Re: [PATCH 2/2] ARM: layerscape: ppa: Fix starting PPA

[Sascha Hauer <s.hauer@pengutronix.de>]

On Mon, Nov 20, 2023 at 03:54:40PM +0100, Ahmad Fatoum wrote:
> Hello Sascha,
> 
> On 20.11.23 15:44, Sascha Hauer wrote:
> > With dba1c26f70 we replaced request_sdram_region() for the PPA with
> > reserve_sdram_region(). The effect is that the region is marked as
> > reserved and mapped non executable. While this is desired for EL2, it
> > also has the effect that we can't start the PPA anymore from EL3.
> > 
> > Map the region cached/executable to start the PPA, then map it
> > uncached/non executable once we are in EL2.
> > 
> > Fixes: dba1c26f70 ("arm: layerscape: ppa: reserve SDRAM region for PPA")
> > Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
> > ---
> >  arch/arm/mach-layerscape/ppa.c | 18 +++++++++---------
> >  1 file changed, 9 insertions(+), 9 deletions(-)
> > 
> > diff --git a/arch/arm/mach-layerscape/ppa.c b/arch/arm/mach-layerscape/ppa.c
> > index 521e6b89da..43145a7ece 100644
> > --- a/arch/arm/mach-layerscape/ppa.c
> > +++ b/arch/arm/mach-layerscape/ppa.c
> > @@ -4,6 +4,7 @@
> >  
> >  #include <common.h>
> >  #include <init.h>
> > +#include <mmu.h>
> >  #include <firmware.h>
> >  #include <memory.h>
> >  #include <linux/sizes.h>
> > @@ -54,17 +55,11 @@ static int ppa_init(void *ppa, size_t ppa_size, void *sec_firmware_addr)
> >  	int ret;
> >  	u32 *boot_loc_ptr_l, *boot_loc_ptr_h;
> >  	struct ccsr_scfg __iomem *scfg = (void *)(LSCH2_SCFG_ADDR);
> > -	int el = current_el();
> >  	struct fit_handle *fit;
> >  	void *conf;
> >  	const void *buf;
> >  	unsigned long firmware_size;
> >  
> > -	if (el < 3) {
> > -		printf("EL%d, skip ppa init\n", el);
> > -		return 0;
> > -	}
> > -
> >  	boot_loc_ptr_l = &scfg->scratchrw[1];
> >  	boot_loc_ptr_h = &scfg->scratchrw[0];
> >  
> > @@ -115,6 +110,7 @@ int ls1046a_ppa_init(resource_size_t ppa_start, resource_size_t ppa_size)
> >  	struct resource *res;
> >  	void *ppa_fw;
> >  	size_t ppa_fw_size;
> > +	int el = current_el();
> >  	int ret;
> >  
> >  	res = reserve_sdram_region("ppa", ppa_start, ppa_size);
> > @@ -126,9 +122,13 @@ int ls1046a_ppa_init(resource_size_t ppa_start, resource_size_t ppa_size)
> >  
> >  	get_builtin_firmware(ppa_ls1046a_bin, &ppa_fw, &ppa_fw_size);
> >  
> > -	ret = ppa_init(ppa_fw, ppa_fw_size, (void *)ppa_start);
> > -	if (ret)
> > -		return ret;
> > +	if (el == 3) {
> > +		remap_range((void *)ppa_start, ppa_size, MAP_CACHED);
> > +		ret = ppa_init(ppa_fw, ppa_fw_size, (void *)ppa_start);
> > +		remap_range((void *)ppa_start, ppa_size, MAP_UNCACHED);
> > +		if (ret)
> > +			return ret;
> 
> Isn't this still racy? I think the EL2 mapping should already have the
> PPA area uncached _before_ jumping into PPA.

It likely is, but to fix that we would have to create a second set of
pagetables and use one for EL3 with PPA executable and another one for EL2
with PPA non executable.

As of now this is purely academic anyway as the PPA doesn't protect its
SDRAM range in any way, so speculating into it doesn't do any harm.

Sascha

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |