Date: Mon, 4 Mar 2013 08:31:40 -0800 From: Tim Kientzle <kientzle@freebsd.org> To: Warner Losh <imp@bsdimp.com> Cc: freebsd-arm@freebsd.org Subject: Re: GENERIC kernel issues Message-ID: <9C4AD429-7134-4433-A713-5DFA12628AE9@freebsd.org> In-Reply-To: <549B1B40-99E7-47D4-BA13-1F08507B7B58@bsdimp.com> References: <DF7B73D4-BE50-4E75-8D5B-FE19A4764F31@freebsd.org> <549B1B40-99E7-47D4-BA13-1F08507B7B58@bsdimp.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mar 3, 2013, at 6:39 PM, Warner Losh wrote: >=20 > On Mar 3, 2013, at 12:43 PM, Tim Kientzle wrote: >=20 >> I spent some time yesterday putting together a kernel >> configuration for a GENERIC ARM kernel that would >> support both RaspberryPi and BeagleBone. >>=20 >> Just to see how far I could get. >>=20 >> Here's a list of the problems I've found so far: >>=20 >> ** Multiple MMU support. If you put these two lines into an >> ARM kernel config, the build will fail in the MMU code: >>=20 >> cpu CPU_ARM1176 >> cpu CPU_CORTEXA >>=20 >> Basically, this turns on the support for multiple MMUs but the >> ARMv6/ARMv7 MMU definitions don't play nicely with run-time >> MMU selection. >=20 > Having looked at the defines, it could be done with variables, but I = fear that will slow things down to do a simple #define -> variable. We = may need two sets of code for performance=85 The MMU management I really know nothing about. If you do =85 ;-) >> ** PHYSADDR/KERNPHYSADDR hardwiring. Ian has made a >> lot of progress and I'm working on some related changes to >> address this. I think we understand how to eliminate these >> constants and replace them with run-time detection of the >> load address. I'm still not sure what changes might be needed >> to the loader to make this work. >>=20 >> ** PIPT vs. VIVT cache management. This is currently set at compile >> time; we'll need to have a way to set this at run time based on the >> CPU. (I have some skeletal code to select CPU at the top of >> initarm by inspecting the FDT. I presume this switch will be routine >> once a robust version of that is in place.) >=20 > Generally we should be doing this, both for the Core and the SoC. I = don't think we do this generally, and we should. It is one of the big = advantages of FDT: It tells you what's going on so you don't have to = guess.... >=20 >> ** TI processor detection. This is currently hardwired at build = time, >> so we cannot currently build a kernel that supports both AM335x >> and OMAP4, for example. >=20 > Can't we do this with compat field in the FDT? I have prototype code that does exactly this, but it needs a lot of cleanup before it can be committed. Tim
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?9C4AD429-7134-4433-A713-5DFA12628AE9>