diff options
| -rw-r--r-- | ChangeLog | 16 | ||||
| -rw-r--r-- | crypt/crypt_util.c | 24 | ||||
| -rw-r--r-- | nptl/ChangeLog | 8 | ||||
| -rw-r--r-- | nptl/sysdeps/pthread/unwind-forcedunwind.c | 14 | ||||
| -rw-r--r-- | sysdeps/x86_64/dl-trampoline.S | 25 |
5 files changed, 71 insertions, 16 deletions
diff --git a/ChangeLog b/ChangeLog index 9e03d17257..d149838948 100644 --- a/ChangeLog +++ b/ChangeLog @@ -1,3 +1,19 @@ +2011-07-22 Ulrich Drepper <drepper@gmail.com> + + * sysdeps/x86_64/dl-trampoline.S (_dl_runtime_profile): Fix test for + OSXSAVE. + (_dl_x86_64_save_sse): Likewise. + + * crypt/crypt_util.c (__init_des_r): Optimize memset calls. + + * crypt/crypt_util.c (__init_des_r): Add read barrier as well. + +2011-07-21 Andreas Schwab <schwab@redhat.com> + + * sysdeps/x86_64/dl-trampoline.S (_dl_runtime_profile): Fix last + change. + (_dl_x86_64_save_sse): Use correct AVX check. + 2011-07-21 Liubov Dmitrieva <liubov.dmitrieva@gmail.com> * sysdeps/x86_64/multiarch/strcpy-sse2-unaligned.S: Fix overfow diff --git a/crypt/crypt_util.c b/crypt/crypt_util.c index 7999a30672..04b262d9c8 100644 --- a/crypt/crypt_util.c +++ b/crypt/crypt_util.c @@ -1,7 +1,8 @@ /* * UFC-crypt: ultra fast crypt(3) implementation * - * Copyright (C) 1991-1993,1996-1998,2000,2010 Free Software Foundation, Inc. + * Copyright (C) 1991-1993,1996-1998,2000,2010,2011 + * Free Software Foundation, Inc. * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public @@ -475,7 +476,8 @@ __init_des_r(__data) small_tables_done: __libc_lock_unlock(_ufc_tables_lock); #endif - } + } else + atomic_read_barrier (); /* * Create the sb tables: @@ -490,10 +492,20 @@ small_tables_done: * */ - _ufc_clearmem((char*)__data->sb0, (int)sizeof(__data->sb0)); - _ufc_clearmem((char*)__data->sb1, (int)sizeof(__data->sb1)); - _ufc_clearmem((char*)__data->sb2, (int)sizeof(__data->sb2)); - _ufc_clearmem((char*)__data->sb3, (int)sizeof(__data->sb3)); + if (__data->sb0 + sizeof (__data->sb0) == __data->sb1 + && __data->sb1 + sizeof (__data->sb1) == __data->sb2 + && __data->sb2 + sizeof (__data->sb2) == __data->sb3) + _ufc_clearmem(__data->sb0, + (int)sizeof(__data->sb0) + + (int)sizeof(__data->sb1) + + (int)sizeof(__data->sb2) + + (int)sizeof(__data->sb3)); + else { + _ufc_clearmem(__data->sb0, (int)sizeof(__data->sb0)); + _ufc_clearmem(__data->sb1, (int)sizeof(__data->sb1)); + _ufc_clearmem(__data->sb2, (int)sizeof(__data->sb2)); + _ufc_clearmem(__data->sb3, (int)sizeof(__data->sb3)); + } for(sg = 0; sg < 4; sg++) { int j1, j2; diff --git a/nptl/ChangeLog b/nptl/ChangeLog index 2acc676fd8..5920800a9f 100644 --- a/nptl/ChangeLog +++ b/nptl/ChangeLog @@ -1,3 +1,11 @@ +2011-07-22 Ulrich Drepper <drepper@gmail.com> + + * sysdeps/pthread/unwind-forcedunwind.c (_Unwind_Resume): Add read + barrier. + (__gcc_personality_v0): Likewise. + (_Unwind_ForcedUnwind): Likewise. + (_Unwind_GetCFA): Likewise. + 2011-07-14 Roland McGrath <roland@hack.frob.com> * allocatestack.c (__reclaim_stacks): Use uintptr_t cast rather than diff --git a/nptl/sysdeps/pthread/unwind-forcedunwind.c b/nptl/sysdeps/pthread/unwind-forcedunwind.c index ae43f335e7..3b27827c6b 100644 --- a/nptl/sysdeps/pthread/unwind-forcedunwind.c +++ b/nptl/sysdeps/pthread/unwind-forcedunwind.c @@ -1,4 +1,4 @@ -/* Copyright (C) 2003, 2005, 2006, 2009 Free Software Foundation, Inc. +/* Copyright (C) 2003, 2005, 2006, 2009, 2011 Free Software Foundation, Inc. This file is part of the GNU C Library. Contributed by Jakub Jelinek <jakub@redhat.com>. @@ -96,6 +96,8 @@ _Unwind_Resume (struct _Unwind_Exception *exc) { if (__builtin_expect (libgcc_s_handle == NULL, 0)) pthread_cancel_init (); + else + atomic_read_barrier (); void (*resume) (struct _Unwind_Exception *exc) = libgcc_s_resume; PTR_DEMANGLE (resume); @@ -105,11 +107,13 @@ _Unwind_Resume (struct _Unwind_Exception *exc) _Unwind_Reason_Code __gcc_personality_v0 (int version, _Unwind_Action actions, _Unwind_Exception_Class exception_class, - struct _Unwind_Exception *ue_header, - struct _Unwind_Context *context) + struct _Unwind_Exception *ue_header, + struct _Unwind_Context *context) { if (__builtin_expect (libgcc_s_handle == NULL, 0)) pthread_cancel_init (); + else + atomic_read_barrier (); _Unwind_Reason_Code (*personality) (int, _Unwind_Action, _Unwind_Exception_Class, struct _Unwind_Exception *, @@ -124,6 +128,8 @@ _Unwind_ForcedUnwind (struct _Unwind_Exception *exc, _Unwind_Stop_Fn stop, { if (__builtin_expect (libgcc_s_handle == NULL, 0)) pthread_cancel_init (); + else + atomic_read_barrier (); _Unwind_Reason_Code (*forcedunwind) (struct _Unwind_Exception *, _Unwind_Stop_Fn, void *) @@ -137,6 +143,8 @@ _Unwind_GetCFA (struct _Unwind_Context *context) { if (__builtin_expect (libgcc_s_handle == NULL, 0)) pthread_cancel_init (); + else + atomic_read_barrier (); _Unwind_Word (*getcfa) (struct _Unwind_Context *) = libgcc_s_getcfa; PTR_DEMANGLE (getcfa); diff --git a/sysdeps/x86_64/dl-trampoline.S b/sysdeps/x86_64/dl-trampoline.S index 1b97929aaa..4aa63ee8d5 100644 --- a/sysdeps/x86_64/dl-trampoline.S +++ b/sysdeps/x86_64/dl-trampoline.S @@ -141,7 +141,8 @@ L(have_avx): movq %r11,%rbx # Restore rbx xorl %eax, %eax // AVX and XSAVE supported? - testl $((1 << 28) | (1 << 27)), %ecx + andl $((1 << 28) | (1 << 27)), %ecx + cmpl $((1 << 28) | (1 << 27)), %ecx je 2f xorl %ecx, %ecx // Get XFEATURE_ENABLED_MASK @@ -150,7 +151,8 @@ L(have_avx): cmpl $0x6, %eax // Nonzero if SSE and AVX state saving is enabled. sete %al -2: movl %eax, L(have_avx)(%rip) +2: leal -1(%eax,%eax), %eax + movl %eax, L(have_avx)(%rip) cmpl $0, %eax 1: js L(no_avx) @@ -183,11 +185,20 @@ _dl_x86_64_save_sse: movl $1, %eax cpuid movq %r11,%rbx # Restore rbx - movl $1, %eax - testl $(1 << 28), %ecx - jne 2f - negl %eax -2: movl %eax, L(have_avx)(%rip) + xorl %eax, %eax + // AVX and XSAVE supported? + andl $((1 << 28) | (1 << 27)), %ecx + cmpl $((1 << 28) | (1 << 27)), %ecx + je 2f + xorl %ecx, %ecx + // Get XFEATURE_ENABLED_MASK + xgetbv + andl $0x6, %eax + cmpl $0x6, %eax + // Nonzero if SSE and AVX state saving is enabled. + sete %al +2: leal -1(%eax,%eax), %eax + movl %eax, L(have_avx)(%rip) cmpl $0, %eax 1: js L(no_avx5) |