On Thu, Jan 26, 2023 at 11:04 AM Sudeep Holla <sudeep.holla@xxxxxxx> wrote: > > On Thu, Jan 26, 2023 at 09:25:12AM -0600, Rob Herring wrote: > > On Thu, Jan 26, 2023 at 8:46 AM Sudeep Holla <sudeep.holla@xxxxxxx> wrote: > > > > > > On Thu, Jan 26, 2023 at 09:43:44AM +0000, Cristian Marussi wrote: > > > > On Wed, Jan 25, 2023 at 02:11:13PM +0000, Sudeep Holla wrote: > > > > > On Wed, Jan 25, 2023 at 01:43:48PM +0000, Cristian Marussi wrote: > > > > > > so now that the catch-all protocol@ patternProperty is gone in favour > > > > > > of the 'protocol-node' definition and $refs, does that mean that any > > > > > > current and future SCMI officially published protocol <N> has to be > > > > > > added to the above explicit protocol list, even though it does not > > > > > > have any special additional required property beside reg ? > > > > > > (like protocol@18 above...) > > > > > > > > > > > > > > > > If there are no consumers, should we just not add and deal with it > > > > > entirely within the kernel. I know we rely today on presence of node > > > > > before we initialise, but hey we have exception for system power protocol > > > > > for other reasons, why not add this one too. > > > > > > > > > > In short we shouldn't have to add a node if there are no consumers. It > > > > > was one of the topic of discussion initially when SCMI binding was added > > > > > and they exist only for the consumers otherwise we don't need it as > > > > > everything is discoverable from the interface. > > > > > > > > It is fine for me the no-consumers/no-node argument (which anyway would > > > > require a few changes in the core init logic anyway to work this way...), > > > > BUT is it not that ANY protocol (even future-ones) does have, potentially, > > > > consumers indeed, since each protocol-node can potentially have a dedicated > > > > channel and related DT channel-descriptor ? (when multiple channels are > > > > allowed by the transport) > > > > > > > > I mean, as an example, you dont strictly need protos 0x18/0x12 nodes for > > > > anything (if we patch the core init as said) UNLESS you want to dedicate > > > > a channel to those protocols; so I'm just checking here if these kind of > > > > scenarios will still be allowed with this binding change, or if I am > > > > missing something. > > > > > > Ah, good point on the transport information. Yes we will need a node if > > > a protocol has a dedicated transport. No one has used so far other than > > > Juno perf, but we never know. We can always extended the bindings if > > > needed. > > > > > > Sorry for missing the dedicated transport part. > > > > So I need to add back 'protocol@.*' or not? > > IMO it is better to know what exactly gets added under each of these protocol > sub-nodes and so better to have entry specific to each known protocols. I > liked that fact with this change as I have seen some crazy vendor extensions > adding all sorts of non-sense defining some vendor protocol. For example [1], > in which case we can catch those better than existing schema which matches > all. So let's not add protocol@.* if possible or until that becomes the only > cleaner way to maintain this. TBC, 'protocol@.*' would not allow anything but the properties defined in the /$defs/protocol-node. So [1] would throw errors without a schema addition. We should either do that along with dropping 'protocol@18' or we keep protocol 0x18 node and add all other providerless protocols. I don't think we need the latter to just check unit-address vs. reg. I want to come up with a better way to do that (dtc does some, but only for defined bus types). Rob
