On Thursday 21 May 2009, Zach Welch wrote:
>
> > (NOTE that the armel build turned up *LOTS* of unrelated bugs, not fixed
> > here. Biggest: abusing "u8 *ptr" by "*((u32 *)ptr) = ..." loses badly,
> > since ARM doesn't guarantee unaligned reads work. That idiom is used
> > all over the place in JTAG buffer conversions.)
>
> Yeah. My effort to add warnings pointed these out to me, and I just
> corrected those with calls to memcpy. Clearly, this kind of work not
> only needs to be done, but it almost begs to be encapsulated in a new
> inline functions in src/helper/binarybuffer.[ch].... doesn't it?
Encapsulated, yes; and then there are all the byteswapping routines,
some of which are not needed at all. Some <endian.h> checks like
static __inline__ void arm_le_to_h_u32(u8 *in)
{
- *((u32 *)in)=le_to_h_u32(in);
+#if __BYTE_ORDER != __LITTLE_ENDIAN
+#error woe is me
+#endif
}
could help ... that is, with a memcpy or something similar being
used *only* when host-order is big-endian.
I'd be more comfortable seeing something like the <asm/byteorder.h>
stack used ... it's got a full suite of cpu_to_{le,be}{16,32,64}()
calls, and vice versa; or for that case, le32_to_cpus() for the
in-situ conversions. They morph to NOPs half the time. :)
_______________________________________________
Openocd-development mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/openocd-development
