1 | /* SPDX-License-Identifier: MIT */ |
2 | /* |
3 | * Copyright © 2016 Intel Corporation |
4 | * |
5 | * Permission is hereby granted, free of charge, to any person obtaining a |
6 | * copy of this software and associated documentation files (the "Software"), |
7 | * to deal in the Software without restriction, including without limitation |
8 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
9 | * and/or sell copies of the Software, and to permit persons to whom the |
10 | * Software is furnished to do so, subject to the following conditions: |
11 | * |
12 | * The above copyright notice and this permission notice (including the next |
13 | * paragraph) shall be included in all copies or substantial portions of the |
14 | * Software. |
15 | * |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING |
21 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS |
22 | * IN THE SOFTWARE. |
23 | * |
24 | */ |
25 | |
26 | #ifndef __I915_VMA_TYPES_H__ |
27 | #define __I915_VMA_TYPES_H__ |
28 | |
29 | #include <linux/rbtree.h> |
30 | |
31 | #include <drm/drm_mm.h> |
32 | |
33 | #include "gem/i915_gem_object_types.h" |
34 | |
35 | /** |
36 | * DOC: Global GTT views |
37 | * |
38 | * Background and previous state |
39 | * |
40 | * Historically objects could exists (be bound) in global GTT space only as |
41 | * singular instances with a view representing all of the object's backing pages |
42 | * in a linear fashion. This view will be called a normal view. |
43 | * |
44 | * To support multiple views of the same object, where the number of mapped |
45 | * pages is not equal to the backing store, or where the layout of the pages |
46 | * is not linear, concept of a GGTT view was added. |
47 | * |
48 | * One example of an alternative view is a stereo display driven by a single |
49 | * image. In this case we would have a framebuffer looking like this |
50 | * (2x2 pages): |
51 | * |
52 | * 12 |
53 | * 34 |
54 | * |
55 | * Above would represent a normal GGTT view as normally mapped for GPU or CPU |
56 | * rendering. In contrast, fed to the display engine would be an alternative |
57 | * view which could look something like this: |
58 | * |
59 | * 1212 |
60 | * 3434 |
61 | * |
62 | * In this example both the size and layout of pages in the alternative view is |
63 | * different from the normal view. |
64 | * |
65 | * Implementation and usage |
66 | * |
67 | * GGTT views are implemented using VMAs and are distinguished via enum |
68 | * i915_gtt_view_type and struct i915_gtt_view. |
69 | * |
70 | * A new flavour of core GEM functions which work with GGTT bound objects were |
71 | * added with the _ggtt_ infix, and sometimes with _view postfix to avoid |
72 | * renaming in large amounts of code. They take the struct i915_gtt_view |
73 | * parameter encapsulating all metadata required to implement a view. |
74 | * |
75 | * As a helper for callers which are only interested in the normal view, |
76 | * globally const i915_gtt_view_normal singleton instance exists. All old core |
77 | * GEM API functions, the ones not taking the view parameter, are operating on, |
78 | * or with the normal GGTT view. |
79 | * |
80 | * Code wanting to add or use a new GGTT view needs to: |
81 | * |
82 | * 1. Add a new enum with a suitable name. |
83 | * 2. Extend the metadata in the i915_gtt_view structure if required. |
84 | * 3. Add support to i915_get_vma_pages(). |
85 | * |
86 | * New views are required to build a scatter-gather table from within the |
87 | * i915_get_vma_pages function. This table is stored in the vma.gtt_view and |
88 | * exists for the lifetime of an VMA. |
89 | * |
90 | * Core API is designed to have copy semantics which means that passed in |
91 | * struct i915_gtt_view does not need to be persistent (left around after |
92 | * calling the core API functions). |
93 | * |
94 | */ |
95 | |
96 | struct i915_vma_resource; |
97 | |
98 | struct intel_remapped_plane_info { |
99 | /* in gtt pages */ |
100 | u32 offset:31; |
101 | u32 linear:1; |
102 | union { |
103 | /* in gtt pages for !linear */ |
104 | struct { |
105 | u16 width; |
106 | u16 height; |
107 | u16 src_stride; |
108 | u16 dst_stride; |
109 | }; |
110 | |
111 | /* in gtt pages for linear */ |
112 | u32 size; |
113 | }; |
114 | } __packed; |
115 | |
116 | struct intel_remapped_info { |
117 | struct intel_remapped_plane_info plane[4]; |
118 | /* in gtt pages */ |
119 | u32 plane_alignment; |
120 | } __packed; |
121 | |
122 | struct intel_rotation_info { |
123 | struct intel_remapped_plane_info plane[2]; |
124 | } __packed; |
125 | |
126 | struct intel_partial_info { |
127 | u64 offset; |
128 | unsigned int size; |
129 | } __packed; |
130 | |
131 | enum i915_gtt_view_type { |
132 | I915_GTT_VIEW_NORMAL = 0, |
133 | I915_GTT_VIEW_ROTATED = sizeof(struct intel_rotation_info), |
134 | I915_GTT_VIEW_PARTIAL = sizeof(struct intel_partial_info), |
135 | I915_GTT_VIEW_REMAPPED = sizeof(struct intel_remapped_info), |
136 | }; |
137 | |
138 | static inline void assert_i915_gem_gtt_types(void) |
139 | { |
140 | BUILD_BUG_ON(sizeof(struct intel_rotation_info) != 2 * sizeof(u32) + 8 * sizeof(u16)); |
141 | BUILD_BUG_ON(sizeof(struct intel_partial_info) != sizeof(u64) + sizeof(unsigned int)); |
142 | BUILD_BUG_ON(sizeof(struct intel_remapped_info) != 5 * sizeof(u32) + 16 * sizeof(u16)); |
143 | |
144 | /* Check that rotation/remapped shares offsets for simplicity */ |
145 | BUILD_BUG_ON(offsetof(struct intel_remapped_info, plane[0]) != |
146 | offsetof(struct intel_rotation_info, plane[0])); |
147 | BUILD_BUG_ON(offsetofend(struct intel_remapped_info, plane[1]) != |
148 | offsetofend(struct intel_rotation_info, plane[1])); |
149 | |
150 | /* As we encode the size of each branch inside the union into its type, |
151 | * we have to be careful that each branch has a unique size. |
152 | */ |
153 | switch ((enum i915_gtt_view_type)0) { |
154 | case I915_GTT_VIEW_NORMAL: |
155 | case I915_GTT_VIEW_PARTIAL: |
156 | case I915_GTT_VIEW_ROTATED: |
157 | case I915_GTT_VIEW_REMAPPED: |
158 | /* gcc complains if these are identical cases */ |
159 | break; |
160 | } |
161 | } |
162 | |
163 | struct i915_gtt_view { |
164 | enum i915_gtt_view_type type; |
165 | union { |
166 | /* Members need to contain no holes/padding */ |
167 | struct intel_partial_info partial; |
168 | struct intel_rotation_info rotated; |
169 | struct intel_remapped_info remapped; |
170 | }; |
171 | }; |
172 | |
173 | /** |
174 | * DOC: Virtual Memory Address |
175 | * |
176 | * A VMA represents a GEM BO that is bound into an address space. Therefore, a |
177 | * VMA's presence cannot be guaranteed before binding, or after unbinding the |
178 | * object into/from the address space. |
179 | * |
180 | * To make things as simple as possible (ie. no refcounting), a VMA's lifetime |
181 | * will always be <= an objects lifetime. So object refcounting should cover us. |
182 | */ |
183 | struct i915_vma { |
184 | struct drm_mm_node node; |
185 | |
186 | struct i915_address_space *vm; |
187 | const struct i915_vma_ops *ops; |
188 | |
189 | struct drm_i915_gem_object *obj; |
190 | |
191 | struct sg_table *pages; |
192 | void __iomem *iomap; |
193 | void *private; /* owned by creator */ |
194 | |
195 | struct i915_fence_reg *fence; |
196 | |
197 | u64 size; |
198 | struct i915_page_sizes page_sizes; |
199 | |
200 | /* mmap-offset associated with fencing for this vma */ |
201 | struct i915_mmap_offset *mmo; |
202 | |
203 | u32 guard; /* padding allocated around vma->pages within the node */ |
204 | u32 fence_size; |
205 | u32 fence_alignment; |
206 | u32 display_alignment; |
207 | |
208 | /** |
209 | * Count of the number of times this vma has been opened by different |
210 | * handles (but same file) for execbuf, i.e. the number of aliases |
211 | * that exist in the ctx->handle_vmas LUT for this vma. |
212 | */ |
213 | atomic_t open_count; |
214 | atomic_t flags; |
215 | /** |
216 | * How many users have pinned this object in GTT space. |
217 | * |
218 | * This is a tightly bound, fairly small number of users, so we |
219 | * stuff inside the flags field so that we can both check for overflow |
220 | * and detect a no-op i915_vma_pin() in a single check, while also |
221 | * pinning the vma. |
222 | * |
223 | * The worst case display setup would have the same vma pinned for |
224 | * use on each plane on each crtc, while also building the next atomic |
225 | * state and holding a pin for the length of the cleanup queue. In the |
226 | * future, the flip queue may be increased from 1. |
227 | * Estimated worst case: 3 [qlen] * 4 [max crtcs] * 7 [max planes] = 84 |
228 | * |
229 | * For GEM, the number of concurrent users for pwrite/pread is |
230 | * unbounded. For execbuffer, it is currently one but will in future |
231 | * be extended to allow multiple clients to pin vma concurrently. |
232 | * |
233 | * We also use suballocated pages, with each suballocation claiming |
234 | * its own pin on the shared vma. At present, this is limited to |
235 | * exclusive cachelines of a single page, so a maximum of 64 possible |
236 | * users. |
237 | */ |
238 | #define I915_VMA_PIN_MASK 0x3ff |
239 | #define I915_VMA_OVERFLOW 0x200 |
240 | |
241 | /** Flags and address space this VMA is bound to */ |
242 | #define I915_VMA_GLOBAL_BIND_BIT 10 |
243 | #define I915_VMA_LOCAL_BIND_BIT 11 |
244 | |
245 | #define I915_VMA_GLOBAL_BIND ((int)BIT(I915_VMA_GLOBAL_BIND_BIT)) |
246 | #define I915_VMA_LOCAL_BIND ((int)BIT(I915_VMA_LOCAL_BIND_BIT)) |
247 | |
248 | #define I915_VMA_BIND_MASK (I915_VMA_GLOBAL_BIND | I915_VMA_LOCAL_BIND) |
249 | |
250 | #define I915_VMA_ERROR_BIT 12 |
251 | #define I915_VMA_ERROR ((int)BIT(I915_VMA_ERROR_BIT)) |
252 | |
253 | #define I915_VMA_GGTT_BIT 13 |
254 | #define I915_VMA_CAN_FENCE_BIT 14 |
255 | #define I915_VMA_USERFAULT_BIT 15 |
256 | #define I915_VMA_GGTT_WRITE_BIT 16 |
257 | |
258 | #define I915_VMA_GGTT ((int)BIT(I915_VMA_GGTT_BIT)) |
259 | #define I915_VMA_CAN_FENCE ((int)BIT(I915_VMA_CAN_FENCE_BIT)) |
260 | #define I915_VMA_USERFAULT ((int)BIT(I915_VMA_USERFAULT_BIT)) |
261 | #define I915_VMA_GGTT_WRITE ((int)BIT(I915_VMA_GGTT_WRITE_BIT)) |
262 | |
263 | #define I915_VMA_SCANOUT_BIT 17 |
264 | #define I915_VMA_SCANOUT ((int)BIT(I915_VMA_SCANOUT_BIT)) |
265 | |
266 | struct i915_active active; |
267 | |
268 | #define I915_VMA_PAGES_BIAS 24 |
269 | #define I915_VMA_PAGES_ACTIVE (BIT(24) | 1) |
270 | atomic_t pages_count; /* number of active binds to the pages */ |
271 | |
272 | /** |
273 | * Whether we hold a reference on the vm dma_resv lock to temporarily |
274 | * block vm freeing until the vma is destroyed. |
275 | * Protected by the vm mutex. |
276 | */ |
277 | bool vm_ddestroy; |
278 | |
279 | /** |
280 | * Support different GGTT views into the same object. |
281 | * This means there can be multiple VMA mappings per object and per VM. |
282 | * i915_gtt_view_type is used to distinguish between those entries. |
283 | * The default one of zero (I915_GTT_VIEW_NORMAL) is default and also |
284 | * assumed in GEM functions which take no ggtt view parameter. |
285 | */ |
286 | struct i915_gtt_view gtt_view; |
287 | |
288 | /** This object's place on the active/inactive lists */ |
289 | struct list_head vm_link; |
290 | |
291 | struct list_head obj_link; /* Link in the object's VMA list */ |
292 | struct rb_node obj_node; |
293 | struct hlist_node obj_hash; |
294 | |
295 | /** This vma's place in the eviction list */ |
296 | struct list_head evict_link; |
297 | |
298 | struct list_head closed_link; |
299 | |
300 | /** The async vma resource. Protected by the vm_mutex */ |
301 | struct i915_vma_resource *resource; |
302 | }; |
303 | |
304 | #endif |
305 | |