We the extra debug printouts the crash looks like this: ELF [tid:42, /usr/lib/libgcc_s.so.1]: Instantiated ELF [tid:42, /usr/lib/libgcc_s.so.1]: The base set to: 0x00000000000a7000 and end: 0x00000000000c0450 ELF [tid:42, /usr/lib/libgcc_s.so.1]: Loading segments ELF [tid:42, /usr/lib/libgcc_s.so.1]: Loaded and mapped PT_LOAD segment at: 0x00000000000a7000 of size: 0x3000 ELF [tid:42, /usr/lib/libgcc_s.so.1]: Loaded and mapped PT_LOAD segment at: 0x00000000000aa000 of size: 0x11000 ELF [tid:42, /usr/lib/libgcc_s.so.1]: Loaded and mapped PT_LOAD segment at: 0x00000000000bb000 of size: 0x4000 ELF [tid:42, /usr/lib/libgcc_s.so.1]: Loaded and mapped PT_LOAD segment at: 0x00000000000bf000 of size: 0x2000 ELF [tid:42, /usr/lib/libgcc_s.so.1]: Loading DT_NEEDED object: libc.so.6 /usr/lib/libgcc_s.so.1: ignoring missing symbol __cpu_indicator_init /usr/lib/libgcc_s.so.1: ignoring missing symbol __cpu_model ELF [tid:42, /usr/lib/libgcc_s.so.1]: Relocated 10 symbols in DT_RELA ELF [tid:42, /usr/lib/libgcc_s.so.1]: Relocated 45 PLT symbols in DT_JMPREL ELF [tid:42, /usr/lib/libgcc_s.so.1]: versioned symbol table ELF [tid:42, /java_isolated.so]: Loading DT_NEEDED object: libc.so.6 ELF [tid:42, /java_isolated.so]: Relocated 469 symbols in DT_RELA ELF [tid:42, /java_isolated.so]: Relocated 246 PLT symbols in DT_JMPREL ELF [tid:42, /java_isolated.so]: versioned symbol table ELF [tid:253, /usr/lib/libgcc_s.so.1]: Executing DT_INIT function ELF [tid:253, /usr/lib/libgcc_s.so.1]: Finished executing DT_INIT function ELF [tid:253, /usr/lib/libgcc_s.so.1]: Executing 2 DT_INIT_ARRAYSZ functions trying to execute or access missing symbol [backtrace] 0x00000000403aeb19 <page_fault+297> 0x00000000403ad8d6 <???+1077598422> 0x00000000403609fa <elf::program::init_library(int, char**)+362> 0x0000000040226237 <osv::application::main()+71> 0x000000004044406c <???+1078214764> 0x00000000404782b9 <???+1078428345> 0x000000004040f29b <thread_main_c+43> 0x00000000403ae852 <???+1077602386>
It looks like one of the 2 functions in DT_INIT_ARRAYSZ references one of the missing symbols (per versym). Probably __cpu_indicator_init. So I wonder that instead of skipping deprecated symbols, we should tweak the versioning logic to make it find symbols IF referenced by the object itself. Other part of the Drepper's text - https://www.akkadia.org/drepper/symbol-versioning - seems to suggest that: "vna_other Contains version index unique for the file which is used in the version symbol table. If the highest bit (bit 15) is set this is a hidden symbol which cannot be referenced *from outside the object*." On Saturday, November 30, 2019 at 1:16:46 PM UTC-5, Waldek Kozaczuk wrote: > > Or is the issue that given that __cpu_indicator_init is a constructor > function that is in symver and deprecated > 4.8 we should also skip > calling deprecated INIT functions ? > > On Sat, Nov 30, 2019 at 00:34 Waldek Kozaczuk <jwkozac...@gmail.com> > wrote: > >> Here is the change made to libgcc.so 4 years ago that seems to be causing >> this problem - >> https://github.com/gcc-mirror/gcc/commit/4b5fb32aba30762e0d8c9e75d1b46b47bee5eeb4 >> . >> >> These 2 missing (hidden with versioning) are present in libgcc.a: >> readelf -s /usr/lib/gcc/x86_64-linux-gnu/9/libgcc.a | grep >> __cpu_indicator_init >> 13: 0000000000000290 905 FUNC GLOBAL HIDDEN 4 >> __cpu_indicator_init >> >> We stopped linking libgcc.a as whole-archive year ago but even if we >> still did it would not help us given this symbol is hidden. >> >> I wonder how we should solve it. >> >> >> On Thursday, November 28, 2019 at 3:18:48 PM UTC-5, Waldek Kozaczuk wrote: >>> >>> This is related to https://github.com/cloudius-systems/osv/issues/743. >>> >>> I have been trying to eliminate ancient libgcc_s.so from >>> usr.manifest.skel and replace it with the version from host. When I do that >>> most (C++ only?) apps fail like so: >>> >>> Cmdline: /java_isolated.so -cp /tests/java/tests.jar >>> io.osv.TestDomainPermissions >>> /usr/lib/libgcc_s.so.1: ignoring missing symbol __cpu_indicator_init >>> /usr/lib/libgcc_s.so.1: ignoring missing symbol __cpu_model >>> trying to execute or access missing symbol >>> [backtrace] >>> 0x000000004039d6ab <page_fault+267> >>> 0x000000004039c486 <???+1077527686> >>> 0x00000000403528ca <elf::program::init_library(int, char**)+362> >>> 0x0000000040223cdd <osv::application::main()+61> >>> 0x000000004042deb8 <???+1078124216> >>> 0x000000004045fdb5 <???+1078328757> >>> 0x00000000403f9ce7 <thread_main_c+39> >>> 0x000000004039d402 <???+1077531650> >>> >>> I somehow discovered it has to do with versioning. So when I undo my >>> commit that fixed libnuma (see >>> https://github.com/cloudius-systems/osv/commit/ed1eed7a567ec17138c65f0a5628c2311603c712 >>> ). >>> >>> So if I manually undo these changes the error disappears. It is >>> obviously related to these warnings: >>> /usr/lib/libgcc_s.so.1: ignoring missing symbol __cpu_indicator_init >>> /usr/lib/libgcc_s.so.1: ignoring missing symbol __cpu_model >>> >>> What is actually going on? And what is the right fix? >>> >>> I found couple of the things that might be related: >>> >>> https://bugzilla.redhat.com/show_bug.cgi?id=1688766 >>> >>> http://cgit.openembedded.org/openembedded-core/tree/meta/recipes-devtools/gcc/gcc-9.2/0032-libgcc_s-Use-alias-for-__cpu_indicator_init-instead-.patch?h=master >>> >>> Waldek >>> >>> >>> >>> >>> -- >> You received this message because you are subscribed to a topic in the >> Google Groups "OSv Development" group. >> To unsubscribe from this topic, visit >> https://groups.google.com/d/topic/osv-dev/HZHzeG_5IqM/unsubscribe. >> To unsubscribe from this group and all its topics, send an email to >> osv-dev+unsubscr...@googlegroups.com. >> To view this discussion on the web visit >> https://groups.google.com/d/msgid/osv-dev/e5b7c460-b7d5-438f-9001-5546f65c4894%40googlegroups.com >> >> <https://groups.google.com/d/msgid/osv-dev/e5b7c460-b7d5-438f-9001-5546f65c4894%40googlegroups.com?utm_medium=email&utm_source=footer> >> . >> > -- You received this message because you are subscribed to the Google Groups "OSv Development" group. To unsubscribe from this group and stop receiving emails from it, send an email to osv-dev+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/osv-dev/ca3133df-a5bf-42bb-84da-b1963aef3d3d%40googlegroups.com.
