src/gallium/drivers/panfrost/pan_screen.h

   1 /**************************************************************************
   2  *
   3  * Copyright 2018-2019 Alyssa Rosenzweig
   4  * Copyright 2018-2019 Collabora, Ltd.
   5  * All Rights Reserved.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a
   8  * copy of this software and associated documentation files (the
   9  * "Software"), to deal in the Software without restriction, including
  10  * without limitation the rights to use, copy, modify, merge, publish,
  11  * distribute, sub license, and/or sell copies of the Software, and to
  12  * permit persons to whom the Software is furnished to do so, subject to
  13  * the following conditions:
  14  *
  15  * The above copyright notice and this permission notice (including the
  16  * next paragraph) shall be included in all copies or substantial portions
  17  * of the Software.
  18  *
  19  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  20  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  21  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  22  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  23  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  24  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  25  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  26  *
  27  **************************************************************************/
  28
  29 #ifndef PAN_SCREEN_H
  30 #define PAN_SCREEN_H
  31
  32 #include "pipe/p_screen.h"
  33 #include "pipe/p_defines.h"
  34 #include "renderonly/renderonly.h"
  35 #include "util/u_dynarray.h"
  36 #include "util/bitset.h"
  37
  38 #include <panfrost-misc.h>
  39 #include "pan_allocate.h"
  40
  41 struct panfrost_context;
  42 struct panfrost_resource;
  43 struct panfrost_screen;
  44
  45 /* Driver limits */
  46 #define PAN_MAX_CONST_BUFFERS 16
  47
  48 /* Flags for allocated memory */
  49
  50 /* This memory region is executable */
  51 #define PAN_ALLOCATE_EXECUTE            (1 << 0)
  52
  53 /* This memory region should be lazily allocated and grow-on-page-fault. Must
  54  * be used in conjunction with INVISIBLE */
  55 #define PAN_ALLOCATE_GROWABLE           (1 << 1)
  56
  57 /* This memory region should not be mapped to the CPU */
  58 #define PAN_ALLOCATE_INVISIBLE          (1 << 2)
  59
  60 /* This memory region will be used for varyings and needs to have the cache
  61  * bits twiddled accordingly */
  62 #define PAN_ALLOCATE_COHERENT_LOCAL     (1 << 3)
  63
  64 /* This region may not be used immediately and will not mmap on allocate
  65  * (semantically distinct from INVISIBLE, which cannot never be mmaped) */
  66 #define PAN_ALLOCATE_DELAY_MMAP         (1 << 4)
  67
  68 /* Transient slab size. This is a balance between fragmentation against cache
  69  * locality and ease of bookkeeping */
  70
  71 #define TRANSIENT_SLAB_PAGES (32) /* 128kb */
  72 #define TRANSIENT_SLAB_SIZE (4096 * TRANSIENT_SLAB_PAGES)
  73
  74 /* Maximum number of transient slabs so we don't need dynamic arrays. Most
  75  * interesting Mali boards are 4GB RAM max, so if the entire RAM was filled
  76  * with transient slabs, you could never exceed (4GB / TRANSIENT_SLAB_SIZE)
  77  * allocations anyway. By capping, we can use a fixed-size bitset for tracking
  78  * free slabs, eliminating quite a bit of complexity. We can pack the free
  79  * state of 8 slabs into a single byte, so for 128kb transient slabs the bitset
  80  * occupies a cheap 4kb of memory */
  81
  82 #define MAX_TRANSIENT_SLABS (1024*1024 / TRANSIENT_SLAB_PAGES)
  83
  84 /* How many power-of-two levels in the BO cache do we want? 2^12
  85  * minimum chosen as it is the page size that all allocations are
  86  * rounded to */
  87
  88 #define MIN_BO_CACHE_BUCKET (12) /* 2^12 = 4KB */
  89 #define MAX_BO_CACHE_BUCKET (22) /* 2^22 = 4MB */
  90
  91 /* Fencepost problem, hence the off-by-one */
  92 #define NR_BO_CACHE_BUCKETS (MAX_BO_CACHE_BUCKET - MIN_BO_CACHE_BUCKET + 1)
  93
  94 struct panfrost_screen {
  95         struct pipe_screen base;
  96         int fd;
  97
  98         /* Properties of the GPU in use */
  99         unsigned gpu_id;
 100         bool require_sfbd;
 101
 102         struct renderonly *ro;
 103
 104         /* Transient memory management is based on borrowing fixed-size slabs
 105          * off the screen (loaning them out to the batch). Dynamic array
 106          * container of panfrost_bo */
 107
 108         struct util_dynarray transient_bo;
 109
 110         /* Set of free transient BOs */
 111         BITSET_DECLARE(free_transient, MAX_TRANSIENT_SLABS);
 112
 113         /* The BO cache is a set of buckets with power-of-two sizes ranging
 114          * from 2^12 (4096, the page size) to 2^(12 + MAX_BO_CACHE_BUCKETS).
 115          * Each bucket is a linked list of free panfrost_bo objects. */
 116
 117         struct list_head bo_cache[NR_BO_CACHE_BUCKETS];
 118
 119         /* While we're busy building up the job for frame N, the GPU is
 120          * still busy executing frame N-1. So hold a reference to
 121          * yesterjob */
 122         int last_fragment_flushed;
 123         struct panfrost_job *last_job;
 124 };
 125
 126 static inline struct panfrost_screen *
 127 pan_screen(struct pipe_screen *p)
 128 {
 129         return (struct panfrost_screen *)p;
 130 }
 131
 132 /* Get a transient BO off the screen given a
 133  * particular index */
 134
 135 static inline struct panfrost_bo *
 136 pan_bo_for_index(struct panfrost_screen *screen, unsigned index)
 137 {
 138         return *(util_dynarray_element(&screen->transient_bo,
 139                                 struct panfrost_bo *, index));
 140 }
 141
 142 void
 143 panfrost_drm_allocate_slab(struct panfrost_screen *screen,
 144                            struct panfrost_memory *mem,
 145                            size_t pages,
 146                            bool same_va,
 147                            int extra_flags,
 148                            int commit_count,
 149                            int extent);
 150 void
 151 panfrost_drm_free_slab(struct panfrost_screen *screen,
 152                        struct panfrost_memory *mem);
 153 struct panfrost_bo *
 154 panfrost_drm_create_bo(struct panfrost_screen *screen, size_t size,
 155                        uint32_t flags);
 156 void
 157 panfrost_drm_mmap_bo(struct panfrost_screen *screen, struct panfrost_bo *bo);
 158 void
 159 panfrost_drm_release_bo(struct panfrost_screen *screen, struct panfrost_bo *bo, bool cacheable);
 160 struct panfrost_bo *
 161 panfrost_drm_import_bo(struct panfrost_screen *screen, int fd);
 162 int
 163 panfrost_drm_export_bo(struct panfrost_screen *screen, const struct panfrost_bo *bo);
 164 int
 165 panfrost_drm_submit_vs_fs_job(struct panfrost_context *ctx, bool has_draws,
 166                               bool is_scanout);
 167 void
 168 panfrost_drm_force_flush_fragment(struct panfrost_context *ctx,
 169                                   struct pipe_fence_handle **fence);
 170 unsigned
 171 panfrost_drm_query_gpu_version(struct panfrost_screen *screen);
 172 int
 173 panfrost_drm_init_context(struct panfrost_context *ctx);
 174 void
 175 panfrost_drm_fence_reference(struct pipe_screen *screen,
 176                              struct pipe_fence_handle **ptr,
 177                              struct pipe_fence_handle *fence);
 178 boolean
 179 panfrost_drm_fence_finish(struct pipe_screen *pscreen,
 180                           struct pipe_context *ctx,
 181                           struct pipe_fence_handle *fence,
 182                           uint64_t timeout);
 183 struct panfrost_bo *
 184 panfrost_bo_cache_fetch(
 185                 struct panfrost_screen *screen,
 186                 size_t size, uint32_t flags);
 187
 188 bool
 189 panfrost_bo_cache_put(
 190                 struct panfrost_screen *screen,
 191                 struct panfrost_bo *bo);
 192
 193 void
 194 panfrost_bo_cache_evict_all(
 195                 struct panfrost_screen *screen);
 196
 197 #endif /* PAN_SCREEN_H */