[9] RFR (S): 8155635: C2: Mixed unsafe oop accesses break alias analysis

[Vladimir Kozlov]

Good.

Thanks,
Vladimir

On 8/24/16 11:47 AM, Vladimir Ivanov wrote:
> http://cr.openjdk.java.net/~vlivanov/8155635/webrev.01/
> https://bugs.openjdk.java.net/browse/JDK-8155635
>
> SplitIf transformation can produce untyped pointers when slitting AddP
> nodes for unsafe accesses through a Phi which merges non-null & null values:
>      AddP ... (Phi (ConP #NULL) (CheckCastPP Oop:...:NotNull))
>
> LibraryCallKit::inline_unsafe_access() produces such code shape for mixed oop accesses.
>
> The fix removes special case for mixed unsafe oop accesses in inline_unsafe_access(), so they are treated as if they are
> always on-heap. Off-heap oop accesses are dangerous and most likely crash in GC barriers anyway.
>
> It's a repeated attempt to fix the bug (previous thread [1]).
>
> Initially, I tried to convert between on-heap & raw pointers, but based on Vladimir K feedback decided to do a
> conservative fix. The only known problematic case is Unsafe.putObject.
>
> Testing: failing test, JPRT, RBT (hs-tier0-comp).
>
> Thanks!
>
> Best regards,
> Vladimir Ivanov
>
> [1] http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/2016-April/022743.html