Re: RFC: ABI support for special memory area

["H.J. Lu" <hjl.tools@gmail.com>]

On Fri, Mar 31, 2017 at 6:44 AM, Suprateeka R Hegde
<hegdesmailbox@gmail.com> wrote:
> On 30-Mar-2017 10:10 PM, H.J. Lu wrote:
>>> However, I am just thinking that your earlier approach --
>>> __gnu_mbind_setup -- is better when shared libraries with GNU_MBIND
>>> segments are dlopen'ed. They dont have to iterate all over again to
>>> reach their PHDR. Or what is the recommendation for such dlopen'ed
>>> libraries?
>>
>> It is true that dl_iterate_phdr is called by every shared object, dlopened or
>> not, to locate its own PHDR.
>
> Lets put a one liner on best practices or guideline kind of. You have
> already made it clear in the example code. I am just thinking of putting
> them in words too.
>
> Lets say something like, each load module is expected to process only
> its special memory segments. To mean that shlibs/exe need not do any
> book-keeping to avoid multiple executions of the special memory setup
> for the same load module.
>
>>> And this dl_iterate_phdr(3) not being part of any standards, may change
>>> in a totally incompatible way in the future.
>>>
>>
>> dl_iterate_phdr isn't in any standard.  But it is in glibc.  Given that my
>> proposal is a GNU extension, it isn't a major issue.  Working with
>> existing glibc is a big plus.
>
> Awesome. Looks great. Thanks a lot for the new approach.
>
> --
> Supra

Here is the updated proposal.

Thanks.


-- 
H.J.
--
ABI support for special memory area

To section attributes, add

#define SHF_GNU_MBIND     0x00100000

for sections used to place data or text into a special memory area.
The section names should start with ".mbind" so that they won't be
grouped together with normal sections by link editor.  The sh_info
field indicates the special memory type.  SHF_GNU_MBIND is only
applicable to SHF_ALLOC sections.

The following memory types in the sh_info field are defined:

/* The highest bandwidth memory.   */
#define GNU_MBIND_HBW       0

To the "Program Header" section, add an inclusive range of segment types
for GNU_MBIND segments:

#define PT_GNU_MBIND_NUM    4096
#define PT_GNU_MBIND_LO     (PT_LOOS + 0x474e555)
#define PT_GNU_MBIND_HI     (PT_GNU_MBIND_LO + PT_GNU_MBIND_NUM - 1)

The array element specifies the location and size of a special memory area.
Each GNU_MBIND segment contains one GNU_MBIND section and the segment
type is PT_GNU_MBIND_LO plus the sh_info value.  If the sh_info value is
greater than PT_GNU_MBIND_NUM, no GNU_MBIND segment will be created.  Each
GNU_MBIND segment must be aligned at page boundary.  The interpretation of
the special memory area information is implementation-dependent.
Implementations may ignore GNU_MBIND segment.

Run-time support

Each load module is expected to process only its special memory segments.
There is no need for executable and shared objects to do any book-keeping
to avoid multiple executions of the special memory setup for the same
load module.

dl_iterate_phdr in the the GNU C library:

int dl_iterate_phdr (int (*callback) (struct dl_phdr_info *info,
                                      size_t size, void *data),
                     void *data);

is called via the .init_array section to process GNU_MBIND segments in
executable and shared objects:

static int
callback (struct dl_phdr_info *info, size_t size, void *data)
{
  Compute the load address of the current module.
  if info->dlpi_addr == the load address of the current module
    {
      check ELF program headers and process GNU_MBIND segments
      return 1;
    }

  return 0;
}

static void
call_gnu_mbind_setup (void)
{
  dl_iterate_phdr (callback, NULL);
}

static void (*init_array) (void)
 __attribute__ ((section (".init_array"), used))
 = &call_gnu_mbind_setup;