Get rid of AIX linker option: -bnorwexec #63
Conversation
|
The IBM copyright end date needs to be updated. |
High-level: this linker option is oxymoronic with JIT environment (where executable code are generated on the fly anyway). At least for the time being before we figured out why default codeCache is allocated through malloc (instead of mmap or shmget, for which that linker option appeared to have no effect by my tests). Even after we figured it out eventually, it is a legitimate question what value this option provides for JIT environment. this also has effect on FFI thunks (which are not generated in JIT codecache). Issue ibmruntimes/openj9-openjdk-jdk21#162
|
|
| @@ -86,7 +86,7 @@ AC_DEFUN([FLAGS_SETUP_LDFLAGS_HELPER], | |||
| LDFLAGS_CXX_PARTIAL_LINKING="$MACHINE_FLAG -r" | |||
|
|
|||
| if test "x$OPENJDK_TARGET_OS" = xaix; then | |||
| BASIC_LDFLAGS="-Wl,-b64 -Wl,-brtl -Wl,-bnorwexec -Wl,-bnolibpath -Wl,-bnoexpall \ | |||
| BASIC_LDFLAGS="-Wl,-b64 -Wl,-brtl -Wl,-bnolibpath -Wl,-bnoexpall \ | |||
There was a problem hiding this comment.
We don't use clang on AIX, so this should make any difference (yet?).
There was a problem hiding this comment.
yes, i think we switched to xlC16's xlclang already. for java23, OpenXL(LLVM based) is expected.
There was a problem hiding this comment.
No, we're still using the xlc toolchain; see the most recent acceptance build: https://openj9-jenkins.osuosl.org/job/Build_JDKnext_ppc64_aix_OpenJDK/672.
[2024-05-29T04:26:54.412Z] configure: Using default toolchain xlc (IBM XL C/C++)
There was a problem hiding this comment.
see later:
BUILD_ENV_VARS_LIST = [PATH+XLC=/opt/IBM/xlC/16.1.0/bin:/opt/IBM/xlc/16.1.0/bin, CC=xlclang, CXX=xlclang++]
That is clang too. plus, the head guard is ==xaix, and it is an linker option (linker is the same between xlC and OpenXL). It has the same effect.
There was a problem hiding this comment.
The presence of "clang" in the names of the tools isn't relevant, the toolchain name, "xlc", is the reason the changed code is unused (see line 82):
elif test "x$TOOLCHAIN_TYPE" = xclang; then
High-level: this linker option is oxymoronic with JIT environment (where executable code are generated on the fly anyway).
At least for the time being before we figured out why default codeCache is allocated through malloc (instead of mmap or shmget, for which that linker option appeared to have no effect by my tests).
Even after we figured it out eventually, it is a legitimate question what value this option provides for JIT environment.
this also has effect on FFI thunks (which are not generated in JIT codecache).
Issue ibmruntimes/openj9-openjdk-jdk21#162