[PATCH 3/4 v2] lib: add crc16_le helper

[Joakim Tjernlund]

Matthieu CASTET <matthieu.castet at parrot.com> wrote on 2010/08/19 09:42:49:
>
> Hi,
>
> Joakim Tjernlund a écrit :
> > Matthieu CASTET <matthieu.castet at parrot.com> wrote on 2010/08/18 18:10:11:
> >>> if so, 0x8005 looks
> >>> wrong too. see CRCPOLY_LE resp. CRCPOLY_BE for an idea.
> >> Why ?
> >
> > Because changing endian of crc also require the POLY
> > to be bit reversed as well. See:
> >  #define CRCPOLY_LE 0xedb88320
> >  #define CRCPOLY_BE 0x04c11db7
> > So I assume you need to do that as well for crc16_be, otherwise it
> > won't be a BE version of the standard crc16 LE
> Well it match the crc provided in onfi nand [1]. But it may be not a
> real BE version.
>
>
> >
> >>> What is this crc sum used for?
> >> Onfi flash parsing in mtd.
> >
> > And this is a one time operation of fairly small amount of
> > data? I ask because you impl. is really slow.
> Yes it checks only 253 bytes at startup time (one time only). So it is
> not speed critical.

OK.

>
> > Why can't you use the crc16 LE version?
> >
> Because it doesn't compute the crc provided by onfi spec. Or is there
> some magic maths to convert LE version to our version ?

I suspect the onfi spec got it wrong and isn't computing a real
crc16_be. I suggest you rename this to onfi_crc16_be and move it
inside the onfi subsystem. I supposed you are stuck with
the onfi version?
Just for the record, I think the crc16_be should be like this:

u16 crc16_le(u16 crc, u8 const *p, size_t len)
{
   int i;
   while (len--) {
      crc ^= *p++ << 8;
      i = 8;
      do {
         crc = (crc << 1) ^ ((crc & 0x8000) ? 0xa001 : 0);
      } while(--i);
   }
   return crc;
}

Don't know of an easy way to go between LE and BE versions but
have a look at lib/crc32.c, there is a unit test in there
that suggests that on can byte reverse the data to go between
LE and BE.

[Deleted ugly code]