Merge branch 'drm-next-3.18' of git://people.freedesktop.org/~agd5f/linux into drm-next

radeon userptr support.

* 'drm-next-3.18' of git://people.freedesktop.org/~agd5f/linux:
  drm/radeon: allow userptr write access under certain conditions
  drm/radeon: add userptr flag to register MMU notifier v3
  drm/radeon: add userptr flag to directly validate the BO to GTT
  drm/radeon: add userptr flag to limit it to anonymous memory v2
  drm/radeon: add userptr support v8

Conflicts:
	drivers/gpu/drm/radeon/radeon_prime.c

drivers/gpu/drm/Kconfig

@@ -116,6 +116,7 @@ config DRM_RADEON
 	select HWMON
 	select BACKLIGHT_CLASS_DEVICE
 	select INTERVAL_TREE
+	select MMU_NOTIFIER
 	help
 	  Choose this option if you have an ATI Radeon graphics card.  There
 	  are both PCI and AGP versions.  You don't need to choose this to

drivers/gpu/drm/radeon/Makefile

@@ -80,7 +80,7 @@ radeon-y += radeon_device.o radeon_asic.o radeon_kms.o \
 	r600_dpm.o rs780_dpm.o rv6xx_dpm.o rv770_dpm.o rv730_dpm.o rv740_dpm.o \
 	rv770_smc.o cypress_dpm.o btc_dpm.o sumo_dpm.o sumo_smc.o trinity_dpm.o \
 	trinity_smc.o ni_dpm.o si_smc.o si_dpm.o kv_smc.o kv_dpm.o ci_smc.o \
-	ci_dpm.o dce6_afmt.o radeon_vm.o radeon_ucode.o radeon_ib.o
+	ci_dpm.o dce6_afmt.o radeon_vm.o radeon_ucode.o radeon_ib.o radeon_mn.o
 
 # add async DMA block
 radeon-y += \

drivers/gpu/drm/radeon/radeon.h

@@ -65,6 +65,7 @@
 #include <linux/list.h>
 #include <linux/kref.h>
 #include <linux/interval_tree.h>
+#include <linux/hashtable.h>
 
 #include <ttm/ttm_bo_api.h>
 #include <ttm/ttm_bo_driver.h>
@@ -488,6 +489,9 @@ struct radeon_bo {
 
 	struct ttm_bo_kmap_obj		dma_buf_vmap;
 	pid_t				pid;
+
+	struct radeon_mn		*mn;
+	struct interval_tree_node	mn_it;
 };
 #define gem_to_radeon_bo(gobj) container_of((gobj), struct radeon_bo, gem_base)
 
@@ -1728,6 +1732,11 @@ void radeon_test_ring_sync(struct radeon_device *rdev,
 			   struct radeon_ring *cpB);
 void radeon_test_syncing(struct radeon_device *rdev);
 
+/*
+ * MMU Notifier
+ */
+int radeon_mn_register(struct radeon_bo *bo, unsigned long addr);
+void radeon_mn_unregister(struct radeon_bo *bo);
 
 /*
  * Debugfs
@@ -2141,6 +2150,8 @@ int radeon_gem_info_ioctl(struct drm_device *dev, void *data,
 			  struct drm_file *filp);
 int radeon_gem_create_ioctl(struct drm_device *dev, void *data,
 			    struct drm_file *filp);
+int radeon_gem_userptr_ioctl(struct drm_device *dev, void *data,
+			     struct drm_file *filp);
 int radeon_gem_pin_ioctl(struct drm_device *dev, void *data,
 			 struct drm_file *file_priv);
 int radeon_gem_unpin_ioctl(struct drm_device *dev, void *data,
@@ -2373,6 +2384,9 @@ struct radeon_device {
 	/* tracking pinned memory */
 	u64 vram_pin_size;
 	u64 gart_pin_size;
+
+	struct mutex	mn_lock;
+	DECLARE_HASHTABLE(mn_hash, 7);
 };
 
 bool radeon_is_px(struct drm_device *dev);
@@ -2874,6 +2888,10 @@ extern void radeon_legacy_set_clock_gating(struct radeon_device *rdev, int enabl
 extern void radeon_atom_set_clock_gating(struct radeon_device *rdev, int enable);
 extern void radeon_ttm_placement_from_domain(struct radeon_bo *rbo, u32 domain);
 extern bool radeon_ttm_bo_is_radeon_bo(struct ttm_buffer_object *bo);
+extern int radeon_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
+				     uint32_t flags);
+extern bool radeon_ttm_tt_has_userptr(struct ttm_tt *ttm);
+extern bool radeon_ttm_tt_is_readonly(struct ttm_tt *ttm);
 extern void radeon_vram_location(struct radeon_device *rdev, struct radeon_mc *mc, u64 base);
 extern void radeon_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc);
 extern int radeon_resume_kms(struct drm_device *dev, bool resume, bool fbcon);

drivers/gpu/drm/radeon/radeon_cs.c

@@ -78,7 +78,8 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
 	struct radeon_cs_chunk *chunk;
 	struct radeon_cs_buckets buckets;
 	unsigned i, j;
-	bool duplicate;
+	bool duplicate, need_mmap_lock = false;
+	int r;
 
 	if (p->chunk_relocs_idx == -1) {
 		return 0;
@@ -165,6 +166,19 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
 			p->relocs[i].allowed_domains = domain;
 		}
 
+		if (radeon_ttm_tt_has_userptr(p->relocs[i].robj->tbo.ttm)) {
+			uint32_t domain = p->relocs[i].prefered_domains;
+			if (!(domain & RADEON_GEM_DOMAIN_GTT)) {
+				DRM_ERROR("Only RADEON_GEM_DOMAIN_GTT is "
+					  "allowed for userptr BOs\n");
+				return -EINVAL;
+			}
+			need_mmap_lock = true;
+			domain = RADEON_GEM_DOMAIN_GTT;
+			p->relocs[i].prefered_domains = domain;
+			p->relocs[i].allowed_domains = domain;
+		}
+
 		p->relocs[i].tv.bo = &p->relocs[i].robj->tbo;
 		p->relocs[i].handle = r->handle;
 
@@ -177,8 +191,15 @@ static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
 	if (p->cs_flags & RADEON_CS_USE_VM)
 		p->vm_bos = radeon_vm_get_bos(p->rdev, p->ib.vm,
 					      &p->validated);
+	if (need_mmap_lock)
+		down_read(&current->mm->mmap_sem);
+
+	r = radeon_bo_list_validate(p->rdev, &p->ticket, &p->validated, p->ring);
 
-	return radeon_bo_list_validate(p->rdev, &p->ticket, &p->validated, p->ring);
+	if (need_mmap_lock)
+		up_read(&current->mm->mmap_sem);
+
+	return r;
 }
 
 static int radeon_cs_get_ring(struct radeon_cs_parser *p, u32 ring, s32 priority)

drivers/gpu/drm/radeon/radeon_device.c

@@ -1270,6 +1270,8 @@ int radeon_device_init(struct radeon_device *rdev,
 	init_rwsem(&rdev->pm.mclk_lock);
 	init_rwsem(&rdev->exclusive_lock);
 	init_waitqueue_head(&rdev->irq.vblank_queue);
+	mutex_init(&rdev->mn_lock);
+	hash_init(rdev->mn_hash);
 	r = radeon_gem_init(rdev);
 	if (r)
 		return r;

drivers/gpu/drm/radeon/radeon_drv.c

@@ -114,6 +114,9 @@ int radeon_gem_object_open(struct drm_gem_object *obj,
 				struct drm_file *file_priv);
 void radeon_gem_object_close(struct drm_gem_object *obj,
 				struct drm_file *file_priv);
+struct dma_buf *radeon_gem_prime_export(struct drm_device *dev,
+					struct drm_gem_object *gobj,
+					int flags);
 extern int radeon_get_crtc_scanoutpos(struct drm_device *dev, int crtc,
 				      unsigned int flags,
 				      int *vpos, int *hpos, ktime_t *stime,
@@ -573,7 +576,7 @@ static struct drm_driver kms_driver = {
 
 	.prime_handle_to_fd = drm_gem_prime_handle_to_fd,
 	.prime_fd_to_handle = drm_gem_prime_fd_to_handle,
-	.gem_prime_export = drm_gem_prime_export,
+	.gem_prime_export = radeon_gem_prime_export,
 	.gem_prime_import = drm_gem_prime_import,
 	.gem_prime_pin = radeon_gem_prime_pin,
 	.gem_prime_unpin = radeon_gem_prime_unpin,

drivers/gpu/drm/radeon/radeon_gem.c

@@ -272,6 +272,94 @@ int radeon_gem_create_ioctl(struct drm_device *dev, void *data,
 	return 0;
 }
 
+int radeon_gem_userptr_ioctl(struct drm_device *dev, void *data,
+			     struct drm_file *filp)
+{
+	struct radeon_device *rdev = dev->dev_private;
+	struct drm_radeon_gem_userptr *args = data;
+	struct drm_gem_object *gobj;
+	struct radeon_bo *bo;
+	uint32_t handle;
+	int r;
+
+	if (offset_in_page(args->addr | args->size))
+		return -EINVAL;
+
+	/* reject unknown flag values */
+	if (args->flags & ~(RADEON_GEM_USERPTR_READONLY |
+	    RADEON_GEM_USERPTR_ANONONLY | RADEON_GEM_USERPTR_VALIDATE |
+	    RADEON_GEM_USERPTR_REGISTER))
+		return -EINVAL;
+
+	if (args->flags & RADEON_GEM_USERPTR_READONLY) {
+		/* readonly pages not tested on older hardware */
+		if (rdev->family < CHIP_R600)
+			return -EINVAL;
+
+	} else if (!(args->flags & RADEON_GEM_USERPTR_ANONONLY) ||
+		   !(args->flags & RADEON_GEM_USERPTR_REGISTER)) {
+
+		/* if we want to write to it we must require anonymous
+		   memory and install a MMU notifier */
+		return -EACCES;
+	}
+
+	down_read(&rdev->exclusive_lock);
+
+	/* create a gem object to contain this object in */
+	r = radeon_gem_object_create(rdev, args->size, 0,
+				     RADEON_GEM_DOMAIN_CPU, 0,
+				     false, &gobj);
+	if (r)
+		goto handle_lockup;
+
+	bo = gem_to_radeon_bo(gobj);
+	r = radeon_ttm_tt_set_userptr(bo->tbo.ttm, args->addr, args->flags);
+	if (r)
+		goto release_object;
+
+	if (args->flags & RADEON_GEM_USERPTR_REGISTER) {
+		r = radeon_mn_register(bo, args->addr);
+		if (r)
+			goto release_object;
+	}
+
+	if (args->flags & RADEON_GEM_USERPTR_VALIDATE) {
+		down_read(&current->mm->mmap_sem);
+		r = radeon_bo_reserve(bo, true);
+		if (r) {
+			up_read(&current->mm->mmap_sem);
+			goto release_object;
+		}
+
+		radeon_ttm_placement_from_domain(bo, RADEON_GEM_DOMAIN_GTT);
+		r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
+		radeon_bo_unreserve(bo);
+		up_read(&current->mm->mmap_sem);
+		if (r)
+			goto release_object;
+	}
+
+	r = drm_gem_handle_create(filp, gobj, &handle);
+	/* drop reference from allocate - handle holds it now */
+	drm_gem_object_unreference_unlocked(gobj);
+	if (r)
+		goto handle_lockup;
+
+	args->handle = handle;
+	up_read(&rdev->exclusive_lock);
+	return 0;
+
+release_object:
+	drm_gem_object_unreference_unlocked(gobj);
+
+handle_lockup:
+	up_read(&rdev->exclusive_lock);
+	r = radeon_gem_handle_lockup(rdev, r);
+
+	return r;
+}
+
 int radeon_gem_set_domain_ioctl(struct drm_device *dev, void *data,
 				struct drm_file *filp)
 {
@@ -315,6 +403,10 @@ int radeon_mode_dumb_mmap(struct drm_file *filp,
 		return -ENOENT;
 	}
 	robj = gem_to_radeon_bo(gobj);
+	if (radeon_ttm_tt_has_userptr(robj->tbo.ttm)) {
+		drm_gem_object_unreference_unlocked(gobj);
+		return -EPERM;
+	}
 	*offset_p = radeon_bo_mmap_offset(robj);
 	drm_gem_object_unreference_unlocked(gobj);
 	return 0;
@@ -532,6 +624,11 @@ int radeon_gem_op_ioctl(struct drm_device *dev, void *data,
 		return -ENOENT;
 	}
 	robj = gem_to_radeon_bo(gobj);
+
+	r = -EPERM;
+	if (radeon_ttm_tt_has_userptr(robj->tbo.ttm))
+		goto out;
+
 	r = radeon_bo_reserve(robj, false);
 	if (unlikely(r))
 		goto out;

drivers/gpu/drm/radeon/radeon_kms.c

@@ -885,5 +885,6 @@ const struct drm_ioctl_desc radeon_ioctls_kms[] = {
 	DRM_IOCTL_DEF_DRV(RADEON_GEM_BUSY, radeon_gem_busy_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(RADEON_GEM_VA, radeon_gem_va_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
 	DRM_IOCTL_DEF_DRV(RADEON_GEM_OP, radeon_gem_op_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(RADEON_GEM_USERPTR, radeon_gem_userptr_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW),
 };
 int radeon_max_kms_ioctl = ARRAY_SIZE(radeon_ioctls_kms);

drivers/gpu/drm/radeon/radeon_mn.c

+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ */
+/*
+ * Authors:
+ *    Christian König <christian.koenig@amd.com>
+ */
+
+#include <linux/firmware.h>
+#include <linux/module.h>
+#include <linux/mmu_notifier.h>
+#include <drm/drmP.h>
+#include <drm/drm.h>
+
+#include "radeon.h"
+
+struct radeon_mn {
+	/* constant after initialisation */
+	struct radeon_device	*rdev;
+	struct mm_struct	*mm;
+	struct mmu_notifier	mn;
+
+	/* only used on destruction */
+	struct work_struct	work;
+
+	/* protected by rdev->mn_lock */
+	struct hlist_node	node;
+
+	/* objects protected by lock */
+	struct mutex		lock;
+	struct rb_root		objects;
+};
+
+/**
+ * radeon_mn_destroy - destroy the rmn
+ *
+ * @work: previously sheduled work item
+ *
+ * Lazy destroys the notifier from a work item
+ */
+static void radeon_mn_destroy(struct work_struct *work)
+{
+	struct radeon_mn *rmn = container_of(work, struct radeon_mn, work);
+	struct radeon_device *rdev = rmn->rdev;
+	struct radeon_bo *bo, *next;
+
+	mutex_lock(&rdev->mn_lock);
+	mutex_lock(&rmn->lock);
+	hash_del(&rmn->node);
+	rbtree_postorder_for_each_entry_safe(bo, next, &rmn->objects, mn_it.rb) {
+		interval_tree_remove(&bo->mn_it, &rmn->objects);
+		bo->mn = NULL;
+	}
+	mutex_unlock(&rmn->lock);
+	mutex_unlock(&rdev->mn_lock);
+	mmu_notifier_unregister(&rmn->mn, rmn->mm);
+	kfree(rmn);
+}
+
+/**
+ * radeon_mn_release - callback to notify about mm destruction
+ *
+ * @mn: our notifier
+ * @mn: the mm this callback is about
+ *
+ * Shedule a work item to lazy destroy our notifier.
+ */
+static void radeon_mn_release(struct mmu_notifier *mn,
+			      struct mm_struct *mm)
+{
+	struct radeon_mn *rmn = container_of(mn, struct radeon_mn, mn);
+	INIT_WORK(&rmn->work, radeon_mn_destroy);
+	schedule_work(&rmn->work);
+}
+
+/**
+ * radeon_mn_invalidate_range_start - callback to notify about mm change
+ *
+ * @mn: our notifier
+ * @mn: the mm this callback is about
+ * @start: start of updated range
+ * @end: end of updated range
+ *
+ * We block for all BOs between start and end to be idle and
+ * unmap them by move them into system domain again.
+ */
+static void radeon_mn_invalidate_range_start(struct mmu_notifier *mn,
+					     struct mm_struct *mm,
+					     unsigned long start,
+					     unsigned long end)
+{
+	struct radeon_mn *rmn = container_of(mn, struct radeon_mn, mn);
+	struct interval_tree_node *it;
+
+	/* notification is exclusive, but interval is inclusive */
+	end -= 1;
+
+	mutex_lock(&rmn->lock);
+
+	it = interval_tree_iter_first(&rmn->objects, start, end);
+	while (it) {
+		struct radeon_bo *bo;
+		int r;
+
+		bo = container_of(it, struct radeon_bo, mn_it);
+		it = interval_tree_iter_next(it, start, end);
+
+		r = radeon_bo_reserve(bo, true);
+		if (r) {
+			DRM_ERROR("(%d) failed to reserve user bo\n", r);
+			continue;
+		}
+
+		if (bo->tbo.sync_obj) {
+			r = radeon_fence_wait(bo->tbo.sync_obj, false);
+			if (r)
+				DRM_ERROR("(%d) failed to wait for user bo\n", r);
+		}
+
+		radeon_ttm_placement_from_domain(bo, RADEON_GEM_DOMAIN_CPU);
+		r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
+		if (r)
+			DRM_ERROR("(%d) failed to validate user bo\n", r);
+
+		radeon_bo_unreserve(bo);
+	}
+	
+	mutex_unlock(&rmn->lock);
+}
+
+static const struct mmu_notifier_ops radeon_mn_ops = {
+	.release = radeon_mn_release,
+	.invalidate_range_start = radeon_mn_invalidate_range_start,
+};
+
+/**
+ * radeon_mn_get - create notifier context
+ *
+ * @rdev: radeon device pointer
+ *
+ * Creates a notifier context for current->mm.
+ */
+static struct radeon_mn *radeon_mn_get(struct radeon_device *rdev)
+{
+	struct mm_struct *mm = current->mm;
+	struct radeon_mn *rmn;
+	int r;
+
+	down_write(&mm->mmap_sem);
+	mutex_lock(&rdev->mn_lock);
+
+	hash_for_each_possible(rdev->mn_hash, rmn, node, (unsigned long)mm)
+		if (rmn->mm == mm)
+			goto release_locks;
+
+	rmn = kzalloc(sizeof(*rmn), GFP_KERNEL);
+	if (!rmn) {
+		rmn = ERR_PTR(-ENOMEM);
+		goto release_locks;
+	}
+
+	rmn->rdev = rdev;
+	rmn->mm = mm;
+	rmn->mn.ops = &radeon_mn_ops;
+	mutex_init(&rmn->lock);
+	rmn->objects = RB_ROOT;
+	
+	r = __mmu_notifier_register(&rmn->mn, mm);
+	if (r)
+		goto free_rmn;
+
+	hash_add(rdev->mn_hash, &rmn->node, (unsigned long)mm);
+
+release_locks:
+	mutex_unlock(&rdev->mn_lock);
+	up_write(&mm->mmap_sem);
+
+	return rmn;
+
+free_rmn:
+	mutex_unlock(&rdev->mn_lock);
+	up_write(&mm->mmap_sem);
+	kfree(rmn);
+
+	return ERR_PTR(r);
+}
+
+/**
+ * radeon_mn_register - register a BO for notifier updates
+ *
+ * @bo: radeon buffer object
+ * @addr: userptr addr we should monitor
+ *
+ * Registers an MMU notifier for the given BO at the specified address.
+ * Returns 0 on success, -ERRNO if anything goes wrong.
+ */
+int radeon_mn_register(struct radeon_bo *bo, unsigned long addr)
+{
+	unsigned long end = addr + radeon_bo_size(bo) - 1;
+	struct radeon_device *rdev = bo->rdev;
+	struct radeon_mn *rmn;
+	struct interval_tree_node *it;
+
+	rmn = radeon_mn_get(rdev);
+	if (IS_ERR(rmn))
+		return PTR_ERR(rmn);
+
+	mutex_lock(&rmn->lock);
+
+	it = interval_tree_iter_first(&rmn->objects, addr, end);
+	if (it) {
+		mutex_unlock(&rmn->lock);
+		return -EEXIST;
+	}
+
+	bo->mn = rmn;
+	bo->mn_it.start = addr;
+	bo->mn_it.last = end;
+	interval_tree_insert(&bo->mn_it, &rmn->objects);
+
+	mutex_unlock(&rmn->lock);
+
+	return 0;
+}
+
+/**
+ * radeon_mn_unregister - unregister a BO for notifier updates
+ *
+ * @bo: radeon buffer object
+ *
+ * Remove any registration of MMU notifier updates from the buffer object.
+ */
+void radeon_mn_unregister(struct radeon_bo *bo)
+{
+	struct radeon_device *rdev = bo->rdev;
+	struct radeon_mn *rmn;
+
+	mutex_lock(&rdev->mn_lock);
+	rmn = bo->mn;
+	if (rmn == NULL) {
+		mutex_unlock(&rdev->mn_lock);
+		return;
+	}
+
+	mutex_lock(&rmn->lock);
+	interval_tree_remove(&bo->mn_it, &rmn->objects);
+	bo->mn = NULL;
+	mutex_unlock(&rmn->lock);
+	mutex_unlock(&rdev->mn_lock);
+}

drivers/gpu/drm/radeon/radeon_object.c

@@ -75,6 +75,7 @@ static void radeon_ttm_bo_destroy(struct ttm_buffer_object *tbo)
 	bo = container_of(tbo, struct radeon_bo, tbo);
 
 	radeon_update_memory_usage(bo, bo->tbo.mem.mem_type, -1);
+	radeon_mn_unregister(bo);
 
 	mutex_lock(&bo->rdev->gem.mutex);
 	list_del_init(&bo->list);
@@ -264,6 +265,9 @@ int radeon_bo_pin_restricted(struct radeon_bo *bo, u32 domain, u64 max_offset,
 {
 	int r, i;
 
+	if (radeon_ttm_tt_has_userptr(bo->tbo.ttm))
+		return -EPERM;
+
 	if (bo->pin_count) {
 		bo->pin_count++;
 		if (gpu_addr)

drivers/gpu/drm/radeon/radeon_prime.c

@@ -111,3 +111,13 @@ struct reservation_object *radeon_gem_prime_res_obj(struct drm_gem_object *obj)
 
 	return bo->tbo.resv;
 }
+
+struct dma_buf *radeon_gem_prime_export(struct drm_device *dev,
+					struct drm_gem_object *gobj,
+					int flags)
+{
+	struct radeon_bo *bo = gem_to_radeon_bo(gobj);
+	if (radeon_ttm_tt_has_userptr(bo->tbo.ttm))
+		return ERR_PTR(-EPERM);
+	return drm_gem_prime_export(dev, gobj, flags);
+}

drivers/gpu/drm/radeon/radeon_ttm.c

@@ -39,6 +39,8 @@
 #include <linux/seq_file.h>
 #include <linux/slab.h>
 #include <linux/swiotlb.h>
+#include <linux/swap.h>
+#include <linux/pagemap.h>
 #include <linux/debugfs.h>
 #include "radeon_reg.h"
 #include "radeon.h"
@@ -515,8 +517,102 @@ struct radeon_ttm_tt {
 	struct ttm_dma_tt		ttm;
 	struct radeon_device		*rdev;
 	u64				offset;
+
+	uint64_t			userptr;
+	struct mm_struct		*usermm;
+	uint32_t			userflags;
 };
 
+/* prepare the sg table with the user pages */
+static int radeon_ttm_tt_pin_userptr(struct ttm_tt *ttm)
+{
+	struct radeon_device *rdev = radeon_get_rdev(ttm->bdev);
+	struct radeon_ttm_tt *gtt = (void *)ttm;
+	unsigned pinned = 0, nents;
+	int r;
+
+	int write = !(gtt->userflags & RADEON_GEM_USERPTR_READONLY);
+	enum dma_data_direction direction = write ?
+		DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
+
+	if (current->mm != gtt->usermm)
+		return -EPERM;
+
+	if (gtt->userflags & RADEON_GEM_USERPTR_ANONONLY) {
+		/* check that we only pin down anonymous memory
+		   to prevent problems with writeback */
+		unsigned long end = gtt->userptr + ttm->num_pages * PAGE_SIZE;
+		struct vm_area_struct *vma;
+		vma = find_vma(gtt->usermm, gtt->userptr);
+		if (!vma || vma->vm_file || vma->vm_end < end)
+			return -EPERM;
+	}
+
+	do {
+		unsigned num_pages = ttm->num_pages - pinned;
+		uint64_t userptr = gtt->userptr + pinned * PAGE_SIZE;
+		struct page **pages = ttm->pages + pinned;
+
+		r = get_user_pages(current, current->mm, userptr, num_pages,
+				   write, 0, pages, NULL);
+		if (r < 0)
+			goto release_pages;
+
+		pinned += r;
+
+	} while (pinned < ttm->num_pages);
+
+	r = sg_alloc_table_from_pages(ttm->sg, ttm->pages, ttm->num_pages, 0,
+				      ttm->num_pages << PAGE_SHIFT,
+				      GFP_KERNEL);
+	if (r)
+		goto release_sg;
+
+	r = -ENOMEM;
+	nents = dma_map_sg(rdev->dev, ttm->sg->sgl, ttm->sg->nents, direction);
+	if (nents != ttm->sg->nents)
+		goto release_sg;
+
+	drm_prime_sg_to_page_addr_arrays(ttm->sg, ttm->pages,
+					 gtt->ttm.dma_address, ttm->num_pages);
+
+	return 0;
+
+release_sg:
+	kfree(ttm->sg);
+
+release_pages:
+	release_pages(ttm->pages, pinned, 0);
+	return r;
+}
+
+static void radeon_ttm_tt_unpin_userptr(struct ttm_tt *ttm)
+{
+	struct radeon_device *rdev = radeon_get_rdev(ttm->bdev);
+	struct radeon_ttm_tt *gtt = (void *)ttm;
+	struct scatterlist *sg;
+	int i;
+
+	int write = !(gtt->userflags & RADEON_GEM_USERPTR_READONLY);
+	enum dma_data_direction direction = write ?
+		DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
+
+	/* free the sg table and pages again */
+	dma_unmap_sg(rdev->dev, ttm->sg->sgl, ttm->sg->nents, direction);
+
+	for_each_sg(ttm->sg->sgl, sg, ttm->sg->nents, i) {
+		struct page *page = sg_page(sg);
+
+		if (!(gtt->userflags & RADEON_GEM_USERPTR_READONLY))
+			set_page_dirty(page);
+
+		mark_page_accessed(page);
+		page_cache_release(page);
+	}
+
+	sg_free_table(ttm->sg);
+}
+
 static int radeon_ttm_backend_bind(struct ttm_tt *ttm,
 				   struct ttm_mem_reg *bo_mem)
 {
@@ -525,6 +621,11 @@ static int radeon_ttm_backend_bind(struct ttm_tt *ttm,
 		RADEON_GART_PAGE_WRITE;
 	int r;
 
+	if (gtt->userptr) {
+		radeon_ttm_tt_pin_userptr(ttm);
+		flags &= ~RADEON_GART_PAGE_WRITE;
+	}
+
 	gtt->offset = (unsigned long)(bo_mem->start << PAGE_SHIFT);
 	if (!ttm->num_pages) {
 		WARN(1, "nothing to bind %lu pages for mreg %p back %p!\n",
@@ -547,6 +648,10 @@ static int radeon_ttm_backend_unbind(struct ttm_tt *ttm)
 	struct radeon_ttm_tt *gtt = (void *)ttm;
 
 	radeon_gart_unbind(gtt->rdev, gtt->offset, ttm->num_pages);
+
+	if (gtt->userptr)
+		radeon_ttm_tt_unpin_userptr(ttm);
+
 	return 0;
 }
 
@@ -603,6 +708,16 @@ static int radeon_ttm_tt_populate(struct ttm_tt *ttm)
 	if (ttm->state != tt_unpopulated)
 		return 0;
 
+	if (gtt->userptr) {
+		ttm->sg = kcalloc(1, sizeof(struct sg_table), GFP_KERNEL);
+		if (!ttm->sg)
+			return -ENOMEM;
+
+		ttm->page_flags |= TTM_PAGE_FLAG_SG;
+		ttm->state = tt_unbound;
+		return 0;
+	}
+
 	if (slave && ttm->sg) {
 		drm_prime_sg_to_page_addr_arrays(ttm->sg, ttm->pages,
 						 gtt->ttm.dma_address, ttm->num_pages);
@@ -652,6 +767,12 @@ static void radeon_ttm_tt_unpopulate(struct ttm_tt *ttm)
 	unsigned i;
 	bool slave = !!(ttm->page_flags & TTM_PAGE_FLAG_SG);
 
+	if (gtt->userptr) {
+		kfree(ttm->sg);
+		ttm->page_flags &= ~TTM_PAGE_FLAG_SG;
+		return;
+	}
+
 	if (slave)
 		return;
 
@@ -680,6 +801,40 @@ static void radeon_ttm_tt_unpopulate(struct ttm_tt *ttm)
 	ttm_pool_unpopulate(ttm);
 }
 
+int radeon_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
+			      uint32_t flags)
+{
+	struct radeon_ttm_tt *gtt = (void *)ttm;
+
+	if (gtt == NULL)
+		return -EINVAL;
+
+	gtt->userptr = addr;
+	gtt->usermm = current->mm;
+	gtt->userflags = flags;
+	return 0;
+}
+
+bool radeon_ttm_tt_has_userptr(struct ttm_tt *ttm)
+{
+	struct radeon_ttm_tt *gtt = (void *)ttm;
+
+	if (gtt == NULL)
+		return false;
+
+	return !!gtt->userptr;
+}
+
+bool radeon_ttm_tt_is_readonly(struct ttm_tt *ttm)
+{
+	struct radeon_ttm_tt *gtt = (void *)ttm;
+
+	if (gtt == NULL)
+		return false;
+
+	return !!(gtt->userflags & RADEON_GEM_USERPTR_READONLY);
+}
+
 static struct ttm_bo_driver radeon_bo_driver = {
 	.ttm_tt_create = &radeon_ttm_tt_create,
 	.ttm_tt_populate = &radeon_ttm_tt_populate,

drivers/gpu/drm/radeon/radeon_vm.c

@@ -892,6 +892,9 @@ int radeon_vm_bo_update(struct radeon_device *rdev,
 	bo_va->flags &= ~RADEON_VM_PAGE_VALID;
 	bo_va->flags &= ~RADEON_VM_PAGE_SYSTEM;
 	bo_va->flags &= ~RADEON_VM_PAGE_SNOOPED;
+	if (bo_va->bo && radeon_ttm_tt_is_readonly(bo_va->bo->tbo.ttm))
+		bo_va->flags &= ~RADEON_VM_PAGE_WRITEABLE;
+
 	if (mem) {
 		addr = mem->start << PAGE_SHIFT;
 		if (mem->mem_type != TTM_PL_SYSTEM) {

include/uapi/drm/radeon_drm.h

@@ -511,6 +511,7 @@ typedef struct {
 #define DRM_RADEON_GEM_BUSY		0x2a
 #define DRM_RADEON_GEM_VA		0x2b
 #define DRM_RADEON_GEM_OP		0x2c
+#define DRM_RADEON_GEM_USERPTR		0x2d
 
 #define DRM_IOCTL_RADEON_CP_INIT    DRM_IOW( DRM_COMMAND_BASE + DRM_RADEON_CP_INIT, drm_radeon_init_t)
 #define DRM_IOCTL_RADEON_CP_START   DRM_IO(  DRM_COMMAND_BASE + DRM_RADEON_CP_START)
@@ -554,6 +555,7 @@ typedef struct {
 #define DRM_IOCTL_RADEON_GEM_BUSY	DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_BUSY, struct drm_radeon_gem_busy)
 #define DRM_IOCTL_RADEON_GEM_VA		DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_VA, struct drm_radeon_gem_va)
 #define DRM_IOCTL_RADEON_GEM_OP		DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_OP, struct drm_radeon_gem_op)
+#define DRM_IOCTL_RADEON_GEM_USERPTR	DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_USERPTR, struct drm_radeon_gem_userptr)
 
 typedef struct drm_radeon_init {
 	enum {
@@ -808,6 +810,23 @@ struct drm_radeon_gem_create {
 	uint32_t	flags;
 };
 
+/*
+ * This is not a reliable API and you should expect it to fail for any
+ * number of reasons and have fallback path that do not use userptr to
+ * perform any operation.
+ */
+#define RADEON_GEM_USERPTR_READONLY	(1 << 0)
+#define RADEON_GEM_USERPTR_ANONONLY	(1 << 1)
+#define RADEON_GEM_USERPTR_VALIDATE	(1 << 2)
+#define RADEON_GEM_USERPTR_REGISTER	(1 << 3)
+
+struct drm_radeon_gem_userptr {
+	uint64_t		addr;
+	uint64_t		size;
+	uint32_t		flags;
+	uint32_t		handle;
+};
+
 #define RADEON_TILING_MACRO				0x1
 #define RADEON_TILING_MICRO				0x2
 #define RADEON_TILING_SWAP_16BIT			0x4