[Bug 237208] java/openjdk11: port to powerpc64

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=237208

--- Comment #9 from Gustavo Romero <gustavo.romero at protonmail.com> ---
(In reply to Piotr Kubaj from comment #4)

Hi Piotr and Mikael,

Is the SIGILL JVM crash still happening?

Piotr, the illegal instruction executed in the arraycopy stubs looks to be a
'mtdscr', i.e. an instruction used to adjust the depth of the data prefetch
prior to the bulk copy.

However I can't make sense of that yet because availability of m{t,f}dscr is
proved at runtime before the emission by the compiler.

Logs show that m{t,f}dscr is not available indeed: 

CPU:total 16 (initial active 16) ppc64 fsqrt isel lxarxeh cmpb popcntb popcntw
fcfids vand lqarx aes vpmsumb mfdscr vsx ldbrx stdbrx sha darn

That JVM on POWER9 reports it has VSX and DARN support but not DSCR? The
instruction that caused the crash in question is also using DSCR SPR number 3,
which is not privileged either.

Does:

int main()
{
 asm("mtspr 3, 0;");
}


crashes with a SIGILL?

-- 
You are receiving this mail because:
You are on the CC list for the bug.