On 13-07-21, 22:34, Arnd Bergmann wrote: > On Tue, Jul 13, 2021 at 9:35 PM Rob Herring <robh+dt@xxxxxxxxxx> wrote: > > There shouldn't be. We have nodes be multiple providers (e.g clocks > > and resets) already. > > I think this would be a little different, but it can still work. There is in > fact already some precedent of doing this, with Jean-Philippe's virtio-iommu > binding, which is documented in both > > Documentation/devicetree/bindings/virtio/iommu.txt > Documentation/devicetree/bindings/virtio/mmio.txt > > Unfortunately, those are still slightly different from where I think we should > be going here, but it's probably close enough to fit into the general > system. > > What we have with virtio-iommu is two special hacks: > - on virtio-mmio, a node with 'compatible="virtio,mmio"' may optionally > have an '#iommu-cells=<1>', in which case we assume it's an iommu. > - for virtio-pci, the node has the standard PCI 'reg' property but a special > 'compatible="virtio,pci-iommu"' property that I think is different from any > other PCI node. > > I think for other virtio devices, we should come up with a way to define a > binding per device (i2c, gpio, ...) without needing to cram this into the > "virtio,mmio" binding or coming up with special compatible strings for > PCI devices. > > Having a child device for the virtio device type gives a better separation > here, since it lets you have two nodes with 'compatible' strings that each > make sense for their respective parent buses: The parent is either a PCI > device or a plain mmio based device, and the child is a virtio device with > its own namespace for compatible values. As you say, the downside is > that this requires an extra node that is redundant because there is always > a 1:1 relation with its parent. > > Having a combined node gets rid of the redundancy but if we want to > identify the device for the purpose of defining a custom binding, it would have > to have two compatible strings, something like > > compatible="virtio,mmio", "virtio,device34"; > > for a virtio-mmio device of device-id 34 (i2c), or a PCI device with > > compatible="pci1af4,1041", "virtio,device34"; > > which also does not quite feel right. I agree that even if the device is discoverable at runtime, we should still have some sort of stuff in DT to distinguish the devices, and "virtio,deviceDID" sounds good enough for that, considering that we already do it for USB, etc. And I am fine with both the ways, a new node or just using the parent node. So whatever you guys decide is fine. > > Direct userspace access to I2C, GPIO, etc. has its issues, we're going > > to repeat that with guests? > > Passing through the i2c or gpio access from a Linux host is just one > way to use it, you could do the same with an emulated i2c device > from qemu, and you could have a fake i2c device behind a virtio-i2c > controller. Or it can be firmware controlled device as well, as Rob said earlier. I think that's what Vincent will be doing for SCMI. Though all I have tested until now is direct userspace access. -- viresh
