Re: [PATCHv2 12/12] arm64: docs: document pointer authentication

[Mark Rutland]

On Tue, Nov 28, 2017 at 04:07:26PM +0100, Andrew Jones wrote:
> Hi Mark,

Hi Drew,

> On Mon, Nov 27, 2017 at 04:38:06PM +0000, Mark Rutland wrote:
> > +Architecture overview
> > +---------------------
> > +
> > +The ARMv8.3 Pointer Authentication extension adds primitives that can be
> > +used to mitigate certain classes of attack where an attacker can corrupt
> > +the contents of some memory (e.g. the stack).
> > +
> > +The extension uses a Pointer Authentication Code (PAC) to determine
> > +whether pointers have been modified unexpectedly. A PAC is derived from
> > +a pointer, another value (such as the stack pointer), and a secret key
> > +held in system registers.
> > +
> > +The extension adds instructions to insert a valid PAC into a pointer,
> > +and to verify/remove the PAC from a pointer. The PAC occupies a number
> > +of high-order bits of the pointer, which varies dependent on the
> > +configured virtual address size and whether pointer tagging is in use.
> > +
> > +A subset of these instructions have been allocated from the HINT
> > +encoding space. In the absence of the extension (or when disabled),
> > +these instructions behave as NOPs. Applications and libraries using
> > +these instructions operate correctly regardless of the presence of the
> > +extension.
> 
> Correctly, but obviously without the additional security. So I assume
> it's expected that applications that demand this security to probe for
> it themselves, presumably by the checking the HWCAP. Is that correct?

Yes. Applications which wish to mandate pointer authentication
(presumably using instructions outside of the HINT space) must check the
relevant HWCAP first.

[...]

> > +Virtualization
> > +--------------
> > +
> > +When CONFIG_ARM64_POINTER_AUTHENTICATION is selected, and uniform HW
> > +support is present, KVM will context switch all keys used by vCPUs.
> > +Otherwise, the feature is disabled. When disabled, accesses to keys, or
> > +use of instructions enabled within the guest will trap to EL2, and an
> > +UNDEFINED exception will be injected into the guest.
> 
> If host applications will just run, with the instructions behaving like
> NOPs, when the extension is either not present or not enabled, then
> shouldn't guest applications also just run?

The enabled/disabled wording is probably the confusing bit here.

At EL1 we have conditional enables for instructions using
AP{I,D}{A,B}Key, which behave as NOPs when disabled.

At EL2 we have a single conditional trap for all instructions using
pointer authentication, that traps to EL2 when instructions are not
NOP'd by EL1.

So "disabled by EL2" is actually "trapped by EL2", and "disabled by EL1"
is "NOP'd by EL1".

> I.e. instead of injecting UNDEF, just treat the instructions as NOPs.
> Or did I misunderstand the trapping?

I think the documentation explained it poorly. Did the above help?

> Does use of the instructions at EL0 trap to EL1 or EL2?

If disabled by EL1, the instructions behave as NOPs (regardless of the
EL2 traps).

If enabled by EL1, but trapped by EL2, the instructions trap to EL2.

If enabled by EL1, and not trapped by EL2, the instructions work as
usual.

I'll see if I can document this better.

Thanks,
Mark.