i915_active.c 7.0 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11
/*
 * SPDX-License-Identifier: MIT
 *
 * Copyright © 2019 Intel Corporation
 */

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

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

12 13 14 15 16 17 18 19 20 21 22
/*
 * 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 kmem_cache *slab_cache;
} global;

23
struct active_node {
24
	struct i915_active_request base;
25 26 27 28 29
	struct i915_active *ref;
	struct rb_node node;
	u64 timeline;
};

30 31 32 33 34 35
static void
__active_park(struct i915_active *ref)
{
	struct active_node *it, *n;

	rbtree_postorder_for_each_entry_safe(it, n, &ref->tree, node) {
36
		GEM_BUG_ON(i915_active_request_isset(&it->base));
37
		kmem_cache_free(global.slab_cache, it);
38 39 40 41
	}
	ref->tree = RB_ROOT;
}

42 43 44 45
static void
__active_retire(struct i915_active *ref)
{
	GEM_BUG_ON(!ref->count);
46 47 48 49 50 51 52
	if (--ref->count)
		return;

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

	ref->retire(ref);
53 54 55
}

static void
56
node_retire(struct i915_active_request *base, struct i915_request *rq)
57 58 59 60 61
{
	__active_retire(container_of(base, struct active_node, base)->ref);
}

static void
62
last_retire(struct i915_active_request *base, struct i915_request *rq)
63 64 65 66
{
	__active_retire(container_of(base, struct i915_active, last));
}

67
static struct i915_active_request *
68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87
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).
	 */
88
	old = i915_active_request_raw(&ref->last, BKL(ref));
89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109
	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;
	}

110
	node = kmem_cache_alloc(global.slab_cache, GFP_KERNEL);
111 112

	/* kmalloc may retire the ref->last (thanks shrinker)! */
113
	if (unlikely(!i915_active_request_raw(&ref->last, BKL(ref)))) {
114
		kmem_cache_free(global.slab_cache, node);
115 116 117 118 119 120
		goto out;
	}

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

121
	i915_active_request_init(&node->base, NULL, node_retire);
122 123 124 125 126 127 128 129 130 131 132 133 134 135
	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.
	 */
136
	if (i915_active_request_isset(&node->base)) {
137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156
		/* 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;
157
	i915_active_request_init(&ref->last, NULL, last_retire);
158 159 160 161 162 163 164
	ref->count = 0;
}

int i915_active_ref(struct i915_active *ref,
		    u64 timeline,
		    struct i915_request *rq)
{
165
	struct i915_active_request *active;
166 167 168 169
	int err = 0;

	/* Prevent reaping in case we malloc/wait while building the tree */
	i915_active_acquire(ref);
170 171

	active = active_instance(ref, timeline);
172 173 174 175
	if (IS_ERR(active)) {
		err = PTR_ERR(active);
		goto out;
	}
176

177
	if (!i915_active_request_isset(active))
178
		ref->count++;
179
	__i915_active_request_set(active, rq);
180 181

	GEM_BUG_ON(!ref->count);
182 183 184
out:
	i915_active_release(ref);
	return err;
185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206
}

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;

207
	ret = i915_active_request_retire(&ref->last, BKL(ref));
208 209 210 211
	if (ret)
		goto out_release;

	rbtree_postorder_for_each_entry_safe(it, n, &ref->tree, node) {
212
		ret = i915_active_request_retire(&it->base, BKL(ref));
213 214 215 216 217 218 219 220 221
		if (ret)
			break;
	}

out_release:
	i915_active_release(ref);
	return ret;
}

222 223
int i915_request_await_active_request(struct i915_request *rq,
				      struct i915_active_request *active)
224 225
{
	struct i915_request *barrier =
226
		i915_active_request_raw(active, &rq->i915->drm.struct_mutex);
227 228 229 230 231 232 233

	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;
234
	int err = 0;
235

236 237 238 239 240 241 242
	/* 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;
243 244

	rbtree_postorder_for_each_entry_safe(it, n, &ref->tree, node) {
245 246 247
		err = i915_request_await_active_request(rq, &it->base);
		if (err)
			goto out;
248 249
	}

250 251 252
out:
	i915_active_release(ref);
	return err;
253 254
}

255
#if IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)
256 257
void i915_active_fini(struct i915_active *ref)
{
258
	GEM_BUG_ON(i915_active_request_isset(&ref->last));
259 260
	GEM_BUG_ON(!RB_EMPTY_ROOT(&ref->tree));
	GEM_BUG_ON(ref->count);
261
}
262
#endif
263

264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283
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 */
}

284 285 286
#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
#include "selftests/i915_active.c"
#endif
287 288 289 290 291 292 293 294 295 296 297 298 299 300

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

	return 0;
}

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