Re: [PATCH v3 1/1] x86: Add Isolated Memory Regions for Quark X1000

[Pavel Machek <pavel@ucw.cz>]

On Tue 2015-02-24 22:40:15, Bryan O'Donoghue wrote:
> On 23/02/15 22:18, Pavel Machek wrote:
> >On Mon 2015-01-26 14:15:27, Bryan O'Donoghue wrote:
> 
> >
> >Do the applications normally need to manipulate IMRs?
> 
> 
> Applications could in theory manipulate IMRs - you might want to place an
> IMR around an EFI capsule in memory for example - before calling a capsule
> update.
> 
> This code will place an IMR around the kernel .text - .rodata which ensures
> that no unwarranted DMA access can rewrite write-only kernel addresses -
> something the MMU would not fault on - on non-IMR enabled processors.
> 
> >Would it be
> >possible to do all IMR manipulations in the bootloader?
> >
> 
> Possible yes - in practical terms for Galileo or the SMARC+Quark from
> Kontron for example - you'd be forcing a bootloader change - which most
> users will not pick up.
> 
> 
> Considering IMRs can reset the system if they aren't sanitized, it's good
> practice for the kernel to go and make sure that every unlocked IMR is
> torn-down and reset.

Ok, makes sense.
										Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html