Hi, On 18 January 2016 at 10:42, Álvaro Fernández Rojas <noltari@xxxxxxxxx> wrote: > I can refine it to support a custom offset for each cpu instead of a generic one, but defining a custom offset for new SoCs such as BCM6368 or BCM6328 would actually break them, since that way the address wouldn't be remapped to 0xb0000000. > See: https://github.com/torvalds/linux/blob/master/arch/mips/include/asm/io.h#L213 > In those CPUs the remapping is done automatically (from 0x10000000 to 0xb0000000), since the registers are located in the low 512MB of address space (0x1fffffffULL). > > However, the older CPUs such as BCM6358 (or BCM3368) need that custom ioremap, since those registers aren't located in the low 512MB of address space. > If you want, I can do something like this: https://gist.github.com/Noltari/bc5fe029c52cf053a454 > And after that we could add other CPUs such as the BCM3368, which needs a different offset: "{ "brcm,bcm3368", 0xfff80000 }" > > What do you think? Should we keep a generic offset (0xfff00000) or should we add SoC specific compatible strings with each custom offset? How about using the arm approach [1] and adding support for registering static mappings? That way we avoid adding a custom plat_ioremap() and keep the amount of bmips code to a minimum. Jonas [1]: https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=ed8fd2186a4e4f3b98434093b56f9b793d48443e https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=101eeda38c0ab8a4f916176e325d9e036d981a24
