AuroraMiddleware/v8
1
0
Fork 0
Code Issues 2 Pull Requests Releases Wiki Activity

[heap] Fix off-thread merging tests

Browse Source 
Fix off-thread space merging unit tests, by

  a) Doing a GC on each test teardown (avoiding leaking GC state between
     tests), and
  b) Removing the AllocationStep when merging, which could cause a
     deadlock if incremental marking triggered a free-list refill.

Bug: v8:10142
Change-Id: Id2670115f96ab3b13613ea357bd44a639d0151e1
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2142260
Commit-Queue: Leszek Swirski <leszeks@chromium.org>
Auto-Submit: Leszek Swirski <leszeks@chromium.org>
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Cr-Commit-Position: refs/heads/master@{#67077}
This commit is contained in:
Leszek Swirski 2020-04-09 09:23:06 +02:00 committed by Commit Bot
parent 66cfc31581
commit db8491c38c
2 changed files with 37 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
src/heap/spaces.cc
View File

@ -1762,14 +1762,12 @@ void PagedSpace::MergeLocalSpace(LocalSpace* other) {
!p->IsFlagSet(Page::NEVER_ALLOCATE_ON_PAGE),
p->AvailableInFreeList() == p->AvailableInFreeListFromAllocatedBytes());
if (merging_from_off_thread) {
// TODO(leszeks): Allocation groups are currently not handled properly by
// the sampling allocation profiler. We'll have to come up with a better
// solution for allocation stepping before shipping.
AllocationStepAfterMerge(
p->area_start(),
static_cast<int>(p->HighWaterMark() - p->area_start()));
}
// TODO(leszeks): Here we should allocation step, but:
// 1. Allocation groups are currently not handled properly by the sampling
// allocation profiler, and
// 2. Observers might try to take the space lock, which isn't reentrant.
// We'll have to come up with a better solution for allocation stepping
// before shipping, which will likely be using LocalHeap.
}
if (merging_from_off_thread) {
@ -4401,7 +4399,9 @@ void OldLargeObjectSpace::MergeOffThreadSpace(
other->RemovePage(page, size);
AddPage(page, size);
AllocationStepAfterMerge(object.address(), size);
// TODO(leszeks): Here we should AllocationStep, see the TODO in
// PagedSpace::MergeOffThreadSpace.
if (heap()->incremental_marking()->black_allocation()) {
heap()->incremental_marking()->marking_state()->WhiteToBlack(object);
}

33
test/unittests/heap/spaces-unittest.cc
View File

@ -15,7 +15,15 @@
namespace v8 {
namespace internal {
using SpacesTest = TestWithIsolate;
class SpacesTest : public TestWithIsolate {
protected:
void TearDown() final {
// Clean up the heap between tests.
i_isolate()->heap()->CollectAllAvailableGarbage(
GarbageCollectionReason::kTesting);
TestWithIsolate::TearDown();
}
};
TEST_F(SpacesTest, CompactionSpaceMerge) {
Heap* heap = i_isolate()->heap();
@ -167,10 +175,15 @@ TEST_F(SpacesTest, OffThreadSpaceMergeDuringIncrementalMarking) {
expected_merged_pages += pages_in_off_thread_space;
}
heap->FinalizeIncrementalMarkingAtomically(GarbageCollectionReason::kTesting);
// Check the page size before finalizing marking, since the GC will see the
// empty pages and will evacuate them.
// TODO(leszeks): Maybe allocate real objects, and hold on to them with
// Handles, to make sure incremental marking finalization doesn't clear them
// away.
EXPECT_EQ(pages_in_old_space + expected_merged_pages,
old_space->CountTotalPages());
heap->FinalizeIncrementalMarkingAtomically(GarbageCollectionReason::kTesting);
}
class LargeOffThreadAllocationThread final : public base::Thread {
@ -266,7 +279,10 @@ TEST_F(SpacesTest, OffThreadLargeObjectSpaceMergeDuringIncrementalMarking) {
threads[i]->Join();
}
int pages_in_old_space = lo_space->PageCount();
heap->StartIncrementalMarking(Heap::kNoGCFlags,
GarbageCollectionReason::kTesting);
int pages_in_lo_space = lo_space->PageCount();
int expected_merged_pages = 0;
for (int i = 0; i < kNumThreads; ++i) {
@ -276,7 +292,14 @@ TEST_F(SpacesTest, OffThreadLargeObjectSpaceMergeDuringIncrementalMarking) {
expected_merged_pages += pages_in_off_thread_space;
}
EXPECT_EQ(pages_in_old_space + expected_merged_pages, lo_space->PageCount());
// Check the page size before finalizing marking, since the GC will see the
// empty pages and will evacuate them.
// TODO(leszeks): Maybe allocate real objects, and hold on to them with
// Handles, to make sure incremental marking finalization doesn't clear them
// away.
EXPECT_EQ(pages_in_lo_space + expected_merged_pages, lo_space->PageCount());
heap->FinalizeIncrementalMarkingAtomically(GarbageCollectionReason::kTesting);
}
TEST_F(SpacesTest, WriteBarrierFromHeapObject) {