1 /*
2 * Copyright © 2018 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #ifndef INTEL_GEM_H
25 #define INTEL_GEM_H
26
27 #include "drm-uapi/i915_drm.h"
28
29 #include <assert.h>
30 #include <errno.h>
31 #include <stdbool.h>
32 #include <stdint.h>
33 #include <stdlib.h>
34 #include <unistd.h>
35 #include <sys/ioctl.h>
36
37 static inline uint64_t
intel_canonical_address(uint64_t v)38 intel_canonical_address(uint64_t v)
39 {
40 /* From the Broadwell PRM Vol. 2a, MI_LOAD_REGISTER_MEM::MemoryAddress:
41 *
42 * "This field specifies the address of the memory location where the
43 * register value specified in the DWord above will read from. The
44 * address specifies the DWord location of the data. Range =
45 * GraphicsVirtualAddress[63:2] for a DWord register GraphicsAddress
46 * [63:48] are ignored by the HW and assumed to be in correct
47 * canonical form [63:48] == [47]."
48 */
49 const int shift = 63 - 47;
50 return (int64_t)(v << shift) >> shift;
51 }
52
53 /**
54 * This returns a 48-bit address with the high 16 bits zeroed.
55 *
56 * It's the opposite of intel_canonicalize_address.
57 */
58 static inline uint64_t
intel_48b_address(uint64_t v)59 intel_48b_address(uint64_t v)
60 {
61 const int shift = 63 - 47;
62 return (uint64_t)(v << shift) >> shift;
63 }
64
65 /**
66 * Call ioctl, restarting if it is interrupted
67 */
68 static inline int
intel_ioctl(int fd,unsigned long request,void * arg)69 intel_ioctl(int fd, unsigned long request, void *arg)
70 {
71 int ret;
72
73 do {
74 ret = ioctl(fd, request, arg);
75 } while (ret == -1 && (errno == EINTR || errno == EAGAIN));
76 return ret;
77 }
78
79 static inline uint64_t
intel_read_gpu_timestamp(int fd)80 intel_read_gpu_timestamp(int fd)
81 {
82 struct drm_i915_reg_read reg_read = {};
83 const uint64_t render_ring_timestamp = 0x2358;
84 reg_read.offset = render_ring_timestamp | I915_REG_READ_8B_WA;
85
86 if (intel_ioctl(fd, DRM_IOCTL_I915_REG_READ, ®_read) < 0)
87 return 0;
88
89 return reg_read.val;
90 }
91
92 /**
93 * A wrapper around DRM_IOCTL_I915_QUERY
94 *
95 * Unfortunately, the error semantics of this ioctl are rather annoying so
96 * it's better to have a common helper.
97 */
98 static inline int
intel_i915_query_flags(int fd,uint64_t query_id,uint32_t flags,void * buffer,int32_t * buffer_len)99 intel_i915_query_flags(int fd, uint64_t query_id, uint32_t flags,
100 void *buffer, int32_t *buffer_len)
101 {
102 struct drm_i915_query_item item = {
103 .query_id = query_id,
104 .length = *buffer_len,
105 .flags = flags,
106 .data_ptr = (uintptr_t)buffer,
107 };
108
109 struct drm_i915_query args = {
110 .num_items = 1,
111 .flags = 0,
112 .items_ptr = (uintptr_t)&item,
113 };
114
115 int ret = intel_ioctl(fd, DRM_IOCTL_I915_QUERY, &args);
116 if (ret != 0)
117 return -errno;
118 else if (item.length < 0)
119 return item.length;
120
121 *buffer_len = item.length;
122 return 0;
123 }
124
125 static inline int
intel_i915_query(int fd,uint64_t query_id,void * buffer,int32_t * buffer_len)126 intel_i915_query(int fd, uint64_t query_id, void *buffer,
127 int32_t *buffer_len)
128 {
129 return intel_i915_query_flags(fd, query_id, 0, buffer, buffer_len);
130 }
131
132 /**
133 * Query for the given data, allocating as needed
134 *
135 * The caller is responsible for freeing the returned pointer.
136 */
137 static inline void *
intel_i915_query_alloc(int fd,uint64_t query_id,int32_t * query_length)138 intel_i915_query_alloc(int fd, uint64_t query_id, int32_t *query_length)
139 {
140 if (query_length)
141 *query_length = 0;
142
143 int32_t length = 0;
144 int ret = intel_i915_query(fd, query_id, NULL, &length);
145 if (ret < 0)
146 return NULL;
147
148 void *data = calloc(1, length);
149 assert(data != NULL); /* This shouldn't happen in practice */
150 if (data == NULL)
151 return NULL;
152
153 ret = intel_i915_query(fd, query_id, data, &length);
154 assert(ret == 0); /* We should have caught the error above */
155 if (ret < 0) {
156 free(data);
157 return NULL;
158 }
159
160 if (query_length)
161 *query_length = length;
162
163 return data;
164 }
165
166 bool intel_gem_supports_syncobj_wait(int fd);
167
168 int intel_gem_count_engines(const struct drm_i915_query_engine_info *info,
169 enum drm_i915_gem_engine_class engine_class);
170 int intel_gem_create_context_engines(int fd,
171 const struct drm_i915_query_engine_info *info,
172 int num_engines, uint16_t *engine_classes);
173
174 #endif /* INTEL_GEM_H */
175