[aarch64-port-dev ] JDK 13 AArch64 issues to backport

[Andrew Dinn]

On 19/06/2019 17:54, Andrew Haley wrote:
> On 6/19/19 5:37 PM, Andrew Dinn wrote:
>> Am I supposed to backport them now or is anyone else going to step in?
> 
> First, please determine whether any of the first set are critical.

I don't think any of these backports are 'critical' in the sense of a
must-have. Several of them do cause VM crashes/produce invalid results
but that is mitigated by the fact that this only happens in rare
circumstances.

So, modulo one quibble (see below) I agree with your assessment of all
the bugs slated for backport to both jdk8u and jdk11u. They are not
critical. However, you omitted to consider the ones which were only for
one of the two repos.

The two issues that are to me most egregious are in the set listed only
for jdk11u:

>   8215792: AArch64: String.indexOf generates incorrect result
>   8218966: AArch64: String.compareTo() can read memory after string

These both produce invalid results with certain long Strings. I think it
is quite possible for these to turn up in real applications (indeed, one
was found during Graal testing of text processing). So, I think it is
important to fix them.

There are two patches to restore lost performance that are also well
worth considering. The first was for jdk8u only:

>   8216350: AArch64: monitor unlock fast path not called

The second is the quibble in the jdk8u/11u set:

>   8217368: AArch64: C2 recursive stack locking optimisation not triggered

You are right that this does not imply a correctness issue. However it
does harm locking performance which may matter a lot to some users. So,
I would still urge backporting these two performance fixes unless you
are really concerned to keep the backport set absolutely minimal.

regards,


Andrew Dinn
-----------
Senior Principal Software Engineer
Red Hat UK Ltd
Registered in England and Wales under Company Registration No. 03798903
Directors: Michael Cunningham, Michael ("Mike") O'Neill, Eric Shander