public inbox for libc-alpha@sourceware.org
 help / color / mirror / Atom feed
From: Adhemerval Zanella <adhemerval.zanella@linaro.org>
To: Florian Weimer <fweimer@redhat.com>
Cc: libc-alpha@sourceware.org
Subject: Re: [PATCH v4 3/3] csu: Implement and use _dl_early_allocate during static startup
Date: Mon, 9 May 2022 13:52:54 -0300	[thread overview]
Message-ID: <b8bd67ad-4a3f-bc91-287d-080cf1a617c0@linaro.org> (raw)
In-Reply-To: <871qx2adqe.fsf@oldenburg.str.redhat.com>



On 09/05/2022 13:10, Florian Weimer wrote:
> * Adhemerval Zanella:
> 
>>> diff --git a/scripts/tst-elf-edit.py b/scripts/tst-elf-edit.py
>>> index a514179bbf..0e19ce1e73 100644
>>> --- a/scripts/tst-elf-edit.py
>>> +++ b/scripts/tst-elf-edit.py
> 
>>> @@ -172,24 +181,35 @@ def elf_edit(f, align):
>>>  
>>>      ehdr = Elf_Ehdr(e_ident)
>>>      ehdr.read(f)
>>> -    if ehdr.e_type != ET_DYN:
>>> -       error('{}: not a shared library'.format(f.name))
>>> +    if ehdr.e_type not in (ET_EXEC, ET_DYN):
>>> +       error('{}: not an executable or shared library'.format(f.name))
>>>  
>>>      phdr = Elf_Phdr(e_ident)
>>> +    maximize_tls_size_done = False
>>>      for i in range(0, ehdr.e_phnum):
>>>          f.seek(ehdr.e_phoff + i * phdr.len)
>>>          phdr.read(f)
>>> -        if phdr.p_type == PT_LOAD:
>>> -            elf_edit_align(phdr, align)
>>> +        if phdr.p_type == PT_LOAD and opts.align is not None:
>>
>> I think you can omit the None check.
> 
> I think this will raise an exception in elf_edit_align because the -a
> argument is now optional.
> 
>>> +void *
>>> +_dl_early_allocate (size_t size)
>>> +{
>>> +  void *result;
>>> +
>>> +  if (__curbrk != NULL)
>>> +    /* If the break has been initialized, brk must have run before,
>>> +       so just call it once more.  */
>>> +    {
>>> +      result = __sbrk (size);
>>> +      if (result == (void *) -1)
>>> +        result = NULL;
>>> +    }
>>> +  else
>>> +    {
>>> +      /* If brk has not been invoked, there is no need to update
>>> +         __curbrk.  The first call to brk will take care of that.  */
>>> +      void *previous = (void *) INTERNAL_SYSCALL_CALL (brk, 0);
>>> +      result = (void *) INTERNAL_SYSCALL_CALL (brk, previous + size);
>>> +      if (result == previous)
>>> +        result = NULL;
>>> +      else
>>> +        result = previous;
>>
>> You will need to factor it be arch-specific since alpha return -ENOMEM
>> in case of failure and sparc has different calling convention (similar 
>> to what it does for clone).  Maybe add a
>>
>> static inline void *
>> brk_call (void *addr)
>> {
>>   void *r = syscall;
>>   r = check_error (r) ? -1 : 0;
>> }
>>
>> And then refactor Linux brk.c versions to use brk_call as well.
> 
> Thanks.  I'm testing a new version with this change.
> 
>>> +  /* If brk fails, fall back to mmap.  This can happen due to
>>> +     unfortunate ASLR layout decisions and kernel bugs, particularly
>>> +     for static PIE.  */
>>> +  if (result == NULL)
>>> +    {
>>> +      long int ret;
>>> +      int prot = PROT_READ | PROT_WRITE;
>>> +      int flags = MAP_PRIVATE | MAP_ANONYMOUS;
>>> +#ifdef __NR_mmap2
>>> +      ret = MMAP_CALL_INTERNAL (mmap2, 0, size, prot, flags, -1, 0);
>>> +#else
>>> +      ret = MMAP_CALL_INTERNAL (mmap, 0, size, prot, flags, -1, 0);
>>> +#endif
>>> +      if (INTERNAL_SYSCALL_ERROR_P (ret))
>>
>> Maybe move it to mmap_call.h and make it a static inline:
>>
>> static inline void *
>> mmap64_call (void *addr, size_t len, int prot, int flags, int fd, 
>> 	   off64_t offset)
>> {
>>   long int ret;
>> #ifdef __NR_mmap2
>>   ret = MMAP_CALL_INTERNAL (mmap2, addr, len, prot, flags, fd,
>> 			    (off_t) (offset / MMAP2_PAGE_UNIT));
>> #else
>>   ret = MMAP_CALL_INTERNAL (mmap, addr, len, prot, flags, fd,
>> 			    offset);
>> #endif
>>   return INTERNAL_SYSCALL_ERROR_P (ret) ? NULL : (void *) ret;
>> }
>>
>> static inline void *
>> mmap_call_internal (size_t len)
>> {
>>   int prot = PROT_READ | PROT_WRITE;
>>   int flags = MAP_PRIVATE | MAP_ANONYMOUS;
>>   return mmap64_call (0, len, PROT_READ | PROT_WRITE,
>> 		      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
>> }
> 
> We have several customization points for mmap and mmap2.
> MMAP2_PAGE_UNIT can expand to a static global variable page_unit.  This
> is difficult to encapsulate properly in an inline function.  Adding
> __brk_call was simple enough, but this looks way more complex.

The idea is to allow __mmap64 use the mmap64_call and have the syscall
logic in one place.  Something like:

void *
__mmap64 (void *addr, size_t len, int prot, int flags, int fd, off64_t offset)
{
  MMAP_CHECK_PAGE_UNIT ();

  if (offset & MMAP_OFF_MASK)
    return (void *) INLINE_SYSCALL_ERROR_RETURN_VALUE (EINVAL);

  MMAP_PREPARE (addr, len, prot, flags, fd, offset);
  return mmap64_call (addr, len, prot, flags, fd, offset);
}


  reply	other threads:[~2022-05-09 16:52 UTC|newest]

Thread overview: 17+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-05-05 15:06 [PATCH v4 1/3] ia64: Always define IA64_USE_NEW_STUB as a flag macro Florian Weimer
2022-05-05 15:07 ` [PATCH v4 2/3] Linux: Implement a useful version of _startup_fatal Florian Weimer
2022-05-05 15:28   ` Andreas Schwab
2022-05-05 15:43     ` Florian Weimer
2022-05-05 17:29   ` Adhemerval Zanella
2022-05-05 15:08 ` [PATCH v4 3/3] csu: Implement and use _dl_early_allocate during static startup Florian Weimer
2022-05-05 17:48   ` Adhemerval Zanella
2022-05-05 18:03     ` Florian Weimer
2022-05-05 18:25       ` Adhemerval Zanella
2022-05-06 10:00         ` Florian Weimer
2022-05-06 13:04           ` Adhemerval Zanella
2022-05-06 13:19   ` Adhemerval Zanella
2022-05-09 16:10     ` Florian Weimer
2022-05-09 16:52       ` Adhemerval Zanella [this message]
2022-05-09 17:08         ` Florian Weimer
2022-05-09 17:37           ` Adhemerval Zanella
2022-05-05 17:28 ` [PATCH v4 1/3] ia64: Always define IA64_USE_NEW_STUB as a flag macro Adhemerval Zanella

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=b8bd67ad-4a3f-bc91-287d-080cf1a617c0@linaro.org \
    --to=adhemerval.zanella@linaro.org \
    --cc=fweimer@redhat.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).