On Fri, Nov 13, 2026 at 5:42 AM Harry Wentland <[email protected]> wrote: > > On 11/11/26 20:43, Dave Airlie via B4 Relay wrote: > > From: Dave Airlie <[email protected]> > > > > Split the caps view accessor into per-capability queries so DM and the > > DAL modules no longer read through struct dc_caps: > > > > - scalar getters for is_apu, ips/dmcub support, max links/streams/ > > planes/slave planes/cursor size/v_total, vtotal_limited_by_fp2, > > i2c speed, PCON/DSC/fused-io support, SubVP processing delay and > > MALL size > > - dc_dmub_supports_psr()/dc_dmub_supports_aux_backlight() for the > > DMUB capability bits > > - dc_get_color_caps() copies out the color pipeline capabilities > > - dc_get_plane_caps() returns the static plane capability array > > > > The KUnit tests keep installing fake capabilities through the > > dc_test_caps() view of their own fake DC. > > > > No production code outside the dc directory reads struct dc members > > through a caps view anymore. > > > > Part of making the struct dc sub-structures opaque outside DC. > > > > Signed-off-by: Dave Airlie <[email protected]> > > --- > > <snip> > > > > > unsigned int dc_get_disable_ips(const struct dc *dc) > > diff --git a/drivers/gpu/drm/amd/display/dc/inc/dc.h > > b/drivers/gpu/drm/amd/display/dc/inc/dc.h > > index 31baba820f6c..7a49f2ee1d77 100644 > > --- a/drivers/gpu/drm/amd/display/dc/inc/dc.h > > +++ b/drivers/gpu/drm/amd/display/dc/inc/dc.h > > @@ -2351,8 +2351,38 @@ struct dc_context *dc_get_ctx(const struct dc *dc); > > /* The ASIC display engine version. */ > > enum dce_version dc_get_dce_version(const struct dc *dc); > > > > -/* Capabilities detected at dc_create() time. Read-only. */ > > -const struct dc_caps *dc_get_caps(const struct dc *dc); > > +/* > > + * Capabilities detected at dc_create() time. Read-only; prefer the > > + * purpose-built queries below over reaching into the caps structure. > > + */ > > +bool dc_is_apu(const struct dc *dc); > > +bool dc_get_ips_support(const struct dc *dc); > > +bool dc_get_dmcub_support(const struct dc *dc); > > +unsigned int dc_get_max_links(const struct dc *dc); > > +unsigned int dc_get_max_streams(const struct dc *dc); > > +unsigned int dc_get_max_planes(const struct dc *dc); > > +unsigned int dc_get_max_slave_planes(const struct dc *dc); > > +unsigned int dc_get_max_cursor_size(const struct dc *dc); > > +unsigned int dc_get_max_v_total(const struct dc *dc); > > +bool dc_get_vtotal_limited_by_fp2(const struct dc *dc); > > +unsigned int dc_get_i2c_speed_in_khz(const struct dc *dc); > > +bool dc_get_dp_hdmi21_pcon_support(const struct dc *dc); > > +bool dc_get_edp_dsc_support(const struct dc *dc); > > +bool dc_get_fused_io_support(const struct dc *dc); > > +unsigned int dc_get_subvp_fw_processing_delay_us(const struct dc *dc); > > +unsigned int dc_get_mall_size_total(const struct dc *dc); > > +bool dc_dmub_supports_psr(const struct dc *dc); > > +bool dc_dmub_supports_aux_backlight(const struct dc *dc); > > + > > +/* Copy out the color pipeline capabilities. */ > > +void dc_get_color_caps(const struct dc *dc, struct dc_color_caps *caps); > > + > > +/* > > + * Return the static plane capability array and its entry count. > > + * Valid until the DC is destroyed. > > + */ > > +const struct dc_plane_cap *dc_get_plane_caps(const struct dc *dc, > > + unsigned int *count); > > > > This creates some API bloat. > > What about creating a new public struct header and throwing dc_caps > (or a subset of it) in there? > > The intent to decouple how DC stores and reports caps is valid.
Yes I think it's probably a good idea, the last couple of patches were definitely pushing the limits of where I wanted to go, dc_caps is at least a reasonably self contained if not a bit haphazard. Dave.
