RFR: 8251152: ARM32: jtreg c2 Test8202414 test crash

Filipp Zhinkin filipp.zhinkin at gmail.com
Wed Sep 2 08:06:59 UTC 2020


Hi,

updated webrev:
http://cr.openjdk.java.net/~bulasevich/fzhinkin/8251152/webrev.1/

Tests are throwing SkippedException now, as Igor suggested.
Seems like I should also update a year in
JdkInternalMiscUnsafeUnalignedAccess' copyright, but I'm not sure if there
should be a line designating Oracle as intellectual property owner along
with SAP. Should I add it too?

Thanks,
Filipp.

On Tue, 1 Sep 2020 at 22:48, Filipp Zhinkin <filipp.zhinkin at gmail.com>
wrote:

> Hi Igor,
>
> On Tue, 1 Sep 2020 at 19:46, Igor Ignatyev <igor.ignatyev at oracle.com>
> wrote:
>
>> Hi Filipp,
>>
>> 1st of all, welcome back!
>>
>
> thanks!
>
>
>>
>> it would be better to throw jtreg.SkippedException at L#46 so jtreg will
>> reported the test as skipped (as opposed to just passed).
>
> Thanks, I'll update Test8202414 as well
> as compiler/unsafe/JdkInternalMiscUnsafeUnalignedAccess (which also skips
> execution the same way).
>
>
>> alternative, you could use '@requires vm.simpleArch != "arm"' to exclude
>> the test from arm32 execution.
>>
>
> I was thinking about adding something like vm.unalignedAccess.enabled, but
> it seems to be too complicated solution for two tests (Test8202414
> and compiler/unsafe/JdkInternalMiscUnsafeUnalignedAccess).
> I don't want to use 'simpleArch', because if some new platform missing
> unaligned access support will be added in the future then someone will have
> to spend time to find out why a test crashes.
>
> Thanks,
> Filipp.
>
>
>>
>> Thanks,
>> -- Igor
>>
>>
>> > On Sep 1, 2020, at 8:29 AM, Filipp Zhinkin <filipp.zhinkin at gmail.com>
>> wrote:
>> >
>> > Hi,
>> >
>> > Test8202414 crashes on ARM32 while writing to memory using an unaligned
>> > address.
>> > ARM32 supports unaligned memory accesses for some load/store
>> instructions
>> > under certain conditions, but LDRD (which is used when we're calling
>> > Unsafe::putLong) is always causing alignment fault when called with an
>> > unaligned address [1].
>> >
>> > The fix is simply skipping the test execution if a platform does not
>> > support unaligned memory accesses.
>> >
>> > Bug: https://bugs.openjdk.java.net/browse/JDK-8251152
>> > Webrev:
>> http://cr.openjdk.java.net/~bulasevich/fzhinkin/8251152/webrev.0/
>> >
>> > [1] ARM Architecture Reference Manual ARMv7-A and ARMv7-R edition,
>> §A3.2.1
>> > Unaligned data access
>> https://developer.arm.com/documentation/ddi0406/cd
>> >
>> > Thanks,
>> > Filipp.
>>
>>


More information about the hotspot-compiler-dev mailing list