> From: David Brownell [mailto:david-b@xxxxxxxxxxx] <snip> > On Wednesday 03 September 2008, Woodruff, Richard wrote: > > Fixed translations do have some benefits. You can ensure that you > > are using section or super section descriptors to cover large areas. > > This does result in better TLB usage. Along with freeing up TLB > > entries you also generally avoid TLB misses on IO calls which > > touch a variety of internal spaces as part of the IRQ sequence. > > ... which is exactly why some linux/arch/... code makes sure ioremap() > can return fixed mappings instead of always requiring dynamic ones. That does sound like a nice consistent way to do it. Which implementation of this would you recommend looking at? Do any ARMs you know of do it today? > True enough, but that decision can be in the hands of whoever > builds a kernel. If you want to deploy an A8-optimized system, > the kernel will be only one part of the equation. As long as the option exists to make the optimized build, yours is a pretty reasonable stance. Practically, I wonder about user communities you are trying to support. If there are tons of active users at each generation then it seems to make sense to cater to them all. If generations are quick flashes then coupling seems less compelling. Regards, Richard W. -- To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html