多くのスレッドに次のようなトレースがあることに気付いたとき、Chrome のスレッド スタックを見ていました。
0, wow64cpu.dll!TurboDispatchJumpAddressEnd+0x6c0
1, wow64cpu.dll!TurboDispatchJumpAddressEnd+0x4a8
2, wow64.dll!Wow64SystemServiceEx+0x1ce
3, wow64.dll!Wow64LdrpInitialize+0x429
4, ntdll.dll!RtlIsDosDeviceName_U+0x24c87
5, ntdll.dll!LdrInitializeThunk+0xe
6, ntdll.dll!ZwWaitForSingleObject+0x15
7, kernel32.dll!WaitForSingleObjectEx+0x43
8, kernel32.dll!WaitForSingleObject+0x12
9, chrome.dll!ovly_debug_event+0x16574
10, chrome.dll!ovly_debug_event+0x14904
11, chrome.dll!ovly_debug_event+0x14826
12, chrome.dll!ovly_debug_event+0x16d19
13, chrome.dll!ovly_debug_event+0x1bea1b
14, chrome.dll!ovly_debug_event+0xe8ff4
15, chrome.dll!ovly_debug_event+0x16b50
16, chrome.dll!ovly_debug_event+0x16ab2
17, kernel32.dll!BaseThreadInitThunk+0x12
18, ntdll.dll!RtlInitializeExceptionChain+0x63
19, ntdll.dll!RtlInitializeExceptionChain+0x36
クロムソースには、ほとんど空の関数としてsel_ldr.c
宣言されているように見える次のコードがあります。ovly_debug_event
void _ovly_debug_event (void) {
#ifdef __GNUC__
/*
* The asm volatile is here as instructed by the GCC docs.
* It's not enough to declare a function noinline.
* GCC will still look inside the function to see if it's worth calling.
*/
__asm__ volatile ("");
#elif NACL_WINDOWS
/*
* Visual Studio inlines empty functions even with noinline attribute,
* so we need a compile memory barrier to make this function not to be
* inlined. Also, it guarantees that nacl_global_xlate_base initialization
* is not reordered. This is important for gdb since it sets breakpoint on
* this function and reads nacl_global_xlate_base value.
*/
_ReadWriteBarrier();
#endif
}
static void StopForDebuggerInit (uintptr_t mem_start) {
/* Put xlate_base in a place where gdb can find it. */
nacl_global_xlate_base = mem_start;
NaClSandboxMemoryStartForValgrind(mem_start);
_ovly_debug_event();
}
これは疑問を投げかけます: なぜ chrome は、デバッグ専用で chrome ではほとんど空の関数に多くの時間を費やしているように見えるのでしょうか?