drm/radeon: add userptr support v8

This patch adds an IOCTL for turning a pointer supplied by
userspace into a buffer object.

It imposes several restrictions upon the memory being mapped:

1. It must be page aligned (both start/end addresses, i.e ptr and size).

2. It must be normal system memory, not a pointer into another map of IO
space (e.g. it must not be a GTT mmapping of another object).

3. The BO is mapped into GTT, so the maximum amount of memory mapped at
all times is still the GTT limit.

4. The BO is only mapped readonly for now, so no write support.

5. List of backing pages is only acquired once, so they represent a
snapshot of the first use.

Exporting and sharing as well as mapping of buffer objects created by
this function is forbidden and results in an -EPERM.

v2: squash all previous changes into first public version
v3: fix tabs, map readonly, don't use MM callback any more
v4: set TTM_PAGE_FLAG_SG so that TTM never messes with the pages,
    pin/unpin pages on bind/unbind instead of populate/unpopulate
v5: rebased on 3.17-wip, IOCTL renamed to userptr, reject any unknown
    flags, better handle READONLY flag, improve permission check
v6: fix ptr cast warning, use set_page_dirty/mark_page_accessed on unpin
v7: add warning about it's availability in the API definition
v8: drop access_ok check, fix VM mapping bits
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v4)
Reviewed-by: Jérôme Glisse <jglisse@redhat.com> (v4)
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/radeon/radeon.h

@@ -2138,6 +2138,8 @@ 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,
@@ -2871,6 +2873,10 @@ 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;
@@ -164,6 +165,19 @@ 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;
 
@@ -176,8 +190,15 @@ 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_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,
@@ -568,7 +571,7 @@ 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 +272,65 @@ 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;
+
+	/* we only support read only mappings for now */
+	if (!(args->flags & RADEON_GEM_USERPTR_READONLY))
+		return -EACCES;
+
+	/* reject unknown flag values */
+	if (args->flags & ~RADEON_GEM_USERPTR_READONLY)
+		return -EINVAL;
+
+	/* readonly pages not tested on older hardware */
+	if (rdev->family < CHIP_R600)
+		return -EINVAL;
+
+	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;
+
+	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 +374,10 @@ 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 +595,11 @@ 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_object.c

@@ -264,6 +264,9 @@ 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

@@ -103,3 +103,13 @@ void radeon_gem_prime_unpin(struct drm_gem_object *obj)
 	radeon_bo_unpin(bo);
 	radeon_bo_unreserve(bo);
 }
+
+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 +517,92 @@ 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;
+
+	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 +611,11 @@ 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 +638,10 @@ 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 +698,16 @@ 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 +757,12 @@ 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 +791,40 @@ 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

@@ -888,6 +888,9 @@ 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 +810,20 @@ 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)
+
+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