Wolfgang Denk wrote:
In message 48188584.3060109@ge.com you wrote:
The solution in the PowerPC world is to add a "eieio" or "sync" instruction[1] appropriately[2], which prevents the bus interface unit from reordering the memory operations.
No, the solution for ALL architectures is to use the appropriate accessor macros which take care of such things (and probably contain calls like sync or similar).
Best regards,
Wolfgang Denk
Agreed.
If my theory is correct, Adrian has turned up a problem with the ARM configuration that apparently only shows up for the PXA250. Or I'm blowing smoke - nothing to see, move along.
The hairs on the back of my neck say it is a sync/bus interface unit problem.
I don't know if it is significant, not being very intimately familiar with ARM (and, in particular, the PXA250), but my quick grep turns up ./include/asm-arm/io.h and a quick browsing shows no syncs. The hairs on the back of my neck are starting to ache. ;-)
static inline void sync(void) { }
#define __arch_getb(a) (*(volatile unsigned char *)(a))
gvb