[PATCH] add --reverse option to objcopy

[PATCH] add --reverse option to objcopy

[Nathan Froyd]

[-- Attachment #1: Type: text/plain, Size: 1109 bytes --]

The attached patch is a forward-port of a patch originally proposed by
Thomas de Lellis:

http://sourceware.org/ml/binutils/2000-03/msg00319.html

The rationale from the original message:

  This patch solves problems trying to use the existing tools, to
  generate a ROM images for certain target systems where the byte order
  implemented in hardware did not match the endian the processor
  used. The ROM programmer did not have this functionality built in
  unfortunately.

  A new option "--reverse" has been added swap bytes within
  an output section... Given bytes in the order:

  01 02 03 04 05 06 07 08

  Allows users to re-order them as:

  02 01 04 03 06 05 08 07 (2 byte reversal)
  04 03 02 01 08 07 06 05 (4 byte reversal)
  03 04 01 02 07 08 05 06 (combination 2 & 4 byte reversal)

There was no comment on his original message, and we have customers who
still find this functionality useful.

Changes from the original patch:

* The original patch's documentation has been converted into texinfo;
* Patch has been cleaned up a little bit;
* A DejaGNU testcase has been added.

OK?

-Nathan

[-- Attachment #2: reverse.patch --]
[-- Type: text/plain, Size: 7778 bytes --]

binutils/
2007-04-13  Nathan Froyd  <froydnj@codesourcery.com>
	    Phil Edwards  <phil@codesourcery.com>
	    Thomas de Lellis <tdel@windriver.com>

        * objcopy.c (reverse_endian):  New variable.
        (command_line_switch, copy_main):  Add OPTION_REVERSE_ENDIAN.
        (copy_options, copy_usage):  Add "reverse" entry.
        (copy_section):  Reverse bytes within output sections.
        * doc/binutils.texi:  Document new objcopy option.

binutils/testsuite/
2007-04-13  Nathan Froyd  <froydnj@codesourcery.com>
	    Phil Edwards  <phil@codesourcery.com>
	    Thomas de Lellis <tdel@windriver.com>

	* binutils-all/objcopy.exp: Add test for --reverse.

--- binutils/doc/binutils.texi	(revision 168708)
+++ binutils/doc/binutils.texi	(local)
@@ -987,6 +987,7 @@ objcopy [@option{-F} @var{bfdname}|@opti
         [@option{--add-section} @var{sectionname}=@var{filename}]
         [@option{--rename-section} @var{oldname}=@var{newname}[,@var{flags}]]
         [@option{--change-leading-char}] [@option{--remove-leading-char}]
+        [@option{--reverse=}@var{num}]
         [@option{--srec-len=}@var{ival}] [@option{--srec-forceS3}]
         [@option{--redefine-sym} @var{old}=@var{new}]
         [@option{--redefine-syms=}@var{filename}]
@@ -1052,6 +1053,7 @@ Note---@command{objcopy} is not able to 
 files.  If the input format has an endianness (some formats do not),
 @command{objcopy} can only copy the inputs into file formats that have the
 same endianness or which have no endianness (e.g., @samp{srec}).
+(However, see the @option{--reverse} option.)
 
 @c man end
 
@@ -1342,6 +1344,29 @@ different conventions for symbol names. 
 when appropriate, regardless of the object file format of the output
 file.
 
+@item --reverse=@var{num}
+Reverse the bytes in a section with output contents.  A section length must
+be evenly divisible by the value given in order for the swap to be able to
+take place. Reversing takes place before the interleaving is performed.
+
+This option is used typically in generating ROM images for problematic
+target systems.  For example, on some target boards, the 32-bit words
+fetched from 8-bit ROMs are re-assembled in little-endian byte order
+regardless of the CPU byte order.  Depending on the programming model, the
+endianness of the ROM may need to be modified.
+
+Consider a simple file with a section containing the following eight
+bytes:  @code{12345678}.
+
+Using @samp{--reverse=2} for the above example, the bytes in the output
+file would be ordered @code{21436587}.
+
+Using @samp{--reverse=4} for the above example, the bytes in the output
+file would be ordered @code{43218765}.
+
+By using @samp{--reverse=2} followed by @samp{--reverse=4}, then bytes
+ordered @code{34127856} would be generated.
+
 @item --srec-len=@var{ival}
 Meaningful only for srec output.  Set the maximum length of the Srecords
 being produced to @var{ival}.  This length covers both address, data and
--- binutils/objcopy.c	(revision 168708)
+++ binutils/objcopy.c	(local)
@@ -218,6 +218,11 @@ static char *prefix_alloc_sections_strin
 /* True if --extract-symbol was passed on the command line.  */
 static bfd_boolean extract_symbol = FALSE;
 
+/* If `reverse_endian' is nonzero, then reverse the order of every chunk
+   of <reverse_endian> bytes within each output section.  */
+static int reverse_endian = 0;
+
+
 /* 150 isn't special; it's just an arbitrary non-ASCII char value.  */
 enum command_line_switch
   {
@@ -265,7 +270,8 @@ enum command_line_switch
     OPTION_WRITABLE_TEXT,
     OPTION_PURE,
     OPTION_IMPURE,
-    OPTION_EXTRACT_SYMBOL
+    OPTION_EXTRACT_SYMBOL,
+    OPTION_REVERSE_ENDIAN
   };
 
 /* Options to handle if running as "strip".  */
@@ -358,6 +364,7 @@ static struct option copy_options[] =
   {"remove-leading-char", no_argument, 0, OPTION_REMOVE_LEADING_CHAR},
   {"remove-section", required_argument, 0, 'R'},
   {"rename-section", required_argument, 0, OPTION_RENAME_SECTION},
+  {"reverse", required_argument, 0, OPTION_REVERSE_ENDIAN},
   {"set-section-flags", required_argument, 0, OPTION_SET_SECTION_FLAGS},
   {"set-start", required_argument, 0, OPTION_SET_START},
   {"srec-len", required_argument, 0, OPTION_SREC_LEN},
@@ -471,6 +478,7 @@ copy_usage (FILE *stream, int exit_statu
      --rename-section <old>=<new>[,<flags>] Rename section <old> to <new>\n\
      --change-leading-char         Force output format's leading character style\n\
      --remove-leading-char         Remove leading character from global symbols\n\
+     --reverse=<num>               Reverse <num> bytes at a time, in output sections with content\n\
      --redefine-sym <old>=<new>    Redefine symbol name <old> to <new>\n\
      --redefine-syms <file>        --redefine-sym for all symbol pairs \n\
                                      listed in <file>\n\
@@ -2383,6 +2391,32 @@ copy_section (bfd *ibfd, sec_ptr isectio
       if (!bfd_get_section_contents (ibfd, isection, memhunk, 0, size))
 	RETURN_NONFATAL (bfd_get_filename (ibfd));
 
+      if (reverse_endian)
+	{
+	  /* We don't handle leftover bytes (too many possible behaviors,
+	     and we don't know what the user wants).  The section length
+	     must be a multiple of the number of bytes to swap.  */
+	  if ((size % reverse_endian) == 0)
+	    {
+	      unsigned long i, j;
+	      bfd_byte b;
+
+	      for (i = 0; i < size; i += reverse_endian)
+		for (j = 0; j < (unsigned long)(reverse_endian / 2); j++)
+		  {
+		    bfd_byte *m = (bfd_byte*)memhunk;
+
+		    b = m[i + j];
+		    m[i + j] = m[(i + reverse_endian) - (j + 1)];
+		    m[(i + reverse_endian) - (j + 1)] = b;
+		  }
+	    }
+	  else
+	    /* User must pad the section up in order to do this.  */
+	    fatal (_("cannot reverse bytes: length of section %s must be evenly divisible by %d"),
+		bfd_section_name (ibfd, isection), reverse_endian);
+	}
+
       if (copy_byte >= 0)
 	{
 	  /* Keep only every `copy_byte'th byte in MEMHUNK.  */
@@ -3256,6 +3290,12 @@ copy_main (int argc, char *argv[])
 	  extract_symbol = TRUE;
 	  break;
 
+	case OPTION_REVERSE_ENDIAN:
+	  reverse_endian = atoi (optarg);
+	  if ((reverse_endian <= 0) || ((reverse_endian % 2) != 0))
+	    fatal (_("number of bytes to reverse must be positive and even"));
+	  break;
+
 	case 0:
 	  /* We've been given a long option.  */
 	  break;
--- binutils/testsuite/binutils-all/objcopy.exp	(revision 168708)
+++ binutils/testsuite/binutils-all/objcopy.exp	(local)
@@ -113,6 +113,40 @@ proc objcopy_test {testname srcfile} {
 
 objcopy_test "simple copy" bintest.s
 
+# Test reversing bytes in a section.
+
+set reversed ${tempfile}-reversed
+set got [binutils_run $OBJCOPY "$OBJCOPYFLAGS -j .data --reverse=4 $tempfile $reversed"]
+
+if ![string match "" $got] then {
+    fail "objcopy --reverse"
+} else {
+    if [is_remote host] {
+        remote_upload host ${reversed} tmpdir/copy-reversed.o
+        set reversed tmpdir/copy-reversed.o
+    }
+
+    set origdata [binutils_run $OBJDUMP "$OBJDUMPFLAGS -s -j .data $tempfile"]
+    set revdata [binutils_run $OBJDUMP "$OBJDUMPFLAGS -s -j .data $reversed"]
+
+    set want "^ \[0-9\]+ (\[0-9\]+)"
+    set found_orig [regexp -lineanchor $want $origdata -> origdata]
+    set found_rev [regexp -lineanchor $want $revdata -> revdata]
+
+    if {$found_orig == 0 || $found_rev == 0} then {
+        fail "objcopy --reverse"
+    } else {
+        scan $origdata "%2x%2x%2x%2x" b1 b2 b3 b4
+        scan $revdata "%2x%2x%2x%2x" c4 c3 c2 c1
+
+        if {$b1 == $c1 && $b2 == $c2 && $b3 == $c3 && $b4 == $c4} then {
+            pass "objcopy --reverse"
+        } else {
+            fail "objcopy --reverse"
+        }
+    }
+}
+
 # Test generating S records.
 
 # We make the srec filename 8.3 compatible. Note that the header string

Re: [PATCH] add --reverse option to objcopy

[H. J. Lu]

On Fri, Apr 13, 2007 at 07:49:22AM -0700, Nathan Froyd wrote:
> The attached patch is a forward-port of a patch originally proposed by
> Thomas de Lellis:
> 
> http://sourceware.org/ml/binutils/2000-03/msg00319.html
> 
> The rationale from the original message:
> 
>   This patch solves problems trying to use the existing tools, to
>   generate a ROM images for certain target systems where the byte order
>   implemented in hardware did not match the endian the processor
>   used. The ROM programmer did not have this functionality built in
>   unfortunately.
> 
>   A new option "--reverse" has been added swap bytes within
>   an output section... Given bytes in the order:

I was very curious what --reverse would do when I looked at the
subject. IMHO, --reverse-endian is clearer.


H.J.

Re: [PATCH] add --reverse option to objcopy

[Nick Clifton]

Hi Nathan,

> OK?

Yes, but there are a couple of niggles:

   * H.J. pointed out that the name of the switch is slightly confusing.
     --reverse-endian or --reverse-bytes might be a more reasonable name.

   * I think that this paragraph in the documentation is rather
     confusing:

+By using @samp{--reverse=2} followed by @samp{--reverse=4}, then bytes
+ordered @code{34127856} would be generated.

     To me this implied that you can use the --reverse switch twice on 
the same command line in order to achieve the effect you describe.  If I 
have read the code correctly however, only the last instance of the 
switch on the command line will be honored, and so you need to run the 
objcopy command *twice* in order to achieve the desired effect.

     This leads on to the third point which is:

   * The code should check to see if --reverse if issued more than once 
on the command line and (if they have different values) warn the user 
that the earlier value(s) are being ignored.

Cheers
   Nick

Re: [PATCH] add --reverse option to objcopy

[Nathan Froyd]

[-- Attachment #1: Type: text/plain, Size: 1215 bytes --]

On Fri, Apr 20, 2007 at 02:32:56PM +0100, Nick Clifton wrote:
>   * H.J. pointed out that the name of the switch is slightly confusing.
>     --reverse-endian or --reverse-bytes might be a more reasonable
>     name.

I have changed the option to --reverse-bytes in the code and the
documentation.

>   * I think that this paragraph in the documentation is rather
>     confusing:
> 
> +By using @samp{--reverse=2} followed by @samp{--reverse=4}, then bytes
> +ordered @code{34127856} would be generated.
> 
>     To me this implied that you can use the --reverse switch twice on 
> the same command line in order to achieve the effect you describe.  If I 
> have read the code correctly however, only the last instance of the 
> switch on the command line will be honored, and so you need to run the 
> objcopy command *twice* in order to achieve the desired effect.

Yes, that's not particularly clear.  I have altered the paragraph to
make it more obvious that you need to run the command twice.

>   * The code should check to see if --reverse if issued more than once 
> on the command line and (if they have different values) warn the user 
> that the earlier value(s) are being ignored.

Done.

OK?

-Nathan

[-- Attachment #2: reverse.patch --]
[-- Type: text/plain, Size: 8539 bytes --]

binutils/
2007-04-20  Nathan Froyd  <froydnj@codesourcery.com>
	    Phil Edwards  <phil@codesourcery.com>
	    Thomas de Lellis <tdel@windriver.com>

        * objcopy.c (reverse_bytes):  New variable.
        (command_line_switch, copy_main):  Add OPTION_REVERSE_ENDIAN.
        (copy_options, copy_usage):  Add "reverse-bytes" entry.
        (copy_section):  Reverse bytes within output sections.
        * doc/binutils.texi:  Document new objcopy option.

binutils/testsuite/
2007-04-20  Nathan Froyd  <froydnj@codesourcery.com>
	    Phil Edwards  <phil@codesourcery.com>
	    Thomas de Lellis <tdel@windriver.com>

	* binutils-all/objcopy.exp: Add test for --reverse-bytes.

Index: binutils/objcopy.c
===================================================================
RCS file: /cvs/src/src/binutils/objcopy.c,v
retrieving revision 1.107
diff -u -r1.107 objcopy.c
--- binutils/objcopy.c	12 Apr 2007 19:20:46 -0000	1.107
+++ binutils/objcopy.c	20 Apr 2007 16:58:18 -0000
@@ -218,6 +218,11 @@
 /* True if --extract-symbol was passed on the command line.  */
 static bfd_boolean extract_symbol = FALSE;
 
+/* If `reverse_bytes' is nonzero, then reverse the order of every chunk
+   of <reverse_bytes> bytes within each output section.  */
+static int reverse_bytes = 0;
+
+
 /* 150 isn't special; it's just an arbitrary non-ASCII char value.  */
 enum command_line_switch
   {
@@ -265,7 +270,8 @@
     OPTION_WRITABLE_TEXT,
     OPTION_PURE,
     OPTION_IMPURE,
-    OPTION_EXTRACT_SYMBOL
+    OPTION_EXTRACT_SYMBOL,
+    OPTION_REVERSE_BYTES
   };
 
 /* Options to handle if running as "strip".  */
@@ -358,6 +364,7 @@
   {"remove-leading-char", no_argument, 0, OPTION_REMOVE_LEADING_CHAR},
   {"remove-section", required_argument, 0, 'R'},
   {"rename-section", required_argument, 0, OPTION_RENAME_SECTION},
+  {"reverse-bytes", required_argument, 0, OPTION_REVERSE_BYTES},
   {"set-section-flags", required_argument, 0, OPTION_SET_SECTION_FLAGS},
   {"set-start", required_argument, 0, OPTION_SET_START},
   {"srec-len", required_argument, 0, OPTION_SREC_LEN},
@@ -471,6 +478,7 @@
      --rename-section <old>=<new>[,<flags>] Rename section <old> to <new>\n\
      --change-leading-char         Force output format's leading character style\n\
      --remove-leading-char         Remove leading character from global symbols\n\
+     --reverse-bytes=<num>         Reverse <num> bytes at a time, in output sections with content\n\
      --redefine-sym <old>=<new>    Redefine symbol name <old> to <new>\n\
      --redefine-syms <file>        --redefine-sym for all symbol pairs \n\
                                      listed in <file>\n\
@@ -2383,6 +2391,32 @@
       if (!bfd_get_section_contents (ibfd, isection, memhunk, 0, size))
 	RETURN_NONFATAL (bfd_get_filename (ibfd));
 
+      if (reverse_bytes)
+	{
+	  /* We don't handle leftover bytes (too many possible behaviors,
+	     and we don't know what the user wants).  The section length
+	     must be a multiple of the number of bytes to swap.  */
+	  if ((size % reverse_bytes) == 0)
+	    {
+	      unsigned long i, j;
+	      bfd_byte b;
+
+	      for (i = 0; i < size; i += reverse_bytes)
+		for (j = 0; j < (unsigned long)(reverse_bytes / 2); j++)
+		  {
+		    bfd_byte *m = (bfd_byte*)memhunk;
+
+		    b = m[i + j];
+		    m[i + j] = m[(i + reverse_bytes) - (j + 1)];
+		    m[(i + reverse_bytes) - (j + 1)] = b;
+		  }
+	    }
+	  else
+	    /* User must pad the section up in order to do this.  */
+	    fatal (_("cannot reverse bytes: length of section %s must be evenly divisible by %d"),
+		bfd_section_name (ibfd, isection), reverse_bytes);
+	}
+
       if (copy_byte >= 0)
 	{
 	  /* Keep only every `copy_byte'th byte in MEMHUNK.  */
@@ -3256,6 +3290,20 @@
 	  extract_symbol = TRUE;
 	  break;
 
+	case OPTION_REVERSE_BYTES:
+          {
+            int prev = reverse_bytes;
+
+            reverse_bytes = atoi (optarg);
+            if ((reverse_bytes <= 0) || ((reverse_bytes % 2) != 0))
+              fatal (_("number of bytes to reverse must be positive and even"));
+            break;
+
+            if (prev && prev != reverse_bytes)
+              non_fatal (_("Warning: ignoring previous --reverse-bytes value of %d"),
+                         prev);
+          }
+
 	case 0:
 	  /* We've been given a long option.  */
 	  break;
Index: binutils/doc/binutils.texi
===================================================================
RCS file: /cvs/src/src/binutils/doc/binutils.texi,v
retrieving revision 1.104
diff -u -r1.104 binutils.texi
--- binutils/doc/binutils.texi	28 Mar 2007 09:08:53 -0000	1.104
+++ binutils/doc/binutils.texi	20 Apr 2007 16:58:18 -0000
@@ -987,6 +987,7 @@
         [@option{--add-section} @var{sectionname}=@var{filename}]
         [@option{--rename-section} @var{oldname}=@var{newname}[,@var{flags}]]
         [@option{--change-leading-char}] [@option{--remove-leading-char}]
+        [@option{--reverse-bytes=}@var{num}]
         [@option{--srec-len=}@var{ival}] [@option{--srec-forceS3}]
         [@option{--redefine-sym} @var{old}=@var{new}]
         [@option{--redefine-syms=}@var{filename}]
@@ -1052,6 +1053,7 @@
 files.  If the input format has an endianness (some formats do not),
 @command{objcopy} can only copy the inputs into file formats that have the
 same endianness or which have no endianness (e.g., @samp{srec}).
+(However, see the @option{--reverse-bytes} option.)
 
 @c man end
 
@@ -1342,6 +1344,30 @@
 when appropriate, regardless of the object file format of the output
 file.
 
+@item --reverse-bytes=@var{num}
+Reverse the bytes in a section with output contents.  A section length must
+be evenly divisible by the value given in order for the swap to be able to
+take place. Reversing takes place before the interleaving is performed.
+
+This option is used typically in generating ROM images for problematic
+target systems.  For example, on some target boards, the 32-bit words
+fetched from 8-bit ROMs are re-assembled in little-endian byte order
+regardless of the CPU byte order.  Depending on the programming model, the
+endianness of the ROM may need to be modified.
+
+Consider a simple file with a section containing the following eight
+bytes:  @code{12345678}.
+
+Using @samp{--reverse-bytes=2} for the above example, the bytes in the
+output file would be ordered @code{21436587}.
+
+Using @samp{--reverse-bytes=4} for the above example, the bytes in the
+output file would be ordered @code{43218765}.
+
+By using @samp{--reverse-bytes=2} for the above example, followed by
+@samp{--reverse-bytes=4} on the output file, the bytes in the second
+output file would be ordered @code{34127856}.
+
 @item --srec-len=@var{ival}
 Meaningful only for srec output.  Set the maximum length of the Srecords
 being produced to @var{ival}.  This length covers both address, data and
Index: binutils/testsuite/binutils-all/objcopy.exp
===================================================================
RCS file: /cvs/src/src/binutils/testsuite/binutils-all/objcopy.exp,v
retrieving revision 1.41
diff -u -r1.41 objcopy.exp
--- binutils/testsuite/binutils-all/objcopy.exp	12 Apr 2007 19:20:46 -0000	1.41
+++ binutils/testsuite/binutils-all/objcopy.exp	20 Apr 2007 16:58:18 -0000
@@ -113,5 +113,39 @@
 
 objcopy_test "simple copy" bintest.s
 
+# Test reversing bytes in a section.
+
+set reversed ${tempfile}-reversed
+set got [binutils_run $OBJCOPY "$OBJCOPYFLAGS -j .data --reverse-bytes=4 $tempfile $reversed"]
+
+if ![string match "" $got] then {
+    fail "objcopy --reverse-bytes"
+} else {
+    if [is_remote host] {
+        remote_upload host ${reversed} tmpdir/copy-reversed.o
+        set reversed tmpdir/copy-reversed.o
+    }
+
+    set origdata [binutils_run $OBJDUMP "$OBJDUMPFLAGS -s -j .data $tempfile"]
+    set revdata [binutils_run $OBJDUMP "$OBJDUMPFLAGS -s -j .data $reversed"]
+
+    set want "^ \[0-9\]+ (\[0-9\]+)"
+    set found_orig [regexp -lineanchor $want $origdata -> origdata]
+    set found_rev [regexp -lineanchor $want $revdata -> revdata]
+
+    if {$found_orig == 0 || $found_rev == 0} then {
+        fail "objcopy --reverse-bytes"
+    } else {
+        scan $origdata "%2x%2x%2x%2x" b1 b2 b3 b4
+        scan $revdata "%2x%2x%2x%2x" c4 c3 c2 c1
+
+        if {$b1 == $c1 && $b2 == $c2 && $b3 == $c3 && $b4 == $c4} then {
+            pass "objcopy --reverse-bytes"
+        } else {
+            fail "objcopy --reverse-bytes"
+        }
+    }
+}
+
 # Test generating S records.
 
 # We make the srec filename 8.3 compatible. Note that the header string

Re: [PATCH] add --reverse option to objcopy

[Nick Clifton]

Hi Nathan,

 > binutils/
 > 2007-04-20  Nathan Froyd  <froydnj@codesourcery.com>
 >	    Phil Edwards  <phil@codesourcery.com>
 >	    Thomas de Lellis <tdel@windriver.com>
 >
 >        * objcopy.c (reverse_bytes):  New variable.
 >        (command_line_switch, copy_main):  Add OPTION_REVERSE_ENDIAN.
 >        (copy_options, copy_usage):  Add "reverse-bytes" entry.
 >        (copy_section):  Reverse bytes within output sections.
 >        * doc/binutils.texi:  Document new objcopy option.
 >
 > binutils/testsuite/
 > 2007-04-20  Nathan Froyd  <froydnj@codesourcery.com>
 > 	    Phil Edwards  <phil@codesourcery.com>
 > 	    Thomas de Lellis <tdel@windriver.com>
 >
 >	* binutils-all/objcopy.exp: Add test for --reverse-bytes.

Approved - please apply.

Cheers
   Nick