RFR (XS) 8238896: Massive x86_32 crashes after JDK-7175279 (Don't use x87 FPU on x86-64)
Aleksey Shipilev
shade at redhat.com
Wed Feb 12 21:25:22 UTC 2020
Bug:
https://bugs.openjdk.java.net/browse/JDK-8238896
Fix:
https://cr.openjdk.java.net/~shade/8238896/webrev.01/
Two issues:
a) (obvious) missing "break" after the move [1];
b) (non-obvious) semantics-breaking move [2] that runs afoul of is_single_fpu and is_single_xmm
inclusiveness; see the warning comment about this in c1_LIR.hpp [3], "So you must always check for
is_???_xmm prior to is_???_fpu to distinguish between fpu- and xmm-registers"
This touches only x86_32 code, so I would not bother with jdk-submit.
Testing: Linux x86_64 fastdebug build; Linux x86_32 fastdebug tier1 (clean), tier2 (running)
--
Thanks,
-Aleksey
[1] https://hg.openjdk.java.net/jdk/jdk/rev/db30f31b9a8e#l10.251
[2] https://hg.openjdk.java.net/jdk/jdk/rev/db30f31b9a8e#l10.77
[3] https://hg.openjdk.java.net/jdk/jdk/file/7ef41e83066b/src/hotspot/share/c1/c1_LIR.hpp#l371
More information about the hotspot-compiler-dev
mailing list