On Wed, 12 May 2021 14:53:39 GMT, Jorn Vernee <jver...@openjdk.org> wrote:
>> src/hotspot/cpu/x86/universalUpcallHandler_x86_64.cpp line 472:
>>
>>> 470: __ block_comment("} preserve_callee_saved_regs ");
>>> 471:
>>> 472: // TODO mxcsr
>>
>> Anything left to do with mxcsr?
>
> I guess this slipped through with the initial PR.
>
> JNI code loads the default value of mxcsr here and saves/restores the
> incoming native value:
> https://github.com/openjdk/jdk/blob/master/src/hotspot/cpu/x86/stubGenerator_x86_64.cpp#L265-L298
>
> However, it does nothing on Windows, while the Windows x64 ABI also defines
> mxcsr to have a non-volatile portion (i.e. that should be saved and
> restored):
> https://docs.microsoft.com/en-us/cpp/build/x64-calling-convention?view=msvc-160#mxcsr
>
> I think this made me a bit unsure about the need for saving and restoring
> mxcsr, and wanted to find an actual test case that causes a problem, rather
> than just copy-pasting the code (or at least understand it a bit better).
> But, at this time I haven't gotten to that yet.
Thanks for the pointers. It would be safer for now to align the behavior with
the call stub.
-------------
PR: https://git.openjdk.java.net/jdk/pull/3699
