Re: [PATCH] backends: sparc uses NONE relocation type in ET_EXEC and ET_DYN files
> Using the NONE relocation doesn't do anything, but is harmless. > > This fixes several (self) tests on sparc that use elflint to > check files are valid ELF. FWIW looks good to me. > > Signed-off-by: Mark Wielaard > --- > backends/ChangeLog | 4 > backends/sparc_reloc.def | 2 +- > 2 files changed, 5 insertions(+), 1 deletion(-) > > diff --git a/backends/ChangeLog b/backends/ChangeLog > index f28ab89b..40ec7c0c 100644 > --- a/backends/ChangeLog > +++ b/backends/ChangeLog > @@ -1,3 +1,7 @@ > +2023-01-19 Mark Wielaard > + > + * sparc_reloc.def (NONE): Add EXEC and DYN. > + > 2022-12-21 Shahab Vahedi > > * Makefile.am (modules): Add arc. > diff --git a/backends/sparc_reloc.def b/backends/sparc_reloc.def > index 7cd5ce96..f0eeb5c6 100644 > --- a/backends/sparc_reloc.def > +++ b/backends/sparc_reloc.def > @@ -28,7 +28,7 @@ > > /* NAME, REL|EXEC|DYN*/ > > -RELOC_TYPE (NONE, REL) > +RELOC_TYPE (NONE, REL|EXEC|DYN) > RELOC_TYPE (8, REL) > RELOC_TYPE (16, REL) > RELOC_TYPE (32, REL|DYN)
Re: [PATCHv2] strip: keep .ctf section in stripped file
> Hi, > > On Tue, 2022-12-20 at 22:35 +0100, Mark Wielaard wrote: >> On Wed, Jun 01, 2022 at 10:55:27AM -0500, Guillermo E. Martinez via >> Elfutils-devel wrote: >> > This is the second version patch to avoid remove the CTF section in >> > stripped files. Changes from v1: >> > >> > - Add description in tests/run-strip-remove-keep-ctf.sh >> > mentioning how to regenerate test input file (testfile-ctf) >> > >> > Please let me know your thoughts. >> > >> > [...] >> > >> > CTF debug format was designed to be present in stripped files, so >> > this section should not be removed, so a new --remove-ctf option >> > is added to indicate explicitly that .ctf section will be stripped >> > out from binary file. >> >> Sorry, I see I never reviewed this v2 variant. I know we tried to >> coordinate with binutils so eu-strip and binutils strip would do the >> same thing. And that Jose had an idea for a new section flag to >> automatically detect what section should/shouldn't be stripped (into a >> separate .debug file). What was the conclusion of that? > > Any update on this? > How should .ctf sections be dealt with by strip like tools? I think the conclusion was that having a NOSTRIP flag in the object file is far from trivial, because what "strip" means for several people and tools is not clear. So I guess we are stuck with the old keep-a-list method. Or we may add NOSTRIP meaning "_never ever_ strip this section, period, no matter what."
Re: [PATCH] strip: keep .ctf section in stripped file
[Added Nick in CC.]
> Hi Guillermo,
>
> On Mon, May 30, 2022 at 09:26:19PM -0500, Guillermo E. Martinez via
> Elfutils-devel wrote:
>> Hello elfutils team,
>>
>> This patch is meant to avoid remove the CTF section in
>> stripped files. Please let me know your thoughts.
>>
>> CTF debug format was designed to be present in stripped files, so
>> this section should not be removed, so a new --remove-ctf option
>> is added to indicate explicitly that .ctf section will be stripped
>> out from binary file.
>
> First, very nice patch. My only real concern with it is that it should
> describe how the testfile-ctf.bz2 is generated. We don't want really
> random test binaries in the testsuite. There should be at least some
> method to regenerate them, even if we don't automate that. See
> e.g. tests/run-readelf-n.sh which also uses binary test files, but has
> a little description on how each of them was generated.
>
> I am CCing Nick Clifton who works on binutils and annobin to check how
> binutils strip handles this. And because we were recently discussing
> putting some annobin data in a special section and how to indicate
> that this section should be explicitly kept or removed. It would be
> great if we could come to some kind of standard way of marking such
> sections so we don't need special arguments for each such section or
> at least have a more generic SECTION_STRIP_P macro.
AFAIK the binutils `strip' does not strip .ctf sections.
> Note, that this might not be possible, these sections might be to
> different/specific that generalizing over them is impossible. But it
> would be good to at least try and discuss it. If only so that elfutils
> eu-strip and binutils strip agree on how to handle/coordinate on such
> special sections.
What about using an OS-specific section flag in elf.h, something like:
#define SHF_GNU_PERSISTENT 0x0ff1 /* Section must not be stripped. */
>
> Cheers,
>
> Mark
>
>> Signed-off-by: Guillermo E. Martinez
>> ---
>> ChangeLog | 23
>> libebl/eblsectionstripp.c | 4 +-
>> libebl/libebl.h| 2 +-
>> libelf/elf-knowledge.h | 7 +-
>> libelf/elf32_checksum.c| 2 +-
>> src/elfcmp.c | 4 +-
>> src/strip.c| 71 --
>> tests/Makefile.am | 6 +-
>> tests/run-strip-remove-keep-ctf.sh | 207 +
>> tests/testfile-ctf.bz2 | Bin 0 -> 3317 bytes
>> 10 files changed, 304 insertions(+), 22 deletions(-)
>> create mode 100755 tests/run-strip-remove-keep-ctf.sh
>> create mode 100755 tests/testfile-ctf.bz2
>>
>> diff --git a/ChangeLog b/ChangeLog
>> index f1a14b5c..2b608866 100644
>> --- a/ChangeLog
>> +++ b/ChangeLog
>> @@ -1,3 +1,26 @@
>> +2022-05-21 Guillermo E. Martinez
>> +
>> +* libebl/eblsectionstripp.c (ebl_section_strip_p): Use
>> +remove_ctf argument.
>> +(SECTION_STRIP_P): Likewise.
>> +* libebl/libebl.h (ebl_section_strip_p): Likewise.
>> +* libelf/elf-knowledge.h (SECTION_STRIP_P): Update macro
>> +definition to use remove_ctf to determine whether .ctf
>> +section is stripped out.
>> +* libelf/elf32_checksum.c (elfw2): Use false value for
>> +remove_ctf parameter.
>> +* src/elfcmp.c (main): Likewise.
>> +* src/strip.c (options): Add --remove-ctf option set by
>> +remove_ctf variable.
>> +(set_remove_special_section_opt): Add new function.
>> +(erratic_special_section_opt): Likewise.
>> +(parse_opt): Parse new --remove-ctf option.
>> +(handle_elf): Adjust .comment and use remove_ctf argument.
>> +* tests/Makefile.am (TEST): Add run-strip-remove-keep-ctf.sh
>> +and testfile-ctf.bz2.
>> +* tests/run-strip-remove-keep-ctf.sh: Add new testcase.
>> +* tests/testfile-ctf.bz2: Add new test harness.
>> +
>> 2022-05-02 Mark Wielaard
>>
>> * Makefile.am (AM_DISTCHECK_CONFIGURE_FLAGS): Remove
>> diff --git a/libebl/eblsectionstripp.c b/libebl/eblsectionstripp.c
>> index a5624ffe..f26cc170 100644
>> --- a/libebl/eblsectionstripp.c
>> +++ b/libebl/eblsectionstripp.c
>> @@ -37,7 +37,7 @@
>> bool
>> ebl_section_strip_p (Ebl *ebl, const GElf_Shdr *shdr,
>> const char *name, bool remove_comment,
>> - bool only_remove_debug)
>> + bool only_remove_debug, bool remove_ctf)
>> {
>>/* If only debug information should be removed check the name. There
>> is unfortunately no other way. */
>> @@ -66,5 +66,5 @@ ebl_section_strip_p (Ebl *ebl, const GElf_Shdr *shdr,
>>return false;
>> }
>>
>> - return SECTION_STRIP_P (shdr, name, remove_comment);
>> + return SECTION_STRIP_P (shdr, name, remove_comment, remove_ctf);
>> }
>> diff --git a/libebl/libebl.h b/libebl/libebl.h
>> index 731001d3..067b769e 100644
>> --- a/libebl/libebl.h
>> +++ b/libebl/libebl.h
>> @@ -205,7 +205,7 @@ extern bool ebl_relative_reloc_p (Ebl *ebl, int rel
Re: [PATCH] strip: keep .ctf section in stripped file
> [Added Nick in CC.] > >> Hi Guillermo, >> >> On Mon, May 30, 2022 at 09:26:19PM -0500, Guillermo E. Martinez via >> Elfutils-devel wrote: >>> Hello elfutils team, >>> >>> This patch is meant to avoid remove the CTF section in >>> stripped files. Please let me know your thoughts. >>> >>> CTF debug format was designed to be present in stripped files, so >>> this section should not be removed, so a new --remove-ctf option >>> is added to indicate explicitly that .ctf section will be stripped >>> out from binary file. >> >> First, very nice patch. My only real concern with it is that it should >> describe how the testfile-ctf.bz2 is generated. We don't want really >> random test binaries in the testsuite. There should be at least some >> method to regenerate them, even if we don't automate that. See >> e.g. tests/run-readelf-n.sh which also uses binary test files, but has >> a little description on how each of them was generated. >> >> I am CCing Nick Clifton who works on binutils and annobin to check how >> binutils strip handles this. And because we were recently discussing >> putting some annobin data in a special section and how to indicate >> that this section should be explicitly kept or removed. It would be >> great if we could come to some kind of standard way of marking such >> sections so we don't need special arguments for each such section or >> at least have a more generic SECTION_STRIP_P macro. >> >> Note, that this might not be possible, these sections might be to >> different/specific that generalizing over them is impossible. But it >> would be good to at least try and discuss it. If only so that elfutils >> eu-strip and binutils strip agree on how to handle/coordinate on such >> special sections. > > What about using an OS-specific section flag in elf.h, something like: > > #define SHF_GNU_PERSISTENT 0x0ff1 /* Section must not be stripped. */ It now occurs to me that it would be probably better to use a less abstract name for the flag, like SHF_GNU_NOSTRIP (I'm not that good on naming stuff.) :)
Re: [PATCH] strip: keep .ctf section in stripped file
> Hi Guys, > > I apologise - I know that Jose CC'ed me on this topic, but I have lost > the email, so I am creating this post instead. > >> What about using an OS-specific section flag in elf.h, something like: >> #define SHF_GNU_PERSISTENT 0x0ff1 /* Section must not be >> stripped. */ > > I rather like this idea. It would certainly make things easier. > Assuming that is that there is a convenient way to set the flag. Ie > we need to teach the assembler about it, and probably the linker too > (so that it can be used in linker scripts). > > It might also be a nice idea to have a complimentary flag to indicate > sections that expect to be stripped, even if they might otherwise be > retained. For example: > > #define SHF_GNU_TREAT_AS_DEBUG 0x0ff2 /* Treat as a debug info >section. Strip with other debug sections and move to a separate >debug info file when creating such files. */ > > This could be added to note sections for example, where the notes > will not be needed at run-time. I am, _very slowly_, working on this. Along with other 100 gazillion things to do, but unless someone else beats me to it, I will send patches at some point.
