Re: [RFC PATCH v2 02/13] KVM: guest_memfd: Support private/shared conversion of device memory

[Aneesh Kumar K.V <aneesh.kumar@kernel.org>]

Ackerley Tng <ackerleytng@google.com> writes:

> "Aneesh Kumar K.V (Arm)" <aneesh.kumar@kernel.org> writes:
>
>>
>> [...snip...]
>>
>> +static int kvm_gmem_device_convert(struct inode *inode, pgoff_t start,
>> +		pgoff_t nr_pages, const struct guest_memfd_device_request *req,
>> +		const struct kvm_memory_slot *slot)
>> +{
>> +	struct gmem_inode *gi = GMEM_I(inode);
>> +	u64 attrs = req ? KVM_MEMORY_ATTRIBUTE_PRIVATE : 0;
>> +	bool prepared = false;
>> +	int ret;
>> +
>> +	MA_STATE(mas, &gi->attributes, start, start + nr_pages - 1);
>> +
>> +	filemap_invalidate_lock(inode->i_mapping);
>> +	if (__kvm_gmem_range_has_attributes(inode, start, nr_pages, attrs)) {
>> +		ret = 0;
>> +		goto out;
>> +	}
>> +
>> +	if (req && !__kvm_gmem_range_has_attributes(inode, start, nr_pages, 0)) {
>> +		ret = -EEXIST;
>> +		goto out;
>> +	}
>> +
>> +	ret = kvm_gmem_mas_preallocate(&mas, attrs, start, nr_pages);
>> +	if (ret)
>> +		goto out;
>> +
>> +	kvm_gmem_invalidate_start(inode, start, start + nr_pages, KVM_FILTER_SHARED);
>> +	if (req && READ_ONCE(gi->device.kvm->vm_dead))
>> +		ret = -EIO;
>> +	else
>> +		ret = gi->provider_ops->prepare_conversion(gi->provider, req);
>> +	if (!ret) {
>> +		prepared = true;
>> +		if (req)
>> +			ret = kvm_gmem_device_make_private(inode, slot, req, nr_pages);
>
> Is there a way to avoid making private from within the conversion flow?
>
> kvm_gmem_device_map() below already does it as part of vcpu_run.
>
> If we do the above, we'd be aligned with regular conversions, where
> to-shared conversions eagerly update arch state (make_shared) but
> to-private notifications to the firmware are done lazily.
>

For the TDISP flow, the vdev must be mapped before the guest can issue
evidence accept. The flow I am currently trying to follow is: the guest
requests validation of the MMIO range, triggering a VM exit; VFIO
validates the range and then converts it to private via
kvm_gmem_set_attributes_ioctl().

In my new series, I have added a backend operation that abstracts this
across folios, RAM PFNs, and device PFNs.

-aneesh