[PATCH v3 00/20] arm64: Unmap the kernel whilst running in userspace (KPTI)

[Ard Biesheuvel]

On 5 January 2018 at 16:06, Greg Kroah-Hartman
<gregkh at linuxfoundation.org> wrote:
> On Thu, Jan 04, 2018 at 10:23:40AM -0800, Florian Fainelli wrote:
>> On 01/03/2018 10:50 PM, Greg Kroah-Hartman wrote:
>> > On Wed, Jan 03, 2018 at 09:17:26PM -0800, Florian Fainelli wrote:
>> >> On 12/11/2017 09:59 AM, Catalin Marinas wrote:
>> >>> On Wed, Dec 06, 2017 at 12:35:19PM +0000, Will Deacon wrote:
>> >>>> Patches are also pushed here:
>> >>>>
>> >>>>   git://git.kernel.org/pub/scm/linux/kernel/git/will/linux.git kpti
>> >>>>
>> >>>> Feedback and testing welcome. At this point, I'd like to start thinking
>> >>>> about getting this merged for 4.16.
>> >>>
>> >>> For the record, the fixed up version was pushed by Will here:
>> >>>
>> >>> git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux.git kpti
>> >>>
>> >>> and I queued it for 4.16 in the arm64 for-next/core branch (same tree as
>> >>> above).
>> >>
>> >> Greg proposed the x86/KPTI patches for the stable-4.9.75 queue, is there
>> >> a plan to get the ARM64/KPTI patches backported towards stable trees as
>> >> well?
>> >
>> > Stable tree patches have to get into Linus's tree first before I can do
>> > anything :)
>> >
>> > Anyway, once that happens, yes, there is a plan, but it's a bit
>> > "different", and I'll talk about it once these are merged.
>>
>> Great, thanks! Bonus question, if someone is using any of the affected
>> devices in AArch32, should we be expecting to see ARM/Linux changes as
>> well, that is, is there a plan to come up with a kpti implementation for
>> ARM?
>
> I have not heard of anyone working on this for any arm32 platforms,
> as of this time, sorry.
>
> Which makes me worry about my android tv, glad I don't connect it to the
> network :(
>

The only ARM variant that is currently known to be affected by
Meltdown/variant 3 (which is what KPTI addresses) is the Cortex-A75,
which is a 64-bit core. That still means 32-bit guests running under
KVM will be affected, as well as a 32-bit kernel running on the bare
metal, but in practice, 32-bit ARM simply doesn't need KPTI. (My KASLR
patches for ARM are a bit in limbo atm, but those would benefit from
unmapping the kernel while running in userland as well)

As for variants 1/2 aka Spectre, I suppose ARM will need to implement
the same nospec/retpoline primitives that are being proposed for other
arches, but that work is not as fleshed out yet.