
 
                visited = 1 << state.uaccess;
                if (insn->visited) {
-                       if (!insn->hint && !insn_state_match(insn, &state))
+                       if (!insn->hint && !insn_cfi_match(insn, &state.cfi))
                                return 1;
 
                        if (insn->visited & visited)
                                return 0;
                }
 
-               if (insn->hint) {
-                       if (insn->restore) {
-                               struct instruction *save_insn, *i;
-
-                               i = insn;
-                               save_insn = NULL;
-                               sym_for_each_insn_continue_reverse(file, func, i) {
-                                       if (i->save) {
-                                               save_insn = i;
-                                               break;
-                                       }
-                               }
-
-                               if (!save_insn) {
-                                       WARN_FUNC("no corresponding CFI save for CFI restore",
-                                                 sec, insn->offset);
-                                       return 1;
-                               }
-
-                               if (!save_insn->visited) {
-                                       /*
-                                        * Oops, no state to copy yet.
-                                        * Hopefully we can reach this
-                                        * instruction from another branch
-                                        * after the save insn has been
-                                        * visited.
-                                        */
-                                       if (insn == first)
-                                               return 0;
-
-                                       WARN_FUNC("objtool isn't smart enough to handle this CFI save/restore combo",
-                                                 sec, insn->offset);
-                                       return 1;
-                               }
-
-                               insn->state = save_insn->state;
-                       }
-
-                       state = insn->state;
+               if (state.noinstr)
+                       state.instr += insn->instr;
 
-               } else
-                       insn->state = state;
+               if (insn->hint)
+                       state.cfi = insn->cfi;
+               else
+                       insn->cfi = state.cfi;
 
                insn->visited |= visited;
 
-               if (!insn->ignore_alts) {
+               if (!insn->ignore_alts && !list_empty(&insn->alts)) {
                        bool skip_orig = false;
 
                        list_for_each_entry(alt, &insn->alts, list) {
@@ -2132,10 +2334,16 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,
                                }
                        }
 
+                       if (insn->alt_group)
+                               fill_alternative_cfi(file, insn);
+
                        if (skip_orig)
                                return 0;
                }
 
+               if (handle_insn_ops(insn, &state))
+                       return 1;
+
                switch (insn->type) {
 
                case INSN_RETURN:
@@ -2202,32 +2410,6 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,
                        }
                        return 0;
 
-               case INSN_STACK:
-                       if (update_insn_state(insn, &state))
-                               return 1;
-
-                       if (insn->stack_op.dest.type == OP_DEST_PUSHF) {
-                               if (!state.uaccess_stack) {
-                                       state.uaccess_stack = 1;
-                               } else if (state.uaccess_stack >> 31) {
-                                       WARN_FUNC("PUSHF stack exhausted", sec, insn->offset);
-                                       return 1;
-                               }
-                               state.uaccess_stack <<= 1;
-                               state.uaccess_stack  |= state.uaccess;
-                       }
-
-                       if (insn->stack_op.src.type == OP_SRC_POPF) {
-                               if (state.uaccess_stack) {
-                                       state.uaccess = state.uaccess_stack & 1;
-                                       state.uaccess_stack >>= 1;
-                                       if (state.uaccess_stack == 1)
-                                               state.uaccess_stack = 0;
-                               }
-                       }
-
-                       break;
-
                case INSN_STAC:
                        if (state.uaccess) {
                                WARN_FUNC("recursive UACCESS enable", sec, insn->offset);
@@ -2273,7 +2455,7 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,
                        return 0;
 
                if (!next_insn) {
-                       if (state.cfa.base == CFI_UNDEFINED)
+                       if (state.cfi.cfa.base == CFI_UNDEFINED)
                                return 0;
                        WARN("%s: unexpected end of section", sec->name);
                        return 1;
@@ -2285,24 +2467,34 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,
        return 0;
 }
 
-static int validate_unwind_hints(struct objtool_file *file)
+static int validate_unwind_hints(struct objtool_file *file, struct section *sec)
 {
        struct instruction *insn;
-       int ret, warnings = 0;
        struct insn_state state;
+       int ret, warnings = 0;
 
        if (!file->hints)
                return 0;
 
-       clear_insn_state(&state);
+       init_insn_state(&state, sec);
 
-       for_each_insn(file, insn) {
+       if (sec) {
+               insn = find_insn(file, sec, 0);
+               if (!insn)
+                       return 0;
+       } else {
+               insn = list_first_entry(&file->insn_list, typeof(*insn), list);
+       }
+
+       while (&insn->list != &file->insn_list && (!sec || insn->sec == sec)) {
                if (insn->hint && !insn->visited) {
                        ret = validate_branch(file, insn->func, insn, state);
                        if (ret && backtrace)
                                BT_FUNC("<=== (hint)", insn);
                        warnings += ret;
                }
+
+               insn = list_next_entry(insn, list);
        }
 
        return warnings;
@@ -2417,43 +2609,69 @@ static bool ignore_unreachable_insn(struct instruction *insn)
        return false;
 }
 
-static int validate_section(struct objtool_file *file, struct section *sec)
+static int validate_symbol(struct objtool_file *file, struct section *sec,
+                          struct symbol *sym, struct insn_state *state)
 {
-       struct symbol *func;
        struct instruction *insn;
-       struct insn_state state;
-       int ret, warnings = 0;
+       int ret;
+
+       if (!sym->len) {
+               WARN("%s() is missing an ELF size annotation", sym->name);
+               return 1;
+       }
+
+       if (sym->pfunc != sym || sym->alias != sym)
+               return 0;
 
-       clear_insn_state(&state);
+       insn = find_insn(file, sec, sym->offset);
+       if (!insn || insn->ignore || insn->visited)
+               return 0;
+
+       state->uaccess = sym->uaccess_safe;
+
+       ret = validate_branch(file, insn->func, insn, *state);
+       if (ret && backtrace)
+               BT_FUNC("<=== (sym)", insn);
+       return ret;
+}
 
-       state.cfa = initial_func_cfi.cfa;
-       memcpy(&state.regs, &initial_func_cfi.regs,
-              CFI_NUM_REGS * sizeof(struct cfi_reg));
-       state.stack_size = initial_func_cfi.cfa.offset;
+static int validate_section(struct objtool_file *file, struct section *sec)
+{
+       struct insn_state state;
+       struct symbol *func;
+       int warnings = 0;
 
        list_for_each_entry(func, &sec->symbol_list, list) {
                if (func->type != STT_FUNC)
                        continue;
 
-               if (!func->len) {
-                       WARN("%s() is missing an ELF size annotation",
-                            func->name);
-                       warnings++;
-               }
+               init_insn_state(&state, sec);
+               state.cfi.cfa = initial_func_cfi.cfa;
+               memcpy(&state.cfi.regs, &initial_func_cfi.regs,
+                      CFI_NUM_REGS * sizeof(struct cfi_reg));
+               state.cfi.stack_size = initial_func_cfi.cfa.offset;
 
-               if (func->pfunc != func || func->alias != func)
-                       continue;
+               warnings += validate_symbol(file, sec, func, &state);
+       }
 
-               insn = find_insn(file, sec, func->offset);
-               if (!insn || insn->ignore || insn->visited)
-                       continue;
+       return warnings;
+}
 
-               state.uaccess = func->uaccess_safe;
+static int validate_vmlinux_functions(struct objtool_file *file)
+{
+       struct section *sec;
+       int warnings = 0;
 
-               ret = validate_branch(file, func, insn, state);
-               if (ret && backtrace)
-                       BT_FUNC("<=== (func)", insn);
-               warnings += ret;
+       sec = find_section_by_name(file->elf, ".noinstr.text");
+       if (sec) {
+               warnings += validate_section(file, sec);
+               warnings += validate_unwind_hints(file, sec);
+       }
+
+       sec = find_section_by_name(file->elf, ".entry.text");
+       if (sec) {
+               warnings += validate_section(file, sec);
+               warnings += validate_unwind_hints(file, sec);
        }
 
        return warnings;
@@ -2464,8 +2682,12 @@ static int validate_functions(struct objtool_file *file)
        struct section *sec;
        int warnings = 0;
 
-       for_each_sec(file, sec)
+       for_each_sec(file, sec) {
+               if (!(sec->sh.sh_flags & SHF_EXECINSTR))
+                       continue;
+
                warnings += validate_section(file, sec);
+       }
 
        return warnings;
 }
@@ -2496,7 +2718,7 @@ int check(const char *_objname, bool orc)
 
        objname = _objname;
 
-       file.elf = elf_read(objname, orc ? O_RDWR : O_RDONLY);
+       file.elf = elf_open_read(objname, orc ? O_RDWR : O_RDONLY);
        if (!file.elf)
                return 1;
 
@@ -2516,6 +2738,15 @@ int check(const char *_objname, bool orc)
        if (list_empty(&file.insn_list))
                goto out;
 
+       if (vmlinux && !validate_dup) {
+               ret = validate_vmlinux_functions(&file);
+               if (ret < 0)
+                       goto out;
+
+               warnings += ret;
+               goto out;
+       }
+
        if (retpoline) {
                ret = validate_retpoline(&file);
                if (ret < 0)
@@ -2528,7 +2759,7 @@ int check(const char *_objname, bool orc)
                goto out;
        warnings += ret;
 
-       ret = validate_unwind_hints(&file);
+       ret = validate_unwind_hints(&file, NULL);
        if (ret < 0)
                goto out;
        warnings += ret;
index f0ce8ff..906b521 100644 (file)
@@ -7,22 +7,16 @@
 #define _CHECK_H
 
 #include <stdbool.h>
-#include "elf.h"
 #include "cfi.h"
 #include "arch.h"
-#include "orc.h"
-#include <linux/hashtable.h>
 
 struct insn_state {
-       struct cfi_reg cfa;
-       struct cfi_reg regs[CFI_NUM_REGS];
-       int stack_size;
-       unsigned char type;
-       bool bp_scratch;
-       bool drap, end, uaccess, df;
+       struct cfi_state cfi;
        unsigned int uaccess_stack;
-       int drap_reg, drap_offset;
-       struct cfi_reg vals[CFI_NUM_REGS];
+       bool uaccess;
+       bool df;
+       bool noinstr;
+       s8 instr;
 };
 
 struct instruction {
@@ -33,29 +27,24 @@ struct instruction {
        unsigned int len;
        enum insn_type type;
        unsigned long immediate;
-       bool alt_group, dead_end, ignore, hint, save, restore, ignore_alts;
+       bool dead_end, ignore, ignore_alts;
+       bool hint;
        bool retpoline_safe;
+       s8 instr;
        u8 visited;
+       u8 ret_offset;
+       int alt_group;
        struct symbol *call_dest;
        struct instruction *jump_dest;
        struct instruction *first_jump_src;
        struct rela *jump_table;
        struct list_head alts;
        struct symbol *func;
-       struct stack_op stack_op;
-       struct insn_state state;
+       struct list_head stack_ops;
+       struct cfi_state cfi;
        struct orc_entry orc;
 };
 
-struct objtool_file {
-       struct elf *elf;
-       struct list_head insn_list;
-       DECLARE_HASHTABLE(insn_hash, 20);
-       bool ignore_unreachables, c_file, hints, rodata;
-};
-
-int check(const char *objname, bool orc);
-
 struct instruction *find_insn(struct objtool_file *file,
                              struct section *sec, unsigned long offset);
 
index c4857fa..8422567 100644 (file)
@@ -27,6 +27,22 @@ static inline u32 str_hash(const char *str)
        return jhash(str, strlen(str), 0);
 }
 
+static inline int elf_hash_bits(void)
+{
+       return vmlinux ? ELF_HASH_BITS : 16;
+}
+
+#define elf_hash_add(hashtable, node, key) \
+       hlist_add_head(node, &hashtable[hash_min(key, elf_hash_bits())])
+
+static void elf_hash_init(struct hlist_head *table)
+{
+       __hash_init(table, 1U << elf_hash_bits());
+}
+
+#define elf_hash_for_each_possible(name, obj, member, key)                     \
+       hlist_for_each_entry(obj, &name[hash_min(key, elf_hash_bits())], member)
+
 static void rb_add(struct rb_root *tree, struct rb_node *node,
                   int (*cmp)(struct rb_node *, const struct rb_node *))
 {
@@ -45,7 +61,7 @@ static void rb_add(struct rb_root *tree, struct rb_node *node,
        rb_insert_color(node, tree);
 }
 
-static struct rb_node *rb_find_first(struct rb_root *tree, const void *key,
+static struct rb_node *rb_find_first(const struct rb_root *tree, const void *key,
                               int (*cmp)(const void *key, const struct rb_node *))
 {
        struct rb_node *node = tree->rb_node;
@@ -111,11 +127,11 @@ static int symbol_by_offset(const void *key, const struct rb_node *node)
        return 0;
 }
 
-struct section *find_section_by_name(struct elf *elf, const char *name)
+struct section *find_section_by_name(const struct elf *elf, const char *name)
 {
        struct section *sec;
 
-       hash_for_each_possible(elf->section_name_hash, sec, name_hash, str_hash(name))
+       elf_hash_for_each_possible(elf->section_name_hash, sec, name_hash, str_hash(name))
                if (!strcmp(sec->name, name))
                        return sec;
 
@@ -127,7 +143,7 @@ static struct section *find_section_by_index(struct elf *elf,
 {
        struct section *sec;
 
-       hash_for_each_possible(elf->section_hash, sec, hash, idx)
+       elf_hash_for_each_possible(elf->section_hash, sec, hash, idx)
                if (sec->idx == idx)
                        return sec;
 
@@ -138,7 +154,7 @@ static struct symbol *find_symbol_by_index(struct elf *elf, unsigned int idx)
 {
        struct symbol *sym;
 
-       hash_for_each_possible(elf->symbol_hash, sym, hash, idx)
+       elf_hash_for_each_possible(elf->symbol_hash, sym, hash, idx)
                if (sym->idx == idx)
                        return sym;
 
@@ -173,7 +189,7 @@ struct symbol *find_func_by_offset(struct section *sec, unsigned long offset)
        return NULL;
 }
 
-struct symbol *find_symbol_containing(struct section *sec, unsigned long offset)
+struct symbol *find_symbol_containing(const struct section *sec, unsigned long offset)
 {
        struct rb_node *node;
 
@@ -201,18 +217,18 @@ struct symbol *find_func_containing(struct section *sec, unsigned long offset)
        return NULL;
 }
 
-struct symbol *find_symbol_by_name(struct elf *elf, const char *name)
+struct symbol *find_symbol_by_name(const struct elf *elf, const char *name)
 {
        struct symbol *sym;
 
-       hash_for_each_possible(elf->symbol_name_hash, sym, name_hash, str_hash(name))
+       elf_hash_for_each_possible(elf->symbol_name_hash, sym, name_hash, str_hash(name))
                if (!strcmp(sym->name, name))
                        return sym;
 
        return NULL;
 }
 
-struct rela *find_rela_by_dest_range(struct elf *elf, struct section *sec,
+struct rela *find_rela_by_dest_range(const struct elf *elf, struct section *sec,
                                     unsigned long offset, unsigned int len)
 {
        struct rela *rela, *r = NULL;
@@ -224,7 +240,7 @@ struct rela *find_rela_by_dest_range(struct elf *elf, struct section *sec,
        sec = sec->rela;
 
        for_offset_range(o, offset, offset + len) {
-               hash_for_each_possible(elf->rela_hash, rela, hash,
+               elf_hash_for_each_possible(elf->rela_hash, rela, hash,
                                       sec_offset_hash(sec, o)) {
                        if (rela->sec != sec)
                                continue;
@@ -241,7 +257,7 @@ struct rela *find_rela_by_dest_range(struct elf *elf, struct section *sec,
        return NULL;
 }
 
-struct rela *find_rela_by_dest(struct elf *elf, struct section *sec, unsigned long offset)
+struct rela *find_rela_by_dest(const struct elf *elf, struct section *sec, unsigned long offset)
 {
        return find_rela_by_dest_range(elf, sec, offset, 1);
 }
@@ -309,8 +325,8 @@ static int read_sections(struct elf *elf)
                sec->len = sec->sh.sh_size;
 
                list_add_tail(&sec->list, &elf->sections);
-               hash_add(elf->section_hash, &sec->hash, sec->idx);
-               hash_add(elf->section_name_hash, &sec->name_hash, str_hash(sec->name));
+               elf_hash_add(elf->section_hash, &sec->hash, sec->idx);
+               elf_hash_add(elf->section_name_hash, &sec->name_hash, str_hash(sec->name));
        }
 
        if (stats)
@@ -327,12 +343,14 @@ static int read_sections(struct elf *elf)
 
 static int read_symbols(struct elf *elf)
 {
-       struct section *symtab, *sec;
+       struct section *symtab, *symtab_shndx, *sec;
        struct symbol *sym, *pfunc;
        struct list_head *entry;
        struct rb_node *pnode;
        int symbols_nr, i;
        char *coldstr;
+       Elf_Data *shndx_data = NULL;
+       Elf32_Word shndx;
 
        symtab = find_section_by_name(elf, ".symtab");
        if (!symtab) {
@@ -340,6 +358,10 @@ static int read_symbols(struct elf *elf)
                return -1;
        }
 
+       symtab_shndx = find_section_by_name(elf, ".symtab_shndx");
+       if (symtab_shndx)
+               shndx_data = symtab_shndx->data;
+
        symbols_nr = symtab->sh.sh_size / symtab->sh.sh_entsize;
 
        for (i = 0; i < symbols_nr; i++) {
@@ -353,8 +375,9 @@ static int read_symbols(struct elf *elf)
 
                sym->idx = i;
 
-               if (!gelf_getsym(symtab->data, i, &sym->sym)) {
-                       WARN_ELF("gelf_getsym");
+               if (!gelf_getsymshndx(symtab->data, shndx_data, i, &sym->sym,
+                                     &shndx)) {
+                       WARN_ELF("gelf_getsymshndx");
                        goto err;
                }
 
@@ -368,10 +391,13 @@ static int read_symbols(struct elf *elf)
                sym->type = GELF_ST_TYPE(sym->sym.st_info);
                sym->bind = GELF_ST_BIND(sym->sym.st_info);
 
-               if (sym->sym.st_shndx > SHN_UNDEF &&
-                   sym->sym.st_shndx < SHN_LORESERVE) {
-                       sym->sec = find_section_by_index(elf,
-                                                        sym->sym.st_shndx);
+               if ((sym->sym.st_shndx > SHN_UNDEF &&
+                    sym->sym.st_shndx < SHN_LORESERVE) ||
+                   (shndx_data && sym->sym.st_shndx == SHN_XINDEX)) {
+                       if (sym->sym.st_shndx != SHN_XINDEX)
+                               shndx = sym->sym.st_shndx;
+
+                       sym->sec = find_section_by_index(elf, shndx);
                        if (!sym->sec) {
                                WARN("couldn't find section for symbol %s",
                                     sym->name);
@@ -394,8 +420,8 @@ static int read_symbols(struct elf *elf)
                else
                        entry = &sym->sec->symbol_list;
                list_add(&sym->list, entry);
-               hash_add(elf->symbol_hash, &sym->hash, sym->idx);
-               hash_add(elf->symbol_name_hash, &sym->name_hash, str_hash(sym->name));
+               elf_hash_add(elf->symbol_hash, &sym->hash, sym->idx);
+               elf_hash_add(elf->symbol_name_hash, &sym->name_hash, str_hash(sym->name));
        }
 
        if (stats)
@@ -456,6 +482,14 @@ err:
        return -1;
 }
 
+void elf_add_rela(struct elf *elf, struct rela *rela)
+{
+       struct section *sec = rela->sec;
+
+       list_add_tail(&rela->list, &sec->rela_list);
+       elf_hash_add(elf->rela_hash, &rela->hash, rela_hash(rela));
+}
+
 static int read_relas(struct elf *elf)
 {
        struct section *sec;
@@ -503,8 +537,7 @@ static int read_relas(struct elf *elf)
                                return -1;
                        }
 
-                       list_add_tail(&rela->list, &sec->rela_list);
-                       hash_add(elf->rela_hash, &rela->hash, rela_hash(rela));
+                       elf_add_rela(elf, rela);
                        nr_rela++;
                }
                max_rela = max(max_rela, nr_rela);
@@ -519,7 +552,7 @@ static int read_relas(struct elf *elf)
        return 0;
 }
 
-struct elf *elf_read(const char *name, int flags)
+struct elf *elf_open_read(const char *name, int flags)
 {
        struct elf *elf;
        Elf_Cmd cmd;
@@ -531,15 +564,16 @@ struct elf *elf_read(const char *name, int flags)
                perror("malloc");
                return NULL;
        }
-       memset(elf, 0, sizeof(*elf));
+       memset(elf, 0, offsetof(struct elf, sections));
 
-       hash_init(elf->symbol_hash);
-       hash_init(elf->symbol_name_hash);
-       hash_init(elf->section_hash);
-       hash_init(elf->section_name_hash);
-       hash_init(elf->rela_hash);
        INIT_LIST_HEAD(&elf->sections);
 
+       elf_hash_init(elf->symbol_hash);
+       elf_hash_init(elf->symbol_name_hash);
+       elf_hash_init(elf->section_hash);
+       elf_hash_init(elf->section_name_hash);
+       elf_hash_init(elf->rela_hash);
+
        elf->fd = open(name, flags);
        if (elf->fd == -1) {
                fprintf(stderr, "objtool: Can't open '%s': %s\n",
@@ -676,8 +710,8 @@ struct section *elf_create_section(struct elf *elf, const char *name,
        shstrtab->changed = true;
 
        list_add_tail(&sec->list, &elf->sections);
-       hash_add(elf->section_hash, &sec->hash, sec->idx);
-       hash_add(elf->section_name_hash, &sec->name_hash, str_hash(sec->name));
+       elf_hash_add(elf->section_hash, &sec->hash, sec->idx);
+       elf_hash_add(elf->section_name_hash, &sec->name_hash, str_hash(sec->name));
 
        return sec;
 }
@@ -745,7 +779,7 @@ int elf_rebuild_rela_section(struct section *sec)
        return 0;
 }
 
-int elf_write(struct elf *elf)
+int elf_write(const struct elf *elf)
 {
        struct section *sec;
        Elf_Scn *s;
index 12e01ac..f4fe1d6 100644 (file)
@@ -39,7 +39,7 @@ struct section {
        char *name;
        int idx;
        unsigned int len;
-       bool changed, text, rodata;
+       bool changed, text, rodata, noinstr;
 };
 
 struct symbol {
@@ -70,17 +70,19 @@ struct rela {
        bool jump_table_start;
 };
 
+#define ELF_HASH_BITS  20
+
 struct elf {
        Elf *elf;
        GElf_Ehdr ehdr;
        int fd;
        char *name;
        struct list_head sections;
-       DECLARE_HASHTABLE(symbol_hash, 20);
-       DECLARE_HASHTABLE(symbol_name_hash, 20);
-       DECLARE_HASHTABLE(section_hash, 16);
-       DECLARE_HASHTABLE(section_name_hash, 16);
-       DECLARE_HASHTABLE(rela_hash, 20);
+       DECLARE_HASHTABLE(symbol_hash, ELF_HASH_BITS);
+       DECLARE_HASHTABLE(symbol_name_hash, ELF_HASH_BITS);
+       DECLARE_HASHTABLE(section_hash, ELF_HASH_BITS);
+       DECLARE_HASHTABLE(section_name_hash, ELF_HASH_BITS);
+       DECLARE_HASHTABLE(rela_hash, ELF_HASH_BITS);
 };
 
 #define OFFSET_STRIDE_BITS     4
@@ -112,22 +114,23 @@ static inline u32 rela_hash(struct rela *rela)
        return sec_offset_hash(rela->sec, rela->offset);
 }
 
-struct elf *elf_read(const char *name, int flags);
-struct section *find_section_by_name(struct elf *elf, const char *name);
+struct elf *elf_open_read(const char *name, int flags);
+struct section *elf_create_section(struct elf *elf, const char *name, size_t entsize, int nr);
+struct section *elf_create_rela_section(struct elf *elf, struct section *base);
+void elf_add_rela(struct elf *elf, struct rela *rela);
+int elf_write(const struct elf *elf);
+void elf_close(struct elf *elf);
+
+struct section *find_section_by_name(const struct elf *elf, const char *name);
 struct symbol *find_func_by_offset(struct section *sec, unsigned long offset);
 struct symbol *find_symbol_by_offset(struct section *sec, unsigned long offset);
-struct symbol *find_symbol_by_name(struct elf *elf, const char *name);
-struct symbol *find_symbol_containing(struct section *sec, unsigned long offset);
-struct rela *find_rela_by_dest(struct elf *elf, struct section *sec, unsigned long offset);
-struct rela *find_rela_by_dest_range(struct elf *elf, struct section *sec,
+struct symbol *find_symbol_by_name(const struct elf *elf, const char *name);
+struct symbol *find_symbol_containing(const struct section *sec, unsigned long offset);
+struct rela *find_rela_by_dest(const struct elf *elf, struct section *sec, unsigned long offset);
+struct rela *find_rela_by_dest_range(const struct elf *elf, struct section *sec,
                                     unsigned long offset, unsigned int len);
 struct symbol *find_func_containing(struct section *sec, unsigned long offset);
-struct section *elf_create_section(struct elf *elf, const char *name, size_t
-                                  entsize, int nr);
-struct section *elf_create_rela_section(struct elf *elf, struct section *base);
 int elf_rebuild_rela_section(struct section *sec);
-int elf_write(struct elf *elf);
-void elf_close(struct elf *elf);
 
 #define for_each_sec(file, sec)                                                \
        list_for_each_entry(sec, &file->elf->sections, list)
index 0b3528f..58fdda5 100644 (file)
@@ -58,7 +58,9 @@ static void cmd_usage(void)
 
        printf("\n");
 
-       exit(129);
+       if (!help)
+               exit(129);
+       exit(0);
 }
 
 static void handle_options(int *argc, const char ***argv)
diff --git a/tools/objtool/objtool.h b/tools/objtool/objtool.h
new file mode 100644 (file)
index 0000000..528028a
--- /dev/null
@@ -0,0 +1,27 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Copyright (C) 2020 Matt Helsley <mhelsley@vmware.com>
+ */
+
+#ifndef _OBJTOOL_H
+#define _OBJTOOL_H
+
+#include <stdbool.h>
+#include <linux/list.h>
+#include <linux/hashtable.h>
+
+#include "elf.h"
+
+struct objtool_file {
+       struct elf *elf;
+       struct list_head insn_list;
+       DECLARE_HASHTABLE(insn_hash, 20);
+       bool ignore_unreachables, c_file, hints, rodata;
+};
+
+int check(const char *objname, bool orc);
+int orc_dump(const char *objname);
+int create_orc(struct objtool_file *file);
+int create_orc_sections(struct objtool_file *file);
+
+#endif /* _OBJTOOL_H */
diff --git a/tools/objtool/orc.h b/tools/objtool/orc.h
deleted file mode 100644 (file)
index ee28322..0000000
+++ /dev/null
@@ -1,18 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0-or-later */
-/*
- * Copyright (C) 2017 Josh Poimboeuf <jpoimboe@redhat.com>
- */
-
-#ifndef _ORC_H
-#define _ORC_H
-
-#include <asm/orc_types.h>
-
-struct objtool_file;
-
-int create_orc(struct objtool_file *file);
-int create_orc_sections(struct objtool_file *file);
-
-int orc_dump(const char *objname);
-
-#endif /* _ORC_H */
index ba4cbb1..fca46e0 100644 (file)
@@ -4,7 +4,8 @@
  */
 
 #include <unistd.h>
-#include "orc.h"
+#include <asm/orc_types.h>
+#include "objtool.h"
 #include "warn.h"
 
 static const char *reg_name(unsigned int reg)
index 4c0dabd..c954998 100644 (file)
@@ -6,7 +6,6 @@
 #include <stdlib.h>
 #include <string.h>
 
-#include "orc.h"
 #include "check.h"
 #include "warn.h"
 
@@ -16,10 +15,10 @@ int create_orc(struct objtool_file *file)
 
        for_each_insn(file, insn) {
                struct orc_entry *orc = &insn->orc;
-               struct cfi_reg *cfa = &insn->state.cfa;
-               struct cfi_reg *bp = &insn->state.regs[CFI_BP];
+               struct cfi_reg *cfa = &insn->cfi.cfa;
+               struct cfi_reg *bp = &insn->cfi.regs[CFI_BP];
 
-               orc->end = insn->state.end;
+               orc->end = insn->cfi.end;
 
                if (cfa->base == CFI_UNDEFINED) {
                        orc->sp_reg = ORC_REG_UNDEFINED;
@@ -75,7 +74,7 @@ int create_orc(struct objtool_file *file)
 
                orc->sp_offset = cfa->offset;
                orc->bp_offset = bp->offset;
-               orc->type = insn->state.type;
+               orc->type = insn->cfi.type;
        }
 
        return 0;
@@ -130,8 +129,7 @@ static int create_orc_entry(struct elf *elf, struct section *u_sec, struct secti
        rela->offset = idx * sizeof(int);
        rela->sec = ip_relasec;
 
-       list_add_tail(&rela->list, &ip_relasec->rela_list);
-       hash_add(elf->rela_hash, &rela->hash, rela_hash(rela));
+       elf_add_rela(elf, rela);
 
        return 0;
 }
diff --git a/tools/objtool/weak.c b/tools/objtool/weak.c
new file mode 100644 (file)
index 0000000..942ea5e
--- /dev/null
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (C) 2020 Matt Helsley <mhelsley@vmware.com>
+ * Weak definitions necessary to compile objtool without
+ * some subcommands (e.g. check, orc).
+ */
+
+#include <stdbool.h>
+#include <errno.h>
+#include "objtool.h"
+
+#define __weak __attribute__((weak))
+
+#define UNSUPPORTED(name)                                              \
+({                                                                     \
+       fprintf(stderr, "error: objtool: " name " not implemented\n");  \
+       return ENOSYS;                                                  \
+})
+
+const char __weak *objname;
+
+int __weak check(const char *_objname, bool orc)
+{
+       UNSUPPORTED("check subcommand");
+}
+
+int __weak orc_dump(const char *_objname)
+{
+       UNSUPPORTED("orc");
+}
+
+int __weak create_orc(struct objtool_file *file)
+{
+       UNSUPPORTED("orc");
+}
+
+int __weak create_orc_sections(struct objtool_file *file)
+{
+       UNSUPPORTED("orc");
+}