Created attachment 199048 [details] Enable the BSD crtbegin/end by default I would like to enable the new crtbegin.o and crtend.o by default in -current. Can I get an exp-run with the attached base patch to make sure I'm not breaking building of any ports? It should apply to base r340213 or later.
New failures on amd64: + {"origin"=>"audio/dpf-plugins-lv2", "phase"=>"build", "errortype"=>"coredump"} + {"origin"=>"audio/fluidplug-lv2", "phase"=>"build", "errortype"=>"coredump"} + {"origin"=>"audio/vamp-plugin-sdk", "phase"=>"build", "errortype"=>"coredump"} + {"origin"=>"audio/zam-plugins-lv2", "phase"=>"build", "errortype"=>"coredump"} + {"origin"=>"audio/zynaddsubfx", "phase"=>"build", "errortype"=>"???"} + {"origin"=>"devel/libsoup", "phase"=>"configure", "errortype"=>"configure_error"} + {"origin"=>"devel/pear-PHP_UML", "flavor"=>"php56", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"devel/pear-PHP_UML", "flavor"=>"php70", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"devel/pear-PHP_UML", "flavor"=>"php71", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"devel/pear-PHP_UML", "flavor"=>"php73", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"devel/pear-PHP_UML", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"devel/pear-TheSeer_fXSL", "flavor"=>"php56", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"devel/pear-TheSeer_fXSL", "flavor"=>"php70", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"devel/pear-TheSeer_fXSL", "flavor"=>"php71", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"devel/pear-TheSeer_fXSL", "flavor"=>"php73", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"devel/pear-TheSeer_fXSL", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"emulators/higan", "phase"=>"extract", "errortype"=>"coredump"} + {"origin"=>"games/ezquake", "phase"=>"extract", "errortype"=>"process_failed"} + {"origin"=>"games/spring", "phase"=>"build", "errortype"=>"coredump"} + {"origin"=>"graphics/librsvg2", "phase"=>"build", "errortype"=>"???"} + {"origin"=>"graphics/tesseract-data", "phase"=>"fetch", "errortype"=>"fetch"} + {"origin"=>"korean/nanumfonts-otf", "phase"=>"extract", "errortype"=>"coredump"} + {"origin"=>"korean/nanumfonts-ttf", "phase"=>"extract", "errortype"=>"coredump"} + {"origin"=>"lang/afnix", "phase"=>"build", "errortype"=>"coredump"} + {"origin"=>"misc/qt5-doc", "phase"=>"extract", "errortype"=>"coredump"} + {"origin"=>"multimedia/vlc", "phase"=>"build", "errortype"=>"coredump"} + {"origin"=>"multimedia/vlc-qt4", "phase"=>"build", "errortype"=>"coredump"} + {"origin"=>"multimedia/vlc3", "phase"=>"build", "errortype"=>"coredump"} + {"origin"=>"net-mgmt/icingaweb2", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"security/libhijack", "phase"=>"build", "errortype"=>"clang"} + {"origin"=>"textproc/py-enchant", "flavor"=>"py36", "phase"=>"configure", "errortype"=>"process_failed"} + {"origin"=>"www/pear-XML_GRDDL", "flavor"=>"php56", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"www/pear-XML_GRDDL", "flavor"=>"php70", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"www/pear-XML_GRDDL", "flavor"=>"php71", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"www/pear-XML_GRDDL", "flavor"=>"php73", "phase"=>"stage", "errortype"=>"process_failed"} + {"origin"=>"www/pear-XML_GRDDL", "phase"=>"stage", "errortype"=>"process_failed"} Around 2000 ports skipped due to those new failures Failure logs on amd64: http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/dpf-plugins-lv2-1.1.1_3.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/FluidPlug-lv2-g20161020_2.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/vamp-plugin-sdk-2.6.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/zam-plugins-lv2-3.10.13_2.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/zynaddsubfx-3.0.3.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/libsoup-2.62.2.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php56-pear-PHP_UML-1.6.2.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php70-pear-PHP_UML-1.6.2.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php71-pear-PHP_UML-1.6.2.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php73-pear-PHP_UML-1.6.2.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php72-pear-PHP_UML-1.6.2.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php56-pear-TheSeer_fXSL-1.1.0.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php70-pear-TheSeer_fXSL-1.1.0.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php71-pear-TheSeer_fXSL-1.1.0.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php73-pear-TheSeer_fXSL-1.1.0.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php72-pear-TheSeer_fXSL-1.1.0.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/higan-106_2.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/ezquake-1.9_9,2.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/spring-98.0_14.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/librsvg2-2.40.20.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/tesseract-data-4.0.0.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/ko-nanumfonts-otf-20160201.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/ko-nanumfonts-ttf-20160201.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/afnix-2.6.3_3.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/qt5-doc-5.11.2.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/vlc-2.2.8_14,4.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/vlc-qt4-2.2.8_14,4.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/vlc3-3.0.4_1,4.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/icingaweb2-2.6.1.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/libhijack-0.8.0.1.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/py36-enchant-2.0.0.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php56-pear-XML_GRDDL-0.2.0.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php70-pear-XML_GRDDL-0.2.0.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php71-pear-XML_GRDDL-0.2.0.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php73-pear-XML_GRDDL-0.2.0.log http://package18.nyi.freebsd.org/data/headamd64PR233056-default/2018-11-09_21h50m43s/logs/errors/php72-pear-XML_GRDDL-0.2.0.log
The libhijack failure is presumably due to r340107 not the crtbegin patch. /wrkdirs/usr/ports/security/libhijack/work/libhijack-0.8.0-1-g50804aa/libhijack/rtld.c:264:6: error: no member named 'textsize' in 'struct Struct_Obj_Entry' soe.textsize = round_page(aux->loadables->phdr.phdr->p_vaddr + ~~~ ^ Some of the failures I looked at (at least spring, ezquake) were segfaults extracting a .7z file so the p7zip port might be a good place to start investigating. That said I did try building p7zip on my desktop with BSD_CRTBEGIN enabled and it did not fail in a trivial test.
It looks like there's an issue with __cxa_finalize. The devel/pear-TheSeer_fXSL port fails with the following from php: (gdb) bt #0 0x000000080182a9f0 in ?? () #1 0x0000000800ac0b05 in __cxa_finalize (dso=0x0) at /tank/users/andrew/freebsd/lib/libc/stdlib/atexit.c:244 #2 0x0000000800a506e1 in __cxa_thread_atexit_hidden (dtor_func=0x1, obj=0x1, dso_symbol=0x0) at /tank/users/andrew/freebsd/lib/libc/stdlib/cxa_thread_atexit_impl.c:90 #3 __cxa_thread_atexit_impl (dtor_func=0x1, obj=0x1, dso_symbol=0x0) at /tank/users/andrew/freebsd/lib/libc/stdlib/cxa_thread_atexit_impl.c:81 #4 0x00000000005a24bb in main ()
A commit references this bug: Author: andrew Date: Tue Nov 13 15:28:27 UTC 2018 New revision: 340395 URL: https://svnweb.freebsd.org/changeset/base/340395 Log: Run __cxa_finalize in shared objects in the destructor path. When we have .dtors call them before .dtor handling, otherwise call from a destructor. PR: 233056 MFC with: r339738 Sponsored by: DARPA, AFRL Changes: head/lib/csu/common/crtbegin.c
New failure on i386: + {"origin"=>"lang/erlang-runtime21", "phase"=>"build", "errortype"=>"coredump"} Failure log: http://package18.nyi.freebsd.org/data/headi386PR233056-default/2018-11-19_22h45m32s/logs/errors/erlang-runtime21-21.1.1.log
(In reply to Antoine Brodin from comment #5) Is the i386 failure from before or after r340395?
(In reply to Ed Maste from comment #6) It's from r340408
(In reply to Antoine Brodin from comment #7) Thanks Antoine. The i386 failure seems like it might be a transient issue: | set -e; LLVM_PROFILE_FILE="obj/i386-portbld-freebsd13.0/opt/smp/erlc-%m.profraw" \ | ERL_FLAGS="-emu_type prof +S 1" erlc -W -DPGO \ | -o obj/i386-portbld-freebsd13.0/opt/smp test/estone_SUITE.erl > obj/i386-portbld-freebsd13.0/opt/smp/PROFILE_LOG | dlsym: Resource temporarily unavailable So presumably the issue is solved by andrew's __cxa_finalize change, with one port needing further investigation/confirmation. Would you like to run another amd64 exp-run before we enable this in HEAD?
(In reply to Ed Maste from comment #8) r340408 is higher than r340395
(In reply to Antoine Brodin from comment #9) > r340408 is higher than r340395 Yes, so the amd64 results in this PR are from before andrew's fix and the i386 results are after the fix.
On amd64, there were 0 new failure at r340409
I'm getting the same error with lang/erlang-runtime21 when I use the gnu crtbegin/end code.
The BSD crtbegin/end has been enabled by default so close.
Re-open, there is a regression with lang/erlang-runtime21 on head-i386. I can't reproduce the failure with gnu crtbegin / crtend.
(In reply to Antoine Brodin from comment #14) I do not know if tier 2 architectures count as required for coverage, but if I understand right the BSD crtbegin/crtend option was disabled for at least powerpc64 because it turned out to be incomplete and did not work without involving some of the the old gcc-related code to some extent. This might be another reason to not close this bugzilla entry.
A commit references this bug: Author: andrew Date: Mon Sep 16 13:41:30 UTC 2019 New revision: 352394 URL: https://svnweb.freebsd.org/changeset/base/352394 Log: MFC r339738, r339744, r339770, r339773, r339864-r339866, r339907-r339908, r339912-r339913, r339916, r339954, r340213, r340395, r340840, r340910-r340911, r341424, r342773, r342974, r351027, r352073: MFC the BSD crtbegin to stable/12 but keep it disabled. r339738: Implement a BSD licensed crtbegin/crtend These are needed for .ctors/.dtors and .jcr handling. The former needs all the function pointers to be called in the correct order from the .init/.fini section. The latter just needs to call a gcj specific function if it exists with a pointer to the start of the .jcr section. This is currently disabled until __dso_handle support is added. Reviewed by: emaste Sponsored by: DARPA, AFRL Differential Revision: https://reviews.freebsd.org/D17587 r339744: Add a missing include for src.opts.mk. Without it MK_TESTS isn't defined. MFC with: r339738 Sponsored by: DARPA, AFRL r339770: Drop the csu tests WARNS to 5 to fix the powerpc64 build. MFC with: r339738 Sponsored by: DARPA, AFRL r339773: Add __dso_handle to the BSD crtbegin. This is used to identify shared objects. MFC with: r339738 Sponsored by: DARPA, AFRL r339864: Check __dso_handle is NULL in non-DSO objects. It should only be non-NULL when accessed from a shared object. MFC with: r339738 Sponsored by: DARPA, AFRL r339865: Include the csu test directories in BSD.tests.dist MFC with: r339738 Sponsored by: DARPA, AFRL r339866: Make the .ctors, .dtors, and .jcr markers as static. They shouldn't be accessible from out of the files they are defined in. MFC with: r339738 Sponsored by: DARPA, AFRL r339907: The jcr argument to _Jv_RegisterClasses is used, stop marking it otherwise. MFC with: r339738 Sponsored by: DARPA, AFRL r339908: Run the csu tests on a DSO. This builds the tests into a shared library, then runs these from the base test programs. With this we can check crtbeginS.o and crtendS.o are working as expected. MFC with: r339738 Sponsored by: DARPA, AFRL r339912: Fix the location of the static keyword. MFC with: r339738 Sponsored by: DARPA, AFRL r339913: Disable the .preinit_array test in DSOs, ld.bfd fails to link objects with the section. MFC with: r339738 Sponsored by: DARPA, AFRL r339916: Build the csu tests on all architectures. The tests haven't been run them, but this is enough to build them so I can get feedback on if the various crt.h headers are correct. MFC with: r339738 Sponsored by: DARPA, AFRL r339954: Add __used to __CTOR_LIST__ and __DTOR_LIST__ Enabling BSD_CRTBEGIN on amd64 resulted in error: unused variable '__CTOR_LIST__'. __CTOR_LIST__ is indeed unused in crtbegin.c; it marks the beginning of the .ctors array and is used in crtend.c. Annotate __DTOR_LIST__ as well for consistency. Discussed with: andrew MFC with: r339738 Sponsored by: The FreeBSD Foundation r340213: Add the (untested) mips and sparc64 .init call sequences. The BSD crtbegin/crtend code now builds on all architectures, however further work is needed to check if it works correctly. MFC with: r339738 Sponsored by: DARPA, AFRL r340395: Run __cxa_finalize in shared objects in the destructor path. When we have .dtors call them before .dtor handling, otherwise call from a destructor. PR: 233056 MFC with: r339738 Sponsored by: DARPA, AFRL r340840: Mark the function called by the MIPS .init/.fini sequence with .local. As with r328939 we need to mark local symbols as such. Without this the assembly parser treats the symbols as global and created relocations against these private symbols. MFC with: r339738 Sponsored by: DARPA, AFRL r340910: Add the missing 0 at the end of the .jcr section. Without this the dynamic library test was failing as it was calling _Jv_RegisterClasses multiple times. r340911: Re-enable the dynamiclib tests. These should be fixed by r340910. r341424: Disable the BSD CRT code on powerpc and sparc64, they need extra crt*.o files that haven't been implemented. Reported by: sbruno MFC with: r339738 Sponsored by: DARPA, AFRL r342773: Add explicit csu test dependency lib/csu/tests/dynamiclib requires libh_csu.so be built first. I'm not sure this is the most correct/best way to address this but it solves the issue in my testing. PR: 233734 Sponsored by: The FreeBSD Foundation r342974: Create crtsavres.o for powerpc builds Summary: GCC expects to link in a crtsavres.o on powerpc platforms. On powerpc64 this is an empty file, but on powerpc and powerpcspe this does contain some save/restore functions, which may not actually be necessary for newer modern GCC and clang. This appeases the in-tree gcc, though, and is needed in order to switch to the BSD CRTRBEGIN. PR: 233751 Reviewed By: andrew Differential Revision: https://reviews.freebsd.org/D18826 r351027: Enable BSD_CRTBEGIN on powerpc In r342974 jhibbits added support to build crtsavres.o. This was the blocker for BSD_CRTBEGIN to be enabled there. As such enable this option again. Reviewed by: jhibbits Sponsored by: DARPA, AFRL r352073: csu: Add the riscv .init call sequence Reviewed by: br Sponsored by: Axiado Differential Revision: https://reviews.freebsd.org/D21537 Changes: _U stable/12/ stable/12/etc/mtree/BSD.tests.dist stable/12/gnu/lib/Makefile stable/12/gnu/lib/csu/Makefile stable/12/lib/csu/Makefile stable/12/lib/csu/Makefile.inc stable/12/lib/csu/aarch64/crt.h stable/12/lib/csu/amd64/crt.h stable/12/lib/csu/arm/crt.h stable/12/lib/csu/common/crtbegin.c stable/12/lib/csu/common/crtend.c stable/12/lib/csu/i386/crt.h stable/12/lib/csu/mips/crt.h stable/12/lib/csu/powerpc/Makefile stable/12/lib/csu/powerpc/crt.h stable/12/lib/csu/powerpc/crtsavres.S stable/12/lib/csu/powerpc64/Makefile stable/12/lib/csu/powerpc64/crt.h stable/12/lib/csu/riscv/crt.h stable/12/lib/csu/sparc64/crt.h stable/12/lib/csu/tests/ stable/12/lib/csu/tests/Makefile stable/12/lib/csu/tests/Makefile.tests stable/12/lib/csu/tests/cxx_constructors.cc stable/12/lib/csu/tests/dso/ stable/12/lib/csu/tests/dynamic/Makefile stable/12/lib/csu/tests/dynamiclib/ stable/12/lib/csu/tests/fini_test.c stable/12/lib/csu/tests/init_test.c stable/12/lib/csu/tests/static/Makefile stable/12/share/mk/src.opts.mk stable/12/tools/build/options/WITHOUT_BSD_CRTBEGIN stable/12/tools/build/options/WITH_BSD_CRTBEGIN