[sanitizer] Implement NanoTime & MonotonicNanoTime for Windows

[cryptoad]

Summary

Implement MonotonicNanoTime using QueryPerformanceCounter.

This function is used by Scudo & the 64-bit Primary allocator. Implementing it
now means that the release-to-OS mechanism of the Primary will kick in (it
never did since the function returned 0 always), but ReleaseMemoryPagesToOS is
still not currently implemented for Windows.

Performance wise, this adds a syscall & a 64-bit division per call to
MonotonicNanoTime so the impact might not be negligible, but I don't think
there is a way around it.

[cryptoad]

Actually, I guess we don't need to make it relative to a base, we can just use the current counter.
Which would also work as is for NanoTime.

[cryptoad]

Implement NanoTime using QueryPerformanceCounter. MonotonicNanoTime will
just call NanoTime.

[alekseyshl]

Should we turn return-to-os off by default on Windows then? So that we won't waste time walking free lists.

lib/sanitizer_common/sanitizer_win.cc

505–514

Why function static, why not at the top level?

512

ULL

513

I wonder what frequency.QuadPart values typically are.

[cryptoad]

U -> ULL

[cryptoad]

lib/sanitizer_common/sanitizer_win.cc

505–514

It's only used here, I didn't feel like it warranted being outside of the scope of the function.

513

On my test machine: 0x2fcd8 (ticks/s).

[cryptoad]

@rnk : how does this look to you?

[alekseyshl]

What about my top level comment? Should we turn return-to-os off by default on Windows to donot waste time walking free lists?

[cryptoad]

In D42579#995205, @alekseyshl wrote:

What about my top level comment? Should we turn return-to-os off by default on Windows to donot waste time walking free lists?

I didn't see it, my apologies.
Yeah I think it's worth turning it off by default for now.
And on Fuchsia as well (cc: @flowerhack) as they don't have a return-to-os function either yet.
Want me to do it in a separate one 1st?

[amccarth]

rnk's been out sick, but this looks right to me.

Even with the multiplication by 10^9, you'd have to run for more than a year before you would risk an overflow (assuming the 195,800 ticks/second frequency seen by cryptoad).

[amccarth]

LGTM. Wrong action selected in last update.

[alekseyshl]

In D42579#995214, @cryptoad wrote:

In D42579#995205, @alekseyshl wrote:

What about my top level comment? Should we turn return-to-os off by default on Windows to donot waste time walking free lists?

I didn't see it, my apologies.
Yeah I think it's worth turning it off by default for now.
And on Fuchsia as well (cc: @flowerhack) as they don't have a return-to-os function either yet.
Want me to do it in a separate one 1st?

Whatever works, this patch or separate one, I'm fine with either.

[cryptoad]

As per review comments, disable release-to-os by default on Windows (& Fuchsia).
Neither of those platforms implement ReleaseMemoryPagesToOS, but since we
are implementing MonotonicNanoTime, this would trigger spurious scanning of
the memory pages.

lib/sanitizer_common/sanitizer_flags.inc

COMMON_FLAG(uptr, soft_rss_limit_mb, 0,
             " If non-zero, a background thread is spawned at startup"
             " which periodically reads RSS. If the limit is reached"
             " all subsequent malloc/new calls will fail or return NULL"
             " (depending on the value of allocator_may_return_null)"
             " until the RSS goes below the soft limit."
             " This limit does not affect memory allocations other than"
             " malloc/new.")
 COMMON_FLAG(bool, heap_profile, false, "Experimental heap profiler, asan-only")
-COMMON_FLAG(s32, allocator_release_to_os_interval_ms, 5000,
+COMMON_FLAG(s32, allocator_release_to_os_interval_ms,
+            (SANITIZER_FUCHSIA || SANITIZER_WINDOWS) ? -1 : 5000,
             "Only affects a 64-bit allocator. If set, tries to release unused "
             "memory to the OS, but not more often than this interval (in "
             "milliseconds). Negative values mean do not attempt to release "
             "memory to the OS.\n")
 COMMON_FLAG(bool, can_use_proc_maps_statm, true,
             "If false, do not attempt to read /proc/maps/statm."
             " Mostly useful for testing sanitizers.")
 COMMON_FLAG(

lib/sanitizer_common/sanitizer_win.cc

void SleepForSeconds(int seconds) {
   Sleep(seconds * 1000);
 }
 
 void SleepForMillis(int millis) {
   Sleep(millis);
 }
 
 u64 NanoTime() {
-  return 0;
+  static LARGE_INTEGER frequency = {0};
+  LARGE_INTEGER counter;
+  if (UNLIKELY(frequency.QuadPart == 0)) {
+    QueryPerformanceFrequency(&frequency);
+    CHECK_NE(frequency.QuadPart, 0);
+  }
+  QueryPerformanceCounter(&counter);
+  counter.QuadPart *= 1000ULL * 1000000ULL;
+  counter.QuadPart /= frequency.QuadPart;
+  return counter.QuadPart;
 }
 
-u64 MonotonicNanoTime() {
-  return 0;
-}
+u64 MonotonicNanoTime() { return NanoTime(); }
 
 void Abort() {
   internal__exit(3);
 }
 
 #if !SANITIZER_GO
 // Read the file to extract the ImageBase field from the PE header. If ASLR is
 // disabled and this virtual address is available, the loader will typically