| Summary: | linker failed to remove all dead code dependency properly | ||
|---|---|---|---|
| Product: | binutils | Reporter: | Michael Matz <matz> |
| Component: | ld | Assignee: | Alan Modra <amodra> |
| Status: | RESOLVED FIXED | ||
| Severity: | normal | CC: | grabli_2005, hjl.tools |
| Priority: | P2 | ||
| Version: | 2.24 | ||
| Target Milestone: | --- | ||
| Host: | Target: | ||
| Build: | Last reconfirmed: | ||
Works for me: [hjl@gnu-6 pr12772]$ cat x.cc #include <string.h> #include <stdlib.h> #include <unistd.h> extern "C" int __attribute__((visibility("default"))) func(char *ptr) { static const char g_const[] = { 1, 1, 0 }; memcpy(ptr, g_const, 3); return 0; } extern "C" char *func2(char *ptr) { static char buf[4096]; if (!getcwd(buf, sizeof(buf))) buf[0] = 0; #define STR "test_string" memcpy(ptr, STR, sizeof(STR)); return buf; } [hjl@gnu-6 pr12772]$ make g++ -c -fpic -O3 -fdata-sections -ffunction-sections -fvisibility-inlines-hidden -fvisibility=hidden -o x.o x.cc g++ -shared -Wl,-gc-sections -o x.so x.o readelf -r x.so Relocation section '.rela.dyn' at offset 0x3f0 contains 4 entries: Offset Info Type Sym. Value Sym. Name + Addend 000000200640 000000000008 R_X86_64_RELATIVE 0000000000200640 0000002007f8 000200000006 R_X86_64_GLOB_DAT 0000000000000000 __gmon_start__ + 0 000000200800 000300000006 R_X86_64_GLOB_DAT 0000000000000000 _Jv_RegisterClasses + 0 000000200808 000400000006 R_X86_64_GLOB_DAT 0000000000000000 __cxa_finalize + 0 Relocation section '.rela.plt' at offset 0x450 contains 1 entries: Offset Info Type Sym. Value Sym. Name + Addend 000000200828 000400000007 R_X86_64_JUMP_SLO 0000000000000000 __cxa_finalize + 0 [hjl@gnu-6 pr12772]$ Works in what meaning?
I have same readelf -r output:
Relocation section '.rela.dyn' at offset 0x440 contains 4 entries:
Offset Info Type Sym. Value Sym. Name + Addend
0000000018c8 000000000008 R_X86_64_RELATIVE 00000000000018c8
000000001868 000100000006 R_X86_64_GLOB_DAT 0000000000000000 __gmon_start__ + 0
000000001870 000500000006 R_X86_64_GLOB_DAT 0000000000000000 __cxa_finalize + 0
000000001878 000600000006 R_X86_64_GLOB_DAT 0000000000000000 _Jv_RegisterClasses + 0
Relocation section '.rela.plt' at offset 0x4a0 contains 1 entries:
Offset Info Type Sym. Value Sym. Name + Addend
000000001898 000500000007 R_X86_64_JUMP_SLO 0000000000000000 __cxa_finalize + 0
But readelf -s shows:
Symbol table '.dynsym' contains 12 entries:
Num: Value Size Type Bind Vis Ndx Name
0: 0000000000000000 0 NOTYPE LOCAL DEFAULT UND
1: 00000000000004c8 0 SECTION LOCAL DEFAULT 10
2: 0000000000000000 0 NOTYPE WEAK DEFAULT UND __gmon_start__
3: 0000000000000000 0 NOTYPE WEAK DEFAULT UND _Jv_RegisterClasses
4: 0000000000000000 0 FUNC WEAK DEFAULT UND __cxa_finalize@GLIBC_2.2.5 (2)
5: 0000000000000000 0 FUNC GLOBAL DEFAULT UND getcwd@GLIBC_2.2.5 (2)
6: 0000000000201020 0 NOTYPE GLOBAL DEFAULT ABS _end
7: 0000000000201010 0 NOTYPE GLOBAL DEFAULT ABS _edata
8: 0000000000201010 0 NOTYPE GLOBAL DEFAULT ABS __bss_start
9: 00000000000004c8 0 FUNC GLOBAL DEFAULT 10 _init
10: 000000000000062c 16 FUNC GLOBAL DEFAULT 13 _fini
11: 00000000000005d0 23 FUNC GLOBAL DEFAULT 12 func
I not obvious to me that relocations is in blame here, but as in my original report - getcwd dependency not discarded properly.
*It`s not obvious to me Yeah, sorry, my title was wrong. The relocations really are gone. But the symbol table entry will stay, even though nothing refers to it anymore (ultimately this then leads to libraries being included that aren't used by anything). This optimisation could be done in a target gc_sweep_symbol hook. You can't do so in the generic ELF code because you'd need to look at plt, got and dyn_reloc fields of the sym. gold linker already fix this issue http://sourceware.org/bugzilla/show_bug.cgi?id=12818 any troubles to fix this in ld? This bug leads to some difficulties in android. Application loads many useless .so libraries (and very heavy) and debug analysys become harder when there lot of garbage symbols and I don`t know which of them really used. CVSROOT: /cvs/src Module name: src Changes by: amodra@sourceware.org 2011-12-07 12:15:54 Modified files: bfd : ChangeLog elflink.c Log message: PR ld/12772 * elflink.c (elf_gc_sweep_symbol): Discard unmarked symbols defined in shared libraries. Patches: http://sourceware.org/cgi-bin/cvsweb.cgi/src/bfd/ChangeLog.diff?cvsroot=src&r1=1.5533&r2=1.5534 http://sourceware.org/cgi-bin/cvsweb.cgi/src/bfd/elflink.c.diff?cvsroot=src&r1=1.431&r2=1.432 Fixed mainline The master branch has been updated by Maciej W. Rozycki <macro@sourceware.org>: https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=81ff47b3a54633819fac4d973e34f1ff0c65606e commit 81ff47b3a54633819fac4d973e34f1ff0c65606e Author: Maciej W. Rozycki <macro@imgtec.com> Date: Mon Jan 16 22:10:57 2017 +0000 PR ld/20828: Fix linker script symbols wrongly forced local with section GC Fix a generic ELF linker regression introduced with a chain of changes made to unused input section garbage collection: - commit 1a766c6843ce ("Also hide symbols without PLT nor GOT references."), <https://sourceware.org/ml/binutils/2011-09/msg00076.html>, - commit 1d5316ab67e1 ("PR ld/13177: garbage collector retains zombie references to external libraries"), <https://sourceware.org/ml/binutils/2011-10/msg00161.html>, - commit 6673f753c019 ("Fix PR 12772, garbage collection of dynamic syms"), <https://sourceware.org/ml/binutils/2011-12/msg00077.html>, causing the garbage collection of unused symbols present in a DSO involved in a link to make identically named symbols ordinarily defined (i.e. not hidden or PROVIDEd) by a linker script local, even though the latter symbols are supposed to be global as if no DSO defined them as well. This is because linker script assignments are processed very late as `lang_process' proceeds, down in the call to `ldemul_before_allocation', which is made after the call to `lang_gc_sections' to do input section garbage collecting. Consequently if unused, then any such DSO-defined symbol has already been garbage-collected and internally marked local. It would ordinarily be removed from dynamic symbol table output, however a linker script assignment correctly replaces its original definition with the new one and enters it into the dynamic symbol table produced as it is supposed to be exported. The original local marking is however retained making the symbol local in the dynamic symbol table and therefore not available externally. This also causes a sorting problem with the MIPS target, which does not expect non-section local dynamic symbols to be output and produces an invalid binary. Fix the problem then, by removing the `forced_local' marking for the offending case and add suitable test cases. First to verify that unused symbols ordinarily defined with linker script assignments remain exported in the context of input section garbage collection whether or not a DSO defining identically named symbols is present in the link. Second that a linker version script still correctly retains or removes such symbols as requested. bfd/ PR ld/20828 * elflink.c (bfd_elf_record_link_assignment): Clear any `forced_local' marking for DSO symbols that are not being provided. ld/ PR ld/20828 * testsuite/ld-elf/pr20828-1.sd: New test. * testsuite/ld-elf/pr20828-2a.sd: New test. * testsuite/ld-elf/pr20828-2b.sd: New test. * testsuite/ld-elf/pr20828.ld: New test linker script. * testsuite/ld-elf/pr20828.ver: New test version script. * testsuite/ld-elf/pr20828.s: New test source. * testsuite/ld-elf/shared.exp: Run the new test. The binutils-2_28-branch branch has been updated by Maciej W. Rozycki <macro@sourceware.org>: https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=4030096d6a689125c026ce1e6300717c2050d105 commit 4030096d6a689125c026ce1e6300717c2050d105 Author: Maciej W. Rozycki <macro@imgtec.com> Date: Mon Jan 16 22:10:57 2017 +0000 PR ld/20828: Fix linker script symbols wrongly forced local with section GC Fix a generic ELF linker regression introduced with a chain of changes made to unused input section garbage collection: - commit 1a766c6843ce ("Also hide symbols without PLT nor GOT references."), <https://sourceware.org/ml/binutils/2011-09/msg00076.html>, - commit 1d5316ab67e1 ("PR ld/13177: garbage collector retains zombie references to external libraries"), <https://sourceware.org/ml/binutils/2011-10/msg00161.html>, - commit 6673f753c019 ("Fix PR 12772, garbage collection of dynamic syms"), <https://sourceware.org/ml/binutils/2011-12/msg00077.html>, causing the garbage collection of unused symbols present in a DSO involved in a link to make identically named symbols ordinarily defined (i.e. not hidden or PROVIDEd) by a linker script local, even though the latter symbols are supposed to be global as if no DSO defined them as well. This is because linker script assignments are processed very late as `lang_process' proceeds, down in the call to `ldemul_before_allocation', which is made after the call to `lang_gc_sections' to do input section garbage collecting. Consequently if unused, then any such DSO-defined symbol has already been garbage-collected and internally marked local. It would ordinarily be removed from dynamic symbol table output, however a linker script assignment correctly replaces its original definition with the new one and enters it into the dynamic symbol table produced as it is supposed to be exported. The original local marking is however retained making the symbol local in the dynamic symbol table and therefore not available externally. This also causes a sorting problem with the MIPS target, which does not expect non-section local dynamic symbols to be output and produces an invalid binary. Fix the problem then, by removing the `forced_local' marking for the offending case and add suitable test cases. First to verify that unused symbols ordinarily defined with linker script assignments remain exported in the context of input section garbage collection whether or not a DSO defining identically named symbols is present in the link. Second that a linker version script still correctly retains or removes such symbols as requested. bfd/ PR ld/20828 * elflink.c (bfd_elf_record_link_assignment): Clear any `forced_local' marking for DSO symbols that are not being provided. ld/ PR ld/20828 * testsuite/ld-elf/pr20828-1.sd: New test. * testsuite/ld-elf/pr20828-2a.sd: New test. * testsuite/ld-elf/pr20828-2b.sd: New test. * testsuite/ld-elf/pr20828.ld: New test linker script. * testsuite/ld-elf/pr20828.ver: New test version script. * testsuite/ld-elf/pr20828.s: New test source. * testsuite/ld-elf/shared.exp: Run the new test. (cherry picked from commit 81ff47b3a54633819fac4d973e34f1ff0c65606e) |
(this got initially reported into Novell bugzilla #694266). Compile this with the following compile command on x86_64-linux: # g++ -O3 -Wl,-gc-sections -fpic -shared \ -fdata-sections -ffunction-sections \ -fvisibility-inlines-hidden -fvisibility=hidden \ -o test.so test.cpp # cat test.cpp #include <string.h> #include <stdlib.h> #include <unistd.h> extern "C" int __attribute__((visibility("default"))) func(char *ptr) { static const char g_const[] = { 1, 1, 0 }; memcpy(ptr, g_const, 3); return 0; } extern "C" char *func2(char *ptr) { static char buf[4096]; if (!getcwd(buf, sizeof(buf))) buf[0] = 0; #define STR "test_string" memcpy(ptr, STR, sizeof(STR)); return buf; } Note that func2 will be hidden, and as it's unreferenced the .text.func2 section will be discarded, as will be the .bss._ZZ5func2E3buf section: # ... ./ld/ld-new: Removing unused section '.text.func2' in file 'show-link-discard-fail.o' ./ld/ld-new: Removing unused section '.bss._ZZ5func2E3buf' in file 'show-link-discard-fail.o' But the relocation to getcwd (or better it's default decorated variant getcwd@@GLIBC_2.2.5) will stay and transferred into the output file. It should have been discarded too. In fact the gc_sweep_hook on x86_64 for instance does lower the PLT count for this reloc, and hence seems to assume that it really won't be emitted. But alas, it is.