Re: bootwrapper can't be compiled for cubieboard2

[Andre Przywara <andre.przywara@linaro.org>]

On 08/30/2013 10:06 AM, Ian Campbell wrote:
> On Fri, 2013-08-30 at 15:12 +0800, Josh Zhao wrote:
>> Hi Ian,
>> I try compile the bootwrapper for cubieboard2 like this:
>
> Hi Josh,
>
> cubieboards2 support is not really in place yet, so I'm afraid there
> will currently be rough edges like this.
>
> That said I think this worked for me, but based on the discussions I
> think the better approach is to merge Andre's u-boot patches to support
> entry in HYP mode into the cubieboard branch. I'll only work on
> bootwrapper any further if some interesting platform comes along for
> which there is no possibility of updating the bootloader, so for now you
> should consider it dead/dormant.
>
> Anyhow, I've tried inserting Andre's patches into the cubieboard2 uboot
> and stumbled across a problem -- it turns out that u-boot has enabled
> the MMU (SCTLR.M), apparently because it is required for the caches to
> be on (see arch/arm/lib/cache-cp15.c:cache_enable). unfortunately SCTLR
> is banked so when nonsec-virt.S:_secure_monitor enabled SCR.NS
> everything falls apart. Naïvely enabling SCTLR.M while in monitor mode
> didn't seem to help.

Ah, good point. My whole code approach was relying on the fact that 
u-boot runs with MMU _off_. With the MMU differences in HYP mode we may 
run into troubles here, it may work if we switch late enough, though.

> Calling cleanup_before_linux() early in armv7_switch_nonsec() seemed to
> do the job -- Andre, do you think that is the right fix?

Do you have a pointer to a u-boot repository with the cubieboard 
patches? Then I'd take a look.

> Not sure why this didn't affect vexpress etc -- perhaps caches aren't
> enabled for that platform?

AFAIK it's rather normal for u-boot to run with at least some caches 
off. These only get turned on if really necessary.
So I'd refrain from enabling the MMU just to support some cache levels 
on a bootloader.

Thanks for looking at this!
Andre.

> Even with this fixed things don't quite work for me, I don't quite
> remember how/why (I last looked at this 7 days ago). I've been trying to
> set aside Friday as "cubieboard hacking day", but I've ended up with a
> bunch of meetings today, hopefully I'll still manage to investigate a
> bit.
>
> Ian.
>
>