diff options
| author | Alex Light <allight@google.com> | 2019-05-06 18:16:24 +0000 |
|---|---|---|
| committer | Treehugger Robot <treehugger-gerrit@google.com> | 2019-05-09 09:07:24 +0000 |
| commit | 4060786d8fa8c0c63c751a837decce4f95a33112 (patch) | |
| tree | 7ce0fcebc6399c5c672569dcf7aa3dc67d503978 /test/ProfileTestMultiDex | |
| parent | 3ff45bf767f4f1baf858e2220e36acffa97b0383 (diff) | |
Revert^2 "Correctly handle thread deopt with thread-specific JVMTI events"
This reverts commit b2a8964f4218c2c52dacf599ebf5cf69f8753bf0.
It turns out that transitioning from instrumentation trampolines and
interpreter trampolines interacts in a racy way with single thread
deoptimization. This caused tests that perform this transition
repeatedly to be flaky when run with the --trace. Since it is expected
to be rare that one traces at the same time they are performing JVMTI
functions (since JVMTI is a superset of trace behaviors) we solved
this by simply not allowing the transition.
Reason for revert: Prevented unsafe transition between entry-exit
trampolines and interpreter only.
Bug: 131865028
Bug: 132283660
Test: ./test.py --host --trace --ntrace
Test: echo "#!/bin/bash" > run-one-test.sh
echo "./art/test/run-test --dev --jit --trace --64 1956-pop-frame-jit-calling 2>&1" >> run-one-test.sh
chmod u+x run-one-test.sh
./art/tools/parallel_run.py -j20 ./run-one-test.sh
Change-Id: Id496b272f353a5a5e000574c107a97d67405d54b
Diffstat (limited to 'test/ProfileTestMultiDex')
0 files changed, 0 insertions, 0 deletions