Merge tag 'amd-drm-next-6.11-2024-06-22' of https://gitlab.freedesktop.org/agd5f/linux into drm-next

amd-drm-next-6.11-2024-06-22:

amdgpu:
- HPD fixes
- PSR fixes
- DCC updates
- DCN 4.0.1 fixes
- FAMS fixes
- Misc code cleanups
- SR-IOV fixes
- GPUVM TLB flush cleanups
- Make VCN less verbose
- ACPI backlight fixes
- MES fixes
- Firmware loading cleanups
- Replay fixes
- LTTPR fixes
- Trap handler fixes
- Cursor and overlay fixes
- Primary plane zpos fixes
- DML 2.1 fixes
- RAS updates
- USB4 fixes
- MALL fixes
- Reserved VMID fix
- Silence UBSAN warnings

amdkfd:
- Misc code cleanups

From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240622152523.2267072-1-alexander.deucher@amd.com
Signed-off-by: Dave Airlie <airlied@redhat.com>

1 files changed, 466 insertions, 164 deletions

diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
index ac18cddeb191..c96407379224 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
@@ -77,9 +77,11 @@
 #include <linux/types.h>
 #include <linux/pm_runtime.h>
 #include <linux/pci.h>
+#include <linux/power_supply.h>
 #include <linux/firmware.h>
 #include <linux/component.h>
 #include <linux/dmi.h>
+#include <linux/sort.h>
 
 #include <drm/display/drm_dp_mst_helper.h>
 #include <drm/display/drm_hdmi_helper.h>
@@ -375,6 +377,20 @@ static inline void reverse_planes_order(struct dc_surface_update *array_of_surfa
 		swap(array_of_surface_update[i], array_of_surface_update[j]);
 }
 
+/*
+ * DC will program planes with their z-order determined by their ordering
+ * in the dc_surface_updates array. This comparator is used to sort them
+ * by descending zpos.
+ */
+static int dm_plane_layer_index_cmp(const void *a, const void *b)
+{
+	const struct dc_surface_update *sa = (struct dc_surface_update *)a;
+	const struct dc_surface_update *sb = (struct dc_surface_update *)b;
+
+	/* Sort by descending dc_plane layer_index (i.e. normalized_zpos) */
+	return sb->surface->layer_index - sa->surface->layer_index;
+}
+
 /**
  * update_planes_and_stream_adapter() - Send planes to be updated in DC
  *
@@ -399,7 +415,8 @@ static inline bool update_planes_and_stream_adapter(struct dc *dc,
 						    struct dc_stream_update *stream_update,
 						    struct dc_surface_update *array_of_surface_update)
 {
-	reverse_planes_order(array_of_surface_update, planes_count);
+	sort(array_of_surface_update, planes_count,
+	     sizeof(*array_of_surface_update), dm_plane_layer_index_cmp, NULL);
 
 	/*
 	 * Previous frame finished and HW is ready for optimization.
@@ -774,9 +791,9 @@ static void dmub_hpd_callback(struct amdgpu_device *adev,
 		aconnector = to_amdgpu_dm_connector(connector);
 		if (link && aconnector->dc_link == link) {
 			if (notify->type == DMUB_NOTIFICATION_HPD)
-				DRM_INFO("DMUB HPD callback: link_index=%u\n", link_index);
-			else if (notify->type == DMUB_NOTIFICATION_HPD_IRQ)
 				DRM_INFO("DMUB HPD IRQ callback: link_index=%u\n", link_index);
+			else if (notify->type == DMUB_NOTIFICATION_HPD_IRQ)
+				DRM_INFO("DMUB HPD RX IRQ callback: link_index=%u\n", link_index);
 			else
 				DRM_WARN("DMUB Unknown HPD callback type %d, link_index=%u\n",
 						notify->type, link_index);
@@ -788,10 +805,13 @@ static void dmub_hpd_callback(struct amdgpu_device *adev,
 	drm_connector_list_iter_end(&iter);
 
 	if (hpd_aconnector) {
-		if (notify->type == DMUB_NOTIFICATION_HPD)
+		if (notify->type == DMUB_NOTIFICATION_HPD) {
+			if (hpd_aconnector->dc_link->hpd_status == (notify->hpd_status == DP_HPD_PLUG))
+				DRM_WARN("DMUB reported hpd status unchanged. link_index=%u\n", link_index);
 			handle_hpd_irq_helper(hpd_aconnector);
-		else if (notify->type == DMUB_NOTIFICATION_HPD_IRQ)
+		} else if (notify->type == DMUB_NOTIFICATION_HPD_IRQ) {
 			handle_hpd_rx_irq(hpd_aconnector);
+		}
 	}
 }
 
@@ -859,7 +879,31 @@ static void dm_dmub_outbox1_low_irq(void *interrupt_params)
 	struct dmcub_trace_buf_entry entry = { 0 };
 	u32 count = 0;
 	struct dmub_hpd_work *dmub_hpd_wrk;
-	struct dc_link *plink = NULL;
+	static const char *const event_type[] = {
+		"NO_DATA",
+		"AUX_REPLY",
+		"HPD",
+		"HPD_IRQ",
+		"SET_CONFIGC_REPLY",
+		"DPIA_NOTIFICATION",
+	};
+
+	do {
+		if (dc_dmub_srv_get_dmub_outbox0_msg(dm->dc, &entry)) {
+			trace_amdgpu_dmub_trace_high_irq(entry.trace_code, entry.tick_count,
+							entry.param0, entry.param1);
+
+			DRM_DEBUG_DRIVER("trace_code:%u, tick_count:%u, param0:%u, param1:%u\n",
+				 entry.trace_code, entry.tick_count, entry.param0, entry.param1);
+		} else
+			break;
+
+		count++;
+
+	} while (count <= DMUB_TRACE_MAX_READ);
+
+	if (count > DMUB_TRACE_MAX_READ)
+		DRM_DEBUG_DRIVER("Warning : count > DMUB_TRACE_MAX_READ");
 
 	if (dc_enable_dmub_notifications(adev->dm.dc) &&
 		irq_params->irq_src == DC_IRQ_SOURCE_DMCUB_OUTBOX) {
@@ -871,7 +915,8 @@ static void dm_dmub_outbox1_low_irq(void *interrupt_params)
 				continue;
 			}
 			if (!dm->dmub_callback[notify.type]) {
-				DRM_DEBUG_DRIVER("DMUB notification skipped, no handler: type=%d\n", notify.type);
+				DRM_WARN("DMUB notification skipped due to no handler: type=%s\n",
+					event_type[notify.type]);
 				continue;
 			}
 			if (dm->dmub_thread_offload[notify.type] == true) {
@@ -889,37 +934,12 @@ static void dm_dmub_outbox1_low_irq(void *interrupt_params)
 				}
 				INIT_WORK(&dmub_hpd_wrk->handle_hpd_work, dm_handle_hpd_work);
 				dmub_hpd_wrk->adev = adev;
-				if (notify.type == DMUB_NOTIFICATION_HPD) {
-					plink = adev->dm.dc->links[notify.link_index];
-					if (plink) {
-						plink->hpd_status =
-							notify.hpd_status == DP_HPD_PLUG;
-					}
-				}
 				queue_work(adev->dm.delayed_hpd_wq, &dmub_hpd_wrk->handle_hpd_work);
 			} else {
 				dm->dmub_callback[notify.type](adev, &notify);
 			}
 		} while (notify.pending_notification);
 	}
-
-
-	do {
-		if (dc_dmub_srv_get_dmub_outbox0_msg(dm->dc, &entry)) {
-			trace_amdgpu_dmub_trace_high_irq(entry.trace_code, entry.tick_count,
-							entry.param0, entry.param1);
-
-			DRM_DEBUG_DRIVER("trace_code:%u, tick_count:%u, param0:%u, param1:%u\n",
-				 entry.trace_code, entry.tick_count, entry.param0, entry.param1);
-		} else
-			break;
-
-		count++;
-
-	} while (count <= DMUB_TRACE_MAX_READ);
-
-	if (count > DMUB_TRACE_MAX_READ)
-		DRM_DEBUG_DRIVER("Warning : count > DMUB_TRACE_MAX_READ");
 }
 
 static int dm_set_clockgating_state(void *handle,
@@ -957,8 +977,8 @@ static void amdgpu_dm_fbc_init(struct drm_connector *connector)
 
 
 	list_for_each_entry(mode, &connector->modes, head) {
-		if (max_size < mode->htotal * mode->vtotal)
-			max_size = mode->htotal * mode->vtotal;
+		if (max_size < (unsigned long) mode->htotal * mode->vtotal)
+			max_size = (unsigned long) mode->htotal * mode->vtotal;
 	}
 
 	if (max_size) {
@@ -2855,7 +2875,8 @@ static int dm_suspend(void *handle)
 
 		dm->cached_dc_state = dc_state_create_copy(dm->dc->current_state);
 
-		dm_gpureset_toggle_interrupts(adev, dm->cached_dc_state, false);
+		if (dm->cached_dc_state)
+			dm_gpureset_toggle_interrupts(adev, dm->cached_dc_state, false);
 
 		amdgpu_dm_commit_zero_streams(dm->dc);
 
@@ -3159,7 +3180,7 @@ static int dm_resume(void *handle)
 		 * this is the case when traversing through already created end sink
 		 * MST connectors, should be skipped
 		 */
-		if (aconnector && aconnector->mst_root)
+		if (aconnector->mst_root)
 			continue;
 
 		mutex_lock(&aconnector->hpd_lock);
@@ -4571,6 +4592,7 @@ amdgpu_dm_register_backlight_device(struct amdgpu_dm_connector *aconnector)
 	struct drm_device *drm = aconnector->base.dev;
 	struct amdgpu_display_manager *dm = &drm_to_adev(drm)->dm;
 	struct backlight_properties props = { 0 };
+	struct amdgpu_dm_backlight_caps caps = { 0 };
 	char bl_name[16];
 
 	if (aconnector->bl_idx == -1)
@@ -4583,8 +4605,16 @@ amdgpu_dm_register_backlight_device(struct amdgpu_dm_connector *aconnector)
 		return;
 	}
 
+	amdgpu_acpi_get_backlight_caps(&caps);
+	if (caps.caps_valid) {
+		if (power_supply_is_system_supplied() > 0)
+			props.brightness = caps.ac_level;
+		else
+			props.brightness = caps.dc_level;
+	} else
+		props.brightness = AMDGPU_MAX_BL_LEVEL;
+
 	props.max_brightness = AMDGPU_MAX_BL_LEVEL;
-	props.brightness = AMDGPU_MAX_BL_LEVEL;
 	props.type = BACKLIGHT_RAW;
 
 	snprintf(bl_name, sizeof(bl_name), "amdgpu_bl%d",
@@ -6392,13 +6422,13 @@ static void apply_dsc_policy_for_stream(struct amdgpu_dm_connector *aconnector,
 	dc_dsc_policy_set_enable_dsc_when_not_needed(
 		aconnector->dsc_settings.dsc_force_enable == DSC_CLK_FORCE_ENABLE);
 
-	if (aconnector->dc_link && sink->sink_signal == SIGNAL_TYPE_EDP &&
+	if (sink->sink_signal == SIGNAL_TYPE_EDP &&
 	    !aconnector->dc_link->panel_config.dsc.disable_dsc_edp &&
 	    dc->caps.edp_dsc_support && aconnector->dsc_settings.dsc_force_enable != DSC_CLK_FORCE_DISABLE) {
 
 		apply_dsc_policy_for_edp(aconnector, sink, stream, dsc_caps, max_dsc_target_bpp_limit_override);
 
-	} else if (aconnector->dc_link && sink->sink_signal == SIGNAL_TYPE_DISPLAY_PORT) {
+	} else if (sink->sink_signal == SIGNAL_TYPE_DISPLAY_PORT) {
 		if (sink->link->dpcd_caps.dongle_type == DISPLAY_DONGLE_NONE) {
 			if (dc_dsc_compute_config(aconnector->dc_link->ctx->dc->res_pool->dscs[0],
 						dsc_caps,
@@ -7073,7 +7103,8 @@ static void create_eml_sink(struct amdgpu_dm_connector *aconnector)
 		aconnector->dc_sink = aconnector->dc_link->local_sink ?
 		aconnector->dc_link->local_sink :
 		aconnector->dc_em_sink;
-		dc_sink_retain(aconnector->dc_sink);
+		if (aconnector->dc_sink)
+			dc_sink_retain(aconnector->dc_sink);
 	}
 }
 
@@ -7900,7 +7931,8 @@ static int amdgpu_dm_connector_get_modes(struct drm_connector *connector)
 				drm_add_modes_noedid(connector, 1920, 1080);
 	} else {
 		amdgpu_dm_connector_ddc_get_modes(connector, edid);
-		amdgpu_dm_connector_add_common_modes(encoder, connector);
+		if (encoder)
+			amdgpu_dm_connector_add_common_modes(encoder, connector);
 		amdgpu_dm_connector_add_freesync_modes(connector, edid);
 	}
 	amdgpu_dm_fbc_init(connector);
@@ -8745,8 +8777,24 @@ static void amdgpu_dm_commit_planes(struct drm_atomic_state *state,
 	 * Disable the cursor first if we're disabling all the planes.
 	 * It'll remain on the screen after the planes are re-enabled
 	 * if we don't.
+	 *
+	 * If the cursor is transitioning from native to overlay mode, the
+	 * native cursor needs to be disabled first.
 	 */
-	if (acrtc_state->active_planes == 0)
+	if (acrtc_state->cursor_mode == DM_CURSOR_OVERLAY_MODE &&
+	    dm_old_crtc_state->cursor_mode == DM_CURSOR_NATIVE_MODE) {
+		struct dc_cursor_position cursor_position = {0};
+
+		if (!dc_stream_set_cursor_position(acrtc_state->stream,
+						   &cursor_position))
+			drm_err(dev, "DC failed to disable native cursor\n");
+
+		bundle->stream_update.cursor_position =
+				&acrtc_state->stream->cursor_position;
+	}
+
+	if (acrtc_state->active_planes == 0 &&
+	    dm_old_crtc_state->cursor_mode == DM_CURSOR_NATIVE_MODE)
 		amdgpu_dm_commit_cursors(state);
 
 	/* update planes when needed */
@@ -8760,7 +8808,8 @@ static void amdgpu_dm_commit_planes(struct drm_atomic_state *state,
 		struct dm_plane_state *dm_new_plane_state = to_dm_plane_state(new_plane_state);
 
 		/* Cursor plane is handled after stream updates */
-		if (plane->type == DRM_PLANE_TYPE_CURSOR) {
+		if (plane->type == DRM_PLANE_TYPE_CURSOR &&
+		    acrtc_state->cursor_mode == DM_CURSOR_NATIVE_MODE) {
 			if ((fb && crtc == pcrtc) ||
 			    (old_plane_state->fb && old_plane_state->crtc == pcrtc)) {
 				cursor_update = true;
@@ -9116,7 +9165,8 @@ static void amdgpu_dm_commit_planes(struct drm_atomic_state *state,
 	 * to be disabling a single plane - those pipes are being disabled.
 	 */
 	if (acrtc_state->active_planes &&
-	    (!updated_planes_and_streams || amdgpu_ip_version(dm->adev, DCE_HWIP, 0) == 0))
+	    (!updated_planes_and_streams || amdgpu_ip_version(dm->adev, DCE_HWIP, 0) == 0) &&
+	    acrtc_state->cursor_mode == DM_CURSOR_NATIVE_MODE)
 		amdgpu_dm_commit_cursors(state);
 
 cleanup:
@@ -9744,6 +9794,8 @@ static void amdgpu_dm_atomic_commit_tail(struct drm_atomic_state *state)
 		for (j = 0; j < status->plane_count; j++)
 			dummy_updates[j].surface = status->plane_states[0];
 
+		sort(dummy_updates, status->plane_count,
+		     sizeof(*dummy_updates), dm_plane_layer_index_cmp, NULL);
 
 		mutex_lock(&dm->dc_lock);
 		dc_exit_ips_for_hw_access(dm->dc);
@@ -10434,7 +10486,8 @@ static bool should_reset_plane(struct drm_atomic_state *state,
 {
 	struct drm_plane *other;
 	struct drm_plane_state *old_other_state, *new_other_state;
-	struct drm_crtc_state *new_crtc_state;
+	struct drm_crtc_state *old_crtc_state, *new_crtc_state;
+	struct dm_crtc_state *old_dm_crtc_state, *new_dm_crtc_state;
 	struct amdgpu_device *adev = drm_to_adev(plane->dev);
 	int i;
 
@@ -10456,14 +10509,38 @@ static bool should_reset_plane(struct drm_atomic_state *state,
 
 	new_crtc_state =
 		drm_atomic_get_new_crtc_state(state, new_plane_state->crtc);
+	old_crtc_state =
+		drm_atomic_get_old_crtc_state(state, old_plane_state->crtc);
 
 	if (!new_crtc_state)
 		return true;
 
+	/*
+	 * A change in cursor mode means a new dc pipe needs to be acquired or
+	 * released from the state
+	 */
+	old_dm_crtc_state = to_dm_crtc_state(old_crtc_state);
+	new_dm_crtc_state = to_dm_crtc_state(new_crtc_state);
+	if (plane->type == DRM_PLANE_TYPE_CURSOR &&
+	    old_dm_crtc_state != NULL &&
+	    old_dm_crtc_state->cursor_mode != new_dm_crtc_state->cursor_mode) {
+		return true;
+	}
+
 	/* CRTC Degamma changes currently require us to recreate planes. */
 	if (new_crtc_state->color_mgmt_changed)
 		return true;
 
+	/*
+	 * On zpos change, planes need to be reordered by removing and re-adding
+	 * them one by one to the dc state, in order of descending zpos.
+	 *
+	 * TODO: We can likely skip bandwidth validation if the only thing that
+	 * changed about the plane was it'z z-ordering.
+	 */
+	if (new_crtc_state->zpos_changed)
+		return true;
+
 	if (drm_atomic_crtc_needs_modeset(new_crtc_state))
 		return true;
 
@@ -10611,6 +10688,68 @@ static int dm_check_cursor_fb(struct amdgpu_crtc *new_acrtc,
 	return 0;
 }
 
+/*
+ * Helper function for checking the cursor in native mode
+ */
+static int dm_check_native_cursor_state(struct drm_crtc *new_plane_crtc,
+					struct drm_plane *plane,
+					struct drm_plane_state *new_plane_state,
+					bool enable)
+{
+
+	struct amdgpu_crtc *new_acrtc;
+	int ret;
+
+	if (!enable || !new_plane_crtc ||
+	    drm_atomic_plane_disabling(plane->state, new_plane_state))
+		return 0;
+
+	new_acrtc = to_amdgpu_crtc(new_plane_crtc);
+
+	if (new_plane_state->src_x != 0 || new_plane_state->src_y != 0) {
+		DRM_DEBUG_ATOMIC("Cropping not supported for cursor plane\n");
+		return -EINVAL;
+	}
+
+	if (new_plane_state->fb) {
+		ret = dm_check_cursor_fb(new_acrtc, new_plane_state,
+						new_plane_state->fb);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+static bool dm_should_update_native_cursor(struct drm_atomic_state *state,
+					   struct drm_crtc *old_plane_crtc,
+					   struct drm_crtc *new_plane_crtc,
+					   bool enable)
+{
+	struct drm_crtc_state *old_crtc_state, *new_crtc_state;
+	struct dm_crtc_state *dm_old_crtc_state, *dm_new_crtc_state;
+
+	if (!enable) {
+		if (old_plane_crtc == NULL)
+			return true;
+
+		old_crtc_state = drm_atomic_get_old_crtc_state(
+			state, old_plane_crtc);
+		dm_old_crtc_state = to_dm_crtc_state(old_crtc_state);
+
+		return dm_old_crtc_state->cursor_mode == DM_CURSOR_NATIVE_MODE;
+	} else {
+		if (new_plane_crtc == NULL)
+			return true;
+
+		new_crtc_state = drm_atomic_get_new_crtc_state(
+			state, new_plane_crtc);
+		dm_new_crtc_state = to_dm_crtc_state(new_crtc_state);
+
+		return dm_new_crtc_state->cursor_mode == DM_CURSOR_NATIVE_MODE;
+	}
+}
+
 static int dm_update_plane_state(struct dc *dc,
 				 struct drm_atomic_state *state,
 				 struct drm_plane *plane,
@@ -10626,8 +10765,7 @@ static int dm_update_plane_state(struct dc *dc,
 	struct drm_crtc_state *old_crtc_state, *new_crtc_state;
 	struct dm_crtc_state *dm_new_crtc_state, *dm_old_crtc_state;
 	struct dm_plane_state *dm_new_plane_state, *dm_old_plane_state;
-	struct amdgpu_crtc *new_acrtc;
-	bool needs_reset;
+	bool needs_reset, update_native_cursor;
 	int ret = 0;
 
 
@@ -10636,24 +10774,16 @@ static int dm_update_plane_state(struct dc *dc,
 	dm_new_plane_state = to_dm_plane_state(new_plane_state);
 	dm_old_plane_state = to_dm_plane_state(old_plane_state);
 
-	if (plane->type == DRM_PLANE_TYPE_CURSOR) {
-		if (!enable || !new_plane_crtc ||
-			drm_atomic_plane_disabling(plane->state, new_plane_state))
-			return 0;
-
-		new_acrtc = to_amdgpu_crtc(new_plane_crtc);
-
-		if (new_plane_state->src_x != 0 || new_plane_state->src_y != 0) {
-			DRM_DEBUG_ATOMIC("Cropping not supported for cursor plane\n");
-			return -EINVAL;
-		}
+	update_native_cursor = dm_should_update_native_cursor(state,
+							      old_plane_crtc,
+							      new_plane_crtc,
+							      enable);
 
-		if (new_plane_state->fb) {
-			ret = dm_check_cursor_fb(new_acrtc, new_plane_state,
-						 new_plane_state->fb);
-			if (ret)
-				return ret;
-		}
+	if (plane->type == DRM_PLANE_TYPE_CURSOR && update_native_cursor) {
+		ret = dm_check_native_cursor_state(new_plane_crtc, plane,
+						    new_plane_state, enable);
+		if (ret)
+			return ret;
 
 		return 0;
 	}
@@ -10719,20 +10849,14 @@ static int dm_update_plane_state(struct dc *dc,
 
 		ret = amdgpu_dm_plane_helper_check_state(new_plane_state, new_crtc_state);
 		if (ret)
-			return ret;
+			goto out;
 
 		WARN_ON(dm_new_plane_state->dc_state);
 
 		dc_new_plane_state = dc_create_plane_state(dc);
-		if (!dc_new_plane_state)
-			return -ENOMEM;
-
-		/* Block top most plane from being a video plane */
-		if (plane->type == DRM_PLANE_TYPE_OVERLAY) {
-			if (amdgpu_dm_plane_is_video_format(new_plane_state->fb->format->format) && *is_top_most_overlay)
-				return -EINVAL;
-
-			*is_top_most_overlay = false;
+		if (!dc_new_plane_state) {
+			ret = -ENOMEM;
+			goto out;
 		}
 
 		DRM_DEBUG_ATOMIC("Enabling DRM plane: %d on DRM crtc %d\n",
@@ -10745,13 +10869,13 @@ static int dm_update_plane_state(struct dc *dc,
 			new_crtc_state);
 		if (ret) {
 			dc_plane_state_release(dc_new_plane_state);
-			return ret;
+			goto out;
 		}
 
 		ret = dm_atomic_get_state(state, &dm_state);
 		if (ret) {
 			dc_plane_state_release(dc_new_plane_state);
-			return ret;
+			goto out;
 		}
 
 		/*
@@ -10768,7 +10892,8 @@ static int dm_update_plane_state(struct dc *dc,
 				dm_state->context)) {
 
 			dc_plane_state_release(dc_new_plane_state);
-			return -EINVAL;
+			ret = -EINVAL;
+			goto out;
 		}
 
 		dm_new_plane_state->dc_state = dc_new_plane_state;
@@ -10783,6 +10908,16 @@ static int dm_update_plane_state(struct dc *dc,
 		*lock_and_validation_needed = true;
 	}
 
+out:
+	/* If enabling cursor overlay failed, attempt fallback to native mode */
+	if (enable && ret == -EINVAL && plane->type == DRM_PLANE_TYPE_CURSOR) {
+		ret = dm_check_native_cursor_state(new_plane_crtc, plane,
+						    new_plane_state, enable);
+		if (ret)
+			return ret;
+
+		dm_new_crtc_state->cursor_mode = DM_CURSOR_NATIVE_MODE;
+	}
 
 	return ret;
 }
@@ -10816,99 +10951,64 @@ dm_get_plane_scale(struct drm_plane_state *plane_state,
 	*out_plane_scale_h = plane_state->crtc_h * 1000 / plane_src_h;
 }
 
-static int dm_check_crtc_cursor(struct drm_atomic_state *state,
-				struct drm_crtc *crtc,
-				struct drm_crtc_state *new_crtc_state)
+/*
+ * The normalized_zpos value cannot be used by this iterator directly. It's only
+ * calculated for enabled planes, potentially causing normalized_zpos collisions
+ * between enabled/disabled planes in the atomic state. We need a unique value
+ * so that the iterator will not generate the same object twice, or loop
+ * indefinitely.
+ */
+static inline struct __drm_planes_state *__get_next_zpos(
+	struct drm_atomic_state *state,
+	struct __drm_planes_state *prev)
 {
-	struct drm_plane *cursor = crtc->cursor, *plane, *underlying;
-	struct drm_plane_state *old_plane_state, *new_plane_state;
-	struct drm_plane_state *new_cursor_state, *new_underlying_state;
-	int i;
-	int cursor_scale_w, cursor_scale_h, underlying_scale_w, underlying_scale_h;
-	bool any_relevant_change = false;
-
-	/* On DCE and DCN there is no dedicated hardware cursor plane. We get a
-	 * cursor per pipe but it's going to inherit the scaling and
-	 * positioning from the underlying pipe. Check the cursor plane's
-	 * blending properties match the underlying planes'.
-	 */
-
-	/* If no plane was enabled or changed scaling, no need to check again */
-	for_each_oldnew_plane_in_state(state, plane, old_plane_state, new_plane_state, i) {
-		int new_scale_w, new_scale_h, old_scale_w, old_scale_h;
-
-		if (!new_plane_state || !new_plane_state->fb || new_plane_state->crtc != crtc)
-			continue;
-
-		if (!old_plane_state || !old_plane_state->fb || old_plane_state->crtc != crtc) {
-			any_relevant_change = true;
-			break;
-		}
-
-		if (new_plane_state->fb == old_plane_state->fb &&
-		    new_plane_state->crtc_w == old_plane_state->crtc_w &&
-		    new_plane_state->crtc_h == old_plane_state->crtc_h)
-			continue;
-
-		dm_get_plane_scale(new_plane_state, &new_scale_w, &new_scale_h);
-		dm_get_plane_scale(old_plane_state, &old_scale_w, &old_scale_h);
+	unsigned int highest_zpos = 0, prev_zpos = 256;
+	uint32_t highest_id = 0, prev_id = UINT_MAX;
+	struct drm_plane_state *new_plane_state;
+	struct drm_plane *plane;
+	int i, highest_i = -1;
 
-		if (new_scale_w != old_scale_w || new_scale_h != old_scale_h) {
-			any_relevant_change = true;
-			break;
-		}
+	if (prev != NULL) {
+		prev_zpos = prev->new_state->zpos;
+		prev_id = prev->ptr->base.id;
 	}
 
-	if (!any_relevant_change)
-		return 0;
-
-	new_cursor_state = drm_atomic_get_plane_state(state, cursor);
-	if (IS_ERR(new_cursor_state))
-		return PTR_ERR(new_cursor_state);
-
-	if (!new_cursor_state->fb)
-		return 0;
-
-	dm_get_plane_scale(new_cursor_state, &cursor_scale_w, &cursor_scale_h);
-
-	/* Need to check all enabled planes, even if this commit doesn't change
-	 * their state
-	 */
-	i = drm_atomic_add_affected_planes(state, crtc);
-	if (i)
-		return i;
-
-	for_each_new_plane_in_state_reverse(state, underlying, new_underlying_state, i) {
-		/* Narrow down to non-cursor planes on the same CRTC as the cursor */
-		if (new_underlying_state->crtc != crtc || underlying == crtc->cursor)
+	for_each_new_plane_in_state(state, plane, new_plane_state, i) {
+		/* Skip planes with higher zpos than the previously returned */
+		if (new_plane_state->zpos > prev_zpos ||
+		    (new_plane_state->zpos == prev_zpos &&
+		     plane->base.id >= prev_id))
 			continue;
 
-		/* Ignore disabled planes */
-		if (!new_underlying_state->fb)
-			continue;
-
-		dm_get_plane_scale(new_underlying_state,
-				   &underlying_scale_w, &underlying_scale_h);
-
-		if (cursor_scale_w != underlying_scale_w ||
-		    cursor_scale_h != underlying_scale_h) {
-			drm_dbg_atomic(crtc->dev,
-				       "Cursor [PLANE:%d:%s] scaling doesn't match underlying [PLANE:%d:%s]\n",
-				       cursor->base.id, cursor->name, underlying->base.id, underlying->name);
-			return -EINVAL;
+		/* Save the index of the plane with highest zpos */
+		if (new_plane_state->zpos > highest_zpos ||
+		    (new_plane_state->zpos == highest_zpos &&
+		     plane->base.id > highest_id)) {
+			highest_zpos = new_plane_state->zpos;
+			highest_id = plane->base.id;
+			highest_i = i;
 		}
-
-		/* If this plane covers the whole CRTC, no need to check planes underneath */
-		if (new_underlying_state->crtc_x <= 0 &&
-		    new_underlying_state->crtc_y <= 0 &&
-		    new_underlying_state->crtc_x + new_underlying_state->crtc_w >= new_crtc_state->mode.hdisplay &&
-		    new_underlying_state->crtc_y + new_underlying_state->crtc_h >= new_crtc_state->mode.vdisplay)
-			break;
 	}
 
-	return 0;
+	if (highest_i < 0)
+		return NULL;
+
+	return &state->planes[highest_i];
 }
 
+/*
+ * Use the uniqueness of the plane's (zpos, drm obj ID) combination to iterate
+ * by descending zpos, as read from the new plane state. This is the same
+ * ordering as defined by drm_atomic_normalize_zpos().
+ */
+#define for_each_oldnew_plane_in_descending_zpos(__state, plane, old_plane_state, new_plane_state) \
+	for (struct __drm_planes_state *__i = __get_next_zpos((__state), NULL); \
+	     __i != NULL; __i = __get_next_zpos((__state), __i))		\
+		for_each_if(((plane) = __i->ptr,				\
+			     (void)(plane) /* Only to avoid unused-but-set-variable warning */, \
+			     (old_plane_state) = __i->old_state,		\
+			     (new_plane_state) = __i->new_state, 1))
+
 static int add_affected_mst_dsc_crtcs(struct drm_atomic_state *state, struct drm_crtc *crtc)
 {
 	struct drm_connector *connector;
@@ -10940,6 +11040,165 @@ static int add_affected_mst_dsc_crtcs(struct drm_atomic_state *state, struct drm
 }
 
 /**
+ * DOC: Cursor Modes - Native vs Overlay
+ *
+ * In native mode, the cursor uses a integrated cursor pipe within each DCN hw
+ * plane. It does not require a dedicated hw plane to enable, but it is
+ * subjected to the same z-order and scaling as the hw plane. It also has format
+ * restrictions, a RGB cursor in native mode cannot be enabled within a non-RGB
+ * hw plane.
+ *
+ * In overlay mode, the cursor uses a separate DCN hw plane, and thus has its
+ * own scaling and z-pos. It also has no blending restrictions. It lends to a
+ * cursor behavior more akin to a DRM client's expectations. However, it does
+ * occupy an extra DCN plane, and therefore will only be used if a DCN plane is
+ * available.
+ */
+
+/**
+ * dm_crtc_get_cursor_mode() - Determine the required cursor mode on crtc
+ * @adev: amdgpu device
+ * @state: DRM atomic state
+ * @dm_crtc_state: amdgpu state for the CRTC containing the cursor
+ * @cursor_mode: Returns the required cursor mode on dm_crtc_state
+ *
+ * Get whether the cursor should be enabled in native mode, or overlay mode, on
+ * the dm_crtc_state.
+ *
+ * The cursor should be enabled in overlay mode if there exists an underlying
+ * plane - on which the cursor may be blended - that is either YUV formatted, or
+ * scaled differently from the cursor.
+ *
+ * Since zpos info is required, drm_atomic_normalize_zpos must be called before
+ * calling this function.
+ *
+ * Return: 0 on success, or an error code if getting the cursor plane state
+ * failed.
+ */
+static int dm_crtc_get_cursor_mode(struct amdgpu_device *adev,
+				   struct drm_atomic_state *state,
+				   struct dm_crtc_state *dm_crtc_state,
+				   enum amdgpu_dm_cursor_mode *cursor_mode)
+{
+	struct drm_plane_state *old_plane_state, *plane_state, *cursor_state;
+	struct drm_crtc_state *crtc_state = &dm_crtc_state->base;
+	struct drm_plane *plane;
+	bool consider_mode_change = false;
+	bool entire_crtc_covered = false;
+	bool cursor_changed = false;
+	int underlying_scale_w, underlying_scale_h;
+	int cursor_scale_w, cursor_scale_h;
+	int i;
+
+	/* Overlay cursor not supported on HW before DCN */
+	if (amdgpu_ip_version(adev, DCE_HWIP, 0) == 0) {
+		*cursor_mode = DM_CURSOR_NATIVE_MODE;
+		return 0;
+	}
+
+	/* Init cursor_mode to be the same as current */
+	*cursor_mode = dm_crtc_state->cursor_mode;
+
+	/*
+	 * Cursor mode can change if a plane's format changes, scale changes, is
+	 * enabled/disabled, or z-order changes.
+	 */
+	for_each_oldnew_plane_in_state(state, plane, old_plane_state, plane_state, i) {
+		int new_scale_w, new_scale_h, old_scale_w, old_scale_h;
+
+		/* Only care about planes on this CRTC */
+		if ((drm_plane_mask(plane) & crtc_state->plane_mask) == 0)
+			continue;
+
+		if (plane->type == DRM_PLANE_TYPE_CURSOR)
+			cursor_changed = true;
+
+		if (drm_atomic_plane_enabling(old_plane_state, plane_state) ||
+		    drm_atomic_plane_disabling(old_plane_state, plane_state) ||
+		    old_plane_state->fb->format != plane_state->fb->format) {
+			consider_mode_change = true;
+			break;
+		}
+
+		dm_get_plane_scale(plane_state, &new_scale_w, &new_scale_h);
+		dm_get_plane_scale(old_plane_state, &old_scale_w, &old_scale_h);
+		if (new_scale_w != old_scale_w || new_scale_h != old_scale_h) {
+			consider_mode_change = true;
+			break;
+		}
+	}
+
+	if (!consider_mode_change && !crtc_state->zpos_changed)
+		return 0;
+
+	/*
+	 * If no cursor change on this CRTC, and not enabled on this CRTC, then
+	 * no need to set cursor mode. This avoids needlessly locking the cursor
+	 * state.
+	 */
+	if (!cursor_changed &&
+	    !(drm_plane_mask(crtc_state->crtc->cursor) & crtc_state->plane_mask)) {
+		return 0;
+	}
+
+	cursor_state = drm_atomic_get_plane_state(state,
+						  crtc_state->crtc->cursor);
+	if (IS_ERR(cursor_state))
+		return PTR_ERR(cursor_state);
+
+	/* Cursor is disabled */
+	if (!cursor_state->fb)
+		return 0;
+
+	/* For all planes in descending z-order (all of which are below cursor
+	 * as per zpos definitions), check their scaling and format
+	 */
+	for_each_oldnew_plane_in_descending_zpos(state, plane, old_plane_state, plane_state) {
+
+		/* Only care about non-cursor planes on this CRTC */
+		if ((drm_plane_mask(plane) & crtc_state->plane_mask) == 0 ||
+		    plane->type == DRM_PLANE_TYPE_CURSOR)
+			continue;
+
+		/* Underlying plane is YUV format - use overlay cursor */
+		if (amdgpu_dm_plane_is_video_format(plane_state->fb->format->format)) {
+			*cursor_mode = DM_CURSOR_OVERLAY_MODE;
+			return 0;
+		}
+
+		dm_get_plane_scale(plane_state,
+				   &underlying_scale_w, &underlying_scale_h);
+		dm_get_plane_scale(cursor_state,
+				   &cursor_scale_w, &cursor_scale_h);
+
+		/* Underlying plane has different scale - use overlay cursor */
+		if (cursor_scale_w != underlying_scale_w &&
+		    cursor_scale_h != underlying_scale_h) {
+			*cursor_mode = DM_CURSOR_OVERLAY_MODE;
+			return 0;
+		}
+
+		/* If this plane covers the whole CRTC, no need to check planes underneath */
+		if (plane_state->crtc_x <= 0 && plane_state->crtc_y <= 0 &&
+		    plane_state->crtc_x + plane_state->crtc_w >= crtc_state->mode.hdisplay &&
+		    plane_state->crtc_y + plane_state->crtc_h >= crtc_state->mode.vdisplay) {
+			entire_crtc_covered = true;
+			break;
+		}
+	}
+
+	/* If planes do not cover the entire CRTC, use overlay mode to enable
+	 * cursor over holes
+	 */
+	if (entire_crtc_covered)
+		*cursor_mode = DM_CURSOR_NATIVE_MODE;
+	else
+		*cursor_mode = DM_CURSOR_OVERLAY_MODE;
+
+	return 0;
+}
+
+/**
  * amdgpu_dm_atomic_check() - Atomic check implementation for AMDgpu DM.
  *
  * @dev: The DRM device
@@ -11108,8 +11367,23 @@ static int amdgpu_dm_atomic_check(struct drm_device *dev,
 		goto fail;
 	}
 
+	/*
+	 * Determine whether cursors on each CRTC should be enabled in native or
+	 * overlay mode.
+	 */
+	for_each_new_crtc_in_state(state, crtc, new_crtc_state, i) {
+		dm_new_crtc_state = to_dm_crtc_state(new_crtc_state);
+
+		ret = dm_crtc_get_cursor_mode(adev, state, dm_new_crtc_state,
+					      &dm_new_crtc_state->cursor_mode);
+		if (ret) {
+			drm_dbg(dev, "Failed to determine cursor mode\n");
+			goto fail;
+		}
+	}
+
 	/* Remove exiting planes if they are modified */
-	for_each_oldnew_plane_in_state_reverse(state, plane, old_plane_state, new_plane_state, i) {
+	for_each_oldnew_plane_in_descending_zpos(state, plane, old_plane_state, new_plane_state) {
 		if (old_plane_state->fb && new_plane_state->fb &&
 		    get_mem_type(old_plane_state->fb) !=
 		    get_mem_type(new_plane_state->fb))
@@ -11154,7 +11428,7 @@ static int amdgpu_dm_atomic_check(struct drm_device *dev,
 	}
 
 	/* Add new/modified planes */
-	for_each_oldnew_plane_in_state_reverse(state, plane, old_plane_state, new_plane_state, i) {
+	for_each_oldnew_plane_in_descending_zpos(state, plane, old_plane_state, new_plane_state) {
 		ret = dm_update_plane_state(dc, state, plane,
 					    old_plane_state,
 					    new_plane_state,
@@ -11188,11 +11462,29 @@ static int amdgpu_dm_atomic_check(struct drm_device *dev,
 			drm_dbg_atomic(dev, "MPO enablement requested on crtc:[%p]\n", crtc);
 	}
 
-	/* Check cursor planes scaling */
+	/* Check cursor planes restrictions */
 	for_each_new_crtc_in_state(state, crtc, new_crtc_state, i) {
-		ret = dm_check_crtc_cursor(state, crtc, new_crtc_state);
+		enum amdgpu_dm_cursor_mode required_cursor_mode;
+
+		/* Overlay cusor not subject to native cursor restrictions */
+		dm_new_crtc_state = to_dm_crtc_state(new_crtc_state);
+		if (dm_new_crtc_state->cursor_mode == DM_CURSOR_OVERLAY_MODE)
+			continue;
+
+		/* If HW can only do native cursor, check restrictions again */
+		ret = dm_crtc_get_cursor_mode(adev, state, dm_new_crtc_state,
+					      &required_cursor_mode);
+
 		if (ret) {
-			drm_dbg_atomic(dev, "dm_check_crtc_cursor() failed\n");
+			drm_dbg_driver(crtc->dev,
+				       "[CRTC:%d:%s] Checking cursor mode failed\n",
+				       crtc->base.id, crtc->name);
+			goto fail;
+		} else if (required_cursor_mode == DM_CURSOR_OVERLAY_MODE) {
+			drm_dbg_driver(crtc->dev,
+				       "[CRTC:%d:%s] Cannot enable native cursor due to scaling or YUV restrictions\n",
+				       crtc->base.id, crtc->name);
+			ret = -EINVAL;
 			goto fail;
 		}
 	}
@@ -11806,6 +12098,12 @@ void amdgpu_dm_trigger_timing_sync(struct drm_device *dev)
 	mutex_unlock(&adev->dm.dc_lock);
 }
 
+static inline void amdgpu_dm_exit_ips_for_hw_access(struct dc *dc)
+{
+	if (dc->ctx->dmub_srv && !dc->ctx->dmub_srv->idle_exit_counter)
+		dc_exit_ips_for_hw_access(dc);
+}
+
 void dm_write_reg_func(const struct dc_context *ctx, uint32_t address,
 		       u32 value, const char *func_name)
 {
@@ -11816,6 +12114,8 @@ void dm_write_reg_func(const struct dc_context *ctx, uint32_t address,
 		return;
 	}
 #endif
+
+	amdgpu_dm_exit_ips_for_hw_access(ctx->dc);
 	cgs_write_register(ctx->cgs_device, address, value);
 	trace_amdgpu_dc_wreg(&ctx->perf_trace->write_count, address, value);
 }
@@ -11839,6 +12139,8 @@ uint32_t dm_read_reg_func(const struct dc_context *ctx, uint32_t address,
 		return 0;
 	}
 
+	amdgpu_dm_exit_ips_for_hw_access(ctx->dc);
+
 	value = cgs_read_register(ctx->cgs_device, address);
 
 	trace_amdgpu_dc_rreg(&ctx->perf_trace->read_count, address, value);