xref: /linux/drivers/accel/ivpu/ivpu_gem.c (revision 5fc6c6f258b34fd0d2ff2a63b8a407a4dcbca750)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2020-2023 Intel Corporation
4  */
5 
6 #include <linux/dma-buf.h>
7 #include <linux/highmem.h>
8 #include <linux/module.h>
9 #include <linux/set_memory.h>
10 #include <linux/xarray.h>
11 
12 #include <drm/drm_cache.h>
13 #include <drm/drm_debugfs.h>
14 #include <drm/drm_file.h>
15 #include <drm/drm_utils.h>
16 
17 #include "ivpu_drv.h"
18 #include "ivpu_gem.h"
19 #include "ivpu_hw.h"
20 #include "ivpu_mmu.h"
21 #include "ivpu_mmu_context.h"
22 
23 MODULE_IMPORT_NS("DMA_BUF");
24 
25 static const struct drm_gem_object_funcs ivpu_gem_funcs;
26 
ivpu_dbg_bo(struct ivpu_device * vdev,struct ivpu_bo * bo,const char * action)27 static inline void ivpu_dbg_bo(struct ivpu_device *vdev, struct ivpu_bo *bo, const char *action)
28 {
29 	ivpu_dbg(vdev, BO,
30 		 "%6s: bo %8p vpu_addr %9llx size %8zu ctx %d has_pages %d dma_mapped %d mmu_mapped %d wc %d imported %d\n",
31 		 action, bo, bo->vpu_addr, ivpu_bo_size(bo), bo->ctx_id,
32 		 (bool)bo->base.pages, (bool)bo->base.sgt, bo->mmu_mapped, bo->base.map_wc,
33 		 (bool)drm_gem_is_imported(&bo->base.base));
34 }
35 
ivpu_bo_lock(struct ivpu_bo * bo)36 static inline int ivpu_bo_lock(struct ivpu_bo *bo)
37 {
38 	return dma_resv_lock(bo->base.base.resv, NULL);
39 }
40 
ivpu_bo_unlock(struct ivpu_bo * bo)41 static inline void ivpu_bo_unlock(struct ivpu_bo *bo)
42 {
43 	dma_resv_unlock(bo->base.base.resv);
44 }
45 
46 /*
47  * ivpu_bo_pin() - pin the backing physical pages and map them to VPU.
48  *
49  * This function pins physical memory pages, then maps the physical pages
50  * to IOMMU address space and finally updates the VPU MMU page tables
51  * to allow the VPU to translate VPU address to IOMMU address.
52  */
ivpu_bo_pin(struct ivpu_bo * bo)53 int __must_check ivpu_bo_pin(struct ivpu_bo *bo)
54 {
55 	struct ivpu_device *vdev = ivpu_bo_to_vdev(bo);
56 	struct sg_table *sgt;
57 	int ret = 0;
58 
59 	ivpu_dbg_bo(vdev, bo, "pin");
60 
61 	sgt = drm_gem_shmem_get_pages_sgt(&bo->base);
62 	if (IS_ERR(sgt)) {
63 		ret = PTR_ERR(sgt);
64 		ivpu_err(vdev, "Failed to map BO in IOMMU: %d\n", ret);
65 		return ret;
66 	}
67 
68 	ivpu_bo_lock(bo);
69 
70 	if (!bo->mmu_mapped) {
71 		drm_WARN_ON(&vdev->drm, !bo->ctx);
72 		ret = ivpu_mmu_context_map_sgt(vdev, bo->ctx, bo->vpu_addr, sgt,
73 					       ivpu_bo_is_snooped(bo));
74 		if (ret) {
75 			ivpu_err(vdev, "Failed to map BO in MMU: %d\n", ret);
76 			goto unlock;
77 		}
78 		bo->mmu_mapped = true;
79 	}
80 
81 unlock:
82 	ivpu_bo_unlock(bo);
83 
84 	return ret;
85 }
86 
87 static int
ivpu_bo_alloc_vpu_addr(struct ivpu_bo * bo,struct ivpu_mmu_context * ctx,const struct ivpu_addr_range * range)88 ivpu_bo_alloc_vpu_addr(struct ivpu_bo *bo, struct ivpu_mmu_context *ctx,
89 		       const struct ivpu_addr_range *range)
90 {
91 	struct ivpu_device *vdev = ivpu_bo_to_vdev(bo);
92 	int idx, ret;
93 
94 	if (!drm_dev_enter(&vdev->drm, &idx))
95 		return -ENODEV;
96 
97 	ivpu_bo_lock(bo);
98 
99 	ret = ivpu_mmu_context_insert_node(ctx, range, ivpu_bo_size(bo), &bo->mm_node);
100 	if (!ret) {
101 		bo->ctx = ctx;
102 		bo->vpu_addr = bo->mm_node.start;
103 	} else {
104 		ivpu_err(vdev, "Failed to add BO to context %u: %d\n", ctx->id, ret);
105 	}
106 
107 	ivpu_bo_unlock(bo);
108 
109 	drm_dev_exit(idx);
110 
111 	return ret;
112 }
113 
ivpu_bo_unbind_locked(struct ivpu_bo * bo)114 static void ivpu_bo_unbind_locked(struct ivpu_bo *bo)
115 {
116 	struct ivpu_device *vdev = ivpu_bo_to_vdev(bo);
117 
118 	lockdep_assert(dma_resv_held(bo->base.base.resv) || !kref_read(&bo->base.base.refcount));
119 
120 	if (bo->mmu_mapped) {
121 		drm_WARN_ON(&vdev->drm, !bo->ctx);
122 		drm_WARN_ON(&vdev->drm, !bo->vpu_addr);
123 		drm_WARN_ON(&vdev->drm, !bo->base.sgt);
124 		ivpu_mmu_context_unmap_sgt(vdev, bo->ctx, bo->vpu_addr, bo->base.sgt);
125 		bo->mmu_mapped = false;
126 	}
127 
128 	if (bo->ctx) {
129 		ivpu_mmu_context_remove_node(bo->ctx, &bo->mm_node);
130 		bo->ctx = NULL;
131 	}
132 
133 	if (drm_gem_is_imported(&bo->base.base))
134 		return;
135 
136 	if (bo->base.sgt) {
137 		dma_unmap_sgtable(vdev->drm.dev, bo->base.sgt, DMA_BIDIRECTIONAL, 0);
138 		sg_free_table(bo->base.sgt);
139 		kfree(bo->base.sgt);
140 		bo->base.sgt = NULL;
141 	}
142 }
143 
ivpu_bo_unbind_all_bos_from_context(struct ivpu_device * vdev,struct ivpu_mmu_context * ctx)144 void ivpu_bo_unbind_all_bos_from_context(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx)
145 {
146 	struct ivpu_bo *bo;
147 
148 	if (drm_WARN_ON(&vdev->drm, !ctx))
149 		return;
150 
151 	mutex_lock(&vdev->bo_list_lock);
152 	list_for_each_entry(bo, &vdev->bo_list, bo_list_node) {
153 		ivpu_bo_lock(bo);
154 		if (bo->ctx == ctx) {
155 			ivpu_dbg_bo(vdev, bo, "unbind");
156 			ivpu_bo_unbind_locked(bo);
157 		}
158 		ivpu_bo_unlock(bo);
159 	}
160 	mutex_unlock(&vdev->bo_list_lock);
161 }
162 
ivpu_gem_create_object(struct drm_device * dev,size_t size)163 struct drm_gem_object *ivpu_gem_create_object(struct drm_device *dev, size_t size)
164 {
165 	struct ivpu_bo *bo;
166 
167 	if (size == 0 || !PAGE_ALIGNED(size))
168 		return ERR_PTR(-EINVAL);
169 
170 	bo = kzalloc(sizeof(*bo), GFP_KERNEL);
171 	if (!bo)
172 		return ERR_PTR(-ENOMEM);
173 
174 	bo->base.base.funcs = &ivpu_gem_funcs;
175 	bo->base.pages_mark_dirty_on_put = true; /* VPU can dirty a BO anytime */
176 
177 	INIT_LIST_HEAD(&bo->bo_list_node);
178 
179 	return &bo->base.base;
180 }
181 
ivpu_gem_prime_import(struct drm_device * dev,struct dma_buf * dma_buf)182 struct drm_gem_object *ivpu_gem_prime_import(struct drm_device *dev,
183 					     struct dma_buf *dma_buf)
184 {
185 	struct device *attach_dev = dev->dev;
186 	struct dma_buf_attachment *attach;
187 	struct sg_table *sgt;
188 	struct drm_gem_object *obj;
189 	int ret;
190 
191 	attach = dma_buf_attach(dma_buf, attach_dev);
192 	if (IS_ERR(attach))
193 		return ERR_CAST(attach);
194 
195 	get_dma_buf(dma_buf);
196 
197 	sgt = dma_buf_map_attachment_unlocked(attach, DMA_BIDIRECTIONAL);
198 	if (IS_ERR(sgt)) {
199 		ret = PTR_ERR(sgt);
200 		goto fail_detach;
201 	}
202 
203 	obj = drm_gem_shmem_prime_import_sg_table(dev, attach, sgt);
204 	if (IS_ERR(obj)) {
205 		ret = PTR_ERR(obj);
206 		goto fail_unmap;
207 	}
208 
209 	obj->import_attach = attach;
210 	obj->resv = dma_buf->resv;
211 
212 	return obj;
213 
214 fail_unmap:
215 	dma_buf_unmap_attachment_unlocked(attach, sgt, DMA_BIDIRECTIONAL);
216 fail_detach:
217 	dma_buf_detach(dma_buf, attach);
218 	dma_buf_put(dma_buf);
219 
220 	return ERR_PTR(ret);
221 }
222 
ivpu_bo_alloc(struct ivpu_device * vdev,u64 size,u32 flags,u32 ctx_id)223 static struct ivpu_bo *ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 flags, u32 ctx_id)
224 {
225 	struct drm_gem_shmem_object *shmem;
226 	struct ivpu_bo *bo;
227 
228 	switch (flags & DRM_IVPU_BO_CACHE_MASK) {
229 	case DRM_IVPU_BO_CACHED:
230 	case DRM_IVPU_BO_WC:
231 		break;
232 	default:
233 		return ERR_PTR(-EINVAL);
234 	}
235 
236 	shmem = drm_gem_shmem_create(&vdev->drm, size);
237 	if (IS_ERR(shmem))
238 		return ERR_CAST(shmem);
239 
240 	bo = to_ivpu_bo(&shmem->base);
241 	bo->ctx_id = ctx_id;
242 	bo->base.map_wc = flags & DRM_IVPU_BO_WC;
243 	bo->flags = flags;
244 
245 	mutex_lock(&vdev->bo_list_lock);
246 	list_add_tail(&bo->bo_list_node, &vdev->bo_list);
247 	mutex_unlock(&vdev->bo_list_lock);
248 
249 	ivpu_dbg_bo(vdev, bo, "alloc");
250 
251 	return bo;
252 }
253 
ivpu_gem_bo_open(struct drm_gem_object * obj,struct drm_file * file)254 static int ivpu_gem_bo_open(struct drm_gem_object *obj, struct drm_file *file)
255 {
256 	struct ivpu_file_priv *file_priv = file->driver_priv;
257 	struct ivpu_device *vdev = file_priv->vdev;
258 	struct ivpu_bo *bo = to_ivpu_bo(obj);
259 	struct ivpu_addr_range *range;
260 
261 	if (bo->ctx) {
262 		ivpu_warn(vdev, "Can't add BO to ctx %u: already in ctx %u\n",
263 			  file_priv->ctx.id, bo->ctx->id);
264 		return -EALREADY;
265 	}
266 
267 	if (bo->flags & DRM_IVPU_BO_SHAVE_MEM)
268 		range = &vdev->hw->ranges.shave;
269 	else if (bo->flags & DRM_IVPU_BO_DMA_MEM)
270 		range = &vdev->hw->ranges.dma;
271 	else
272 		range = &vdev->hw->ranges.user;
273 
274 	return ivpu_bo_alloc_vpu_addr(bo, &file_priv->ctx, range);
275 }
276 
ivpu_gem_bo_free(struct drm_gem_object * obj)277 static void ivpu_gem_bo_free(struct drm_gem_object *obj)
278 {
279 	struct ivpu_device *vdev = to_ivpu_device(obj->dev);
280 	struct ivpu_bo *bo = to_ivpu_bo(obj);
281 
282 	ivpu_dbg_bo(vdev, bo, "free");
283 
284 	mutex_lock(&vdev->bo_list_lock);
285 	list_del(&bo->bo_list_node);
286 	mutex_unlock(&vdev->bo_list_lock);
287 
288 	drm_WARN_ON(&vdev->drm, !drm_gem_is_imported(&bo->base.base) &&
289 		    !dma_resv_test_signaled(obj->resv, DMA_RESV_USAGE_READ));
290 	drm_WARN_ON(&vdev->drm, ivpu_bo_size(bo) == 0);
291 	drm_WARN_ON(&vdev->drm, bo->base.vaddr);
292 
293 	ivpu_bo_unbind_locked(bo);
294 	drm_WARN_ON(&vdev->drm, bo->mmu_mapped);
295 	drm_WARN_ON(&vdev->drm, bo->ctx);
296 
297 	drm_WARN_ON(obj->dev, refcount_read(&bo->base.pages_use_count) > 1);
298 	drm_gem_shmem_free(&bo->base);
299 }
300 
301 static const struct drm_gem_object_funcs ivpu_gem_funcs = {
302 	.free = ivpu_gem_bo_free,
303 	.open = ivpu_gem_bo_open,
304 	.print_info = drm_gem_shmem_object_print_info,
305 	.pin = drm_gem_shmem_object_pin,
306 	.unpin = drm_gem_shmem_object_unpin,
307 	.get_sg_table = drm_gem_shmem_object_get_sg_table,
308 	.vmap = drm_gem_shmem_object_vmap,
309 	.vunmap = drm_gem_shmem_object_vunmap,
310 	.mmap = drm_gem_shmem_object_mmap,
311 	.vm_ops = &drm_gem_shmem_vm_ops,
312 };
313 
ivpu_bo_create_ioctl(struct drm_device * dev,void * data,struct drm_file * file)314 int ivpu_bo_create_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
315 {
316 	struct ivpu_file_priv *file_priv = file->driver_priv;
317 	struct ivpu_device *vdev = file_priv->vdev;
318 	struct drm_ivpu_bo_create *args = data;
319 	u64 size = PAGE_ALIGN(args->size);
320 	struct ivpu_bo *bo;
321 	int ret;
322 
323 	if (args->flags & ~DRM_IVPU_BO_FLAGS)
324 		return -EINVAL;
325 
326 	if (size == 0)
327 		return -EINVAL;
328 
329 	bo = ivpu_bo_alloc(vdev, size, args->flags, file_priv->ctx.id);
330 	if (IS_ERR(bo)) {
331 		ivpu_err(vdev, "Failed to allocate BO: %pe (ctx %u size %llu flags 0x%x)",
332 			 bo, file_priv->ctx.id, args->size, args->flags);
333 		return PTR_ERR(bo);
334 	}
335 
336 	ret = drm_gem_handle_create(file, &bo->base.base, &args->handle);
337 	if (ret)
338 		ivpu_err(vdev, "Failed to create handle for BO: %pe (ctx %u size %llu flags 0x%x)",
339 			 bo, file_priv->ctx.id, args->size, args->flags);
340 	else
341 		args->vpu_addr = bo->vpu_addr;
342 
343 	drm_gem_object_put(&bo->base.base);
344 
345 	return ret;
346 }
347 
348 struct ivpu_bo *
ivpu_bo_create(struct ivpu_device * vdev,struct ivpu_mmu_context * ctx,struct ivpu_addr_range * range,u64 size,u32 flags)349 ivpu_bo_create(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx,
350 	       struct ivpu_addr_range *range, u64 size, u32 flags)
351 {
352 	struct iosys_map map;
353 	struct ivpu_bo *bo;
354 	int ret;
355 
356 	if (drm_WARN_ON(&vdev->drm, !range))
357 		return NULL;
358 
359 	drm_WARN_ON(&vdev->drm, !PAGE_ALIGNED(range->start));
360 	drm_WARN_ON(&vdev->drm, !PAGE_ALIGNED(range->end));
361 	drm_WARN_ON(&vdev->drm, !PAGE_ALIGNED(size));
362 
363 	bo = ivpu_bo_alloc(vdev, size, flags, IVPU_GLOBAL_CONTEXT_MMU_SSID);
364 	if (IS_ERR(bo)) {
365 		ivpu_err(vdev, "Failed to allocate BO: %pe (vpu_addr 0x%llx size %llu flags 0x%x)",
366 			 bo, range->start, size, flags);
367 		return NULL;
368 	}
369 
370 	ret = ivpu_bo_alloc_vpu_addr(bo, ctx, range);
371 	if (ret)
372 		goto err_put;
373 
374 	ret = ivpu_bo_pin(bo);
375 	if (ret)
376 		goto err_put;
377 
378 	if (flags & DRM_IVPU_BO_MAPPABLE) {
379 		ivpu_bo_lock(bo);
380 		ret = drm_gem_shmem_vmap_locked(&bo->base, &map);
381 		ivpu_bo_unlock(bo);
382 
383 		if (ret)
384 			goto err_put;
385 	}
386 
387 	return bo;
388 
389 err_put:
390 	drm_gem_object_put(&bo->base.base);
391 	return NULL;
392 }
393 
ivpu_bo_create_global(struct ivpu_device * vdev,u64 size,u32 flags)394 struct ivpu_bo *ivpu_bo_create_global(struct ivpu_device *vdev, u64 size, u32 flags)
395 {
396 	return ivpu_bo_create(vdev, &vdev->gctx, &vdev->hw->ranges.global, size, flags);
397 }
398 
ivpu_bo_free(struct ivpu_bo * bo)399 void ivpu_bo_free(struct ivpu_bo *bo)
400 {
401 	struct iosys_map map = IOSYS_MAP_INIT_VADDR(bo->base.vaddr);
402 
403 	if (bo->flags & DRM_IVPU_BO_MAPPABLE) {
404 		ivpu_bo_lock(bo);
405 		drm_gem_shmem_vunmap_locked(&bo->base, &map);
406 		ivpu_bo_unlock(bo);
407 	}
408 
409 	drm_gem_object_put(&bo->base.base);
410 }
411 
ivpu_bo_info_ioctl(struct drm_device * dev,void * data,struct drm_file * file)412 int ivpu_bo_info_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
413 {
414 	struct drm_ivpu_bo_info *args = data;
415 	struct drm_gem_object *obj;
416 	struct ivpu_bo *bo;
417 	int ret = 0;
418 
419 	obj = drm_gem_object_lookup(file, args->handle);
420 	if (!obj)
421 		return -ENOENT;
422 
423 	bo = to_ivpu_bo(obj);
424 
425 	ivpu_bo_lock(bo);
426 	args->flags = bo->flags;
427 	args->mmap_offset = drm_vma_node_offset_addr(&obj->vma_node);
428 	args->vpu_addr = bo->vpu_addr;
429 	args->size = obj->size;
430 	ivpu_bo_unlock(bo);
431 
432 	drm_gem_object_put(obj);
433 	return ret;
434 }
435 
ivpu_bo_wait_ioctl(struct drm_device * dev,void * data,struct drm_file * file)436 int ivpu_bo_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
437 {
438 	struct drm_ivpu_bo_wait *args = data;
439 	struct drm_gem_object *obj;
440 	unsigned long timeout;
441 	long ret;
442 
443 	timeout = drm_timeout_abs_to_jiffies(args->timeout_ns);
444 
445 	/* Add 1 jiffy to ensure the wait function never times out before intended timeout_ns */
446 	timeout += 1;
447 
448 	obj = drm_gem_object_lookup(file, args->handle);
449 	if (!obj)
450 		return -EINVAL;
451 
452 	ret = dma_resv_wait_timeout(obj->resv, DMA_RESV_USAGE_READ, true, timeout);
453 	if (ret == 0) {
454 		ret = -ETIMEDOUT;
455 	} else if (ret > 0) {
456 		ret = 0;
457 		args->job_status = to_ivpu_bo(obj)->job_status;
458 	}
459 
460 	drm_gem_object_put(obj);
461 
462 	return ret;
463 }
464 
ivpu_bo_print_info(struct ivpu_bo * bo,struct drm_printer * p)465 static void ivpu_bo_print_info(struct ivpu_bo *bo, struct drm_printer *p)
466 {
467 	ivpu_bo_lock(bo);
468 
469 	drm_printf(p, "%-9p %-3u 0x%-12llx %-10lu 0x%-8x %-4u",
470 		   bo, bo->ctx_id, bo->vpu_addr, bo->base.base.size,
471 		   bo->flags, kref_read(&bo->base.base.refcount));
472 
473 	if (bo->base.pages)
474 		drm_printf(p, " has_pages");
475 
476 	if (bo->mmu_mapped)
477 		drm_printf(p, " mmu_mapped");
478 
479 	if (drm_gem_is_imported(&bo->base.base))
480 		drm_printf(p, " imported");
481 
482 	drm_printf(p, "\n");
483 
484 	ivpu_bo_unlock(bo);
485 }
486 
ivpu_bo_list(struct drm_device * dev,struct drm_printer * p)487 void ivpu_bo_list(struct drm_device *dev, struct drm_printer *p)
488 {
489 	struct ivpu_device *vdev = to_ivpu_device(dev);
490 	struct ivpu_bo *bo;
491 
492 	drm_printf(p, "%-9s %-3s %-14s %-10s %-10s %-4s %s\n",
493 		   "bo", "ctx", "vpu_addr", "size", "flags", "refs", "attribs");
494 
495 	mutex_lock(&vdev->bo_list_lock);
496 	list_for_each_entry(bo, &vdev->bo_list, bo_list_node)
497 		ivpu_bo_print_info(bo, p);
498 	mutex_unlock(&vdev->bo_list_lock);
499 }
500 
ivpu_bo_list_print(struct drm_device * dev)501 void ivpu_bo_list_print(struct drm_device *dev)
502 {
503 	struct drm_printer p = drm_info_printer(dev->dev);
504 
505 	ivpu_bo_list(dev, &p);
506 }
507