atomicops_internals_x86_gcc.cc 3.31 KB
Newer Older
gejun's avatar
gejun committed
1 2 3 4 5 6 7 8 9 10
// Copyright (c) 2006-2008 The Chromium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

// This module gets enough CPU information to optimize the
// atomicops module on x86.

#include <stdint.h>
#include <string.h>

11
#include "butil/atomicops.h"
gejun's avatar
gejun committed
12 13 14 15 16

// This file only makes sense with atomicops_internals_x86_gcc.h -- it
// depends on structs that are defined in that file.  If atomicops.h
// doesn't sub-include that file, then we aren't needed, and shouldn't
// try to do anything.
17
#ifdef BUTIL_ATOMICOPS_INTERNALS_X86_GCC_H_
gejun's avatar
gejun committed
18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99

// Inline cpuid instruction.  In PIC compilations, %ebx contains the address
// of the global offset table.  To avoid breaking such executables, this code
// must preserve that register's value across cpuid instructions.
#if defined(__i386__)
#define cpuid(a, b, c, d, inp) \
  asm("mov %%ebx, %%edi\n"     \
      "cpuid\n"                \
      "xchg %%edi, %%ebx\n"    \
      : "=a" (a), "=D" (b), "=c" (c), "=d" (d) : "a" (inp))
#elif defined(__x86_64__)
#define cpuid(a, b, c, d, inp) \
  asm("mov %%rbx, %%rdi\n"     \
      "cpuid\n"                \
      "xchg %%rdi, %%rbx\n"    \
      : "=a" (a), "=D" (b), "=c" (c), "=d" (d) : "a" (inp))
#endif

#if defined(cpuid)        // initialize the struct only on x86

// Set the flags so that code will run correctly and conservatively, so even
// if we haven't been initialized yet, we're probably single threaded, and our
// default values should hopefully be pretty safe.
struct AtomicOps_x86CPUFeatureStruct AtomicOps_Internalx86CPUFeatures = {
  false,          // bug can't exist before process spawns multiple threads
};

namespace {

// Initialize the AtomicOps_Internalx86CPUFeatures struct.
void AtomicOps_Internalx86CPUFeaturesInit() {
  uint32_t eax;
  uint32_t ebx;
  uint32_t ecx;
  uint32_t edx;

  // Get vendor string (issue CPUID with eax = 0)
  cpuid(eax, ebx, ecx, edx, 0);
  char vendor[13];
  memcpy(vendor, &ebx, 4);
  memcpy(vendor + 4, &edx, 4);
  memcpy(vendor + 8, &ecx, 4);
  vendor[12] = 0;

  // get feature flags in ecx/edx, and family/model in eax
  cpuid(eax, ebx, ecx, edx, 1);

  int family = (eax >> 8) & 0xf;        // family and model fields
  int model = (eax >> 4) & 0xf;
  if (family == 0xf) {                  // use extended family and model fields
    family += (eax >> 20) & 0xff;
    model += ((eax >> 16) & 0xf) << 4;
  }

  // Opteron Rev E has a bug in which on very rare occasions a locked
  // instruction doesn't act as a read-acquire barrier if followed by a
  // non-locked read-modify-write instruction.  Rev F has this bug in
  // pre-release versions, but not in versions released to customers,
  // so we test only for Rev E, which is family 15, model 32..63 inclusive.
  if (strcmp(vendor, "AuthenticAMD") == 0 &&       // AMD
      family == 15 &&
      32 <= model && model <= 63) {
    AtomicOps_Internalx86CPUFeatures.has_amd_lock_mb_bug = true;
  } else {
    AtomicOps_Internalx86CPUFeatures.has_amd_lock_mb_bug = false;
  }
}

class AtomicOpsx86Initializer {
 public:
  AtomicOpsx86Initializer() {
    AtomicOps_Internalx86CPUFeaturesInit();
  }
};

// A global to get use initialized on startup via static initialization :/
AtomicOpsx86Initializer g_initer;

}  // namespace

#endif  // if x86

100
#endif  // ifdef BUTIL_ATOMICOPS_INTERNALS_X86_GCC_H_