Add AVX support to ld.so auditing for x86-64.

19 files changed, 1018 insertions, 126 deletions

diff --git a/ChangeLog b/ChangeLog
index 04760b065c..e90d19fc78 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,37 @@
+2009-07-07  H.J. Lu  <hongjiu.lu@intel.com>
+
+	* config.h.in: Add HAVE_AVX_SUPPORT entry.
+	* config.make.in: Add config-cflags-avx entry.
+	* configure.in: Substitute libc_cv_cc_avx.
+	* elf/Makefile: Add rules to build and run tst-audit4 and tst-audit5.
+	* elf/tst-audit4.c: New file.
+	* elf/tst-audit5.c: New file.
+	* elf/tst-auditmod4a.c: New file.
+	* elf/tst-auditmod4b.c: New file.
+	* elf/tst-auditmod5a.c: New file.
+	* elf/tst-auditmod5b.c: New file.
+	* sysdeps/x86_64/Makefile (gen-as-const-headers): Add
+	link-defines.sym.
+	* sysdeps/x86_64/bits/link.h (La_x86_64_ymm): New.
+	(La_x86_64_vector): Likewise.
+	(La_x86_64_regs): Append lr_vector.
+	(La_x86_64_retval): Append lr_vector0/lrv_vector1.
+	* sysdeps/x86_64/dl-trampoline.S (_dl_runtime_profile): Move
+	saving and restoring SSE registers to ...
+	* sysdeps/x86_64/dl-trampoline.h: This.  New file.
+	* sysdeps/x86_64/dl-trampoline.S: Include <config.h> and
+	<link-defines.h>.
+	(_dl_runtime_profile): Use LR_SIZE to allocate space for
+	La_x86_64_regs.  Allocate extra space and jump to memory at
+	save_and_restore_vector if HAVE_AVX_SUPPORT is defined.
+	(save_and_restore_vector_sse): New.
+	(save_and_restore_vector_avx): Likewise.
+	(check_avx): Likewise.
+	(save_and_restore_vector): Likewise.
+	* sysdeps/x86_64/elf/configure.in: Set libc_cv_cc_avx and
+	HAVE_AVX_SUPPORT.
+	* sysdeps/x86_64/link-defines.sym: New file.
+
 2009-07-10  Ulrich Drepper  <drepper@redhat.com>
 
 	* elf/do-lookup.h: Removed after folding content into...
diff --git a/config.h.in b/config.h.in
index 5f16874584..18bf01a38c 100644
--- a/config.h.in
+++ b/config.h.in
@@ -135,6 +135,9 @@
 /* Define if gcc supports SSE4.  */
 #undef	HAVE_SSE4_SUPPORT
 
+/* Define if gcc supports AVX.  */
+#undef	HAVE_AVX_SUPPORT
+
 /* Define if the compiler's exception support is based on libunwind.  */
 #undef	HAVE_CC_WITH_LIBUNWIND
 
diff --git a/config.make.in b/config.make.in
index 5fb5c8110c..d65706ceac 100644
--- a/config.make.in
+++ b/config.make.in
@@ -35,6 +35,7 @@ cflags-cpu = @libc_cv_cc_submachine@
 asflags-cpu = @libc_cv_cc_submachine@
 
 config-cflags-sse4 = @libc_cv_cc_sse4@
+config-cflags-avx = @libc_cv_cc_avx@
 
 defines = @DEFINES@
 sysincludes = @SYSINCLUDES@
diff --git a/configure b/configure
index 4e49f7068b..48e6952bbd 100755
--- a/configure
+++ b/configure
@@ -657,6 +657,7 @@ xcoff
 elf
 ldd_rewrite_script
 use_ldconfig
+libc_cv_cc_avx
 libc_cv_cc_sse4
 libc_cv_cpp_asm_debuginfo
 libc_cv_forced_unwind
@@ -8772,6 +8773,7 @@ fi
 
 
 
+
 if test $elf = yes; then
   cat >>confdefs.h <<\_ACEOF
 #define HAVE_ELF 1
diff --git a/configure.in b/configure.in
index 61c87418e6..4584afe605 100644
--- a/configure.in
+++ b/configure.in
@@ -2277,6 +2277,7 @@ AC_SUBST(libc_cv_forced_unwind)
 dnl sysdeps/CPU/configure.in checks set this via arch-specific asm tests
 AC_SUBST(libc_cv_cpp_asm_debuginfo)
 AC_SUBST(libc_cv_cc_sse4)
+AC_SUBST(libc_cv_cc_avx)
 
 AC_SUBST(use_ldconfig)
 AC_SUBST(ldd_rewrite_script)
diff --git a/elf/Makefile b/elf/Makefile
index 3e656ae0ad..e4b977e9e3 100644
--- a/elf/Makefile
+++ b/elf/Makefile
@@ -89,8 +89,10 @@ distribute	:= rtld-Rules \
 		   unload4mod1.c unload4mod2.c unload4mod3.c unload4mod4.c \
 		   unload6mod1.c unload6mod2.c unload6mod3.c \
 		   unload7mod1.c unload7mod2.c \
-		   tst-audit1.c tst-audit2.c tst-audit3.c \
+		   tst-audit1.c tst-audit2.c tst-audit3.c tst-audit4.c \
 		   tst-auditmod1.c tst-auditmod3a.c tst-auditmod3b.c \
+		   tst-auditmod4a.c tst-auditmod4b.c \
+		   tst-audit5.c tst-auditmod5a.c tst-auditmod5b.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 \
@@ -198,7 +200,7 @@ 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,$(config-machine))
-tests += tst-audit3
+tests += tst-audit3 tst-audit4 tst-audit5
 endif
 endif
 ifeq (yesyes,$(have-fpie)$(build-shared))
@@ -238,6 +240,8 @@ modules-names = testobj1 testobj2 testobj3 testobj4 testobj5 testobj6 \
 		tst-dlopenrpathmod tst-deep1mod1 tst-deep1mod2 tst-deep1mod3 \
 		tst-dlmopen1mod tst-auditmod1 \
 		tst-auditmod3a tst-auditmod3b \
+		tst-auditmod4a tst-auditmod4b \
+		tst-auditmod5a tst-auditmod5b \
 		unload3mod1 unload3mod2 unload3mod3 unload3mod4 \
 		unload4mod1 unload4mod2 unload4mod3 unload4mod4 \
 		unload6mod1 unload6mod2 unload6mod3 \
@@ -973,6 +977,14 @@ $(objpfx)tst-audit3: $(objpfx)tst-auditmod3a.so
 $(objpfx)tst-audit3.out: $(objpfx)tst-auditmod3b.so
 tst-audit3-ENV = LD_AUDIT=$(objpfx)tst-auditmod3b.so
 
+$(objpfx)tst-audit4: $(objpfx)tst-auditmod4a.so
+$(objpfx)tst-audit4.out: $(objpfx)tst-auditmod4b.so
+tst-audit4-ENV = LD_AUDIT=$(objpfx)tst-auditmod4b.so
+
+$(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-global1: $(libdl)
 $(objpfx)tst-global1.out: $(objpfx)testobj6.so $(objpfx)testobj2.so
 
@@ -1115,3 +1127,9 @@ $(objpfx)tst-unique1.out: $(objpfx)tst-unique1mod1.so \
 
 $(objpfx)tst-unique2: $(libdl) $(objpfx)tst-unique2mod1.so
 $(objpfx)tst-unique2.out: $(objpfx)tst-unique2mod2.so
+
+ifeq (yes,$(config-cflags-avx))
+CFLAGS-tst-audit4.c += -mavx
+CFLAGS-tst-auditmod4a.c += -mavx
+CFLAGS-tst-auditmod4b.c += -mavx
+endif
diff --git a/elf/tst-audit4.c b/elf/tst-audit4.c
new file mode 100644
index 0000000000..b17d4a61a7
--- /dev/null
+++ b/elf/tst-audit4.c
@@ -0,0 +1,35 @@
+/* Test case for x86-64 preserved registers in dynamic linker.  */
+
+#ifdef __AVX__
+#include <stdlib.h>
+#include <string.h>
+#include <cpuid.h>
+#include <immintrin.h>
+
+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))
+    {
+      __m256i ymm = _mm256_setzero_si256 ();
+      __m256i ret = audit_test (ymm, ymm, ymm, ymm, ymm, ymm, ymm, ymm);
+
+      ymm =  _mm256_set1_epi32 (0x12349876);
+      if (memcmp (&ymm, &ret, sizeof (ret)))
+	abort ();
+    }
+  return 0;
+}
+#else
+int
+main (void)
+{
+  return 0;
+}
+#endif
diff --git a/elf/tst-audit5.c b/elf/tst-audit5.c
new file mode 100644
index 0000000000..0094fee61f
--- /dev/null
+++ b/elf/tst-audit5.c
@@ -0,0 +1,21 @@
+/* Test case for x86-64 preserved registers in dynamic linker.  */
+
+#include <stdlib.h>
+#include <string.h>
+
+#include <emmintrin.h>
+
+extern __m128i audit_test (__m128i, __m128i, __m128i, __m128i,
+			   __m128i, __m128i, __m128i, __m128i);
+int
+main (void)
+{
+  __m128i xmm = _mm_setzero_si128 ();
+  __m128i ret = audit_test (xmm, xmm, xmm, xmm, xmm, xmm, xmm, xmm);
+
+  xmm = _mm_set1_epi32 (0x12349876);
+  if (memcmp (&xmm, &ret, sizeof (ret)))
+    abort ();
+
+  return 0;
+}
diff --git a/elf/tst-auditmod4a.c b/elf/tst-auditmod4a.c
new file mode 100644
index 0000000000..c9c24c04a8
--- /dev/null
+++ b/elf/tst-auditmod4a.c
@@ -0,0 +1,48 @@
+/* Test case for x86-64 preserved registers in dynamic linker.  */
+
+#ifdef __AVX__
+#include <stdlib.h>
+#include <string.h>
+#include <immintrin.h>
+
+__m256i
+audit_test (__m256i x0, __m256i x1, __m256i x2, __m256i x3,
+	    __m256i x4, __m256i x5, __m256i x6, __m256i x7)
+{
+  __m256i ymm;
+
+  ymm = _mm256_set1_epi32 (1);
+  if (memcmp (&ymm, &x0, sizeof (ymm)))
+    abort ();
+
+  ymm = _mm256_set1_epi32 (2);
+  if (memcmp (&ymm, &x1, sizeof (ymm)))
+    abort ();
+
+  ymm = _mm256_set1_epi32 (3);
+  if (memcmp (&ymm, &x2, sizeof (ymm)))
+    abort ();
+
+  ymm = _mm256_set1_epi32 (4);
+  if (memcmp (&ymm, &x3, sizeof (ymm)))
+    abort ();
+
+  ymm = _mm256_set1_epi32 (5);
+  if (memcmp (&ymm, &x4, sizeof (ymm)))
+    abort ();
+
+  ymm = _mm256_set1_epi32 (6);
+  if (memcmp (&ymm, &x5, sizeof (ymm)))
+    abort ();
+
+  ymm = _mm256_set1_epi32 (7);
+  if (memcmp (&ymm, &x6, sizeof (ymm)))
+    abort ();
+
+  ymm = _mm256_set1_epi32 (8);
+  if (memcmp (&ymm, &x7, sizeof (ymm)))
+    abort ();
+
+  return _mm256_setzero_si256 ();
+}
+#endif
diff --git a/elf/tst-auditmod4b.c b/elf/tst-auditmod4b.c
new file mode 100644
index 0000000000..a6d3c6a6c5
--- /dev/null
+++ b/elf/tst-auditmod4b.c
@@ -0,0 +1,206 @@
+/* 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)
+    {
+      __m256i zero = _mm256_setzero_si256 ();
+      if (memcmp (&regs->lr_vector[0], &zero, sizeof (zero))
+	  || memcmp (&regs->lr_vector[1], &zero, sizeof (zero))
+	  || memcmp (&regs->lr_vector[2], &zero, sizeof (zero))
+	  || memcmp (&regs->lr_vector[3], &zero, sizeof (zero))
+	  || memcmp (&regs->lr_vector[4], &zero, sizeof (zero))
+	  || memcmp (&regs->lr_vector[5], &zero, sizeof (zero))
+	  || memcmp (&regs->lr_vector[6], &zero, sizeof (zero))
+	  || memcmp (&regs->lr_vector[7], &zero, sizeof (zero)))
+	abort ();
+
+      for (int i = 0; i < 8; i++)
+	regs->lr_vector[i].ymm[0]
+	  = (La_x86_64_ymm) _mm256_set1_epi32 (i + 1);
+
+      __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)
+    {
+      __m256i zero = _mm256_setzero_si256 ();
+      if (memcmp (&outregs->lrv_vector0, &zero, sizeof (zero)))
+	abort ();
+
+      for (int i = 0; i < 8; i++)
+	{
+	  __m256i ymm = _mm256_set1_epi32 (i + 1);
+	  if (memcmp (&inregs->lr_vector[i], &ymm, sizeof (ymm)) != 0)
+	    abort ();
+	}
+
+      outregs->lrv_vector0.ymm[0]
+	= (La_x86_64_ymm) _mm256_set1_epi32 (0x12349876);
+
+      __m256i 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-auditmod5a.c b/elf/tst-auditmod5a.c
new file mode 100644
index 0000000000..8511a70747
--- /dev/null
+++ b/elf/tst-auditmod5a.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 (1);
+  if (memcmp (&xmm, &x0, sizeof (xmm)))
+    abort ();
+
+  xmm =  _mm_set1_epi32 (2);
+  if (memcmp (&xmm, &x1, sizeof (xmm)))
+    abort ();
+
+  xmm =  _mm_set1_epi32 (3);
+  if (memcmp (&xmm, &x2, sizeof (xmm)))
+    abort ();
+
+  xmm =  _mm_set1_epi32 (4);
+  if (memcmp (&xmm, &x3, sizeof (xmm)))
+    abort ();
+
+  xmm =  _mm_set1_epi32 (5);
+  if (memcmp (&xmm, &x4, sizeof (xmm)))
+    abort ();
+
+  xmm =  _mm_set1_epi32 (6);
+  if (memcmp (&xmm, &x5, sizeof (xmm)))
+    abort ();
+
+  xmm =  _mm_set1_epi32 (7);
+  if (memcmp (&xmm, &x6, sizeof (xmm)))
+    abort ();
+
+  xmm =  _mm_set1_epi32 (8);
+  if (memcmp (&xmm, &x7, sizeof (xmm)))
+    abort ();
+
+  return _mm_setzero_si128 ();
+}
diff --git a/elf/tst-auditmod5b.c b/elf/tst-auditmod5b.c
new file mode 100644
index 0000000000..7e1e941126
--- /dev/null
+++ b/elf/tst-auditmod5b.c
@@ -0,0 +1,178 @@
+/* Verify that changing xmm 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>
+#include <emmintrin.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)
+{