On Fri, Jul 10, 2026 at 01:27:37PM +0300, Cristian Ciocaltea wrote:
> On 7/8/26 1:11 PM, Cristian Ciocaltea wrote:
> > Hi Maxime,
> > 
> > On 7/7/26 7:10 PM, Maxime Ripard wrote:
> >> On Fri, Jul 03, 2026 at 10:31:55PM +0300, Cristian Ciocaltea wrote:
> >>> Hi Dmitry,
> >>>
> >>> Thanks for your quick review!
> >>>
> >>> On 7/3/26 5:05 PM, Dmitry Baryshkov wrote:
> >>>> On Thu, Jul 02, 2026 at 05:46:15PM +0300, Cristian Ciocaltea wrote:
> >>>>> In preparation for adding HDMI 2.x source capabilities, introduce struct
> >>>>> drm_connector_hdmi_caps and a new drmm_connector_hdmi_init_with_caps()
> >>>>> helper.
> >>>>>
> >>>>> The existing drmm_connector_hdmi_init() helper currently takes
> >>>>> individual capability arguments such as supported_formats and max_bpc.
> >>>>> Adding more HDMI-specific arguments to that function would not scale
> >>>>> well, so move those values into a dedicated capabilities structure and
> >>>>> implement the existing helper as a wrapper around the new caps-based
> >>>>> interface.
> >>>>
> >>>> I think, it was an intention of Maxime: make sure that every driver is
> >>>> forced to provide some values here. With the struct-based init it is
> >>>> easy to overlook or to ommit a value.
> >>>
> >>> Agreed that the struct-based init loses the compile-time guarantee that 
> >>> every
> >>> argument is explicitly provided - that's a real downside.  
> >>>
> >>> I'd argue it's recoverable, though: the init helper validates the 
> >>> mandatory
> >>> fields, so a driver that omits a required value gets rejected at init time
> >>> rather than silently misconfigured.  The "you must provide sane values" 
> >>> property
> >>> is expected to be preserved, just enforced at runtime instead of by the
> >>> compiler. 
> >>
> >> Yeah, I don't think we can win with C here. Rust might, but we're
> >> probably a long way from that.
> >>
> >>> The main motivation for the struct is scalability/maintainability as we 
> >>> add HDMI
> >>> 2.x capabilities: new fields go into the struct rather than growing the 
> >>> helper's
> >>> argument list, so existing callers don't need churny signature updates on 
> >>> every
> >>> extension.
> >>>
> >>> FWIW, in the previous revision we discussed addressing the concern with a
> >>> callback instead.  Sadly, I had to discard that approach, as it proved not
> >>> flexible enough, e.g. drm_bridge_connector_init() computes caps 
> >>> dynamically, and
> >>> would have required either stateful callbacks, or storing 
> >>> redundant/temporary
> >>> cap data in driver-private structures just to satisfy the callback.
> >>
> >> I just realized something reviewing your patch: we don't necessarily
> >> need an extra argument or a callback, we can just put these fields into
> >> drm_hdmi_connector_funcs directly, and then validate them in init.
> > 
> > If I understand correctly, we should drop the drm_connector_hdmi_caps struct
> > introduced by this patch and move all its fields into 
> > drm_hdmi_connector_funcs.
> > 
> > In that case, how should we proceed with drmm_connector_hdmi_init()? 
> 
> Actually, this brings us to the callback issue: we cannot compute caps
> dynamically, as it only works with static data, since funcs is supposed to be
> immutable.

Does it? The core and helpers must consider it immutable but it doesn't
have to. drm_bridge_connector for example could totally allocate it and
dynamically create it based on the bridge capabilities.

> Moreover, we would end up with some redundancy, as most of these input fields,
> after validation, would be stored (altered or not) directly in
> drm_connector/drm_connector_hdmi structs.

I mean, part of the conversion would obviously be to remove the
redundant fields.

Attachment: signature.asc
Description: PGP signature

Reply via email to