1 // SPDX-License-Identifier: GPL-2.0 OR MIT
2 /* Copyright 2017-2019 Qiang Yu <yuq825@gmail.com> */
3
4 #include <linux/mm.h>
5 #include <linux/iosys-map.h>
6 #include <linux/sync_file.h>
7 #include <linux/pagemap.h>
8 #include <linux/shmem_fs.h>
9 #include <linux/dma-mapping.h>
10
11 #include <drm/drm_file.h>
12 #include <drm/drm_syncobj.h>
13 #include <drm/drm_utils.h>
14
15 #include <drm/lima_drm.h>
16
17 #include "lima_drv.h"
18 #include "lima_gem.h"
19 #include "lima_vm.h"
20
lima_heap_alloc(struct lima_bo * bo,struct lima_vm * vm)21 int lima_heap_alloc(struct lima_bo *bo, struct lima_vm *vm)
22 {
23 struct page **pages;
24 struct address_space *mapping = bo->base.base.filp->f_mapping;
25 struct device *dev = bo->base.base.dev->dev;
26 size_t old_size = bo->heap_size;
27 size_t new_size = bo->heap_size ? bo->heap_size * 2 :
28 (lima_heap_init_nr_pages << PAGE_SHIFT);
29 struct sg_table sgt;
30 int i, ret;
31
32 if (bo->heap_size >= bo->base.base.size)
33 return -ENOSPC;
34
35 new_size = min(new_size, bo->base.base.size);
36
37 mutex_lock(&bo->base.pages_lock);
38
39 if (bo->base.pages) {
40 pages = bo->base.pages;
41 } else {
42 pages = kvmalloc_array(bo->base.base.size >> PAGE_SHIFT,
43 sizeof(*pages), GFP_KERNEL | __GFP_ZERO);
44 if (!pages) {
45 mutex_unlock(&bo->base.pages_lock);
46 return -ENOMEM;
47 }
48
49 bo->base.pages = pages;
50 bo->base.pages_use_count = 1;
51
52 mapping_set_unevictable(mapping);
53 }
54
55 for (i = old_size >> PAGE_SHIFT; i < new_size >> PAGE_SHIFT; i++) {
56 struct page *page = shmem_read_mapping_page(mapping, i);
57
58 if (IS_ERR(page)) {
59 mutex_unlock(&bo->base.pages_lock);
60 return PTR_ERR(page);
61 }
62 pages[i] = page;
63 }
64
65 mutex_unlock(&bo->base.pages_lock);
66
67 ret = sg_alloc_table_from_pages(&sgt, pages, i, 0,
68 new_size, GFP_KERNEL);
69 if (ret)
70 return ret;
71
72 if (bo->base.sgt) {
73 dma_unmap_sgtable(dev, bo->base.sgt, DMA_BIDIRECTIONAL, 0);
74 sg_free_table(bo->base.sgt);
75 } else {
76 bo->base.sgt = kmalloc(sizeof(*bo->base.sgt), GFP_KERNEL);
77 if (!bo->base.sgt) {
78 sg_free_table(&sgt);
79 return -ENOMEM;
80 }
81 }
82
83 ret = dma_map_sgtable(dev, &sgt, DMA_BIDIRECTIONAL, 0);
84 if (ret) {
85 sg_free_table(&sgt);
86 kfree(bo->base.sgt);
87 bo->base.sgt = NULL;
88 return ret;
89 }
90
91 *bo->base.sgt = sgt;
92
93 if (vm) {
94 ret = lima_vm_map_bo(vm, bo, old_size >> PAGE_SHIFT);
95 if (ret)
96 return ret;
97 }
98
99 bo->heap_size = new_size;
100 return 0;
101 }
102
lima_gem_create_handle(struct drm_device * dev,struct drm_file * file,u32 size,u32 flags,u32 * handle)103 int lima_gem_create_handle(struct drm_device *dev, struct drm_file *file,
104 u32 size, u32 flags, u32 *handle)
105 {
106 int err;
107 gfp_t mask;
108 struct drm_gem_shmem_object *shmem;
109 struct drm_gem_object *obj;
110 struct lima_bo *bo;
111 bool is_heap = flags & LIMA_BO_FLAG_HEAP;
112
113 shmem = drm_gem_shmem_create(dev, size);
114 if (IS_ERR(shmem))
115 return PTR_ERR(shmem);
116
117 obj = &shmem->base;
118
119 /* Mali Utgard GPU can only support 32bit address space */
120 mask = mapping_gfp_mask(obj->filp->f_mapping);
121 mask &= ~__GFP_HIGHMEM;
122 mask |= __GFP_DMA32;
123 mapping_set_gfp_mask(obj->filp->f_mapping, mask);
124
125 if (is_heap) {
126 bo = to_lima_bo(obj);
127 err = lima_heap_alloc(bo, NULL);
128 if (err)
129 goto out;
130 } else {
131 struct sg_table *sgt = drm_gem_shmem_get_pages_sgt(shmem);
132
133 if (IS_ERR(sgt)) {
134 err = PTR_ERR(sgt);
135 goto out;
136 }
137 }
138
139 err = drm_gem_handle_create(file, obj, handle);
140
141 out:
142 /* drop reference from allocate - handle holds it now */
143 drm_gem_object_put(obj);
144
145 return err;
146 }
147
lima_gem_free_object(struct drm_gem_object * obj)148 static void lima_gem_free_object(struct drm_gem_object *obj)
149 {
150 struct lima_bo *bo = to_lima_bo(obj);
151
152 if (!list_empty(&bo->va))
153 dev_err(obj->dev->dev, "lima gem free bo still has va\n");
154
155 drm_gem_shmem_free(&bo->base);
156 }
157
lima_gem_object_open(struct drm_gem_object * obj,struct drm_file * file)158 static int lima_gem_object_open(struct drm_gem_object *obj, struct drm_file *file)
159 {
160 struct lima_bo *bo = to_lima_bo(obj);
161 struct lima_drm_priv *priv = to_lima_drm_priv(file);
162 struct lima_vm *vm = priv->vm;
163
164 return lima_vm_bo_add(vm, bo, true);
165 }
166
lima_gem_object_close(struct drm_gem_object * obj,struct drm_file * file)167 static void lima_gem_object_close(struct drm_gem_object *obj, struct drm_file *file)
168 {
169 struct lima_bo *bo = to_lima_bo(obj);
170 struct lima_drm_priv *priv = to_lima_drm_priv(file);
171 struct lima_vm *vm = priv->vm;
172
173 lima_vm_bo_del(vm, bo);
174 }
175
lima_gem_pin(struct drm_gem_object * obj)176 static int lima_gem_pin(struct drm_gem_object *obj)
177 {
178 struct lima_bo *bo = to_lima_bo(obj);
179
180 if (bo->heap_size)
181 return -EINVAL;
182
183 return drm_gem_shmem_pin(&bo->base);
184 }
185
lima_gem_vmap(struct drm_gem_object * obj,struct iosys_map * map)186 static int lima_gem_vmap(struct drm_gem_object *obj, struct iosys_map *map)
187 {
188 struct lima_bo *bo = to_lima_bo(obj);
189
190 if (bo->heap_size)
191 return -EINVAL;
192
193 return drm_gem_shmem_vmap(&bo->base, map);
194 }
195
lima_gem_mmap(struct drm_gem_object * obj,struct vm_area_struct * vma)196 static int lima_gem_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma)
197 {
198 struct lima_bo *bo = to_lima_bo(obj);
199
200 if (bo->heap_size)
201 return -EINVAL;
202
203 return drm_gem_shmem_mmap(&bo->base, vma);
204 }
205
206 static const struct drm_gem_object_funcs lima_gem_funcs = {
207 .free = lima_gem_free_object,
208 .open = lima_gem_object_open,
209 .close = lima_gem_object_close,
210 .print_info = drm_gem_shmem_object_print_info,
211 .pin = lima_gem_pin,
212 .unpin = drm_gem_shmem_object_unpin,
213 .get_sg_table = drm_gem_shmem_object_get_sg_table,
214 .vmap = lima_gem_vmap,
215 .vunmap = drm_gem_shmem_object_vunmap,
216 .mmap = lima_gem_mmap,
217 .vm_ops = &drm_gem_shmem_vm_ops,
218 };
219
lima_gem_create_object(struct drm_device * dev,size_t size)220 struct drm_gem_object *lima_gem_create_object(struct drm_device *dev, size_t size)
221 {
222 struct lima_bo *bo;
223
224 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
225 if (!bo)
226 return ERR_PTR(-ENOMEM);
227
228 mutex_init(&bo->lock);
229 INIT_LIST_HEAD(&bo->va);
230 bo->base.map_wc = true;
231 bo->base.base.funcs = &lima_gem_funcs;
232
233 return &bo->base.base;
234 }
235
lima_gem_get_info(struct drm_file * file,u32 handle,u32 * va,u64 * offset)236 int lima_gem_get_info(struct drm_file *file, u32 handle, u32 *va, u64 *offset)
237 {
238 struct drm_gem_object *obj;
239 struct lima_bo *bo;
240 struct lima_drm_priv *priv = to_lima_drm_priv(file);
241 struct lima_vm *vm = priv->vm;
242
243 obj = drm_gem_object_lookup(file, handle);
244 if (!obj)
245 return -ENOENT;
246
247 bo = to_lima_bo(obj);
248
249 *va = lima_vm_get_va(vm, bo);
250
251 *offset = drm_vma_node_offset_addr(&obj->vma_node);
252
253 drm_gem_object_put(obj);
254 return 0;
255 }
256
lima_gem_sync_bo(struct lima_sched_task * task,struct lima_bo * bo,bool write,bool explicit)257 static int lima_gem_sync_bo(struct lima_sched_task *task, struct lima_bo *bo,
258 bool write, bool explicit)
259 {
260 int err;
261
262 err = dma_resv_reserve_fences(lima_bo_resv(bo), 1);
263 if (err)
264 return err;
265
266 /* explicit sync use user passed dep fence */
267 if (explicit)
268 return 0;
269
270 return drm_sched_job_add_implicit_dependencies(&task->base,
271 &bo->base.base,
272 write);
273 }
274
lima_gem_add_deps(struct drm_file * file,struct lima_submit * submit)275 static int lima_gem_add_deps(struct drm_file *file, struct lima_submit *submit)
276 {
277 int i, err;
278
279 for (i = 0; i < ARRAY_SIZE(submit->in_sync); i++) {
280 struct dma_fence *fence = NULL;
281
282 if (!submit->in_sync[i])
283 continue;
284
285 err = drm_syncobj_find_fence(file, submit->in_sync[i],
286 0, 0, &fence);
287 if (err)
288 return err;
289
290 err = drm_sched_job_add_dependency(&submit->task->base, fence);
291 if (err) {
292 dma_fence_put(fence);
293 return err;
294 }
295 }
296
297 return 0;
298 }
299
lima_gem_submit(struct drm_file * file,struct lima_submit * submit)300 int lima_gem_submit(struct drm_file *file, struct lima_submit *submit)
301 {
302 int i, err = 0;
303 struct ww_acquire_ctx ctx;
304 struct lima_drm_priv *priv = to_lima_drm_priv(file);
305 struct lima_vm *vm = priv->vm;
306 struct drm_syncobj *out_sync = NULL;
307 struct dma_fence *fence;
308 struct lima_bo **bos = submit->lbos;
309
310 if (submit->out_sync) {
311 out_sync = drm_syncobj_find(file, submit->out_sync);
312 if (!out_sync)
313 return -ENOENT;
314 }
315
316 for (i = 0; i < submit->nr_bos; i++) {
317 struct drm_gem_object *obj;
318 struct lima_bo *bo;
319
320 obj = drm_gem_object_lookup(file, submit->bos[i].handle);
321 if (!obj) {
322 err = -ENOENT;
323 goto err_out0;
324 }
325
326 bo = to_lima_bo(obj);
327
328 /* increase refcnt of gpu va map to prevent unmapped when executing,
329 * will be decreased when task done
330 */
331 err = lima_vm_bo_add(vm, bo, false);
332 if (err) {
333 drm_gem_object_put(obj);
334 goto err_out0;
335 }
336
337 bos[i] = bo;
338 }
339
340 err = drm_gem_lock_reservations((struct drm_gem_object **)bos,
341 submit->nr_bos, &ctx);
342 if (err)
343 goto err_out0;
344
345 err = lima_sched_task_init(
346 submit->task, submit->ctx->context + submit->pipe,
347 bos, submit->nr_bos, vm);
348 if (err)
349 goto err_out1;
350
351 err = lima_gem_add_deps(file, submit);
352 if (err)
353 goto err_out2;
354
355 for (i = 0; i < submit->nr_bos; i++) {
356 err = lima_gem_sync_bo(
357 submit->task, bos[i],
358 submit->bos[i].flags & LIMA_SUBMIT_BO_WRITE,
359 submit->flags & LIMA_SUBMIT_FLAG_EXPLICIT_FENCE);
360 if (err)
361 goto err_out2;
362 }
363
364 fence = lima_sched_context_queue_task(submit->task);
365
366 for (i = 0; i < submit->nr_bos; i++) {
367 dma_resv_add_fence(lima_bo_resv(bos[i]), fence,
368 submit->bos[i].flags & LIMA_SUBMIT_BO_WRITE ?
369 DMA_RESV_USAGE_WRITE : DMA_RESV_USAGE_READ);
370 }
371
372 drm_gem_unlock_reservations((struct drm_gem_object **)bos,
373 submit->nr_bos, &ctx);
374
375 for (i = 0; i < submit->nr_bos; i++)
376 drm_gem_object_put(&bos[i]->base.base);
377
378 if (out_sync) {
379 drm_syncobj_replace_fence(out_sync, fence);
380 drm_syncobj_put(out_sync);
381 }
382
383 dma_fence_put(fence);
384
385 return 0;
386
387 err_out2:
388 lima_sched_task_fini(submit->task);
389 err_out1:
390 drm_gem_unlock_reservations((struct drm_gem_object **)bos,
391 submit->nr_bos, &ctx);
392 err_out0:
393 for (i = 0; i < submit->nr_bos; i++) {
394 if (!bos[i])
395 break;
396 lima_vm_bo_del(vm, bos[i]);
397 drm_gem_object_put(&bos[i]->base.base);
398 }
399 if (out_sync)
400 drm_syncobj_put(out_sync);
401 return err;
402 }
403
lima_gem_wait(struct drm_file * file,u32 handle,u32 op,s64 timeout_ns)404 int lima_gem_wait(struct drm_file *file, u32 handle, u32 op, s64 timeout_ns)
405 {
406 bool write = op & LIMA_GEM_WAIT_WRITE;
407 long ret, timeout;
408
409 if (!op)
410 return 0;
411
412 timeout = drm_timeout_abs_to_jiffies(timeout_ns);
413
414 ret = drm_gem_dma_resv_wait(file, handle, write, timeout);
415 if (ret == -ETIME)
416 ret = timeout ? -ETIMEDOUT : -EBUSY;
417
418 return ret;
419 }
420