Support mixed SSE/AVX audit and check AVX only once.

This patch fixes mixed SSE/AVX audit and checks AVX only once in
_dl_runtime_profile. When an AVX or SSE register value in pltenter is
modified, we have to make sure that the SSE part value is the same in both
lr_xmm and lr_vector fields so that pltexit will get the correct value
from either lr_xmm or lr_vector fields. AVX-enabled pltenter should
update both lr_xmm and lr_vector fields to support stacked AVX/SSE
pltenter functions.

8 files changed, 758 insertions, 2 deletions

diff --git a/elf/Makefile b/elf/Makefile
index 3baad9621d..d57c7fe7ed 100644
--- a/elf/Makefile
+++ b/elf/Makefile
@@ -93,6 +93,9 @@ distribute	:= rtld-Rules \
 		   tst-auditmod1.c tst-auditmod3a.c tst-auditmod3b.c \
 		   tst-auditmod4a.c tst-auditmod4b.c \
 		   tst-audit5.c tst-auditmod5a.c tst-auditmod5b.c \
+		   tst-audit6.c tst-auditmod6a.c tst-auditmod6b.c \
+		   tst-auditmod6c.c \
+		   tst-audit7.c tst-auditmod7a.c tst-auditmod7b.c \
 		   order2mod1.c order2mod2.c order2mod3.c order2mod4.c \
 		   tst-stackguard1.c tst-stackguard1-static.c \
 		   tst-array5.c tst-array5-static.c tst-array5dep.c \
@@ -200,7 +203,7 @@ tests += loadtest restest1 preloadtest loadfail multiload origtest resolvfail \
 test-srcs = tst-pathopt
 tests-execstack-yes = tst-execstack tst-execstack-needed tst-execstack-prog
 ifeq (x86_64,$(config-machine))
-tests += tst-audit3 tst-audit4 tst-audit5
+tests += tst-audit3 tst-audit4 tst-audit5 tst-audit6 tst-audit7
 endif
 endif
 ifeq (yesyes,$(have-fpie)$(build-shared))
@@ -255,7 +258,9 @@ endif
 ifeq (x86_64,$(config-machine))
 modules-names += tst-auditmod3a tst-auditmod3b \
 		tst-auditmod4a tst-auditmod4b \
-		tst-auditmod5a tst-auditmod5b
+		tst-auditmod5a tst-auditmod5b \
+		tst-auditmod6a tst-auditmod6b tst-auditmod6c \
+		tst-auditmod7a tst-auditmod7b
 endif
 modules-execstack-yes = tst-execstack-mod
 extra-test-objs += $(addsuffix .os,$(strip $(modules-names)))
@@ -987,6 +992,15 @@ $(objpfx)tst-audit5: $(objpfx)tst-auditmod5a.so
 $(objpfx)tst-audit5.out: $(objpfx)tst-auditmod5b.so
 tst-audit5-ENV = LD_AUDIT=$(objpfx)tst-auditmod5b.so
 
+$(objpfx)tst-audit6: $(objpfx)tst-auditmod6a.so
+$(objpfx)tst-audit6.out: $(objpfx)tst-auditmod6b.so \
+			 $(objpfx)tst-auditmod6c.so
+tst-audit6-ENV = LD_AUDIT=$(objpfx)tst-auditmod6b.so:$(objpfx)tst-auditmod6c.so
+
+$(objpfx)tst-audit7: $(objpfx)tst-auditmod7a.so
+$(objpfx)tst-audit7.out: $(objpfx)tst-auditmod7b.so
+tst-audit7-ENV = LD_AUDIT=$(objpfx)tst-auditmod7b.so
+
 $(objpfx)tst-global1: $(libdl)
 $(objpfx)tst-global1.out: $(objpfx)testobj6.so $(objpfx)testobj2.so
 
@@ -1134,4 +1148,7 @@ ifeq (yes,$(config-cflags-avx))
 CFLAGS-tst-audit4.c += -mavx
 CFLAGS-tst-auditmod4a.c += -mavx
 CFLAGS-tst-auditmod4b.c += -mavx
+CFLAGS-tst-auditmod6b.c += -mavx
+CFLAGS-tst-auditmod6c.c += -mavx
+CFLAGS-tst-auditmod7b.c += -mavx
 endif
diff --git a/elf/tst-audit6.c b/elf/tst-audit6.c
new file mode 100644
index 0000000000..1f6dcb16e9
--- /dev/null
+++ b/elf/tst-audit6.c
@@ -0,0 +1,28 @@
+/* Test case for x86-64 preserved registers in dynamic linker.  */
+
+#include <stdlib.h>
+#include <string.h>
+#include <cpuid.h>
+#include <emmintrin.h>
+
+extern __m128i audit_test (__m128i, __m128i, __m128i, __m128i,
+			   __m128i, __m128i, __m128i, __m128i);
+
+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))
+    {
+      __m128i xmm = _mm_setzero_si128 ();
+      __m128i ret = audit_test (xmm, xmm, xmm, xmm, xmm, xmm, xmm, xmm);
+
+      xmm = _mm_set1_epi32 (0x98abcdef);
+      if (memcmp (&xmm, &ret, sizeof (ret)))
+	abort ();
+    }
+  return 0;
+}
diff --git a/elf/tst-audit7.c b/elf/tst-audit7.c
new file mode 100644
index 0000000000..1d2a7de439
--- /dev/null
+++ b/elf/tst-audit7.c
@@ -0,0 +1 @@
+#include "tst-audit6.c"
diff --git a/elf/tst-auditmod6a.c b/elf/tst-auditmod6a.c
new file mode 100644
index 0000000000..c3a850ce98
--- /dev/null
+++ b/elf/tst-auditmod6a.c
@@ -0,0 +1,46 @@
+/* Test case for x86-64 preserved registers in dynamic linker.  */
+
+#include <stdlib.h>
+#include <string.h>
+#include <emmintrin.h>
+
+__m128i
+audit_test (__m128i x0, __m128i x1, __m128i x2, __m128i x3,
+	    __m128i x4, __m128i x5, __m128i x6, __m128i x7)
+{
+  __m128i xmm;
+
+  xmm =  _mm_set1_epi32 (0x100);
+  if (memcmp (&xmm, &x0, sizeof (xmm)))
+    abort ();
+
+  xmm =  _mm_set1_epi32 (0x101);
+  if (memcmp (&xmm, &x1, sizeof (xmm)))
+    abort ();
+
+  xmm =  _mm_set1_epi32 (0x102);
+  if (memcmp (&xmm, &x2, sizeof (xmm)))
+    abort ();
+
+  xmm =  _mm_set1_epi32 (0x103);
+  if (memcmp (&xmm, &x3, sizeof (xmm)))
+    abort ();
+
+  xmm =  _mm_set1_epi32 (0x104);
+  if (memcmp (&xmm, &x4, sizeof (xmm)))
+    abort ();
+
+  xmm =  _mm_set1_epi32 (0x105);
+  if (memcmp (&xmm, &x5, sizeof (xmm)))
+    abort ();
+
+  xmm =  _mm_set1_epi32 (0x106);
+  if (memcmp (&xmm, &x6, sizeof (xmm)))
+    abort ();
+
+  xmm =  _mm_set1_epi32 (0x107);
+  if (memcmp (&xmm, &x7, sizeof (xmm)))
+    abort ();
+
+  return _mm_setzero_si128 ();
+}
diff --git a/elf/tst-auditmod6b.c b/elf/tst-auditmod6b.c
new file mode 100644
index 0000000000..f756b50227
--- /dev/null
+++ b/elf/tst-auditmod6b.c
@@ -0,0 +1,220 @@
+/* Verify that changing AVX registers in audit library won't affect
+   function parameter passing/return.  */
+
+#include <dlfcn.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+#include <bits/wordsize.h>
+#include <gnu/lib-names.h>
+
+unsigned int
+la_version (unsigned int v)
+{
+  setlinebuf (stdout);
+
+  printf ("version: %u\n", v);
+
+  char buf[20];
+  sprintf (buf, "%u", v);
+
+  return v;
+}
+
+void
+la_activity (uintptr_t *cookie, unsigned int flag)
+{
+  if (flag == LA_ACT_CONSISTENT)
+    printf ("activity: consistent\n");
+  else if (flag == LA_ACT_ADD)
+    printf ("activity: add\n");
+  else if (flag == LA_ACT_DELETE)
+    printf ("activity: delete\n");
+  else
+    printf ("activity: unknown activity %u\n", flag);
+}
+
+char *
+la_objsearch (const char *name, uintptr_t *cookie, unsigned int flag)
+{
+  char buf[100];
+  const char *flagstr;
+  if (flag == LA_SER_ORIG)
+    flagstr = "LA_SET_ORIG";
+  else if (flag == LA_SER_LIBPATH)
+    flagstr = "LA_SER_LIBPATH";
+  else if (flag == LA_SER_RUNPATH)
+    flagstr = "LA_SER_RUNPATH";
+  else if (flag == LA_SER_CONFIG)
+    flagstr = "LA_SER_CONFIG";
+  else if (flag == LA_SER_DEFAULT)
+    flagstr = "LA_SER_DEFAULT";
+  else if (flag == LA_SER_SECURE)
+    flagstr = "LA_SER_SECURE";
+  else
+    {
+       sprintf (buf, "unknown flag %d", flag);
+       flagstr = buf;
+    }
+  printf ("objsearch: %s, %s\n", name, flagstr);
+
+  return (char *) name;
+}
+
+unsigned int
+la_objopen (struct link_map *l, Lmid_t lmid, uintptr_t *cookie)
+{
+  printf ("objopen: %ld, %s\n", lmid, l->l_name);
+
+  return 3;
+}
+
+void
+la_preinit (uintptr_t *cookie)
+{
+  printf ("preinit\n");
+}
+
+unsigned int
+la_objclose  (uintptr_t *cookie)
+{
+  printf ("objclose\n");
+  return 0;
+}
+
+uintptr_t
+la_symbind64 (Elf64_Sym *sym, unsigned int ndx, uintptr_t *refcook,
+	      uintptr_t *defcook, unsigned int *flags, const char *symname)
+{
+  printf ("symbind64: symname=%s, st_value=%#lx, ndx=%u, flags=%u\n",
+	  symname, (long int) sym->st_value, ndx, *flags);
+
+  return sym->st_value;
+}
+
+#define pltenter la_x86_64_gnu_pltenter
+#define pltexit la_x86_64_gnu_pltexit
+#define La_regs La_x86_64_regs
+#define La_retval La_x86_64_retval
+#define int_retval lrv_rax
+
+#include <tst-audit.h>
+
+#ifdef __AVX__
+#include <immintrin.h>
+#include <cpuid.h>
+
+static int avx = -1;
+
+static int
+__attribute ((always_inline))
+check_avx (void)
+{
+  if (avx == -1)
+    {
+      unsigned int eax, ebx, ecx, edx;
+
+      if (__get_cpuid (1, &eax, &ebx, &ecx, &edx)
+	  && (ecx & bit_AVX))
+	avx = 1;
+      else
+	avx = 0;
+    }
+  return avx;
+}
+#else
+#include <emmintrin.h>
+#endif
+
+ElfW(Addr)
+pltenter (ElfW(Sym) *sym, unsigned int ndx, uintptr_t *refcook,
+	  uintptr_t *defcook, La_regs *regs, unsigned int *flags,
+	  const char *symname, long int *framesizep)
+{
+  printf ("pltenter: symname=%s, st_value=%#lx, ndx=%u, flags=%u\n",
+	  symname, (long int) sym->st_value, ndx, *flags);
+
+#ifdef __AVX__
+  if (check_avx () && strcmp (symname, "audit_test") == 0)
+    {
+      int i;
+
+      __m128i xmm = _mm_setzero_si128 ();
+      for (i = 0; i < 8; i++)
+	if (memcmp (&regs->lr_xmm[i], &xmm, sizeof (xmm))
+	    || memcmp (&regs->lr_vector[i], &xmm, sizeof (xmm)))
+	  abort ();
+
+      for (i = 0; i < 8; i += 2)
+	{
+	  regs->lr_xmm[i] = (La_x86_64_xmm) _mm_set1_epi32 (i + 1);
+	  regs->lr_vector[i].xmm[0] = regs->lr_xmm[i];
+	  regs->lr_vector[i + 1].ymm[0]
+	    = (La_x86_64_ymm) _mm256_set1_epi32 (i + 2);
+	  regs->lr_xmm[i + 1] = regs->lr_vector[i + 1].xmm[0];
+	}
+
+      __m256i ymm = _mm256_set1_epi32 (-1);
+      asm volatile ("vmovdqa %0, %%ymm0" : : "x" (ymm) : "xmm0" );
+      asm volatile ("vmovdqa %0, %%ymm1" : : "x" (ymm) : "xmm1" );
+      asm volatile ("vmovdqa %0, %%ymm2" : : "x" (ymm) : "xmm2" );
+      asm volatile ("vmovdqa %0, %%ymm3" : : "x" (ymm) : "xmm3" );
+      asm volatile ("vmovdqa %0, %%ymm4" : : "x" (ymm) : "xmm4" );
+      asm volatile ("vmovdqa %0, %%ymm5" : : "x" (ymm) : "xmm5" );
+      asm volatile ("vmovdqa %0, %%ymm6" : : "x" (ymm) : "xmm6" );
+      asm volatile ("vmovdqa %0, %%ymm7" : : "x" (ymm) : "xmm7" );
+
+      *framesizep = 1024;
+    }
+#endif
+
+  return sym->st_value;
+}
+
+unsigned int
+pltexit (ElfW(Sym) *sym, unsigned int ndx, uintptr_t *refcook,
+	 uintptr_t *defcook, const La_regs *inregs, La_retval *outregs,
+	 const char *symname)
+{
+  printf ("pltexit: symname=%s, st_value=%#lx, ndx=%u, retval=%tu\n",
+	  symname, (long int) sym->st_value, ndx, outregs->int_retval);
+
+#ifdef __AVX__
+  if (check_avx () && strcmp (symname, "audit_test") == 0)
+    {
+      int i;
+
+      __m128i xmm = _mm_setzero_si128 ();
+      if (memcmp (&outregs->lrv_xmm0, &xmm, sizeof (xmm))
+	  || memcmp (&outregs->lrv_vector0, &xmm, sizeof (xmm)))
+	abort ();
+
+      __m256i ymm;
+
+      for (i = 0; i < 8; i += 2)
+	{
+	  xmm = _mm_set1_epi32 (i + 0x100);
+	  if (memcmp (&inregs->lr_xmm[i], &xmm, sizeof (xmm))
+	      || memcmp (&inregs->lr_vector[i], &xmm, sizeof (xmm)))
+	    abort ();
+
+	  ymm = _mm256_set1_epi32 (i + 0x101);
+	  if (memcmp (&inregs->lr_xmm[i + 1],
+		      &inregs->lr_vector[i + 1].xmm[0], sizeof (xmm))
+	      || memcmp (&inregs->lr_vector[i + 1], &ymm, sizeof (ymm)))
+	    abort ();
+	}
+
+      outregs->lrv_vector0.ymm[0]
+	= (La_x86_64_ymm) _mm256_set1_epi32 (0x12349876);
+
+      ymm = _mm256_set1_epi32 (-1);
+      asm volatile ("vmovdqa %0, %%ymm0" : : "x" (ymm) : "xmm0" );
+      asm volatile ("vmovdqa %0, %%ymm1" : : "x" (ymm) : "xmm1" );
+    }
+#endif
+
+  return 0;
+}
diff --git a/elf/tst-auditmod6c.c b/elf/tst-auditmod6c.c
new file mode 100644
index 0000000000..49cbf05492
--- /dev/null
+++ b/elf/tst-auditmod6c.c
@@ -0,0 +1,225 @@
+/* Verify that changing AVX registers in audit library won't affect
+   function parameter passing/return.  */
+
+#include <dlfcn.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+#include <bits/wordsize.h>
+#include <gnu/lib-names.h>
+
+unsigned int
+la_version (unsigned int v)
+{
+  setlinebuf (stdout);
+
+  printf ("version: %u\n", v);
+
+  char buf[20];
+  sprintf (buf, "%u", v);
+
+  return v;
+}
+
+void
+la_activity (uintptr_t *cookie, unsigned int flag)
+{
+  if (flag == LA_ACT_CONSISTENT)
+    printf ("activity: consistent\n");
+  else if (flag == LA_ACT_ADD)
+    printf ("activity: add\n");
+  else if (flag == LA_ACT_DELETE)
+    printf ("activity: delete\n");
+  else
+    printf ("activity: unknown activity %u\n", flag);
+}
+
+char *
+la_objsearch (const char *name, uintptr_t *cookie, unsigned int flag)
+{
+  char buf[100];
+  const char *flagstr;
+  if (flag == LA_SER_ORIG)
+    flagstr = "LA_SET_ORIG";
+  else if (flag == LA_SER_LIBPATH)
+    flagstr = "LA_SER_LIBPATH";
+  else if (flag == LA_SER_RUNPATH)
+    flagstr = "LA_SER_RUNPATH";
+  else if (flag == LA_SER_CONFIG)
+    flagstr = "LA_SER_CONFIG";
+  else if (flag == LA_SER_DEFAULT)
+    flagstr = "LA_SER_DEFAULT";
+  else if (flag == LA_SER_SECURE)
+    flagstr = "LA_SER_SECURE";
+  else
+    {
+       sprintf (buf, "unknown flag %d", flag);
+       flagstr = buf;
+    }
+  printf ("objsearch: %s, %s\n", name, flagstr);
+
+  return (char *) name;
+}
+
+unsigned int
+la_objopen (struct link_map *l, Lmid_t lmid, uintptr_t *cookie)
+{
+  printf ("objopen: %ld, %s\n", lmid, l->l_name);
+
+  return 3;
+}
+
+void
+la_preinit (uintptr_t *cookie)
+{
+  printf ("preinit\n");
+}
+
+unsigned int
+la_objclose  (uintptr_t *cookie)
+{
+  printf ("objclose\n");
+  return 0;
+}
+
+uintptr_t
+la_symbind64 (Elf64_Sym *sym, unsigned int ndx, uintptr_t *refcook,
+	      uintptr_t *defcook, unsigned int *flags, const char *symname)
+{
+  printf ("symbind64: symname=%s, st_value=%#lx, ndx=%u, flags=%u\n",
+	  symname, (long int) sym->st_value, ndx, *flags);
+
+  return sym->st_value;
+}
+
+#define pltenter la_x86_64_gnu_pltenter
+#define pltexit la_x86_64_gnu_pltexit
+#define La_regs La_x86_64_regs
+#define La_retval La_x86_64_retval
+#define int_retval lrv_rax
+
+#include <tst-audit.h>
+
+#ifdef __AVX__
+#include <immintrin.h>
+#include <cpuid.h>
+
+static int avx = -1;
+
+static int
+__attribute ((always_inline))
+check_avx (void)
+{
+  if (avx == -1)
+    {
+      unsigned int eax, ebx, ecx, edx;
+
+      if (__get_cpuid (1, &eax, &ebx, &ecx, &edx)
+	  && (ecx & bit_AVX))
+	avx = 1;
+      else
+	avx = 0;
+    }
+  return avx;
+}
+#else
+#include <emmintrin.h>
+#endif
+
+ElfW(Addr)
+pltenter (ElfW(Sym) *sym, unsigned int ndx, uintptr_t *refcook,
+	  uintptr_t *defcook, La_regs *regs, unsigned int *flags,
+	  const char *symname, long int *framesizep)
+{
+  printf ("pltenter: symname=%s, st_value=%#lx, ndx=%u, flags=%u\n",
+	  symname, (long int) sym->st_value, ndx, *flags);
+
+#ifdef __AVX__
+  if (check_avx () && strcmp (symname, "audit_test") == 0)
+    {
+      int i;
+      __m128i xmm;
+      __m256i ymm;
+
+      for (i = 0; i < 8; i += 2)
+	{
+	  xmm = _mm_set1_epi32 (i + 1);
+	  if (memcmp (&regs->lr_xmm[i], &xmm, sizeof (xmm))
+	      || memcmp (&regs->lr_vector[i], &xmm, sizeof (xmm)))
+	    abort ();
+	  regs->lr_xmm[i] = (La_x86_64_xmm) _mm_set1_epi32 (i + 0x100);
+	  regs->lr_vector[i].xmm[0] = regs->lr_xmm[i];
+
+	  ymm = _mm256_set1_epi32 (i + 2);
+	  if (memcmp (&regs->lr_xmm[i + 1],
+		      &regs->lr_vector[i + 1].xmm[0], sizeof (xmm))
+	      || memcmp (&regs->lr_vector[i + 1], &ymm, sizeof (ymm)))
+	    abort ();
+	  regs->lr_vector[i + 1].ymm[0]
+	    = (La_x86_64_ymm) _mm256_set1_epi32 (i + 0x101);
+	  regs->lr_xmm[i + 1] = regs->lr_vector[i + 1].xmm[0];
+	}
+
+      ymm = _mm256_set1_epi32 (-1);
+      asm volatile ("vmovdqa %0, %%ymm0" : : "x" (ymm) : "xmm0" );
+      asm volatile ("vmovdqa %0, %%ymm1" : : "x" (ymm) : "xmm1" );
+      asm volatile ("vmovdqa %0, %%ymm2" : : "x" (ymm) : "xmm2" );
+      asm volatile ("vmovdqa %0, %%ymm3" : : "x" (ymm) : "xmm3" );
+      asm volatile ("vmovdqa %0, %%ymm4" : : "x" (ymm) : "xmm4" );
+      asm volatile ("vmovdqa %0, %%ymm5" : : "x" (ymm) : "xmm5" );
+      asm volatile ("vmovdqa %0, %%ymm6" : : "x" (ymm) : "xmm6" );
+      asm volatile ("vmovdqa %0, %%ymm7" : : "x" (ymm) : "xmm7" );
+
+      *framesizep = 1024;
+    }
+#endif
+
+  return sym->st_value;
+}
+
+unsigned int
+pltexit (ElfW(Sym) *sym, unsigned int ndx, uintptr_t *refcook,
+	 uintptr_t *defcook, const La_regs *inregs, La_retval *outregs,
+	 const char *symname)
+{
+  printf ("pltexit: symname=%s, st_value=%#lx, ndx=%u, retval=%tu\n",
+	  symname, (long int) sym->st_value, ndx, outregs->int_retval);
+
+#ifdef __AVX__
+  if (check_avx () && strcmp (symname, "audit_test") == 0)
+    {
+      int i;
+
+      __m256i ymm = _mm256_set1_epi32 (0x12349876);;
+      if (memcmp (&outregs->lrv_vector0, &ymm, sizeof (ymm)))
+	abort ();
+
+      __m128i xmm;
+
+      for (i = 0; i < 8; i += 2)
+	{
+	  xmm = _mm_set1_epi32 (i + 0x100);
+	  if (memcmp (&inregs->lr_xmm[i], &xmm, sizeof (xmm))
+	      || memcmp (&inregs->lr_vector[i], &xmm, sizeof (xmm)))
+	    abort ();
+
+	  ymm = _mm256_set1_epi32 (i + 0x101);
+	  if (memcmp (&inregs->lr_xmm[i + 1],
+		      &inregs->lr_vector[i + 1].xmm[0], sizeof (xmm))
+	      || memcmp (&inregs->lr_vector[i + 1], &ymm, sizeof (ymm)))
+	    abort ();
+	}
+
+      outregs->lrv_vector0.ymm[0]
+	= (La_x86_64_ymm) _mm256_set1_epi32 (0x98abcdef);
+
+      ymm = _mm256_set1_epi32 (-1);
+      asm volatile ("vmovdqa %0, %%ymm0" : : "x" (ymm) : "xmm0" );
+      asm volatile ("vmovdqa %0, %%ymm1" : : "x" (ymm) : "xmm1" );
+    }
+#endif
+
+  return 0;
+}
diff --git a/elf/tst-auditmod7a.c b/elf/tst-auditmod7a.c
new file mode 100644
index 0000000000..b379df75d6
--- /dev/null
+++ b/elf/tst-auditmod7a.c
@@ -0,0 +1 @@
+#include "tst-auditmod6a.c"
diff --git a/elf/tst-auditmod7b.c b/elf/tst-auditmod7b.c
new file mode 100644
index 0000000000..eb237586fe
--- /dev/null
+++ b/elf/tst-auditmod7b.c
@@ -0,0 +1,218 @@
+/* Verify that changing AVX registers in audit library won't affect
+   function parameter passing/return.  */
+
+#include <dlfcn.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+#include <bits/wordsize.h>
+#include <gnu/lib-names.h>
+
+unsigned int
+la_version (unsigned int v)
+{
+  setlinebuf (stdout);
+
+  printf ("version: %u\n", v);
+
+  char buf[20];
+  sprintf (buf, "%u", v);
+
+  return v;
+}
+
+void
+la_activity (uintptr_t *cookie, unsigned int flag)
+{
+  if (flag == LA_ACT_CONSISTENT)
+    printf ("activity: consistent\n");
+  else if (flag == LA_ACT_ADD)
+    printf ("activity: add\n");
+  else if (flag == LA_ACT_DELETE)
+    printf ("activity: delete\n");
+  else
+    printf ("activity: unknown activity %u\n", flag);
+}
+
+char *
+la_objsearch (const char *name, uintptr_t *cookie, unsigned int flag)
+{
+  char buf[100];
+  const char *flagstr;
+  if (flag == LA_SER_ORIG)
+    flagstr = "LA_SET_ORIG";
+  else if (flag == LA_SER_LIBPATH)
+    flagstr = "LA_SER_LIBPATH";
+  else if (flag == LA_SER_RUNPATH)
+    flagstr = "LA_SER_RUNPATH";
+  else if (flag == LA_SER_CONFIG)
+    flagstr = "LA_SER_CONFIG";
+  else if (flag == LA_SER_DEFAULT)
+    flagstr = "LA_SER_DEFAULT";
+  else if (flag == LA_SER_SECURE)
+    flagstr = "LA_SER_SECURE";
+  else
+    {
+       sprintf (buf, "unknown flag %d", flag);
+       flagstr = buf;
+    }
+  printf ("objsearch: %s, %s\n", name, flagstr);
+
+  return (char *) name;
+}
+
+unsigned int
+la_objopen (struct link_map *l, Lmid_t lmid, uintptr_t *cookie)
+{
+  printf ("objopen: %ld, %s\n", lmid, l->l_name);
+
+  return 3;
+}
+
+void
+la_preinit (uintptr_t *cookie)
+{
+  printf ("preinit\n");
+}
+
+unsigned int
+la_objclose  (uintptr_t *cookie)
+{
+  printf ("objclose\n");
+  return 0;
+}
+
+uintptr_t
+la_symbind64 (Elf64_Sym *sym, unsigned int ndx, uintptr_t *refcook,
+	      uintptr_t *defcook, unsigned int *flags, const char *symname)
+{
+  printf ("symbind64: symname=%s, st_value=%#lx, ndx=%u, flags=%u\n",
+	  symname, (long int) sym->st_value, ndx, *flags);
+
+  return sym->st_value;
+}
+
+#define pltenter la_x86_64_gnu_pltenter
+#define pltexit la_x86_64_gnu_pltexit
+#define La_regs La_x86_64_regs
+#define La_retval La_x86_64_retval
+#define int_retval lrv_rax
+
+#include <tst-audit.h>
+
+#ifdef __AVX__
+#include <immintrin.h>
+#include <cpuid.h>
+
+static int avx = -1;
+
+static int
+__attribute ((always_inline))
+check_avx (void)
+{
+  if (avx == -1)
+    {
+      unsigned int eax, ebx, ecx, edx;
+
+      if (__get_cpuid (1, &eax, &ebx, &ecx, &edx)
+	  && (ecx & bit_AVX))
+	avx = 1;
+      else
+	avx = 0;
+    }
+  return avx;
+}
+#else
+#include <emmintrin.h>
+#endif
+
+ElfW(Addr)
+pltenter (ElfW(Sym) *sym, unsigned int ndx, uintptr_t *refcook,
+	  uintptr_t *defcook, La_regs *regs, unsigned int *flags,
+	  const char *symname, long int *framesizep)
+{
+  printf ("pltenter: symname=%s, st_value=%#lx, ndx=%u, flags=%u\n",
+	  symname, (long int) sym->st_value, ndx, *flags);
+
+#ifdef __AVX__
+  if (check_avx () && strcmp (symname, "audit_test") == 0)
+    {
+      int i;
+
+      __m128i xmm = _mm_setzero_si128 ();
+      for (i = 0; i < 8; i++)
+	if (memcmp (&regs->lr_xmm[i], &xmm, sizeof (xmm))
+	    || memcmp (&regs->lr_vector[i], &xmm, sizeof (xmm)))
+	  abort ();
+
+      for (i = 0; i < 8; i += 2)
+	{
+	  regs->lr_xmm[i] = (La_x86_64_xmm) _mm_set1_epi32 (i + 0x100);
+	  regs->lr_vector[i + 1].ymm[0]
+	    = (La_x86_64_ymm) _mm256_set1_epi32 (i + 0x101);
+	}
+
+      __m256i ymm = _mm256_set1_epi32 (-1);
+      asm volatile ("vmovdqa %0, %%ymm0" : : "x" (ymm) : "xmm0" );
+      asm volatile ("vmovdqa %0, %%ymm1" : : "x" (ymm) : "xmm1" );
+      asm volatile ("vmovdqa %0, %%ymm2" : : "x" (ymm) : "xmm2" );
+      asm volatile ("vmovdqa %0, %%ymm3" : : "x" (ymm) : "xmm3" );
+      asm volatile ("vmovdqa %0, %%ymm4" : : "x" (ymm) : "xmm4" );
+      asm volatile ("vmovdqa %0, %%ymm5" : : "x" (ymm) : "xmm5" );
+      asm volatile ("vmovdqa %0, %%ymm6" : : "x" (ymm) : "xmm6" );
+      asm volatile ("vmovdqa %0, %%ymm7" : : "x" (ymm) : "xmm7" );
+
+      *framesizep = 1024;
+    }
+#endif
+
+  return sym->st_value;
+}
+
+unsigned int
+pltexit (ElfW(Sym) *sym, unsigned int ndx, uintptr_t *refcook,
+	 uintptr_t *defcook, const La_regs *inregs, La_retval *outregs,
+	 const char *symname)
+{
+  printf ("pltexit: symname=%s, st_value=%#lx, ndx=%u, retval=%tu\n",
+	  symname, (long int) sym->st_value, ndx, outregs->int_retval);
+
+#ifdef __AVX__
+  if (check_avx () && strcmp (symname, "audit_test") == 0)
+    {
+      int i;
+
+      __m128i xmm = _mm_setzero_si128 ();
+      if (memcmp (&outregs->lrv_xmm0, &xmm, sizeof (xmm))
+	  || memcmp (&outregs->lrv_vector0, &xmm, sizeof (xmm)))
+	abort ();
+
+      __m256i ymm;
+
+      for (i = 0; i < 8; i += 2)
+	{
+	  xmm = _mm_set1_epi32 (i + 0x100);
+	  if (memcmp (&inregs->lr_xmm[i], &xmm, sizeof (xmm))
+	      || memcmp (&inregs->lr_vector[i], &xmm, sizeof (xmm)))
+	    abort ();
+
+	  ymm = _mm256_set1_epi32 (i + 0x101);
+	  if (memcmp (&inregs->lr_xmm[i + 1],
+		      &inregs->lr_vector[i + 1].xmm[0], sizeof (xmm))
+	      || memcmp (&inregs->lr_vector[i + 1], &ymm, sizeof (ymm)))
+	    abort ();
+	}
+
+      outregs->lrv_vector0.ymm[0]
+	= (La_x86_64_ymm) _mm256_set1_epi32 (0x98abcdef);
+
+      ymm = _mm256_set1_epi32 (-1);
+      asm volatile ("vmovdqa %0, %%ymm0" : : "x" (ymm) : "xmm0" );
+      asm volatile ("vmovdqa %0, %%ymm1" : : "x" (ymm) : "xmm1" );
+    }
+#endif
+
+  return 0;
+}