1/*
  2 * Copyright © 2017 Intel Corporation
  3 *
  4 * Permission is hereby granted, free of charge, to any person obtaining a
  5 * copy of this software and associated documentation files (the "Software"),
  6 * to deal in the Software without restriction, including without limitation
  7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  8 * and/or sell copies of the Software, and to permit persons to whom the
  9 * Software is furnished to do so, subject to the following conditions:
 10 *
 11 * The above copyright notice and this permission notice (including the next
 12 * paragraph) shall be included in all copies or substantial portions of the
 13 * Software.
 14 *
 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 21 * IN THE SOFTWARE.
 22 *
 23 */
 24
 25#include "display/intel_frontbuffer.h"
 26#include "gt/intel_gt.h"
 27#include "i915_drv.h"
 28#include "i915_gem_clflush.h"
 29#include "i915_gem_context.h"
 30#include "i915_gem_object.h"
 31#include "i915_globals.h"
 32#include "i915_trace.h"
 33
 34static struct i915_global_object {
 35	struct i915_global base;
 36	struct kmem_cache *slab_objects;
 37} global;
 38
 39struct drm_i915_gem_object *i915_gem_object_alloc(void)
 40{
 41	return kmem_cache_zalloc(global.slab_objects, GFP_KERNEL);
 42}
 43
 44void i915_gem_object_free(struct drm_i915_gem_object *obj)
 45{
 46	return kmem_cache_free(global.slab_objects, obj);
 47}
 48
 49void i915_gem_object_init(struct drm_i915_gem_object *obj,
 50			  const struct drm_i915_gem_object_ops *ops)
 51{
 52	mutex_init(&obj->mm.lock);
 53
 54	spin_lock_init(&obj->vma.lock);
 55	INIT_LIST_HEAD(&obj->vma.list);
 56
 57	INIT_LIST_HEAD(&obj->mm.link);
 58
 59	INIT_LIST_HEAD(&obj->lut_list);
 60
 61	init_rcu_head(&obj->rcu);
 62
 63	obj->ops = ops;
 64
 65	obj->mm.madv = I915_MADV_WILLNEED;
 66	INIT_RADIX_TREE(&obj->mm.get_page.radix, GFP_KERNEL | __GFP_NOWARN);
 67	mutex_init(&obj->mm.get_page.lock);
 68}
 69
 70/**
 71 * Mark up the object's coherency levels for a given cache_level
 72 * @obj: #drm_i915_gem_object
 73 * @cache_level: cache level
 74 */
 75void i915_gem_object_set_cache_coherency(struct drm_i915_gem_object *obj,
 76					 unsigned int cache_level)
 77{
 78	obj->cache_level = cache_level;
 79
 80	if (cache_level != I915_CACHE_NONE)
 81		obj->cache_coherent = (I915_BO_CACHE_COHERENT_FOR_READ |
 82				       I915_BO_CACHE_COHERENT_FOR_WRITE);
 83	else if (HAS_LLC(to_i915(obj->base.dev)))
 84		obj->cache_coherent = I915_BO_CACHE_COHERENT_FOR_READ;
 85	else
 86		obj->cache_coherent = 0;
 87
 88	obj->cache_dirty =
 89		!(obj->cache_coherent & I915_BO_CACHE_COHERENT_FOR_WRITE);
 90}
 91
 92void i915_gem_close_object(struct drm_gem_object *gem, struct drm_file *file)
 93{
 94	struct drm_i915_gem_object *obj = to_intel_bo(gem);
 95	struct drm_i915_file_private *fpriv = file->driver_priv;
 96	struct i915_lut_handle *lut, *ln;
 97	LIST_HEAD(close);
 98
 99	i915_gem_object_lock(obj);
100	list_for_each_entry_safe(lut, ln, &obj->lut_list, obj_link) {
101		struct i915_gem_context *ctx = lut->ctx;
102
103		if (ctx->file_priv != fpriv)
104			continue;
105
106		i915_gem_context_get(ctx);
107		list_move(&lut->obj_link, &close);
108	}
109	i915_gem_object_unlock(obj);
110
111	list_for_each_entry_safe(lut, ln, &close, obj_link) {
112		struct i915_gem_context *ctx = lut->ctx;
113		struct i915_vma *vma;
114
115		/*
116		 * We allow the process to have multiple handles to the same
117		 * vma, in the same fd namespace, by virtue of flink/open.
118		 */
119
120		mutex_lock(&ctx->mutex);
121		vma = radix_tree_delete(&ctx->handles_vma, lut->handle);
122		if (vma) {
123			GEM_BUG_ON(vma->obj != obj);
124			GEM_BUG_ON(!atomic_read(&vma->open_count));
125			if (atomic_dec_and_test(&vma->open_count) &&
126			    !i915_vma_is_ggtt(vma))
127				i915_vma_close(vma);
128		}
129		mutex_unlock(&ctx->mutex);
130
131		i915_gem_context_put(lut->ctx);
132		i915_lut_handle_free(lut);
133		i915_gem_object_put(obj);
134	}
135}
136
137static void __i915_gem_free_object_rcu(struct rcu_head *head)
138{
139	struct drm_i915_gem_object *obj =
140		container_of(head, typeof(*obj), rcu);
141	struct drm_i915_private *i915 = to_i915(obj->base.dev);
142
143	dma_resv_fini(&obj->base._resv);
144	i915_gem_object_free(obj);
145
146	GEM_BUG_ON(!atomic_read(&i915->mm.free_count));
147	atomic_dec(&i915->mm.free_count);
148}
149
150static void __i915_gem_free_objects(struct drm_i915_private *i915,
151				    struct llist_node *freed)
152{
153	struct drm_i915_gem_object *obj, *on;
154	intel_wakeref_t wakeref;
155
156	wakeref = intel_runtime_pm_get(&i915->runtime_pm);
157	llist_for_each_entry_safe(obj, on, freed, freed) {
158		struct i915_vma *vma, *vn;
159
160		trace_i915_gem_object_destroy(obj);
161
162		mutex_lock(&i915->drm.struct_mutex);
163
164		list_for_each_entry_safe(vma, vn, &obj->vma.list, obj_link) {
165			GEM_BUG_ON(i915_vma_is_active(vma));
166			vma->flags &= ~I915_VMA_PIN_MASK;
167			i915_vma_destroy(vma);
168		}
169		GEM_BUG_ON(!list_empty(&obj->vma.list));
170		GEM_BUG_ON(!RB_EMPTY_ROOT(&obj->vma.tree));
171
172		mutex_unlock(&i915->drm.struct_mutex);
173
174		GEM_BUG_ON(atomic_read(&obj->bind_count));
175		GEM_BUG_ON(obj->userfault_count);
176		GEM_BUG_ON(!list_empty(&obj->lut_list));
177
178		atomic_set(&obj->mm.pages_pin_count, 0);
179		__i915_gem_object_put_pages(obj, I915_MM_NORMAL);
180		GEM_BUG_ON(i915_gem_object_has_pages(obj));
181		bitmap_free(obj->bit_17);
182
183		if (obj->base.import_attach)
184			drm_prime_gem_destroy(&obj->base, NULL);
185
186		drm_gem_free_mmap_offset(&obj->base);
187
188		if (obj->ops->release)
189			obj->ops->release(obj);
190
191		/* But keep the pointer alive for RCU-protected lookups */
192		call_rcu(&obj->rcu, __i915_gem_free_object_rcu);
193	}
194	intel_runtime_pm_put(&i915->runtime_pm, wakeref);
195}
196
197void i915_gem_flush_free_objects(struct drm_i915_private *i915)
198{
199	struct llist_node *freed = llist_del_all(&i915->mm.free_list);
200
201	if (unlikely(freed))
202		__i915_gem_free_objects(i915, freed);
203}
204
205static void __i915_gem_free_work(struct work_struct *work)
206{
207	struct drm_i915_private *i915 =
208		container_of(work, struct drm_i915_private, mm.free_work);
209
210	i915_gem_flush_free_objects(i915);
211}
212
213void i915_gem_free_object(struct drm_gem_object *gem_obj)
214{
215	struct drm_i915_gem_object *obj = to_intel_bo(gem_obj);
216	struct drm_i915_private *i915 = to_i915(obj->base.dev);
217
218	GEM_BUG_ON(i915_gem_object_is_framebuffer(obj));
219
220	/*
221	 * Before we free the object, make sure any pure RCU-only
222	 * read-side critical sections are complete, e.g.
223	 * i915_gem_busy_ioctl(). For the corresponding synchronized
224	 * lookup see i915_gem_object_lookup_rcu().
225	 */
226	atomic_inc(&i915->mm.free_count);
227
228	/*
229	 * This serializes freeing with the shrinker. Since the free
230	 * is delayed, first by RCU then by the workqueue, we want the
231	 * shrinker to be able to free pages of unreferenced objects,
232	 * or else we may oom whilst there are plenty of deferred
233	 * freed objects.
234	 */
235	i915_gem_object_make_unshrinkable(obj);
236
237	/*
238	 * Since we require blocking on struct_mutex to unbind the freed
239	 * object from the GPU before releasing resources back to the
240	 * system, we can not do that directly from the RCU callback (which may
241	 * be a softirq context), but must instead then defer that work onto a
242	 * kthread. We use the RCU callback rather than move the freed object
243	 * directly onto the work queue so that we can mix between using the
244	 * worker and performing frees directly from subsequent allocations for
245	 * crude but effective memory throttling.
246	 */
247	if (llist_add(&obj->freed, &i915->mm.free_list))
248		queue_work(i915->wq, &i915->mm.free_work);
249}
250
251static bool gpu_write_needs_clflush(struct drm_i915_gem_object *obj)
252{
253	return !(obj->cache_level == I915_CACHE_NONE ||
254		 obj->cache_level == I915_CACHE_WT);
255}
256
257void
258i915_gem_object_flush_write_domain(struct drm_i915_gem_object *obj,
259				   unsigned int flush_domains)
260{
261	struct i915_vma *vma;
262
263	assert_object_held(obj);
264
265	if (!(obj->write_domain & flush_domains))
266		return;
267
268	switch (obj->write_domain) {
269	case I915_GEM_DOMAIN_GTT:
270		for_each_ggtt_vma(vma, obj)
271			intel_gt_flush_ggtt_writes(vma->vm->gt);
272
273		intel_frontbuffer_flush(obj->frontbuffer, ORIGIN_CPU);
274
275		for_each_ggtt_vma(vma, obj) {
276			if (vma->iomap)
277				continue;
278
279			i915_vma_unset_ggtt_write(vma);
280		}
281
282		break;
283
284	case I915_GEM_DOMAIN_WC:
285		wmb();
286		break;
287
288	case I915_GEM_DOMAIN_CPU:
289		i915_gem_clflush_object(obj, I915_CLFLUSH_SYNC);
290		break;
291
292	case I915_GEM_DOMAIN_RENDER:
293		if (gpu_write_needs_clflush(obj))
294			obj->cache_dirty = true;
295		break;
296	}
297
298	obj->write_domain = 0;
299}
300
301void i915_gem_init__objects(struct drm_i915_private *i915)
302{
303	INIT_WORK(&i915->mm.free_work, __i915_gem_free_work);
304}
305
306static void i915_global_objects_shrink(void)
307{
308	kmem_cache_shrink(global.slab_objects);
309}
310
311static void i915_global_objects_exit(void)
312{
313	kmem_cache_destroy(global.slab_objects);
314}
315
316static struct i915_global_object global = { {
317	.shrink = i915_global_objects_shrink,
318	.exit = i915_global_objects_exit,
319} };
320
321int __init i915_global_objects_init(void)
322{
323	global.slab_objects =
324		KMEM_CACHE(drm_i915_gem_object, SLAB_HWCACHE_ALIGN);
325	if (!global.slab_objects)
326		return -ENOMEM;
327
328	i915_global_register(&global.base);
329	return 0;
330}
331
332#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
333#include "selftests/huge_gem_object.c"
334#include "selftests/huge_pages.c"
335#include "selftests/i915_gem_object.c"
336#include "selftests/i915_gem_coherency.c"
337#endif