[PATCH 0/7] Reliable CPU compatibility diagnostics in ld.so

[PATCH 0/7] Reliable CPU compatibility diagnostics in ld.so

[Florian Weimer]

This series enables reliable CPU compatibility diagnostics by building
the early parts of the dynamic loader with an older ISA level than the
rest of glibc.

I tested this with our POWER9 build flags (running on POWER8), and with
-march=x86-64-v4 on a x86-64-v3 system.  In both cases, an accurate
erorr is printed.  Also built with build-many-glibcs.py, and tested on
i686-linux-gnu and x86_64-linux-gnu,.

Florian Weimer (7):
  x86: HAVE_X86_LAHF_SAHF, HAVE_X86_MOVBE and -march=x86-64-vN (bug
    28782)
  Generate gcc-macros.h
  elf: Split dl-printf.c from dl-misc.c
  Add --early-cflags configure option
  powerpc64le: Use <gcc-macros.h> in early HWCAP check
  x86: Add x86-64-vN check to early startup
  s390x: Use <gcc-macros.h> in early HWCAP check

 Makeconfig                                    |   9 +
 config.make.in                                |   1 +
 configure                                     |  13 +
 configure.ac                                  |   6 +
 elf/Makefile                                  |  10 +-
 elf/dl-misc.c                                 | 281 +----------------
 elf/dl-printf.c                               | 292 ++++++++++++++++++
 manual/install.texi                           |   7 +
 sysdeps/powerpc/powerpc64/le/dl-hwcap-check.h |   9 +-
 sysdeps/s390/s390-64/dl-hwcap-check.h         |   5 +-
 sysdeps/x86/Makefile                          |   1 +
 sysdeps/x86/configure                         |   4 +-
 sysdeps/x86/configure.ac                      |   4 +-
 sysdeps/x86/dl-get-cpu-features.c             |  31 +-
 14 files changed, 384 insertions(+), 289 deletions(-)
 create mode 100644 elf/dl-printf.c


base-commit: 4997a533ae4b51ef66a6b68862b7578a7acb82df
-- 
2.34.1

[PATCH 1/7] x86: HAVE_X86_LAHF_SAHF, HAVE_X86_MOVBE and -march=x86-64-vN (bug 28782)

[Florian Weimer]

HAVE_X86_LAHF_SAHF is implied by x86-64-v2, and HAVE_X86_MOVBE by
x86-64-v3.

The individual flag does not appear in -fverbose-asm flag output
even if the ISA level implies it.
---
 sysdeps/x86/configure    | 4 ++--
 sysdeps/x86/configure.ac | 4 ++--
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/sysdeps/x86/configure b/sysdeps/x86/configure
index 62676bb686..7bdbfdc6dc 100644
--- a/sysdeps/x86/configure
+++ b/sysdeps/x86/configure
@@ -155,7 +155,7 @@ else
   (eval $ac_try) 2>&5
   ac_status=$?
   $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
-  test $ac_status = 0; }; } | grep -q "\-msahf"; then
+  test $ac_status = 0; }; } | grep -qE '(-msahf\b|-march=x86-64-v)'; then
       libc_cv_have_x86_lahf_sahf=yes
     fi
 fi
@@ -176,7 +176,7 @@ else
   (eval $ac_try) 2>&5
   ac_status=$?
   $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
-  test $ac_status = 0; }; } | grep -q "\-mmovbe"; then
+  test $ac_status = 0; }; } | grep -qE '(-mmovbe\b|-march=x86-64-v([3-9]|[1-9][0-9]))'; then
       libc_cv_have_x86_movbe=yes
     fi
 fi
diff --git a/sysdeps/x86/configure.ac b/sysdeps/x86/configure.ac
index 04a12ab680..10d5c2e0e5 100644
--- a/sysdeps/x86/configure.ac
+++ b/sysdeps/x86/configure.ac
@@ -110,7 +110,7 @@ if test $libc_cv_include_x86_isa_level = yes; then
   AC_CACHE_CHECK([for LAHF/SAHF instruction support],
 		 libc_cv_have_x86_lahf_sahf, [dnl
     libc_cv_have_x86_lahf_sahf=no
-    if AC_TRY_COMMAND(${CC-cc} $CFLAGS $CPPFLAGS -fverbose-asm -S -o - -x c /dev/null) | grep -q "\-msahf"; then
+    if AC_TRY_COMMAND(${CC-cc} $CFLAGS $CPPFLAGS -fverbose-asm -S -o - -x c /dev/null) | grep -qE '(-msahf\b|-march=x86-64-v)'; then
       libc_cv_have_x86_lahf_sahf=yes
     fi])
   if test $libc_cv_have_x86_lahf_sahf = yes; then
@@ -119,7 +119,7 @@ if test $libc_cv_include_x86_isa_level = yes; then
   AC_CACHE_CHECK([for MOVBE instruction support],
 		 libc_cv_have_x86_movbe, [dnl
     libc_cv_have_x86_movbe=no
-    if AC_TRY_COMMAND(${CC-cc} $CFLAGS $CPPFLAGS -fverbose-asm -S -o - -x c /dev/null) | grep -q "\-mmovbe"; then
+    if AC_TRY_COMMAND(${CC-cc} $CFLAGS $CPPFLAGS -fverbose-asm -S -o - -x c /dev/null) | grep -qE '(-mmovbe\b|-march=x86-64-v(@<:@3-9@:>@|@<:@1-9@:>@@<:@0-9@:>@))'; then
       libc_cv_have_x86_movbe=yes
     fi])
   if test $libc_cv_have_x86_movbe = yes; then
-- 
2.34.1

[PATCH 2/7] Generate gcc-macros.h

[Florian Weimer]

The file can be used to check the effect of the default compiler
flags on code generation even in areas of the build that uses
non-default compiler flags.
---
 Makeconfig | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/Makeconfig b/Makeconfig
index 9b6fc6b08f..2e79077343 100644
--- a/Makeconfig
+++ b/Makeconfig
@@ -1202,6 +1202,15 @@ $(common-objpfx)dl-tunable-list.stmp: \
 	touch $@
 endif
 
+# Dump the GCC macros used by the default compiler flags to a header
+# file, so that they can be inspected when using different compiler
+# flags.  Add the GCCMACRO prefix to make these macro names unique.
+$(common-objpfx)gcc-macros.h.in: $(common-objpfx)config.status
+	$(CC) $(CFLAGS) $(CPPFLAGS) -E -dM -x c -o $@ /dev/null
+$(common-objpfx)gcc-macros.h: $(common-objpfx)gcc-macros.h.in
+	sed 's/^#define /#define GCCMACRO/' < $< > $@
+before-compile += $(common-objpfx)gcc-macros.h
+
 # Generate version maps, but wait until sysdep-subdirs is known
 ifeq ($(sysd-sorted-done),t)
 ifeq ($(build-shared),yes)
-- 
2.34.1

[PATCH 3/7] elf: Split dl-printf.c from dl-misc.c

[Florian Weimer]

This allows to use different compiler flags for the diagnostics
code.
---
 elf/Makefile    |   2 +-
 elf/dl-misc.c   | 281 +---------------------------------------------
 elf/dl-printf.c | 292 ++++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 297 insertions(+), 278 deletions(-)
 create mode 100644 elf/dl-printf.c

diff --git a/elf/Makefile b/elf/Makefile
index b86d116be9..0c4be43deb 100644
--- a/elf/Makefile
+++ b/elf/Makefile
@@ -31,7 +31,7 @@ routines	= $(all-dl-routines) dl-support dl-iteratephdr \
 # The core dynamic linking functions are in libc for the static and
 # profiled libraries.
 dl-routines	= $(addprefix dl-,load lookup object reloc deps \
-				  runtime init fini debug misc \
+				  runtime init fini debug misc printf \
 				  version profile tls origin scope \
 				  execstack open close trampoline \
 				  exception sort-maps lookup-direct \
diff --git a/elf/dl-misc.c b/elf/dl-misc.c
index 452b79de4a..6f40c28820 100644
--- a/elf/dl-misc.c
+++ b/elf/dl-misc.c
@@ -16,24 +16,16 @@
    License along with the GNU C Library; if not, see
    <https://www.gnu.org/licenses/>.  */
 
-#include <assert.h>
+#include <_itoa.h>
 #include <fcntl.h>
 #include <ldsodefs.h>
-#include <limits.h>
 #include <link.h>
-#include <stdarg.h>
-#include <stdlib.h>
-#include <string.h>
-#include <unistd.h>
+#include <not-cancel.h>
 #include <stdint.h>
+#include <stdlib.h>
 #include <sys/mman.h>
-#include <sys/param.h>
 #include <sys/stat.h>
-#include <sys/uio.h>
-#include <sysdep.h>
-#include <_itoa.h>
-#include <dl-writev.h>
-#include <not-cancel.h>
+#include <unistd.h>
 
 /* Read the whole contents of FILE into new mmap'd space with given
    protections.  *SIZEP gets the size of the file.  On error MAP_FAILED
@@ -70,270 +62,6 @@ _dl_sysdep_read_whole_file (const char *file, size_t *sizep, int prot)
   return result;
 }
 
-
-/* Bare-bones printf implementation.  This function only knows about
-   the formats and flags needed and can handle only up to 64 stripes in
-   the output.  */
-static void
-_dl_debug_vdprintf (int fd, int tag_p, const char *fmt, va_list arg)
-{
-# define NIOVMAX 64
-  struct iovec iov[NIOVMAX];
-  int niov = 0;
-  pid_t pid = 0;
-  char pidbuf[12];
-
-  while (*fmt != '\0')
-    {
-      const char *startp = fmt;
-
-      if (tag_p > 0)
-	{
-	  /* Generate the tag line once.  It consists of the PID and a
-	     colon followed by a tab.  */
-	  if (pid == 0)
-	    {
-	      char *p;
-	      pid = __getpid ();
-	      assert (pid >= 0 && sizeof (pid_t) <= 4);
-	      p = _itoa (pid, &pidbuf[10], 10, 0);
-	      while (p > pidbuf)
-		*--p = ' ';
-	      pidbuf[10] = ':';
-	      pidbuf[11] = '\t';
-	    }
-
-	  /* Append to the output.  */
-	  assert (niov < NIOVMAX);
-	  iov[niov].iov_len = 12;
-	  iov[niov++].iov_base = pidbuf;
-
-	  /* No more tags until we see the next newline.  */
-	  tag_p = -1;
-	}
-
-      /* Skip everything except % and \n (if tags are needed).  */
-      while (*fmt != '\0' && *fmt != '%' && (! tag_p || *fmt != '\n'))
-	++fmt;
-
-      /* Append constant string.  */
-      assert (niov < NIOVMAX);
-      if ((iov[niov].iov_len = fmt - startp) != 0)
-	iov[niov++].iov_base = (char *) startp;
-
-      if (*fmt == '%')
-	{
-	  /* It is a format specifier.  */
-	  char fill = ' ';
-	  int width = -1;
-	  int prec = -1;
-#if LONG_MAX != INT_MAX
-	  int long_mod = 0;
-#endif
-
-	  /* Recognize zero-digit fill flag.  */
-	  if (*++fmt == '0')
-	    {
-	      fill = '0';
-	      ++fmt;
-	    }
-
-	  /* See whether with comes from a parameter.  Note that no other
-	     way to specify the width is implemented.  */
-	  if (*fmt == '*')
-	    {
-	      width = va_arg (arg, int);
-	      ++fmt;
-	    }
-
-	  /* Handle precision.  */
-	  if (*fmt == '.' && fmt[1] == '*')
-	    {
-	      prec = va_arg (arg, int);
-	      fmt += 2;
-	    }
-
-	  /* Recognize the l modifier.  It is only important on some
-	     platforms where long and int have a different size.  We
-	     can use the same code for size_t.  */
-	  if (*fmt == 'l' || *fmt == 'Z')
-	    {
-#if LONG_MAX != INT_MAX
-	      long_mod = 1;
-#endif
-	      ++fmt;
-	    }
-
-	  switch (*fmt)
-	    {
-	      /* Integer formatting.  */
-	    case 'd':
-	    case 'u':
-	    case 'x':
-	      {
-		/* We have to make a difference if long and int have a
-		   different size.  */
-#if LONG_MAX != INT_MAX
-		unsigned long int num = (long_mod
-					 ? va_arg (arg, unsigned long int)
-					 : va_arg (arg, unsigned int));
-#else
-		unsigned long int num = va_arg (arg, unsigned int);
-#endif
-		bool negative = false;
-		if (*fmt == 'd')
-		  {
-#if LONG_MAX != INT_MAX
-		    if (long_mod)
-		      {
-			if ((long int) num < 0)
-			  negative = true;
-		      }
-		    else
-		      {
-			if ((int) num < 0)
-			  {
-			    num = (unsigned int) num;
-			    negative = true;
-			  }
-		      }
-#else
-		    if ((int) num < 0)
-		      negative = true;
-#endif
-		  }
-
-		/* We use alloca() to allocate the buffer with the most
-		   pessimistic guess for the size.  Using alloca() allows
-		   having more than one integer formatting in a call.  */
-		char *buf = (char *) alloca (1 + 3 * sizeof (unsigned long int));
-		char *endp = &buf[1 + 3 * sizeof (unsigned long int)];
-		char *cp = _itoa (num, endp, *fmt == 'x' ? 16 : 10, 0);
-
-		/* Pad to the width the user specified.  */
-		if (width != -1)
-		  while (endp - cp < width)
-		    *--cp = fill;
-
-		if (negative)
-		  *--cp = '-';
-
-		iov[niov].iov_base = cp;
-		iov[niov].iov_len = endp - cp;
-		++niov;
-	      }
-	      break;
-
-	    case 's':
-	      /* Get the string argument.  */
-	      iov[niov].iov_base = va_arg (arg, char *);
-	      iov[niov].iov_len = strlen (iov[niov].iov_base);
-	      if (prec != -1)
-		iov[niov].iov_len = MIN ((size_t) prec, iov[niov].iov_len);
-	      ++niov;
-	      break;
-
-	    case '%':
-	      iov[niov].iov_base = (void *) fmt;
-	      iov[niov].iov_len = 1;
-	      ++niov;
-	      break;
-
-	    default:
-	      assert (! "invalid format specifier");
-	    }
-	  ++fmt;
-	}
-      else if (*fmt == '\n')
-	{
-	  /* See whether we have to print a single newline character.  */
-	  if (fmt == startp)
-	    {
-	      iov[niov].iov_base = (char *) startp;
-	      iov[niov++].iov_len = 1;
-	    }
-	  else
-	    /* No, just add it to the rest of the string.  */
-	    ++iov[niov - 1].iov_len;
-
-	  /* Next line, print a tag again.  */
-	  tag_p = 1;
-	  ++fmt;
-	}
-    }
-
-  /* Finally write the result.  */
-  _dl_writev (fd, iov, niov);
-}
-
-
-/* Write to debug file.  */
-void
-_dl_debug_printf (const char *fmt, ...)
-{
-  va_list arg;
-
-  va_start (arg, fmt);
-  _dl_debug_vdprintf (GLRO(dl_debug_fd), 1, fmt, arg);
-  va_end (arg);
-}
-
-
-/* Write to debug file but don't start with a tag.  */
-void
-_dl_debug_printf_c (const char *fmt, ...)
-{
-  va_list arg;
-
-  va_start (arg, fmt);
-  _dl_debug_vdprintf (GLRO(dl_debug_fd), -1, fmt, arg);
-  va_end (arg);
-}
-
-
-/* Write the given file descriptor.  */
-void
-_dl_dprintf (int fd, const char *fmt, ...)
-{
-  va_list arg;
-
-  va_start (arg, fmt);
-  _dl_debug_vdprintf (fd, 0, fmt, arg);
-  va_end (arg);
-}
-
-void
-_dl_printf (const char *fmt, ...)
-{
-  va_list arg;
-
-  va_start (arg, fmt);
-  _dl_debug_vdprintf (STDOUT_FILENO, 0, fmt, arg);
-  va_end (arg);
-}
-
-void
-_dl_error_printf (const char *fmt, ...)
-{
-  va_list arg;
-
-  va_start (arg, fmt);
-  _dl_debug_vdprintf (STDERR_FILENO, 0, fmt, arg);
-  va_end (arg);
-}
-
-void
-_dl_fatal_printf (const char *fmt, ...)
-{
-  va_list arg;
-
-  va_start (arg, fmt);
-  _dl_debug_vdprintf (STDERR_FILENO, 0, fmt, arg);
-  va_end (arg);
-  _exit (127);
-}
-rtld_hidden_def (_dl_fatal_printf)
-
 /* Test whether given NAME matches any of the names of the given object.  */
 int
 _dl_name_match_p (const char *name, const struct link_map *map)
@@ -354,7 +82,6 @@ _dl_name_match_p (const char *name, const struct link_map *map)
   return 0;
 }
 
-
 unsigned long int
 _dl_higher_prime_number (unsigned long int n)
 {
diff --git a/elf/dl-printf.c b/elf/dl-printf.c
new file mode 100644
index 0000000000..d3264ba96c
--- /dev/null
+++ b/elf/dl-printf.c
@@ -0,0 +1,292 @@
+/* printf implementation for the dynamic loader.
+   Copyright (C) 1997-2022 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
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <https://www.gnu.org/licenses/>.  */
+
+#include <_itoa.h>
+#include <assert.h>
+#include <dl-writev.h>
+#include <ldsodefs.h>
+#include <limits.h>
+#include <stdarg.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/uio.h>
+#include <unistd.h>
+
+/* Bare-bones printf implementation.  This function only knows about
+   the formats and flags needed and can handle only up to 64 stripes in
+   the output.  */
+static void
+_dl_debug_vdprintf (int fd, int tag_p, const char *fmt, va_list arg)
+{
+# define NIOVMAX 64
+  struct iovec iov[NIOVMAX];
+  int niov = 0;
+  pid_t pid = 0;
+  char pidbuf[12];
+
+  while (*fmt != '\0')
+    {
+      const char *startp = fmt;
+
+      if (tag_p > 0)
+	{
+	  /* Generate the tag line once.  It consists of the PID and a
+	     colon followed by a tab.  */
+	  if (pid == 0)
+	    {
+	      char *p;
+	      pid = __getpid ();
+	      assert (pid >= 0 && sizeof (pid_t) <= 4);
+	      p = _itoa (pid, &pidbuf[10], 10, 0);
+	      while (p > pidbuf)
+		*--p = ' ';
+	      pidbuf[10] = ':';
+	      pidbuf[11] = '\t';
+	    }
+
+	  /* Append to the output.  */
+	  assert (niov < NIOVMAX);
+	  iov[niov].iov_len = 12;
+	  iov[niov++].iov_base = pidbuf;
+
+	  /* No more tags until we see the next newline.  */
+	  tag_p = -1;
+	}
+
+      /* Skip everything except % and \n (if tags are needed).  */
+      while (*fmt != '\0' && *fmt != '%' && (! tag_p || *fmt != '\n'))
+	++fmt;
+
+      /* Append constant string.  */
+      assert (niov < NIOVMAX);
+      if ((iov[niov].iov_len = fmt - startp) != 0)
+	iov[niov++].iov_base = (char *) startp;
+
+      if (*fmt == '%')
+	{
+	  /* It is a format specifier.  */
+	  char fill = ' ';
+	  int width = -1;
+	  int prec = -1;
+#if LONG_MAX != INT_MAX
+	  int long_mod = 0;
+#endif
+
+	  /* Recognize zero-digit fill flag.  */
+	  if (*++fmt == '0')
+	    {
+	      fill = '0';
+	      ++fmt;
+	    }
+
+	  /* See whether with comes from a parameter.  Note that no other
+	     way to specify the width is implemented.  */
+	  if (*fmt == '*')
+	    {
+	      width = va_arg (arg, int);
+	      ++fmt;
+	    }
+
+	  /* Handle precision.  */
+	  if (*fmt == '.' && fmt[1] == '*')
+	    {
+	      prec = va_arg (arg, int);
+	      fmt += 2;
+	    }
+
+	  /* Recognize the l modifier.  It is only important on some
+	     platforms where long and int have a different size.  We
+	     can use the same code for size_t.  */
+	  if (*fmt == 'l' || *fmt == 'Z')
+	    {
+#if LONG_MAX != INT_MAX
+	      long_mod = 1;
+#endif
+	      ++fmt;
+	    }
+
+	  switch (*fmt)
+	    {
+	      /* Integer formatting.  */
+	    case 'd':
+	    case 'u':
+	    case 'x':
+	      {
+		/* We have to make a difference if long and int have a
+		   different size.  */
+#if LONG_MAX != INT_MAX
+		unsigned long int num = (long_mod
+					 ? va_arg (arg, unsigned long int)
+					 : va_arg (arg, unsigned int));
+#else
+		unsigned long int num = va_arg (arg, unsigned int);
+#endif
+		bool negative = false;
+		if (*fmt == 'd')
+		  {
+#if LONG_MAX != INT_MAX
+		    if (long_mod)
+		      {
+			if ((long int) num < 0)
+			  negative = true;
+		      }
+		    else
+		      {
+			if ((int) num < 0)
+			  {
+			    num = (unsigned int) num;
+			    negative = true;
+			  }
+		      }
+#else
+		    if ((int) num < 0)
+		      negative = true;
+#endif
+		  }
+
+		/* We use alloca() to allocate the buffer with the most
+		   pessimistic guess for the size.  Using alloca() allows
+		   having more than one integer formatting in a call.  */
+		char *buf = (char *) alloca (1 + 3 * sizeof (unsigned long int));
+		char *endp = &buf[1 + 3 * sizeof (unsigned long int)];
+		char *cp = _itoa (num, endp, *fmt == 'x' ? 16 : 10, 0);
+
+		/* Pad to the width the user specified.  */
+		if (width != -1)
+		  while (endp - cp < width)
+		    *--cp = fill;
+
+		if (negative)
+		  *--cp = '-';
+
+		iov[niov].iov_base = cp;
+		iov[niov].iov_len = endp - cp;
+		++niov;
+	      }
+	      break;
+
+	    case 's':
+	      /* Get the string argument.  */
+	      iov[niov].iov_base = va_arg (arg, char *);
+	      iov[niov].iov_len = strlen (iov[niov].iov_base);
+	      if (prec != -1)
+		iov[niov].iov_len = MIN ((size_t) prec, iov[niov].iov_len);
+	      ++niov;
+	      break;
+
+	    case '%':
+	      iov[niov].iov_base = (void *) fmt;
+	      iov[niov].iov_len = 1;
+	      ++niov;
+	      break;
+
+	    default:
+	      assert (! "invalid format specifier");
+	    }
+	  ++fmt;
+	}
+      else if (*fmt == '\n')
+	{
+	  /* See whether we have to print a single newline character.  */
+	  if (fmt == startp)
+	    {
+	      iov[niov].iov_base = (char *) startp;
+	      iov[niov++].iov_len = 1;
+	    }
+	  else
+	    /* No, just add it to the rest of the string.  */
+	    ++iov[niov - 1].iov_len;
+
+	  /* Next line, print a tag again.  */
+	  tag_p = 1;
+	  ++fmt;
+	}
+    }
+
+  /* Finally write the result.  */
+  _dl_writev (fd, iov, niov);
+}
+
+
+/* Write to debug file.  */
+void
+_dl_debug_printf (const char *fmt, ...)
+{
+  va_list arg;
+
+  va_start (arg, fmt);
+  _dl_debug_vdprintf (GLRO(dl_debug_fd), 1, fmt, arg);
+  va_end (arg);
+}
+
+
+/* Write to debug file but don't start with a tag.  */
+void
+_dl_debug_printf_c (const char *fmt, ...)
+{
+  va_list arg;
+
+  va_start (arg, fmt);
+  _dl_debug_vdprintf (GLRO(dl_debug_fd), -1, fmt, arg);
+  va_end (arg);
+}
+
+
+/* Write the given file descriptor.  */
+void
+_dl_dprintf (int fd, const char *fmt, ...)
+{
+  va_list arg;
+
+  va_start (arg, fmt);
+  _dl_debug_vdprintf (fd, 0, fmt, arg);
+  va_end (arg);
+}
+
+void
+_dl_printf (const char *fmt, ...)
+{
+  va_list arg;
+
+  va_start (arg, fmt);
+  _dl_debug_vdprintf (STDOUT_FILENO, 0, fmt, arg);
+  va_end (arg);
+}
+
+void
+_dl_error_printf (const char *fmt, ...)
+{
+  va_list arg;
+
+  va_start (arg, fmt);
+  _dl_debug_vdprintf (STDERR_FILENO, 0, fmt, arg);
+  va_end (arg);
+}
+
+void
+_dl_fatal_printf (const char *fmt, ...)
+{
+  va_list arg;
+
+  va_start (arg, fmt);
+  _dl_debug_vdprintf (STDERR_FILENO, 0, fmt, arg);
+  va_end (arg);
+  _exit (127);
+}
+rtld_hidden_def (_dl_fatal_printf)
-- 
2.34.1

[PATCH 4/7] Add --early-cflags configure option

[Florian Weimer]

---
 config.make.in      |  1 +
 configure           | 13 +++++++++++++
 configure.ac        |  6 ++++++
 elf/Makefile        |  8 ++++++++
 manual/install.texi |  7 +++++++
 5 files changed, 35 insertions(+)

diff --git a/config.make.in b/config.make.in
index e8630a8d0c..e06f1ff49f 100644
--- a/config.make.in
+++ b/config.make.in
@@ -110,6 +110,7 @@ CFLAGS = @CFLAGS@
 CPPFLAGS-config = @CPPFLAGS@
 CPPUNDEFS = @CPPUNDEFS@
 extra-nonshared-cflags = @extra_nonshared_cflags@
+early-cflags = @early_cflags@
 ASFLAGS-config = @ASFLAGS_config@
 AR = @AR@
 NM = @NM@
diff --git a/configure b/configure
index 3f956cf777..6eedd8c3a7 100755
--- a/configure
+++ b/configure
@@ -681,6 +681,7 @@ force_install
 bindnow
 hardcoded_path_in_tests
 enable_timezone_tools
+early_cflags
 extra_nonshared_cflags
 use_default_link
 sysheaders
@@ -761,6 +762,7 @@ with_selinux
 with_headers
 with_default_link
 with_nonshared_cflags
+with_early_cflags
 with_timeoutfactor
 enable_sanity_checks
 enable_shared
@@ -1480,6 +1482,8 @@ Optional Packages:
   --with-default-link     do not use explicit linker scripts
   --with-nonshared-cflags=CFLAGS
                           build nonshared libraries with additional CFLAGS
+  --with-early-cflags=CFLAGS
+                          build early initialization with additional CFLAGS
   --with-timeoutfactor=NUM
                           specify an integer to scale the timeout
   --with-cpu=CPU          select code for CPU variant
@@ -3386,6 +3390,15 @@ fi
 
 
 
+# Check whether --with-early-cflags was given.
+if test "${with_early_cflags+set}" = set; then :
+  withval=$with_early_cflags; early_cflags=$withval
+else
+  early_cflags=
+fi
+
+
+
 
 # Check whether --with-timeoutfactor was given.
 if test "${with_timeoutfactor+set}" = set; then :
diff --git a/configure.ac b/configure.ac
index 277d3527d2..445b0b7f15 100644
--- a/configure.ac
+++ b/configure.ac
@@ -162,6 +162,12 @@ AC_ARG_WITH([nonshared-cflags],
 	    [extra_nonshared_cflags=$withval],
 	    [extra_nonshared_cflags=])
 AC_SUBST(extra_nonshared_cflags)
+AC_ARG_WITH([early-cflags],
+	    AS_HELP_STRING([--with-early-cflags=CFLAGS],
+			   [build early initialization with additional CFLAGS]),
+	    [early_cflags=$withval],
+	    [early_cflags=])
+AC_SUBST(early_cflags)
 
 AC_ARG_WITH([timeoutfactor],
 	    AS_HELP_STRING([--with-timeoutfactor=NUM],
diff --git a/elf/Makefile b/elf/Makefile
index 0c4be43deb..9fcb4ebd13 100644
--- a/elf/Makefile
+++ b/elf/Makefile
@@ -99,6 +99,14 @@ CFLAGS-.o += $(call elide-stack-protector,.o,$(elide-routines.os))
 CFLAGS-.op += $(call elide-stack-protector,.op,$(elide-routines.os))
 CFLAGS-.os += $(call elide-stack-protector,.os,$(all-rtld-routines))
 
+# Add the requested compiler flags to the early startup code.
+CFLAGS-dl-printf.os += $(early-cflags)
+CFLAGS-dl-sysdep.os += $(early-cflags)
+CFLAGS-dl-tunables.os += $(early-cflags)
+CFLAGS-dl-write.os += $(early-cflags)
+CFLAGS-dl-writev.os += $(early-cflags)
+CFLAGS-rtld.os += $(early-cflags)
+
 ifeq ($(unwind-find-fde),yes)
 routines += unwind-dw2-fde-glibc
 shared-only-routines += unwind-dw2-fde-glibc
diff --git a/manual/install.texi b/manual/install.texi
index b1afc56f5a..27d5893c5e 100644
--- a/manual/install.texi
+++ b/manual/install.texi
@@ -131,6 +131,13 @@ that the objects in @file{libc_nonshared.a} are compiled with this flag
 (although this will not affect the generated code in this particular
 case and potentially change debugging information and metadata only).
 
+@item --with-early-cflags=@var{cflags}
+Use additional compiler flags @var{cflags} to build the early startup
+code of the dynamic linker.  These flags can be used to enable early
+dynamic linker diagnostics to run on CPUs which are not compatible with
+the rest of @theglibc{}, for example, due to compiler flags which target
+a later instruction set architecture (ISA).
+
 @item --with-timeoutfactor=@var{NUM}
 Specify an integer @var{NUM} to scale the timeout of test programs.
 This factor can be changed at run time using @env{TIMEOUTFACTOR}
-- 
2.34.1

[PATCH 5/7] powerpc64le: Use <gcc-macros.h> in early HWCAP check

[Florian Weimer]

This is required so that the checks still work if $(early-cflags)
selects a different ISA level.
---
 sysdeps/powerpc/powerpc64/le/dl-hwcap-check.h | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/sysdeps/powerpc/powerpc64/le/dl-hwcap-check.h b/sysdeps/powerpc/powerpc64/le/dl-hwcap-check.h
index 713a7f0313..b43c182683 100644
--- a/sysdeps/powerpc/powerpc64/le/dl-hwcap-check.h
+++ b/sysdeps/powerpc/powerpc64/le/dl-hwcap-check.h
@@ -19,17 +19,18 @@
 #ifndef _DL_HWCAP_CHECK_H
 #define _DL_HWCAP_CHECK_H
 
+#include <gcc-macros.h>
 #include <ldsodefs.h>
 
 static inline void
 dl_hwcap_check (void)
 {
-#ifdef _ARCH_PWR9
+#ifdef GCCMACRO_ARCH_PWR9
   if ((GLRO (dl_hwcap2) & PPC_FEATURE2_ARCH_3_00) == 0)
     _dl_fatal_printf ("\
 Fatal glibc error: CPU lacks ISA 3.00 support (POWER9 or later required)\n");
 #endif
-#ifdef __FLOAT128_HARDWARE__
+#ifdef GCCMACRO__FLOAT128_HARDWARE__
   if ((GLRO (dl_hwcap2) & PPC_FEATURE2_HAS_IEEE128) == 0)
     _dl_fatal_printf ("\
 Fatal glibc error: CPU lacks float128 support (POWER 9 or later required)\n");
@@ -37,12 +38,12 @@ Fatal glibc error: CPU lacks float128 support (POWER 9 or later required)\n");
    /* This check is not actually reached when building for POWER10 and
       running on POWER9 because there are faulting PCREL instructions
       before this point.  */
-#if defined _ARCH_PWR10 || defined __PCREL__
+#if defined GCCMACRO_ARCH_PWR10 || defined GCCMACRO__PCREL__
   if ((GLRO (dl_hwcap2) & PPC_FEATURE2_ARCH_3_1) == 0)
     _dl_fatal_printf ("\
 Fatal glibc error: CPU lacks ISA 3.10 support (POWER10 or later required)\n");
 #endif
-#ifdef __MMA__
+#ifdef GCCMACRO__MMA__
   if ((GLRO (dl_hwcap2) & PPC_FEATURE2_MMA) == 0)
     _dl_fatal_printf ("\
 Fatal glibc error: CPU lacks MMA support (POWER10 or later required)\n");
-- 
2.34.1

[PATCH 6/7] x86: Add x86-64-vN check to early startup

[Florian Weimer]

This ISA level covers the glibc build itself.  <dl-hwcap-check.h>
cannot be used because this check (by design) happens before
DL_PLATFORM_INIT and the x86 CPU flags initialization.
---
 sysdeps/x86/Makefile              |  1 +
 sysdeps/x86/dl-get-cpu-features.c | 31 ++++++++++++++++++++++++++++++-
 2 files changed, 31 insertions(+), 1 deletion(-)

diff --git a/sysdeps/x86/Makefile b/sysdeps/x86/Makefile
index 402986ff68..0c5ab42667 100644
--- a/sysdeps/x86/Makefile
+++ b/sysdeps/x86/Makefile
@@ -7,6 +7,7 @@ sysdep_routines += get-cpuid-feature-leaf
 sysdep-dl-routines += dl-get-cpu-features
 sysdep_headers += sys/platform/x86.h bits/platform/x86.h
 
+CFLAGS-dl-get-cpu-features.os += $(early-cflags)
 CFLAGS-get-cpuid-feature-leaf.o += $(no-stack-protector)
 
 tests += tst-get-cpu-features tst-get-cpu-features-static \
diff --git a/sysdeps/x86/dl-get-cpu-features.c b/sysdeps/x86/dl-get-cpu-features.c
index 6339c9df4e..4ec0e5d2af 100644
--- a/sysdeps/x86/dl-get-cpu-features.c
+++ b/sysdeps/x86/dl-get-cpu-features.c
@@ -20,6 +20,7 @@
 
 #ifdef SHARED
 # include <cpu-features.c>
+# include <gcc-macros.h>
 
 /* NB: Normally, DL_PLATFORM_INIT calls init_cpu_features to initialize
    CPU features in dynamic executable.  But when loading ld.so inside of
@@ -36,7 +37,35 @@ _dl_x86_init_cpu_features (void)
 {
   struct cpu_features *cpu_features = __get_cpu_features ();
   if (cpu_features->basic.kind == arch_kind_unknown)
-    init_cpu_features (cpu_features);
+    {
+      init_cpu_features (cpu_features);
+
+# if IS_IN (rtld)
+      /* See isa-level.c.  */
+#  if defined GCCMACRO__GCC_HAVE_SYNC_COMPARE_AND_SWAP_16		\
+  && defined HAVE_X86_LAHF_SAHF && defined GCCMACRO__POPCNT__		\
+  && defined GCCMACRO__SSE3__ && defined GCCMACRO__SSSE3__		\
+  && defined GCCMACRO__SSE4_1__ && defined GCCMACRO__SSE4_2__
+      if (!(cpu_features->isa_1 & GNU_PROPERTY_X86_ISA_1_V2))
+	_dl_fatal_printf ("\
+Fatal glibc error: CPU does not support x86-64-v%d\n", 2);
+#   if defined GCCMACRO__AVX__ && defined GCCMACRO__AVX2__ \
+  && defined GCCMACRO__F16C__ && defined GCCMACRO__FMA__   \
+  && defined GCCMACRO__LZCNT__ && defined HAVE_X86_MOVBE
+      if (!(cpu_features->isa_1 & GNU_PROPERTY_X86_ISA_1_V3))
+	_dl_fatal_printf ("\
+Fatal glibc error: CPU does not support x86-64-v%d\n", 3);
+#    if defined GCCMACRO__AVX512F__ && defined GCCMACRO__AVX512BW__ \
+     && defined GCCMACRO__AVX512CD__ && defined GCCMACRO__AVX512DQ__ \
+     && defined GCCMACRO__AVX512VL__
+      if (!(cpu_features->isa_1 & GNU_PROPERTY_X86_ISA_1_V4))
+	_dl_fatal_printf ("\
+Fatal glibc error: CPU does not support x86-64-v%d\n", 4);
+#    endif /* ISA level 4 */
+#   endif /* ISA level 3 */
+#  endif /* ISA level 2 */
+# endif /* IS_IN (rtld) */
+    }
 }
 
 __ifunc (__x86_cpu_features, __x86_cpu_features, NULL, void,
-- 
2.34.1

[PATCH 7/7] s390x: Use <gcc-macros.h> in early HWCAP check

[Florian Weimer]

This is required so that the checks still work if $(early-cflags)
selects a different ISA level.

---
 sysdeps/s390/s390-64/dl-hwcap-check.h | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/sysdeps/s390/s390-64/dl-hwcap-check.h b/sysdeps/s390/s390-64/dl-hwcap-check.h
index 53e02250b8..f769932325 100644
--- a/sysdeps/s390/s390-64/dl-hwcap-check.h
+++ b/sysdeps/s390/s390-64/dl-hwcap-check.h
@@ -19,17 +19,18 @@
 #ifndef _DL_HWCAP_CHECK_H
 #define _DL_HWCAP_CHECK_H
 
+#include <gcc-macros.h>
 #include <ldsodefs.h>
 
 static inline void
 dl_hwcap_check (void)
 {
 #if defined __ARCH__
-# if __ARCH__ >= 13
+# if GCCMACRO__ARCH__ >= 13
   if (!(GLRO(dl_hwcap) & HWCAP_S390_VXRS_EXT2))
     _dl_fatal_printf ("\
 Fatal glibc error: CPU lacks VXRS_EXT2 support (z15 or later required)\n");
-# elif __ARCH__ >= 12
+# elif GCCMACRO__ARCH__ >= 12
   if (!(GLRO(dl_hwcap) & HWCAP_S390_VXE))
     _dl_fatal_printf ("\
 Fatal glibc error: CPU lacks VXE support (z14 or later required)\n");
-- 
2.34.1

Re: [PATCH 1/7] x86: HAVE_X86_LAHF_SAHF, HAVE_X86_MOVBE and -march=x86-64-vN (bug 28782)

[H.J. Lu]

On Fri, Jan 14, 2022 at 4:40 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> HAVE_X86_LAHF_SAHF is implied by x86-64-v2, and HAVE_X86_MOVBE by
> x86-64-v3.
>
> The individual flag does not appear in -fverbose-asm flag output
> even if the ISA level implies it.
> ---
>  sysdeps/x86/configure    | 4 ++--
>  sysdeps/x86/configure.ac | 4 ++--
>  2 files changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/sysdeps/x86/configure b/sysdeps/x86/configure
> index 62676bb686..7bdbfdc6dc 100644
> --- a/sysdeps/x86/configure
> +++ b/sysdeps/x86/configure
> @@ -155,7 +155,7 @@ else
>    (eval $ac_try) 2>&5
>    ac_status=$?
>    $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
> -  test $ac_status = 0; }; } | grep -q "\-msahf"; then
> +  test $ac_status = 0; }; } | grep -qE '(-msahf\b|-march=x86-64-v)'; then
>        libc_cv_have_x86_lahf_sahf=yes
>      fi
>  fi
> @@ -176,7 +176,7 @@ else
>    (eval $ac_try) 2>&5
>    ac_status=$?
>    $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
> -  test $ac_status = 0; }; } | grep -q "\-mmovbe"; then
> +  test $ac_status = 0; }; } | grep -qE '(-mmovbe\b|-march=x86-64-v([3-9]|[1-9][0-9]))'; then
>        libc_cv_have_x86_movbe=yes
>      fi
>  fi
> diff --git a/sysdeps/x86/configure.ac b/sysdeps/x86/configure.ac
> index 04a12ab680..10d5c2e0e5 100644
> --- a/sysdeps/x86/configure.ac
> +++ b/sysdeps/x86/configure.ac
> @@ -110,7 +110,7 @@ if test $libc_cv_include_x86_isa_level = yes; then
>    AC_CACHE_CHECK([for LAHF/SAHF instruction support],
>                  libc_cv_have_x86_lahf_sahf, [dnl
>      libc_cv_have_x86_lahf_sahf=no
> -    if AC_TRY_COMMAND(${CC-cc} $CFLAGS $CPPFLAGS -fverbose-asm -S -o - -x c /dev/null) | grep -q "\-msahf"; then
> +    if AC_TRY_COMMAND(${CC-cc} $CFLAGS $CPPFLAGS -fverbose-asm -S -o - -x c /dev/null) | grep -qE '(-msahf\b|-march=x86-64-v)'; then
>        libc_cv_have_x86_lahf_sahf=yes
>      fi])
>    if test $libc_cv_have_x86_lahf_sahf = yes; then
> @@ -119,7 +119,7 @@ if test $libc_cv_include_x86_isa_level = yes; then
>    AC_CACHE_CHECK([for MOVBE instruction support],
>                  libc_cv_have_x86_movbe, [dnl
>      libc_cv_have_x86_movbe=no
> -    if AC_TRY_COMMAND(${CC-cc} $CFLAGS $CPPFLAGS -fverbose-asm -S -o - -x c /dev/null) | grep -q "\-mmovbe"; then
> +    if AC_TRY_COMMAND(${CC-cc} $CFLAGS $CPPFLAGS -fverbose-asm -S -o - -x c /dev/null) | grep -qE '(-mmovbe\b|-march=x86-64-v(@<:@3-9@:>@|@<:@1-9@:>@@<:@0-9@:>@))'; then
>        libc_cv_have_x86_movbe=yes
>      fi])
>    if test $libc_cv_have_x86_movbe = yes; then
> --
> 2.34.1
>
>

LGTM.

Reviewed-by: H.J. Lu <hjl.tools@gmail.com>

Thanks.

-- 
H.J.

Re: [PATCH 2/7] Generate gcc-macros.h

[H.J. Lu]

On Fri, Jan 14, 2022 at 4:42 AM Florian Weimer via Libc-alpha
<libc-alpha@sourceware.org> wrote:
>
> The file can be used to check the effect of the default compiler
> flags on code generation even in areas of the build that uses
> non-default compiler flags.
> ---
>  Makeconfig | 9 +++++++++
>  1 file changed, 9 insertions(+)
>
> diff --git a/Makeconfig b/Makeconfig
> index 9b6fc6b08f..2e79077343 100644
> --- a/Makeconfig
> +++ b/Makeconfig
> @@ -1202,6 +1202,15 @@ $(common-objpfx)dl-tunable-list.stmp: \
>         touch $@
>  endif
>
> +# Dump the GCC macros used by the default compiler flags to a header
> +# file, so that they can be inspected when using different compiler
> +# flags.  Add the GCCMACRO prefix to make these macro names unique.
> +$(common-objpfx)gcc-macros.h.in: $(common-objpfx)config.status
> +       $(CC) $(CFLAGS) $(CPPFLAGS) -E -dM -x c -o $@ /dev/null
> +$(common-objpfx)gcc-macros.h: $(common-objpfx)gcc-macros.h.in
> +       sed 's/^#define /#define GCCMACRO/' < $< > $@
> +before-compile += $(common-objpfx)gcc-macros.h
> +
>  # Generate version maps, but wait until sysdep-subdirs is known
>  ifeq ($(sysd-sorted-done),t)
>  ifeq ($(build-shared),yes)
> --
> 2.34.1
>
>

LGTM.

Reviewed-by: H.J. Lu <hjl.tools@gmail.com>

Thanks.


-- 
H.J.

Re: [PATCH 3/7] elf: Split dl-printf.c from dl-misc.c

[H.J. Lu]

On Fri, Jan 14, 2022 at 4:42 AM Florian Weimer via Libc-alpha
<libc-alpha@sourceware.org> wrote:
>
> This allows to use different compiler flags for the diagnostics
> code.
> ---
>  elf/Makefile    |   2 +-
>  elf/dl-misc.c   | 281 +---------------------------------------------
>  elf/dl-printf.c | 292 ++++++++++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 297 insertions(+), 278 deletions(-)
>  create mode 100644 elf/dl-printf.c
>
> diff --git a/elf/Makefile b/elf/Makefile
> index b86d116be9..0c4be43deb 100644
> --- a/elf/Makefile
> +++ b/elf/Makefile
> @@ -31,7 +31,7 @@ routines      = $(all-dl-routines) dl-support dl-iteratephdr \
>  # The core dynamic linking functions are in libc for the static and
>  # profiled libraries.
>  dl-routines    = $(addprefix dl-,load lookup object reloc deps \
> -                                 runtime init fini debug misc \
> +                                 runtime init fini debug misc printf \
>                                   version profile tls origin scope \
>                                   execstack open close trampoline \
>                                   exception sort-maps lookup-direct \

Should we put one item per-line now?


-- 
H.J.

Re: [PATCH 4/7] Add --early-cflags configure option

[H.J. Lu]

On Fri, Jan 14, 2022 at 4:43 AM Florian Weimer via Libc-alpha
<libc-alpha@sourceware.org> wrote:
>
> ---
>  config.make.in      |  1 +
>  configure           | 13 +++++++++++++
>  configure.ac        |  6 ++++++
>  elf/Makefile        |  8 ++++++++
>  manual/install.texi |  7 +++++++
>  5 files changed, 35 insertions(+)
>
> diff --git a/config.make.in b/config.make.in
> index e8630a8d0c..e06f1ff49f 100644
> --- a/config.make.in
> +++ b/config.make.in
> @@ -110,6 +110,7 @@ CFLAGS = @CFLAGS@
>  CPPFLAGS-config = @CPPFLAGS@
>  CPPUNDEFS = @CPPUNDEFS@
>  extra-nonshared-cflags = @extra_nonshared_cflags@
> +early-cflags = @early_cflags@
>  ASFLAGS-config = @ASFLAGS_config@
>  AR = @AR@
>  NM = @NM@
> diff --git a/configure b/configure
> index 3f956cf777..6eedd8c3a7 100755
> --- a/configure
> +++ b/configure
> @@ -681,6 +681,7 @@ force_install
>  bindnow
>  hardcoded_path_in_tests
>  enable_timezone_tools
> +early_cflags
>  extra_nonshared_cflags
>  use_default_link
>  sysheaders
> @@ -761,6 +762,7 @@ with_selinux
>  with_headers
>  with_default_link
>  with_nonshared_cflags
> +with_early_cflags
>  with_timeoutfactor
>  enable_sanity_checks
>  enable_shared
> @@ -1480,6 +1482,8 @@ Optional Packages:
>    --with-default-link     do not use explicit linker scripts
>    --with-nonshared-cflags=CFLAGS
>                            build nonshared libraries with additional CFLAGS
> +  --with-early-cflags=CFLAGS
> +                          build early initialization with additional CFLAGS
>    --with-timeoutfactor=NUM
>                            specify an integer to scale the timeout
>    --with-cpu=CPU          select code for CPU variant
> @@ -3386,6 +3390,15 @@ fi
>
>
>
> +# Check whether --with-early-cflags was given.
> +if test "${with_early_cflags+set}" = set; then :
> +  withval=$with_early_cflags; early_cflags=$withval
> +else
> +  early_cflags=
> +fi
> +
> +
> +
>
>  # Check whether --with-timeoutfactor was given.
>  if test "${with_timeoutfactor+set}" = set; then :
> diff --git a/configure.ac b/configure.ac
> index 277d3527d2..445b0b7f15 100644
> --- a/configure.ac
> +++ b/configure.ac
> @@ -162,6 +162,12 @@ AC_ARG_WITH([nonshared-cflags],
>             [extra_nonshared_cflags=$withval],
>             [extra_nonshared_cflags=])
>  AC_SUBST(extra_nonshared_cflags)
> +AC_ARG_WITH([early-cflags],
> +           AS_HELP_STRING([--with-early-cflags=CFLAGS],
> +                          [build early initialization with additional CFLAGS]),
> +           [early_cflags=$withval],
> +           [early_cflags=])
> +AC_SUBST(early_cflags)
>
>  AC_ARG_WITH([timeoutfactor],
>             AS_HELP_STRING([--with-timeoutfactor=NUM],
> diff --git a/elf/Makefile b/elf/Makefile
> index 0c4be43deb..9fcb4ebd13 100644
> --- a/elf/Makefile
> +++ b/elf/Makefile
> @@ -99,6 +99,14 @@ CFLAGS-.o += $(call elide-stack-protector,.o,$(elide-routines.os))
>  CFLAGS-.op += $(call elide-stack-protector,.op,$(elide-routines.os))
>  CFLAGS-.os += $(call elide-stack-protector,.os,$(all-rtld-routines))
>
> +# Add the requested compiler flags to the early startup code.
> +CFLAGS-dl-printf.os += $(early-cflags)
> +CFLAGS-dl-sysdep.os += $(early-cflags)
> +CFLAGS-dl-tunables.os += $(early-cflags)
> +CFLAGS-dl-write.os += $(early-cflags)
> +CFLAGS-dl-writev.os += $(early-cflags)
> +CFLAGS-rtld.os += $(early-cflags)
> +
>  ifeq ($(unwind-find-fde),yes)
>  routines += unwind-dw2-fde-glibc
>  shared-only-routines += unwind-dw2-fde-glibc
> diff --git a/manual/install.texi b/manual/install.texi
> index b1afc56f5a..27d5893c5e 100644
> --- a/manual/install.texi
> +++ b/manual/install.texi
> @@ -131,6 +131,13 @@ that the objects in @file{libc_nonshared.a} are compiled with this flag
>  (although this will not affect the generated code in this particular
>  case and potentially change debugging information and metadata only).
>
> +@item --with-early-cflags=@var{cflags}
> +Use additional compiler flags @var{cflags} to build the early startup
> +code of the dynamic linker.  These flags can be used to enable early
> +dynamic linker diagnostics to run on CPUs which are not compatible with
> +the rest of @theglibc{}, for example, due to compiler flags which target
> +a later instruction set architecture (ISA).
> +

Isn't it limited to rtld?  If yes, I prefer --with-early-rtld-cflags

>  @item --with-timeoutfactor=@var{NUM}
>  Specify an integer @var{NUM} to scale the timeout of test programs.
>  This factor can be changed at run time using @env{TIMEOUTFACTOR}
> --
> 2.34.1
>
>


-- 
H.J.

Re: [PATCH 3/7] elf: Split dl-printf.c from dl-misc.c

[Florian Weimer]

* H. J. Lu:

> On Fri, Jan 14, 2022 at 4:42 AM Florian Weimer via Libc-alpha
> <libc-alpha@sourceware.org> wrote:
>>
>> This allows to use different compiler flags for the diagnostics
>> code.
>> ---
>>  elf/Makefile    |   2 +-
>>  elf/dl-misc.c   | 281 +---------------------------------------------
>>  elf/dl-printf.c | 292 ++++++++++++++++++++++++++++++++++++++++++++++++
>>  3 files changed, 297 insertions(+), 278 deletions(-)
>>  create mode 100644 elf/dl-printf.c
>>
>> diff --git a/elf/Makefile b/elf/Makefile
>> index b86d116be9..0c4be43deb 100644
>> --- a/elf/Makefile
>> +++ b/elf/Makefile
>> @@ -31,7 +31,7 @@ routines      = $(all-dl-routines) dl-support dl-iteratephdr \
>>  # The core dynamic linking functions are in libc for the static and
>>  # profiled libraries.
>>  dl-routines    = $(addprefix dl-,load lookup object reloc deps \
>> -                                 runtime init fini debug misc \
>> +                                 runtime init fini debug misc printf \
>>                                   version profile tls origin scope \
>>                                   execstack open close trampoline \
>>                                   exception sort-maps lookup-direct \
>
> Should we put one item per-line now?

And drop the addprefix?  I can submit a separate patch for that.

Thanks,
Florian

Re: [PATCH 4/7] Add --early-cflags configure option

[Florian Weimer]

* H. J. Lu:

>> +@item --with-early-cflags=@var{cflags}
>> +Use additional compiler flags @var{cflags} to build the early startup
>> +code of the dynamic linker.  These flags can be used to enable early
>> +dynamic linker diagnostics to run on CPUs which are not compatible with
>> +the rest of @theglibc{}, for example, due to compiler flags which target
>> +a later instruction set architecture (ISA).
>
> Isn't it limited to rtld?  If yes, I prefer --with-early-rtld-cflags

It's currently restricted to ld.so.  But we might enhance it to cover
static linking in the future, too.

Thanks,
Florian

Re: [PATCH 3/7] elf: Split dl-printf.c from dl-misc.c

[H.J. Lu]

On Fri, Jan 14, 2022 at 6:27 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> * H. J. Lu:
>
> > On Fri, Jan 14, 2022 at 4:42 AM Florian Weimer via Libc-alpha
> > <libc-alpha@sourceware.org> wrote:
> >>
> >> This allows to use different compiler flags for the diagnostics
> >> code.
> >> ---
> >>  elf/Makefile    |   2 +-
> >>  elf/dl-misc.c   | 281 +---------------------------------------------
> >>  elf/dl-printf.c | 292 ++++++++++++++++++++++++++++++++++++++++++++++++
> >>  3 files changed, 297 insertions(+), 278 deletions(-)
> >>  create mode 100644 elf/dl-printf.c
> >>
> >> diff --git a/elf/Makefile b/elf/Makefile
> >> index b86d116be9..0c4be43deb 100644
> >> --- a/elf/Makefile
> >> +++ b/elf/Makefile
> >> @@ -31,7 +31,7 @@ routines      = $(all-dl-routines) dl-support dl-iteratephdr \
> >>  # The core dynamic linking functions are in libc for the static and
> >>  # profiled libraries.
> >>  dl-routines    = $(addprefix dl-,load lookup object reloc deps \
> >> -                                 runtime init fini debug misc \
> >> +                                 runtime init fini debug misc printf \
> >>                                   version profile tls origin scope \
> >>                                   execstack open close trampoline \
> >>                                   exception sort-maps lookup-direct \
> >
> > Should we put one item per-line now?
>
> And drop the addprefix?  I can submit a separate patch for that.

Yes.  It will make adding the next file a bit easier.

> Thanks,
> Florian
>


-- 
H.J.

Re: [PATCH 4/7] Add --early-cflags configure option

[H.J. Lu]

On Fri, Jan 14, 2022 at 6:29 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> * H. J. Lu:
>
> >> +@item --with-early-cflags=@var{cflags}
> >> +Use additional compiler flags @var{cflags} to build the early startup
> >> +code of the dynamic linker.  These flags can be used to enable early
> >> +dynamic linker diagnostics to run on CPUs which are not compatible with
> >> +the rest of @theglibc{}, for example, due to compiler flags which target
> >> +a later instruction set architecture (ISA).
> >
> > Isn't it limited to rtld?  If yes, I prefer --with-early-rtld-cflags
>
> It's currently restricted to ld.so.  But we might enhance it to cover
> static linking in the future, too.

But static linking may need a different early CFLAGS.

-- 
H.J.

Re: [PATCH 4/7] Add --early-cflags configure option

[Florian Weimer]

* H. J. Lu:

> On Fri, Jan 14, 2022 at 6:29 AM Florian Weimer <fweimer@redhat.com> wrote:
>>
>> * H. J. Lu:
>>
>> >> +@item --with-early-cflags=@var{cflags}
>> >> +Use additional compiler flags @var{cflags} to build the early startup
>> >> +code of the dynamic linker.  These flags can be used to enable early
>> >> +dynamic linker diagnostics to run on CPUs which are not compatible with
>> >> +the rest of @theglibc{}, for example, due to compiler flags which target
>> >> +a later instruction set architecture (ISA).
>> >
>> > Isn't it limited to rtld?  If yes, I prefer --with-early-rtld-cflags
>>
>> It's currently restricted to ld.so.  But we might enhance it to cover
>> static linking in the future, too.
>
> But static linking may need a different early CFLAGS.

Fair point, I will send an updated patch.

Thanks,
Florian

Re: [PATCH 7/7] s390x: Use <gcc-macros.h> in early HWCAP check

[Stefan Liebler]

On 14/01/2022 13:41, Florian Weimer wrote:
> This is required so that the checks still work if $(early-cflags)
> selects a different ISA level.
> 
> ---
>  sysdeps/s390/s390-64/dl-hwcap-check.h | 5 +++--
>  1 file changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/sysdeps/s390/s390-64/dl-hwcap-check.h b/sysdeps/s390/s390-64/dl-hwcap-check.h
> index 53e02250b8..f769932325 100644
> --- a/sysdeps/s390/s390-64/dl-hwcap-check.h
> +++ b/sysdeps/s390/s390-64/dl-hwcap-check.h
> @@ -19,17 +19,18 @@
>  #ifndef _DL_HWCAP_CHECK_H
>  #define _DL_HWCAP_CHECK_H
>  
> +#include <gcc-macros.h>
>  #include <ldsodefs.h>
>  
>  static inline void
>  dl_hwcap_check (void)
>  {
>  #if defined __ARCH__
> -# if __ARCH__ >= 13
> +# if GCCMACRO__ARCH__ >= 13
>    if (!(GLRO(dl_hwcap) & HWCAP_S390_VXRS_EXT2))
>      _dl_fatal_printf ("\
>  Fatal glibc error: CPU lacks VXRS_EXT2 support (z15 or later required)\n");
> -# elif __ARCH__ >= 12
> +# elif GCCMACRO__ARCH__ >= 12
>    if (!(GLRO(dl_hwcap) & HWCAP_S390_VXE))
>      _dl_fatal_printf ("\
>  Fatal glibc error: CPU lacks VXE support (z14 or later required)\n");



Hi Florian,

I'm not quite sure if all your patches are already committed. I've just
give it a try with commit f8b765bec44e6c464a7eabf80e58c6851ca15ac3:

- configure glibc with --with-rtld-early-cflags=-march=zEC12 and
CFLAGS=-march=z15 on a z15.

- Rebooted with novx-kernel-parameter => vector-related HWCAPs are
disabled and executing vector-instructions leads to a crash

- run a helloworld-program: crash due to vector-instruction in
_dl_setup_hash, which is called in _dl_start_final before
_dl_sysdep_start is called which runs dl_hwcap_check.

I've checked the build-log and see that the following files are compiled
with -march=zEC12:
- dl-printf.c
- ../sysdeps/unix/sysv/linux/dl-write.c
- dl-tunables.c
- ../sysdeps/unix/sysv/linux/dl-sysdep.c
- rtld.c

Bye
Stefan

Re: [PATCH 7/7] s390x: Use <gcc-macros.h> in early HWCAP check

[Florian Weimer]

[-- Attachment #1: Type: text/plain, Size: 902 bytes --]

* Stefan Liebler:

> I'm not quite sure if all your patches are already committed. I've just
> give it a try with commit f8b765bec44e6c464a7eabf80e58c6851ca15ac3:
>
> - configure glibc with --with-rtld-early-cflags=-march=zEC12 and
> CFLAGS=-march=z15 on a z15.
>
> - Rebooted with novx-kernel-parameter => vector-related HWCAPs are
> disabled and executing vector-instructions leads to a crash
>
> - run a helloworld-program: crash due to vector-instruction in
> _dl_setup_hash, which is called in _dl_start_final before
> _dl_sysdep_start is called which runs dl_hwcap_check.
>
> I've checked the build-log and see that the following files are compiled
> with -march=zEC12:
> - dl-printf.c
> - ../sysdeps/unix/sysv/linux/dl-write.c
> - dl-tunables.c
> - ../sysdeps/unix/sysv/linux/dl-sysdep.c
> - rtld.c

Sorry, I missed that requirement.

Would you mind testing the attached patch?

Thanks,
Florian

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-elf-Move-_dl_setup_hash-to-its-own-file.patch --]
[-- Type: text/x-patch, Size: 5233 bytes --]

From dae7882e11191cab17afb22dba4ebba69f433d93 Mon Sep 17 00:00:00 2001
Message-Id: <dae7882e11191cab17afb22dba4ebba69f433d93.1642510437.git.fweimer@redhat.com>
From: Florian Weimer <fweimer@redhat.com>
Date: Tue, 18 Jan 2022 13:53:11 +0100
Subject: [PATCH] elf: Move _dl_setup_hash to its own file
To: libc-alpha@sourceware.org

And compile it with the early CFLAGS.  _dl_setup_hash is called
very early for the ld.so link map, so it should be compiled
differently.
---
 elf/Makefile        |  2 ++
 elf/dl-lookup.c     | 45 --------------------------------
 elf/dl-setup_hash.c | 63 +++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 65 insertions(+), 45 deletions(-)
 create mode 100644 elf/dl-setup_hash.c

diff --git a/elf/Makefile b/elf/Makefile
index c6c4710e16..692a65b061 100644
--- a/elf/Makefile
+++ b/elf/Makefile
@@ -74,6 +74,7 @@ dl-routines = \
   dl-reloc \
   dl-runtime \
   dl-scope \
+  dl-setup_hash \
   dl-sort-maps \
   dl-thread_gscope_wait \
   dl-tls \
@@ -169,6 +170,7 @@ CFLAGS-.os += $(call elide-stack-protector,.os,$(all-rtld-routines))
 
 # Add the requested compiler flags to the early startup code.
 CFLAGS-dl-printf.os += $(rtld-early-cflags)
+CFLAGS-dl-setup_hash.os += $(rtld-early-cflags)
 CFLAGS-dl-sysdep.os += $(rtld-early-cflags)
 CFLAGS-dl-tunables.os += $(rtld-early-cflags)
 CFLAGS-dl-write.os += $(rtld-early-cflags)
diff --git a/elf/dl-lookup.c b/elf/dl-lookup.c
index f43ae150b7..cbf46fda62 100644
--- a/elf/dl-lookup.c
+++ b/elf/dl-lookup.c
@@ -953,51 +953,6 @@ _dl_lookup_symbol_x (const char *undef_name, struct link_map *undef_map,
 }
 
 
-/* Cache the location of MAP's hash table.  */
-
-void
-_dl_setup_hash (struct link_map *map)
-{
-  Elf_Symndx *hash;
-
-  if (__glibc_likely (map->l_info[ELF_MACHINE_GNU_HASH_ADDRIDX] != NULL))
-    {
-      Elf32_Word *hash32
-	= (void *) D_PTR (map, l_info[ELF_MACHINE_GNU_HASH_ADDRIDX]);
-      map->l_nbuckets = *hash32++;
-      Elf32_Word symbias = *hash32++;
-      Elf32_Word bitmask_nwords = *hash32++;
-      /* Must be a power of two.  */
-      assert ((bitmask_nwords & (bitmask_nwords - 1)) == 0);
-      map->l_gnu_bitmask_idxbits = bitmask_nwords - 1;
-      map->l_gnu_shift = *hash32++;
-
-      map->l_gnu_bitmask = (ElfW(Addr) *) hash32;
-      hash32 += __ELF_NATIVE_CLASS / 32 * bitmask_nwords;
-
-      map->l_gnu_buckets = hash32;
-      hash32 += map->l_nbuckets;
-      map->l_gnu_chain_zero = hash32 - symbias;
-
-      /* Initialize MIPS xhash translation table.  */
-      ELF_MACHINE_XHASH_SETUP (hash32, symbias, map);
-
-      return;
-    }
-
-  if (!map->l_info[DT_HASH])
-    return;
-  hash = (void *) D_PTR (map, l_info[DT_HASH]);
-
-  map->l_nbuckets = *hash++;
-  /* Skip nchain.  */
-  hash++;
-  map->l_buckets = hash;
-  hash += map->l_nbuckets;
-  map->l_chain = hash;
-}
-
-
 static void
 _dl_debug_bindings (const char *undef_name, struct link_map *undef_map,
 		    const ElfW(Sym) **ref, struct sym_val *value,
diff --git a/elf/dl-setup_hash.c b/elf/dl-setup_hash.c
new file mode 100644
index 0000000000..6dd57c5c94
--- /dev/null
+++ b/elf/dl-setup_hash.c
@@ -0,0 +1,63 @@
+/* Cache the location of a link map hash table.
+   Copyright (C) 1995-2022 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
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <https://www.gnu.org/licenses/>.  */
+
+#include <assert.h>
+#include <link.h>
+#include <ldsodefs.h>
+
+void
+_dl_setup_hash (struct link_map *map)
+{
+  Elf_Symndx *hash;
+
+  if (__glibc_likely (map->l_info[ELF_MACHINE_GNU_HASH_ADDRIDX] != NULL))
+    {
+      Elf32_Word *hash32
+        = (void *) D_PTR (map, l_info[ELF_MACHINE_GNU_HASH_ADDRIDX]);
+      map->l_nbuckets = *hash32++;
+      Elf32_Word symbias = *hash32++;
+      Elf32_Word bitmask_nwords = *hash32++;
+      /* Must be a power of two.  */
+      assert ((bitmask_nwords & (bitmask_nwords - 1)) == 0);
+      map->l_gnu_bitmask_idxbits = bitmask_nwords - 1;
+      map->l_gnu_shift = *hash32++;
+
+      map->l_gnu_bitmask = (ElfW(Addr) *) hash32;
+      hash32 += __ELF_NATIVE_CLASS / 32 * bitmask_nwords;
+
+      map->l_gnu_buckets = hash32;
+      hash32 += map->l_nbuckets;
+      map->l_gnu_chain_zero = hash32 - symbias;
+
+      /* Initialize MIPS xhash translation table.  */
+      ELF_MACHINE_XHASH_SETUP (hash32, symbias, map);
+
+      return;
+    }
+
+  if (!map->l_info[DT_HASH])
+    return;
+  hash = (void *) D_PTR (map, l_info[DT_HASH]);
+
+  map->l_nbuckets = *hash++;
+  /* Skip nchain.  */
+  hash++;
+  map->l_buckets = hash;
+  hash += map->l_nbuckets;
+  map->l_chain = hash;
+}
-- 
2.34.1

Re: [PATCH 7/7] s390x: Use <gcc-macros.h> in early HWCAP check

[Stefan Liebler]

On 18/01/2022 13:54, Florian Weimer wrote:
> * Stefan Liebler:
> 
>> I'm not quite sure if all your patches are already committed. I've just
>> give it a try with commit f8b765bec44e6c464a7eabf80e58c6851ca15ac3:
>>
>> - configure glibc with --with-rtld-early-cflags=-march=zEC12 and
>> CFLAGS=-march=z15 on a z15.
>>
>> - Rebooted with novx-kernel-parameter => vector-related HWCAPs are
>> disabled and executing vector-instructions leads to a crash
>>
>> - run a helloworld-program: crash due to vector-instruction in
>> _dl_setup_hash, which is called in _dl_start_final before
>> _dl_sysdep_start is called which runs dl_hwcap_check.
>>
>> I've checked the build-log and see that the following files are compiled
>> with -march=zEC12:
>> - dl-printf.c
>> - ../sysdeps/unix/sysv/linux/dl-write.c
>> - dl-tunables.c
>> - ../sysdeps/unix/sysv/linux/dl-sysdep.c
>> - rtld.c
> 
> Sorry, I missed that requirement.
> 
> Would you mind testing the attached patch?
> 
> Thanks,
> Florian

Sure. Now it works fine and I get the expected:
Fatal glibc error: CPU lacks VXRS_EXT2 support (z15 or later required)

Thanks,
Stefan

Re: [PATCH 7/7] s390x: Use <gcc-macros.h> in early HWCAP check

[Florian Weimer]

* Stefan Liebler:

> On 18/01/2022 13:54, Florian Weimer wrote:
>> * Stefan Liebler:
>> 
>>> I'm not quite sure if all your patches are already committed. I've just
>>> give it a try with commit f8b765bec44e6c464a7eabf80e58c6851ca15ac3:
>>>
>>> - configure glibc with --with-rtld-early-cflags=-march=zEC12 and
>>> CFLAGS=-march=z15 on a z15.
>>>
>>> - Rebooted with novx-kernel-parameter => vector-related HWCAPs are
>>> disabled and executing vector-instructions leads to a crash
>>>
>>> - run a helloworld-program: crash due to vector-instruction in
>>> _dl_setup_hash, which is called in _dl_start_final before
>>> _dl_sysdep_start is called which runs dl_hwcap_check.
>>>
>>> I've checked the build-log and see that the following files are compiled
>>> with -march=zEC12:
>>> - dl-printf.c
>>> - ../sysdeps/unix/sysv/linux/dl-write.c
>>> - dl-tunables.c
>>> - ../sysdeps/unix/sysv/linux/dl-sysdep.c
>>> - rtld.c
>> 
>> Sorry, I missed that requirement.
>> 
>> Would you mind testing the attached patch?
>> 
>> Thanks,
>> Florian
>
> Sure. Now it works fine and I get the expected:
> Fatal glibc error: CPU lacks VXRS_EXT2 support (z15 or later required)

Thanks.  Should I push it with your Reviewed-by: and Tested-by:?

Florian

Re: [PATCH 7/7] s390x: Use <gcc-macros.h> in early HWCAP check

[Stefan Liebler]

On 18/01/2022 14:33, Florian Weimer wrote:
> * Stefan Liebler:
> 
>> On 18/01/2022 13:54, Florian Weimer wrote:
>>> * Stefan Liebler:
>>>
>>>> I'm not quite sure if all your patches are already committed. I've just
>>>> give it a try with commit f8b765bec44e6c464a7eabf80e58c6851ca15ac3:
>>>>
>>>> - configure glibc with --with-rtld-early-cflags=-march=zEC12 and
>>>> CFLAGS=-march=z15 on a z15.
>>>>
>>>> - Rebooted with novx-kernel-parameter => vector-related HWCAPs are
>>>> disabled and executing vector-instructions leads to a crash
>>>>
>>>> - run a helloworld-program: crash due to vector-instruction in
>>>> _dl_setup_hash, which is called in _dl_start_final before
>>>> _dl_sysdep_start is called which runs dl_hwcap_check.
>>>>
>>>> I've checked the build-log and see that the following files are compiled
>>>> with -march=zEC12:
>>>> - dl-printf.c
>>>> - ../sysdeps/unix/sysv/linux/dl-write.c
>>>> - dl-tunables.c
>>>> - ../sysdeps/unix/sysv/linux/dl-sysdep.c
>>>> - rtld.c
>>>
>>> Sorry, I missed that requirement.
>>>
>>> Would you mind testing the attached patch?
>>>
>>> Thanks,
>>> Florian
>>
>> Sure. Now it works fine and I get the expected:
>> Fatal glibc error: CPU lacks VXRS_EXT2 support (z15 or later required)
> 
> Thanks.  Should I push it with your Reviewed-by: and Tested-by:?
> 
> Florian
> 

Yes, this if fine for me.
I've had a look at the patch and it just extracted the _dl_setup_hash
function into a new file which is then build with the rtld-early-cflags.

Reviewed-by: Stefan Liebler <stli@linux.ibm.com>
Tested-by: Stefan Liebler <stli@linux.ibm.com>

Re: [PATCH 7/7] s390x: Use <gcc-macros.h> in early HWCAP check

[Joseph Myers]

The "elf: Move _dl_setup_hash to its own file" change (commit 
c90363403b57b3b7919061851cb3e6d9c85e784a) appears to have broken the build 
for MIPS (all ABIs).

In file included from ../sysdeps/gnu/ldsodefs.h:46,
                 from ../sysdeps/unix/sysv/linux/ldsodefs.h:25,
                 from ../sysdeps/unix/sysv/linux/mips/ldsodefs.h:22,
                 from dl-setup_hash.c:21:
dl-setup_hash.c: In function '_dl_setup_hash':
../sysdeps/mips/ldsodefs.h:39:33: error: implicit declaration of function 'DT_MIPS'; did you mean 'EM_MIPS'? [-Werror=implicit-function-declaration]
   39 |       (hash32) += (map)->l_info[DT_MIPS (SYMTABNO)]->d_un.d_val - (symbias); \
      |                                 ^~~~~~~
dl-setup_hash.c:48:7: note: in expansion of macro 'ELF_MACHINE_XHASH_SETUP'
   48 |       ELF_MACHINE_XHASH_SETUP (hash32, symbias, map);
      |       ^~~~~~~~~~~~~~~~~~~~~~~
../sysdeps/mips/ldsodefs.h:39:42: error: 'SYMTABNO' undeclared (first use in this function)
   39 |       (hash32) += (map)->l_info[DT_MIPS (SYMTABNO)]->d_un.d_val - (symbias); \
      |                                          ^~~~~~~~
dl-setup_hash.c:48:7: note: in expansion of macro 'ELF_MACHINE_XHASH_SETUP'
   48 |       ELF_MACHINE_XHASH_SETUP (hash32, symbias, map);
      |       ^~~~~~~~~~~~~~~~~~~~~~~
../sysdeps/mips/ldsodefs.h:39:42: note: each undeclared identifier is reported only once for each function it appears in
   39 |       (hash32) += (map)->l_info[DT_MIPS (SYMTABNO)]->d_un.d_val - (symbias); \
      |                                          ^~~~~~~~
dl-setup_hash.c:48:7: note: in expansion of macro 'ELF_MACHINE_XHASH_SETUP'
   48 |       ELF_MACHINE_XHASH_SETUP (hash32, symbias, map);
      |       ^~~~~~~~~~~~~~~~~~~~~~~
cc1: all warnings being treated as errors

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: [PATCH 7/7] s390x: Use <gcc-macros.h> in early HWCAP check

[Florian Weimer]

* Joseph Myers:

> The "elf: Move _dl_setup_hash to its own file" change (commit 
> c90363403b57b3b7919061851cb3e6d9c85e784a) appears to have broken the build 
> for MIPS (all ABIs).

Sorry, I didn't run a full build of a glibcs this time. 8-/

I did check that ELF_MACHINE_XHASH_SETUP was defined in mips
<ldsodefs.h>, but missed its macro dependency.

I'm build-testing a patch right now, will post it once it finishes
building.

Thanks,
Florian