RFR: Add function entry/exit instruments for aarch64

Man Cao manc at openjdk.java.net
Thu Apr 30 19:23:16 UTC 2020


On Thu, 30 Apr 2020 16:49:18 GMT, Arthur Eubanks <aeubanks at openjdk.org> wrote:

>> @aeubanks thanks for review.
>> 1, I list generate_native_entry in commit msg because I think the generate_native_entry is the real   entry in JVM we
>> want to instrument, which will call jvmti interface notify_method_entry/exit.
>> 2, I was just wondering which case will go generate_native_wrapper, thus, I removed the instruments in
>> generate_native_wrapper, but all cases still passed. as for RacyNativeLoopTest.java, I add a  breakpoint (int 3) before
>> tsan instruments in generate_native_wrapper, it couldn't stop the execution. but the breakpoint in
>> generate_native_entry could.
>
> Hmm interesting...
> If that's the case then going with your proposed changes is fine in a future commit. But I'll sponsor this to unblock
> for now.

For why generate_native_wrapper was not called, it looks like it is only called for compiled code and intrinsics.
-XX:+ThreadSanitizer disables all compilation, and the those native methods in AbstractNativeLoop are not intrinsics.

-------------

PR: https://git.openjdk.java.net/tsan/pull/7


More information about the tsan-dev mailing list