On Mon, Jun 26, 2023 at 05:30:07PM +0200, Krzysztof Kozlowski wrote:
> On 26/06/2023 15:54, Nícolas F. R. A. Prado wrote:
> > On Fri, Jun 23, 2023 at 06:21:31PM +0200, Krzysztof Kozlowski wrote:
> >> On 21/06/2023 20:00, Nícolas F. R. A. Prado wrote:
> >>>>
> >>>> But anyway this variant comes with some set of regs and reg-names. Other
> >>>> variant comes with different set. In all cases they should be defined,
> >>>> even by "defined" means not allowed.
> >>>
> >>> I'm not sure what you mean. Are you suggesting to disable reg-names on mt8173?
> >>
> >> That's one of the options if for some reason you don't want to define them.
> >>
> >>>
> >>>>
> >>>>>
> >>>>> But in a separate series we could drop vdecsys from mt8173's reg as well,
> >>>>> passing it as a syscon instead, which would solve the warning on that platform,
> >>>>> though some more driver changes would be needed to be able to handle it for that
> >>>>> SoC. The newer SoCs like mt8192, mt8195, etc, should also get vdecsys dropped
> >>>>> from their regs to have a correct memory description.
> >>>>>
> >>>>
> >>>> Sure, but I don't understand how does it affect defining and making
> >>>> specific regs/reg-names or keeping them loose.
> >>>
> >>> We need some way to tell in the driver whether the first reg is VDEC_SYS or not.
> >>> Since so far reg-names have not been used for the vcodec, the simplest, and
> >>> cleanest, way to do it, is to add reg-names when VDEC_SYS is not present. When
> >>> the other SoCs are updated to no longer have the first reg as VDEC_SYS, they
> >>> would also have reg-names added to their binding, to clearly indicate that.
> >>
> >> Don't use reg-names for that. The order of entries is anyway strict.
> >
> > Since the order of entries is strict, if I remove VDEC_SYS from mt8183, I also
> > need to remove it from mt8173, is that what you mean?
>
> It's different compatible, so it can have different entries.
>
>
> > I would still check for
> > the presence of reg-names in the driver to differentiate whether the old or new
> > binding is used, you just don't want different reg-names between compatibles in
> > the binding?
>
> I wrote already what I want:
>
> In all cases they should be defined, even by "defined" means not allowed.
>
> Now of course the best would be if the reg-names are always the same, at
> least in respect of order of items. This is what we try to do for all
> devices.
>
> >
> >>
> >>>
> >>> For example, for mt8173 we currently have
> >>>
> >>> vcodec_dec: vcodec@16000000 {
> >>> compatible = "mediatek,mt8173-vcodec-dec";
> >>> reg = <0 0x16000000 0 0x100>, /* VDEC_SYS */
> >>> <0 0x16020000 0 0x1000>, /* VDEC_MISC */
> >>> <0 0x16021000 0 0x800>, /* VDEC_LD */
> >>> <0 0x16021800 0 0x800>, /* VDEC_TOP */
> >>> <0 0x16022000 0 0x1000>, /* VDEC_CM */
> >>> <0 0x16023000 0 0x1000>, /* VDEC_AD */
> >>> <0 0x16024000 0 0x1000>, /* VDEC_AV */
> >>> <0 0x16025000 0 0x1000>, /* VDEC_PP */
> >>> <0 0x16026800 0 0x800>, /* VDEC_HWD */
> >>> <0 0x16027000 0 0x800>, /* VDEC_HWQ */
> >>> <0 0x16027800 0 0x800>, /* VDEC_HWB */
> >>> <0 0x16028400 0 0x400>; /* VDEC_HWG */
> >>>
> >>> In a future series, when removing VDEC_SYS from it, it would become
> >>>
> >>> vcodec_dec: vcodec@16020000 {
> >>> compatible = "mediatek,mt8173-vcodec-dec";
> >>> reg = <0 0x16020000 0 0x1000>, /* VDEC_MISC */
> >>> <0 0x16021000 0 0x800>, /* VDEC_LD */
> >>> <0 0x16021800 0 0x800>, /* VDEC_TOP */
> >>> <0 0x16022000 0 0x1000>, /* VDEC_CM */
> >>> <0 0x16023000 0 0x1000>, /* VDEC_AD */
> >>> <0 0x16024000 0 0x1000>, /* VDEC_AV */
> >>> <0 0x16025000 0 0x1000>, /* VDEC_PP */
> >>> <0 0x16026800 0 0x800>, /* VDEC_HWD */
> >>> <0 0x16027000 0 0x800>, /* VDEC_HWQ */
> >>> <0 0x16027800 0 0x800>, /* VDEC_HWB */
> >>> <0 0x16028400 0 0x400>; /* VDEC_HWG */
> >>> reg-names = "misc", "ld", "top", "cm", "ad", "av", "pp",
> >>> "hwd", "hwq", "hwb", "hwg";
> >>
> >> So you want to use reg-names to avoid ABI break. This is not the reason
> >> not to define reg-names for other case.
> >
> > There will be an ABI break anyway when the first reg is removed (as shown
> > above), I'm just trying to avoid churn: adding a reg-name that will be removed
> > later.
>
> So remove the reg-name now and there will be no "later"?
OK, I'll send a v4 with VDEC_SYS also removed from mt8173.
Thanks,
Nícolas
