| ToDo list for elfutils -*-outline-*- | |
| ---------------------- | |
| Time-stamp: <2009-02-05 22:08:01 drepper> | |
| * mkinstalldirs | |
| Remove everywhere. Use mkdir -p. | |
| * libelf: | |
| ** verify section | |
| Currently the elf_update function trusts the user blindly if the | |
| ELF_F_LAYOUT flag is set. This is OK if the data is prepared by a | |
| ELF_C_NULL call but not if the user prepared the data herself | |
| ** break out archive handling from elf_begin | |
| The handling of archives (especially of the symbol tables) must be | |
| broken out of elf_begin. There are several different forms of | |
| archives and only when having the archive handling separately this | |
| remains maintainable. | |
| ** shdrs in read-only files | |
| When reading (ELF_C_READ*) then there is no need to malloc Shdr | |
| structure in elfXX_getshdr if file is mmaped and unaligned access | |
| is allowed or the structure is aligned. Use ELF_F_MALLOCED flag | |
| to differentiate. | |
| ** shdrs after elf_cntl (ELF_C_FDREAD) | |
| Similar to the above. After ELF_C_FDREAD the file is completely | |
| in memory. See also this mailing list thread: | |
| https://fedorahosted.org/pipermail/elfutils-devel/2012-July/002368.html | |
| * libdw | |
| ** More memory access checks needed | |
| All accesses to the debug sections should make sure the offsets are | |
| valid. This is currently especially a problem with leb128 accesses. | |
| ** Low level macro information operations | |
| in 5.11.3 are not implemented. gcc currently does not emit this | |
| information so I cannot test it. | |
| ** Rename dwarf_getabbrev | |
| * libcpu | |
| ** x86 | |
| *** Opcodes | |
| crc32 | |
| extractps | |
| pextrb | |
| pextrd/pextrq | |
| pextrw | |
| pinsrq | |
| popcnt 64-bit reg | |
| * nm: | |
| ** add demangler support | |
| Use demangler from libiberty. | |
| ** add support to read debugging symbols | |
| Implement -l option for BSD and POSIX format | |
| * strip: | |
| ** support SHT_SYMTAB_SHNDX | |
| should be removed if not needed anymore | |
| * ld: | |
| ** sanity check .rel sh_info content | |
| the sh_info of all .rel sections with the same name must point to | |
| sections which also have the same name | |
| ** use ld.so.conf | |
| to locate shared libraries also use /etc/ld.so.conf | |
| ** handle object files for different architectures | |
| ld.so is expected to ignore object files for different architectures and | |
| continue looking for a matching file (e.g., ignore 32-bit binaries on | |
| 64-bit platforms and vice versa). We probably need the same in ld. | |
| ** reuse after elf_end | |
| Some files are closed using elf_end. They are removed from memory only | |
| if no reference is left (especially for archives this is a problem). | |
| The old mapping should be reused in that case. The problem is worse | |
| for files which are not mapped read-only (archives again). | |
| ** size for STT_SECTION entries | |
| The STT_SECTION entries have zero for the size but can easily get | |
| the size of the section. | |
| ** .eh_frame_hdr | |
| Not implemented at all in the moment except for recognition of the option | |
| itself. | |
| ** variables with aliases in executables | |
| When linking an executable with a references against a variable in a | |
| DSO, create symbol table entries for all the aliases of the variable | |
| in the DSO and create a relocation for one of them (a non-weak | |
| definition) | |
| * elflint | |
| ** additional checks | |
| 1st GOT entry == _DYNAMIC | |
| check versioning info: | |
| always BASE in verdef | |
| sh_size/sh_entsize matches last offset != 0 | |
| check whether any relocation is for a merge-able section | |
| check TLS relocation dependencies | |
| Check content of .eh_frame_hdr, .eh_frame, .gcc_except_table | |
| *** for x86 | |
| check that R_386_TLS_GD is followed by R_386_PLT32 for __tls_get_addr | |
| ** relax | |
| prelink generated files | |
| * elfcmp | |
| ** treat relocation sections special | |
| Differences in the relocation sections can be ignored if all | |
| the same symbols with the same targets are present and the order | |
| of overlapping relocations doesn't change. There really never | |
| should be overlapping relocations but who knows. | |
| * mcs | |
| Sun has it. Can modify sections which are not in segments. | |
| -a string | |
| Append string to the comment section of the ELF object | |
| files. If string contains embedded blanks, it must be | |
| enclosed in quotation marks. | |
| -c Compress the contents of the comment section of the | |
| ELF object files. All duplicate entries are removed. | |
| The ordering of the remaining entries is not dis- | |
| turbed. | |
| -d Delete the contents of the comment section from the | |
| ELF object files. The section header for the comment | |
| section is also removed. | |
| -n name | |
| Specify the name of the comment section to access if | |
| other than .comment. By default, mcs deals with the | |
| section named .comment. This option can be used to | |
| specify another section. mcs can take multiple -n | |
| options to allow for specification of multiple sec- | |
| tion comments. | |
| -p Print the contents of the comment section on the stan- | |
| dard output. Each section printed is tagged by the | |
| name of the file from which it was extracted, using | |
| the format file[member_name]: for archive files and | |
| file: for other files. | |
| -V Print on standard error the version number of mcs. | |
| Local Variables: | |
| eval:(hide-sublevels 3) | |
| End: |