From mboxrd@z Thu Jan 1 00:00:00 1970 From: Paul Chavent Date: Tue, 01 May 2012 17:24:48 +0200 Subject: [Buildroot] Running linux with mmu disabled on arm (AT91). Message-ID: <4FA00040.8000207@fnac.net> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: buildroot@busybox.net Hi. I'd like to try linux on an AT91SAM9G20 (i have a linuxstamp board), with MMU disabled (the Hyok-Sung Choi & Hee-Chul Yun paper demonstrate that it could be possible). All the code seems to be present in the kernel... So i would like to share my experience. (1) I had to make two little changes in the kernel code : - I'm not able to change the REMAP_VECTOR_TO_DRAM. I have submitted the problem to the kbuild mailing list (http://www.spinics.net/lists/linux-kbuild/msg06153.html). - The soc_detect code that init the at91_soc_initdata structure is never called. Later this structure is used unitialized. I'm currently using the joined patch as a workaround. Someone can review it please ? (2) I used to run the module with a "classic system" based on linux + eglibc, and i needed only one toolchain (arm-xxx-linux-gnueabi) for the kernel and the userspace apps. I found that for the nommu case i wasn't able to build the kernel with the userspace toolchain. So I had to build 2 toolchain, based on binutils 2.22, gcc 4.7, uclibc 0.9.33.1 and linux 3.2.14 : - an arm-xxx-eabi toolchain for the kernel - an arm-xxx-uclinux-uclibceabi for the userspace apps Do you confirm that it is not possible to compile the kernel with arm-xxx-uclinux-uclibceabi ? Or, may i have misconfigured the toolchain (i join the toolchain build procedure) ? (3) The arm-xxx-uclinux-uclibceabi with elf2flt seems to produce running bins only if it is build with the uClibc DOPIC not set. Is it required to disable DOPIC ? Moreover, i can't run bins produced with a arm-xxx-linux-uclibceabi toolchain and -Wl,-elf2flt (not uclinux one). Is it the expected behavior ? (4) The elf2flt needs to be updated to be aware of the exidx section, and some reloc types. I join the patch i currently use. Given these observations, i've been able to run a linux with an hello world as the /init process. My next tests will be to run threaded programs, c++ programs, then busybox. Thanks for your attention. Paul. -------------- next part -------------- diff -abBruN linux-3.2.14.orig/arch/arm/Kconfig-nommu linux-3.2.14.mod/arch/arm/Kconfig-nommu --- linux-3.2.14.orig/arch/arm/Kconfig-nommu 2012-04-02 18:53:31.000000000 +0200 +++ linux-3.2.14.mod/arch/arm/Kconfig-nommu 2012-04-15 12:52:05.000000000 +0200 @@ -34,8 +34,7 @@ used instead of the auto-probing which utilizes the register. config REMAP_VECTORS_TO_RAM - bool 'Install vectors to the beginning of RAM' if DRAM_BASE - depends on DRAM_BASE + bool 'Install vectors to the beginning of RAM' help The kernel needs to change the hardware exception vectors. In nommu mode, the hardware exception vectors are normally diff -abBruN linux-3.2.14.orig/arch/arm/mm/nommu.c linux-3.2.14.mod/arch/arm/mm/nommu.c --- linux-3.2.14.orig/arch/arm/mm/nommu.c 2012-04-02 18:53:31.000000000 +0200 +++ linux-3.2.14.mod/arch/arm/mm/nommu.c 2012-04-16 15:50:47.000000000 +0200 @@ -31,12 +31,23 @@ { } +static void __init devicemaps_init(struct machine_desc *mdesc) +{ + /* + * Ask the machine support to map in the statically mapped devices. + */ + if (mdesc->map_io) + mdesc->map_io(); +} + /* * paging_init() sets up the page tables, initialises the zone memory * maps, and sets up the zero page, bad page and bad page tables. */ void __init paging_init(struct machine_desc *mdesc) { + devicemaps_init(mdesc); + bootmem_init(); } -------------- next part --------------