[PATCH] pad out code segment for NaCl targets

[Roland McGrath <mcgrathr at google dot com>]

This is a fully working version of my second draft RFC patch for this.  The
fake-section nonsense is horribly kludgery, but I haven't heard any better
suggestions and I seem to have gotten it working well enough.

OK for trunk and 2.23?  (The 2.23 version is trivially different because of
the absence of the elf_seg_map macro.)

As this touches only NaCl targets, for which I'm the sole authority, I'll
consider myself approved enough if nobody objects in a few days.  But a
positive approval from a maintainer would be much appreciated.


Thanks,
Roland


bfd/
2013-08-20  Roland McGrath  <mcgrathr@google.com>

	* elf-nacl.c (nacl_modify_segment_map): Fix logic reordering the
	elf_segment_map list.  If an executable segment is page-aligned
	but does not end with a full page, then append a fake section into
	the segment map entry that pads out the page.
	(nacl_final_write_processing): New function.  Write the code fill
	laid out in nacl_modify_segment_map.
	* elf-nacl.h: Declare it.
	* elf32-arm.c (elf32_arm_nacl_final_write_processing): New function.
	(elf_backend_final_write_processing): Define it for NaCl backend.
	* elf32-i386.c (elf_backend_final_write_processing): Likewise.
	* elf64-x86-64.c (elf_backend_final_write_processing): Likewise.

	* elf-nacl.c (segment_eligible_for_headers): Rename MAXPAGESIZE
	parameter to MINPAGESIZE.
	(nacl_modify_segment_map): Use minpagesize instead of maxpagesize.

	* elf32-arm.c (ELF_MINPAGESIZE, ELF_COMMONPAGESIZE): #undef for NaCl.

ld/testsuite/
2013-08-20  Roland McGrath  <mcgrathr@google.com>

	* ld-scripts/rgn-at3.d: XFAIL for *-*-nacl* targets.
	* ld-scripts/rgn-over8-ok.d: Likewise.

--- a/bfd/elf-nacl.c
+++ b/bfd/elf-nacl.c
@@ -1,5 +1,5 @@
 /* Native Client support for ELF
-   Copyright 2012 Free Software Foundation, Inc.
+   Copyright 2012, 2013 Free Software Foundation, Inc.

    This file is part of BFD, the Binary File Descriptor library.

@@ -20,6 +20,7 @@

 #include "sysdep.h"
 #include "bfd.h"
+#include "libbfd.h"
 #include "elf-bfd.h"
 #include "elf-nacl.h"
 #include "elf/common.h"
@@ -48,11 +49,11 @@ segment_executable (struct elf_segment_map *seg)
    allow space for the headers.  */
 static bfd_boolean
 segment_eligible_for_headers (struct elf_segment_map *seg,
-                              bfd_vma maxpagesize, bfd_vma sizeof_headers)
+			      bfd_vma minpagesize, bfd_vma sizeof_headers)
 {
   bfd_boolean any_contents = FALSE;
   unsigned int i;
-  if (seg->count == 0 || seg->sections[0]->lma % maxpagesize < sizeof_headers)
+  if (seg->count == 0 || seg->sections[0]->lma % minpagesize < sizeof_headers)
     return FALSE;
   for (i = 0; i < seg->count; ++i)
     {
@@ -76,7 +77,7 @@ nacl_modify_segment_map (bfd *abfd, struct
bfd_link_info *info)
   struct elf_segment_map **last_load = NULL;
   bfd_boolean moved_headers = FALSE;
   int sizeof_headers = info == NULL ? 0 : bfd_sizeof_headers (abfd, info);
-  bfd_vma maxpagesize = get_elf_backend_data (abfd)->maxpagesize;
+  bfd_vma minpagesize = get_elf_backend_data (abfd)->minpagesize;

   if (info != NULL && info->user_phdrs)
     /* The linker script used PHDRS explicitly, so don't change what the
@@ -89,20 +90,91 @@ nacl_modify_segment_map (bfd *abfd, struct
bfd_link_info *info)

       if (seg->p_type == PT_LOAD)
 	{
+	  bfd_boolean executable = segment_executable (seg);
+
+	  if (executable
+	      && seg->count > 0
+	      && seg->sections[0]->vma % minpagesize == 0)
+	    {
+	      asection *lastsec = seg->sections[seg->count - 1];
+	      bfd_vma end = lastsec->vma + lastsec->size;
+	      if (end % minpagesize != 0)
+		{
+		  /* This is an executable segment that starts on a page
+		     boundary but does not end on a page boundary.  Fill
+		     it out to a whole page with code fill (the tail of
+		     the segment will not be within any section).  Thus
+		     the entire code segment can be mapped from the file
+		     as whole pages and that mapping will contain only
+		     valid instructions.
+
+		     To accomplish this, we must fake out the code in
+		     assign_file_positions_for_load_sections (elf.c) so
+		     that it advances past the rest of the final page,
+		     rather than trying to put the next (unaligned, or
+		     unallocated) section.  We do this by appending a
+		     dummy section record to this element in the segment
+		     map.  No such output section ever actually exists,
+		     but this gets the layout logic to advance the file
+		     positions past this partial page.  Since we are
+		     lying to BFD like this, nothing will ever know to
+		     write the section contents.  So we do that by hand
+		     after the fact, in nacl_final_write_processing, below.  */
+
+		  struct elf_segment_map *newseg;
+		  asection *sec;
+		  struct bfd_elf_section_data *secdata;
+
+		  BFD_ASSERT (!seg->p_size_valid);
+
+		  secdata = bfd_zalloc (abfd, sizeof *secdata);
+		  if (secdata == NULL)
+		    return FALSE;
+
+		  sec = bfd_zalloc (abfd, sizeof *sec);
+		  if (sec == NULL)
+		    return FALSE;
+
+		  /* Fill in only the fields that actually affect the logic
+		     in assign_file_positions_for_load_sections.  */
+		  sec->vma = end;
+		  sec->lma = lastsec->lma + lastsec->size;
+		  sec->size = minpagesize - (end % minpagesize);
+		  sec->flags = (SEC_ALLOC | SEC_LOAD
+				| SEC_READONLY | SEC_CODE | SEC_LINKER_CREATED);
+		  sec->used_by_bfd = secdata;
+
+		  secdata->this_hdr.sh_type = SHT_PROGBITS;
+		  secdata->this_hdr.sh_flags = SHF_ALLOC | SHF_EXECINSTR;
+		  secdata->this_hdr.sh_addr = sec->vma;
+		  secdata->this_hdr.sh_size = sec->size;
+
+		  newseg = bfd_alloc (abfd,
+				      sizeof *newseg + ((seg->count + 1)
+							* sizeof (asection *)));
+		  if (newseg == NULL)
+		    return FALSE;
+		  memcpy (newseg, seg,
+			  sizeof *newseg + (seg->count * sizeof (asection *)));
+		  newseg->sections[newseg->count++] = sec;
+		  *m = seg = newseg;
+		}
+	    }
+
 	  /* First, we're just finding the earliest PT_LOAD.
 	     By the normal rules, this will be the lowest-addressed one.
 	     We only have anything interesting to do if it's executable.  */
 	  last_load = m;
 	  if (first_load == NULL)
 	    {
-              if (!segment_executable (*m))
-                return TRUE;
+	      if (!executable)
+		goto next;
 	      first_load = m;
 	    }
 	  /* Now that we've noted the first PT_LOAD, we're looking for
 	     the first non-executable PT_LOAD with a nonempty p_filesz.  */
 	  else if (!moved_headers
-                   && segment_eligible_for_headers (seg, maxpagesize,
+		   && segment_eligible_for_headers (seg, minpagesize,
 						    sizeof_headers))
 	    {
 	      /* This is the one we were looking for!
@@ -127,6 +199,7 @@ nacl_modify_segment_map (bfd *abfd, struct
bfd_link_info *info)
 	    }
 	}

+    next:
       m = &seg->next;
     }

@@ -183,7 +256,7 @@ nacl_modify_program_headers (bfd *abfd, struct
bfd_link_info *info)
       m = &(*m)->next;
       ++p;

-      while ((*m) != NULL)
+      while (*m != NULL)
 	{
 	  if (p->p_type == PT_LOAD && p->p_vaddr < first_load_phdr->p_vaddr)
 	    {
@@ -207,12 +280,24 @@ nacl_modify_program_headers (bfd *abfd, struct
bfd_link_info *info)
 	  struct elf_segment_map *first_next = first_seg->next;
 	  struct elf_segment_map *next_next = next_seg->next;

-          first_seg->next = next_next;
+	  if (next_load_seg == &first_seg->next)
+	    {
 	      *first_load_seg = next_seg;
+	      next_seg->next = first_seg;
+	      first_seg->next = next_next;
+	    }
+	  else
+	    {
+	      *first_load_seg = first_next;
+	      *next_load_seg = next_next;

-          next_seg->next = first_next;
+	      first_seg->next = *next_load_seg;
 	      *next_load_seg = first_seg;

+	      next_seg->next = *first_load_seg;
+	      *first_load_seg = next_seg;
+	    }
+
 	  move_phdr = *next_load_phdr;
 	  memmove (first_load_phdr + 1, first_load_phdr,
 		   (next_load_phdr - first_load_phdr) * sizeof move_phdr);
@@ -222,3 +307,39 @@ nacl_modify_program_headers (bfd *abfd, struct
bfd_link_info *info)

   return TRUE;
 }
+
+void
+nacl_final_write_processing (bfd *abfd, bfd_boolean linker ATTRIBUTE_UNUSED)
+{
+  struct elf_segment_map *seg;
+  for (seg = elf_seg_map (abfd); seg != NULL; seg = seg->next)
+    if (seg->p_type == PT_LOAD
+	&& seg->count > 1
+	&& seg->sections[seg->count - 1]->owner == NULL)
+      {
+	/* This is a fake section added in nacl_modify_segment_map, above.
+	   It's not a real BFD section, so nothing wrote its contents.
+	   Now write out its contents.  */
+
+	asection *sec = seg->sections[seg->count - 1];
+	char *fill;
+
+	BFD_ASSERT (sec->flags & SEC_LINKER_CREATED);
+	BFD_ASSERT (sec->flags & SEC_CODE);
+	BFD_ASSERT (sec->size > 0);
+
+	fill = abfd->arch_info->fill (sec->size, bfd_big_endian (abfd), TRUE);
+
+	if (fill == NULL
+	    || bfd_seek (abfd, sec->filepos, SEEK_SET) != 0
+	    || bfd_bwrite (fill, sec->size, abfd) != sec->size)
+	  {
+	    /* We don't have a proper way to report an error here.  So
+	       instead fudge things so that elf_write_shdrs_and_ehdr will
+	       fail.  */
+	    elf_elfheader (abfd)->e_shoff = (file_ptr) -1;
+	  }
+
+	free (fill);
+      }
+}
--- a/bfd/elf-nacl.h
+++ b/bfd/elf-nacl.h
@@ -1,5 +1,5 @@
 /* Native Client support for ELF
-   Copyright 2012 Free Software Foundation, Inc.
+   Copyright 2012, 2013 Free Software Foundation, Inc.

    This file is part of BFD, the Binary File Descriptor library.

@@ -22,3 +22,4 @@

 bfd_boolean nacl_modify_segment_map (bfd *, struct bfd_link_info *);
 bfd_boolean nacl_modify_program_headers (bfd *, struct bfd_link_info *);
+void nacl_final_write_processing (bfd *, bfd_boolean linker);
--- a/bfd/elf32-arm.c
+++ b/bfd/elf32-arm.c
@@ -15860,6 +15860,14 @@ elf32_arm_nacl_modify_segment_map (bfd *abfd,
struct bfd_link_info *info)
 	  && nacl_modify_segment_map (abfd, info));
 }

+static void
+elf32_arm_nacl_final_write_processing (bfd *abfd, bfd_boolean linker)
+{
+  elf32_arm_final_write_processing (abfd, linker);
+  nacl_final_write_processing (abfd, linker);
+}
+
+
 #undef	elf32_bed
 #define elf32_bed			elf32_arm_nacl_bed
 #undef  bfd_elf32_bfd_link_hash_table_create
@@ -15871,9 +15879,14 @@ elf32_arm_nacl_modify_segment_map (bfd *abfd,
struct bfd_link_info *info)
 #define	elf_backend_modify_segment_map		elf32_arm_nacl_modify_segment_map
 #undef	elf_backend_modify_program_headers
 #define	elf_backend_modify_program_headers	nacl_modify_program_headers
+#undef  elf_backend_final_write_processing
+#define elf_backend_final_write_processing	elf32_arm_nacl_final_write_processing

 #undef	ELF_MAXPAGESIZE
 #define ELF_MAXPAGESIZE			0x10000
+#undef	ELF_MINPAGESIZE
+#undef	ELF_COMMONPAGESIZE
+

 #include "elf32-target.h"

@@ -15882,6 +15895,13 @@ elf32_arm_nacl_modify_segment_map (bfd *abfd,
struct bfd_link_info *info)
 #undef	elf_backend_modify_segment_map
 #define elf_backend_modify_segment_map		elf32_arm_modify_segment_map
 #undef	elf_backend_modify_program_headers
+#undef  elf_backend_final_write_processing
+#define elf_backend_final_write_processing	elf32_arm_final_write_processing
+#undef	ELF_MINPAGESIZE
+#define ELF_MINPAGESIZE			0x1000
+#undef	ELF_COMMONPAGESIZE
+#define ELF_COMMONPAGESIZE		0x1000
+

 /* VxWorks Targets.  */

--- a/bfd/elf32-i386.c
+++ b/bfd/elf32-i386.c
@@ -5302,12 +5302,15 @@ static const struct elf_i386_backend_data
elf_i386_nacl_arch_bed =
 #define	elf_backend_modify_segment_map		nacl_modify_segment_map
 #undef	elf_backend_modify_program_headers
 #define	elf_backend_modify_program_headers	nacl_modify_program_headers
+#undef	elf_backend_final_write_processing
+#define elf_backend_final_write_processing	nacl_final_write_processing

 #include "elf32-target.h"

 /* Restore defaults.  */
 #undef	elf_backend_modify_segment_map
 #undef	elf_backend_modify_program_headers
+#undef	elf_backend_final_write_processing

 /* VxWorks support.  */

--- a/bfd/elf64-x86-64.c
+++ b/bfd/elf64-x86-64.c
@@ -5397,6 +5397,8 @@ static const struct elf_x86_64_backend_data
elf_x86_64_nacl_arch_bed =
 #define	elf_backend_modify_segment_map		nacl_modify_segment_map
 #undef	elf_backend_modify_program_headers
 #define	elf_backend_modify_program_headers	nacl_modify_program_headers
+#undef	elf_backend_final_write_processing
+#define elf_backend_final_write_processing	nacl_final_write_processing

 #include "elf64-target.h"

@@ -5441,6 +5443,7 @@ static const struct elf_x86_64_backend_data
elf_x86_64_nacl_arch_bed =
 #undef	elf_backend_size_info
 #undef	elf_backend_modify_segment_map
 #undef	elf_backend_modify_program_headers
+#undef	elf_backend_final_write_processing

 /* Intel L1OM support.  */

--- a/ld/testsuite/ld-scripts/rgn-at3.d
+++ b/ld/testsuite/ld-scripts/rgn-at3.d
@@ -2,9 +2,13 @@
 # source: rgn-at.s
 # ld: -T rgn-at3.t
 # objdump: -w -h
-# xfail: rx-*-*
+# xfail: rx-*-* *-*-nacl*
 #   FAILS on the RX because the linker has to set LMA == VMA for the
 #   Renesas loader.
+#   FAILs on NaCl targets because the linker extends the first segment
+#   to fill out the page, making its p_vaddr+p_memsz cover the sh_addr
+#   of .bss too, which makes BFD compute its LMA from the p_paddr of the
+#   text segment.

 .*:     file format .*

--- a/ld/testsuite/ld-scripts/rgn-over8-ok.d
+++ b/ld/testsuite/ld-scripts/rgn-over8-ok.d
@@ -2,9 +2,13 @@
 # source: rgn-over8.s
 # ld: -T rgn-over8.t
 # objdump: -w -h
-# xfail: rx-*-*
+# xfail: rx-*-* *-*-nacl*
 #   FAILS on the RX because the linker has to set LMA == VMA for the
 #   Renesas loader.
+#   FAILs on NaCl targets because the linker extends the first segment
+#   to fill out the page, making its p_vaddr+p_memsz cover the sh_addr
+#   of .bss too, which makes BFD compute its LMA from the p_paddr of the
+#   text segment.

 .*:     file format .*