[BOLT] Instrumentation: AArch64 instrumentation support in runtime

[Elvina]

Summary

This patch adds support for AArch64 in instrumentation runtime library, including AArch64 system calls.
Also this commit divides syscalls into target-specific files.

Elvina Yakubova,
Advanced Software Technology Lab, Huawei

[Elvina]

clang-format

[rafauler]

bolt/CMakeLists.txt

41–42

We don't support runtime on windows, so we shouldn't delete these checks. Just change the x86 check to check for both x86 OR AArch64

bolt/runtime/instr.cpp

1661

I have the impression that the correct offset here is 256 instead of 288, but I can't test this. That's why I asked for a test for indirect calls in the other diff.

1702

same

1707

clang format is doing the wrong thing here putting SAVE_ALL in the same line as other assembly instructions.

I think we should use:

// clang-format off

and

// clang-format on

surrounding this code, and put SAVE_ALL in a separate line like before this diff.

[Elvina]

bolt/runtime/instr.cpp

1661

We have several stp/str instructions before that, when we're adding instrumentation snippets in createIndirectCallIst and createInstrumentedIndCallHandlerEntryBB functions, that's where 32 is added from

1707

Thanks for suggesting it, that's exactly what I needed :)

[Elvina]

Fixed Cmake, added clang-format plugs

[treapster]

bolt/runtime/CMakeLists.txt

32

Drop if, use -mgeneral-regs-only

[Elvina]

bolt/runtime/CMakeLists.txt

32

thanks for noticing, forgot about it

[rafauler]

I wonder why is this diff not showing up on top of the D151899 diff stack anymore? Isn't it dependent on that stack?

[Elvina]

In D151942#4474686, @rafauler wrote:

I wonder why is this diff not showing up on top of the D151899 diff stack anymore? Isn't it dependent on that stack?

Yes, it is, added it.

[Elvina]

Gentle ping

[Elvina]

add check for Darwin OS, so we'll not get error during building with clang on Linux (see here https://github.com/llvm/llvm-project/issues/63255#issuecomment-1589716433)

[rafauler]

bolt/runtime/CMakeLists.txt

30

Dropping -mno-sse is causing problems for me on my x86 machine.

test runtime/X86/instrumentation-xmm.c is failing

[Elvina]

bolt/runtime/CMakeLists.txt

30

Thanks for noticing!
Turned out general-regs-only doesn't disable sse on clang https://github.com/llvm/llvm-project/blob/main/clang/lib/Driver/ToolChains/Arch/X86.cpp#L245

[Elvina]

Setting -mno-sse flag back

[rafauler]

Thanks Elvina! LGTM

[Elvina]

rebase + add syscalls for AArch64

[yota9]

LGTM

[Elvina]

@rafauler Thanks for reviewing this!

[phosek]

This change broke our builder that's cross-compiling the toolchain from x86_64-linux-gnu to aarch64-linux-gnu. It looks like the bolt_rt build is now trying to execute the cross-compiled clang and fails because it's no possible to run the aarch64-linux-gnu binary on x86_64-linux-gnu.

bolt/CMakeLists.txt

35–38

This probably also needs to check that CMAKE_CROSSCOMPILING isn't set (that is NOT CMAKE_CROSSCOMPILING).

[rafauler]

bolt/CMakeLists.txt

35–38

Would this work? D158906

[Elvina]

@rafauler Thanks for fixing this. I noticed the issue just now

bolt/CMakeLists.txt

 foreach (tgt ${BOLT_TARGETS_TO_BUILD})
   if (NOT tgt IN_LIST LLVM_TARGETS_TO_BUILD)
     message(FATAL_ERROR "BOLT target '${tgt}' is not in LLVM_TARGETS_TO_BUILD")
   endif()
   message(STATUS "Targeting ${tgt} in llvm-bolt")
 endforeach()
 
 set(BOLT_ENABLE_RUNTIME_default OFF)
-if (CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64"
+if ((CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64"
+    OR CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64")
     AND (CMAKE_SYSTEM_NAME STREQUAL "Linux"
-      OR CMAKE_SYSTEM_NAME STREQUAL "Darwin")
-    AND "X86" IN_LIST BOLT_TARGETS_TO_BUILD)
+      OR CMAKE_SYSTEM_NAME STREQUAL "Darwin"))
   set(BOLT_ENABLE_RUNTIME_default ON)
 endif()
 option(BOLT_ENABLE_RUNTIME "Enable BOLT runtime" ${BOLT_ENABLE_RUNTIME_default})
 if (BOLT_ENABLE_RUNTIME)
   # Some systems prevent reading /proc/self/map_files
   execute_process(COMMAND ls /proc/self/map_files
     RESULT_VARIABLE LS OUTPUT_QUIET ERROR_QUIET)
   if (LS)
     set(BOLT_ENABLE_RUNTIME OFF)
     message(WARNING
       "BOLT runtime is disabled as /proc/self/map_files is unreadable.")
   endif()

bolt/runtime/CMakeLists.txt

add_library(bolt_rt_hugify STATIC
   )
 set_target_properties(bolt_rt_hugify PROPERTIES ARCHIVE_OUTPUT_DIRECTORY "${LLVM_LIBRARY_DIR}")
 
 set(BOLT_RT_FLAGS
   -ffreestanding
   -fno-exceptions
   -fno-rtti
   -fno-stack-protector
-  -mno-sse
-  -fPIC)
+  -fPIC
+  -mgeneral-regs-only)
+if (CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64")
+  set(BOLT_RT_FLAGS ${BOLT_RT_FLAGS} "-mno-sse")
+endif()
 
 # Don't let the compiler think it can create calls to standard libs
 target_compile_options(bolt_rt_instr PRIVATE ${BOLT_RT_FLAGS})
 target_include_directories(bolt_rt_instr PRIVATE ${CMAKE_CURRENT_BINARY_DIR})
 target_compile_options(bolt_rt_hugify PRIVATE ${BOLT_RT_FLAGS})
 target_include_directories(bolt_rt_hugify PRIVATE ${CMAKE_CURRENT_BINARY_DIR})
 
 install(TARGETS bolt_rt_instr DESTINATION "lib${LLVM_LIBDIR_SUFFIX}")
 install(TARGETS bolt_rt_hugify DESTINATION "lib${LLVM_LIBDIR_SUFFIX}")
 
-if (CMAKE_CXX_COMPILER_ID MATCHES ".*Clang.*")
+if (CMAKE_CXX_COMPILER_ID MATCHES ".*Clang.*" AND CMAKE_SYSTEM_NAME STREQUAL "Darwin")
   add_library(bolt_rt_instr_osx STATIC
     instr.cpp
     ${CMAKE_CURRENT_BINARY_DIR}/config.h
   )
   set_target_properties(bolt_rt_instr_osx PROPERTIES ARCHIVE_OUTPUT_DIRECTORY "${LLVM_LIBRARY_DIR}")
   target_include_directories(bolt_rt_instr_osx PRIVATE ${CMAKE_CURRENT_BINARY_DIR})
   target_compile_options(bolt_rt_instr_osx PRIVATE
     -target x86_64-apple-darwin19.6.0
     ${BOLT_RT_FLAGS})
   install(TARGETS bolt_rt_instr_osx DESTINATION "lib${LLVM_LIBDIR_SUFFIX}")
 endif()

bolt/runtime/common.h

 //===- bolt/runtime/common.h ------------------------------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
-#if !defined(__x86_64__)
-#error "For x86_64 only"
-#endif
-
 #if defined(__linux__)
 
 #include <cstddef>
 #include <cstdint>
 
 #include "config.h"
 
 #ifdef HAVE_ELF_H

 
 typedef long long int64_t;
 typedef int int32_t;
 
 #else
 #error "For Linux or MacOS only"
 #endif
 
-// Save all registers while keeping 16B stack alignment
-#define SAVE_ALL                                                               \
-  "push %%rax\n"                                                               \
-  "push %%rbx\n"                                                               \
-  "push %%rcx\n"                                                               \
-  "push %%rdx\n"                                                               \
-  "push %%rdi\n"                                                               \
-  "push %%rsi\n"                                                               \
-  "push %%rbp\n"                                                               \
-  "push %%r8\n"                                                                \
-  "push %%r9\n"                                                                \
-  "push %%r10\n"                                                               \
-  "push %%r11\n"                                                               \
-  "push %%r12\n"                                                               \
-  "push %%r13\n"                                                               \
-  "push %%r14\n"                                                               \
-  "push %%r15\n"                                                               \
-  "sub $8, %%rsp\n"
-
-// Mirrors SAVE_ALL
-#define RESTORE_ALL                                                            \
-  "add $8, %%rsp\n"                                                            \
-  "pop %%r15\n"                                                                \
-  "pop %%r14\n"                                                                \
-  "pop %%r13\n"                                                                \
-  "pop %%r12\n"                                                                \
-  "pop %%r11\n"                                                                \
-  "pop %%r10\n"                                                                \
-  "pop %%r9\n"                                                                 \
-  "pop %%r8\n"                                                                 \
-  "pop %%rbp\n"                                                                \
-  "pop %%rsi\n"                                                                \
-  "pop %%rdi\n"                                                                \
-  "pop %%rdx\n"                                                                \
-  "pop %%rcx\n"                                                                \
-  "pop %%rbx\n"                                                                \
-  "pop %%rax\n"
-
 #define PROT_READ 0x1  /* Page can be read.  */
 #define PROT_WRITE 0x2 /* Page can be written.  */
 #define PROT_EXEC 0x4  /* Page can be executed.  */
 #define PROT_NONE 0x0  /* Page can not be accessed.  */
 #define PROT_GROWSDOWN                                                         \
   0x01000000 /* Extend change to start of                                      \
                 growsdown vma (mprotect only).  */
 #define PROT_GROWSUP                                                           \
int memcmp(const void *s1, const void *s2, size_t n) {
   }
   return 0;
 }
 } // extern "C"
 
 // Anonymous namespace covering everything but our library entry point
 namespace {
 
-constexpr uint32_t BufSize = 10240;
-
-#define _STRINGIFY(x) #x
-#define STRINGIFY(x) _STRINGIFY(x)
-
-uint64_t __read(uint64_t fd, const void *buf, uint64_t count) {
-  uint64_t ret;
-#if defined(__APPLE__)
-#define READ_SYSCALL 0x2000003
-#else
-#define READ_SYSCALL 0
-#endif
-  __asm__ __volatile__("movq $" STRINGIFY(READ_SYSCALL) ", %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       : "D"(fd), "S"(buf), "d"(count)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-uint64_t __write(uint64_t fd, const void *buf, uint64_t count) {
-  uint64_t ret;
-#if defined(__APPLE__)
-#define WRITE_SYSCALL 0x2000004
-#else
-#define WRITE_SYSCALL 1
-#endif
-  __asm__ __volatile__("movq $" STRINGIFY(WRITE_SYSCALL) ", %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       : "D"(fd), "S"(buf), "d"(count)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-void *__mmap(uint64_t addr, uint64_t size, uint64_t prot, uint64_t flags,
-             uint64_t fd, uint64_t offset) {
-#if defined(__APPLE__)
-#define MMAP_SYSCALL 0x20000c5
-#else
-#define MMAP_SYSCALL 9
-#endif
-  void *ret;
-  register uint64_t r8 asm("r8") = fd;
-  register uint64_t r9 asm("r9") = offset;
-  register uint64_t r10 asm("r10") = flags;
-  __asm__ __volatile__("movq $" STRINGIFY(MMAP_SYSCALL) ", %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       : "D"(addr), "S"(size), "d"(prot), "r"(r10), "r"(r8),
-                         "r"(r9)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-uint64_t __munmap(void *addr, uint64_t size) {
-#if defined(__APPLE__)
-#define MUNMAP_SYSCALL 0x2000049
-#else
-#define MUNMAP_SYSCALL 11
-#endif
-  uint64_t ret;
-  __asm__ __volatile__("movq $" STRINGIFY(MUNMAP_SYSCALL) ", %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       : "D"(addr), "S"(size)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
+struct dirent64 {
+  uint64_t d_ino;          /* Inode number */
+  int64_t d_off;           /* Offset to next linux_dirent */
+  unsigned short d_reclen; /* Length of this linux_dirent */
+  unsigned char d_type;
+  char d_name[]; /* Filename (null-terminated) */
+                 /* length is actually (d_reclen - 2 -
+                   offsetof(struct linux_dirent, d_name)) */
+};
 
-#define SIG_BLOCK 0
-#define SIG_UNBLOCK 1
-#define SIG_SETMASK 2
+/* Length of the entries in `struct utsname' is 65.  */
+#define _UTSNAME_LENGTH 65
 
-static const uint64_t MaskAllSignals[] = {-1ULL};
+struct UtsNameTy {
+  char sysname[_UTSNAME_LENGTH];  /* Operating system name (e.g., "Linux") */
+  char nodename[_UTSNAME_LENGTH]; /* Name within "some implementation-defined
+                      network" */
+  char release[_UTSNAME_LENGTH]; /* Operating system release (e.g., "2.6.28") */
+  char version[_UTSNAME_LENGTH]; /* Operating system version */
+  char machine[_UTSNAME_LENGTH]; /* Hardware identifier */
+  char domainname[_UTSNAME_LENGTH]; /* NIS or YP domain name */
+};
 
-uint64_t __sigprocmask(int how, const void *set, void *oldset) {
-#if defined(__APPLE__)
-#define SIGPROCMASK_SYSCALL 0x2000030
-#else
-#define SIGPROCMASK_SYSCALL 14
-#endif
-  uint64_t ret;
-  register long r10 asm("r10") = sizeof(uint64_t);
-  __asm__ __volatile__("movq $" STRINGIFY(SIGPROCMASK_SYSCALL) ", %%rax\n"
-                                                               "syscall\n"
-                       : "=a"(ret)
-                       : "D"(how), "S"(set), "d"(oldset), "r"(r10)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
+struct timespec {
+  uint64_t tv_sec;  /* seconds */
+  uint64_t tv_nsec; /* nanoseconds */
+};
 
-uint64_t __getpid() {
-  uint64_t ret;
-#if defined(__APPLE__)
-#define GETPID_SYSCALL 20
+#if defined(__aarch64__)
+#include "sys_aarch64.h"
 #else
-#define GETPID_SYSCALL 39
+#include "sys_x86_64.h"
 #endif
-  __asm__ __volatile__("movq $" STRINGIFY(GETPID_SYSCALL) ", %%rax\n"
-                                                          "syscall\n"
-                       : "=a"(ret)
-                       :
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
 
-uint64_t __exit(uint64_t code) {
-#if defined(__APPLE__)
-#define EXIT_SYSCALL 0x2000001
-#else
-#define EXIT_SYSCALL 231
-#endif
-  uint64_t ret;
-  __asm__ __volatile__("movq $" STRINGIFY(EXIT_SYSCALL) ", %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       : "D"(code)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
+constexpr uint32_t BufSize = 10240;
 
 // Helper functions for writing strings to the .fdata file. We intentionally
 // avoid using libc names to make it clear it is our impl.
 
 /// Write number Num using Base to the buffer in OutBuf, returns a pointer to
 /// the end of the string.
 char *intToStr(char *OutBuf, uint64_t Num, uint32_t Base) {
   const char *Chars = "0123456789abcdef";
static bool scanUInt32(const char *&Buf, const char *End, uint32_t &Ret) {
   }
   if (OldBuf != Buf && Result <= 0xFFFFFFFFu) {
     Ret = static_cast<uint32_t>(Result);
     return true;
   }
   return false;
 }
 
-#if !defined(__APPLE__)
-// We use a stack-allocated buffer for string manipulation in many pieces of
-// this code, including the code that prints each line of the fdata file. This
-// buffer needs to accomodate large function names, but shouldn't be arbitrarily
-// large (dynamically allocated) for simplicity of our memory space usage.
-
-// Declare some syscall wrappers we use throughout this code to avoid linking
-// against system libc.
-uint64_t __open(const char *pathname, uint64_t flags, uint64_t mode) {
-  uint64_t ret;
-  __asm__ __volatile__("movq $2, %%rax\n"
-                       "syscall"
-                       : "=a"(ret)
-                       : "D"(pathname), "S"(flags), "d"(mode)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-struct dirent {
-  unsigned long d_ino;     /* Inode number */
-  unsigned long d_off;     /* Offset to next linux_dirent */
-  unsigned short d_reclen; /* Length of this linux_dirent */
-  char d_name[];           /* Filename (null-terminated) */
-                           /* length is actually (d_reclen - 2 -
-                             offsetof(struct linux_dirent, d_name)) */
-};
-
-long __getdents(unsigned int fd, dirent *dirp, size_t count) {
-  long ret;
-  __asm__ __volatile__("movq $78, %%rax\n"
-                       "syscall"
-                       : "=a"(ret)
-                       : "D"(fd), "S"(dirp), "d"(count)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-uint64_t __readlink(const char *pathname, char *buf, size_t bufsize) {
-  uint64_t ret;
-  __asm__ __volatile__("movq $89, %%rax\n"
-                       "syscall"
-                       : "=a"(ret)
-                       : "D"(pathname), "S"(buf), "d"(bufsize)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-uint64_t __lseek(uint64_t fd, uint64_t pos, uint64_t whence) {
-  uint64_t ret;
-  __asm__ __volatile__("movq $8, %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       : "D"(fd), "S"(pos), "d"(whence)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-int __close(uint64_t fd) {
-  uint64_t ret;
-  __asm__ __volatile__("movq $3, %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       : "D"(fd)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-int __madvise(void *addr, size_t length, int advice) {
-  int ret;
-  __asm__ __volatile__("movq $28, %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       : "D"(addr), "S"(length), "d"(advice)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-#define _UTSNAME_LENGTH 65
-
-struct UtsNameTy {
-  char sysname[_UTSNAME_LENGTH];  /* Operating system name (e.g., "Linux") */
-  char nodename[_UTSNAME_LENGTH]; /* Name within "some implementation-defined
-                      network" */
-  char release[_UTSNAME_LENGTH]; /* Operating system release (e.g., "2.6.28") */
-  char version[_UTSNAME_LENGTH]; /* Operating system version */
-  char machine[_UTSNAME_LENGTH]; /* Hardware identifier */
-  char domainname[_UTSNAME_LENGTH]; /* NIS or YP domain name */
-};
-
-int __uname(struct UtsNameTy *Buf) {
-  int Ret;
-  __asm__ __volatile__("movq $63, %%rax\n"
-                       "syscall\n"
-                       : "=a"(Ret)
-                       : "D"(Buf)
-                       : "cc", "rcx", "r11", "memory");
-  return Ret;
-}
-
-struct timespec {
-  uint64_t tv_sec;  /* seconds */
-  uint64_t tv_nsec; /* nanoseconds */
-};
-
-uint64_t __nanosleep(const timespec *req, timespec *rem) {
-  uint64_t ret;
-  __asm__ __volatile__("movq $35, %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       : "D"(req), "S"(rem)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-int64_t __fork() {
-  uint64_t ret;
-  __asm__ __volatile__("movq $57, %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       :
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-int __mprotect(void *addr, size_t len, int prot) {
-  int ret;
-  __asm__ __volatile__("movq $10, %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       : "D"(addr), "S"(len), "d"(prot)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-uint64_t __getppid() {
-  uint64_t ret;
-  __asm__ __volatile__("movq $110, %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       :
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-int __setpgid(uint64_t pid, uint64_t pgid) {
-  int ret;
-  __asm__ __volatile__("movq $109, %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       : "D"(pid), "S"(pgid)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-uint64_t __getpgid(uint64_t pid) {
-  uint64_t ret;
-  __asm__ __volatile__("movq $121, %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       : "D"(pid)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-int __kill(uint64_t pid, int sig) {
-  int ret;
-  __asm__ __volatile__("movq $62, %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       : "D"(pid), "S"(sig)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-int __fsync(int fd) {
-  int ret;
-  __asm__ __volatile__("movq $74, %%rax\n"
-                       "syscall\n"
-                       : "=a"(ret)
-                       : "D"(fd)
-                       : "cc", "rcx", "r11", "memory");
-  return ret;
-}
-
-//              %rdi      %rsi         %rdx        %r10         %r8
-// sys_prctl  int option  unsigned    unsigned    unsigned    unsigned
-//                        long arg2   long arg3   long arg4   long arg5
-int __prctl(int Option, unsigned long Arg2, unsigned long Arg3,
-            unsigned long Arg4, unsigned long Arg5) {
-  int Ret;
-  register long rdx asm("rdx") = Arg3;
-  register long r8 asm("r8") = Arg5;
-  register long r10 asm("r10") = Arg4;
-  __asm__ __volatile__("movq $157, %%rax\n"
-                       "syscall\n"
-                       : "=a"(Ret)
-                       : "D"(Option), "S"(Arg2), "d"(rdx), "r"(r10), "r"(r8)
-                       :);
-  return Ret;
-}
-
-#endif
-
 void reportError(const char *Msg, uint64_t Size) {
   __write(2, Msg, Size);
   __exit(1);
 }
 
 void assert(bool Assertion, const char *Msg) {
   if (Assertion)
     return;
   char Buf[BufSize];
   char *Ptr = Buf;
   Ptr = strCopy(Ptr, "Assertion failed: ");
   Ptr = strCopy(Ptr, Msg, BufSize - 40);
   Ptr = strCopy(Ptr, "\n");
   reportError(Buf, Ptr - Buf);
 }
 
+#define SIG_BLOCK 0
+#define SIG_UNBLOCK 1
+#define SIG_SETMASK 2
+
+static const uint64_t MaskAllSignals[] = {-1ULL};
+
 class Mutex {
   volatile bool InUse{false};
 
 public:
   bool acquire() { return !__atomic_test_and_set(&InUse, __ATOMIC_ACQUIRE); }
   void release() { __atomic_clear(&InUse, __ATOMIC_RELEASE); }
 };
 

bolt/runtime/instr.cpp

 // destructors for global objects. That's because those need support from the
 // linker in initialization/finalization code, and we want to keep our linker
 // very simple. Similarly, we don't create any global objects that are zero
 // initialized, since those would need to go .bss, which our simple linker also
 // don't support (TODO?).
 //
 //===----------------------------------------------------------------------===//
 
-#if defined (__x86_64__)
 #include "common.h"
 
 // Enables a very verbose logging to stderr useful when debugging
 //#define ENABLE_DEBUG
 
 #ifdef ENABLE_DEBUG
 #define DEBUG(X)                                                               \
   { X; }
if (TargetPath[0] != '\0')
     return TargetPath;
 
   unsigned long CurAddr = (unsigned long)__get_pc();
   uint64_t FDdir = __open(DirPath, O_RDONLY,
                           /*mode=*/0666);
   assert(static_cast<int64_t>(FDdir) >= 0,
          "failed to open /proc/self/map_files");
 
-  while (long Nread = __getdents(FDdir, (struct dirent *)Buf, BufSize)) {
+  while (long Nread = __getdents64(FDdir, (struct dirent64 *)Buf, BufSize)) {
     assert(static_cast<int64_t>(Nread) != -1, "failed to get folder entries");
 
-    struct dirent *d;
+    struct dirent64 *d;
     for (long Bpos = 0; Bpos < Nread; Bpos += d->d_reclen) {
-      d = (struct dirent *)(Buf + Bpos);
+      d = (struct dirent64 *)(Buf + Bpos);
 
       uint64_t StartAddress, EndAddress;
       if (!parseAddressRange(d->d_name, StartAddress, EndAddress))
         continue;
       if (CurAddr < StartAddress || CurAddr > EndAddress)
         continue;
       char FindBuf[NameMax];
       char *C = strCopy(FindBuf, DirPath, NameMax);

 instrumentIndirectCall(uint64_t Target, uint64_t IndCallID) {
   GlobalIndCallCounters[IndCallID].incrementVal(Target, *GlobalAlloc);
 }
 
 /// We receive as in-stack arguments the identifier of the indirect call site
 /// as well as the target address for the call
 extern "C" __attribute((naked)) void __bolt_instr_indirect_call()
 {
+#if defined(__aarch64__)
+  // clang-format off
+  __asm__ __volatile__(SAVE_ALL
+                       "ldp x0, x1, [sp, #288]\n"
+                       "bl instrumentIndirectCall\n"
+                       RESTORE_ALL
+                       "ret\n"
+                       :::);
+  // clang-format on
+#else
+  // clang-format off
   __asm__ __volatile__(SAVE_ALL
                        "mov 0xa0(%%rsp), %%rdi\n"
                        "mov 0x98(%%rsp), %%rsi\n"
                        "call instrumentIndirectCall\n"
                        RESTORE_ALL
                        "ret\n"
                        :::);
+  // clang-format on
+#endif
 }
 
 extern "C" __attribute((naked)) void __bolt_instr_indirect_tailcall()
 {
+#if defined(__aarch64__)
+  // clang-format off
+  __asm__ __volatile__(SAVE_ALL
+                       "ldp x0, x1, [sp, #288]\n"
+                       "bl instrumentIndirectCall\n"
+                       RESTORE_ALL
+                       "ret\n"
+                       :::);
+  // clang-format on
+#else
+  // clang-format off
   __asm__ __volatile__(SAVE_ALL
                        "mov 0x98(%%rsp), %%rdi\n"
                        "mov 0x90(%%rsp), %%rsi\n"
                        "call instrumentIndirectCall\n"
                        RESTORE_ALL
                        "ret\n"
                        :::);
+  // clang-format on
+#endif
 }
 
 /// This is hooking ELF's entry, it needs to save all machine state.
 extern "C" __attribute((naked)) void __bolt_instr_start()
 {
+#if defined(__aarch64__)
+  // clang-format off
+  __asm__ __volatile__(SAVE_ALL
+                       "bl __bolt_instr_setup\n"
+                       RESTORE_ALL
+                       "adrp x16, __bolt_start_trampoline\n"
+                       "add x16, x16, #:lo12:__bolt_start_trampoline\n"
+                       "br x16\n"
+                       :::);
+  // clang-format on
+#else
+  // clang-format off
   __asm__ __volatile__(SAVE_ALL
                        "call __bolt_instr_setup\n"
                        RESTORE_ALL
                        "jmp __bolt_start_trampoline\n"
                        :::);
+  // clang-format on
+#endif
 }
 
 /// This is hooking into ELF's DT_FINI
 extern "C" void __bolt_instr_fini() {
-  __bolt_fini_trampoline();
+#if defined(__aarch64__)
+  // clang-format off
+  __asm__ __volatile__(SAVE_ALL
+                       "adrp x16, __bolt_fini_trampoline\n"
+                       "add x16, x16, #:lo12:__bolt_fini_trampoline\n"
+                       "blr x16\n"
+                       RESTORE_ALL
+                       :::);
+  // clang-format on
+#else
+  __asm__ __volatile__("call __bolt_fini_trampoline\n" :::);
+#endif
   if (__bolt_instr_sleep_time == 0)
     __bolt_instr_data_dump();
   DEBUG(report("Finished.\n"));
 }
 
 #endif
 
 #if defined(__APPLE__)

 __attribute__((section("__TEXT,__fini")))
 __attribute__((force_align_arg_pointer))
 void _bolt_instr_fini() {
   report("Bye!\n");
   __bolt_instr_data_dump();
 }
 
 #endif
-#endif

bolt/runtime/sys_aarch64.h

+#ifndef LLVM_TOOLS_LLVM_BOLT_SYS_AARCH64
+#define LLVM_TOOLS_LLVM_BOLT_SYS_AARCH64
+
+// Save all registers while keeping 16B stack alignment
+#define SAVE_ALL                                                               \
+  "stp x0, x1, [sp, #-16]!\n"                                                  \
+  "stp x2, x3, [sp, #-16]!\n"                                                  \
+  "stp x4, x5, [sp, #-16]!\n"                                                  \
+  "stp x6, x7, [sp, #-16]!\n"                                                  \
+  "stp x8, x9, [sp, #-16]!\n"                                                  \
+  "stp x10, x11, [sp, #-16]!\n"                                                \
+  "stp x12, x13, [sp, #-16]!\n"                                                \
+  "stp x14, x15, [sp, #-16]!\n"                                                \
+  "stp x16, x17, [sp, #-16]!\n"                                                \
+  "stp x18, x19, [sp, #-16]!\n"                                                \
+  "stp x20, x21, [sp, #-16]!\n"                                                \
+  "stp x22, x23, [sp, #-16]!\n"                                                \
+  "stp x24, x25, [sp, #-16]!\n"                                                \
+  "stp x26, x27, [sp, #-16]!\n"                                                \
+  "stp x28, x29, [sp, #-16]!\n"                                                \
+  "str x30, [sp,#-16]!\n"
+// Mirrors SAVE_ALL
+#define RESTORE_ALL                                                            \
+  "ldr x30, [sp], #16\n"                                                       \
+  "ldp x28, x29, [sp], #16\n"                                                  \
+  "ldp x26, x27, [sp], #16\n"                                                  \
+  "ldp x24, x25, [sp], #16\n"                                                  \
+  "ldp x22, x23, [sp], #16\n"                                                  \
+  "ldp x20, x21, [sp], #16\n"                                                  \
+  "ldp x18, x19, [sp], #16\n"                                                  \
+  "ldp x16, x17, [sp], #16\n"                                                  \
+  "ldp x14, x15, [sp], #16\n"                                                  \
+  "ldp x12, x13, [sp], #16\n"                                                  \
+  "ldp x10, x11, [sp], #16\n"                                                  \
+  "ldp x8, x9, [sp], #16\n"                                                    \
+  "ldp x6, x7, [sp], #16\n"                                                    \
+  "ldp x4, x5, [sp], #16\n"                                                    \
+  "ldp x2, x3, [sp], #16\n"                                                    \
+  "ldp x0, x1, [sp], #16\n"
+
+// Anonymous namespace covering everything but our library entry point
+namespace {
+
+uint64_t __read(uint64_t fd, const void *buf, uint64_t count) {
+  uint64_t ret;
+  register uint64_t x0 __asm__("x0") = fd;
+  register const void *x1 __asm__("x1") = buf;
+  register uint64_t x2 __asm__("x2") = count;
+  register uint32_t w8 __asm__("w8") = 63;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(x2), "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+uint64_t __write(uint64_t fd, const void *buf, uint64_t count) {
+  uint64_t ret;
+  register uint64_t x0 __asm__("x0") = fd;
+  register const void *x1 __asm__("x1") = buf;
+  register uint64_t x2 __asm__("x2") = count;
+  register uint32_t w8 __asm__("w8") = 64;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(x2), "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+void *__mmap(uint64_t addr, uint64_t size, uint64_t prot, uint64_t flags,
+             uint64_t fd, uint64_t offset) {
+  void *ret;
+  register uint64_t x0 __asm__("x0") = addr;
+  register uint64_t x1 __asm__("x1") = size;
+  register uint64_t x2 __asm__("x2") = prot;
+  register uint64_t x3 __asm__("x3") = flags;
+  register uint64_t x4 __asm__("x4") = fd;
+  register uint64_t x5 __asm__("x5") = offset;
+  register uint32_t w8 __asm__("w8") = 222;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(x2), "r"(x3), "r"(x4), "r"(x5), "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+uint64_t __munmap(void *addr, uint64_t size) {
+  uint64_t ret;
+  register void *x0 __asm__("x0") = addr;
+  register uint64_t x1 __asm__("x1") = size;
+  register uint32_t w8 __asm__("w8") = 215;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+uint64_t __exit(uint64_t code) {
+  uint64_t ret;
+  register uint64_t x0 __asm__("x0") = code;
+  register uint32_t w8 __asm__("w8") = 94;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret), "+r"(x0)
+                       : "r"(w8)
+                       : "cc", "memory", "x1");
+  return ret;
+}
+
+uint64_t __open(const char *pathname, uint64_t flags, uint64_t mode) {
+  uint64_t ret;
+  register int x0 __asm__("x0") = -100;
+  register const char *x1 __asm__("x1") = pathname;
+  register uint64_t x2 __asm__("x2") = flags;
+  register uint64_t x3 __asm__("x3") = mode;
+  register uint32_t w8 __asm__("w8") = 56;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(x2), "r"(x3), "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+long __getdents64(unsigned int fd, dirent64 *dirp, size_t count) {
+  long ret;
+  register unsigned int x0 __asm__("x0") = fd;
+  register dirent64 *x1 __asm__("x1") = dirp;
+  register size_t x2 __asm__("x2") = count;
+  register uint32_t w8 __asm__("w8") = 61;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(x2), "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+uint64_t __readlink(const char *pathname, char *buf, size_t bufsize) {
+  uint64_t ret;
+  register int x0 __asm__("x0") = -100;
+  register const char *x1 __asm__("x1") = pathname;
+  register char *x2 __asm__("x2") = buf;
+  register size_t x3 __asm__("x3") = bufsize;
+  register uint32_t w8 __asm__("w8") = 78; // readlinkat
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(x2), "r"(x3), "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+uint64_t __lseek(uint64_t fd, uint64_t pos, uint64_t whence) {
+  uint64_t ret;
+  register uint64_t x0 __asm__("x0") = fd;
+  register uint64_t x1 __asm__("x1") = pos;
+  register uint64_t x2 __asm__("x2") = whence;
+  register uint32_t w8 __asm__("w8") = 62;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(x2), "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+int __close(uint64_t fd) {
+  int ret;
+  register uint64_t x0 __asm__("x0") = fd;
+  register uint32_t w8 __asm__("w8") = 57;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %w0, w0"
+                       : "=r"(ret), "+r"(x0)
+                       : "r"(w8)
+                       : "cc", "memory", "x1");
+  return ret;
+}
+
+int __madvise(void *addr, size_t length, int advice) {
+  int ret;
+  register void *x0 __asm__("x0") = addr;
+  register size_t x1 __asm__("x1") = length;
+  register int x2 __asm__("x2") = advice;
+  register uint32_t w8 __asm__("w8") = 233;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %w0, w0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(x2), "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+int __uname(struct UtsNameTy *buf) {
+  int ret;
+  register UtsNameTy *x0 __asm__("x0") = buf;
+  register uint32_t w8 __asm__("w8") = 160;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %w0, w0"
+                       : "=r"(ret), "+r"(x0)
+                       : "r"(w8)
+                       : "cc", "memory", "x1");
+  return ret;
+}
+
+uint64_t __nanosleep(const timespec *req, timespec *rem) {
+  uint64_t ret;
+  register const timespec *x0 __asm__("x0") = req;
+  register timespec *x1 __asm__("x1") = rem;
+  register uint32_t w8 __asm__("w8") = 101;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+int64_t __fork() {
+  uint64_t ret;
+  // clone instead of fork with flags
+  // "CLONE_CHILD_CLEARTID|CLONE_CHILD_SETTID|SIGCHLD"
+  register uint64_t x0 __asm__("x0") = 0x1200011;
+  register uint64_t x1 __asm__("x1") = 0;
+  register uint64_t x2 __asm__("x2") = 0;
+  register uint64_t x3 __asm__("x3") = 0;
+  register uint64_t x4 __asm__("x4") = 0;
+  register uint32_t w8 __asm__("w8") = 220;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(x2), "r"(x3), "r"(x4), "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+int __mprotect(void *addr, size_t len, int prot) {
+  int ret;
+  register void *x0 __asm__("x0") = addr;
+  register size_t x1 __asm__("x1") = len;
+  register int x2 __asm__("x2") = prot;
+  register uint32_t w8 __asm__("w8") = 226;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %w0, w0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(x2), "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+uint64_t __getpid() {
+  uint64_t ret;
+  register uint32_t w8 __asm__("w8") = 172;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret)
+                       : "r"(w8)
+                       : "cc", "memory", "x0", "x1");
+  return ret;
+}
+
+uint64_t __getppid() {
+  uint64_t ret;
+  register uint32_t w8 __asm__("w8") = 173;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret)
+                       : "r"(w8)
+                       : "cc", "memory", "x0", "x1");
+  return ret;
+}
+
+int __setpgid(uint64_t pid, uint64_t pgid) {
+  int ret;
+  register uint64_t x0 __asm__("x0") = pid;
+  register uint64_t x1 __asm__("x1") = pgid;
+  register uint32_t w8 __asm__("w8") = 154;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %w0, w0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+uint64_t __getpgid(uint64_t pid) {
+  uint64_t ret;
+  register uint64_t x0 __asm__("x0") = pid;
+  register uint32_t w8 __asm__("w8") = 155;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret), "+r"(x0)
+                       : "r"(w8)
+                       : "cc", "memory", "x1");
+  return ret;
+}
+
+int __kill(uint64_t pid, int sig) {
+  int ret;
+  register uint64_t x0 __asm__("x0") = pid;
+  register int x1 __asm__("x1") = sig;
+  register uint32_t w8 __asm__("w8") = 129;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %w0, w0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+int __fsync(int fd) {
+  int ret;
+  register int x0 __asm__("x0") = fd;
+  register uint32_t w8 __asm__("w8") = 82;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %w0, w0"
+                       : "=r"(ret), "+r"(x0)
+                       : "r"(w8)
+                       : "cc", "memory", "x1");
+  return ret;
+}
+
+uint64_t __sigprocmask(int how, const void *set, void *oldset) {
+  uint64_t ret;
+  register int x0 __asm__("x0") = how;
+  register const void *x1 __asm__("x1") = set;
+  register void *x2 __asm__("x2") = oldset;
+  register long x3 asm("x3") = 8;
+  register uint32_t w8 __asm__("w8") = 135;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %0, x0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(x2), "r"(x3), "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+int __prctl(int option, unsigned long arg2, unsigned long arg3,
+            unsigned long arg4, unsigned long arg5) {
+  int ret;
+  register int x0 __asm__("x0") = option;
+  register unsigned long x1 __asm__("x1") = arg2;
+  register unsigned long x2 __asm__("x2") = arg3;
+  register unsigned long x3 __asm__("x3") = arg4;
+  register unsigned long x4 __asm__("x4") = arg5;
+  register uint32_t w8 __asm__("w8") = 167;
+  __asm__ __volatile__("svc #0\n"
+                       "mov %w0, w0"
+                       : "=r"(ret), "+r"(x0), "+r"(x1)
+                       : "r"(x2), "r"(x3), "r"(x4), "r"(w8)
+                       : "cc", "memory");
+  return ret;
+}
+
+} // anonymous namespace
+
+#endif

bolt/runtime/sys_x86_64.h

+#ifndef LLVM_TOOLS_LLVM_BOLT_SYS_X86_64
+#define LLVM_TOOLS_LLVM_BOLT_SYS_X86_64
+
+// Save all registers while keeping 16B stack alignment
+#define SAVE_ALL                                                               \
+  "push %%rax\n"                                                               \
+  "push %%rbx\n"                                                               \
+  "push %%rcx\n"                                                               \
+  "push %%rdx\n"                                                               \
+  "push %%rdi\n"                                                               \
+  "push %%rsi\n"                                                               \
+  "push %%rbp\n"                                                               \
+  "push %%r8\n"                                                                \
+  "push %%r9\n"                                                                \
+  "push %%r10\n"                                                               \
+  "push %%r11\n"                                                               \
+  "push %%r12\n"                                                               \
+  "push %%r13\n"                                                               \
+  "push %%r14\n"                                                               \
+  "push %%r15\n"                                                               \
+  "sub $8, %%rsp\n"
+// Mirrors SAVE_ALL
+#define RESTORE_ALL                                                            \
+  "add $8, %%rsp\n"                                                            \
+  "pop %%r15\n"                                                                \
+  "pop %%r14\n"                                                                \
+  "pop %%r13\n"                                                                \
+  "pop %%r12\n"                                                                \
+  "pop %%r11\n"                                                                \
+  "pop %%r10\n"                                                                \
+  "pop %%r9\n"                                                                 \
+  "pop %%r8\n"                                                                 \
+  "pop %%rbp\n"                                                                \
+  "pop %%rsi\n"                                                                \
+  "pop %%rdi\n"                                                                \
+  "pop %%rdx\n"                                                                \
+  "pop %%rcx\n"                                                                \
+  "pop %%rbx\n"                                                                \
+  "pop %%rax\n"
+
+namespace {
+
+#define _STRINGIFY(x) #x
+#define STRINGIFY(x) _STRINGIFY(x)
+
+uint64_t __read(uint64_t fd, const void *buf, uint64_t count) {
+  uint64_t ret;
+#if defined(__APPLE__)
+#define READ_SYSCALL 0x2000003
+#else
+#define READ_SYSCALL 0
+#endif
+  __asm__ __volatile__("movq $" STRINGIFY(READ_SYSCALL) ", %%rax\n"
+                                                        "syscall\n"
+                       : "=a"(ret)
+                       : "D"(fd), "S"(buf), "d"(count)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+uint64_t __write(uint64_t fd, const void *buf, uint64_t count) {
+  uint64_t ret;
+#if defined(__APPLE__)
+#define WRITE_SYSCALL 0x2000004
+#else
+#define WRITE_SYSCALL 1
+#endif
+  __asm__ __volatile__("movq $" STRINGIFY(WRITE_SYSCALL) ", %%rax\n"
+                                                         "syscall\n"
+                       : "=a"(ret)
+                       : "D"(fd), "S"(buf), "d"(count)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+void *__mmap(uint64_t addr, uint64_t size, uint64_t prot, uint64_t flags,
+             uint64_t fd, uint64_t offset) {
+#if defined(__APPLE__)
+#define MMAP_SYSCALL 0x20000c5
+#else
+#define MMAP_SYSCALL 9
+#endif
+  void *ret;
+  register uint64_t r8 asm("r8") = fd;
+  register uint64_t r9 asm("r9") = offset;
+  register uint64_t r10 asm("r10") = flags;
+  __asm__ __volatile__("movq $" STRINGIFY(MMAP_SYSCALL) ", %%rax\n"
+                                                        "syscall\n"
+                       : "=a"(ret)
+                       : "D"(addr), "S"(size), "d"(prot), "r"(r10), "r"(r8),
+                         "r"(r9)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+uint64_t __munmap(void *addr, uint64_t size) {
+#if defined(__APPLE__)
+#define MUNMAP_SYSCALL 0x2000049
+#else
+#define MUNMAP_SYSCALL 11
+#endif
+  uint64_t ret;
+  __asm__ __volatile__("movq $" STRINGIFY(MUNMAP_SYSCALL) ", %%rax\n"
+                                                          "syscall\n"
+                       : "=a"(ret)
+                       : "D"(addr), "S"(size)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+uint64_t __sigprocmask(int how, const void *set, void *oldset) {
+#if defined(__APPLE__)
+#define SIGPROCMASK_SYSCALL 0x2000030
+#else
+#define SIGPROCMASK_SYSCALL 14
+#endif
+  uint64_t ret;
+  register long r10 asm("r10") = sizeof(uint64_t);
+  __asm__ __volatile__("movq $" STRINGIFY(SIGPROCMASK_SYSCALL) ", %%rax\n"
+                                                               "syscall\n"
+                       : "=a"(ret)
+                       : "D"(how), "S"(set), "d"(oldset), "r"(r10)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+uint64_t __getpid() {
+  uint64_t ret;
+#if defined(__APPLE__)
+#define GETPID_SYSCALL 20
+#else
+#define GETPID_SYSCALL 39
+#endif
+  __asm__ __volatile__("movq $" STRINGIFY(GETPID_SYSCALL) ", %%rax\n"
+                                                          "syscall\n"
+                       : "=a"(ret)
+                       :
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+uint64_t __exit(uint64_t code) {
+#if defined(__APPLE__)
+#define EXIT_SYSCALL 0x2000001
+#else
+#define EXIT_SYSCALL 231
+#endif
+  uint64_t ret;
+  __asm__ __volatile__("movq $" STRINGIFY(EXIT_SYSCALL) ", %%rax\n"
+                                                        "syscall\n"
+                       : "=a"(ret)
+                       : "D"(code)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+#if !defined(__APPLE__)
+// We use a stack-allocated buffer for string manipulation in many pieces of
+// this code, including the code that prints each line of the fdata file. This
+// buffer needs to accomodate large function names, but shouldn't be arbitrarily
+// large (dynamically allocated) for simplicity of our memory space usage.
+
+// Declare some syscall wrappers we use throughout this code to avoid linking
+// against system libc.
+uint64_t __open(const char *pathname, uint64_t flags, uint64_t mode) {
+  uint64_t ret;
+  __asm__ __volatile__("movq $2, %%rax\n"
+                       "syscall"
+                       : "=a"(ret)
+                       : "D"(pathname), "S"(flags), "d"(mode)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+long __getdents64(unsigned int fd, dirent64 *dirp, size_t count) {
+  long ret;
+  __asm__ __volatile__("movq $217, %%rax\n"
+                       "syscall"
+                       : "=a"(ret)
+                       : "D"(fd), "S"(dirp), "d"(count)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+uint64_t __readlink(const char *pathname, char *buf, size_t bufsize) {
+  uint64_t ret;
+  __asm__ __volatile__("movq $89, %%rax\n"
+                       "syscall"
+                       : "=a"(ret)
+                       : "D"(pathname), "S"(buf), "d"(bufsize)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+uint64_t __lseek(uint64_t fd, uint64_t pos, uint64_t whence) {
+  uint64_t ret;
+  __asm__ __volatile__("movq $8, %%rax\n"
+                       "syscall\n"
+                       : "=a"(ret)
+                       : "D"(fd), "S"(pos), "d"(whence)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+int __close(uint64_t fd) {
+  uint64_t ret;
+  __asm__ __volatile__("movq $3, %%rax\n"
+                       "syscall\n"
+                       : "=a"(ret)
+                       : "D"(fd)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+int __madvise(void *addr, size_t length, int advice) {
+  int ret;
+  __asm__ __volatile__("movq $28, %%rax\n"
+                       "syscall\n"
+                       : "=a"(ret)
+                       : "D"(addr), "S"(length), "d"(advice)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+int __uname(struct UtsNameTy *Buf) {
+  int Ret;
+  __asm__ __volatile__("movq $63, %%rax\n"
+                       "syscall\n"
+                       : "=a"(Ret)
+                       : "D"(Buf)
+                       : "cc", "rcx", "r11", "memory");
+  return Ret;
+}
+
+uint64_t __nanosleep(const timespec *req, timespec *rem) {
+  uint64_t ret;
+  __asm__ __volatile__("movq $35, %%rax\n"
+                       "syscall\n"
+                       : "=a"(ret)
+                       : "D"(req), "S"(rem)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+int64_t __fork() {
+  uint64_t ret;
+  __asm__ __volatile__("movq $57, %%rax\n"
+                       "syscall\n"
+                       : "=a"(ret)
+                       :
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+int __mprotect(void *addr, size_t len, int prot) {
+  int ret;
+  __asm__ __volatile__("movq $10, %%rax\n"
+                       "syscall\n"
+                       : "=a"(ret)
+                       : "D"(addr), "S"(len), "d"(prot)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+uint64_t __getppid() {
+  uint64_t ret;
+  __asm__ __volatile__("movq $110, %%rax\n"
+                       "syscall\n"
+                       : "=a"(ret)
+                       :
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+int __setpgid(uint64_t pid, uint64_t pgid) {
+  int ret;
+  __asm__ __volatile__("movq $109, %%rax\n"
+                       "syscall\n"
+                       : "=a"(ret)
+                       : "D"(pid), "S"(pgid)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+uint64_t __getpgid(uint64_t pid) {
+  uint64_t ret;
+  __asm__ __volatile__("movq $121, %%rax\n"
+                       "syscall\n"
+                       : "=a"(ret)
+                       : "D"(pid)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+int __kill(uint64_t pid, int sig) {
+  int ret;
+  __asm__ __volatile__("movq $62, %%rax\n"
+                       "syscall\n"
+                       : "=a"(ret)
+                       : "D"(pid), "S"(sig)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+int __fsync(int fd) {
+  int ret;
+  __asm__ __volatile__("movq $74, %%rax\n"
+                       "syscall\n"
+                       : "=a"(ret)
+                       : "D"(fd)
+                       : "cc", "rcx", "r11", "memory");
+  return ret;
+}
+
+//              %rdi      %rsi         %rdx        %r10         %r8
+// sys_prctl  int option  unsigned    unsigned    unsigned    unsigned
+//                        long arg2   long arg3   long arg4   long arg5
+int __prctl(int Option, unsigned long Arg2, unsigned long Arg3,
+            unsigned long Arg4, unsigned long Arg5) {
+  int Ret;
+  register long rdx asm("rdx") = Arg3;
+  register long r8 asm("r8") = Arg5;
+  register long r10 asm("r10") = Arg4;
+  __asm__ __volatile__("movq $157, %%rax\n"
+                       "syscall\n"
+                       : "=a"(Ret)
+                       : "D"(Option), "S"(Arg2), "d"(rdx), "r"(r10), "r"(r8)
+                       :);
+  return Ret;
+}
+
+#endif
+
+} // anonymous namespace
+
+#endif