[patch 1/2] Implement reglocs for s390/s390x - readelf

[Jan Kratochvil <jan dot kratochvil at redhat dot com>]

Hi Mark,

jankratochvil/s390-core-readelf

On Tue, 29 Jan 2013 11:22:17 +0100, Mark Wielaard wrote:
> On Wed, 2013-01-23 at 22:13 +0100, Jan Kratochvil wrote:
> > Re: [patch] Implement reglocs for s390/s390x
> > https://lists.fedorahosted.org/pipermail/elfutils-devel/2012-October/002715.html
> > Message-ID: <20121014155013.GA5525@host2.jankratochvil.net>
>
> I admit to be a little lost in the review thread. But it seems to me
> this could be split up between formatting changes (like the \n, b, B
> readelf.c one) and the actual s390_corenote.c change. Would it be
> possible to post this as separate patches (and maybe even split off the
> test addition if that depends on the formatting)?

here is the readelf.c part.  But it is not useful without [patch 2/2] as
nothing in elfutils is using the format this patch fixes.

If you use jankratochvil/s390-core with this patch undone one sees (reversed):

   LINUX                  8  S390_LAST_BREAK
-    last_break: 0xfd75ccbe, last_break: 0x06000000
-    last_break: 0x07030000
[...]
-    last_break: 0x10809f00
-    la./test-subr.sh: line 73: 20195 Segmentation fault      LD_LIBRARY_PATH="${built_library_path}${LD_LIBRARY_PATH:+:}$LD_LIBRARY_PATH" $VALGRIND_CMD "$@"
+    last_break: 0xfd75ccbe
+  LINUX                  4  S390_SYSTEM_CALL

This is because S390_LAST_BREAK has only single item and readelf.c expected
single-item is repeating through the whole section data.  But in this case the
item has even offset 4 bytes (and size 4 bytes in a section of 8 bytes;
initial 4 bytes are ignored).  Repeating items with offset were not handled
correctly, which led to the crash.

So it is best not to errorneously consider S390_LAST_BREAK to have repeating
content.


Thanks,
Jan


commit 22881b36d6512394a20c67edfc31ae745e0e50b3
Author: Jan Kratochvil <jan.kratochvil@redhat.com>
Date:   Wed Oct 10 19:32:05 2012 +0200

    src/
    	* readelf.c (handle_core_items): Limit special repeated items handling
    	to single-item formats '\n', 'b' and 'B', assert OFFSET 0 there.
    
    Signed-off-by: Jan Kratochvil <jan.kratochvil@redhat.com>

diff --git a/src/ChangeLog b/src/ChangeLog
index 377c124..4d00013 100644
--- a/src/ChangeLog
+++ b/src/ChangeLog
@@ -1,3 +1,8 @@
+2013-01-29  Jan Kratochvil  <jan.kratochvil@redhat.com>
+
+	* readelf.c (handle_core_items): Limit special repeated items handling
+	to single-item formats '\n', 'b' and 'B', assert OFFSET 0 there.
+
 2012-12-18  Mark Wielaard  <mark@bordewijk.wildebeest.org>
 
 	* readelf.c (ELF_INPUT_SECTION): New argp key value.
diff --git a/src/readelf.c b/src/readelf.c
index 0b46459..c256102 100644
--- a/src/readelf.c
+++ b/src/readelf.c
@@ -7851,6 +7851,24 @@ handle_core_items (Elf *core, const void *desc, size_t descsz,
 {
   if (nitems == 0)
     return 0;
+  unsigned int colno = 0;
+
+  /* FORMAT '\n' makes sense to be present only as a single item as it
+     processes all the data of a note.  FORMATs 'b' and 'B' have a special case
+     if present as a single item but they can be also processed with other
+     items below.  */
+  if (nitems == 1 && (items[0].format == '\n' || items[0].format == 'b'
+		      || items[0].format == 'B'))
+    {
+      assert (items[0].offset == 0);
+      size_t size = descsz;
+      colno = handle_core_item (core, items, desc, colno, &size);
+      /* If SIZE is not zero here there is some remaining data.  But we do not
+	 know how to process it anyway.  */
+      return colno;
+    }
+  for (size_t i = 0; i < nitems; ++i)
+    assert (items[i].format != '\n');
 
   /* Sort to collect the groups together.  */
   const Ebl_Core_Item *sorted_items[nitems];
@@ -7869,23 +7887,7 @@ handle_core_items (Elf *core, const void *desc, size_t descsz,
   qsort (groups, ngroups, sizeof groups[0], &compare_core_item_groups);
 
   /* Write out all the groups.  */
-  unsigned int colno = 0;
-
   const void *last = desc;
-  if (nitems == 1)
-    {
-      size_t size = descsz;
-      /* If this note contains registers as well as items, don't pass
-	 &size to express that we don't wish to repeat.  */
-      colno = handle_core_item (core, sorted_items[0], desc, colno,
-				size != 0 ? &size : NULL);
-
-      if (size == 0)
-	return colno;
-      desc += descsz - size;
-      descsz = size;
-    }
-
   do
     {
       for (size_t i = 0; i < ngroups; ++i)