Tue May 30 15:52:32 1995 Roland McGrath <roland@churchy.gnu.ai.mit.edu>

	* mach/Makefile (server-interfaces): Removed notify and
 	device_reply.  For shlibs with eager binding, libmachuser.so must
 	not refer to any functions not defined in libc.

14 files changed, 193 insertions, 87 deletions

diff --git a/ChangeLog b/ChangeLog
index 356fdf4c47..48af781cae 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,9 @@
+Tue May 30 15:52:32 1995  Roland McGrath  <roland@churchy.gnu.ai.mit.edu>
+
+	* mach/Makefile (server-interfaces): Removed notify and
+ 	device_reply.  For shlibs with eager binding, libmachuser.so must
+ 	not refer to any functions not defined in libc.
+
 Sat May 27 16:23:22 1995  Jim Meyering  (meyering@comco.com)
 
 	* sysdeps/generic/memchr.c: Cast RHS to const unsigned char *
diff --git a/elf/dl-error.c b/elf/dl-error.c
index acb21a0414..b5af2e323f 100644
--- a/elf/dl-error.c
+++ b/elf/dl-error.c
@@ -43,8 +43,14 @@ _dl_catch_error (const char **errstring,
 
   signalled_errstring = signalled_objname = NULL;
   errcode = setjmp (catch_env);
-  (*operate) ();
+  if (errcode == 0)
+    {
+      (*operate) ();
+      return 0;
+    }
+
+  /* We get here only if we longjmp'd out of OPERATE.  */
   *errstring = signalled_errstring;
   *objname = signalled_objname;
-  return *errstring ? errcode : 0;
+  return errcode == -1 ? 0 : errcode;
 }
diff --git a/elf/dl-fini.c b/elf/dl-fini.c
index cbc05252d2..69ff83d488 100644
--- a/elf/dl-fini.c
+++ b/elf/dl-fini.c
@@ -26,5 +26,5 @@ _dl_fini (void)
 
   for (l = _dl_loaded; l; l = l->l_next)
     if (l->l_init_called && l->l_info[DT_FINI])
-      (*(void (*) (void)) l->l_info[DT_FINI]->d_un.d_ptr) ();
+      (*(void (*) (void)) (l->l_addr + l->l_info[DT_FINI]->d_un.d_ptr)) ();
 }
diff --git a/elf/dl-init.c b/elf/dl-init.c
index e3bfc2ccea..36eb32aa78 100644
--- a/elf/dl-init.c
+++ b/elf/dl-init.c
@@ -46,15 +46,17 @@ _dl_init_next (void)
 	{
 	  /* Find each dependency in order, and see if it
 	     needs to run an initializer.  */
+	  const char *strtab
+	    = ((void *) l->l_addr + l->l_info[DT_STRTAB]->d_un.d_ptr);
 	  const Elf32_Dyn *d;
 	  for (d = l->l_ld; d->d_tag != DT_NULL; ++d)
 	    if (d->d_tag == DT_NEEDED)
 	      {
-		struct link_map *needed = _dl_map_object
-		  (l, (const char *) (l->l_addr + d->d_un.d_ptr), NULL);
+		struct link_map *needed
+		  = _dl_map_object (l, strtab + d->d_un.d_val, NULL);
 		Elf32_Addr init;
 		--needed->l_opencount;
-		init = next_init (l); /* Recurse on this dependency.  */
+		init = next_init (needed); /* Recurse on this dependency.  */
 		if (init != 0)
 		  return init;
 	      }
@@ -74,7 +76,7 @@ _dl_init_next (void)
     }
 
   /* Look for the first initializer not yet called.  */
-  l = _dl_loaded;
+  l = _dl_loaded->l_next;	/* Skip the executable itself.  */
   do
     {
       init = next_init (l);
diff --git a/elf/dl-load.c b/elf/dl-load.c
index f8b37ba346..bb1ad972d4 100644
--- a/elf/dl-load.c
+++ b/elf/dl-load.c
@@ -158,11 +158,12 @@ _dl_map_object (struct link_map *loader, const char *name,
 
       size_t namelen = strlen (name) + 1;
 
-      void trypath (const char *dirpath)
+      inline void trypath (const char *dirpath)
 	{
 	  fd = open_path (name, namelen, dirpath, &realname);
 	}
 
+      fd = -1;
       if (loader && loader->l_info[DT_RPATH])
 	trypath ((const char *) (loader->l_addr +
 				 loader->l_info[DT_RPATH]->d_un.d_ptr));
@@ -317,14 +318,14 @@ _dl_map_object (struct link_map *loader, const char *name,
 	    if (ph->p_memsz > ph->p_filesz)
 	      {
 		/* Extra zero pages should appear at the end of this segment,
-		   after the data mapped from the file.  Adjust MAPEND to map
-		   only the data from the file.  We will later allocate zero
-		   pages following the data mapping.  */
-		caddr_t zero = mapat - mapstart + ph->p_filesz;
-		caddr_t zeroend = mapat - mapstart + ph->p_memsz;
-		caddr_t zeropage
-		  = (caddr_t) ((Elf32_Addr) (zero + pagesize - 1)
-			       & ~(pagesize - 1));
+		   after the data mapped from the file.   */
+		caddr_t zero, zeroend, zeropage;
+
+		mapat += ph->p_vaddr - mapstart;
+		zero = mapat + ph->p_filesz;
+		zeroend = mapat + ph->p_memsz;
+		zeropage = (caddr_t) ((Elf32_Addr) (zero + pagesize - 1)
+				      & ~(pagesize - 1));
 
 		if (zeroend < zeropage)
 		  /* All the extra data is in the last page of the segment.
@@ -342,7 +343,7 @@ _dl_map_object (struct link_map *loader, const char *name,
 				      prot|PROT_WRITE) < 0)
 			  lose (errno, "cannot change memory protections");
 		      }
-		    memset (zero, 0, zeroend - zero);
+		    memset (zero, 0, zeropage - zero);
 		    if ((prot & PROT_WRITE) == 0)
 		      mprotect ((caddr_t) ((Elf32_Addr) zero
 					   & ~(pagesize - 1)),
diff --git a/elf/dl-lookup.c b/elf/dl-lookup.c
index a7afcc79bb..4d5d795ee5 100644
--- a/elf/dl-lookup.c
+++ b/elf/dl-lookup.c
@@ -20,6 +20,7 @@ Cambridge, MA 02139, USA.  */
 #include <stddef.h>
 #include <libelf.h>
 #include <link.h>
+#include <assert.h>
 
 /* Search loaded objects' symbol tables for a definition of 
    the symbol UNDEF_NAME.  Don't use a PLT defn in UNDEF_MAP, since
@@ -70,13 +71,13 @@ _dl_lookup_symbol (const char *undef_name, const Elf32_Sym **ref,
 	      continue;
 	    }
 
-	  if (sym == *ref)
+	  if (sym->st_shndx == SHN_UNDEF)
 	    /* This is the same symbol we are looking for the value for.
 	       If it is a PLT entry, it will have a value of its own;
 	       but that is not what we are looking for.  */
-	    continue;
+	      continue;
 
-	  if (strcmp (strtab + sym->st_name, undef_name))
+	  if (sym != *ref && strcmp (strtab + sym->st_name, undef_name))
 	    /* Not the symbol we are looking for.  */
 	    continue;
 
@@ -106,8 +107,8 @@ _dl_lookup_symbol (const char *undef_name, const Elf32_Sym **ref,
       const char msg[] = "undefined symbol: ";
       char buf[sizeof msg + strlen (undef_name)];
       memcpy (buf, msg, sizeof msg - 1);
-      memcpy (&buf[sizeof msg - 1], undef_name, sizeof buf - sizeof msg);
-      _dl_signal_error (0, reference_name, msg);
+      memcpy (&buf[sizeof msg - 1], undef_name, sizeof buf - sizeof msg + 1);
+      _dl_signal_error (0, reference_name, buf);
     }
 
   *ref = weak_value.s;
diff --git a/elf/dl-reloc.c b/elf/dl-reloc.c
index 94ffb71759..ebc31d07fa 100644
--- a/elf/dl-reloc.c
+++ b/elf/dl-reloc.c
@@ -83,7 +83,7 @@ _dl_relocate_object (struct link_map *l, int lazy)
     l->l_prev->l_next = l;
   }
 
-  if (l->l_info[DT_JMPREL] && ! lazy)
+  if (l->l_info[DT_JMPREL] && lazy)
     /* Set up the PLT so its unrelocated entries will
        jump to _dl_runtime_resolve, which will relocate them.  */
     elf_machine_runtime_setup (l);
diff --git a/elf/do-rel.h b/elf/do-rel.h
index 04643e8fac..acef25d3f8 100644
--- a/elf/do-rel.h
+++ b/elf/do-rel.h
@@ -30,37 +30,46 @@ Cambridge, MA 02139, USA.  */
 /* Perform the relocations in MAP on the running program image as specified
    by RELTAG, SZTAG.  *RESOLVE is called to resolve symbol values; it
    modifies its argument pointer to point to the defining symbol, and
-   returns the base load address of the defining object.  */
+   returns the base load address of the defining object.  If LAZY is
+   nonzero, this is the first pass on PLT relocations; they should be set
+   up to call _dl_runtime_resolve, rather than fully resolved now.  */
 
 static inline void
 elf_dynamic_do_rel (struct link_map *map,
 		    int reltag, int sztag, 
-		    Elf32_Addr (*resolve) (const Elf32_Sym **))
+		    Elf32_Addr (*resolve) (const Elf32_Sym **),
+		    int lazy)
 {
   const Elf32_Sym *const symtab
-    = (const Elf32_Sym *) map->l_info[DT_SYMTAB]->d_un.d_ptr;
-  const Elf32_Rel *r = (const Elf32_Rel *) map->l_info[reltag]->d_un.d_ptr;
+    = (const Elf32_Sym *) (map->l_addr + map->l_info[DT_SYMTAB]->d_un.d_ptr);
+  const Elf32_Rel *r
+    = (const Elf32_Rel *) (map->l_addr + map->l_info[reltag]->d_un.d_ptr);
   const Elf32_Rel *end = &r[map->l_info[sztag]->d_un.d_val / sizeof *r];
 
-  for (; r < end; ++r)
-    {
-      const Elf32_Sym *definer = &symtab[ELF32_R_SYM (r->r_info)];
-      Elf32_Addr loadbase;
+  if (lazy)
+    /* Doing lazy PLT relocations; they need very little info.  */
+    for (; r < end; ++r)
+      elf_machine_lazy_rel (map, r);
+  else
+    for (; r < end; ++r)
+      {
+	const Elf32_Sym *definer = &symtab[ELF32_R_SYM (r->r_info)];
+	Elf32_Addr loadbase;
 
-      if (ELF32_R_SYM (r->r_info) == STN_UNDEF)
-	loadbase = 0;		/* This value will not be consulted.  */
-      else
-	{
-	  if (resolve)
-	    loadbase = (*resolve) (&definer);
-	  else
-	    {
-	      assert (definer->st_shndx != SHN_UNDEF);
-	      loadbase = map->l_addr;
-	    }
-	}
-      elf_machine_rel (map, r, loadbase, definer);
-    }
+	if (ELF32_R_SYM (r->r_info) == STN_UNDEF)
+	  loadbase = 0;		/* This value will not be consulted.  */
+	else
+	  {
+	    if (resolve)
+	      loadbase = (*resolve) (&definer);
+	    else
+	      {
+		assert (definer->st_shndx != SHN_UNDEF);
+		loadbase = map->l_addr;
+	      }
+	  }
+	elf_machine_rel (map, r, loadbase, definer);
+      }
 }
 
 #undef elf_dynamic_do_rel
diff --git a/elf/dynamic-link.h b/elf/dynamic-link.h
index fc5c585356..a7316eefe8 100644
--- a/elf/dynamic-link.h
+++ b/elf/dynamic-link.h
@@ -56,9 +56,9 @@ elf_get_dynamic_info (Elf32_Dyn *dyn, Elf32_Dyn *info[DT_NUM])
 #include "do-rel.h"
 #define ELF_DYNAMIC_DO_REL(map, lazy, resolve)				      \
   if ((map)->l_info[DT_REL])						      \
-    elf_dynamic_do_rel ((map), DT_REL, DT_RELSZ, (resolve));		      \
-  if (!(lazy) && (map)->l_info[DT_PLTREL]->d_un.d_val == DT_REL)	      \
-    elf_dynamic_do_rel ((map), DT_JMPREL, DT_PLTRELSZ, (resolve));
+    elf_dynamic_do_rel ((map), DT_REL, DT_RELSZ, (resolve), 0);		      \
+  if ((map)->l_info[DT_PLTREL]->d_un.d_val == DT_REL)			      \
+    elf_dynamic_do_rel ((map), DT_JMPREL, DT_PLTRELSZ, (resolve), (lazy));
 #else
 #define ELF_DYNAMIC_DO_RELA(map, lazy, resolve) /* Nothing to do.  */
 #endif
@@ -68,9 +68,9 @@ elf_get_dynamic_info (Elf32_Dyn *dyn, Elf32_Dyn *info[DT_NUM])
 #include "do-rel.h"
 #define ELF_DYNAMIC_DO_RELA(map, lazy, resolve)				      \
   if ((map)->l_info[DT_RELA])						      \
-    elf_dynamic_do_rela ((map), DT_RELA, DT_RELASZ, (resolve));		      \
-  if (!(lazy) && (map)->l_info[DT_PLTREL]->d_un.d_val == DT_RELA)	      \
-    elf_dynamic_do_rela ((map), DT_JMPREL, DT_PLTRELSZ, (resolve);
+    elf_dynamic_do_rela ((map), DT_RELA, DT_RELASZ, (resolve), 0);	      \
+  if ((map)->l_info[DT_PLTREL]->d_un.d_val == DT_RELA)			      \
+    elf_dynamic_do_rela ((map), DT_JMPREL, DT_PLTRELSZ, (resolve), (lazy));
 #else
 #define ELF_DYNAMIC_DO_RELA(map, lazy, resolve) /* Nothing to do.  */
 #endif
diff --git a/elf/rtld.c b/elf/rtld.c
index 85f258a948..409b9705d8 100644
--- a/elf/rtld.c
+++ b/elf/rtld.c
@@ -93,6 +93,8 @@ _dl_start (void *arg)
 
 void _start (void);
 
+static int rtld_command;	/* Nonzero if we were run directly.  */
+
 static void
 dl_main (const Elf32_Phdr *phdr,
 	 Elf32_Word phent,
@@ -138,6 +140,7 @@ file you run.  This is mostly of use for maintainers to test new versions\n\
 of this helper program; chances are you did not intend to run this program.\n"
 			      );
 
+	  rtld_command = 1;
 	  interpreter_name = _dl_argv[0];
 	  --_dl_argc;
 	  ++_dl_argv;
@@ -164,7 +167,7 @@ of this helper program; chances are you did not intend to run this program.\n"
 	  case PT_DYNAMIC:
 	    /* This tells us where to find the dynamic section,
 	       which tells us everything we need to do.  */
-	    l->l_ld = (void *) ph->p_vaddr;
+	    l->l_ld = (void *) l->l_addr + ph->p_vaddr;
 	    break;
 	  case PT_INTERP:
 	    /* This "interpreter segment" was used by the program loader to
@@ -173,7 +176,7 @@ of this helper program; chances are you did not intend to run this program.\n"
 	       dlopen call or DT_NEEDED entry, for something that wants to link
 	       against the dynamic linker as a shared library, will know that
 	       the shared object is already loaded.  */
-	    interpreter_name = (void *) ph->p_vaddr;
+	    interpreter_name = (void *) l->l_addr + ph->p_vaddr;
 	    break;
 	  }
       assert (interpreter_name); /* How else did we get here?  */
@@ -220,7 +223,7 @@ of this helper program; chances are you did not intend to run this program.\n"
 	    l->l_next->l_prev = l->l_prev;
 	}
 
-      lazy = _dl_secure || *(getenv ("LD_BIND_NOW") ?: "");
+      lazy = !_dl_secure && *(getenv ("LD_BIND_NOW") ?: "") == '\0';
 
       /* Now we have all the objects loaded.  Relocate them all.
 	 We do this in reverse order so that copy relocs of earlier
diff --git a/sysdeps/i386/dl-machine.h b/sysdeps/i386/dl-machine.h
index 1797ae5b87..f387a887c1 100644
--- a/sysdeps/i386/dl-machine.h
+++ b/sysdeps/i386/dl-machine.h
@@ -75,7 +75,7 @@ elf_machine_rel (struct link_map *map,
 		 const Elf32_Rel *reloc,
 		 Elf32_Addr sym_loadaddr, const Elf32_Sym *sym)
 {
-  Elf32_Addr *const reloc_addr = (Elf32_Addr *) reloc->r_offset;
+  Elf32_Addr *const reloc_addr = (void *) (map->l_addr + reloc->r_offset);
   const Elf32_Addr sym_value = sym_loadaddr + sym->st_value;
 
   switch (ELF32_R_TYPE (reloc->r_info))
@@ -102,6 +102,20 @@ elf_machine_rel (struct link_map *map,
     }
 }
 
+static inline void
+elf_machine_lazy_rel (struct link_map *map, const Elf32_Rel *reloc)
+{
+  Elf32_Addr *const reloc_addr = (void *) (map->l_addr + reloc->r_offset);
+  switch (ELF32_R_TYPE (reloc->r_info))
+    {
+    case R_386_JMP_SLOT:
+      *reloc_addr += map->l_addr;
+      break;
+    default:
+      assert (! "unexpected PLT reloc type");
+      break;
+    }
+}
 
 /* The i386 never uses Elf32_Rela relocations.  */
 #define ELF_MACHINE_NO_RELA 1
@@ -113,12 +127,14 @@ elf_machine_rel (struct link_map *map,
 static inline void
 elf_machine_runtime_setup (struct link_map *l)
 {
+  Elf32_Addr *got;
   extern void _dl_runtime_resolve (Elf32_Word);
+
   /* The GOT entries for functions in the PLT have not yet been filled
      in.  Their initial contents will arrange when called to push an
      offset into the .rel.plt section, push _GLOBAL_OFFSET_TABLE_[1],
      and then jump to _GLOBAL_OFFSET_TABLE[2].  */
-  Elf32_Addr *got = (Elf32_Addr *) l->l_info[DT_PLTGOT]->d_un.d_ptr;
+  got = (Elf32_Addr *) (l->l_addr + l->l_info[DT_PLTGOT]->d_un.d_ptr);
   got[1] = (Elf32_Addr) l;	/* Identify this shared object.  */
   /* This function will get called to fix up the GOT entry indicated by
      the offset on the stack, and then jump to the resolved address.  */
@@ -140,9 +156,20 @@ _dl_start_user:\n\
 	# Save the user entry point address in %edi.\n\
 	movl %eax, %edi\n\
 	# Point %ebx at the GOT.
-1:	call 2f\n\
-2:	popl %ebx\n\
-	addl $_GLOBAL_OFFSET_TABLE_+[.-2b], %ebx\n\
+	call 0f\n\
+0:	popl %ebx\n\
+	addl $_GLOBAL_OFFSET_TABLE_+[.-0b], %ebx\n\
+	# See if we were run as a command with the executable file\n\
+	# name as an extra leading argument.\n\
+	movl rtld_command@GOT(%ebx), %eax\n\
+	movl (%eax),%eax\n\
+	testl %eax,%eax\n\
+	jz 0f\n\
+	# Pop the original argument count, decrement it, and replace\n\
+	# the original first argument pointer with the new count.\n\
+	popl %eax\n\
+	decl %eax\n\
+	movl %eax,(%esp)\n\
 	# Call _dl_init_next to return the address of an initializer\n\
 	# function to run.\n\
 0:	call _dl_init_next@PLT\n\
@@ -150,7 +177,7 @@ _dl_start_user:\n\
 	testl %eax,%eax\n\
 	jz 1f\n\
 	# Call the shared object initializer function.\n\
-	# NOTE: We depend only on the registers (%ebx)\n\
+	# NOTE: We depend only on the registers (%ebx and %edi)\n\
 	# and the return address pushed by this call;\n\
 	# the initializer is called with the stack just\n\
 	# as it appears on entry, and it is free to move\n\
@@ -159,8 +186,8 @@ _dl_start_user:\n\
 	call *%eax\n\
 	# Loop to call _dl_init_next for the next initializer.\n\
 	jmp 0b\n\
-	# Pass our finalizer function to the user in %edx, as per ELF ABI.\n\
-	leal _dl_fini@GOT(%ebx), %edx\n\
+1:	# Pass our finalizer function to the user in %edx, as per ELF ABI.\n\
+	movl _dl_fini@GOT(%ebx), %edx\n\
 	# Jump to the user's entry point.\n\
 	jmp *%edi\n\
 ");
diff --git a/sysdeps/i386/dl-runtime.c b/sysdeps/i386/dl-runtime.c
index 1bc569760c..8e218e2a62 100644
--- a/sysdeps/i386/dl-runtime.c
+++ b/sysdeps/i386/dl-runtime.c
@@ -34,15 +34,16 @@ void
 _dl_runtime_resolve (Elf32_Word reloc_offset)
 {
   __label__ return_insn;
-  struct link_map *l = (void *) &(&reloc_offset)[-1];
+  struct link_map *l = (void *) (&reloc_offset)[-1];
 
   const Elf32_Sym *const symtab
-    = (const Elf32_Sym *) l->l_info[DT_SYMTAB]->d_un.d_ptr;
-  const char *strtab
-    = ((void *) l->l_addr + l->l_info[DT_STRTAB]->d_un.d_ptr);
+    = (const Elf32_Sym *) (l->l_addr + l->l_info[DT_SYMTAB]->d_un.d_ptr);
+  const char *strtab =
+    (const char *) (l->l_addr + l->l_info[DT_STRTAB]->d_un.d_ptr);
 
-  const Elf32_Rel *const reloc = (void *) (l->l_info[DT_JMPREL]->d_un.d_ptr
-					   + reloc_offset);
+  const Elf32_Rel *const reloc
+    = (const void *) (l->l_addr + l->l_info[DT_JMPREL]->d_un.d_ptr +
+		      reloc_offset);
 
   const Elf32_Sym *definer;
   Elf32_Addr loadbase;
@@ -83,7 +84,7 @@ _dl_runtime_resolve (Elf32_Word reloc_offset)
      referred to by this PLT entry; once "ret" pops this address, the
      function in the shared object will run with the stack arranged just as
      when the user entered the PLT.  */
-  (&reloc_offset)[0] = *(Elf32_Word *) reloc->r_offset;
+  (&reloc_offset)[0] = *(Elf32_Word *) (l->l_addr + reloc->r_offset);
 
   return;
 
diff --git a/sysdeps/mach/hurd/dl-sysdep.c b/sysdeps/mach/hurd/dl-sysdep.c
index 1dca319433..d4845213f4 100644
--- a/sysdeps/mach/hurd/dl-sysdep.c
+++ b/sysdeps/mach/hurd/dl-sysdep.c
@@ -36,6 +36,8 @@ Cambridge, MA 02139, USA.  */
 
 #include "dl-machine.h"
 
+extern void __mach_init (void);
+
 extern int _dl_argc;
 extern char **_dl_argv;
 extern char **_environ;
@@ -52,8 +54,8 @@ _dl_sysdep_start (void **start_argptr,
       char **p;
 
       /* Cache the information in various global variables.  */
-      _dl_argc = *argdata++;
-      _dl_argv = (void *) argdata;
+      _dl_argc = *argdata;
+      _dl_argv = (void *) &argdata[1];
       _environ = &_dl_argv[_dl_argc + 1];
       for (p = _environ; *p; ++p);
       _dl_hurd_data = (void *) ++p;
@@ -66,6 +68,12 @@ _dl_sysdep_start (void **start_argptr,
 		  _dl_hurd_data->phdrsz / sizeof (Elf32_Phdr),
 		  &_dl_hurd_data->user_entry);
 
+      /* Deallocate the reply port and task port rights acquired by
+	 __mach_init.  We are done with them now, and the user will
+	 reacquire them for himself when he wants them.  */
+      __mig_dealloc_reply_port (MACH_PORT_NULL);
+      __mach_port_deallocate (__mach_task_self (), __mach_task_self_);
+
       {
 	extern void _dl_start_user (void);
 	/* Unwind the stack to ARGDATA and simulate a return from _dl_start
@@ -74,6 +82,9 @@ _dl_sysdep_start (void **start_argptr,
       }
     }
 
+  /* Set up so we can do RPCs.  */
+  __mach_init ();
+
   /* See hurd/hurdstartup.c; this deals with getting information
      from the exec server and slicing up the arguments.
      Then it will call `go', above.  */
diff --git a/sysdeps/mach/hurd/i386/init-first.c b/sysdeps/mach/hurd/i386/init-first.c
index d747e75c7d..74b15c8f2f 100644
--- a/sysdeps/mach/hurd/i386/init-first.c
+++ b/sysdeps/mach/hurd/i386/init-first.c
@@ -95,8 +95,8 @@ init1 (int argc, char *arg0, ...)
   __libc_init (argc, argv, __environ);
 }
 
-static void 
-init (int *data, int retaddr)
+static void
+init (int *data, void *usercode, void **retaddrloc)
 {
   int argc = *data;
   char **argv = (void *) (data + 1);
@@ -115,6 +115,11 @@ init (int *data, int retaddr)
   if (__hurd_threadvar_max < _HURD_THREADVAR_MAX)
     __hurd_threadvar_max = _HURD_THREADVAR_MAX;
 
+
+  /* After possibly switching stacks, call `init1' (above) with the user
+     code as the return address, and the argument data immediately above
+     that on the stack.  */
+
   if (_cthread_init_routine)
     {
       /* Initialize cthreads, which will allocate us a new stack to run on.  */
@@ -136,13 +141,45 @@ init (int *data, int retaddr)
 	/* Copy the Hurd startup data block to the new stack.  */
 	*od = *d;
 
-      data = newsp;
+      /* Push the user code address on the top of the new stack.  It will
+	 be the return address for `init1'; we will jump there with NEWSP
+	 as the stack pointer.  */
+      *--(void **) newsp = usercode;
+      /* Mutate our own return address to run the code below.  */
+      *retaddrloc = &&switch_stacks;
+      /* Force NEWSP into %ecx and &init1 into %eax, which are not restored
+         by function return.  */
+      asm volatile ("# a %0 c %1" : : "a" (&init1), "c" (newsp));
+      return;
+    switch_stacks:
+      /* Our return address was redirected to here, so at this point our
+	 stack is unwound and callers' registers restored.  Only %ecx and
+	 %eax are call-clobbered and thus still have the values we set just
+	 above.  Fetch from there the new stack pointer we will run on, and
+	 jmp to the run-time address of `init1'; when it returns, it will
+	 run the user code with the argument data at the top of the stack.  */
+      asm volatile ("movl %ecx, %esp; jmp *%eax");
+      /* NOTREACHED */
+    }
+  else
+    {
+      /* We are not switching stacks, but we must play some games with
+	 the one we've got, similar to the stack-switching code above.  */
+      *retaddrloc = &&call_init1;
+      /* Force the user code address into %ecx and the run-time address of
+	 `init1' into %eax, for use below.  */
+      asm volatile ("# a %0 c %1" : : "a" (&init1), "c" (usercode));
+      return;
+    call_init1:
+      /* As in the stack-switching case, at this point our stack is unwound
+	 and callers' registers restored, and only %ecx and %eax
+	 communicate values from the lines above.  In this case we have
+	 stashed in %ecx the user code return address.  Push it on the top
+	 of the stack so it acts as init1's return address, and then jump
+	 there.  */
+      asm volatile ("pushl %ecx; jmp *%eax");
+      /* NOTREACHED */
     }
-
-  /* Call `init1' (above) with the user code as the return address,
-     and the argument data immediately above that on the stack.  */
-  *--data = retaddr;
-  asm volatile ("movl %0, %%esp; jmp %*%1" : : "g" (data), "r" (&init1));
 }  
 
 
@@ -151,30 +188,32 @@ init (int *data, int retaddr)
    It is called just before the user _start code from i386/elf/start.S,
    with the stack set up as that code gets it.  */
 
-static void soinit (int argc, ...) __attribute__ ((unused, section (".init")));
+/* NOTE!  The linker notices the magical name `_init' and sets the DT_INIT
+   pointer in the dynamic section based solely on that.  It is convention
+   for this function to be in the `.init' section, but the symbol name is
+   the only thing that really matters!!  */
+/*void _init (int argc, ...) __attribute__ ((unused, section (".init")));*/
 
-static void
-soinit (int argc, ...)
+void
+_init (int argc, ...)
 {
   /* Initialize data structures so we can do RPCs.  */
   __mach_init ();
 
   RUN_HOOK (_hurd_preinit_hook, ());
   
-  init (&argc, (&argc)[-1]);
-
-  (void) &soinit;		/* Avoid gcc optimizing this fn out.  */
+  init (&argc, ((void **) &argc)[-1], &((void **) &argc)[-1]);
 }
 #endif
 
 
 void
-__libc_init_first (int argc, ...)
+__libc_init_first (int argc __attribute__ ((unused)), ...)
 {
 #ifndef PIC
   void doinit (int *data)
     {
-      init (data, (&argc)[-1]);
+      init (data, ((void **) &argc)[-1], &((void **) &data)[-1]);
     }
 
   /* Initialize data structures so we can do RPCs.  */