Anatolij Gustschin wrote:
quoted
quoted
ad->pix_fmt >> > - diu_ops.get_pixel_format(var->bits_per_pixel,
- machine_data->monitor_port);
+ cpu_to_be32(diu_ops.get_pixel_format(machine_data->monitor_port,
+ var->bits_per_pixel));
I think you can drop cpu_to_be32 here, it is somehow confusing. The
descriptor fields are LE32,
Not according to the structure definition:
struct diu_ad {
/* Word 0(32-bit) in DDR memory */
/* __u16 comp; */
/* __u16 pixel_s:2; */
/* __u16 pallete:1; */
/* __u16 red_c:2; */
/* __u16 green_c:2; */
/* __u16 blue_c:2; */
/* __u16 alpha_c:3; */
/* __u16 byte_f:1; */
/* __u16 res0:3; */
__be32 pix_fmt; /* hard coding pixel format */
I see other places where the endianness of pix_fmt is assumed. I'll have
to fix those later.
When I look at the 5121 and 1022 reference manuals, it says that the area
descriptor is little-endian. I think York made this register big-endian
so that it matches the endianness of get_pixel_format(), which doesn't
make much sense. I'll have to fix the whole endianness thing with pix_fmt
in another patch.
> get_pixel_format() returns LE32 value.
I don't think the endianness of get_pixel_format() is fixed. It's
whatever the CPU native endiannes is. That's why I added a cpu_to_be32().
Anyway, I'm not quite sure what the right answer is, but it seems obvious
that cpu_to_be32() is wrong. I'll post another patch which removes it.
--
Timur Tabi
Linux kernel developer at Freescale