Re: [PATCH] [SH] ABI baseline update

[Thomas Schwinge <thomas at codesourcery dot com>]

Hi!

On Tue, 29 May 2012 22:41:09 +0900, Kaz Kojima <kkojima@rr.iij4u.or.jp> wrote:
> The attached patch is to update the abi baseline files for SH.
> Only checked for cross sh4-unknown-linux-gnu with "make check-abi".
> Thomas, could you check if it works for you?

I get very similar results.  I've also been checking the changes against
other architectures' files, though I have not chased down every single
difference.  I primarily used i386's, where I modified its *.abilist
files by moving everything from before GLIBC_2.2 into GLIBC_2.2, then
filtered all of GLIBC_2.2 through Âsort | uniq to minimize these
differences.

> diff -upN ORIG/libc/sysdeps/unix/sysv/linux/sh/nptl/ld.abilist libc/sysdeps/unix/sysv/linux/sh/nptl/ld.abilist
> --- ORIG/libc/sysdeps/unix/sysv/linux/sh/nptl/ld.abilist	1970-01-01 09:00:00.000000000 +0900
> +++ libc/sysdeps/unix/sysv/linux/sh/nptl/ld.abilist	2012-05-29 22:21:20.000000000 +0900

Same for me.  Generally looks good.

> @@ -0,0 +1,16 @@
> +GLIBC_2.2
> + GLIBC_2.2 A
> + __libc_memalign F
> + __libc_stack_end D 0x4
> + _dl_mcount F
> + _r_debug D 0x14
> + calloc F
> + free F
> + malloc F
> + realloc F
> +GLIBC_2.3
> + GLIBC_2.3 A
> + __tls_get_addr F
> +GLIBC_2.4
> + GLIBC_2.4 A
> + __stack_chk_guard D 0x4

What other architectures' files do not have is the __stack_chk_guard
symbol -- why is SH being special there?

> diff -upN ORIG/libc/sysdeps/unix/sysv/linux/sh/nptl/libc.abilist libc/sysdeps/unix/sysv/linux/sh/nptl/libc.abilist
> --- ORIG/libc/sysdeps/unix/sysv/linux/sh/nptl/libc.abilist	2012-05-29 21:37:37.000000000 +0900
> +++ libc/sysdeps/unix/sysv/linux/sh/nptl/libc.abilist	2012-05-29 22:18:43.000000000 +0900

Very similar for me.

> @@ -50,9 +50,7 @@ GLIBC_2.12
>   sys_nerr D 0x4
>  GLIBC_2.13
>   GLIBC_2.13 A
> - __fentry__ F
>   fanotify_init F
> - fanotify_mark F
>   prlimit F
>   prlimit64 F
>  GLIBC_2.14

(I'm seeing this, too.)  I have not yet completely researched why/where
we're losing these symbols (and if that is worrysome or not), but i386
still has them in GLIBC_2.13.

i386 and x86_64 define __fentry__ in their mcount implementations, so it
perhaps is an error that they originally were present in SH's
libc.abilist (and also in x390's, by the way).

fanotify_mark can be found in several syscalls.list files, but not in one
that SH uses (which might be wrong or might be correct).

> @@ -503,7 +503,7 @@ GLIBC_2.2
>   _longjmp F
>   _mcleanup F
>   _mcount F
> - _nl_default_dirname D 0x20
> + _nl_default_dirname D 0x19
>   _nl_domain_bindings D 0x4
>   _nl_msg_cat_cntr D 0x4
>   _null_auth D 0xc

According to
<http://sourceware.org/glibc/wiki/Testing/Testsuite#ABI_check>, we should
be using Â_nl_default_dirname D 0x12Â (which also happens to match the
value that I get).

Additionally, I have a Â__fpscr_values D 0x8Â for GLIBC_2.2 which I also
yet have to trace down why I have it and you don't.

> diff -upN ORIG/libc/sysdeps/unix/sysv/linux/sh/nptl/libm.abilist libc/sysdeps/unix/sysv/linux/sh/nptl/libm.abilist
> --- ORIG/libc/sysdeps/unix/sysv/linux/sh/nptl/libm.abilist	2012-05-29 21:37:37.000000000 +0900
> +++ libc/sysdeps/unix/sysv/linux/sh/nptl/libm.abilist	2012-05-29 22:18:32.000000000 +0900

Same for me, probably good.

> @@ -4,83 +4,56 @@ GLIBC_2.15
>   __acosf_finite F
>   __acosh_finite F
>   __acoshf_finite F
> - __acoshl_finite F
> - __acosl_finite F
>   __asin_finite F
>   __asinf_finite F
> - __asinl_finite F
> [...]

Given these are prefixed with two underscores (and thus internal?) it's
probably fine to have them disappear?

> diff -upN ORIG/libc/sysdeps/unix/sysv/linux/sh/nptl/librt.abilist libc/sysdeps/unix/sysv/linux/sh/nptl/librt.abilist
> --- ORIG/libc/sysdeps/unix/sysv/linux/sh/nptl/librt.abilist	2012-05-29 21:37:37.000000000 +0900
> +++ libc/sysdeps/unix/sysv/linux/sh/nptl/librt.abilist	2012-05-29 22:18:38.000000000 +0900

Same for me, looks good.


GrÃÃe,
 Thomas

Attachment: pgp00000.pgp
Description: PGP signature