PATCH: Save and restore xmm6 in x86-64 _dl_runtime_profile

["H.J. Lu" <hongjiu dot lu at intel dot com>]

_dl_runtime_profile in sysdeps/x86_64/dl-trampoline.S doesn't preserve
xmm6.  This patch fixes it and added a testcase.


H.J.
---
2009-06-30  H.J. Lu  <hongjiu.lu@intel.com>

	* elf/Makefile (distribute): Remove tst-audit.sh.  Add
	tst-audit2.c tst-audit3.c tst-auditmod3a.c tst-auditmod3b.c.
	(tests): Add tst-audit3 for x86_64.
	(modules-names): Add tst-auditmod3a tst-auditmod3b.
	($(objpfx)tst-audit3): New.
	($(objpfx)tst-audit3.out): Likewise.
	(tst-audit3-ENV): Likewise.

	* elf/tst-audit3.c: New.
	* elf/tst-auditmod3a.c: Likewise.
	* elf/tst-auditmod3b.c: Likewise.

	* sysdeps/x86_64/dl-trampoline.S (_dl_runtime_profile): Save
	and restore xmm6.

diff --git a/elf/Makefile b/elf/Makefile
index 56935d5..6c1a04c 100644
--- elf/Makefile
+++ elf/Makefile
@@ -89,7 +89,8 @@ distribute	:= rtld-Rules \
 		   unload4mod1.c unload4mod2.c unload4mod3.c unload4mod4.c \
 		   unload6mod1.c unload6mod2.c unload6mod3.c \
 		   unload7mod1.c unload7mod2.c \
-		   tst-auditmod1.c tst-audit.sh \
+		   tst-audit1.c tst-audit2.c tst-audit3.c \
+		   tst-auditmod1.c tst-auditmod3a.c tst-auditmod3b.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 \
@@ -193,6 +194,9 @@ tests += loadtest restest1 preloadtest loadfail multiload origtest resolvfail \
 #	 reldep9
 test-srcs = tst-pathopt
 tests-execstack-yes = tst-execstack tst-execstack-needed tst-execstack-prog
+ifeq (x86_64,$(config-machine))
+tests += tst-audit3
+endif
 endif
 ifeq (yesyes,$(have-fpie)$(build-shared))
 tests: $(objpfx)tst-pie1.out
@@ -230,6 +234,7 @@ modules-names = testobj1 testobj2 testobj3 testobj4 testobj5 testobj6 \
 		$(modules-execstack-$(have-z-execstack)) \
 		tst-dlopenrpathmod tst-deep1mod1 tst-deep1mod2 tst-deep1mod3 \
 		tst-dlmopen1mod tst-auditmod1 \
+		tst-auditmod3a tst-auditmod3b \
 		unload3mod1 unload3mod2 unload3mod3 unload3mod4 \
 		unload4mod1 unload4mod2 unload4mod3 unload4mod4 \
 		unload6mod1 unload6mod2 unload6mod3 \
@@ -959,6 +964,10 @@ tst-audit1-ENV = LD_AUDIT=$(objpfx)tst-auditmod1.so
 $(objpfx)tst-audit2.out: $(objpfx)tst-auditmod1.so
 tst-audit2-ENV = LD_AUDIT=$(objpfx)tst-auditmod1.so
 
+$(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-global1: $(libdl)
 $(objpfx)tst-global1.out: $(objpfx)testobj6.so $(objpfx)testobj2.so
 
diff --git a/sysdeps/x86_64/dl-trampoline.S b/sysdeps/x86_64/dl-trampoline.S
index d8d9bc1..33e6115 100644
--- sysdeps/x86_64/dl-trampoline.S
+++ sysdeps/x86_64/dl-trampoline.S
@@ -107,7 +107,8 @@ _dl_runtime_profile:
 	movaps %xmm3, 112(%rsp)
 	movaps %xmm4, 128(%rsp)
 	movaps %xmm5, 144(%rsp)
-	movaps %xmm7, 160(%rsp)
+	movaps %xmm6, 160(%rsp)
+	movaps %xmm7, 176(%rsp)
 
 	movq %rsp, %rcx		# La_x86_64_regs pointer to %rcx.
 	movq 48(%rbx), %rdx	# Load return address if needed.
@@ -128,7 +129,8 @@ _dl_runtime_profile:
 	movaps 112(%rsp), %xmm3
 	movaps 128(%rsp), %xmm4
 	movaps 144(%rsp), %xmm5
-	movaps 160(%rsp), %xmm7
+	movaps 160(%rsp), %xmm6
+	movaps 176(%rsp), %xmm7
 
 	movq 16(%rbx), %r10	# Anything in framesize?
 	testq %r10, %r10
--- /dev/null	2009-06-26 17:33:06.493000902 -0700
+++ elf/tst-audit3.c	2009-06-30 14:00:28.000000000 -0700
@@ -0,0 +1,20 @@
+/* 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);
+  
+  if (memcmp (&xmm, &ret, sizeof (ret)))
+    abort ();
+
+  return 0;
+}
--- /dev/null	2009-06-26 17:33:06.493000902 -0700
+++ elf/tst-auditmod3a.c	2009-06-30 14:00:28.000000000 -0700
@@ -0,0 +1,28 @@
+/* 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 = _mm_setzero_si128 ();
+  
+  printf ("Enter %s\n", __PRETTY_FUNCTION__);
+  if (memcmp (&xmm, &x0, sizeof (xmm))
+      || memcmp (&xmm, &x1, sizeof (xmm))
+      || memcmp (&xmm, &x2, sizeof (xmm))
+      || memcmp (&xmm, &x3, sizeof (xmm))
+      || memcmp (&xmm, &x4, sizeof (xmm))
+      || memcmp (&xmm, &x5, sizeof (xmm))
+      || memcmp (&xmm, &x6, sizeof (xmm))
+      || memcmp (&xmm, &x7, sizeof (xmm)))
+    {
+      printf ("Failed %s\n", __PRETTY_FUNCTION__);
+      abort ();
+    }
+
+  return xmm;
+}
--- /dev/null	2009-06-26 17:33:06.493000902 -0700
+++ elf/tst-auditmod3b.c	2009-06-30 14:11:04.000000000 -0700
@@ -0,0 +1,156 @@
+/* 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)
+{
+  printf ("preinit\n");
+}
+
+unsigned int
+la_objclose  (uintptr_t *cookie)
+{
+  printf ("objclose\n");
+  return 0;
+}
+
+uintptr_t
+la_symbind32 (Elf32_Sym *sym, unsigned int ndx, uintptr_t *refcook,
+	      uintptr_t *defcook, unsigned int *flags, const char *symname)
+{
+  printf ("symbind32: symname=%s, st_value=%#lx, ndx=%u, flags=%u\n",
+	  symname, (long int) sym->st_value, ndx, *flags);
+
+  return sym->st_value;
+}
+
+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>
+
+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);
+
+  __m128i xmm = _mm_set1_epi32 (-1);
+  asm volatile ("movdqa %0, %%xmm0" : : "x" (xmm) : "xmm0" );
+  asm volatile ("movdqa %0, %%xmm1" : : "x" (xmm) : "xmm1" );
+  asm volatile ("movdqa %0, %%xmm2" : : "x" (xmm) : "xmm2" );
+  asm volatile ("movdqa %0, %%xmm3" : : "x" (xmm) : "xmm3" );
+  asm volatile ("movdqa %0, %%xmm4" : : "x" (xmm) : "xmm4" );
+  asm volatile ("movdqa %0, %%xmm5" : : "x" (xmm) : "xmm5" );
+  asm volatile ("movdqa %0, %%xmm6" : : "x" (xmm) : "xmm6" );
+  asm volatile ("movdqa %0, %%xmm7" : : "x" (xmm) : "xmm7" );
+
+  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);
+
+  __m128i xmm = _mm_set1_epi32 (-1);
+  asm volatile ("movdqa %0, %%xmm0" : : "x" (xmm) : "xmm0" );
+  asm volatile ("movdqa %0, %%xmm1" : : "x" (xmm) : "xmm1" );
+  asm volatile ("movdqa %0, %%xmm2" : : "x" (xmm) : "xmm2" );
+  asm volatile ("movdqa %0, %%xmm3" : : "x" (xmm) : "xmm3" );
+  asm volatile ("movdqa %0, %%xmm4" : : "x" (xmm) : "xmm4" );
+  asm volatile ("movdqa %0, %%xmm5" : : "x" (xmm) : "xmm5" );
+  asm volatile ("movdqa %0, %%xmm6" : : "x" (xmm) : "xmm6" );
+  asm volatile ("movdqa %0, %%xmm7" : : "x" (xmm) : "xmm7" );
+
+  return 0;
+}