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
|
From 70f54ff98d5c40772d60fc20861ff4a3884df437 Mon Sep 17 00:00:00 2001
From: Roger He <Hongbo.He@amd.com>
Date: Thu, 21 Dec 2017 17:42:53 +0800
Subject: [PATCH 3736/4131] drm/ttm: enable swapout for reserved BOs during
allocation
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
if the bo shares same reservation object then not lock it again
at swapout time to make it possible to swap out.
v2: refine the commmit message
Reviewed-by: Thomas Hellström <thellstrom@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chuming Zhou <david1.zhou@amd.com>
Signed-off-by: Roger He <Hongbo.He@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/ttm/ttm_bo.c | 142 ++++++++++++++++++++-------------------
drivers/gpu/drm/ttm/ttm_memory.c | 45 ++++++-------
2 files changed, 95 insertions(+), 92 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index d3ea59c..6c1ef7a 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -1737,95 +1737,101 @@ EXPORT_SYMBOL(ttm_bo_synccpu_write_release);
* buffer object on the bo_global::swap_lru list.
*/
-static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
+int ttm_bo_swapout(struct ttm_bo_global *glob, struct ttm_operation_ctx *ctx)
{
- struct ttm_bo_global *glob =
- container_of(shrink, struct ttm_bo_global, shrink);
- struct ttm_buffer_object *bo;
- int ret = -EBUSY;
- unsigned i;
+ struct ttm_buffer_object *bo;
+ int ret = -EBUSY;
+ bool locked;
+ unsigned i;
+
+ spin_lock(&glob->lru_lock);
+ for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
+ list_for_each_entry(bo, &glob->swap_lru[i], swap) {
+ if (ttm_bo_evict_swapout_allowable(bo, ctx, &locked)) {
+ ret = 0;
+ break;
+ }
+ }
+ if (!ret)
+ break;
+ }
- spin_lock(&glob->lru_lock);
- for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
- list_for_each_entry(bo, &glob->swap_lru[i], swap) {
- ret = __ttm_bo_reserve(bo, false, true, NULL);
- if (!ret)
- break;
- }
- if (!ret)
- break;
- }
+ if (ret) {
+ spin_unlock(&glob->lru_lock);
+ return ret;
+ }
- if (ret) {
- spin_unlock(&glob->lru_lock);
- return ret;
- }
+ kref_get(&bo->list_kref);
- kref_get(&bo->list_kref);
+ if (!list_empty(&bo->ddestroy)) {
+ ret = ttm_bo_cleanup_refs(bo, false, false, locked);
+ kref_put(&bo->list_kref, ttm_bo_release_list);
+ return ret;
+ }
- if (!list_empty(&bo->ddestroy)) {
- ret = ttm_bo_cleanup_refs_and_unlock(bo, false, false);
- kref_put(&bo->list_kref, ttm_bo_release_list);
- return ret;
- }
+ ttm_bo_del_from_lru(bo);
+ spin_unlock(&glob->lru_lock);
- ttm_bo_del_from_lru(bo);
- spin_unlock(&glob->lru_lock);
+ /**
+ * Move to system cached
+ */
- /**
- * Move to system cached
- */
+ if (bo->mem.mem_type != TTM_PL_SYSTEM ||
+ bo->ttm->caching_state != tt_cached) {
+ struct ttm_operation_ctx ctx = { false, false };
+ struct ttm_mem_reg evict_mem;
- if (bo->mem.mem_type != TTM_PL_SYSTEM ||
- bo->ttm->caching_state != tt_cached) {
- struct ttm_operation_ctx ctx = { false, false };
- struct ttm_mem_reg evict_mem;
+ evict_mem = bo->mem;
+ evict_mem.mm_node = NULL;
+ evict_mem.placement = TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED;
+ evict_mem.mem_type = TTM_PL_SYSTEM;
- evict_mem = bo->mem;
- evict_mem.mm_node = NULL;
- evict_mem.placement = TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED;
- evict_mem.mem_type = TTM_PL_SYSTEM;
-
- ret = ttm_bo_handle_move_mem(bo, &evict_mem, true, &ctx);
- if (unlikely(ret != 0))
- goto out;
- }
+ ret = ttm_bo_handle_move_mem(bo, &evict_mem, true, &ctx);
+ if (unlikely(ret != 0))
+ goto out;
+ }
- /**
- * Make sure BO is idle.
- */
+ /**
+ * Make sure BO is idle.
+ */
- ret = ttm_bo_wait(bo, false, false);
- if (unlikely(ret != 0))
- goto out;
+ ret = ttm_bo_wait(bo, false, false);
+ if (unlikely(ret != 0))
+ goto out;
- ttm_bo_unmap_virtual(bo);
+ ttm_bo_unmap_virtual(bo);
- /**
- * Swap out. Buffer will be swapped in again as soon as
- * anyone tries to access a ttm page.
- */
+ /**
+ * Swap out. Buffer will be swapped in again as soon as
+ * anyone tries to access a ttm page.
+ */
- if (bo->bdev->driver->swap_notify)
- bo->bdev->driver->swap_notify(bo);
+ if (bo->bdev->driver->swap_notify)
+ bo->bdev->driver->swap_notify(bo);
- ret = ttm_tt_swapout(bo->ttm, bo->persistent_swap_storage);
+ ret = ttm_tt_swapout(bo->ttm, bo->persistent_swap_storage);
out:
- /**
- *
- * Unreserve without putting on LRU to avoid swapping out an
- * already swapped buffer.
- */
-
- __ttm_bo_unreserve(bo);
- kref_put(&bo->list_kref, ttm_bo_release_list);
- return ret;
+ /**
+ *
+ * Unreserve without putting on LRU to avoid swapping out an
+ * already swapped buffer.
+ */
+ if (locked)
+ kcl_reservation_object_unlock(bo->resv);
+ kref_put(&bo->list_kref, ttm_bo_release_list);
+ return ret;
}
+EXPORT_SYMBOL(ttm_bo_swapout);
void ttm_bo_swapout_all(struct ttm_bo_device *bdev)
{
- while (ttm_bo_swapout(&bdev->glob->shrink) == 0)
+ struct ttm_operation_ctx ctx = {
+ .interruptible = false,
+ .no_wait_gpu = false
+ };
+
+ while (ttm_bo_swapout(bdev->glob, &ctx) == 0)
;
}
EXPORT_SYMBOL(ttm_bo_swapout_all);
diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c
index c1cbefe..dc25904 100644
--- a/drivers/gpu/drm/ttm/ttm_memory.c
+++ b/drivers/gpu/drm/ttm/ttm_memory.c
@@ -209,37 +209,34 @@ static bool ttm_zones_above_swap_target(struct ttm_mem_global *glob,
* Note that this function is reentrant:
* many threads may try to swap out at any given time.
*/
-
static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
- uint64_t extra)
+ uint64_t extra, struct ttm_operation_ctx *ctx)
{
- int ret;
- struct ttm_mem_shrink *shrink;
+ int ret;
- spin_lock(&glob->lock);
- if (glob->shrink == NULL)
- goto out;
-
- while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
- shrink = glob->shrink;
- spin_unlock(&glob->lock);
- ret = shrink->do_shrink(shrink);
- spin_lock(&glob->lock);
- if (unlikely(ret != 0))
- goto out;
- }
-out:
- spin_unlock(&glob->lock);
-}
+ spin_lock(&glob->lock);
+ while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
+ spin_unlock(&glob->lock);
+ ret = ttm_bo_swapout(glob->bo_glob, ctx);
+ spin_lock(&glob->lock);
+ if (unlikely(ret != 0))
+ break;
+ }
+ spin_unlock(&glob->lock);
+}
static void ttm_shrink_work(struct work_struct *work)
{
- struct ttm_mem_global *glob =
- container_of(work, struct ttm_mem_global, work);
-
- ttm_shrink(glob, true, 0ULL);
+ struct ttm_operation_ctx ctx = {
+ .interruptible = false,
+ .no_wait_gpu = false
+ };
+ struct ttm_mem_global *glob =
+ container_of(work, struct ttm_mem_global, work);
+
+ ttm_shrink(glob, true, 0ULL, &ctx);
}
static int ttm_mem_init_kernel_zone(struct ttm_mem_global *glob,
@@ -526,7 +523,7 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
return -ENOMEM;
if (unlikely(count-- == 0))
return -ENOMEM;
- ttm_shrink(glob, false, memory + (memory >> 2) + 16);
+ ttm_shrink(glob, false, memory + (memory >> 2) + 16, ctx);
}
return 0;
--
2.7.4
|
