public inbox for libc-alpha@sourceware.org
 help / color / mirror / Atom feed
From: Fangrui Song <maskray@google.com>
To: "H.J. Lu" <hjl.tools@gmail.com>
Cc: GNU C Library <libc-alpha@sourceware.org>
Subject: Re: [PATCH] i386: Port elf_machine_{load_address, dynamic} from x86-64
Date: Thu, 23 Sep 2021 23:55:08 -0700	[thread overview]
Message-ID: <20210924065508.ewyzqjczfx4coipf@google.com> (raw)
In-Reply-To: <CAMe9rOoZp63=9FAiVggKoOnRhfmXZEU-q-LT_NZ563NmJM1HHg@mail.gmail.com>


On 2021-09-23, H.J. Lu wrote:
>On Thu, Sep 23, 2021 at 9:26 PM Fangrui Song <maskray@google.com> wrote:
>>
>> This drops reliance on _GLOBAL_OFFSET_TABLE_[0] being the link-time
>> address of _DYNAMIC.
>>
>> The code sequence length does not change.
>> ---
>>  sysdeps/i386/dl-machine.h | 29 +++++++++++------------------
>>  1 file changed, 11 insertions(+), 18 deletions(-)
>>
>> diff --git a/sysdeps/i386/dl-machine.h b/sysdeps/i386/dl-machine.h
>> index 590b41d8d7..9f0eeaf66e 100644
>> --- a/sysdeps/i386/dl-machine.h
>> +++ b/sysdeps/i386/dl-machine.h
>> @@ -34,27 +34,20 @@ elf_machine_matches_host (const Elf32_Ehdr *ehdr)
>>  }
>>
>>
>> -/* Return the link-time address of _DYNAMIC.  Conveniently, this is the
>> -   first element of the GOT, a special entry that is never relocated.  */
>> -static inline Elf32_Addr __attribute__ ((unused, const))
>> -elf_machine_dynamic (void)
>> -{
>> -  /* This produces a GOTOFF reloc that resolves to zero at link time, so in
>> -     fact just loads from the GOT register directly.  By doing it without
>> -     an asm we can let the compiler choose any register.  */
>> -  extern const Elf32_Addr _GLOBAL_OFFSET_TABLE_[] attribute_hidden;
>> -  return _GLOBAL_OFFSET_TABLE_[0];
>> -}
>> -
>>  /* Return the run-time load address of the shared object.  */
>> -static inline Elf32_Addr __attribute__ ((unused))
>> +static inline ElfW(Addr) __attribute__ ((unused))
>>  elf_machine_load_address (void)
>>  {
>> -  /* Compute the difference between the runtime address of _DYNAMIC as seen
>> -     by a GOTOFF reference, and the link-time address found in the special
>> -     unrelocated first GOT entry.  */
>> -  extern Elf32_Dyn bygotoff[] asm ("_DYNAMIC") attribute_hidden;
>> -  return (Elf32_Addr) &bygotoff - elf_machine_dynamic ();
>> +  extern const ElfW(Ehdr) __ehdr_start attribute_hidden;
>> +  return (ElfW(Addr)) &__ehdr_start;
>> +}
>> +
>> +/* Return the link-time address of _DYNAMIC.  */
>> +static inline ElfW(Addr) __attribute__ ((unused))
>> +elf_machine_dynamic (void)
>> +{
>> +  extern ElfW(Dyn) _DYNAMIC[] attribute_hidden;
>> +  return (ElfW(Addr)) _DYNAMIC - elf_machine_load_address ();
>>  }
>>
>>  /* Set up the loaded object described by L so its unrelocated PLT
>> --
>> 2.33.0.685.g46640cef36-goog
>>
>
>what are the code differences before and after?

long ehdr(void) {
   extern char __ehdr_start[] __attribute__((visibility("hidden")));
   return (long)__ehdr_start;
}

long got(void) {
   extern long  _GLOBAL_OFFSET_TABLE_[] __attribute__((visibility("hidden")));
   return _GLOBAL_OFFSET_TABLE_[0];
}


ehdr:
         call    __x86.get_pc_thunk.ax
         addl    $_GLOBAL_OFFSET_TABLE_, %eax
         leal    __ehdr_start@GOTOFF(%eax), %eax
         ret

got:
         call    __x86.get_pc_thunk.ax
         addl    $_GLOBAL_OFFSET_TABLE_, %eax
         movl    _GLOBAL_OFFSET_TABLE_@GOTOFF(%eax), %eax
         ret


In GCC generated elf/rtld.os, the local code sequence related to __ehdr_start/_GLOBAL_OFFSET_TABLE_
does not change its size, but globally the new code triggers some code motion
and eventually makes the file smaller.

FWIW ld.so:.text is 48 bytes smaller.

New code doesn't have memory load. I guess it may allow GCC to optimize more.

  reply	other threads:[~2021-09-24  6:55 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-09-24  4:26 [PATCH] i386: Port elf_machine_{load_address,dynamic} " Fangrui Song
2021-09-24  6:20 ` [PATCH] i386: Port elf_machine_{load_address, dynamic} " H.J. Lu
2021-09-24  6:55   ` Fangrui Song [this message]
2021-09-24 16:05 ` H.J. Lu

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=20210924065508.ewyzqjczfx4coipf@google.com \
    --to=maskray@google.com \
    --cc=hjl.tools@gmail.com \
    --cc=libc-alpha@sourceware.org \
    /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).