On Wed, Mar 31, 2021 at 11:41:20AM +0100, Steven Price wrote:
On 31/03/2021 10:32, David Hildenbrand wrote:
On 31.03.21 11:21, Catalin Marinas wrote:
On Wed, Mar 31, 2021 at 09:34:44AM +0200, David Hildenbrand wrote:
On 30.03.21 12:30, Catalin Marinas wrote:
On Mon, Mar 29, 2021 at 05:06:51PM +0100, Steven Price wrote:
On 28/03/2021 13:21, Catalin Marinas wrote:
However, the bigger issue is that Stage 2 cannot disable
tagging for Stage 1 unless the memory is Non-cacheable or
Device at S2. Is there a way to detect what gets mapped in
the guest as Normal Cacheable memory and make sure it's
only early memory or hotplug but no ZONE_DEVICE (or
something else like on-chip memory)?� If we can't
guarantee that all Cacheable memory given to a guest
supports tags, we should disable the feature altogether.
In stage 2 I believe we only have two types of mapping -
'normal' or DEVICE_nGnRE (see stage2_map_set_prot_attr()).
Filtering out the latter is a case of checking the 'device'
variable, and makes sense to avoid the overhead you
describe.
This should also guarantee that all stage-2 cacheable
memory supports tags,
as kvm_is_device_pfn() is simply !pfn_valid(), and
pfn_valid() should only
be true for memory that Linux considers "normal".
If you think "normal" == "normal System RAM", that's wrong; see
below.
By "normal" I think both Steven and I meant the Normal Cacheable memory
attribute (another being the Device memory attribute).
Sadly there's no good standardised terminology here. Aarch64 provides the
"normal (cacheable)" definition. Memory which is mapped as "Normal
Cacheable" is implicitly MTE capable when shared with a guest (because the
stage 2 mappings don't allow restricting MTE other than mapping it as Device
memory).
So MTE also forces us to have a definition of memory which is "bog standard
memory"[1] separate from the mapping attributes. This is the main memory
which fully supports MTE.
Separate from the "bog standard" we have the "special"[1] memory, e.g.
ZONE_DEVICE memory may be mapped as "Normal Cacheable" at stage 1 but that
memory may not support MTE tags. This memory can only be safely shared with
a guest in the following situations:
1. MTE is completely disabled for the guest
2. The stage 2 mappings are 'device' (e.g. DEVICE_nGnRE)
3. We have some guarantee that guest MTE access are in some way safe.
(1) is the situation today (without this patch series). But it prevents the
guest from using MTE in any form.
(2) is pretty terrible for general memory, but is the get-out clause for
mapping devices into the guest.
(3) isn't something we have any architectural way of discovering. We'd need
to know what the device did with the MTE accesses (and any caches between
the CPU and the device) to ensure there aren't any side-channels or h/w
lockup issues. We'd also need some way of describing this memory to the
guest.
So at least for the time being the approach is to avoid letting a guest with
MTE enabled have access to this sort of memory.
When a slot is added by the VMM, if it asked MTE in guest (I guess
that's an opt-in by the VMM, haven't checked the other patches), can we
reject it if it's is going to be mapped as Normal Cacheable but it is a
ZONE_DEVICE (i.e. !kvm_is_device_pfn() + one of David's suggestions to
check for ZONE_DEVICE)? This way we don't need to do more expensive
checks in set_pte_at().