summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/i915/i915_active.c
blob: 293e5bcc4b6c44131aa9d17f8dfdb11a96426bd8 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
/*
 * SPDX-License-Identifier: MIT
 *
 * Copyright © 2019 Intel Corporation
 */

#include "gt/intel_engine_pm.h"

#include "i915_drv.h"
#include "i915_active.h"
#include "i915_globals.h"

#define BKL(ref) (&(ref)->i915->drm.struct_mutex)

/*
 * Active refs memory management
 *
 * To be more economical with memory, we reap all the i915_active trees as
 * they idle (when we know the active requests are inactive) and allocate the
 * nodes from a local slab cache to hopefully reduce the fragmentation.
 */
static struct i915_global_active {
	struct i915_global base;
	struct kmem_cache *slab_cache;
} global;

struct active_node {
	struct i915_active_request base;
	struct i915_active *ref;
	struct rb_node node;
	u64 timeline;
};

static void
__active_park(struct i915_active *ref)
{
	struct active_node *it, *n;

	rbtree_postorder_for_each_entry_safe(it, n, &ref->tree, node) {
		GEM_BUG_ON(i915_active_request_isset(&it->base));
		kmem_cache_free(global.slab_cache, it);
	}
	ref->tree = RB_ROOT;
}

static void
__active_retire(struct i915_active *ref)
{
	GEM_BUG_ON(!ref->count);
	if (--ref->count)
		return;

	/* return the unused nodes to our slabcache */
	__active_park(ref);

	ref->retire(ref);
}

static void
node_retire(struct i915_active_request *base, struct i915_request *rq)
{
	__active_retire(container_of(base, struct active_node, base)->ref);
}

static void
last_retire(struct i915_active_request *base, struct i915_request *rq)
{
	__active_retire(container_of(base, struct i915_active, last));
}

static struct i915_active_request *
active_instance(struct i915_active *ref, u64 idx)
{
	struct active_node *node;
	struct rb_node **p, *parent;
	struct i915_request *old;

	/*
	 * We track the most recently used timeline to skip a rbtree search
	 * for the common case, under typical loads we never need the rbtree
	 * at all. We can reuse the last slot if it is empty, that is
	 * after the previous activity has been retired, or if it matches the
	 * current timeline.
	 *
	 * Note that we allow the timeline to be active simultaneously in
	 * the rbtree and the last cache. We do this to avoid having
	 * to search and replace the rbtree element for a new timeline, with
	 * the cost being that we must be aware that the ref may be retired
	 * twice for the same timeline (as the older rbtree element will be
	 * retired before the new request added to last).
	 */
	old = i915_active_request_raw(&ref->last, BKL(ref));
	if (!old || old->fence.context == idx)
		goto out;

	/* Move the currently active fence into the rbtree */
	idx = old->fence.context;

	parent = NULL;
	p = &ref->tree.rb_node;
	while (*p) {
		parent = *p;

		node = rb_entry(parent, struct active_node, node);
		if (node->timeline == idx)
			goto replace;

		if (node->timeline < idx)
			p = &parent->rb_right;
		else
			p = &parent->rb_left;
	}

	node = kmem_cache_alloc(global.slab_cache, GFP_KERNEL);

	/* kmalloc may retire the ref->last (thanks shrinker)! */
	if (unlikely(!i915_active_request_raw(&ref->last, BKL(ref)))) {
		kmem_cache_free(global.slab_cache, node);
		goto out;
	}

	if (unlikely(!node))
		return ERR_PTR(-ENOMEM);

	i915_active_request_init(&node->base, NULL, node_retire);
	node->ref = ref;
	node->timeline = idx;

	rb_link_node(&node->node, parent, p);
	rb_insert_color(&node->node, &ref->tree);

replace:
	/*
	 * Overwrite the previous active slot in the rbtree with last,
	 * leaving last zeroed. If the previous slot is still active,
	 * we must be careful as we now only expect to receive one retire
	 * callback not two, and so much undo the active counting for the
	 * overwritten slot.
	 */
	if (i915_active_request_isset(&node->base)) {
		/* Retire ourselves from the old rq->active_list */
		__list_del_entry(&node->base.link);
		ref->count--;
		GEM_BUG_ON(!ref->count);
	}
	GEM_BUG_ON(list_empty(&ref->last.link));
	list_replace_init(&ref->last.link, &node->base.link);
	node->base.request = fetch_and_zero(&ref->last.request);

out:
	return &ref->last;
}

void i915_active_init(struct drm_i915_private *i915,
		      struct i915_active *ref,
		      void (*retire)(struct i915_active *ref))
{
	ref->i915 = i915;
	ref->retire = retire;
	ref->tree = RB_ROOT;
	i915_active_request_init(&ref->last, NULL, last_retire);
	init_llist_head(&ref->barriers);
	ref->count = 0;
}

int i915_active_ref(struct i915_active *ref,
		    u64 timeline,
		    struct i915_request *rq)
{
	struct i915_active_request *active;
	int err = 0;

	/* Prevent reaping in case we malloc/wait while building the tree */
	i915_active_acquire(ref);

	active = active_instance(ref, timeline);
	if (IS_ERR(active)) {
		err = PTR_ERR(active);
		goto out;
	}

	if (!i915_active_request_isset(active))
		ref->count++;
	__i915_active_request_set(active, rq);

	GEM_BUG_ON(!ref->count);
out:
	i915_active_release(ref);
	return err;
}

bool i915_active_acquire(struct i915_active *ref)
{
	lockdep_assert_held(BKL(ref));
	return !ref->count++;
}

void i915_active_release(struct i915_active *ref)
{
	lockdep_assert_held(BKL(ref));
	__active_retire(ref);
}

int i915_active_wait(struct i915_active *ref)
{
	struct active_node *it, *n;
	int ret = 0;

	if (i915_active_acquire(ref))
		goto out_release;

	ret = i915_active_request_retire(&ref->last, BKL(ref));
	if (ret)
		goto out_release;

	rbtree_postorder_for_each_entry_safe(it, n, &ref->tree, node) {
		ret = i915_active_request_retire(&it->base, BKL(ref));
		if (ret)
			break;
	}

out_release:
	i915_active_release(ref);
	return ret;
}

int i915_request_await_active_request(struct i915_request *rq,
				      struct i915_active_request *active)
{
	struct i915_request *barrier =
		i915_active_request_raw(active, &rq->i915->drm.struct_mutex);

	return barrier ? i915_request_await_dma_fence(rq, &barrier->fence) : 0;
}

int i915_request_await_active(struct i915_request *rq, struct i915_active *ref)
{
	struct active_node *it, *n;
	int err = 0;

	/* await allocates and so we need to avoid hitting the shrinker */
	if (i915_active_acquire(ref))
		goto out; /* was idle */

	err = i915_request_await_active_request(rq, &ref->last);
	if (err)
		goto out;

	rbtree_postorder_for_each_entry_safe(it, n, &ref->tree, node) {
		err = i915_request_await_active_request(rq, &it->base);
		if (err)
			goto out;
	}

out:
	i915_active_release(ref);
	return err;
}

#if IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)
void i915_active_fini(struct i915_active *ref)
{
	GEM_BUG_ON(i915_active_request_isset(&ref->last));
	GEM_BUG_ON(!RB_EMPTY_ROOT(&ref->tree));
	GEM_BUG_ON(ref->count);
}
#endif

int i915_active_acquire_preallocate_barrier(struct i915_active *ref,
					    struct intel_engine_cs *engine)
{
	struct drm_i915_private *i915 = engine->i915;
	struct llist_node *pos, *next;
	unsigned long tmp;
	int err;

	GEM_BUG_ON(!engine->mask);
	for_each_engine_masked(engine, i915, engine->mask, tmp) {
		struct intel_context *kctx = engine->kernel_context;
		struct active_node *node;

		node = kmem_cache_alloc(global.slab_cache, GFP_KERNEL);
		if (unlikely(!node)) {
			err = -ENOMEM;
			goto unwind;
		}

		i915_active_request_init(&node->base,
					 (void *)engine, node_retire);
		node->timeline = kctx->ring->timeline->fence_context;
		node->ref = ref;
		ref->count++;

		intel_engine_pm_get(engine);
		llist_add((struct llist_node *)&node->base.link,
			  &ref->barriers);
	}

	return 0;

unwind:
	llist_for_each_safe(pos, next, llist_del_all(&ref->barriers)) {
		struct active_node *node;

		node = container_of((struct list_head *)pos,
				    typeof(*node), base.link);
		engine = (void *)rcu_access_pointer(node->base.request);

		intel_engine_pm_put(engine);
		kmem_cache_free(global.slab_cache, node);
	}
	return err;
}

void i915_active_acquire_barrier(struct i915_active *ref)
{
	struct llist_node *pos, *next;

	i915_active_acquire(ref);

	llist_for_each_safe(pos, next, llist_del_all(&ref->barriers)) {
		struct intel_engine_cs *engine;
		struct active_node *node;
		struct rb_node **p, *parent;

		node = container_of((struct list_head *)pos,
				    typeof(*node), base.link);

		engine = (void *)rcu_access_pointer(node->base.request);
		RCU_INIT_POINTER(node->base.request, ERR_PTR(-EAGAIN));

		parent = NULL;
		p = &ref->tree.rb_node;
		while (*p) {
			parent = *p;
			if (rb_entry(parent,
				     struct active_node,
				     node)->timeline < node->timeline)
				p = &parent->rb_right;
			else
				p = &parent->rb_left;
		}
		rb_link_node(&node->node, parent, p);
		rb_insert_color(&node->node, &ref->tree);

		llist_add((struct llist_node *)&node->base.link,
			  &engine->barrier_tasks);
		intel_engine_pm_put(engine);
	}
	i915_active_release(ref);
}

void i915_request_add_barriers(struct i915_request *rq)
{
	struct intel_engine_cs *engine = rq->engine;
	struct llist_node *node, *next;

	llist_for_each_safe(node, next, llist_del_all(&engine->barrier_tasks))
		list_add_tail((struct list_head *)node, &rq->active_list);
}

int i915_active_request_set(struct i915_active_request *active,
			    struct i915_request *rq)
{
	int err;

	/* Must maintain ordering wrt previous active requests */
	err = i915_request_await_active_request(rq, active);
	if (err)
		return err;

	__i915_active_request_set(active, rq);
	return 0;
}

void i915_active_retire_noop(struct i915_active_request *active,
			     struct i915_request *request)
{
	/* Space left intentionally blank */
}

#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
#include "selftests/i915_active.c"
#endif

static void i915_global_active_shrink(void)
{
	kmem_cache_shrink(global.slab_cache);
}

static void i915_global_active_exit(void)
{
	kmem_cache_destroy(global.slab_cache);
}

static struct i915_global_active global = { {
	.shrink = i915_global_active_shrink,
	.exit = i915_global_active_exit,
} };

int __init i915_global_active_init(void)
{
	global.slab_cache = KMEM_CACHE(active_node, SLAB_HWCACHE_ALIGN);
	if (!global.slab_cache)
		return -ENOMEM;

	i915_global_register(&global.base);
	return 0;
}