On 27.01.2025 11:40 PM, Bjorn Andersson wrote: > On Mon, Jan 27, 2025 at 06:57:41PM +0100, Stephan Gerhold wrote: >> On Mon, Jan 13, 2025 at 09:11:33PM -0800, Bjorn Andersson wrote: > [..] >>> 118 files changed, 8389 insertions(+), 670 deletions(-) >>> --- >> >> This is quite a lot of code and new files for a temporary migration. >> It's also difficult to test these changes fully, since there are >> separate overlays for each SoC and sometimes even each board. >> >> Would it be easier to just duplicate the dwc3-qcom driver for now? >> Making a copy of the current dwc3-qcom.c would be just 1000 lines of >> extra code, compared to more than 7000 for the overlay approach. >> >> The copy (e.g. dwc3-qcom-legacy.c) would keep handling the old bindings >> with the existing code (that is known to work to some extent). We can >> then improve upon the main version without risk of breaking any old >> DTBs. If we decide to drop support for the old DTBs at some point, we >> can just drop dwc3-qcom-legacy. >> >> This approach is also not pretty, but I think the risk and effort would >> be lower than making sure the overlay approach works on all the affected >> targets. >> > > I like this suggestion. > > It's much more isolated and we know the current state of the driver > works with the current dtbs out there - so backwards compatibility would > be handled. I also did end up having to use separate compatibles for the > old and new binding/driver, so this should be quite clean - i.e. > nicer than the overlay-based path... > > The one drawback would be that devices that isn't updated to a new dtb > would not gain the upcoming improved support for role switching, or any > of the improvements in pm_runtime-support (as I assume we'd only care > about the new driver). But I think that's worth the saving in > complexity. If a user explicitly decides not to update, there's only so much we can do Konrad
