[PATCH v4 02/16] arm: mach-k3: Add basic support for J784S4 SoC definition

[Nishanth Menon]

On 09:46-20231006, Manorit Chawdhry wrote:
> Hi Nishanth,
> 
> On 06:26-20231005, Nishanth Menon wrote:
> > On 10:29-20231005, Manorit Chawdhry wrote:
> > > Hi Nishanth,
> > > 
> > > On 07:24-20231004, Nishanth Menon wrote:
> > > > On 10:43-20231004, Manorit Chawdhry wrote:
> > > > 
> > > > > These are required to remove the firewall configurations that are done
> > > > > by ROM, those are not the ones that are being handled by OIDs. The
> > > > 
> > > > I am not sure I understand this clearly. OIDs are setup to open up
> > > > firewalls or close firewalls as the system requires and since it
> > > > is authenticated, not compromiseable.- U-boot by itself (even if
> > > > authenticated), is not a secure entity for it to dictate the firewall
> > > > configuration (u-boot must be assumed to be compromised after
> > > > authentication is complete). So, doing firewall configuration via APIs
> > > > after boot, to me looks broken approach.
> > > > 
> > > 
> > > I know U-boot ain't that secure given the most trusted entity is always
> > > gonna be the software that starts up the system, we can't expect those
> > > to be doing all the work and based on that we have the secure boot
> > > designed to configure firewalls (that are not owned by anymore) and
> > > U-boot R5 being one of the early bootloaders do come as a part of it. 
> > > 
> > > Regarding the OIDs thing, I don't think the OID in question is looked by
> > > ROM and ROM always configures some firewalls for it's usecase that are
> > > present in those arrays. 
> > > 
> > > The OID that we are using in the series that you had shared is looked by
> > > TIFS instead of ROM and TIFS is the entity that is authenticating the
> > > binary along with setting up the firewalls.
> > > 
> > > > > current series that is being worked on is to add additional firewalling
> > > > > support with OIDs that TIFS will be handling.
> > > > > The above patch is
> > > > > essentially added to have the same development experience on GP devices
> > > > > similar to HS after the secure boot is done so that people don't end up
> > > > 
> > > > huh? the code seems to blindly call the remove_fwl_configs(cbass_hc_cfg0_fwls, ARRAY_SIZE(cbass_hc_cfg0_fwls));
> > > > where is the distinction of HS vs GP here? This implementation looks
> > > > completely broken to me at least.. please correct what I missed here.
> > > 
> > > Since this call is used across all SoCs there wasn't any point to make
> > > the differentiation between GP and HS here, remove_fwl_configs
> > > internally handles looking at the firewalls and disabling them if they
> > > are enabled ( Which would be only in the case of HS devices ), for GP it
> > > would automatically by a noop.
> > 
> > Correct me if I understand the security chain here:
> > 
> > ROM sets up firewalls that are needed by itself
> > TIFS (in multicertificate will setup it's own firewalls)
> > R5 SPL comes along and opens up other firewalls
> > Each stage beyond this: such as tispl.bin containing TFA/OPTEE uses OIDs to
> > set up firewalls to protect themselves (enforced by TIFS)
> > A53 SPL and U-boot itself startups but has no ability to change the
> > protection firewalls enforced by x509 OIDs.
> > 
> > Further, firewalls have lockdown bit that enforces the setting (and
> > cannot be over-ridden) till system restart is requested
> > 
> > Is this correct? If so, needs to be clearly documented.
> 
> Yes, this seems correct. Will be updating it in the upstream
> documentation in another series.
> 

Thanks. I suggest:
a) Add documentation as part of firewall series[1] including information
as to how to customize the flow as needed.
b) once the firewall series is merged
c) am69/j784s4 series and add reference in code to the section of
documentation.

[1] https://lore.kernel.org/all/20231004-binman-firewalling-v3-0-e4a102324e1f@ti.com/
-- 
Regards,
Nishanth Menon
Key (0xDDB5849D1736249D) / Fingerprint: F8A2 8693 54EB 8232 17A3  1A34 DDB5 849D 1736 249D