v7-M: Fixing XIP when the kernel is in ROM

[u.kleine-koenig@pengutronix.de (Uwe Kleine-König)]

Hello Ezequiel,

On Tue, Oct 27, 2015 at 07:08:46PM -0300, Ezequiel Garcia wrote:
> On 27 October 2015 at 18:52, Maxime Coquelin <mcoquelin.stm32@gmail.com> wrote:
> > 2015-10-27 22:46 GMT+01:00 Ezequiel Garcia <ezequiel@vanguardiasur.com.ar>:
> >> On 27 October 2015 at 18:33, Maxime Coquelin <mcoquelin.stm32@gmail.com> wrote:
> >>> 2015-10-27 21:33 GMT+01:00 Stefan Agner <stefan@agner.ch>:
> > ...
> >>>> Anyway, I guess Ezequiel's fix is the right thing to do...
> >>> Or, maybe we should put the stack in RAM as Uwe suggest?
> >>>
> >>
> >> It seems the current fix would be used as-is (no need to
> >> set a temporary stack before the SVC call).
> >>
> >> And on top of that, we need a to set a proper stack as soon
> >> as the kernel takes control.
> >>
> >> Am I right?
> > IMHO, yes, that's how I understand it.
> >
> 
> After some deeper code parsing, it seems the stack
> is set in __mmap_switched.
> 
> Although it seems that happens after __v7m_setup,
> ARM noMMU has worked like this since ages now, is v7-M
> really different to need a special treatment?

Yeah, v7-M needs stack in it's setup function, while earlier no-MMU
machines don't. (I didn't check that though.)

So the current situation is that in svc something is written to
__v7m_setup_stack which doesn't seem to hurt in practise on most
machines, but on your's is provokes a hang.

With your patch the same something is written to whereever sp points to
when the kernel is entered. This is hardly better.

So the right fix is to move __v7m_setup_stack to .data I guess.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-K?nig            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |