Re: [PATCH RFC] KVM: TDX: Defer guest memory removal to decrease shutdown time

[Adrian Hunter <adrian.hunter@intel.com>]

On 13/03/25 20:39, Paolo Bonzini wrote:
> On Thu, Mar 13, 2025 at 7:16 PM Adrian Hunter <adrian.hunter@intel.com> wrote:
>> Improve TDX shutdown performance by adding a more efficient shutdown
>> operation at the cost of adding separate branches for the TDX MMU
>> operations for normal runtime and shutdown.  This more efficient method was
>> previously used in earlier versions of the TDX patches, but was removed to
>> simplify the initial upstreaming.  This is an RFC, and still needs a proper
>> upstream commit log. It is intended to be an eventual follow up to base
>> support.
> 
> In the latest code the HKID is released in kvm_arch_pre_destroy_vm().

I am looking at kvm-coco-queue

> That is before kvm_free_memslot() calls kvm_gmem_unbind(), which
> results in fput() and hence kvm_gmem_release().
> 
> So, as long as userspace doesn't remove the memslots and close the
> guestmemfds, shouldn't the TD_TEARDOWN method be usable?

kvm_arch_pre_destroy_vm() is called from kvm_destroy_vm()
which won't happen before kvm_gmem_release() calls
kvm_put_kvm().