xref: /linux/drivers/gpu/drm/v3d/v3d_bo.c (revision 22c55fb9eb92395d999b8404d73e58540d11bdd8)
1 // SPDX-License-Identifier: GPL-2.0+
2 /* Copyright (C) 2015-2018 Broadcom */
3 
4 /**
5  * DOC: V3D GEM BO management support
6  *
7  * Compared to VC4 (V3D 2.x), V3D 3.3 introduces an MMU between the
8  * GPU and the bus, allowing us to use shmem objects for our storage
9  * instead of CMA.
10  *
11  * Physically contiguous objects may still be imported to V3D, but the
12  * driver doesn't allocate physically contiguous objects on its own.
13  * Display engines requiring physically contiguous allocations should
14  * look into Mesa's "renderonly" support (as used by the Mesa pl111
15  * driver) for an example of how to integrate with V3D.
16  */
17 
18 #include <linux/dma-buf.h>
19 #include <linux/vmalloc.h>
20 
21 #include "v3d_drv.h"
22 #include "uapi/drm/v3d_drm.h"
23 
24 static enum drm_gem_object_status v3d_gem_status(struct drm_gem_object *obj)
25 {
26 	struct v3d_bo *bo = to_v3d_bo(obj);
27 	enum drm_gem_object_status res = 0;
28 
29 	if (bo->base.pages)
30 		res |= DRM_GEM_OBJECT_RESIDENT;
31 
32 	return res;
33 }
34 
35 /* Called DRM core on the last userspace/kernel unreference of the
36  * BO.
37  */
38 void v3d_free_object(struct drm_gem_object *obj)
39 {
40 	struct v3d_dev *v3d = to_v3d_dev(obj->dev);
41 	struct v3d_bo *bo = to_v3d_bo(obj);
42 
43 	if (bo->vaddr)
44 		v3d_put_bo_vaddr(bo);
45 
46 	v3d_mmu_remove_ptes(bo);
47 
48 	mutex_lock(&v3d->bo_lock);
49 	v3d->bo_stats.num_allocated--;
50 	v3d->bo_stats.pages_allocated -= obj->size >> V3D_MMU_PAGE_SHIFT;
51 	mutex_unlock(&v3d->bo_lock);
52 
53 	spin_lock(&v3d->mm_lock);
54 	drm_mm_remove_node(&bo->node);
55 	spin_unlock(&v3d->mm_lock);
56 
57 	/* GPU execution may have dirtied any pages in the BO. */
58 	bo->base.pages_mark_dirty_on_put = true;
59 
60 	drm_gem_shmem_free(&bo->base);
61 }
62 
63 static const struct drm_gem_object_funcs v3d_gem_funcs = {
64 	.free = v3d_free_object,
65 	.print_info = drm_gem_shmem_object_print_info,
66 	.pin = drm_gem_shmem_object_pin,
67 	.unpin = drm_gem_shmem_object_unpin,
68 	.get_sg_table = drm_gem_shmem_object_get_sg_table,
69 	.vmap = drm_gem_shmem_object_vmap,
70 	.vunmap = drm_gem_shmem_object_vunmap,
71 	.mmap = drm_gem_shmem_object_mmap,
72 	.status = v3d_gem_status,
73 	.vm_ops = &drm_gem_shmem_vm_ops,
74 };
75 
76 /* gem_create_object function for allocating a BO struct and doing
77  * early setup.
78  */
79 struct drm_gem_object *v3d_create_object(struct drm_device *dev, size_t size)
80 {
81 	struct v3d_bo *bo;
82 	struct drm_gem_object *obj;
83 
84 	if (size == 0)
85 		return ERR_PTR(-EINVAL);
86 
87 	bo = kzalloc(sizeof(*bo), GFP_KERNEL);
88 	if (!bo)
89 		return ERR_PTR(-ENOMEM);
90 	obj = &bo->base.base;
91 
92 	obj->funcs = &v3d_gem_funcs;
93 	bo->base.map_wc = true;
94 	INIT_LIST_HEAD(&bo->unref_head);
95 
96 	return &bo->base.base;
97 }
98 
99 static int
100 v3d_bo_create_finish(struct drm_gem_object *obj)
101 {
102 	struct v3d_dev *v3d = to_v3d_dev(obj->dev);
103 	struct v3d_bo *bo = to_v3d_bo(obj);
104 	struct sg_table *sgt;
105 	u64 align;
106 	int ret;
107 
108 	/* So far we pin the BO in the MMU for its lifetime, so use
109 	 * shmem's helper for getting a lifetime sgt.
110 	 */
111 	sgt = drm_gem_shmem_get_pages_sgt(&bo->base);
112 	if (IS_ERR(sgt))
113 		return PTR_ERR(sgt);
114 
115 	if (!v3d->gemfs)
116 		align = SZ_4K;
117 	else if (obj->size >= SZ_1M)
118 		align = SZ_1M;
119 	else if (obj->size >= SZ_64K)
120 		align = SZ_64K;
121 	else
122 		align = SZ_4K;
123 
124 	spin_lock(&v3d->mm_lock);
125 	/* Allocate the object's space in the GPU's page tables.
126 	 * Inserting PTEs will happen later, but the offset is for the
127 	 * lifetime of the BO.
128 	 */
129 	ret = drm_mm_insert_node_generic(&v3d->mm, &bo->node,
130 					 obj->size >> V3D_MMU_PAGE_SHIFT,
131 					 align >> V3D_MMU_PAGE_SHIFT, 0, 0);
132 	spin_unlock(&v3d->mm_lock);
133 	if (ret)
134 		return ret;
135 
136 	/* Track stats for /debug/dri/n/bo_stats. */
137 	mutex_lock(&v3d->bo_lock);
138 	v3d->bo_stats.num_allocated++;
139 	v3d->bo_stats.pages_allocated += obj->size >> V3D_MMU_PAGE_SHIFT;
140 	mutex_unlock(&v3d->bo_lock);
141 
142 	v3d_mmu_insert_ptes(bo);
143 
144 	return 0;
145 }
146 
147 struct v3d_bo *v3d_bo_create(struct drm_device *dev, struct drm_file *file_priv,
148 			     size_t unaligned_size)
149 {
150 	struct drm_gem_shmem_object *shmem_obj;
151 	struct v3d_dev *v3d = to_v3d_dev(dev);
152 	struct v3d_bo *bo;
153 	int ret;
154 
155 	shmem_obj = drm_gem_shmem_create_with_mnt(dev, unaligned_size,
156 						  v3d->gemfs);
157 	if (IS_ERR(shmem_obj))
158 		return ERR_CAST(shmem_obj);
159 	bo = to_v3d_bo(&shmem_obj->base);
160 	bo->vaddr = NULL;
161 
162 	ret = v3d_bo_create_finish(&shmem_obj->base);
163 	if (ret)
164 		goto free_obj;
165 
166 	return bo;
167 
168 free_obj:
169 	drm_gem_shmem_free(shmem_obj);
170 	return ERR_PTR(ret);
171 }
172 
173 struct drm_gem_object *
174 v3d_prime_import_sg_table(struct drm_device *dev,
175 			  struct dma_buf_attachment *attach,
176 			  struct sg_table *sgt)
177 {
178 	struct drm_gem_object *obj;
179 	int ret;
180 
181 	obj = drm_gem_shmem_prime_import_sg_table(dev, attach, sgt);
182 	if (IS_ERR(obj))
183 		return obj;
184 
185 	ret = v3d_bo_create_finish(obj);
186 	if (ret) {
187 		drm_gem_shmem_free(&to_v3d_bo(obj)->base);
188 		return ERR_PTR(ret);
189 	}
190 
191 	return obj;
192 }
193 
194 void v3d_get_bo_vaddr(struct v3d_bo *bo)
195 {
196 	struct drm_gem_shmem_object *obj = &bo->base;
197 
198 	bo->vaddr = vmap(obj->pages, obj->base.size >> PAGE_SHIFT, VM_MAP,
199 			 pgprot_writecombine(PAGE_KERNEL));
200 }
201 
202 void v3d_put_bo_vaddr(struct v3d_bo *bo)
203 {
204 	vunmap(bo->vaddr);
205 	bo->vaddr = NULL;
206 }
207 
208 int v3d_create_bo_ioctl(struct drm_device *dev, void *data,
209 			struct drm_file *file_priv)
210 {
211 	struct drm_v3d_create_bo *args = data;
212 	struct v3d_bo *bo = NULL;
213 	int ret;
214 
215 	if (args->flags != 0) {
216 		DRM_INFO("unknown create_bo flags: %d\n", args->flags);
217 		return -EINVAL;
218 	}
219 
220 	bo = v3d_bo_create(dev, file_priv, PAGE_ALIGN(args->size));
221 	if (IS_ERR(bo))
222 		return PTR_ERR(bo);
223 
224 	args->offset = bo->node.start << V3D_MMU_PAGE_SHIFT;
225 
226 	ret = drm_gem_handle_create(file_priv, &bo->base.base, &args->handle);
227 	drm_gem_object_put(&bo->base.base);
228 
229 	return ret;
230 }
231 
232 int v3d_mmap_bo_ioctl(struct drm_device *dev, void *data,
233 		      struct drm_file *file_priv)
234 {
235 	struct drm_v3d_mmap_bo *args = data;
236 	struct drm_gem_object *gem_obj;
237 
238 	if (args->flags != 0) {
239 		DRM_INFO("unknown mmap_bo flags: %d\n", args->flags);
240 		return -EINVAL;
241 	}
242 
243 	gem_obj = drm_gem_object_lookup(file_priv, args->handle);
244 	if (!gem_obj) {
245 		DRM_DEBUG("Failed to look up GEM BO %d\n", args->handle);
246 		return -ENOENT;
247 	}
248 
249 	args->offset = drm_vma_node_offset_addr(&gem_obj->vma_node);
250 	drm_gem_object_put(gem_obj);
251 
252 	return 0;
253 }
254 
255 int v3d_get_bo_offset_ioctl(struct drm_device *dev, void *data,
256 			    struct drm_file *file_priv)
257 {
258 	struct drm_v3d_get_bo_offset *args = data;
259 	struct drm_gem_object *gem_obj;
260 	struct v3d_bo *bo;
261 
262 	gem_obj = drm_gem_object_lookup(file_priv, args->handle);
263 	if (!gem_obj) {
264 		DRM_DEBUG("Failed to look up GEM BO %d\n", args->handle);
265 		return -ENOENT;
266 	}
267 	bo = to_v3d_bo(gem_obj);
268 
269 	args->offset = bo->node.start << V3D_MMU_PAGE_SHIFT;
270 
271 	drm_gem_object_put(gem_obj);
272 	return 0;
273 }
274 
275 int
276 v3d_wait_bo_ioctl(struct drm_device *dev, void *data,
277 		  struct drm_file *file_priv)
278 {
279 	int ret;
280 	struct drm_v3d_wait_bo *args = data;
281 	ktime_t start = ktime_get();
282 	u64 delta_ns;
283 	unsigned long timeout_jiffies =
284 		nsecs_to_jiffies_timeout(args->timeout_ns);
285 
286 	if (args->pad != 0)
287 		return -EINVAL;
288 
289 	ret = drm_gem_dma_resv_wait(file_priv, args->handle,
290 				    true, timeout_jiffies);
291 
292 	/* Decrement the user's timeout, in case we got interrupted
293 	 * such that the ioctl will be restarted.
294 	 */
295 	delta_ns = ktime_to_ns(ktime_sub(ktime_get(), start));
296 	if (delta_ns < args->timeout_ns)
297 		args->timeout_ns -= delta_ns;
298 	else
299 		args->timeout_ns = 0;
300 
301 	/* Asked to wait beyond the jiffy/scheduler precision? */
302 	if (ret == -ETIME && args->timeout_ns)
303 		ret = -EAGAIN;
304 
305 	return ret;
306 }
307