1 /*
2 * Copyright © 2021 Raspberry Pi Ltd
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 /**
25 * Gallium query object support for performance counters
26 *
27 * This contains the performance V3D counters queries.
28 */
29
30 #include "v3d_query.h"
31
32 #include "common/v3d_performance_counters.h"
33
34 struct v3d_query_perfcnt
35 {
36 struct v3d_query base;
37
38 unsigned num_queries;
39 struct v3d_perfmon_state *perfmon;
40 };
41
42 static void
kperfmon_destroy(struct v3d_context * v3d,struct v3d_perfmon_state * perfmon)43 kperfmon_destroy(struct v3d_context *v3d, struct v3d_perfmon_state *perfmon)
44 {
45 struct drm_v3d_perfmon_destroy destroyreq;
46
47 destroyreq.id = perfmon->kperfmon_id;
48 int ret = v3d_ioctl(v3d->fd, DRM_IOCTL_V3D_PERFMON_DESTROY, &destroyreq);
49 if (ret != 0)
50 fprintf(stderr, "failed to destroy perfmon %d: %s\n",
51 perfmon->kperfmon_id, strerror(errno));
52 }
53
54 int
v3dX(get_driver_query_group_info_perfcnt)55 v3dX(get_driver_query_group_info_perfcnt)(struct v3d_screen *screen, unsigned index,
56 struct pipe_driver_query_group_info *info)
57 {
58 if (!screen->has_perfmon)
59 return 0;
60
61 if (!info)
62 return 1;
63
64 if (index > 0)
65 return 0;
66
67 info->name = "V3D counters";
68 info->max_active_queries = DRM_V3D_MAX_PERF_COUNTERS;
69 info->num_queries = ARRAY_SIZE(v3d_performance_counters);
70
71 return 1;
72 }
73
74 int
v3dX(get_driver_query_info_perfcnt)75 v3dX(get_driver_query_info_perfcnt)(struct v3d_screen *screen, unsigned index,
76 struct pipe_driver_query_info *info)
77 {
78 if (!screen->has_perfmon)
79 return 0;
80
81 if (!info)
82 return ARRAY_SIZE(v3d_performance_counters);
83
84 if (index >= ARRAY_SIZE(v3d_performance_counters))
85 return 0;
86
87 info->group_id = 0;
88 info->name = v3d_performance_counters[index][V3D_PERFCNT_NAME];
89 info->query_type = PIPE_QUERY_DRIVER_SPECIFIC + index;
90 info->result_type = PIPE_DRIVER_QUERY_RESULT_TYPE_CUMULATIVE;
91 info->type = PIPE_DRIVER_QUERY_TYPE_UINT64;
92 info->flags = PIPE_DRIVER_QUERY_FLAG_BATCH;
93
94 return 1;
95 }
96
97 static void
v3d_destroy_query_perfcnt(struct v3d_context * v3d,struct v3d_query * query)98 v3d_destroy_query_perfcnt(struct v3d_context *v3d, struct v3d_query *query)
99 {
100 struct v3d_query_perfcnt *pquery = (struct v3d_query_perfcnt *)query;
101
102 assert(pquery->perfmon);
103
104 if (v3d->active_perfmon == pquery->perfmon) {
105 fprintf(stderr, "Query is active; end query before destroying\n");
106 return;
107 }
108 if (pquery->perfmon->kperfmon_id)
109 kperfmon_destroy(v3d, pquery->perfmon);
110
111 v3d_fence_unreference(&pquery->perfmon->last_job_fence);
112 free(pquery->perfmon);
113 free(query);
114 }
115
116 static bool
v3d_begin_query_perfcnt(struct v3d_context * v3d,struct v3d_query * query)117 v3d_begin_query_perfcnt(struct v3d_context *v3d, struct v3d_query *query)
118 {
119 struct v3d_query_perfcnt *pquery = (struct v3d_query_perfcnt *)query;
120 struct drm_v3d_perfmon_create createreq = { 0 };
121 int i, ret;
122
123 /* Only one perfmon can be activated per context */
124 if (v3d->active_perfmon) {
125 fprintf(stderr,
126 "Another query is already active; "
127 "finish it before starting a new one\n");
128 return false;
129 }
130
131 assert(pquery->perfmon);
132
133 /* Reset the counters by destroying the previously allocated perfmon */
134 if (pquery->perfmon->kperfmon_id)
135 kperfmon_destroy(v3d, pquery->perfmon);
136
137 for (i = 0; i < pquery->num_queries; i++)
138 createreq.counters[i] = pquery->perfmon->counters[i];
139
140 createreq.ncounters = pquery->num_queries;
141 ret = v3d_ioctl(v3d->fd, DRM_IOCTL_V3D_PERFMON_CREATE, &createreq);
142 if (ret != 0)
143 return false;
144
145 pquery->perfmon->kperfmon_id = createreq.id;
146 pquery->perfmon->job_submitted = false;
147 v3d_fence_unreference(&pquery->perfmon->last_job_fence);
148
149 /* Ensure all pending jobs are flushed before activating the
150 * perfmon
151 */
152 v3d_flush((struct pipe_context *)v3d);
153 v3d->active_perfmon = pquery->perfmon;
154
155 return true;
156 }
157
158 static bool
v3d_end_query_perfcnt(struct v3d_context * v3d,struct v3d_query * query)159 v3d_end_query_perfcnt(struct v3d_context *v3d, struct v3d_query *query)
160 {
161 struct v3d_query_perfcnt *pquery = (struct v3d_query_perfcnt *)query;
162
163 assert(pquery->perfmon);
164
165 if (v3d->active_perfmon != pquery->perfmon) {
166 fprintf(stderr, "This query is not active\n");
167 return false;
168 }
169
170 /* Ensure all pending jobs are flushed before deactivating the
171 * perfmon
172 */
173 v3d_flush((struct pipe_context *)v3d);
174
175 /* Get a copy of latest submitted job's fence to wait for its
176 * completion
177 */
178 if (v3d->active_perfmon->job_submitted) {
179 int fd = -1;
180 drmSyncobjExportSyncFile(v3d->fd, v3d->out_sync, &fd);
181 if (fd == -1) {
182 fprintf(stderr, "export failed\n");
183 v3d->active_perfmon->last_job_fence = NULL;
184 } else {
185 v3d->active_perfmon->last_job_fence =
186 v3d_fence_create(v3d, fd);
187 }
188 }
189
190 v3d->active_perfmon = NULL;
191
192 return true;
193 }
194
195 static bool
v3d_get_query_result_perfcnt(struct v3d_context * v3d,struct v3d_query * query,bool wait,union pipe_query_result * vresult)196 v3d_get_query_result_perfcnt(struct v3d_context *v3d, struct v3d_query *query,
197 bool wait, union pipe_query_result *vresult)
198 {
199 struct v3d_query_perfcnt *pquery = (struct v3d_query_perfcnt *)query;
200 struct drm_v3d_perfmon_get_values req = { 0 };
201 int i, ret;
202
203 assert(pquery->perfmon);
204
205 if (pquery->perfmon->job_submitted) {
206 if (!v3d_fence_wait(v3d->screen,
207 pquery->perfmon->last_job_fence,
208 wait ? OS_TIMEOUT_INFINITE : 0))
209 return false;
210
211 req.id = pquery->perfmon->kperfmon_id;
212 req.values_ptr = (uintptr_t)pquery->perfmon->values;
213 ret = v3d_ioctl(v3d->fd, DRM_IOCTL_V3D_PERFMON_GET_VALUES, &req);
214 if (ret != 0) {
215 fprintf(stderr, "Can't request perfmon counters values\n");
216 return false;
217 }
218 }
219
220 for (i = 0; i < pquery->num_queries; i++)
221 vresult->batch[i].u64 = pquery->perfmon->values[i];
222
223 return true;
224 }
225
226 static const struct v3d_query_funcs perfcnt_query_funcs = {
227 .destroy_query = v3d_destroy_query_perfcnt,
228 .begin_query = v3d_begin_query_perfcnt,
229 .end_query = v3d_end_query_perfcnt,
230 .get_query_result = v3d_get_query_result_perfcnt,
231 };
232
233 struct pipe_query *
v3dX(create_batch_query_perfcnt)234 v3dX(create_batch_query_perfcnt)(struct v3d_context *v3d, unsigned num_queries,
235 unsigned *query_types)
236 {
237 struct v3d_query_perfcnt *pquery = NULL;
238 struct v3d_query *query;
239 struct v3d_perfmon_state *perfmon = NULL;
240 int i;
241
242 /* Validate queries */
243 for (i = 0; i < num_queries; i++) {
244 if (query_types[i] < PIPE_QUERY_DRIVER_SPECIFIC ||
245 query_types[i] >= PIPE_QUERY_DRIVER_SPECIFIC +
246 ARRAY_SIZE(v3d_performance_counters)) {
247 fprintf(stderr, "Invalid query type\n");
248 return NULL;
249 }
250 }
251
252 pquery = calloc(1, sizeof(*pquery));
253 if (!pquery)
254 return NULL;
255
256 perfmon = calloc(1, sizeof(*perfmon));
257 if (!perfmon) {
258 free(pquery);
259 return NULL;
260 }
261
262 for (i = 0; i < num_queries; i++)
263 perfmon->counters[i] = query_types[i] - PIPE_QUERY_DRIVER_SPECIFIC;
264
265 pquery->perfmon = perfmon;
266 pquery->num_queries = num_queries;
267
268 query = &pquery->base;
269 query->funcs = &perfcnt_query_funcs;
270
271 /* Note that struct pipe_query isn't actually defined anywhere. */
272 return (struct pipe_query *)query;
273 }
274