[PATCH v3 07/10] ARM: install HYP vectors at PBL and Barebox entry

[Lucas Stach]

Am Donnerstag, den 05.04.2018, 09:42 +0200 schrieb Sascha Hauer:
> On Mon, Mar 26, 2018 at 09:20:22PM +0200, Lucas Stach wrote:
> > If the CPU was already in HYP mode when entering the PBL, install a
> > simple trap handler to allow to get back from SVC to HYP before
> > switching to HYP mode.
> > 
> > As the vectors are part of the currently running binary, we need to
> > do the same setup when starting the real Barebox binary, as the PBL
> > setup vectors might get overwritten. To do this we trap into HYP mode
> > just before jumping to Barebox and then re-do the vector setup and
> > SVC switch as the first thing in Barebox proper.
> > 
> > > > Signed-off-by: Lucas Stach <l.stach at pengutronix.de>
> > > > Tested-by: Roland Hieber <r.hieber at pengutronix.de>
> > ---
> >  arch/arm/cpu/lowlevel.S   | 3 +++
> >  arch/arm/cpu/start-pbl.c  | 4 ++++
> >  arch/arm/cpu/start.c      | 3 +++
> >  arch/arm/cpu/uncompress.c | 4 ++++
> >  4 files changed, 14 insertions(+)
> > 
> > diff --git a/arch/arm/cpu/lowlevel.S b/arch/arm/cpu/lowlevel.S
> > index 43665981e48b..13dfe496ad17 100644
> > --- a/arch/arm/cpu/lowlevel.S
> > +++ b/arch/arm/cpu/lowlevel.S
> > @@ -8,6 +8,9 @@ ENTRY(arm_cpu_lowlevel_init)
> > > >  	/* save lr, since it may be banked away with a processor mode change */
> > > > > >  	mov	r2, lr
> >  
> > > > +	/* careful: the hyp install corrupts r0 and r1 */
> > > > > > +	bl	__hyp_install
> > +
> > > >  	/* set the cpu to SVC32 mode, mask irq and fiq */
> > > > > >  	mrs	r12, cpsr
> > > > > >  	eor	r12, r12, #HYP_MODE
> > diff --git a/arch/arm/cpu/start-pbl.c b/arch/arm/cpu/start-pbl.c
> > index e851b4a2da5e..e0793579e2e8 100644
> > --- a/arch/arm/cpu/start-pbl.c
> > +++ b/arch/arm/cpu/start-pbl.c
> > @@ -26,6 +26,7 @@
> >  #include <asm/barebox-arm-head.h>
> >  #include <asm-generic/memory_layout.h>
> >  #include <asm/sections.h>
> > +#include <asm/secure.h>
> >  #include <asm/pgtable.h>
> >  #include <asm/cache.h>
> >  #include <asm/unaligned.h>
> > @@ -100,5 +101,8 @@ __noreturn void barebox_single_pbl_start(unsigned long membase,
> > > >  	else
> > > >  		barebox = (void *)barebox_base;
> >  
> > > > +	if (__boot_cpu_mode == HYP_MODE)
> > > > +		armv7_switch_to_hyp();
> > +
> > > >  	barebox(membase, memsize, boarddata);
> >  }
> > diff --git a/arch/arm/cpu/start.c b/arch/arm/cpu/start.c
> > index 171e6ad0eb7a..a0db6436f387 100644
> > --- a/arch/arm/cpu/start.c
> > +++ b/arch/arm/cpu/start.c
> > @@ -24,6 +24,7 @@
> >  #include <asm/barebox-arm-head.h>
> >  #include <asm-generic/memory_layout.h>
> >  #include <asm/sections.h>
> > +#include <asm/secure.h>
> >  #include <asm/unaligned.h>
> >  #include <asm/cache.h>
> >  #include <memory.h>
> > @@ -145,6 +146,8 @@ __noreturn void barebox_non_pbl_start(unsigned long membase,
> > > >  	unsigned long malloc_start, malloc_end;
> > > >  	unsigned long barebox_size = barebox_image_size + MAX_BSS_SIZE;
> >  
> > +	armv7_hyp_install();
> 
> Calling a armv7 specific function in a generic code path is not so nice.
> If this is really necessary then I suggest to #ifdef it here rather than
> in the header file (where armv7_hyp_install() expands to a noop for
> armv8. And what about the older architectures? A quick test revealed
> this works on armv5, but is this intentional or by accident?

Yep, it is coded to be a quick return on older < ARMv7 with the
executed instructions being compatible. I'll still change this to a
IS_ENABLED condition, as this might save some code size if only !ARMv7
arches are enabled in the config.

Thanks,
Lucas