On 20/05/2024 14:14, Nishanth Menon wrote: > On 08:53-20240520, Krzysztof Kozlowski wrote: >> On 19/05/2024 22:27, Francesco Dolcini wrote: > > [...] >>> If it's not the case we'll send the patch later on, however some >>> DT files maintainers (e.g. arch/arm64/boot/dts/ti/) have a policy to >>> just accept DT file in which the binding changes are already merged >>> therefore I was trying to be a little bit proactive here. >> >> TI? Never heard something like this from them... Such requirement would >> seriously slow down any work, so it's not really reasonable. Expectation >> is to post both binding change and an user, so DTS, in case of USB in >> separate patchsets. > > There is a reason we have set that "soft rule": > - Driver subsystem merges have known to be broken from time to time and > the dt maintainer is left holding compatibles that have not made to > master. You mean driver tree took the bindings (because then you take the DTS) but they did not make it up to the master? Yeah, happens, but postponing DTS code for this is quite harsh. > - ARM subsystem merges prefers not to see checkpatch warnings - > typically, this happens with new compatibles in the driver subsystem. Never heard any comment about this. Maybe because I always do checkpatch tests (part of workflow), so then any warning, which there are pretty lot like touching maintainer files, is understood as accepted. > - Off chance that driver subsystem maintainer picks up the dt changes as > well (should not happen, but has happened) DTS changes? Yes, that's why they should be in separate patchsets for few known trees, like USB, net, media. > > We have however flexed the rule when: > a) driver maintainer is willing to provide us an immutable tag that we > can merge in and base the dts on top. You cannot base DTS on top of any driver branch. In the past Arnd and Olof were rejecting this, recently this got more flexible, but still cannot. This implies dependency, which is a no-go. > b) We felt that the chances of the driver not making it is very very low > (typically after 1+ month in next) and the dts change is in the wider > interest of the community. In such case, we have to explicitly take > the action of letting the patch submitter, driver subsystem to let us > know if something bad happens to the PR, also in our PR to SoC > maintainers, we have to call it out along with rationale why this is > OK. This is a bunch of work from a lot of folks, so prefer only to > trigger this path in case of exceptional cases - there have been a > few far in between. > > Again, the default rule (driver in one window, binding in next) has > kept us out of trouble for a few years now at the detriment of pace > of merges, but that took care of a lot of conflicts that we had seen > during initial days of k3 - there are few chains in the lakml list > where this was the direction we ended up in after discussion. > > But, yes - as you mentioned, send the patches of the "user" of the dt > binding and driver gives the subsystem and dt maintainers a chance to > review in the context of usage prior to the driver and binding merge. Anyway, I am fine, but then I expect from the contributor to explain where is the user. Specifically, for completely new bindings I will NAK them if there is no driver or DTS posted. Why? Because experience shown me that people post some bindings and then entirely different DTS, without running full dtbs_check. Therefore I want to see both - bindings and DTS, if applicable. It is easy to fulfill both requirements - mine and yours. Post bindings. Post DTS and link the bindings, with explanation it should be picked up *later*, because that's what TI expects. That's way I can still find both pieces, but you do not risk taking DTS too early. Best regards, Krzysztof
