Merge tag 'drm-fixes-2020-01-24' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "This one has a core mst fix and two i915 fixes. amdgpu just enables
  some hw outside experimental.

  The panfrost fix is a little bigger than I'd like at this stage but it
  fixes a fairly fundamental problem with global shared buffers in that
  driver, and since it's confined to that driver and I've taken a look
  at it, I think it's fine to get into the tree now, so it can get
  stable propagated as well.

  core/mst:
   - Fix SST branch device handling

  amdgpu:
   - enable renoir outside experimental

  i915:
   - Avoid overflow with huge userptr objects
   - uAPI fix to correctly handle negative values in
     engine->uabi_class/instance (cc: stable)

  panfrost:
   - Fix mapping of globally visible BO's (Boris)"

* tag 'drm-fixes-2020-01-24' of git://anongit.freedesktop.org/drm/drm:
  drm/amdgpu: remove the experimental flag for renoir
  drm/panfrost: Add the panfrost_gem_mapping concept
  drm/i915: Align engine->uabi_class/instance with i915_drm.h
  drm/i915/userptr: fix size calculation
  drm/dp_mst: Handle SST-only branch device case

drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c

@@ -1004,7 +1004,7 @@ static const struct pci_device_id pciidlist[] = {
 	{0x1002, 0x734F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI14},
 
 	/* Renoir */
-	{0x1002, 0x1636, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RENOIR|AMD_IS_APU|AMD_EXP_HW_SUPPORT},
+	{0x1002, 0x1636, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RENOIR|AMD_IS_APU},
 
 	/* Navi12 */
 	{0x1002, 0x7360, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_NAVI12|AMD_EXP_HW_SUPPORT},

drivers/gpu/drm/drm_dp_mst_topology.c

@@ -1916,73 +1916,90 @@ static u8 drm_dp_calculate_rad(struct drm_dp_mst_port *port,
 	return parent_lct + 1;
 }
 
-static int drm_dp_port_set_pdt(struct drm_dp_mst_port *port, u8 new_pdt)
+static bool drm_dp_mst_is_dp_mst_end_device(u8 pdt, bool mcs)
+{
+	switch (pdt) {
+	case DP_PEER_DEVICE_DP_LEGACY_CONV:
+	case DP_PEER_DEVICE_SST_SINK:
+		return true;
+	case DP_PEER_DEVICE_MST_BRANCHING:
+		/* For sst branch device */
+		if (!mcs)
+			return true;
+
+		return false;
+	}
+	return true;
+}
+
+static int
+drm_dp_port_set_pdt(struct drm_dp_mst_port *port, u8 new_pdt,
+		    bool new_mcs)
 {
 	struct drm_dp_mst_topology_mgr *mgr = port->mgr;
 	struct drm_dp_mst_branch *mstb;
 	u8 rad[8], lct;
 	int ret = 0;
 
-	if (port->pdt == new_pdt)
+	if (port->pdt == new_pdt && port->mcs == new_mcs)
 		return 0;
 
 	/* Teardown the old pdt, if there is one */
-	switch (port->pdt) {
-	case DP_PEER_DEVICE_DP_LEGACY_CONV:
-	case DP_PEER_DEVICE_SST_SINK:
-		/*
-		 * If the new PDT would also have an i2c bus, don't bother
-		 * with reregistering it
-		 */
-		if (new_pdt == DP_PEER_DEVICE_DP_LEGACY_CONV ||
-		    new_pdt == DP_PEER_DEVICE_SST_SINK) {
-			port->pdt = new_pdt;
-			return 0;
-		}
+	if (port->pdt != DP_PEER_DEVICE_NONE) {
+		if (drm_dp_mst_is_dp_mst_end_device(port->pdt, port->mcs)) {
+			/*
+			 * If the new PDT would also have an i2c bus,
+			 * don't bother with reregistering it
+			 */
+			if (new_pdt != DP_PEER_DEVICE_NONE &&
+			    drm_dp_mst_is_dp_mst_end_device(new_pdt, new_mcs)) {
+				port->pdt = new_pdt;
+				port->mcs = new_mcs;
+				return 0;
+			}
 
-		/* remove i2c over sideband */
-		drm_dp_mst_unregister_i2c_bus(&port->aux);
-		break;
-	case DP_PEER_DEVICE_MST_BRANCHING:
-		mutex_lock(&mgr->lock);
-		drm_dp_mst_topology_put_mstb(port->mstb);
-		port->mstb = NULL;
-		mutex_unlock(&mgr->lock);
-		break;
+			/* remove i2c over sideband */
+			drm_dp_mst_unregister_i2c_bus(&port->aux);
+		} else {
+			mutex_lock(&mgr->lock);
+			drm_dp_mst_topology_put_mstb(port->mstb);
+			port->mstb = NULL;
+			mutex_unlock(&mgr->lock);
+		}
 	}
 
 	port->pdt = new_pdt;
-	switch (port->pdt) {
-	case DP_PEER_DEVICE_DP_LEGACY_CONV:
-	case DP_PEER_DEVICE_SST_SINK:
-		/* add i2c over sideband */
-		ret = drm_dp_mst_register_i2c_bus(&port->aux);
-		break;
+	port->mcs = new_mcs;
 
-	case DP_PEER_DEVICE_MST_BRANCHING:
-		lct = drm_dp_calculate_rad(port, rad);
-		mstb = drm_dp_add_mst_branch_device(lct, rad);
-		if (!mstb) {
-			ret = -ENOMEM;
-			DRM_ERROR("Failed to create MSTB for port %p", port);
-			goto out;
-		}
+	if (port->pdt != DP_PEER_DEVICE_NONE) {
+		if (drm_dp_mst_is_dp_mst_end_device(port->pdt, port->mcs)) {
+			/* add i2c over sideband */
+			ret = drm_dp_mst_register_i2c_bus(&port->aux);
+		} else {
+			lct = drm_dp_calculate_rad(port, rad);
+			mstb = drm_dp_add_mst_branch_device(lct, rad);
+			if (!mstb) {
+				ret = -ENOMEM;
+				DRM_ERROR("Failed to create MSTB for port %p",
+					  port);
+				goto out;
+			}
 
-		mutex_lock(&mgr->lock);
-		port->mstb = mstb;
-		mstb->mgr = port->mgr;
-		mstb->port_parent = port;
+			mutex_lock(&mgr->lock);
+			port->mstb = mstb;
+			mstb->mgr = port->mgr;
+			mstb->port_parent = port;
 
-		/*
-		 * Make sure this port's memory allocation stays
-		 * around until its child MSTB releases it
-		 */
-		drm_dp_mst_get_port_malloc(port);
-		mutex_unlock(&mgr->lock);
+			/*
+			 * Make sure this port's memory allocation stays
+			 * around until its child MSTB releases it
+			 */
+			drm_dp_mst_get_port_malloc(port);
+			mutex_unlock(&mgr->lock);
 
-		/* And make sure we send a link address for this */
-		ret = 1;
-		break;
+			/* And make sure we send a link address for this */
+			ret = 1;
+		}
 	}
 
 out:
@@ -2135,9 +2152,8 @@ drm_dp_mst_port_add_connector(struct drm_dp_mst_branch *mstb,
 		goto error;
 	}
 
-	if ((port->pdt == DP_PEER_DEVICE_DP_LEGACY_CONV ||
-	     port->pdt == DP_PEER_DEVICE_SST_SINK) &&
-	    port->port_num >= DP_MST_LOGICAL_PORT_0) {
+	if (port->pdt != DP_PEER_DEVICE_NONE &&
+	    drm_dp_mst_is_dp_mst_end_device(port->pdt, port->mcs)) {
 		port->cached_edid = drm_get_edid(port->connector,
 						 &port->aux.ddc);
 		drm_connector_set_tile_property(port->connector);
@@ -2201,6 +2217,7 @@ drm_dp_mst_handle_link_address_port(struct drm_dp_mst_branch *mstb,
 	struct drm_dp_mst_port *port;
 	int old_ddps = 0, ret;
 	u8 new_pdt = DP_PEER_DEVICE_NONE;
+	bool new_mcs = 0;
 	bool created = false, send_link_addr = false, changed = false;
 
 	port = drm_dp_get_port(mstb, port_msg->port_number);
@@ -2245,7 +2262,7 @@ drm_dp_mst_handle_link_address_port(struct drm_dp_mst_branch *mstb,
 	port->input = port_msg->input_port;
 	if (!port->input)
 		new_pdt = port_msg->peer_device_type;
-	port->mcs = port_msg->mcs;
+	new_mcs = port_msg->mcs;
 	port->ddps = port_msg->ddps;
 	port->ldps = port_msg->legacy_device_plug_status;
 	port->dpcd_rev = port_msg->dpcd_revision;
@@ -2272,7 +2289,7 @@ drm_dp_mst_handle_link_address_port(struct drm_dp_mst_branch *mstb,
 		}
 	}
 
-	ret = drm_dp_port_set_pdt(port, new_pdt);
+	ret = drm_dp_port_set_pdt(port, new_pdt, new_mcs);
 	if (ret == 1) {
 		send_link_addr = true;
 	} else if (ret < 0) {
@@ -2286,7 +2303,8 @@ drm_dp_mst_handle_link_address_port(struct drm_dp_mst_branch *mstb,
 	 * we're coming out of suspend. In this case, always resend the link
 	 * address if there's an MSTB on this port
 	 */
-	if (!created && port->pdt == DP_PEER_DEVICE_MST_BRANCHING)
+	if (!created && port->pdt == DP_PEER_DEVICE_MST_BRANCHING &&
+	    port->mcs)
 		send_link_addr = true;
 
 	if (port->connector)
@@ -2323,6 +2341,7 @@ drm_dp_mst_handle_conn_stat(struct drm_dp_mst_branch *mstb,
 	struct drm_dp_mst_port *port;
 	int old_ddps, old_input, ret, i;
 	u8 new_pdt;
+	bool new_mcs;
 	bool dowork = false, create_connector = false;
 
 	port = drm_dp_get_port(mstb, conn_stat->port_number);
@@ -2354,7 +2373,6 @@ drm_dp_mst_handle_conn_stat(struct drm_dp_mst_branch *mstb,
 	old_ddps = port->ddps;
 	old_input = port->input;
 	port->input = conn_stat->input_port;
-	port->mcs = conn_stat->message_capability_status;
 	port->ldps = conn_stat->legacy_device_plug_status;
 	port->ddps = conn_stat->displayport_device_plug_status;
 
@@ -2367,8 +2385,8 @@ drm_dp_mst_handle_conn_stat(struct drm_dp_mst_branch *mstb,
 	}
 
 	new_pdt = port->input ? DP_PEER_DEVICE_NONE : conn_stat->peer_device_type;
-
-	ret = drm_dp_port_set_pdt(port, new_pdt);
+	new_mcs = conn_stat->message_capability_status;
+	ret = drm_dp_port_set_pdt(port, new_pdt, new_mcs);
 	if (ret == 1) {
 		dowork = true;
 	} else if (ret < 0) {
@@ -3929,6 +3947,8 @@ drm_dp_mst_detect_port(struct drm_connector *connector,
 	switch (port->pdt) {
 	case DP_PEER_DEVICE_NONE:
 	case DP_PEER_DEVICE_MST_BRANCHING:
+		if (!port->mcs)
+			ret = connector_status_connected;
 		break;
 
 	case DP_PEER_DEVICE_SST_SINK:
@@ -4541,7 +4561,7 @@ drm_dp_delayed_destroy_port(struct drm_dp_mst_port *port)
 	if (port->connector)
 		port->mgr->cbs->destroy_connector(port->mgr, port->connector);
 
-	drm_dp_port_set_pdt(port, DP_PEER_DEVICE_NONE);
+	drm_dp_port_set_pdt(port, DP_PEER_DEVICE_NONE, port->mcs);
 	drm_dp_mst_put_port_malloc(port);
 }
 

drivers/gpu/drm/i915/gem/i915_gem_busy.c

@@ -9,16 +9,16 @@
 #include "i915_gem_ioctls.h"
 #include "i915_gem_object.h"
 
-static __always_inline u32 __busy_read_flag(u8 id)
+static __always_inline u32 __busy_read_flag(u16 id)
 {
-	if (id == (u8)I915_ENGINE_CLASS_INVALID)
+	if (id == (u16)I915_ENGINE_CLASS_INVALID)
 		return 0xffff0000u;
 
 	GEM_BUG_ON(id >= 16);
 	return 0x10000u << id;
 }
 
-static __always_inline u32 __busy_write_id(u8 id)
+static __always_inline u32 __busy_write_id(u16 id)
 {
 	/*
 	 * The uABI guarantees an active writer is also amongst the read
@@ -29,14 +29,14 @@ static __always_inline u32 __busy_write_id(u8 id)
 	 * last_read - hence we always set both read and write busy for
 	 * last_write.
 	 */
-	if (id == (u8)I915_ENGINE_CLASS_INVALID)
+	if (id == (u16)I915_ENGINE_CLASS_INVALID)
 		return 0xffffffffu;
 
 	return (id + 1) | __busy_read_flag(id);
 }
 
 static __always_inline unsigned int
-__busy_set_if_active(const struct dma_fence *fence, u32 (*flag)(u8 id))
+__busy_set_if_active(const struct dma_fence *fence, u32 (*flag)(u16 id))
 {
 	const struct i915_request *rq;
 
@@ -57,7 +57,7 @@ __busy_set_if_active(const struct dma_fence *fence, u32 (*flag)(u8 id))
 		return 0;
 
 	/* Beware type-expansion follies! */
-	BUILD_BUG_ON(!typecheck(u8, rq->engine->uabi_class));
+	BUILD_BUG_ON(!typecheck(u16, rq->engine->uabi_class));
 	return flag(rq->engine->uabi_class);
 }
 

drivers/gpu/drm/i915/gem/i915_gem_userptr.c

@@ -402,7 +402,7 @@ struct get_pages_work {
 
 static struct sg_table *
 __i915_gem_userptr_alloc_pages(struct drm_i915_gem_object *obj,
-			       struct page **pvec, int num_pages)
+			       struct page **pvec, unsigned long num_pages)
 {
 	unsigned int max_segment = i915_sg_segment_size();
 	struct sg_table *st;
@@ -448,9 +448,10 @@ __i915_gem_userptr_get_pages_worker(struct work_struct *_work)
 {
 	struct get_pages_work *work = container_of(_work, typeof(*work), work);
 	struct drm_i915_gem_object *obj = work->obj;
-	const int npages = obj->base.size >> PAGE_SHIFT;
+	const unsigned long npages = obj->base.size >> PAGE_SHIFT;
+	unsigned long pinned;
 	struct page **pvec;
-	int pinned, ret;
+	int ret;
 
 	ret = -ENOMEM;
 	pinned = 0;
@@ -553,7 +554,7 @@ __i915_gem_userptr_get_pages_schedule(struct drm_i915_gem_object *obj)
 
 static int i915_gem_userptr_get_pages(struct drm_i915_gem_object *obj)
 {
-	const int num_pages = obj->base.size >> PAGE_SHIFT;
+	const unsigned long num_pages = obj->base.size >> PAGE_SHIFT;
 	struct mm_struct *mm = obj->userptr.mm->mm;
 	struct page **pvec;
 	struct sg_table *pages;

drivers/gpu/drm/i915/gt/intel_engine_types.h

@@ -274,8 +274,8 @@ struct intel_engine_cs {
 	u8 class;
 	u8 instance;
 
-	u8 uabi_class;
-	u8 uabi_instance;
+	u16 uabi_class;
+	u16 uabi_instance;
 
 	u32 uabi_capabilities;
 	u32 context_size;

drivers/gpu/drm/i915/i915_gem_gtt.c

@@ -1177,6 +1177,7 @@ gen8_ppgtt_insert_pte(struct i915_ppgtt *ppgtt,
 	pd = i915_pd_entry(pdp, gen8_pd_index(idx, 2));
 	vaddr = kmap_atomic_px(i915_pt_entry(pd, gen8_pd_index(idx, 1)));
 	do {
+		GEM_BUG_ON(iter->sg->length < I915_GTT_PAGE_SIZE);
 		vaddr[gen8_pd_index(idx, 0)] = pte_encode | iter->dma;
 
 		iter->dma += I915_GTT_PAGE_SIZE;
@@ -1660,6 +1661,7 @@ static void gen6_ppgtt_insert_entries(struct i915_address_space *vm,
 
 	vaddr = kmap_atomic_px(i915_pt_entry(pd, act_pt));
 	do {
+		GEM_BUG_ON(iter.sg->length < I915_GTT_PAGE_SIZE);
 		vaddr[act_pte] = pte_encode | GEN6_PTE_ADDR_ENCODE(iter.dma);
 
 		iter.dma += I915_GTT_PAGE_SIZE;

drivers/gpu/drm/panfrost/panfrost_drv.c

@@ -78,8 +78,10 @@ static int panfrost_ioctl_get_param(struct drm_device *ddev, void *data, struct
 static int panfrost_ioctl_create_bo(struct drm_device *dev, void *data,
 		struct drm_file *file)
 {
+	struct panfrost_file_priv *priv = file->driver_priv;
 	struct panfrost_gem_object *bo;
 	struct drm_panfrost_create_bo *args = data;
+	struct panfrost_gem_mapping *mapping;
 
 	if (!args->size || args->pad ||
 	    (args->flags & ~(PANFROST_BO_NOEXEC | PANFROST_BO_HEAP)))
@@ -95,7 +97,14 @@ static int panfrost_ioctl_create_bo(struct drm_device *dev, void *data,
 	if (IS_ERR(bo))
 		return PTR_ERR(bo);
 
-	args->offset = bo->node.start << PAGE_SHIFT;
+	mapping = panfrost_gem_mapping_get(bo, priv);
+	if (!mapping) {
+		drm_gem_object_put_unlocked(&bo->base.base);
+		return -EINVAL;
+	}
+
+	args->offset = mapping->mmnode.start << PAGE_SHIFT;
+	panfrost_gem_mapping_put(mapping);
 
 	return 0;
 }
@@ -119,6 +128,11 @@ panfrost_lookup_bos(struct drm_device *dev,
 		  struct drm_panfrost_submit *args,
 		  struct panfrost_job *job)
 {
+	struct panfrost_file_priv *priv = file_priv->driver_priv;
+	struct panfrost_gem_object *bo;
+	unsigned int i;
+	int ret;
+
 	job->bo_count = args->bo_handle_count;
 
 	if (!job->bo_count)
@@ -130,9 +144,32 @@ panfrost_lookup_bos(struct drm_device *dev,
 	if (!job->implicit_fences)
 		return -ENOMEM;
 
-	return drm_gem_objects_lookup(file_priv,
-				      (void __user *)(uintptr_t)args->bo_handles,
-				      job->bo_count, &job->bos);
+	ret = drm_gem_objects_lookup(file_priv,
+				     (void __user *)(uintptr_t)args->bo_handles,
+				     job->bo_count, &job->bos);
+	if (ret)
+		return ret;
+
+	job->mappings = kvmalloc_array(job->bo_count,
+				       sizeof(struct panfrost_gem_mapping *),
+				       GFP_KERNEL | __GFP_ZERO);
+	if (!job->mappings)
+		return -ENOMEM;
+
+	for (i = 0; i < job->bo_count; i++) {
+		struct panfrost_gem_mapping *mapping;
+
+		bo = to_panfrost_bo(job->bos[i]);
+		mapping = panfrost_gem_mapping_get(bo, priv);
+		if (!mapping) {
+			ret = -EINVAL;
+			break;
+		}
+
+		job->mappings[i] = mapping;
+	}
+
+	return ret;
 }
 
 /**
@@ -320,7 +357,9 @@ static int panfrost_ioctl_mmap_bo(struct drm_device *dev, void *data,
 static int panfrost_ioctl_get_bo_offset(struct drm_device *dev, void *data,
 			    struct drm_file *file_priv)
 {
+	struct panfrost_file_priv *priv = file_priv->driver_priv;
 	struct drm_panfrost_get_bo_offset *args = data;
+	struct panfrost_gem_mapping *mapping;
 	struct drm_gem_object *gem_obj;
 	struct panfrost_gem_object *bo;
 
@@ -331,18 +370,26 @@ static int panfrost_ioctl_get_bo_offset(struct drm_device *dev, void *data,
 	}
 	bo = to_panfrost_bo(gem_obj);
 
-	args->offset = bo->node.start << PAGE_SHIFT;
-
+	mapping = panfrost_gem_mapping_get(bo, priv);
 	drm_gem_object_put_unlocked(gem_obj);
+
+	if (!mapping)
+		return -EINVAL;
+
+	args->offset = mapping->mmnode.start << PAGE_SHIFT;
+	panfrost_gem_mapping_put(mapping);
 	return 0;
 }
 
 static int panfrost_ioctl_madvise(struct drm_device *dev, void *data,
 				  struct drm_file *file_priv)
 {
+	struct panfrost_file_priv *priv = file_priv->driver_priv;
 	struct drm_panfrost_madvise *args = data;
 	struct panfrost_device *pfdev = dev->dev_private;
 	struct drm_gem_object *gem_obj;
+	struct panfrost_gem_object *bo;
+	int ret = 0;
 
 	gem_obj = drm_gem_object_lookup(file_priv, args->handle);
 	if (!gem_obj) {
@@ -350,22 +397,48 @@ static int panfrost_ioctl_madvise(struct drm_device *dev, void *data,
 		return -ENOENT;
 	}
 
+	bo = to_panfrost_bo(gem_obj);
+
 	mutex_lock(&pfdev->shrinker_lock);
+	mutex_lock(&bo->mappings.lock);
+	if (args->madv == PANFROST_MADV_DONTNEED) {
+		struct panfrost_gem_mapping *first;
+
+		first = list_first_entry(&bo->mappings.list,
+					 struct panfrost_gem_mapping,
+					 node);
+
+		/*
+		 * If we want to mark the BO purgeable, there must be only one
+		 * user: the caller FD.
+		 * We could do something smarter and mark the BO purgeable only
+		 * when all its users have marked it purgeable, but globally
+		 * visible/shared BOs are likely to never be marked purgeable
+		 * anyway, so let's not bother.
+		 */
+		if (!list_is_singular(&bo->mappings.list) ||
+		    WARN_ON_ONCE(first->mmu != &priv->mmu)) {
+			ret = -EINVAL;
+			goto out_unlock_mappings;
+		}
+	}
+
 	args->retained = drm_gem_shmem_madvise(gem_obj, args->madv);
 
 	if (args->retained) {
-		struct panfrost_gem_object *bo = to_panfrost_bo(gem_obj);
-
 		if (args->madv == PANFROST_MADV_DONTNEED)
 			list_add_tail(&bo->base.madv_list,
 				      &pfdev->shrinker_list);
 		else if (args->madv == PANFROST_MADV_WILLNEED)
 			list_del_init(&bo->base.madv_list);
 	}
+
+out_unlock_mappings:
+	mutex_unlock(&bo->mappings.lock);
 	mutex_unlock(&pfdev->shrinker_lock);
 
 	drm_gem_object_put_unlocked(gem_obj);
-	return 0;
+	return ret;
 }
 
 int panfrost_unstable_ioctl_check(void)

drivers/gpu/drm/panfrost/panfrost_gem.c

@@ -29,6 +29,12 @@ static void panfrost_gem_free_object(struct drm_gem_object *obj)
 	list_del_init(&bo->base.madv_list);
 	mutex_unlock(&pfdev->shrinker_lock);
 
+	/*
+	 * If we still have mappings attached to the BO, there's a problem in
+	 * our refcounting.
+	 */
+	WARN_ON_ONCE(!list_empty(&bo->mappings.list));
+
 	if (bo->sgts) {
 		int i;
 		int n_sgt = bo->base.base.size / SZ_2M;
@@ -46,6 +52,69 @@ static void panfrost_gem_free_object(struct drm_gem_object *obj)
 	drm_gem_shmem_free_object(obj);
 }
 
+struct panfrost_gem_mapping *
+panfrost_gem_mapping_get(struct panfrost_gem_object *bo,
+			 struct panfrost_file_priv *priv)
+{
+	struct panfrost_gem_mapping *iter, *mapping = NULL;
+
+	mutex_lock(&bo->mappings.lock);
+	list_for_each_entry(iter, &bo->mappings.list, node) {
+		if (iter->mmu == &priv->mmu) {
+			kref_get(&iter->refcount);
+			mapping = iter;
+			break;
+		}
+	}
+	mutex_unlock(&bo->mappings.lock);
+
+	return mapping;
+}
+
+static void
+panfrost_gem_teardown_mapping(struct panfrost_gem_mapping *mapping)
+{
+	struct panfrost_file_priv *priv;
+
+	if (mapping->active)
+		panfrost_mmu_unmap(mapping);
+
+	priv = container_of(mapping->mmu, struct panfrost_file_priv, mmu);
+	spin_lock(&priv->mm_lock);
+	if (drm_mm_node_allocated(&mapping->mmnode))
+		drm_mm_remove_node(&mapping->mmnode);
+	spin_unlock(&priv->mm_lock);
+}
+
+static void panfrost_gem_mapping_release(struct kref *kref)
+{
+	struct panfrost_gem_mapping *mapping;
+
+	mapping = container_of(kref, struct panfrost_gem_mapping, refcount);
+
+	panfrost_gem_teardown_mapping(mapping);
+	drm_gem_object_put_unlocked(&mapping->obj->base.base);
+	kfree(mapping);
+}
+
+void panfrost_gem_mapping_put(struct panfrost_gem_mapping *mapping)
+{
+	if (!mapping)
+		return;
+
+	kref_put(&mapping->refcount, panfrost_gem_mapping_release);
+}
+
+void panfrost_gem_teardown_mappings(struct panfrost_gem_object *bo)
+{
+	struct panfrost_gem_mapping *mapping;
+
+	mutex_lock(&bo->mappings.lock);
+	list_for_each_entry(mapping, &bo->mappings.list, node)
+		panfrost_gem_teardown_mapping(mapping);
+	mutex_unlock(&bo->mappings.lock);
+}
+
 int panfrost_gem_open(struct drm_gem_object *obj, struct drm_file *file_priv)
 {
 	int ret;
@@ -54,6 +123,16 @@ int panfrost_gem_open(struct drm_gem_object *obj, struct drm_file *file_priv)
 	struct panfrost_gem_object *bo = to_panfrost_bo(obj);
 	unsigned long color = bo->noexec ? PANFROST_BO_NOEXEC : 0;
 	struct panfrost_file_priv *priv = file_priv->driver_priv;
+	struct panfrost_gem_mapping *mapping;
+
+	mapping = kzalloc(sizeof(*mapping), GFP_KERNEL);
+	if (!mapping)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&mapping->node);
+	kref_init(&mapping->refcount);
+	drm_gem_object_get(obj);
+	mapping->obj = bo;
 
 	/*
 	 * Executable buffers cannot cross a 16MB boundary as the program
@@ -66,37 +145,48 @@ int panfrost_gem_open(struct drm_gem_object *obj, struct drm_file *file_priv)
 	else
 		align = size >= SZ_2M ? SZ_2M >> PAGE_SHIFT : 0;
 
-	bo->mmu = &priv->mmu;
+	mapping->mmu = &priv->mmu;
 	spin_lock(&priv->mm_lock);
-	ret = drm_mm_insert_node_generic(&priv->mm, &bo->node,
+	ret = drm_mm_insert_node_generic(&priv->mm, &mapping->mmnode,
 					 size >> PAGE_SHIFT, align, color, 0);
 	spin_unlock(&priv->mm_lock);
 	if (ret)
-		return ret;
+		goto err;
 
 	if (!bo->is_heap) {
-		ret = panfrost_mmu_map(bo);
-		if (ret) {
-			spin_lock(&priv->mm_lock);
-			drm_mm_remove_node(&bo->node);
-			spin_unlock(&priv->mm_lock);
-		}
+		ret = panfrost_mmu_map(mapping);
+		if (ret)
+			goto err;
 	}
+
+	mutex_lock(&bo->mappings.lock);
+	WARN_ON(bo->base.madv != PANFROST_MADV_WILLNEED);
+	list_add_tail(&mapping->node, &bo->mappings.list);
+	mutex_unlock(&bo->mappings.lock);
+
+err:
+	if (ret)
+		panfrost_gem_mapping_put(mapping);
 	return ret;
 }
 
 void panfrost_gem_close(struct drm_gem_object *obj, struct drm_file *file_priv)
 {
-	struct panfrost_gem_object *bo = to_panfrost_bo(obj);
 	struct panfrost_file_priv *priv = file_priv->driver_priv;
+	struct panfrost_gem_object *bo = to_panfrost_bo(obj);
+	struct panfrost_gem_mapping *mapping = NULL, *iter;
 
-	if (bo->is_mapped)
-		panfrost_mmu_unmap(bo);
+	mutex_lock(&bo->mappings.lock);
+	list_for_each_entry(iter, &bo->mappings.list, node) {
+		if (iter->mmu == &priv->mmu) {
+			mapping = iter;
+			list_del(&iter->node);
+			break;
+		}
+	}
+	mutex_unlock(&bo->mappings.lock);
 
-	spin_lock(&priv->mm_lock);
-	if (drm_mm_node_allocated(&bo->node))
-		drm_mm_remove_node(&bo->node);
-	spin_unlock(&priv->mm_lock);
+	panfrost_gem_mapping_put(mapping);
 }
 
 static int panfrost_gem_pin(struct drm_gem_object *obj)
@@ -136,6 +226,8 @@ struct drm_gem_object *panfrost_gem_create_object(struct drm_device *dev, size_t
 	if (!obj)
 		return NULL;
 
+	INIT_LIST_HEAD(&obj->mappings.list);
+	mutex_init(&obj->mappings.lock);
 	obj->base.base.funcs = &panfrost_gem_funcs;
 
 	return &obj->base.base;

drivers/gpu/drm/panfrost/panfrost_gem.h

@@ -13,23 +13,46 @@ struct panfrost_gem_object {
 	struct drm_gem_shmem_object base;
 	struct sg_table *sgts;
 
-	struct panfrost_mmu *mmu;
-	struct drm_mm_node node;
-	bool is_mapped		:1;
+	/*
+	 * Use a list for now. If searching a mapping ever becomes the
+	 * bottleneck, we should consider using an RB-tree, or even better,
+	 * let the core store drm_gem_object_mapping entries (where we
+	 * could place driver specific data) instead of drm_gem_object ones
+	 * in its drm_file->object_idr table.
+	 *
+	 * struct drm_gem_object_mapping {
+	 *	struct drm_gem_object *obj;
+	 *	void *driver_priv;
+	 * };
+	 */
+	struct {
+		struct list_head list;
+		struct mutex lock;
+	} mappings;
+
 	bool noexec		:1;
 	bool is_heap		:1;
 };
 
+struct panfrost_gem_mapping {
+	struct list_head node;
+	struct kref refcount;
+	struct panfrost_gem_object *obj;
+	struct drm_mm_node mmnode;
+	struct panfrost_mmu *mmu;
+	bool active		:1;
+};
+
 static inline
 struct  panfrost_gem_object *to_panfrost_bo(struct drm_gem_object *obj)
 {
 	return container_of(to_drm_gem_shmem_obj(obj), struct panfrost_gem_object, base);
 }
 
-static inline
-struct  panfrost_gem_object *drm_mm_node_to_panfrost_bo(struct drm_mm_node *node)
+static inline struct panfrost_gem_mapping *
+drm_mm_node_to_panfrost_mapping(struct drm_mm_node *node)
 {
-	return container_of(node, struct panfrost_gem_object, node);
+	return container_of(node, struct panfrost_gem_mapping, mmnode);
 }
 
 struct drm_gem_object *panfrost_gem_create_object(struct drm_device *dev, size_t size);
@@ -49,6 +72,12 @@ int panfrost_gem_open(struct drm_gem_object *obj, struct drm_file *file_priv);
 void panfrost_gem_close(struct drm_gem_object *obj,
 			struct drm_file *file_priv);
 
+struct panfrost_gem_mapping *
+panfrost_gem_mapping_get(struct panfrost_gem_object *bo,
+			 struct panfrost_file_priv *priv);
+void panfrost_gem_mapping_put(struct panfrost_gem_mapping *mapping);
+void panfrost_gem_teardown_mappings(struct panfrost_gem_object *bo);
+
 void panfrost_gem_shrinker_init(struct drm_device *dev);
 void panfrost_gem_shrinker_cleanup(struct drm_device *dev);
 

drivers/gpu/drm/panfrost/panfrost_gem_shrinker.c

@@ -39,11 +39,12 @@ panfrost_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc
 static bool panfrost_gem_purge(struct drm_gem_object *obj)
 {
 	struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
+	struct panfrost_gem_object *bo = to_panfrost_bo(obj);
 
 	if (!mutex_trylock(&shmem->pages_lock))
 		return false;
 
-	panfrost_mmu_unmap(to_panfrost_bo(obj));
+	panfrost_gem_teardown_mappings(bo);
 	drm_gem_shmem_purge_locked(obj);
 
 	mutex_unlock(&shmem->pages_lock);

drivers/gpu/drm/panfrost/panfrost_job.c

@@ -268,9 +268,20 @@ static void panfrost_job_cleanup(struct kref *ref)
 	dma_fence_put(job->done_fence);
 	dma_fence_put(job->render_done_fence);
 
-	if (job->bos) {
+	if (job->mappings) {
 		for (i = 0; i < job->bo_count; i++)
+			panfrost_gem_mapping_put(job->mappings[i]);
+		kvfree(job->mappings);
+	}
+
+	if (job->bos) {
+		struct panfrost_gem_object *bo;
+
+		for (i = 0; i < job->bo_count; i++) {
+			bo = to_panfrost_bo(job->bos[i]);
 			drm_gem_object_put_unlocked(job->bos[i]);
+		}
+
 		kvfree(job->bos);
 	}
 

drivers/gpu/drm/panfrost/panfrost_job.h

@@ -32,6 +32,7 @@ struct panfrost_job {
 
 	/* Exclusive fences we have taken from the BOs to wait for */
 	struct dma_fence **implicit_fences;
+	struct panfrost_gem_mapping **mappings;
 	struct drm_gem_object **bos;
 	u32 bo_count;
 

drivers/gpu/drm/panfrost/panfrost_mmu.c

@@ -269,14 +269,15 @@ static int mmu_map_sg(struct panfrost_device *pfdev, struct panfrost_mmu *mmu,
 	return 0;
 }
 
-int panfrost_mmu_map(struct panfrost_gem_object *bo)
+int panfrost_mmu_map(struct panfrost_gem_mapping *mapping)
 {
+	struct panfrost_gem_object *bo = mapping->obj;
 	struct drm_gem_object *obj = &bo->base.base;
 	struct panfrost_device *pfdev = to_panfrost_device(obj->dev);
 	struct sg_table *sgt;
 	int prot = IOMMU_READ | IOMMU_WRITE;
 
-	if (WARN_ON(bo->is_mapped))
+	if (WARN_ON(mapping->active))
 		return 0;
 
 	if (bo->noexec)
@@ -286,25 +287,28 @@ int panfrost_mmu_map(struct panfrost_gem_object *bo)
 	if (WARN_ON(IS_ERR(sgt)))
 		return PTR_ERR(sgt);
 
-	mmu_map_sg(pfdev, bo->mmu, bo->node.start << PAGE_SHIFT, prot, sgt);
-	bo->is_mapped = true;
+	mmu_map_sg(pfdev, mapping->mmu, mapping->mmnode.start << PAGE_SHIFT,
+		   prot, sgt);
+	mapping->active = true;
 
 	return 0;
 }
 
-void panfrost_mmu_unmap(struct panfrost_gem_object *bo)
+void panfrost_mmu_unmap(struct panfrost_gem_mapping *mapping)
 {
+	struct panfrost_gem_object *bo = mapping->obj;
 	struct drm_gem_object *obj = &bo->base.base;
 	struct panfrost_device *pfdev = to_panfrost_device(obj->dev);
-	struct io_pgtable_ops *ops = bo->mmu->pgtbl_ops;
-	u64 iova = bo->node.start << PAGE_SHIFT;
-	size_t len = bo->node.size << PAGE_SHIFT;
+	struct io_pgtable_ops *ops = mapping->mmu->pgtbl_ops;
+	u64 iova = mapping->mmnode.start << PAGE_SHIFT;
+	size_t len = mapping->mmnode.size << PAGE_SHIFT;
 	size_t unmapped_len = 0;
 
-	if (WARN_ON(!bo->is_mapped))
+	if (WARN_ON(!mapping->active))
 		return;
 
-	dev_dbg(pfdev->dev, "unmap: as=%d, iova=%llx, len=%zx", bo->mmu->as, iova, len);
+	dev_dbg(pfdev->dev, "unmap: as=%d, iova=%llx, len=%zx",
+		mapping->mmu->as, iova, len);
 
 	while (unmapped_len < len) {
 		size_t unmapped_page;
@@ -318,8 +322,9 @@ void panfrost_mmu_unmap(struct panfrost_gem_object *bo)
 		unmapped_len += pgsize;
 	}
 
-	panfrost_mmu_flush_range(pfdev, bo->mmu, bo->node.start << PAGE_SHIFT, len);
-	bo->is_mapped = false;
+	panfrost_mmu_flush_range(pfdev, mapping->mmu,
+				 mapping->mmnode.start << PAGE_SHIFT, len);
+	mapping->active = false;
 }
 
 static void mmu_tlb_inv_context_s1(void *cookie)
@@ -394,10 +399,10 @@ void panfrost_mmu_pgtable_free(struct panfrost_file_priv *priv)
 	free_io_pgtable_ops(mmu->pgtbl_ops);
 }
 
-static struct panfrost_gem_object *
-addr_to_drm_mm_node(struct panfrost_device *pfdev, int as, u64 addr)
+static struct panfrost_gem_mapping *
+addr_to_mapping(struct panfrost_device *pfdev, int as, u64 addr)
 {
-	struct panfrost_gem_object *bo = NULL;
+	struct panfrost_gem_mapping *mapping = NULL;
 	struct panfrost_file_priv *priv;
 	struct drm_mm_node *node;
 	u64 offset = addr >> PAGE_SHIFT;
@@ -418,8 +423,9 @@ addr_to_drm_mm_node(struct panfrost_device *pfdev, int as, u64 addr)
 	drm_mm_for_each_node(node, &priv->mm) {
 		if (offset >= node->start &&
 		    offset < (node->start + node->size)) {
-			bo = drm_mm_node_to_panfrost_bo(node);
-			drm_gem_object_get(&bo->base.base);
+			mapping = drm_mm_node_to_panfrost_mapping(node);
+
+			kref_get(&mapping->refcount);
 			break;
 		}
 	}
@@ -427,7 +433,7 @@ addr_to_drm_mm_node(struct panfrost_device *pfdev, int as, u64 addr)
 	spin_unlock(&priv->mm_lock);
 out:
 	spin_unlock(&pfdev->as_lock);
-	return bo;
+	return mapping;
 }
 
 #define NUM_FAULT_PAGES (SZ_2M / PAGE_SIZE)
@@ -436,28 +442,30 @@ static int panfrost_mmu_map_fault_addr(struct panfrost_device *pfdev, int as,
 				       u64 addr)
 {
 	int ret, i;
+	struct panfrost_gem_mapping *bomapping;
 	struct panfrost_gem_object *bo;
 	struct address_space *mapping;
 	pgoff_t page_offset;
 	struct sg_table *sgt;
 	struct page **pages;
 
-	bo = addr_to_drm_mm_node(pfdev, as, addr);
-	if (!bo)
+	bomapping = addr_to_mapping(pfdev, as, addr);
+	if (!bomapping)
 		return -ENOENT;
 
+	bo = bomapping->obj;
 	if (!bo->is_heap) {
 		dev_WARN(pfdev->dev, "matching BO is not heap type (GPU VA = %llx)",
-			 bo->node.start << PAGE_SHIFT);
+			 bomapping->mmnode.start << PAGE_SHIFT);
 		ret = -EINVAL;
 		goto err_bo;
 	}
-	WARN_ON(bo->mmu->as != as);
+	WARN_ON(bomapping->mmu->as != as);
 
 	/* Assume 2MB alignment and size multiple */
 	addr &= ~((u64)SZ_2M - 1);
 	page_offset = addr >> PAGE_SHIFT;
-	page_offset -= bo->node.start;
+	page_offset -= bomapping->mmnode.start;
 
 	mutex_lock(&bo->base.pages_lock);
 
@@ -509,13 +517,14 @@ static int panfrost_mmu_map_fault_addr(struct panfrost_device *pfdev, int as,
 		goto err_map;
 	}
 
-	mmu_map_sg(pfdev, bo->mmu, addr, IOMMU_WRITE | IOMMU_READ | IOMMU_NOEXEC, sgt);
+	mmu_map_sg(pfdev, bomapping->mmu, addr,
+		   IOMMU_WRITE | IOMMU_READ | IOMMU_NOEXEC, sgt);
 
-	bo->is_mapped = true;
+	bomapping->active = true;
 
 	dev_dbg(pfdev->dev, "mapped page fault @ AS%d %llx", as, addr);
 
-	drm_gem_object_put_unlocked(&bo->base.base);
+	panfrost_gem_mapping_put(bomapping);
 
 	return 0;
 

drivers/gpu/drm/panfrost/panfrost_mmu.h

@@ -4,12 +4,12 @@
 #ifndef __PANFROST_MMU_H__
 #define __PANFROST_MMU_H__
 
-struct panfrost_gem_object;
+struct panfrost_gem_mapping;
 struct panfrost_file_priv;
 struct panfrost_mmu;
 
-int panfrost_mmu_map(struct panfrost_gem_object *bo);
-void panfrost_mmu_unmap(struct panfrost_gem_object *bo);
+int panfrost_mmu_map(struct panfrost_gem_mapping *mapping);
+void panfrost_mmu_unmap(struct panfrost_gem_mapping *mapping);
 
 int panfrost_mmu_init(struct panfrost_device *pfdev);
 void panfrost_mmu_fini(struct panfrost_device *pfdev);

drivers/gpu/drm/panfrost/panfrost_perfcnt.c

@@ -25,7 +25,7 @@
 #define V4_SHADERS_PER_COREGROUP	4
 
 struct panfrost_perfcnt {
-	struct panfrost_gem_object *bo;
+	struct panfrost_gem_mapping *mapping;
 	size_t bosize;
 	void *buf;
 	struct panfrost_file_priv *user;
@@ -49,7 +49,7 @@ static int panfrost_perfcnt_dump_locked(struct panfrost_device *pfdev)
 	int ret;
 
 	reinit_completion(&pfdev->perfcnt->dump_comp);
-	gpuva = pfdev->perfcnt->bo->node.start << PAGE_SHIFT;
+	gpuva = pfdev->perfcnt->mapping->mmnode.start << PAGE_SHIFT;
 	gpu_write(pfdev, GPU_PERFCNT_BASE_LO, gpuva);
 	gpu_write(pfdev, GPU_PERFCNT_BASE_HI, gpuva >> 32);
 	gpu_write(pfdev, GPU_INT_CLEAR,
@@ -89,17 +89,22 @@ static int panfrost_perfcnt_enable_locked(struct panfrost_device *pfdev,
 	if (IS_ERR(bo))
 		return PTR_ERR(bo);
 
-	perfcnt->bo = to_panfrost_bo(&bo->base);
-
 	/* Map the perfcnt buf in the address space attached to file_priv. */
-	ret = panfrost_gem_open(&perfcnt->bo->base.base, file_priv);
+	ret = panfrost_gem_open(&bo->base, file_priv);
 	if (ret)
 		goto err_put_bo;
 
+	perfcnt->mapping = panfrost_gem_mapping_get(to_panfrost_bo(&bo->base),
+						    user);
+	if (!perfcnt->mapping) {
+		ret = -EINVAL;
+		goto err_close_bo;
+	}
+
 	perfcnt->buf = drm_gem_shmem_vmap(&bo->base);
 	if (IS_ERR(perfcnt->buf)) {
 		ret = PTR_ERR(perfcnt->buf);
-		goto err_close_bo;
+		goto err_put_mapping;
 	}
 
 	/*
@@ -154,12 +159,17 @@ static int panfrost_perfcnt_enable_locked(struct panfrost_device *pfdev,
 	if (panfrost_has_hw_issue(pfdev, HW_ISSUE_8186))
 		gpu_write(pfdev, GPU_PRFCNT_TILER_EN, 0xffffffff);
 
+	/* The BO ref is retained by the mapping. */
+	drm_gem_object_put_unlocked(&bo->base);
+
 	return 0;
 
 err_vunmap:
-	drm_gem_shmem_vunmap(&perfcnt->bo->base.base, perfcnt->buf);
+	drm_gem_shmem_vunmap(&bo->base, perfcnt->buf);
+err_put_mapping:
+	panfrost_gem_mapping_put(perfcnt->mapping);
 err_close_bo:
-	panfrost_gem_close(&perfcnt->bo->base.base, file_priv);
+	panfrost_gem_close(&bo->base, file_priv);
 err_put_bo:
 	drm_gem_object_put_unlocked(&bo->base);
 	return ret;
@@ -182,11 +192,11 @@ static int panfrost_perfcnt_disable_locked(struct panfrost_device *pfdev,
 		  GPU_PERFCNT_CFG_MODE(GPU_PERFCNT_CFG_MODE_OFF));
 
 	perfcnt->user = NULL;
-	drm_gem_shmem_vunmap(&perfcnt->bo->base.base, perfcnt->buf);
+	drm_gem_shmem_vunmap(&perfcnt->mapping->obj->base.base, perfcnt->buf);
 	perfcnt->buf = NULL;
-	panfrost_gem_close(&perfcnt->bo->base.base, file_priv);
-	drm_gem_object_put_unlocked(&perfcnt->bo->base.base);
-	perfcnt->bo = NULL;
+	panfrost_gem_close(&perfcnt->mapping->obj->base.base, file_priv);
+	panfrost_gem_mapping_put(perfcnt->mapping);
+	perfcnt->mapping = NULL;
 	pm_runtime_mark_last_busy(pfdev->dev);
 	pm_runtime_put_autosuspend(pfdev->dev);