Possibility of integer overflow in JfrThreadSampler::run()

[Severin Gehwolf]

Hi,

On Wed, 2019-01-09 at 13:23 +0900, Yasumasa Suenaga wrote:
> Hi all,
> 
> I posted this issue in May 2018 [1].
> It still occurs in current jdk/jdk when it was build by GCC 8.
> 
> We can avoid it with -fno-strict-overflow GCC option. So I propose to add it
> to JvmOverrideFiles.gmk:
> 
>   http://cr.openjdk.java.net/~ysuenaga/jfr-integer-overflow/webrev.1/
> 
> JDK-8147466 has been fixed with similar change for jdk7u [2].
> 
> What do you think?

This fix is masking undefined behavior (signed integer overflow) in
code. We should fix the code instead. The previous fix looked better,
but would still overflow if:

max(_interval_java, 10) + last_java_ms > max_jlong

Note that the assertion triggers on a fastdebug build with GCC 8.2 and
running :jdk_jfr test group for me (18 tests fail). Failures go away
with -fno-strict-overflow.

> I will file it to JBS if it is accepted.

Please file a bug.

Thanks,
Severin

> 
> Thanks,
> 
> Yasumasa
> 
> [1] https://mail.openjdk.java.net/pipermail/hotspot-jfr-dev/2018-May/000047.html
> [2] http://hg.openjdk.java.net/jdk7u/jdk7u/jdk/rev/35dcc0db31dc