Merge tag 'drm-misc-fixes-2023-07-13' of ssh://git.freedesktop.org/git/drm/drm-misc into drm-fixes

A couple of nouveau patches addressing improving HDMI support and
firmware handling, a fix for TTM to skip pinned BO when evicting, and a
fix for the fbdev documentation.
Signed-off-by: Dave Airlie <airlied@redhat.com>

From: Maxime Ripard <mripard@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/nq3ke75juephbex5acfyi5t6bxv22nhmfcpfhru55haj2nv3us@gehrlmjbqgjk

drivers/accel/ivpu/ivpu_drv.h

@@ -75,6 +75,7 @@ struct ivpu_wa_table {
 	bool punit_disabled;
 	bool clear_runtime_mem;
 	bool d3hot_after_power_off;
+	bool interrupt_clear_with_0;
 };
 
 struct ivpu_hw_info;

drivers/accel/ivpu/ivpu_hw_mtl.c

@@ -101,6 +101,9 @@ static void ivpu_hw_wa_init(struct ivpu_device *vdev)
 	vdev->wa.punit_disabled = ivpu_is_fpga(vdev);
 	vdev->wa.clear_runtime_mem = false;
 	vdev->wa.d3hot_after_power_off = true;
+
+	if (ivpu_device_id(vdev) == PCI_DEVICE_ID_MTL && ivpu_revision(vdev) < 4)
+		vdev->wa.interrupt_clear_with_0 = true;
 }
 
 static void ivpu_hw_timeouts_init(struct ivpu_device *vdev)
@@ -885,7 +888,7 @@ static void ivpu_hw_mtl_irq_disable(struct ivpu_device *vdev)
 	REGB_WR32(MTL_BUTTRESS_GLOBAL_INT_MASK, 0x1);
 	REGB_WR32(MTL_BUTTRESS_LOCAL_INT_MASK, BUTTRESS_IRQ_DISABLE_MASK);
 	REGV_WR64(MTL_VPU_HOST_SS_ICB_ENABLE_0, 0x0ull);
-	REGB_WR32(MTL_VPU_HOST_SS_FW_SOC_IRQ_EN, 0x0);
+	REGV_WR32(MTL_VPU_HOST_SS_FW_SOC_IRQ_EN, 0x0);
 }
 
 static void ivpu_hw_mtl_irq_wdt_nce_handler(struct ivpu_device *vdev)
@@ -973,12 +976,15 @@ static u32 ivpu_hw_mtl_irqb_handler(struct ivpu_device *vdev, int irq)
 		schedule_recovery = true;
 	}
 
-	/*
-	 * Clear local interrupt status by writing 0 to all bits.
-	 * This must be done after interrupts are cleared at the source.
-	 * Writing 1 triggers an interrupt, so we can't perform read update write.
-	 */
-	REGB_WR32(MTL_BUTTRESS_INTERRUPT_STAT, 0x0);
+	/* This must be done after interrupts are cleared at the source. */
+	if (IVPU_WA(interrupt_clear_with_0))
+		/*
+		 * Writing 1 triggers an interrupt, so we can't perform read update write.
+		 * Clear local interrupt status by writing 0 to all bits.
+		 */
+		REGB_WR32(MTL_BUTTRESS_INTERRUPT_STAT, 0x0);
+	else
+		REGB_WR32(MTL_BUTTRESS_INTERRUPT_STAT, status);
 
 	/* Re-enable global interrupt */
 	REGB_WR32(MTL_BUTTRESS_GLOBAL_INT_MASK, 0x0);

drivers/dma-buf/dma-fence-unwrap.c

@@ -66,18 +66,36 @@ struct dma_fence *__dma_fence_unwrap_merge(unsigned int num_fences,
 {
 	struct dma_fence_array *result;
 	struct dma_fence *tmp, **array;
+	ktime_t timestamp;
 	unsigned int i;
 	size_t count;
 
 	count = 0;
+	timestamp = ns_to_ktime(0);
 	for (i = 0; i < num_fences; ++i) {
-		dma_fence_unwrap_for_each(tmp, &iter[i], fences[i])
-			if (!dma_fence_is_signaled(tmp))
+		dma_fence_unwrap_for_each(tmp, &iter[i], fences[i]) {
+			if (!dma_fence_is_signaled(tmp)) {
 				++count;
+			} else if (test_bit(DMA_FENCE_FLAG_TIMESTAMP_BIT,
+					    &tmp->flags)) {
+				if (ktime_after(tmp->timestamp, timestamp))
+					timestamp = tmp->timestamp;
+			} else {
+				/*
+				 * Use the current time if the fence is
+				 * currently signaling.
+				 */
+				timestamp = ktime_get();
+			}
+		}
 	}
 
+	/*
+	 * If we couldn't find a pending fence just return a private signaled
+	 * fence with the timestamp of the last signaled one.
+	 */
 	if (count == 0)
-		return dma_fence_get_stub();
+		return dma_fence_allocate_private_stub(timestamp);
 
 	array = kmalloc_array(count, sizeof(*array), GFP_KERNEL);
 	if (!array)
@@ -138,7 +156,7 @@ struct dma_fence *__dma_fence_unwrap_merge(unsigned int num_fences,
 	} while (tmp);
 
 	if (count == 0) {
-		tmp = dma_fence_get_stub();
+		tmp = dma_fence_allocate_private_stub(ktime_get());
 		goto return_tmp;
 	}
 

drivers/dma-buf/dma-fence.c

@@ -150,16 +150,17 @@ EXPORT_SYMBOL(dma_fence_get_stub);
 
 /**
  * dma_fence_allocate_private_stub - return a private, signaled fence
+ * @timestamp: timestamp when the fence was signaled
  *
  * Return a newly allocated and signaled stub fence.
  */
-struct dma_fence *dma_fence_allocate_private_stub(void)
+struct dma_fence *dma_fence_allocate_private_stub(ktime_t timestamp)
 {
 	struct dma_fence *fence;
 
 	fence = kzalloc(sizeof(*fence), GFP_KERNEL);
 	if (fence == NULL)
-		return ERR_PTR(-ENOMEM);
+		return NULL;
 
 	dma_fence_init(fence,
 		       &dma_fence_stub_ops,
@@ -169,7 +170,7 @@ struct dma_fence *dma_fence_allocate_private_stub(void)
 	set_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
 		&fence->flags);
 
-	dma_fence_signal(fence);
+	dma_fence_signal_timestamp(fence, timestamp);
 
 	return fence;
 }

drivers/gpu/drm/armada/armada_fbdev.c

@@ -209,10 +209,6 @@ void armada_fbdev_setup(struct drm_device *dev)
 		goto err_drm_client_init;
 	}
 
-	ret = armada_fbdev_client_hotplug(&fbh->client);
-	if (ret)
-		drm_dbg_kms(dev, "client hotplug ret=%d\n", ret);
-
 	drm_client_register(&fbh->client);
 
 	return;

drivers/gpu/drm/bridge/synopsys/dw-hdmi.c

@@ -1426,9 +1426,9 @@ void dw_hdmi_set_high_tmds_clock_ratio(struct dw_hdmi *hdmi,
 	/* Control for TMDS Bit Period/TMDS Clock-Period Ratio */
 	if (dw_hdmi_support_scdc(hdmi, display)) {
 		if (mtmdsclock > HDMI14_MAX_TMDSCLK)
-			drm_scdc_set_high_tmds_clock_ratio(&hdmi->connector, 1);
+			drm_scdc_set_high_tmds_clock_ratio(hdmi->curr_conn, 1);
 		else
-			drm_scdc_set_high_tmds_clock_ratio(&hdmi->connector, 0);
+			drm_scdc_set_high_tmds_clock_ratio(hdmi->curr_conn, 0);
 	}
 }
 EXPORT_SYMBOL_GPL(dw_hdmi_set_high_tmds_clock_ratio);
@@ -2116,7 +2116,7 @@ static void hdmi_av_composer(struct dw_hdmi *hdmi,
 				min_t(u8, bytes, SCDC_MIN_SOURCE_VERSION));
 
 			/* Enabled Scrambling in the Sink */
-			drm_scdc_set_scrambling(&hdmi->connector, 1);
+			drm_scdc_set_scrambling(hdmi->curr_conn, 1);
 
 			/*
 			 * To activate the scrambler feature, you must ensure
@@ -2132,7 +2132,7 @@ static void hdmi_av_composer(struct dw_hdmi *hdmi,
 			hdmi_writeb(hdmi, 0, HDMI_FC_SCRAMBLER_CTRL);
 			hdmi_writeb(hdmi, (u8)~HDMI_MC_SWRSTZ_TMDSSWRST_REQ,
 				    HDMI_MC_SWRSTZ);
-			drm_scdc_set_scrambling(&hdmi->connector, 0);
+			drm_scdc_set_scrambling(hdmi->curr_conn, 0);
 		}
 	}
 
@@ -3553,6 +3553,7 @@ struct dw_hdmi *dw_hdmi_probe(struct platform_device *pdev,
 	hdmi->bridge.ops = DRM_BRIDGE_OP_DETECT | DRM_BRIDGE_OP_EDID
 			 | DRM_BRIDGE_OP_HPD;
 	hdmi->bridge.interlace_allowed = true;
+	hdmi->bridge.ddc = hdmi->ddc;
 #ifdef CONFIG_OF
 	hdmi->bridge.of_node = pdev->dev.of_node;
 #endif

drivers/gpu/drm/bridge/ti-sn65dsi86.c

@@ -170,10 +170,10 @@
  * @pwm_refclk_freq: Cache for the reference clock input to the PWM.
  */
 struct ti_sn65dsi86 {
-	struct auxiliary_device		bridge_aux;
-	struct auxiliary_device		gpio_aux;
-	struct auxiliary_device		aux_aux;
-	struct auxiliary_device		pwm_aux;
+	struct auxiliary_device		*bridge_aux;
+	struct auxiliary_device		*gpio_aux;
+	struct auxiliary_device		*aux_aux;
+	struct auxiliary_device		*pwm_aux;
 
 	struct device			*dev;
 	struct regmap			*regmap;
@@ -468,27 +468,34 @@ static void ti_sn65dsi86_delete_aux(void *data)
 	auxiliary_device_delete(data);
 }
 
-/*
- * AUX bus docs say that a non-NULL release is mandatory, but it makes no
- * sense for the model used here where all of the aux devices are allocated
- * in the single shared structure. We'll use this noop as a workaround.
- */
-static void ti_sn65dsi86_noop(struct device *dev) {}
+static void ti_sn65dsi86_aux_device_release(struct device *dev)
+{
+	struct auxiliary_device *aux = container_of(dev, struct auxiliary_device, dev);
+
+	kfree(aux);
+}
 
 static int ti_sn65dsi86_add_aux_device(struct ti_sn65dsi86 *pdata,
-				       struct auxiliary_device *aux,
+				       struct auxiliary_device **aux_out,
 				       const char *name)
 {
 	struct device *dev = pdata->dev;
+	struct auxiliary_device *aux;
 	int ret;
 
+	aux = kzalloc(sizeof(*aux), GFP_KERNEL);
+	if (!aux)
+		return -ENOMEM;
+
 	aux->name = name;
 	aux->dev.parent = dev;
-	aux->dev.release = ti_sn65dsi86_noop;
+	aux->dev.release = ti_sn65dsi86_aux_device_release;
 	device_set_of_node_from_dev(&aux->dev, dev);
 	ret = auxiliary_device_init(aux);
-	if (ret)
+	if (ret) {
+		kfree(aux);
 		return ret;
+	}
 	ret = devm_add_action_or_reset(dev, ti_sn65dsi86_uninit_aux, aux);
 	if (ret)
 		return ret;
@@ -497,6 +504,8 @@ static int ti_sn65dsi86_add_aux_device(struct ti_sn65dsi86 *pdata,
 	if (ret)
 		return ret;
 	ret = devm_add_action_or_reset(dev, ti_sn65dsi86_delete_aux, aux);
+	if (!ret)
+		*aux_out = aux;
 
 	return ret;
 }

drivers/gpu/drm/drm_client.c

@@ -122,13 +122,34 @@ EXPORT_SYMBOL(drm_client_init);
  * drm_client_register() it is no longer permissible to call drm_client_release()
  * directly (outside the unregister callback), instead cleanup will happen
  * automatically on driver unload.
+ *
+ * Registering a client generates a hotplug event that allows the client
+ * to set up its display from pre-existing outputs. The client must have
+ * initialized its state to able to handle the hotplug event successfully.
  */
 void drm_client_register(struct drm_client_dev *client)
 {
 	struct drm_device *dev = client->dev;
+	int ret;
 
 	mutex_lock(&dev->clientlist_mutex);
 	list_add(&client->list, &dev->clientlist);
+
+	if (client->funcs && client->funcs->hotplug) {
+		/*
+		 * Perform an initial hotplug event to pick up the
+		 * display configuration for the client. This step
+		 * has to be performed *after* registering the client
+		 * in the list of clients, or a concurrent hotplug
+		 * event might be lost; leaving the display off.
+		 *
+		 * Hold the clientlist_mutex as for a regular hotplug
+		 * event.
+		 */
+		ret = client->funcs->hotplug(client);
+		if (ret)
+			drm_dbg_kms(dev, "client hotplug ret=%d\n", ret);
+	}
 	mutex_unlock(&dev->clientlist_mutex);
 }
 EXPORT_SYMBOL(drm_client_register);

drivers/gpu/drm/drm_fbdev_dma.c

@@ -217,7 +217,7 @@ static const struct drm_client_funcs drm_fbdev_dma_client_funcs = {
  * drm_fbdev_dma_setup() - Setup fbdev emulation for GEM DMA helpers
  * @dev: DRM device
  * @preferred_bpp: Preferred bits per pixel for the device.
- *                 @dev->mode_config.preferred_depth is used if this is zero.
+ *                 32 is used if this is zero.
  *
  * This function sets up fbdev emulation for GEM DMA drivers that support
  * dumb buffers with a virtual address and that can be mmap'ed.
@@ -252,10 +252,6 @@ void drm_fbdev_dma_setup(struct drm_device *dev, unsigned int preferred_bpp)
 		goto err_drm_client_init;
 	}
 
-	ret = drm_fbdev_dma_client_hotplug(&fb_helper->client);
-	if (ret)
-		drm_dbg_kms(dev, "client hotplug ret=%d\n", ret);
-
 	drm_client_register(&fb_helper->client);
 
 	return;

drivers/gpu/drm/drm_fbdev_generic.c

@@ -339,10 +339,6 @@ void drm_fbdev_generic_setup(struct drm_device *dev, unsigned int preferred_bpp)
 		goto err_drm_client_init;
 	}
 
-	ret = drm_fbdev_generic_client_hotplug(&fb_helper->client);
-	if (ret)
-		drm_dbg_kms(dev, "client hotplug ret=%d\n", ret);
-
 	drm_client_register(&fb_helper->client);
 
 	return;

drivers/gpu/drm/drm_syncobj.c

@@ -353,10 +353,10 @@ EXPORT_SYMBOL(drm_syncobj_replace_fence);
  */
 static int drm_syncobj_assign_null_handle(struct drm_syncobj *syncobj)
 {
-	struct dma_fence *fence = dma_fence_allocate_private_stub();
+	struct dma_fence *fence = dma_fence_allocate_private_stub(ktime_get());
 
-	if (IS_ERR(fence))
-		return PTR_ERR(fence);
+	if (!fence)
+		return -ENOMEM;
 
 	drm_syncobj_replace_fence(syncobj, fence);
 	dma_fence_put(fence);

drivers/gpu/drm/exynos/exynos_drm_fbdev.c

@@ -215,10 +215,6 @@ void exynos_drm_fbdev_setup(struct drm_device *dev)
 	if (ret)
 		goto err_drm_client_init;
 
-	ret = exynos_drm_fbdev_client_hotplug(&fb_helper->client);
-	if (ret)
-		drm_dbg_kms(dev, "client hotplug ret=%d\n", ret);
-
 	drm_client_register(&fb_helper->client);
 
 	return;

drivers/gpu/drm/gma500/fbdev.c

@@ -328,10 +328,6 @@ void psb_fbdev_setup(struct drm_psb_private *dev_priv)
 		goto err_drm_fb_helper_unprepare;
 	}
 
-	ret = psb_fbdev_client_hotplug(&fb_helper->client);
-	if (ret)
-		drm_dbg_kms(dev, "client hotplug ret=%d\n", ret);
-
 	drm_client_register(&fb_helper->client);
 
 	return;

drivers/gpu/drm/msm/msm_fbdev.c

@@ -246,10 +246,6 @@ void msm_fbdev_setup(struct drm_device *dev)
 		goto err_drm_fb_helper_unprepare;
 	}
 
-	ret = msm_fbdev_client_hotplug(&helper->client);
-	if (ret)
-		drm_dbg_kms(dev, "client hotplug ret=%d\n", ret);
-
 	drm_client_register(&helper->client);
 
 	return;

drivers/gpu/drm/nouveau/dispnv50/disp.c

@@ -910,15 +910,19 @@ nv50_msto_prepare(struct drm_atomic_state *state,
 	struct nouveau_drm *drm = nouveau_drm(msto->encoder.dev);
 	struct nv50_mstc *mstc = msto->mstc;
 	struct nv50_mstm *mstm = mstc->mstm;
-	struct drm_dp_mst_atomic_payload *payload;
+	struct drm_dp_mst_topology_state *old_mst_state;
+	struct drm_dp_mst_atomic_payload *payload, *old_payload;
 
 	NV_ATOMIC(drm, "%s: msto prepare\n", msto->encoder.name);
 
+	old_mst_state = drm_atomic_get_old_mst_topology_state(state, mgr);
+
 	payload = drm_atomic_get_mst_payload_state(mst_state, mstc->port);
+	old_payload = drm_atomic_get_mst_payload_state(old_mst_state, mstc->port);
 
 	// TODO: Figure out if we want to do a better job of handling VCPI allocation failures here?
 	if (msto->disabled) {
-		drm_dp_remove_payload(mgr, mst_state, payload, payload);
+		drm_dp_remove_payload(mgr, mst_state, old_payload, payload);
 
 		nvif_outp_dp_mst_vcpi(&mstm->outp->outp, msto->head->base.index, 0, 0, 0, 0);
 	} else {

drivers/gpu/drm/nouveau/nouveau_chan.c

@@ -90,6 +90,7 @@ nouveau_channel_del(struct nouveau_channel **pchan)
 		if (cli)
 			nouveau_svmm_part(chan->vmm->svmm, chan->inst);
 
+		nvif_object_dtor(&chan->blit);
 		nvif_object_dtor(&chan->nvsw);
 		nvif_object_dtor(&chan->gart);
 		nvif_object_dtor(&chan->vram);

drivers/gpu/drm/nouveau/nouveau_chan.h

@@ -53,6 +53,7 @@ struct nouveau_channel {
 	u32 user_put;
 
 	struct nvif_object user;
+	struct nvif_object blit;
 
 	struct nvif_event kill;
 	atomic_t killed;

drivers/gpu/drm/nouveau/nouveau_drm.c

@@ -375,15 +375,29 @@ nouveau_accel_gr_init(struct nouveau_drm *drm)
 		ret = nvif_object_ctor(&drm->channel->user, "drmNvsw",
 				       NVDRM_NVSW, nouveau_abi16_swclass(drm),
 				       NULL, 0, &drm->channel->nvsw);
+
+		if (ret == 0 && device->info.chipset >= 0x11) {
+			ret = nvif_object_ctor(&drm->channel->user, "drmBlit",
+					       0x005f, 0x009f,
+					       NULL, 0, &drm->channel->blit);
+		}
+
 		if (ret == 0) {
 			struct nvif_push *push = drm->channel->chan.push;
-			ret = PUSH_WAIT(push, 2);
-			if (ret == 0)
+			ret = PUSH_WAIT(push, 8);
+			if (ret == 0) {
+				if (device->info.chipset >= 0x11) {
+					PUSH_NVSQ(push, NV05F, 0x0000, drm->channel->blit.handle);
+					PUSH_NVSQ(push, NV09F, 0x0120, 0,
+							       0x0124, 1,
+							       0x0128, 2);
+				}
 				PUSH_NVSQ(push, NV_SW, 0x0000, drm->channel->nvsw.handle);
+			}
 		}
 
 		if (ret) {
-			NV_ERROR(drm, "failed to allocate sw class, %d\n", ret);
+			NV_ERROR(drm, "failed to allocate sw or blit class, %d\n", ret);
 			nouveau_accel_gr_fini(drm);
 			return;
 		}

drivers/gpu/drm/nouveau/nvkm/engine/disp/g94.c

@@ -295,6 +295,7 @@ g94_sor = {
 	.clock = nv50_sor_clock,
 	.war_2 = g94_sor_war_2,
 	.war_3 = g94_sor_war_3,
+	.hdmi = &g84_sor_hdmi,
 	.dp = &g94_sor_dp,
 };
 

drivers/gpu/drm/nouveau/nvkm/engine/disp/gt215.c

@@ -125,7 +125,7 @@ gt215_sor_hdmi_infoframe_avi(struct nvkm_ior *ior, int head, void *data, u32 siz
 	pack_hdmi_infoframe(&avi, data, size);
 
 	nvkm_mask(device, 0x61c520 + soff, 0x00000001, 0x00000000);
-	if (size)
+	if (!size)
 		return;
 
 	nvkm_wr32(device, 0x61c528 + soff, avi.header);

drivers/gpu/drm/nouveau/nvkm/subdev/acr/base.c

@@ -224,7 +224,7 @@ nvkm_acr_oneinit(struct nvkm_subdev *subdev)
 	u64 falcons;
 	int ret, i;
 
-	if (list_empty(&acr->hsfw)) {
+	if (list_empty(&acr->hsfw) || !acr->func || !acr->func->wpr_layout) {
 		nvkm_debug(subdev, "No HSFW(s)\n");
 		nvkm_acr_cleanup(acr);
 		return 0;

drivers/gpu/drm/omapdrm/omap_fbdev.c

@@ -318,10 +318,6 @@ void omap_fbdev_setup(struct drm_device *dev)
 
 	INIT_WORK(&fbdev->work, pan_worker);
 
-	ret = omap_fbdev_client_hotplug(&helper->client);
-	if (ret)
-		drm_dbg_kms(dev, "client hotplug ret=%d\n", ret);
-
 	drm_client_register(&helper->client);
 
 	return;

drivers/gpu/drm/panel/panel-simple.c

@@ -2178,6 +2178,7 @@ static const struct panel_desc innolux_at043tn24 = {
 		.height = 54,
 	},
 	.bus_format = MEDIA_BUS_FMT_RGB888_1X24,
+	.connector_type = DRM_MODE_CONNECTOR_DPI,
 	.bus_flags = DRM_BUS_FLAG_DE_HIGH | DRM_BUS_FLAG_PIXDATA_DRIVE_POSEDGE,
 };
 
@@ -3202,6 +3203,7 @@ static const struct drm_display_mode powertip_ph800480t013_idf02_mode = {
 	.vsync_start = 480 + 49,
 	.vsync_end = 480 + 49 + 2,
 	.vtotal = 480 + 49 + 2 + 22,
+	.flags = DRM_MODE_FLAG_NVSYNC | DRM_MODE_FLAG_NHSYNC,
 };
 
 static const struct panel_desc powertip_ph800480t013_idf02  = {

drivers/gpu/drm/radeon/radeon_fbdev.c

@@ -383,10 +383,6 @@ void radeon_fbdev_setup(struct radeon_device *rdev)
 		goto err_drm_client_init;
 	}
 
-	ret = radeon_fbdev_client_hotplug(&fb_helper->client);
-	if (ret)
-		drm_dbg_kms(rdev->ddev, "client hotplug ret=%d\n", ret);
-
 	drm_client_register(&fb_helper->client);
 
 	return;

drivers/gpu/drm/scheduler/sched_entity.c

@@ -176,16 +176,32 @@ static void drm_sched_entity_kill_jobs_cb(struct dma_fence *f,
 {
 	struct drm_sched_job *job = container_of(cb, struct drm_sched_job,
 						 finish_cb);
-	int r;
+	unsigned long index;
 
 	dma_fence_put(f);
 
 	/* Wait for all dependencies to avoid data corruptions */
-	while (!xa_empty(&job->dependencies)) {
-		f = xa_erase(&job->dependencies, job->last_dependency++);
-		r = dma_fence_add_callback(f, &job->finish_cb,
-					   drm_sched_entity_kill_jobs_cb);
-		if (!r)
+	xa_for_each(&job->dependencies, index, f) {
+		struct drm_sched_fence *s_fence = to_drm_sched_fence(f);
+
+		if (s_fence && f == &s_fence->scheduled) {
+			/* The dependencies array had a reference on the scheduled
+			 * fence, and the finished fence refcount might have
+			 * dropped to zero. Use dma_fence_get_rcu() so we get
+			 * a NULL fence in that case.
+			 */
+			f = dma_fence_get_rcu(&s_fence->finished);
+
+			/* Now that we have a reference on the finished fence,
+			 * we can release the reference the dependencies array
+			 * had on the scheduled fence.
+			 */
+			dma_fence_put(&s_fence->scheduled);
+		}
+
+		xa_erase(&job->dependencies, index);
+		if (f && !dma_fence_add_callback(f, &job->finish_cb,
+						 drm_sched_entity_kill_jobs_cb))
 			return;
 
 		dma_fence_put(f);
@@ -415,8 +431,17 @@ static struct dma_fence *
 drm_sched_job_dependency(struct drm_sched_job *job,
 			 struct drm_sched_entity *entity)
 {
-	if (!xa_empty(&job->dependencies))
-		return xa_erase(&job->dependencies, job->last_dependency++);
+	struct dma_fence *f;
+
+	/* We keep the fence around, so we can iterate over all dependencies
+	 * in drm_sched_entity_kill_jobs_cb() to ensure all deps are signaled
+	 * before killing the job.
+	 */
+	f = xa_load(&job->dependencies, job->last_dependency);
+	if (f) {
+		job->last_dependency++;
+		return dma_fence_get(f);
+	}
 
 	if (job->sched->ops->prepare_job)
 		return job->sched->ops->prepare_job(job, entity);

drivers/gpu/drm/scheduler/sched_fence.c

@@ -48,8 +48,32 @@ static void __exit drm_sched_fence_slab_fini(void)
 	kmem_cache_destroy(sched_fence_slab);
 }
 
-void drm_sched_fence_scheduled(struct drm_sched_fence *fence)
+static void drm_sched_fence_set_parent(struct drm_sched_fence *s_fence,
+				       struct dma_fence *fence)
 {
+	/*
+	 * smp_store_release() to ensure another thread racing us
+	 * in drm_sched_fence_set_deadline_finished() sees the
+	 * fence's parent set before test_bit()
+	 */
+	smp_store_release(&s_fence->parent, dma_fence_get(fence));
+	if (test_bit(DRM_SCHED_FENCE_FLAG_HAS_DEADLINE_BIT,
+		     &s_fence->finished.flags))
+		dma_fence_set_deadline(fence, s_fence->deadline);
+}
+
+void drm_sched_fence_scheduled(struct drm_sched_fence *fence,
+			       struct dma_fence *parent)
+{
+	/* Set the parent before signaling the scheduled fence, such that,
+	 * any waiter expecting the parent to be filled after the job has
+	 * been scheduled (which is the case for drivers delegating waits
+	 * to some firmware) doesn't have to busy wait for parent to show
+	 * up.
+	 */
+	if (!IS_ERR_OR_NULL(parent))
+		drm_sched_fence_set_parent(fence, parent);
+
 	dma_fence_signal(&fence->scheduled);
 }
 
@@ -181,20 +205,6 @@ struct drm_sched_fence *to_drm_sched_fence(struct dma_fence *f)
 }
 EXPORT_SYMBOL(to_drm_sched_fence);
 
-void drm_sched_fence_set_parent(struct drm_sched_fence *s_fence,
-				struct dma_fence *fence)
-{
-	/*
-	 * smp_store_release() to ensure another thread racing us
-	 * in drm_sched_fence_set_deadline_finished() sees the
-	 * fence's parent set before test_bit()
-	 */
-	smp_store_release(&s_fence->parent, dma_fence_get(fence));
-	if (test_bit(DRM_SCHED_FENCE_FLAG_HAS_DEADLINE_BIT,
-		     &s_fence->finished.flags))
-		dma_fence_set_deadline(fence, s_fence->deadline);
-}
-
 struct drm_sched_fence *drm_sched_fence_alloc(struct drm_sched_entity *entity,
 					      void *owner)
 {

drivers/gpu/drm/scheduler/sched_main.c

@@ -1043,10 +1043,9 @@ static int drm_sched_main(void *param)
 		trace_drm_run_job(sched_job, entity);
 		fence = sched->ops->run_job(sched_job);
 		complete_all(&entity->entity_idle);
-		drm_sched_fence_scheduled(s_fence);
+		drm_sched_fence_scheduled(s_fence, fence);
 
 		if (!IS_ERR_OR_NULL(fence)) {
-			drm_sched_fence_set_parent(s_fence, fence);
 			/* Drop for original kref_init of the fence */
 			dma_fence_put(fence);
 

drivers/gpu/drm/tegra/fbdev.c

@@ -225,10 +225,6 @@ void tegra_fbdev_setup(struct drm_device *dev)
 	if (ret)
 		goto err_drm_client_init;
 
-	ret = tegra_fbdev_client_hotplug(&helper->client);
-	if (ret)
-		drm_dbg_kms(dev, "client hotplug ret=%d\n", ret);
-
 	drm_client_register(&helper->client);
 
 	return;

drivers/gpu/drm/ttm/ttm_bo.c

@@ -458,18 +458,18 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo,
 		goto out;
 	}
 
-bounce:
-	ret = ttm_bo_handle_move_mem(bo, evict_mem, true, ctx, &hop);
-	if (ret == -EMULTIHOP) {
+	do {
+		ret = ttm_bo_handle_move_mem(bo, evict_mem, true, ctx, &hop);
+		if (ret != -EMULTIHOP)
+			break;
+
 		ret = ttm_bo_bounce_temp_buffer(bo, &evict_mem, ctx, &hop);
-		if (ret) {
-			if (ret != -ERESTARTSYS && ret != -EINTR)
-				pr_err("Buffer eviction failed\n");
-			ttm_resource_free(bo, &evict_mem);
-			goto out;
-		}
-		/* try and move to final place now. */
-		goto bounce;
+	} while (!ret);
+
+	if (ret) {
+		ttm_resource_free(bo, &evict_mem);
+		if (ret != -ERESTARTSYS && ret != -EINTR)
+			pr_err("Buffer eviction failed\n");
 	}
 out:
 	return ret;
@@ -517,6 +517,12 @@ static bool ttm_bo_evict_swapout_allowable(struct ttm_buffer_object *bo,
 {
 	bool ret = false;
 
+	if (bo->pin_count) {
+		*locked = false;
+		*busy = false;
+		return false;
+	}
+
 	if (bo->base.resv == ctx->resv) {
 		dma_resv_assert_held(bo->base.resv);
 		if (ctx->allow_res_evict)
@@ -1167,6 +1173,7 @@ int ttm_bo_swapout(struct ttm_buffer_object *bo, struct ttm_operation_ctx *ctx,
 		ret = ttm_bo_handle_move_mem(bo, evict_mem, true, &ctx, &hop);
 		if (unlikely(ret != 0)) {
 			WARN(ret == -EMULTIHOP, "Unexpected multihop in swaput - likely driver bug.\n");
+			ttm_resource_free(bo, &evict_mem);
 			goto out;
 		}
 	}

drivers/gpu/drm/ttm/ttm_resource.c

@@ -86,6 +86,8 @@ static void ttm_lru_bulk_move_pos_tail(struct ttm_lru_bulk_move_pos *pos,
 				       struct ttm_resource *res)
 {
 	if (pos->last != res) {
+		if (pos->first == res)
+			pos->first = list_next_entry(res, lru);
 		list_move(&res->lru, &pos->last->lru);
 		pos->last = res;
 	}
@@ -111,7 +113,8 @@ static void ttm_lru_bulk_move_del(struct ttm_lru_bulk_move *bulk,
 {
 	struct ttm_lru_bulk_move_pos *pos = ttm_lru_bulk_move_pos(bulk, res);
 
-	if (unlikely(pos->first == res && pos->last == res)) {
+	if (unlikely(WARN_ON(!pos->first || !pos->last) ||
+		     (pos->first == res && pos->last == res))) {
 		pos->first = NULL;
 		pos->last = NULL;
 	} else if (pos->first == res) {

include/drm/gpu_scheduler.h

@@ -583,15 +583,14 @@ void drm_sched_entity_set_priority(struct drm_sched_entity *entity,
 bool drm_sched_entity_is_ready(struct drm_sched_entity *entity);
 int drm_sched_entity_error(struct drm_sched_entity *entity);
 
-void drm_sched_fence_set_parent(struct drm_sched_fence *s_fence,
-				struct dma_fence *fence);
 struct drm_sched_fence *drm_sched_fence_alloc(
 	struct drm_sched_entity *s_entity, void *owner);
 void drm_sched_fence_init(struct drm_sched_fence *fence,
 			  struct drm_sched_entity *entity);
 void drm_sched_fence_free(struct drm_sched_fence *fence);
 
-void drm_sched_fence_scheduled(struct drm_sched_fence *fence);
+void drm_sched_fence_scheduled(struct drm_sched_fence *fence,
+			       struct dma_fence *parent);
 void drm_sched_fence_finished(struct drm_sched_fence *fence, int result);
 
 unsigned long drm_sched_suspend_timeout(struct drm_gpu_scheduler *sched);

include/linux/dma-fence.h

@@ -606,7 +606,7 @@ static inline signed long dma_fence_wait(struct dma_fence *fence, bool intr)
 void dma_fence_set_deadline(struct dma_fence *fence, ktime_t deadline);
 
 struct dma_fence *dma_fence_get_stub(void);
-struct dma_fence *dma_fence_allocate_private_stub(void);
+struct dma_fence *dma_fence_allocate_private_stub(ktime_t timestamp);
 u64 dma_fence_context_alloc(unsigned num);
 
 extern const struct dma_fence_ops dma_fence_array_ops;