public inbox for binutils@sourceware.org
 help / color / mirror / Atom feed
From: "Martin Liška" <mliska@suse.cz>
To: Fangrui Song <maskray@google.com>
Cc: binutils@sourceware.org, Alan Modra <amodra@gmail.com>
Subject: Re: [PATCH] readelf: support zstd compressed debug sections [PR 29640]
Date: Fri, 14 Oct 2022 10:26:54 +0200	[thread overview]
Message-ID: <5eeed442-a153-5805-eae9-44b006506fd1@suse.cz> (raw)
In-Reply-To: <56a23340-a416-0ac9-dc17-4be601edacf7@suse.cz>

On 10/14/22 09:47, Martin Liška wrote:
> On 10/14/22 05:35, Fangrui Song wrote:
>> On 2022-10-13, Martin Liška wrote:
>>> On 10/1/22 08:20, Fangrui Song wrote:
>>>> ---
>>>>  binutils/Makefile.am |   6 +--
>>>>  binutils/Makefile.in |   6 +--
>>>>  binutils/readelf.c   | 112 +++++++++++++++++++++++++++----------------
>>>>  3 files changed, 78 insertions(+), 46 deletions(-)
>>>>
>>>> diff --git a/binutils/Makefile.am b/binutils/Makefile.am
>>>> index 751fbacce12..b249af721ae 100644
>>>> --- a/binutils/Makefile.am
>>>> +++ b/binutils/Makefile.am
>>>> @@ -54,8 +54,8 @@ DEBUGINFOD_LIBS = @DEBUGINFOD_LIBS@
>>>>  WARN_CFLAGS = @WARN_CFLAGS@
>>>>  WARN_CFLAGS_FOR_BUILD = @WARN_CFLAGS_FOR_BUILD@
>>>>  NO_WERROR = @NO_WERROR@
>>>> -AM_CFLAGS = $(WARN_CFLAGS) $(ZLIBINC)
>>>> -AM_CFLAGS_FOR_BUILD = $(WARN_CFLAGS_FOR_BUILD) $(ZLIBINC)
>>>> +AM_CFLAGS = $(WARN_CFLAGS) $(ZLIBINC) $(ZSTD_CFLAGS)
>>>> +AM_CFLAGS_FOR_BUILD = $(WARN_CFLAGS_FOR_BUILD) $(ZLIBINC) $(ZSTD_CFLAGS)
>>>>  LIBICONV = @LIBICONV@
>>>>
>>>>  # these two are almost the same program
>>>> @@ -256,7 +256,7 @@ objcopy_SOURCES = objcopy.c not-strip.c rename.c $(WRITE_DEBUG_SRCS) $(BULIBS)
>>>>  strings_SOURCES = strings.c $(BULIBS)
>>>>
>>>>  readelf_SOURCES = readelf.c version.c unwind-ia64.c dwarf.c demanguse.c $(ELFLIBS)
>>>> -readelf_LDADD   = $(LIBCTF_NOBFD) $(LIBINTL) $(LIBIBERTY) $(ZLIB) $(DEBUGINFOD_LIBS) $(MSGPACK_LIBS)
>>>> +readelf_LDADD   = $(LIBCTF_NOBFD) $(LIBINTL) $(LIBIBERTY) $(ZLIB) $(ZSTD_LIBS) $(DEBUGINFOD_LIBS) $(MSGPACK_LIBS)
>>>>
>>>>  elfedit_SOURCES = elfedit.c version.c $(ELFLIBS)
>>>>  elfedit_LDADD = $(LIBINTL) $(LIBIBERTY)
>>>> diff --git a/binutils/Makefile.in b/binutils/Makefile.in
>>>> index 6de4e239408..7d9039e0f2f 100644
>>>> --- a/binutils/Makefile.in
>>>> +++ b/binutils/Makefile.in
>>>> @@ -651,8 +651,8 @@ am__skipyacc =
>>>>  # case both are empty.
>>>>  ZLIB = @zlibdir@ -lz
>>>>  ZLIBINC = @zlibinc@
>>>> -AM_CFLAGS = $(WARN_CFLAGS) $(ZLIBINC)
>>>> -AM_CFLAGS_FOR_BUILD = $(WARN_CFLAGS_FOR_BUILD) $(ZLIBINC)
>>>> +AM_CFLAGS = $(WARN_CFLAGS) $(ZLIBINC) $(ZSTD_CFLAGS)
>>>> +AM_CFLAGS_FOR_BUILD = $(WARN_CFLAGS_FOR_BUILD) $(ZLIBINC) $(ZSTD_CFLAGS)
>>>>
>>>>  # these two are almost the same program
>>>>  AR_PROG = ar
>>>> @@ -790,7 +790,7 @@ size_SOURCES = size.c $(BULIBS)
>>>>  objcopy_SOURCES = objcopy.c not-strip.c rename.c $(WRITE_DEBUG_SRCS) $(BULIBS)
>>>>  strings_SOURCES = strings.c $(BULIBS)
>>>>  readelf_SOURCES = readelf.c version.c unwind-ia64.c dwarf.c demanguse.c $(ELFLIBS)
>>>> -readelf_LDADD = $(LIBCTF_NOBFD) $(LIBINTL) $(LIBIBERTY) $(ZLIB) $(DEBUGINFOD_LIBS) $(MSGPACK_LIBS)
>>>> +readelf_LDADD = $(LIBCTF_NOBFD) $(LIBINTL) $(LIBIBERTY) $(ZLIB) $(ZSTD_LIBS) $(DEBUGINFOD_LIBS) $(MSGPACK_LIBS)
>>>>  elfedit_SOURCES = elfedit.c version.c $(ELFLIBS)
>>>>  elfedit_LDADD = $(LIBINTL) $(LIBIBERTY)
>>>>  strip_new_SOURCES = objcopy.c is-strip.c rename.c $(WRITE_DEBUG_SRCS) $(BULIBS)
>>>> diff --git a/binutils/readelf.c b/binutils/readelf.c
>>>> index 351571c8abb..04cda213f33 100644
>>>> --- a/binutils/readelf.c
>>>> +++ b/binutils/readelf.c
>>>> @@ -44,6 +44,9 @@
>>>>  #include <assert.h>
>>>>  #include <time.h>
>>>>  #include <zlib.h>
>>>> +#ifdef HAVE_ZSTD
>>>> +#include <zstd.h>
>>>> +#endif
>>>>  #include <wchar.h>
>>>>
>>>>  #if defined HAVE_MSGPACK
>>>> @@ -15170,48 +15173,56 @@ get_section_contents (Elf_Internal_Shdr * section, Filedata * filedata)
>>>>                               _("section contents"));
>>>>  }
>>>>
>>>> -/* Uncompresses a section that was compressed using zlib, in place.  */
>>>> +/* Uncompresses a section that was compressed using zlib/zstd, in place.  */
>>>>
>>>>  static bool
>>>> -uncompress_section_contents (unsigned char **buffer,
>>>> -                 uint64_t uncompressed_size,
>>>> -                 uint64_t *size)
>>>> +uncompress_section_contents (bool is_zstd, unsigned char **buffer,
>>>> +                 uint64_t uncompressed_size, uint64_t *size)
>>>>  {
>>>>    uint64_t compressed_size = *size;
>>>>    unsigned char *compressed_buffer = *buffer;
>>>> -  unsigned char *uncompressed_buffer;
>>>> +  unsigned char *uncompressed_buffer = xmalloc (uncompressed_size);
>>>>    z_stream strm;
>>>>    int rc;
>>>>
>>>> -  /* It is possible the section consists of several compressed
>>>> -     buffers concatenated together, so we uncompress in a loop.  */
>>>> -  /* PR 18313: The state field in the z_stream structure is supposed
>>>> -     to be invisible to the user (ie us), but some compilers will
>>>> -     still complain about it being used without initialisation.  So
>>>> -     we first zero the entire z_stream structure and then set the fields
>>>> -     that we need.  */
>>>> -  memset (& strm, 0, sizeof strm);
>>>> -  strm.avail_in = compressed_size;
>>>> -  strm.next_in = (Bytef *) compressed_buffer;
>>>> -  strm.avail_out = uncompressed_size;
>>>> -  uncompressed_buffer = (unsigned char *) xmalloc (uncompressed_size);
>>>> +  if (is_zstd)
>>>> +    {
>>>> +#ifdef HAVE_ZSTD
>>>> +      size_t ret = ZSTD_decompress (uncompressed_buffer, uncompressed_size,
>>>> +                    compressed_buffer, compressed_size);
>>>> +      if (ZSTD_isError (ret))
>>>> +    goto fail;
>>>> +#endif
>>>> +    }
>>>> +  else
>>>> +    {
>>>> +      /* It is possible the section consists of several compressed
>>>> +     buffers concatenated together, so we uncompress in a loop.  */
>>>> +      /* PR 18313: The state field in the z_stream structure is supposed
>>>> +     to be invisible to the user (ie us), but some compilers will
>>>> +     still complain about it being used without initialisation.  So
>>>> +     we first zero the entire z_stream structure and then set the fields
>>>> +     that we need.  */
>>>> +      memset (&strm, 0, sizeof strm);
>>>> +      strm.avail_in = compressed_size;
>>>> +      strm.next_in = (Bytef *)compressed_buffer;
>>>> +      strm.avail_out = uncompressed_size;
>>>>
>>>> -  rc = inflateInit (& strm);
>>>> -  while (strm.avail_in > 0)
>>>> -    {
>>>> -      if (rc != Z_OK)
>>>> -        break;
>>>> -      strm.next_out = ((Bytef *) uncompressed_buffer
>>>> -                       + (uncompressed_size - strm.avail_out));
>>>> -      rc = inflate (&strm, Z_FINISH);
>>>> -      if (rc != Z_STREAM_END)
>>>> -        break;
>>>> -      rc = inflateReset (& strm);
>>>> +      rc = inflateInit (&strm);
>>>> +      while (strm.avail_in > 0)
>>>> +    {
>>>> +      if (rc != Z_OK)
>>>> +        break;
>>>> +      strm.next_out = ((Bytef *)uncompressed_buffer
>>>> +               + (uncompressed_size - strm.avail_out));
>>>> +      rc = inflate (&strm, Z_FINISH);
>>>> +      if (rc != Z_STREAM_END)
>>>> +        break;
>>>> +      rc = inflateReset (&strm);
>>>> +    }
>>>> +      if (inflateEnd (&strm) != Z_OK || rc != Z_OK || strm.avail_out != 0)
>>>> +    goto fail;
>>>>      }
>>>> -  if (inflateEnd (& strm) != Z_OK
>>>> -      || rc != Z_OK
>>>> -      || strm.avail_out != 0)
>>>> -    goto fail;
>>>>
>>>>    *buffer = uncompressed_buffer;
>>>>    *size = uncompressed_size;
>>>> @@ -15254,6 +15265,7 @@ dump_section_as_strings (Elf_Internal_Shdr * section, Filedata * filedata)
>>>>      {
>>>>        uint64_t new_size = num_bytes;
>>>>        uint64_t uncompressed_size = 0;
>>>> +      bool is_zstd = false;
>>>>
>>>>        if ((section->sh_flags & SHF_COMPRESSED) != 0)
>>>>      {
>>>> @@ -15266,7 +15278,13 @@ dump_section_as_strings (Elf_Internal_Shdr * section, Filedata * filedata)
>>>>             by get_compression_header.  */
>>>>          goto error_out;
>>>>
>>>> -      if (chdr.ch_type != ELFCOMPRESS_ZLIB)
>>>> +      if (chdr.ch_type == ELFCOMPRESS_ZLIB)
>>>> +        ;
>>>> +#ifdef HAVE_ZSTD
>>>> +      else if (chdr.ch_type == ELFCOMPRESS_ZSTD)
>>>> +        is_zstd = true;
>>>> +#endif
>>>> +      else
>>>>          {
>>>>            warn (_("section '%s' has unsupported compress type: %d\n"),
>>>>              printable_section_name (filedata, section), chdr.ch_type);
>>>> @@ -15295,8 +15313,8 @@ dump_section_as_strings (Elf_Internal_Shdr * section, Filedata * filedata)
>>>>
>>>>        if (uncompressed_size)
>>>>      {
>>>> -      if (uncompress_section_contents (& start,
>>>> -                       uncompressed_size, & new_size))
>>>> +      if (uncompress_section_contents (is_zstd, &start, uncompressed_size,
>>>> +                       &new_size))
>>>>          num_bytes = new_size;
>>>>        else
>>>>          {
>>>> @@ -15470,6 +15488,7 @@ dump_section_as_bytes (Elf_Internal_Shdr *section,
>>>>      {
>>>>        uint64_t new_size = section_size;
>>>>        uint64_t uncompressed_size = 0;
>>>> +      bool is_zstd = false;
>>>>
>>>>        if ((section->sh_flags & SHF_COMPRESSED) != 0)
>>>>      {
>>>> @@ -15482,7 +15501,13 @@ dump_section_as_bytes (Elf_Internal_Shdr *section,
>>>>             by get_compression_header.  */
>>>>          goto error_out;
>>>>
>>>> -      if (chdr.ch_type != ELFCOMPRESS_ZLIB)
>>>> +      if (chdr.ch_type == ELFCOMPRESS_ZLIB)
>>>> +        ;
>>>> +#ifdef HAVE_ZSTD
>>>> +      else if (chdr.ch_type == ELFCOMPRESS_ZSTD)
>>>> +        is_zstd = true;
>>>> +#endif
>>>> +      else
>>>>          {
>>>>            warn (_("section '%s' has unsupported compress type: %d\n"),
>>>>              printable_section_name (filedata, section), chdr.ch_type);
>>>> @@ -15511,8 +15536,8 @@ dump_section_as_bytes (Elf_Internal_Shdr *section,
>>>>
>>>>        if (uncompressed_size)
>>>>      {
>>>> -      if (uncompress_section_contents (& start, uncompressed_size,
>>>> -                       & new_size))
>>>> +      if (uncompress_section_contents (is_zstd, &start, uncompressed_size,
>>>> +                       &new_size))
>>>>          {
>>>>            section_size = new_size;
>>>>          }
>>>> @@ -15848,6 +15873,7 @@ load_specific_debug_section (enum dwarf_section_display_enum  debug,
>>>>        unsigned char *start = section->start;
>>>>        uint64_t size = sec->sh_size;
>>>>        uint64_t uncompressed_size = 0;
>>>> +      bool is_zstd = false;
>>>>
>>>>        if ((sec->sh_flags & SHF_COMPRESSED) != 0)
>>>>      {
>>>> @@ -15869,7 +15895,13 @@ load_specific_debug_section (enum dwarf_section_display_enum  debug,
>>>>             by get_compression_header.  */
>>>>          return false;
>>>>
>>>> -      if (chdr.ch_type != ELFCOMPRESS_ZLIB)
>>>> +      if (chdr.ch_type == ELFCOMPRESS_ZLIB)
>>>> +        ;
>>>> +#ifdef HAVE_ZSTD
>>>> +      else if (chdr.ch_type == ELFCOMPRESS_ZSTD)
>>>> +        is_zstd = true;
>>>> +#endif
>>>> +      else
>>>>          {
>>>>            warn (_("section '%s' has unsupported compress type: %d\n"),
>>>>              section->name, chdr.ch_type);
>>>> @@ -15898,7 +15930,7 @@ load_specific_debug_section (enum dwarf_section_display_enum  debug,
>>>>
>>>>        if (uncompressed_size)
>>>>      {
>>>> -      if (uncompress_section_contents (&start, uncompressed_size,
>>>> +      if (uncompress_section_contents (is_zstd, &start, uncompressed_size,
>>>>                         &size))
>>>>          {
>>>>            /* Free the compressed buffer, update the section buffer
>>>
>>> Hi.
>>>
>>> I noticed the following issue (dunno if caused by objcopy or the patches readelf) with this patch:
>>>
>>> $ echo 'int main() {return 0;}' | gcc -O2 -g -c -o a.o -x c -
>>> $ /home/marxin/Programming/binutils/objdir/ld/../binutils/objcopy -O elf32-x86-64 a.o a-x32.o --compress-debug-sections=zstd
>>> $ /home/marxin/Programming/binutils/objdir/binutils/readelf -wi a-x32.o >/dev/null
>>> readelf: Error: Unable to decompress section .debug_info
>>> readelf: Error: No comp units in .debug_info section ?
>>>
>>> while zlib is fine:
>>>
>>> $ /home/marxin/Programming/binutils/objdir/ld/../binutils/objcopy -O elf32-x86-64 a.o a-x32.o --compress-debug-sections=zlib
>>> $ /home/marxin/Programming/binutils/objdir/binutils/readelf -wi a-x32.o >/dev/null
>>>
>>> Can you please take a look?
>>
>> objcopy -O elf32-x86-64 a.o a-x32.o --compress-debug-sections=zstd uses
>> ch_type=1 (ELFCOMPRESS_ZLIB), but I don't know where goes wrong.
>>
>> Converting EI_CLASS looks very hacky to me.
> 
> Ok, I've got it. It's unrelated to '-O elf32-x86-64', but to the fact we move from zlib-gabi compression of a.o
> to zstd for the objcopy output file.
> 
> Simpler reproducer:
> 
> $ echo 'int main() {return 0;}' | gcc -O2 -g -c -o a.o -x c - -Wa,--compress-debug-sections=zlib
> $ /home/marxin/Programming/binutils/objdir//binutils/objcopy a.o a2.o --compress-debug-sections=zstd
> $ /home/marxin/Programming/binutils/objdir/binutils/readelf -wi a2.o
> readelf: Error: Unable to decompress section .debug_info
> readelf: Error: No comp units in .debug_info section ?
> 
> As seen the .debug_info section in a2.o claims to be compressed:
> 
> $ readelf -SW a2.o | grep C
>   [ 5] .debug_info       PROGBITS        0000000000000000 000043 000050 00   C  0   0  1
> 
> But it's not as it's not beneficial and the following code in in bfd/compress.c is taken:
> 
>   /* If compression didn't make the section smaller, keep it uncompressed.  */
>   if (compressed_size >= uncompressed_size)
>     {
>       memcpy (buffer, input_buffer, uncompressed_size);
>       sec->compress_status = COMPRESS_SECTION_NONE;
>     }
> 
> Apparently, we miss removal of the compression header :/ Moreover, the following is fine:
> $ /home/marxin/Programming/binutils/objdir//binutils/objcopy a.o a2.o --compress-debug-sections=none
> $ /home/marxin/Programming/binutils/objdir/binutils/readelf -wi a2.o >/dev/null
> 
> Lemme try preparing a fix.
> 
> Martin

There's a patch candidate that works for the simple example (all x86_64 ELF), but fails for:
objcopy -O elf32-x86-64 a.o a-x32.o --compress-debug-sections=zstd

likely due to bfd_update_compression_header where compression header is updated.

diff --git a/bfd/compress.c b/bfd/compress.c
index 364df14142b..262e2f5bed2 100644
--- a/bfd/compress.c
+++ b/bfd/compress.c
@@ -202,8 +202,10 @@ bfd_compress_section_contents (bfd *abfd, sec_ptr sec)
       compressed_size += new_header_size;
     }
 
-  /* If compression didn't make the section smaller, keep it uncompressed.  */
-  if (compressed_size >= uncompressed_size)
+  /* If compression didn't make the section smaller, keep it uncompressed.
+     Do it only if the original section is not compressed, otherwise we would
+     need to drop the section header.  */
+  if (!compressed && compressed_size >= uncompressed_size)
     {
       memcpy (buffer, input_buffer, uncompressed_size);
       sec->compress_status = COMPRESS_SECTION_NONE;

@Alan: Can you please help us?

Thanks,
Martin


  reply	other threads:[~2022-10-14  8:26 UTC|newest]

Thread overview: 14+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-10-01  6:20 Fangrui Song
2022-10-06  2:20 ` Fangrui Song
2022-10-20  8:04   ` Martin Liška
2022-10-20  8:05   ` Martin Liška
2022-10-13 11:38 ` Martin Liška
2022-10-14  3:35   ` Fangrui Song
2022-10-14  7:47     ` Martin Liška
2022-10-14  8:26       ` Martin Liška [this message]
2022-10-14 11:28         ` Alan Modra
2022-10-14 12:09           ` Alan Modra
2022-10-16  4:42             ` Alan Modra
2022-10-16 20:46               ` Fangrui Song
2022-10-21  9:16                 ` Alan Modra
2022-10-17  7:52               ` Martin Liška

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=5eeed442-a153-5805-eae9-44b006506fd1@suse.cz \
    --to=mliska@suse.cz \
    --cc=amodra@gmail.com \
    --cc=binutils@sourceware.org \
    --cc=maskray@google.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).