Re: Security Random Number Generator support

[Neal Liu]

On Wed, 2020-06-03 at 17:34 +0800, Russell King - ARM Linux admin wrote:
> On Wed, Jun 03, 2020 at 08:40:58AM +0100, Marc Zyngier wrote:
> > On 2020-06-03 08:29, Neal Liu wrote:
> > > On Tue, 2020-06-02 at 21:02 +0800, Marc Zyngier wrote:
> > > > On 2020-06-02 13:14, Ard Biesheuvel wrote:
> > > > > On Tue, 2 Jun 2020 at 10:15, Neal Liu <neal.liu@xxxxxxxxxxxx> wrote:
> > > > >>
> > > > >> These patch series introduce a security random number generator
> > > > >> which provides a generic interface to get hardware rnd from Secure
> > > > >> state. The Secure state can be Arm Trusted Firmware(ATF), Trusted
> > > > >> Execution Environment(TEE), or even EL2 hypervisor.
> > > > >>
> > > > >> Patch #1..2 adds sec-rng kernel driver for Trustzone based SoCs.
> > > > >> For security awareness SoCs on ARMv8 with TrustZone enabled,
> > > > >> peripherals like entropy sources is not accessible from normal world
> > > > >> (linux) and rather accessible from secure world (HYP/ATF/TEE) only.
> > > > >> This driver aims to provide a generic interface to Arm Trusted
> > > > >> Firmware or Hypervisor rng service.
> > > > >>
> > > > >>
> > > > >> changes since v1:
> > > > >> - rename mt67xx-rng to mtk-sec-rng since all MediaTek ARMv8 SoCs can
> > > > >> reuse
> > > > >>   this driver.
> > > > >>   - refine coding style and unnecessary check.
> > > > >>
> > > > >>   changes since v2:
> > > > >>   - remove unused comments.
> > > > >>   - remove redundant variable.
> > > > >>
> > > > >>   changes since v3:
> > > > >>   - add dt-bindings for MediaTek rng with TrustZone enabled.
> > > > >>   - revise HWRNG SMC call fid.
> > > > >>
> > > > >>   changes since v4:
> > > > >>   - move bindings to the arm/firmware directory.
> > > > >>   - revise driver init flow to check more property.
> > > > >>
> > > > >>   changes since v5:
> > > > >>   - refactor to more generic security rng driver which
> > > > >>     is not platform specific.
> > > > >>
> > > > >> *** BLURB HERE ***
> > > > >>
> > > > >> Neal Liu (2):
> > > > >>   dt-bindings: rng: add bindings for sec-rng
> > > > >>   hwrng: add sec-rng driver
> > > > >>
> > > > >
> > > > > There is no reason to model a SMC call as a driver, and represent it
> > > > > via a DT node like this.
> > > > 
> > > > +1.
> > > > 
> > > > > It would be much better if this SMC interface is made truly generic,
> > > > > and wired into the arch_get_random() interface, which can be used much
> > > > > earlier.
> > > > 
> > > > Wasn't there a plan to standardize a SMC call to rule them all?
> > > > 
> > > >          M.
> > > 
> > > Could you give us a hint how to make this SMC interface more generic in
> > > addition to my approach?
> > > There is no (easy) way to get platform-independent SMC function ID,
> > > which is why we encode it into device tree, and provide a generic
> > > driver. In this way, different devices can be mapped and then get
> > > different function ID internally.
> > 
> > The idea is simply to have *one* single ID that caters for all
> > implementations, just like we did for PSCI at the time. This
> > requires ARM to edict a standard, which is what I was referring
> > to above.
> 
> This sounds all too familiar.
> 
> This kind of thing is something that ARM have seems to shy away from
> doing - it's a point I brought up many years ago when the whole
> trustzone thing first appeared with its SMC call.  Those around the
> conference table were not interested - ARM seemed to prefer every
> vendor to do off and do their own thing with the SMC interface.

Does that mean it make sense to model a sec-rng driver, and get each
vendor's SMC function id by DT node?

> 
> Then OMAP came along with its SMC interfaces, and so did the pain of
> not having a standardised way to configure the L2C when Linux was
> running in the non-secure world, resulting in stuff like l2c_configure
> etc, where each and every implementation has to supply a function to
> call its platform specific SMC interfaces to configure a piece of
> hardware common across many different platforms.
> 
> ARM have seemed reluctant to standardise on stuff like this, so
> unless someone pushes hard for it from inside ARM, I doubt it will
> ever happen.
>