Re: [PATCH v3 08/12] KVM: Propagate vcpu explicitly to mark_page_dirty_in_slot()

On Thu, Nov 18, 2021, David Woodhouse wrote:

On 18 November 2021 18:50:55 GMT, Sean Christopherson [off-list ref] wrote:

quoted

On Thu, Nov 18, 2021, Sean Christopherson wrote:

quoted

On Thu, Nov 18, 2021, David Woodhouse wrote:

quoted

That leaves the one in TDP MMU handle_changed_spte_dirty_log() which
AFAICT can trigger the same crash seen by butt3rflyh4ck — can't that
happen from a thread where kvm_get_running_vcpu() is NULL too? For that
one I'm not sure.

I think could be trigger in the TDP MMU via kvm_mmu_notifier_release()
-> kvm_mmu_zap_all(), e.g. if the userspace VMM exits while dirty logging is
enabled.  That should be easy to (dis)prove via a selftest.

Scratch that, the dirty log update is guarded by the new_spte being present, so
zapping of any kind won't trigger it.

Currently, I believe the only path that would create a present SPTE without an
active vCPU is mmu_notifer.change_pte, but that squeaks by because its required
to be wrapped with invalidate_range_{start,end}(MMU_NOTIFY_CLEAR), and KVM zaps
in that situation.

Is it sufficient to have *an* active vCPU?  What if a VMM has threads for
active vCPUs but is doing mmap/munmap on a *different* thread? Does that not
suffer the same crash?

It is sufficient for the current physical CPU to have an active vCPU, which is
generally guaranteed in the MMU code because, with a few exceptions, populating
SPTEs is done in vCPU context.

mmap() will never directly trigger SPTE creation, KVM first requires a vCPU to
fault on the new address.  munmap() is a pure zap flow, i.e. won't create a
present SPTE and trigger the writeback of the dirty bit.

That's also why I dislike using kvm_get_running_vcpu(); when it's needed, there's
a valid vCPU from the caller, but it deliberately gets dropped and indirectly
picked back up.