On 08/17/2010 01:50 AM, Sascha Hauer wrote: > On Tue, Aug 17, 2010 at 08:35:54AM +0200, Sascha Hauer wrote: >> On Mon, Aug 16, 2010 at 10:51:38PM -0500, Nishanth Menon wrote: >>> Hi, >>> while just giving the omap3530 boot from internal sram a whirl, found that >>> >>> commit 78104ae181f78d83664fdc7522eb632d9c3b2ec9 >>> Author: Sascha Hauer >>> Date: Fri Apr 16 15:50:30 2010 +0200 >>> >>> arm: reimplement startup code in C >>> >>> Lets translate the startup code to a language we all understand better. >>> Tested on pcm038 (arm v5) and pcm043 (arm v6). >>> >>> Signed-off-by: Sascha Hauer >>> >>> broke the support for OMAP3530 internal boot capability -> it looks like >>> many of the V7Dcache skip logic got dropped off in the patch. will try >>> to fix it ahead.. reverting this commit helps the barebox master.. >> >> We worked on Omap lately. I remember that there was a problem with the >> call to arch_init_lowlevel. Michael, do you have input on this? > > This commit indeed broke omap because the call to arch_init_lowlevel was > missing afterwards. This got fixed with 3aa5c83. Maybe you have a > different problem and stumbled upon this commit while bisecting. You > could try checking out 78104ae181f78d83664fdc7522eb632d9c3b2ec9 and > cherry-pick 3aa5c83. Does this work? will try later tonight (but I doubt it - I had tried master and next - it did not build yesterday due to ehci/otg changes, but those have patches from Michael today), but to give an idea, attached is my revert patch I applied(against master), but if you see there are two parts between arch_init_lowlevel and board_init_lowlevel: cache invalidate and configuration of cache Just glancing, cache invalidate logic that got introduced is the v4/v5(?) logic I guess, not the v7 logic. let me see later tonight if there is a cleaner way to do it.. Regards, Nishanth Menon