Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Medium] drm/rockchip: dw_hdmi_qp: live VRR property mutation in encoder atomic_check() --
commit cf4752529f9d39b6df9beff8735c1aec029921cc Author: Nicolas Frattaroli <[email protected]> drm/rockchip: dw_hdmi_qp: Add VRR support This commit attaches VRR connector properties on RK3588 in dw_hdmi_qp_rockchip_bind() and updates vrr_capable and the VRR display limit properties in dw_hdmi_qp_rockchip_encoder_atomic_check(). > diff --git a/drivers/gpu/drm/rockchip/dw_hdmi_qp-rockchip.c > b/drivers/gpu/drm/rockchip/dw_hdmi_qp-rockchip.c > index f2a1ea8dfefa5..dcde50bca36bf 100644 > --- a/drivers/gpu/drm/rockchip/dw_hdmi_qp-rockchip.c > +++ b/drivers/gpu/drm/rockchip/dw_hdmi_qp-rockchip.c [ ... ] > @@ -195,6 +198,7 @@ dw_hdmi_qp_rockchip_encoder_atomic_check(struct > drm_encoder *encoder, > struct drm_crtc_state *crtc_state, > struct drm_connector_state *conn_state) > { > + struct drm_connector *connector = conn_state->connector; > struct rockchip_crtc_state *s = to_rockchip_crtc_state(crtc_state); > struct rockchip_hdmi_qp *hdmi = to_rockchip_hdmi_qp(encoder); > union phy_configure_opts phy_cfg = {}; > @@ -205,6 +209,12 @@ dw_hdmi_qp_rockchip_encoder_atomic_check(struct > drm_encoder *encoder, > if (ingest_fmt < 0) > return -EINVAL; > > + if (connector->vrr_capable_property) { > + drm_connector_set_vrr_capable_property(connector, > + > drm_display_info_is_vrr_capable(&connector->display_info)); > + drm_connector_hdmi_update_vrr_limits(connector, > &crtc_state->mode); > + } [Severity: Medium] As reported by Sashiko on RFC [1], is it safe to update the live connector VRR properties inside dw_hdmi_qp_rockchip_encoder_atomic_check()? Both drm_connector_set_vrr_capable_property() and drm_connector_hdmi_update_vrr_limits() call drm_object_property_set_value() on connector->base, which directly mutates connector->base.properties->values[] because vrr_capable, vrr_display_min, and vrr_display_max are immutable properties rather than transactional drm_connector_state fields. When userspace performs a DRM_MODE_ATOMIC_TEST_ONLY commit to probe candidate modes, or when an atomic check fails later in the pipeline, this path: drm_mode_atomic_ioctl() drm_atomic_check_only() drm_atomic_helper_check() drm_atomic_helper_check_modeset() mode_fixup() dw_hdmi_qp_rockchip_encoder_atomic_check() immediately overwrites the live user-visible vrr_display_min and vrr_display_max values with limits derived from the uncommitted crtc_state->mode. In addition, mode_fixup() in drivers/gpu/drm/drm_atomic_helper.c skips the encoder atomic_check when a connector has no CRTC or encoder assigned: if (!new_conn_state->crtc || !new_conn_state->best_encoder) continue; When a VRR-capable display is connected at boot or hotplugged while the connector is disabled, DRM_IOCTL_MODE_GETCONNECTOR probes the connector and updates EDID without running dw_hdmi_qp_rockchip_encoder_atomic_check(), leaving vrr_capable at 0 until a modeset occurs. Conversely, when a VRR display is unplugged and its CRTC is disabled (new_conn_state->crtc == NULL), mode_fixup() skips the encoder check and leaves vrr_capable stuck at 1 and vrr_display_min / vrr_display_max holding the disconnected sink's values. Finally, connector->display_info and connector->status (read by drm_display_info_is_vrr_capable() and drm_connector_hdmi_update_vrr_limits()) are protected by dev->mode_config.mutex, whereas atomic_check only holds connection_mutex and crtc->mutex, which can race with concurrent hotplug updates in drm_bridge_connector_handle_hpd(). Should vrr_capable be updated in the connector hotplug / EDID probe path while holding mode_config.mutex, and should the mode-dependent VRR limits be updated during atomic commit rather than atomic_check? [1] https://lore.kernel.org/r/[email protected] -- Sashiko AI review ยท https://sashiko.dev/#/patchset/[email protected]?part=12
