Hi Thibaut,

generally I'm not a fan to include workarounds for bugs which we do not have anymore in mainline U-Boot. Isn't there any other alternative for this? What do other people think?

If nobody objects to the genereal principle, then I have some requests below.

During a few months, offsets of files in multi-file images were miscalculated, this has been fixed by 02b9b22446e3d7ad6a6382be17a1ce79a7de589b, but unfortunatly, some devices (I'm thinking of the Neo FreeRunner) are using a broken version of U-Boot.

This problem can easily be worked around at image creation time by adding one byte of junk at the end of the first (and optionnally second) file, if its size is a multiple of 4.

Hm, as I read it, you add 4 bytes (not one) in case the image is already padded correctly to 32-bit, correct? If so, then please correct the comment.

It's not really clean, but it shouldn't cause any problem. At least, I haven't encountered any using this patch.

Signed-off-by: Thibaut Girka thib@sitedethib.com

tools/mkimage.c | 9 ++++++++- tools/mkimage.h | 1 + 2 files changed, 9 insertions(+), 1 deletions(-)

diff --git a/tools/mkimage.c b/tools/mkimage.c index f5859d7..239c1f0 100644 --- a/tools/mkimage.c +++ b/tools/mkimage.c @@ -243,6 +243,9 @@ main (int argc, char **argv) usage (); params.imagename = *++argv; goto NXTARG;

• 	case 'p':
  

• 		params.pflag++;
  

• 		break;
  case 'v':
  	params.vflag++;
  	break;
  

@@ -383,6 +386,8 @@ NXTARG: ; params.cmdname, file, strerror(errno)); exit (EXIT_FAILURE); }

• 		if (params.pflag && sep && (sbuf.st_size % 4 == 0))
  

• 			sbuf.st_size += 1;
  	size = cpu_to_uimage (sbuf.st_size);
  } else {
  	size = 0;
  

@@ -556,7 +561,8 @@ copy_file (int ifd, const char *datafile, int pad) exit (EXIT_FAILURE); }

• if (pad && ((tail = size % 4) != 0)) {

• tail = size % 4;

• if (pad && (params.pflag || tail != 0)) {

  if (write(ifd, (char *)&zero, 4-tail) != 4-tail) { fprintf (stderr, "%s: Write error on %s: %s\n",

@@ -586,6 +592,7 @@ usage () " -e ==> set entry point to 'ep' (hex)\n" " -n ==> set image name to 'name'\n" " -d ==> use image data from 'datafile'\n"

• 	 "          -p ==> force padding in multi-file images\n"
  

This is no real padding, so please don't make it look like it is. Maybe use "q"(uirk) as an option character and change the description to indicate that this is not a "forced padding" but a "incorrect additional 32-bit padding to work around an old bug (see man-page)". A fix to "doc/mkimage.1" which now exists is also mandatory.

	 "          -x ==> set XIP (execute in place)\n",
params.cmdname);

fprintf (stderr, " %s [-D dtc_options] -f fit-image.its fit-image\n", diff --git a/tools/mkimage.h b/tools/mkimage.h index 9033a7d..6e18750 100644 --- a/tools/mkimage.h +++ b/tools/mkimage.h @@ -58,6 +58,7 @@ struct mkimage_params { int eflag; int fflag; int lflag;

• int pflag; int vflag; int xflag; int os;

Cheers Detlev