On 4/22/22 11:48, Marc Zyngier wrote:
On Thu, 21 Apr 2022 19:04:40 +0100,
Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote:
The KVM_SYSTEM_EVENT_NDATA_VALID mechanism that was introduced
contextually with KVM_SYSTEM_EVENT_SEV_TERM is not a good match
for ARM and RISC-V, which want to communicate information even
for existing KVM_SYSTEM_EVENT_* constants. Userspace is not ready
to filter out bit 31 of type, and fails to process the
KVM_EXIT_SYSTEM_EVENT exit.
Therefore, tie the availability of ndata to a system capability
(which will be added once all architectures are on board).
Userspace written for released versions of Linux has no reason to
check flags, since it was never written, so it is okay to replace
it with ndata and data[0] (on 32-bit kernels) or with data[0]
(on 64-bit kernels).
How is it going to work for new userspace on old kernels, for which
the ndata field is left uninitialised?
New userspace needs to check the capability before accessing ndata.
If it is desirable for Android, crosvm can "quirk" that ARM always has
valid data[0] even in the absence of the capability.
Cat we please get a #define that aliases data[0] to flags? At the next
merge of the KVM headers into their respective trees, all the existing
VMM are going to break if they have a reference to this field (CrosVM
definitely does today -- yes, we're ahead of time).
That would be a union rather than a define but yes, I can do it.
Paolo
