Reviewers: yurys, caseq, Benedikt Meurer,
Message:
Created Revert of Fix CPU profiler deadlock on Windows + AMD CPU.
Description:
Revert of Fix CPU profiler deadlock on Windows + AMD CPU. (patchset #1 id:1
of
https://codereview.chromium.org/1304873011/ )
Reason for revert:
Broke a test
Original issue's description:
Fix CPU profiler deadlock on Windows + AMD CPU.
Implement a lock free version of RolloverProtectedTickClock::Now
to eliminate a deadlock.
BUG=chromium:521420
LOG=Y
Committed: https://crrev.com/d6db8e5902af1f067f0f149844f4e92824ee93f3
Cr-Commit-Position: refs/heads/master@{#30489}
[email protected],[email protected],[email protected]
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=chromium:521420
Please review this at https://codereview.chromium.org/1303423006/
Base URL: https://chromium.googlesource.com/v8/v8.git@master
Affected files (+13, -13 lines):
M src/base/platform/time.cc
Index: src/base/platform/time.cc
diff --git a/src/base/platform/time.cc b/src/base/platform/time.cc
index
1e01d9ac04e056fb64aa184df83e8048e53e73af..7e4c8fc0ac902db962fa3b5a3337f1e876ebd561
100644
--- a/src/base/platform/time.cc
+++ b/src/base/platform/time.cc
@@ -17,7 +17,6 @@
#include <ostream>
#if V8_OS_WIN
-#include "src/base/atomicops.h"
#include "src/base/lazy-instance.h"
#include "src/base/win32-headers.h"
#endif
@@ -435,35 +434,36 @@
class RolloverProtectedTickClock final : public TickClock {
public:
- RolloverProtectedTickClock() : rollover_(0) {}
+ // We initialize rollover_ms_ to 1 to ensure that we will never
+ // return 0 from TimeTicks::HighResolutionNow() and TimeTicks::Now()
below.
+ RolloverProtectedTickClock() : last_seen_now_(0), rollover_ms_(1) {}
virtual ~RolloverProtectedTickClock() {}
int64_t Now() override {
+ LockGuard<Mutex> lock_guard(&mutex_);
// We use timeGetTime() to implement TimeTicks::Now(), which rolls over
// every ~49.7 days. We try to track rollover ourselves, which works if
- // TimeTicks::Now() is called at least every 24 days.
+ // TimeTicks::Now() is called at least every 49 days.
// Note that we do not use GetTickCount() here, since timeGetTime()
gives
// more predictable delta values, as described here:
//
http://blogs.msdn.com/b/larryosterman/archive/2009/09/02/what-s-the-difference-between-gettickcount-and-timegettime.aspx
// timeGetTime() provides 1ms granularity when combined with
// timeBeginPeriod(). If the host application for V8 wants fast
timers, it
// can use timeBeginPeriod() to increase the resolution.
- // We use a lock-free version because the sampler thread calls it
- // while having the rest of the world stopped, that could cause a
deadlock.
- base::Atomic32 rollover = base::Acquire_Load(&rollover_);
- base::Atomic32 now = static_cast<base::Atomic32>(timeGetTime());
- if ((now >> 31) != (rollover & 1)) {
- base::Release_CompareAndSwap(&rollover_, rollover, rollover + 1);
- ++rollover;
+ DWORD now = timeGetTime();
+ if (now < last_seen_now_) {
+ rollover_ms_ += V8_INT64_C(0x100000000); // ~49.7 days.
}
- int64_t rollover_ms = static_cast<int64_t>(rollover) << 31;
- return (rollover_ms | now) * Time::kMicrosecondsPerMillisecond;
+ last_seen_now_ = now;
+ return (now + rollover_ms_) * Time::kMicrosecondsPerMillisecond;
}
bool IsHighResolution() override { return false; }
private:
- base::Atomic32 rollover_;
+ Mutex mutex_;
+ DWORD last_seen_now_;
+ int64_t rollover_ms_;
};
--
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
---
You received this message because you are subscribed to the Google Groups "v8-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
