Refine testing for xmm/ymm register use in x86-64 ld.so.

The test now takes the callgraph into account.  Only code called
during runtime relocation is affected by the limitation.  We now
determine the affected object files as closely as possible from
the outside.  This allowed to remove some the specializations
for some of the string functions as they are only used in other
code paths.

6 files changed, 86 insertions, 12 deletions

diff --git a/ChangeLog b/ChangeLog
index e31b72ff65..60b76547c1 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,19 @@
 2009-07-27  Ulrich Drepper  <drepper@redhat.com>
 
+	* sysdeps/x86_64/tst-xmmymm.sh: Refine testing.  The script now
+	determines which files are used in runtime lookups and only checks
+	those for SSE use.
+	* sysdeps/x86_64/rtld-memchr.c: Removed.  Not needed with refined
+	testing.
+	* sysdeps/x86_64/rtld-rawmemchr.c: Removed.
+	* sysdeps/x86_64/multiarch/rtld-rawmemchr.c: Removed
+	* sysdeps/x86_64/Makefile: Emit warning that tst-xmmymm.sh might
+	take a while.
+
+	* elf/dl-open.c: Move _dl_scope_free to...
+	* elf/dl-scope.c: ...here.  New file.
+	* elf/Makefile (dl-routines): Add scope.
+
 	* resolv/resolv.h (RES_USE_DNSSEC): Define.
 	* resolv/res_debug.c (p_option): Handle RES_USE_EDNS0 and
 	RES_USE_DNSSEC.
diff --git a/sysdeps/x86_64/Makefile b/sysdeps/x86_64/Makefile
index 57cd88432a..e8d0285e26 100644
--- a/sysdeps/x86_64/Makefile
+++ b/sysdeps/x86_64/Makefile
@@ -22,6 +22,7 @@ sysdep-rtld-routines += tlsdesc dl-tlsdesc
 
 tests: $(objpfx)tst-xmmymm.out
 $(objpfx)tst-xmmymm.out: ../sysdeps/x86_64/tst-xmmymm.sh $(objpfx)ld.so
+	@echo "Checking ld.so for SSE register use.  This will take a few seconds..."
 	$(SHELL) -e $< $(objpfx) > $@
 endif
 
diff --git a/sysdeps/x86_64/multiarch/rtld-rawmemchr.c b/sysdeps/x86_64/multiarch/rtld-rawmemchr.c
deleted file mode 100644
index 53a90675ab..0000000000
--- a/sysdeps/x86_64/multiarch/rtld-rawmemchr.c
+++ /dev/null
@@ -1 +0,0 @@
-#include "../rtld-rawmemchr.c"
diff --git a/sysdeps/x86_64/rtld-memchr.c b/sysdeps/x86_64/rtld-memchr.c
deleted file mode 100644
index f63fefbcec..0000000000
--- a/sysdeps/x86_64/rtld-memchr.c
+++ /dev/null
@@ -1 +0,0 @@
-#include <string/memchr.c>
diff --git a/sysdeps/x86_64/rtld-rawmemchr.c b/sysdeps/x86_64/rtld-rawmemchr.c
deleted file mode 100644
index 2b9189393c..0000000000
--- a/sysdeps/x86_64/rtld-rawmemchr.c
+++ /dev/null
@@ -1 +0,0 @@
-#include <string/rawmemchr.c>
diff --git a/sysdeps/x86_64/tst-xmmymm.sh b/sysdeps/x86_64/tst-xmmymm.sh
index 0735276e6d..a576e7da0d 100755
--- a/sysdeps/x86_64/tst-xmmymm.sh
+++ b/sysdeps/x86_64/tst-xmmymm.sh
@@ -1,17 +1,79 @@
-#! /bin/sh
+#! /bin/bash
 objpfx="$1"
 
 tmp=$(mktemp ${objpfx}tst-xmmymm.XXXXXX)
 trap 'rm -f "$tmp"' 1 2 3 15
 
-objdump -d "${objpfx}ld.so" |
-awk 'BEGIN { last="" } /^[[:xdigit:]]* <[_[:alnum:]]*>:$/ { fct=substr($2, 2, length($2)-3) } /,%[xy]mm[[:digit:]]*$/ { if (last != fct) { print fct; last=fct} }' |
-tee "$tmp"
+# List of object files we have to test
+rtldobjs=$(readelf -W -wi ${objpfx}dl-allobjs.os |
+    awk '/^ </ { if ($5 == "(DW_TAG_compile_unit)") c=1; else c=0 } $2 == "DW_AT_name" { if (c == 1) print $NF }' |
+    sed 's,\(.*/\|\)\([_[:alnum:]-]*[.]\).$,\2os,')
+rtldobjs="$rtldobjs $(ar t ${objpfx}rtld-libc.a)"
 
-echo "Functions which incorrectly modify xmm/ymm registers:"
-err=1
-egrep -vs '^_dl_runtime_profile$' "$tmp" || err=0
-if test $err -eq 0; then echo "None"; fi
+# OBJECT symbols can be ignored.
+readelf -sW ${objpfx}dl-allobjs.os ${objpfx}rtld-libc.a |
+egrep " OBJECT  *GLOBAL " |
+awk '{if ($7 != "ABS") print $8 }' |
+sort -u > "$tmp"
+declare -a objects
+objects=($(cat "$tmp"))
+
+objs="dl-runtime.os"
+tocheck="dl-runtime.os"
+
+while test -n "$objs"; do
+  this="$objs"
+  objs=""
+
+  for f in $this; do
+    undef=$(nm -u "$objpfx"../*/"$f" | awk '{print $2}')
+    if test -n "$undef"; then
+      for s in $undef; do
+	for obj in ${objects[*]} "_GLOBAL_OFFSET_TABLE_"; do
+	  if test "$obj" = "$s"; then
+	    continue 2
+	  fi
+	done
+        for o in $rtldobjs; do
+	  ro=$(echo "$objpfx"../*/"$o")
+	  if nm -g --defined-only "$ro" | egrep -qs " $s\$"; then
+	    if ! (echo "$tocheck $objs" | fgrep -qs "$o"); then
+	      echo "$o needed for $s"
+	      objs="$objs $o"
+	    fi
+	    break;
+	  fi
+	done
+      done
+    fi
+  done
+  tocheck="$tocheck$objs"
+done
+
+echo
+echo
+echo "object files needed: $tocheck"
+
+cp /dev/null "$tmp"
+for f in $tocheck; do
+  objdump -d "$objpfx"../*/"$f" |
+  awk 'BEGIN { last="" } /^[[:xdigit:]]* <[_[:alnum:]]*>:$/ { fct=substr($2, 2, length($2)-3) } /,%[xy]mm[[:digit:]]*$/ { if (last != fct) { print fct; last=fct} }' |
+  while read fct; do
+    if test "$fct" != "_dl_runtime_profile"; then
+      echo "function $fct in $f modifies xmm/ymm" >> "$tmp"
+      result=1
+    fi
+  done
+done
+
+if test -s "$tmp"; then
+  echo
+  echo
+  cat "$tmp"
+  result=1
+else
+  result=0
+fi
 
 rm "$tmp"
-exit $err
+exit $result