Fix check for AVX enablement

The AVX bit is set if the CPU supports AVX.  But this doesn't mean the
kernel does.  Add checks according to Intel's documentation.
(cherry picked from commit 5644ef5461b5d3ff266206d8ee70d4b575ea6658)

diff --git a/ChangeLog b/ChangeLog
index 515f22c0462a3ab2beae886b0b0b736f79029e25..9b130a5cd7c5a31705bcb034f7cedc1750f75c0b 100644 (file)
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,12 @@
+2011-07-20  Ulrich Drepper  <drepper@gmail.com>
+
+       [BZ #13007]
+       * sysdeps/x86_64/dl-trampoline.S (_dl_runtime_profile): More complete
+       check for AVX enablement so that we don't crash with old kernels and
+       new hardware.
+       * elf/tst-audit4.c: Add same checks here.
+       * elf/tst-audit6.c: Likewise.
+
 2011-07-20  Ulrich Drepper  <drepper@gmail.com>
 
        [BZ #12852]

diff --git a/elf/tst-audit4.c b/elf/tst-audit4.c
index b17d4a61a792173090214156c8f8aad3486a9697..c4f1d5bdb9cf06e26ea2abc61b985a70ad25dc49 100644 (file)
--- a/elf/tst-audit4.c
+++ b/elf/tst-audit4.c
@@ -6,16 +6,30 @@
 #include <cpuid.h>
 #include <immintrin.h>
 
+
+static int
+avx_enabled (void)
+{
+  unsigned int eax, ebx, ecx, edx;
+
+  if (__get_cpuid (1, &eax, &ebx, &ecx, &edx) == 0
+      || (ecx & (bit_AVX | bit_OSXSAVE)) != (bit_AVX | bit_OSXSAVE))
+    return 0;
+
+  /* Check the OS has AVX and SSE saving enabled.  */
+  asm ("xgetbv" : "=a" (eax), "=d" (edx) : "c" (0));
+
+  return (eax & 6) == 6;
+}
+
+
 extern __m256i audit_test (__m256i, __m256i, __m256i, __m256i,
                           __m256i, __m256i, __m256i, __m256i);
 int
 main (void)
 {
-  unsigned int eax, ebx, ecx, edx;
-
   /* Run AVX test only if AVX is supported.  */
-  if (__get_cpuid (1, &eax, &ebx, &ecx, &edx)
-      && (ecx & bit_AVX))
+  if (avx_enabled ())
     {
       __m256i ymm = _mm256_setzero_si256 ();
       __m256i ret = audit_test (ymm, ymm, ymm, ymm, ymm, ymm, ymm, ymm);

diff --git a/elf/tst-audit6.c b/elf/tst-audit6.c
index 1f6dcb16e9a529397d18cf334ad375ca2ec9c4c5..64209a152e7424a66d47d81c92ed3883b3c4bda0 100644 (file)
--- a/elf/tst-audit6.c
+++ b/elf/tst-audit6.c
@@ -8,14 +8,28 @@
 extern __m128i audit_test (__m128i, __m128i, __m128i, __m128i,
                           __m128i, __m128i, __m128i, __m128i);
 
-int
-main (void)
+
+static int
+avx_enabled (void)
 {
   unsigned int eax, ebx, ecx, edx;
 
+  if (__get_cpuid (1, &eax, &ebx, &ecx, &edx) == 0
+      || (ecx & (bit_AVX | bit_OSXSAVE)) != (bit_AVX | bit_OSXSAVE))
+    return 0;
+
+  /* Check the OS has AVX and SSE saving enabled.  */
+  asm ("xgetbv" : "=a" (eax), "=d" (edx) : "c" (0));
+
+  return (eax & 6) == 6;
+}
+
+
+int
+main (void)
+{
   /* Run AVX test only if AVX is supported.  */
-  if (__get_cpuid (1, &eax, &ebx, &ecx, &edx)
-      && (ecx & bit_AVX))
+  if (avx_enabled ())
     {
       __m128i xmm = _mm_setzero_si128 ();
       __m128i ret = audit_test (xmm, xmm, xmm, xmm, xmm, xmm, xmm, xmm);

diff --git a/sysdeps/x86_64/dl-trampoline.S b/sysdeps/x86_64/dl-trampoline.S
index 5564a11af296203c7fbc64277fdb644a47899248..1b97929aaa159adf92a3f7344444145bf7cb5c76 100644 (file)
--- a/sysdeps/x86_64/dl-trampoline.S
+++ b/sysdeps/x86_64/dl-trampoline.S
@@ -1,5 +1,5 @@
 /* PLT trampolines.  x86-64 version.
-   Copyright (C) 2004, 2005, 2007, 2009 Free Software Foundation, Inc.
+   Copyright (C) 2004, 2005, 2007, 2009, 2011 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -139,10 +139,17 @@ L(have_avx):
        movl    $1, %eax
        cpuid
        movq    %r11,%rbx               # Restore rbx
-       movl    $1, %eax
-       testl   $(1 << 28), %ecx
-       jne     2f
-       negl    %eax
+       xorl    %eax, %eax
+       // AVX and XSAVE supported?
+       testl   $((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:     movl    %eax, L(have_avx)(%rip)
        cmpl    $0, %eax