On Wed, 12 Jun 2024 at 11:41, Mark Rutland <mark.rutland@xxxxxxx> wrote:
>
> Given how badly things go when we get this wrong (e.g. TLB corruption), I'd
> like to say "just mark it incompatible", applying to all instrumentation, not
> just KCSAN.
Ack. I'll start out with just KCSAN (since that's the actual technical
issue right now). But since the SW PAN support is hopefully not
something that we should worry about going forward, I wouldn't mind it
being de-emphasized.
It's not like PAN is something that should necessarily be everywhere.
The real advantage of SMAP on x86 (and then PAN on arm) is that it
catches wild kernel pointers. As long as the HW support is common
enough, people will find bugs on those platforms.
So I think the advantage of SW PAN was "it will find issues early
before HW PAN is widely available". It might be time to lay SW PAN
entirely to rest now.
I'll send out a new version of the arm64 patches with the KCSAN build
failure fixed (with the simple no-op save/restore functions by making
KCSAN and SW PAN mutually incompatible), and with the virtual address
fix you pointed out in the other email.
Linus
