gmane.comp.lib.glibc.alpha

Re: PATCH: Add x32 dummy sysctl

Chris Metcalf — 2012-05-18T11:48:39

In ports, the linux-generic port does have a definition of sysctl() that
returns -1 with ENOSYS, and includes a stub_warning().  For now this avoids
potential issues from older code that might try to call this API as a
fallback if /proc/sys isn't available (for example).

PATCH: Fold copyright years

H.J. Lu — 2012-05-18T11:17:57

Hi,

I checked in this patch to fold some copyright years.

H.J.
--
diff --git a/ChangeLog b/ChangeLog
index c69543c..ce03a18 100644
--- a/ChangeLog
+++ b/ChangeLog
< at >< at > -1,5 +1,11 < at >< at >
 2012-05-18  H.J. Lu  <hongjiu.lu< at >intel.com>
 
+* sysdeps/unix/sysv/linux/x86_64/sigcontextinfo.h: Fold copyright
+years.
+* sysdeps/unix/sysv/linux/x86_64/sys/debugreg.h: Likewise.
+* sysdeps/unix/sysv/linux/x86_64/sys/procfs.h: Likewise.
+* sysdeps/unix/sysv/linux/x86_64/sys/reg.h: Likewise.
+
 * sysdeps/unix/sysv/linux/x86_64/bits/sigcontext.h: Don't
 include <bits/wordsize.h>.  Check __x86_64__ instead of
 __WORDSIZE.
diff --git a/sysdeps/unix/sysv/linux/x86_64/sigcontextinfo.h b/sysdeps/unix/sysv/linux/x86_64/sigcontextinfo.h
index f122702..c0d894c 100644
--- a/sysdeps/unix/sysv/linux/x86_64/sigcontextinfo.h
+++ b/sysdeps/unix/sysv/linux/x86_64/sigcontextinfo.h
< at >< at > -1,4 +1,4 < at >< at >
-/* Copyright (C) 2001, 2002, 2003, 2012 Free Software Foundation, Inc.
+/* Copyright (C) 2001-2012 Free Software Foundation, Inc.
    This fi

Re: [PATCH 1/4 v3] [SH/SH4]: Add support fedisableexcept, feenableexcept, fegetexcept and feupdateenv

Thomas Schwinge — 2012-05-18T10:54:29

Hi!

This turned into a general analysis of (parts of) SH's math/*.out
results.  Also, perhaps there's a testsuite harness oddity -- seach for
»verbose« in the following text if you want to skip the (probably)
SH-specific bits.


On Thu, 17 May 2012 10:07:00 +0900, Kaz Kojima <kkojima< at >rr.iij4u.or.jp> wrote:

Ho, likewise there might be something wrong in my setup, too.  :-)

If you want, you could send me your math/*.out files, and I could compare
with mine.


Yes, that indeed helps, thanks!

math/test-fenv.out:

    [...]
    -61 errors occurred.
    +6 errors occurred.

Similarly, a lot of failures disappear from math/test-float.out:

    [...]
     Test suite completed:
       3832 test cases plus 3388 tests for exception flags executed.
    -  520 errors occurred.
    +  420 errors occurred.

In the following, [old] means that this was already present without Kaz'
recent patch.  When I say new, this is a recent tree including both
Nobuhiro's patches and Kaz' recent patch.

[old] Several exception-reala

Re: [PATCH] Avoid GOT relocations in ld.so on powerpc

Andreas Schwab — 2012-05-18T10:04:34

This is what I have checked in.

Andreas.

* sysdeps/powerpc/powerpc32/dl-start.S (_dl_start_user): Use
INTUSE on _dl_argv, and _rtld_local instead of _rtld_global.
* sysdeps/powerpc/powerpc32/fpu/__longjmp-common.S [IS_IN_rtld]:
Refer to _rtld_local_ro instead of _rtld_global_ro.
* sysdeps/powerpc/powerpc32/fpu/setjmp-common.S [IS_IN_rtld]:
Likewise.
* sysdeps/powerpc/powerpc64/__longjmp-common.S [IS_IN_rtld]:
Likewise.
* sysdeps/powerpc/powerpc64/setjmp-common.S [IS_IN_rtld]:
Likewise.
* sysdeps/powerpc/powerpc64/dl-trampoline.S [SHARED]: Likewise.
* sysdeps/powerpc/powerpc64/dl-machine.h: Use _rtld_local instead
of _rtld_global, and rtld_progname instead of _dl_argv[0].

diff --git a/sysdeps/powerpc/powerpc32/dl-start.S b/sysdeps/powerpc/powerpc32/dl-start.S
index 2546fa5..b2d0194 100644
--- a/sysdeps/powerpc/powerpc32/dl-start.S
+++ b/sysdeps/powerpc/powerpc32/dl-start.S
< at >< at > -1,6 +1,5 < at >< at >
 /* Machine-dependent ELF startup code.  PowerPC version.
-   Copyright (C) 1995-2000, 2002, 2004, 2005, 2

Re: RFC: manual/contrib.texi

Andreas Jaeger — 2012-05-18T07:52:23

What about adding to the introduction something like this:

Some of the mentioned contributions might not be part of the current glibc, 
still we would like to mention those that contributed.

Andreas

Re: [PATCH] Do not provide inlines for GCC build-in functions

Andreas Jaeger — 2012-05-18T07:06:28

As Uros commented in gcc.gnu.org/bugzilla/show_bug.cgi?id=53190 "As a side 
note, mathinline.h really needs some serious TLC, there are many
functions that are much better implemented with gcc builtin functions, not 
to
mention that x87 asm in 32bit case interferes *badly* with -mfpmath=sse.
"

GCC can do a better job optimization the call if a builtin exists, so let's 
use it.

I'll commit this now,

Thanks for the review,
Andreas

Re: [PATCH] Export AT_HWCAP from libc, v4

Andreas Jaeger — 2012-05-18T07:02:19

I agree, this is fine now.

Richard, let's give others a day more to comment - and if nobody objects 
until Friday night (your timezone ;), go ahead and commit,

Andreas

Re: [PATCH] Optimize strstr, strcasestr and memmem

Ondřej Bílka — 2012-05-18T06:18:22

I posted strstr implementation at this list and got no response. 

I use only SSE2 instructions and my algorithm is faster than using SSE4 instructions.
With trivial modifications I could also use plain arithmetic/AVX2

A trick is to check first two characters or zero terminator in parallel.
This is about 25 times faster on core2 vs glibc one, on i7 its only 3
times faster.

I use these trick for my extension of regular expression engine a
partialy autogenerated strstr follows.

Feel powerful in bed

Elsie Shepard — 2012-05-18T05:15:52

http://cabinasdepintura.com.mx/spoken.html Stimulate your body to impress 
women

Re: PATCH: Use 64-bit anonymouse union on x32 TCB pointer_guard

H.J. Lu — 2012-05-18T04:15:35

Good idea. I changed it in my copy.

Re: [PATCH] Do not provide inlines for GCC build-in functions

Carlos O'Donell — 2012-05-18T04:12:25

Looks good to me.

I double checked that GCC 3.3 doesn't have the builtins and GCC 3.4 does.

I find it a bit lame that we go from a math inline to a builtin behind
the users back, but this seems to be common practice, and I don't see
a better solution.

Cheers,
Carlos.

Re: PATCH: Use 64-bit anonymouse union on x32 TCB pointer_guard

Mike Frysinger — 2012-05-18T04:12:17

probably doesn't matter, but should this be uint64_t ?
-mike

PATCH: Use 64-bit anonymouse union on x32 TCB pointer_guard

H.J. Lu — 2012-05-18T03:52:21

Hi,

This patch uses 64-bit anonymouse union on x32 TCB pointer_guard to
support 64-bit PTR_MANGLE/PTR_DEMANGLE.  I verifed that there is no
change in x86-64 TCB offsets.  Tested on Linux/x32.  OK to install?

Thanks.


H.J.
----
* sysdeps/x86_64/tls.h (tcbhead_t): Use anonymouse union on
pointer_guard.  Reordor fields for x32.

diff --git a/nptl/sysdeps/x86_64/tls.h b/nptl/sysdeps/x86_64/tls.h
index f838916..c593bc8 100644
--- a/nptl/sysdeps/x86_64/tls.h
+++ b/nptl/sysdeps/x86_64/tls.h
< at >< at > -52,17 +52,30 < at >< at > typedef struct
   int gscope_flag;
   uintptr_t sysinfo;
   uintptr_t stack_guard;
+# ifdef __ILP32__
+  int rtld_must_xmm_save;
+  union
+    {
+      /* Needed for 64-bit PTR_MANGLE/PTR_DEMANGLE.  */
+      uintptr_t pointer_guard;
+      int64_t __pointer_guard_word;
+    };
+  unsigned long int vgetcpu_cache[2];
+  /* Reservation of some values for the TM ABI.  */
+  void *__private_tm[5];
+# else
   uintptr_t pointer_guard;
   unsigned long int vgetcpu_cache[2];
-# ifndef __ASSUME_PRIVATE_FUTEX
+#

Re: RFC: Add __pointer_guard_t for pointer_guard

H.J. Lu — 2012-05-18T03:10:09

This also works.

Re: RFC: Add __pointer_guard_t for pointer_guard

H.J. Lu — 2012-05-18T02:42:01

The pointer_guard field in tcbhead_t must be 64-bit for x32
even if the pointer size is 32bit.  Another possibility is to add a
4 byte pad after pointer_guard.  The only drawback is its value
can't be changed.

Re: [PATCH] Optimize strstr, strcasestr and memmem

Maxim Kuvyrkov — 2012-05-18T02:25:47


Thanks for the pointer.  For avoidance of doubt, I've benchmarked the patch on a Core2 machine without SSE4.

I will benchmark the SSE4 implementation against the normal + this patch on short needles.  The benchmark that motivated this patch is libosip message parsing, which heavily uses string functions with small strings.

--
Maxim Kuvyrkov
CodeSourcery / Mentor Graphics

Re: RFC: Add __pointer_guard_t for pointer_guard

Mike Frysinger — 2012-05-18T01:23:25

it's not possible to fix the asm code to work on 32bit values on x32 systems ?  
it kind of sucks we have to add a new type which is uintptr_t for most 
targets.
-mike

RFC: Add __pointer_guard_t for pointer_guard

H.J. Lu — 2012-05-18T01:15:09

As described in

http://sourceware.org/ml/libc-alpha/2012-05/msg01253.html

on x32, PTR_DEMANGLE/PTR_MANGLE are used on 64-bit registers, which
requires the pointer_guard field in tcbhead_t to be 64-bit. But
uintptr_t is 32-bit on x32.  This patch adds __pointer_guard_t for
pointer_guard and reorders tcbhead_t for x32.  By pure luck, stack_guard
and __private_tm offsets used by GCC stay the same.  Any comments?

Thanks.


H.J.

* bits/types.h (__pointer_guard_t): New type.
* bits/typesizes.h (__POINTER_GUARD_T_TYPE): New macro.
sysdeps/mach/hurd/bits/typesizes.h (__POINTER_GUARD_T_TYPE):
Likewise.
* sysdeps/unix/bsd/bsd4.4/freebsd/bits/typesizes.h
(__POINTER_GUARD_T_TYPE): Likewise.
* sysdeps/unix/sysv/linux/s390/bits/typesizes.h
(__POINTER_GUARD_T_TYPE): Likewise.
* sysdeps/unix/sysv/linux/sparc/bits/typesizes.h
(__POINTER_GUARD_T_TYPE): Likewise.
* sysdeps/unix/sysv/linux/x86_64/bits/typesizes.h
(__POINTER_GUARD_T_TYPE): Likewise.
* elf/rtld.c (__pointer_chk_guard_local): Replace uintptr_t
wi

Re: PATCH: Add x32 time

H.J. Lu — 2012-05-18T00:25:36

Those are in C.  If you compare gettimeofday.c and time.c, excluding
copyright, there are more codes which are different than common. We
can define libc_ifunc_vdso in include/libc-symbols.h to cover

# include <dl-vdso.h>

void *time_ifunc (void) __asm__ ("__GI_time");

void *
time_ifunc (void)
{
  PREPARE_VERSION (linux26, "LINUX_2.6", 61765110);

  return _dl_vdso_vsym ("__vdso_time", &linux26);
}
__asm (".type __GI_time, %gnu_indirect_function");

It should cover most if not all common code.

Re: Flatten sysdeps/unix/bsd/bsd4.4 into sysdeps/unix/bsd

David Miller — 2012-05-18T00:03:40

From: David Miller <davem< at >davemloft.net>
Date: Thu, 17 May 2012 19:37:49 -0400 (EDT)

I was very curious how much this would help my build times on sparc.
So I tested changing "asm" to just ".S" instead of ".S .s" in the
sysd-rules target.

Before:

davem< at >nunsaram:~/src/GIT/GLIBC/build-sparcv9-multiarch$ time make -s -j128 >build.log 2>&1; echo $?

real    8m28.048s
user    92m39.820s
sys     7m34.810s

After:

davem< at >nunsaram:~/src/GIT/GLIBC/build-sparcv9-multiarch$ time make -s -j128 >build.log 2>&1; echo $?

real    7m35.993s
user    93m26.110s
sys     7m46.160s

Nearly a full minute.  And this is with a hacked GNU make that
attempts to mitigate a large amount of the prefix rule overhead.  For
those of you using standard GNU make binaries, the effect will be
even larger.

I guess the only saving grace for make taking up %60 of the glibc
build time is that it's good for tuning strcmp() et al. :-)

Re: [RFC] FIPS compliance and other crypt(3) improvements

Alexandre Oliva — 2012-05-17T23:54:12


It's the only POSIX-documented error code for crypt.  That's why I went
with it.




Eeek!  Thanks, fixed.


Nope.  I came up with it myself.  I thought of adding _GNU_ in there
somewhere, and bumping the number way up, as an extension without
conflicts, but I didn't get that far.


Other userland programs and libraries test FIPS status reading /proc
files directly, but I though they (and any newer programs) could switch
to a more portable interface.