1 /**************************************************************************
3 * Copyright 2018-2019 Alyssa Rosenzweig
4 * Copyright 2018-2019 Collabora, Ltd.
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:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
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.
27 **************************************************************************/
33 #include "renderonly/renderonly.h"
34 #include "util/u_dynarray.h"
35 #include "util/bitset.h"
37 #include "util/list.h"
39 #include <panfrost-misc.h>
42 #define PAN_MAX_CONST_BUFFERS 16
44 /* Transient slab size. This is a balance between fragmentation against cache
45 * locality and ease of bookkeeping */
47 #define TRANSIENT_SLAB_PAGES (32) /* 128kb */
48 #define TRANSIENT_SLAB_SIZE (4096 * TRANSIENT_SLAB_PAGES)
50 /* Maximum number of transient slabs so we don't need dynamic arrays. Most
51 * interesting Mali boards are 4GB RAM max, so if the entire RAM was filled
52 * with transient slabs, you could never exceed (4GB / TRANSIENT_SLAB_SIZE)
53 * allocations anyway. By capping, we can use a fixed-size bitset for tracking
54 * free slabs, eliminating quite a bit of complexity. We can pack the free
55 * state of 8 slabs into a single byte, so for 128kb transient slabs the bitset
56 * occupies a cheap 4kb of memory */
58 #define MAX_TRANSIENT_SLABS (1024*1024 / TRANSIENT_SLAB_PAGES)
60 /* How many power-of-two levels in the BO cache do we want? 2^12
61 * minimum chosen as it is the page size that all allocations are
64 #define MIN_BO_CACHE_BUCKET (12) /* 2^12 = 4KB */
65 #define MAX_BO_CACHE_BUCKET (22) /* 2^22 = 4MB */
67 /* Fencepost problem, hence the off-by-one */
68 #define NR_BO_CACHE_BUCKETS (MAX_BO_CACHE_BUCKET - MIN_BO_CACHE_BUCKET + 1)
70 struct panfrost_device
{
76 /* Properties of the GPU in use */
79 unsigned thread_tls_alloc
;
82 drmVersionPtr kernel_version
;
84 struct renderonly
*ro
;
86 pthread_mutex_t active_bos_lock
;
87 struct set
*active_bos
;
92 /* List containing all cached BOs sorted in LRU (Least
93 * Recently Used) order. This allows us to quickly evict BOs
94 * that are more than 1 second old.
98 /* The BO cache is a set of buckets with power-of-two sizes
99 * ranging from 2^12 (4096, the page size) to
100 * 2^(12 + MAX_BO_CACHE_BUCKETS).
101 * Each bucket is a linked list of free panfrost_bo objects. */
103 struct list_head buckets
[NR_BO_CACHE_BUCKETS
];
108 panfrost_open_device(void *memctx
, int fd
, struct panfrost_device
*dev
);
111 panfrost_close_device(struct panfrost_device
*dev
);