i915_gem_evict.c 8.7 KB
Newer Older
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
/*
 * Copyright © 2008-2010 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 *
 * Authors:
 *    Eric Anholt <eric@anholt.net>
 *    Chris Wilson <chris@chris-wilson.co.uuk>
 *
 */

29 30
#include <drm/drmP.h>
#include <drm/i915_drm.h>
31 32 33

#include "i915_drv.h"
#include "intel_drv.h"
C
Chris Wilson 已提交
34
#include "i915_trace.h"
35

C
Chris Wilson 已提交
36
static bool ggtt_is_idle(struct drm_i915_private *dev_priv)
37
{
C
Chris Wilson 已提交
38
	struct i915_ggtt *ggtt = &dev_priv->ggtt;
39
	struct intel_engine_cs *engine;
40
	enum intel_engine_id id;
41

42
	for_each_engine(engine, dev_priv, id) {
C
Chris Wilson 已提交
43 44 45 46
		struct intel_timeline *tl;

		tl = &ggtt->base.timeline.engine[engine->id];
		if (i915_gem_active_isset(&tl->last_request))
47 48 49 50 51 52
			return false;
	}

	return true;
}

53
static bool
54 55 56 57
mark_free(struct drm_mm_scan *scan,
	  struct i915_vma *vma,
	  unsigned int flags,
	  struct list_head *unwind)
58
{
59
	if (i915_vma_is_pinned(vma))
60 61
		return false;

62 63 64
	if (WARN_ON(!list_empty(&vma->exec_list)))
		return false;

65
	if (flags & PIN_NONFAULT && !list_empty(&vma->obj->userfault_link))
66 67
		return false;

68
	list_add(&vma->exec_list, unwind);
69
	return drm_mm_scan_add_block(scan, &vma->node);
70 71
}

72 73 74
/**
 * i915_gem_evict_something - Evict vmas to make room for binding a new one
 * @vm: address space to evict from
75
 * @min_size: size of the desired free space
76 77
 * @alignment: alignment constraint of the desired free space
 * @cache_level: cache_level for the desired space
78 79 80
 * @start: start (inclusive) of the range from which to evict objects
 * @end: end (exclusive) of the range from which to evict objects
 * @flags: additional flags to control the eviction algorithm
81 82 83 84 85 86 87
 *
 * This function will try to evict vmas until a free space satisfying the
 * requirements is found. Callers must check first whether any such hole exists
 * already before calling this function.
 *
 * This function is used by the object/vma binding code.
 *
88 89 90 91
 * Since this function is only used to free up virtual address space it only
 * ignores pinned vmas, and not object where the backing storage itself is
 * pinned. Hence obj->pages_pin_count does not protect against eviction.
 *
92 93 94
 * To clarify: This is for freeing up virtual address space, not for freeing
 * memory in e.g. the shrinker.
 */
95
int
96
i915_gem_evict_something(struct i915_address_space *vm,
97 98 99
			 u64 min_size, u64 alignment,
			 unsigned cache_level,
			 u64 start, u64 end,
100
			 unsigned flags)
101
{
102
	struct drm_i915_private *dev_priv = to_i915(vm->dev);
103
	struct drm_mm_scan scan;
104 105 106 107 108 109 110 111
	struct list_head eviction_list;
	struct list_head *phases[] = {
		&vm->inactive_list,
		&vm->active_list,
		NULL,
	}, **phase;
	struct i915_vma *vma, *next;
	int ret;
112

113
	lockdep_assert_held(&vm->dev->struct_mutex);
114
	trace_i915_gem_evict(vm, min_size, alignment, flags);
C
Chris Wilson 已提交
115

116 117 118
	/*
	 * The goal is to evict objects and amalgamate space in LRU order.
	 * The oldest idle objects reside on the inactive list, which is in
119 120
	 * retirement order. The next objects to retire are those in flight,
	 * on the active list, again in retirement order.
121 122 123
	 *
	 * The retirement sequence is thus:
	 *   1. Inactive objects (already retired)
124
	 *   2. Active objects (will stall on unbinding)
125 126 127 128
	 *
	 * On each list, the oldest objects lie at the HEAD with the freshest
	 * object on the TAIL.
	 */
129 130
	drm_mm_scan_init_with_range(&scan, &vm->mm,
				    min_size, alignment, cache_level,
131 132
				    start, end,
				    flags & PIN_HIGH ? DRM_MM_CREATE_TOP : 0);
133

134
	if (flags & PIN_NONBLOCK)
135
		phases[1] = NULL;
136

137 138 139 140 141
search_again:
	INIT_LIST_HEAD(&eviction_list);
	phase = phases;
	do {
		list_for_each_entry(vma, *phase, vm_link)
142
			if (mark_free(&scan, vma, flags, &eviction_list))
143 144
				goto found;
	} while (*++phase);
145 146

	/* Nothing found, clean up and bail out! */
147
	list_for_each_entry_safe(vma, next, &eviction_list, exec_list) {
148
		ret = drm_mm_scan_remove_block(&scan, &vma->node);
149
		BUG_ON(ret);
150

151
		INIT_LIST_HEAD(&vma->exec_list);
152 153
	}

154
	/* Can we unpin some objects such as idle hw contents,
155 156 157
	 * or pending flips? But since only the GGTT has global entries
	 * such as scanouts, rinbuffers and contexts, we can skip the
	 * purge when inspecting per-process local address spaces.
158
	 */
159
	if (!i915_is_ggtt(vm) || flags & PIN_NONBLOCK)
160
		return -ENOSPC;
161

C
Chris Wilson 已提交
162
	if (ggtt_is_idle(dev_priv)) {
163 164 165 166
		/* If we still have pending pageflip completions, drop
		 * back to userspace to give our workqueues time to
		 * acquire our locks and unpin the old scanouts.
		 */
167
		return intel_has_pending_fb_unpin(vm->dev) ? -EAGAIN : -ENOSPC;
168 169
	}

170 171 172 173 174
	/* Not everything in the GGTT is tracked via vma (otherwise we
	 * could evict as required with minimal stalling) so we are forced
	 * to idle the GPU and explicitly retire outstanding requests in
	 * the hopes that we can then remove contexts and the like only
	 * bound by their active reference.
175
	 */
176 177 178 179
	ret = i915_gem_switch_to_kernel_context(dev_priv);
	if (ret)
		return ret;

180 181 182
	ret = i915_gem_wait_for_idle(dev_priv,
				     I915_WAIT_INTERRUPTIBLE |
				     I915_WAIT_LOCKED);
183 184 185 186 187
	if (ret)
		return ret;

	i915_gem_retire_requests(dev_priv);
	goto search_again;
188 189

found:
190
	/* drm_mm doesn't allow any other other operations while
191 192 193 194 195 196
	 * scanning, therefore store to-be-evicted objects on a
	 * temporary list and take a reference for all before
	 * calling unbind (which may remove the active reference
	 * of any of our objects, thus corrupting the list).
	 */
	list_for_each_entry_safe(vma, next, &eviction_list, exec_list) {
197
		if (drm_mm_scan_remove_block(&scan, &vma->node))
198
			__i915_vma_pin(vma);
199 200
		else
			list_del_init(&vma->exec_list);
201
	}
202

203
	/* Unbinding will emit any required flushes */
204
	while (!list_empty(&eviction_list)) {
205 206
		vma = list_first_entry(&eviction_list,
				       struct i915_vma,
207
				       exec_list);
208 209

		list_del_init(&vma->exec_list);
210
		__i915_vma_unpin(vma);
211
		if (ret == 0)
212
			ret = i915_vma_unbind(vma);
213
	}
214
	return ret;
215 216
}

217 218 219 220 221
int
i915_gem_evict_for_vma(struct i915_vma *target)
{
	struct drm_mm_node *node, *next;

222 223
	lockdep_assert_held(&target->vm->dev->struct_mutex);

224 225 226 227 228 229 230 231 232 233 234 235 236
	list_for_each_entry_safe(node, next,
			&target->vm->mm.head_node.node_list,
			node_list) {
		struct i915_vma *vma;
		int ret;

		if (node->start + node->size <= target->node.start)
			continue;
		if (node->start >= target->node.start + target->node.size)
			break;

		vma = container_of(node, typeof(*vma), node);

237 238
		if (i915_vma_is_pinned(vma)) {
			if (!vma->exec_entry || i915_vma_pin_count(vma) > 1)
239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257
				/* Object is pinned for some other use */
				return -EBUSY;

			/* We need to evict a buffer in the same batch */
			if (vma->exec_entry->flags & EXEC_OBJECT_PINNED)
				/* Overlapping fixed objects in the same batch */
				return -EINVAL;

			return -ENOSPC;
		}

		ret = i915_vma_unbind(vma);
		if (ret)
			return ret;
	}

	return 0;
}

258
/**
259 260
 * i915_gem_evict_vm - Evict all idle vmas from a vm
 * @vm: Address space to cleanse
261 262
 * @do_idle: Boolean directing whether to idle first.
 *
263 264
 * This function evicts all idles vmas from a vm. If all unpinned vmas should be
 * evicted the @do_idle needs to be set to true.
265
 *
266 267 268 269 270
 * This is used by the execbuf code as a last-ditch effort to defragment the
 * address space.
 *
 * To clarify: This is for freeing up virtual address space, not for freeing
 * memory in e.g. the shrinker.
271 272
 */
int i915_gem_evict_vm(struct i915_address_space *vm, bool do_idle)
273 274 275 276
{
	struct i915_vma *vma, *next;
	int ret;

277
	lockdep_assert_held(&vm->dev->struct_mutex);
278 279
	trace_i915_gem_evict_vm(vm);

280
	if (do_idle) {
281 282
		struct drm_i915_private *dev_priv = to_i915(vm->dev);

283
		if (i915_is_ggtt(vm)) {
284
			ret = i915_gem_switch_to_kernel_context(dev_priv);
285 286 287
			if (ret)
				return ret;
		}
288

289 290 291
		ret = i915_gem_wait_for_idle(dev_priv,
					     I915_WAIT_INTERRUPTIBLE |
					     I915_WAIT_LOCKED);
292 293 294
		if (ret)
			return ret;

295
		i915_gem_retire_requests(dev_priv);
296
		WARN_ON(!list_empty(&vm->active_list));
297 298
	}

299
	list_for_each_entry_safe(vma, next, &vm->inactive_list, vm_link)
300
		if (!i915_vma_is_pinned(vma))
301 302 303 304
			WARN_ON(i915_vma_unbind(vma));

	return 0;
}