1 /**************************************************************************
2 *
3 * Copyright 2018-2019 Alyssa Rosenzweig
4 * Copyright 2018-2019 Collabora, Ltd.
5 * Copyright © 2015 Intel Corporation
6 * All Rights Reserved.
7 *
8 * Permission is hereby granted, free of charge, to any person obtaining a
9 * copy of this software and associated documentation files (the
10 * "Software"), to deal in the Software without restriction, including
11 * without limitation the rights to use, copy, modify, merge, publish,
12 * distribute, sub license, and/or sell copies of the Software, and to
13 * permit persons to whom the Software is furnished to do so, subject to
14 * the following conditions:
15 *
16 * The above copyright notice and this permission notice (including the
17 * next paragraph) shall be included in all copies or substantial portions
18 * of the Software.
19 *
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
21 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
23 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
24 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
25 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
26 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 *
28 **************************************************************************/
29
30 #ifndef PAN_DEVICE_H
31 #define PAN_DEVICE_H
32
33 #include <xf86drm.h>
34 #include "renderonly/renderonly.h"
35 #include "util/bitset.h"
36 #include "util/list.h"
37 #include "util/sparse_array.h"
38 #include "util/timespec.h"
39 #include "util/u_dynarray.h"
40
41 #include "panfrost/util/pan_ir.h"
42 #include "pan_blend.h"
43 #include "pan_blitter.h"
44 #include "pan_indirect_dispatch.h"
45 #include "pan_pool.h"
46 #include "pan_props.h"
47 #include "pan_util.h"
48
49 #include "kmod/pan_kmod.h"
50
51 #include <genxml/gen_macros.h>
52
53 #if defined(__cplusplus)
54 extern "C" {
55 #endif
56
57 /* Always reserve the lower 32MB */
58 #define PAN_VA_USER_START 0x2000000ull
59
60 /* Max address space size allowed */
61 #define PAN_VA_USER_END (1ull << 48ull)
62
63 /* Driver limits */
64 #define PAN_MAX_CONST_BUFFERS 16
65
66 /* Mali hardware can texture up to 65536 x 65536 x 65536 and render up to 16384
67 * x 16384, but 8192 x 8192 should be enough for anyone. The OpenGL game
68 * "Cathedral" requires a texture of width 8192 to start.
69 */
70 #define PAN_MAX_MIP_LEVELS 14
71
72 #define PAN_MAX_TEXEL_BUFFER_ELEMENTS 65536
73
74 /* How many power-of-two levels in the BO cache do we want? 2^12
75 * minimum chosen as it is the page size that all allocations are
76 * rounded to */
77
78 #define MIN_BO_CACHE_BUCKET (12) /* 2^12 = 4KB */
79 #define MAX_BO_CACHE_BUCKET (22) /* 2^22 = 4MB */
80
81 /* Fencepost problem, hence the off-by-one */
82 #define NR_BO_CACHE_BUCKETS (MAX_BO_CACHE_BUCKET - MIN_BO_CACHE_BUCKET + 1)
83
84 struct panfrost_device {
85 /* For ralloc */
86 void *memctx;
87
88 /* Kmod objects. */
89 struct {
90 /* The pan_kmod_dev object backing this device. */
91 struct pan_kmod_dev *dev;
92
93 /* Cached pan_kmod_dev_props properties queried at device create time. */
94 struct pan_kmod_dev_props props;
95
96 /* VM attached to this device. */
97 struct pan_kmod_vm *vm;
98 } kmod;
99
100 /* For pandecode */
101 struct pandecode_context *decode_ctx;
102
103 /* Properties of the GPU in use */
104 unsigned arch;
105
106 /* Number of shader cores */
107 unsigned core_count;
108
109 /* Range of core IDs, equal to the maximum core ID + 1. Satisfies
110 * core_id_range >= core_count.
111 */
112 unsigned core_id_range;
113
114 /* Maximum tilebuffer size in bytes for optimal performance. */
115 unsigned optimal_tib_size;
116
117 unsigned thread_tls_alloc;
118 struct panfrost_tiler_features tiler_features;
119 const struct panfrost_model *model;
120 bool has_afbc;
121 bool has_afrc;
122
123 /* Table of formats, indexed by a PIPE format */
124 const struct panfrost_format *formats;
125 const struct pan_blendable_format *blendable_formats;
126
127 /* Bitmask of supported compressed texture formats */
128 uint32_t compressed_formats;
129
130 /* debug flags, see pan_util.h how to interpret */
131 unsigned debug;
132
133 struct renderonly *ro;
134
135 pthread_mutex_t bo_map_lock;
136 struct util_sparse_array bo_map;
137
138 struct {
139 pthread_mutex_t lock;
140
141 /* List containing all cached BOs sorted in LRU (Least
142 * Recently Used) order. This allows us to quickly evict BOs
143 * that are more than 1 second old.
144 */
145 struct list_head lru;
146
147 /* The BO cache is a set of buckets with power-of-two sizes
148 * ranging from 2^12 (4096, the page size) to
149 * 2^(12 + MAX_BO_CACHE_BUCKETS).
150 * Each bucket is a linked list of free panfrost_bo objects. */
151
152 struct list_head buckets[NR_BO_CACHE_BUCKETS];
153 } bo_cache;
154
155 struct pan_blitter_cache blitter;
156 struct pan_blend_shader_cache blend_shaders;
157 struct pan_indirect_dispatch_meta indirect_dispatch;
158
159 /* Tiler heap shared across all tiler jobs, allocated against the
160 * device since there's only a single tiler. Since this is invisible to
161 * the CPU, it's okay for multiple contexts to reference it
162 * simultaneously; by keeping on the device struct, we eliminate a
163 * costly per-context allocation. */
164
165 struct panfrost_bo *tiler_heap;
166
167 /* The tiler heap is shared by all contexts, and is written by tiler
168 * jobs and read by fragment job. We need to ensure that a
169 * vertex/tiler job chain from one context is not inserted between
170 * the vertex/tiler and fragment job of another context, otherwise
171 * we end up with tiler heap corruption.
172 */
173 pthread_mutex_t submit_lock;
174
175 /* Sample positions are preloaded into a write-once constant buffer,
176 * such that they can be referenced fore free later. Needed
177 * unconditionally on Bifrost, and useful for sharing with Midgard */
178
179 struct panfrost_bo *sample_positions;
180 };
181
182 static inline int
panfrost_device_fd(const struct panfrost_device * dev)183 panfrost_device_fd(const struct panfrost_device *dev)
184 {
185 return dev->kmod.dev->fd;
186 }
187
188 static inline uint32_t
panfrost_device_gpu_id(const struct panfrost_device * dev)189 panfrost_device_gpu_id(const struct panfrost_device *dev)
190 {
191 return dev->kmod.props.gpu_prod_id;
192 }
193
194 static inline uint32_t
panfrost_device_gpu_rev(const struct panfrost_device * dev)195 panfrost_device_gpu_rev(const struct panfrost_device *dev)
196 {
197 return dev->kmod.props.gpu_revision;
198 }
199
200 static inline int
panfrost_device_kmod_version_major(const struct panfrost_device * dev)201 panfrost_device_kmod_version_major(const struct panfrost_device *dev)
202 {
203 return dev->kmod.dev->driver.version.major;
204 }
205
206 static inline int
panfrost_device_kmod_version_minor(const struct panfrost_device * dev)207 panfrost_device_kmod_version_minor(const struct panfrost_device *dev)
208 {
209 return dev->kmod.dev->driver.version.minor;
210 }
211
212 void panfrost_open_device(void *memctx, int fd, struct panfrost_device *dev);
213
214 void panfrost_close_device(struct panfrost_device *dev);
215
216 bool panfrost_supports_compressed_format(struct panfrost_device *dev,
217 unsigned fmt);
218
219 static inline struct panfrost_bo *
pan_lookup_bo(struct panfrost_device * dev,uint32_t gem_handle)220 pan_lookup_bo(struct panfrost_device *dev, uint32_t gem_handle)
221 {
222 return (struct panfrost_bo *)util_sparse_array_get(&dev->bo_map, gem_handle);
223 }
224
225 static inline bool
pan_is_bifrost(const struct panfrost_device * dev)226 pan_is_bifrost(const struct panfrost_device *dev)
227 {
228 return dev->arch >= 6 && dev->arch <= 7;
229 }
230
231 static inline uint64_t
pan_gpu_time_to_ns(struct panfrost_device * dev,uint64_t gpu_time)232 pan_gpu_time_to_ns(struct panfrost_device *dev, uint64_t gpu_time)
233 {
234 return (gpu_time * NSEC_PER_SEC) / dev->kmod.props.timestamp_frequency;
235 }
236
237 #if defined(__cplusplus)
238 } // extern "C"
239 #endif
240
241 #endif
242