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 #include "v3d_screen.h"
32
33 struct v3d_query_perfcnt
34 {
35 struct v3d_query base;
36
37 unsigned num_queries;
38 struct v3d_perfmon_state *perfmon;
39 };
40
41 static void
kperfmon_destroy(struct v3d_context * v3d,struct v3d_perfmon_state * perfmon)42 kperfmon_destroy(struct v3d_context *v3d, struct v3d_perfmon_state *perfmon)
43 {
44 struct drm_v3d_perfmon_destroy destroyreq;
45
46 destroyreq.id = perfmon->kperfmon_id;
47 int ret = v3d_ioctl(v3d->fd, DRM_IOCTL_V3D_PERFMON_DESTROY, &destroyreq);
48 if (ret != 0)
49 fprintf(stderr, "failed to destroy perfmon %d: %s\n",
50 perfmon->kperfmon_id, strerror(errno));
51 }
52
53 static void
v3d_destroy_query_perfcnt(struct v3d_context * v3d,struct v3d_query * query)54 v3d_destroy_query_perfcnt(struct v3d_context *v3d, struct v3d_query *query)
55 {
56 struct v3d_query_perfcnt *pquery = (struct v3d_query_perfcnt *)query;
57
58 assert(pquery->perfmon);
59
60 if (v3d->active_perfmon == pquery->perfmon) {
61 fprintf(stderr, "Query is active; end query before destroying\n");
62 return;
63 }
64 if (pquery->perfmon->kperfmon_id)
65 kperfmon_destroy(v3d, pquery->perfmon);
66
67 v3d_fence_unreference(&pquery->perfmon->last_job_fence);
68 free(pquery->perfmon);
69 free(query);
70 }
71
72 static bool
v3d_begin_query_perfcnt(struct v3d_context * v3d,struct v3d_query * query)73 v3d_begin_query_perfcnt(struct v3d_context *v3d, struct v3d_query *query)
74 {
75 struct v3d_query_perfcnt *pquery = (struct v3d_query_perfcnt *)query;
76 struct drm_v3d_perfmon_create createreq = { 0 };
77 int i, ret;
78
79 /* Only one perfmon can be activated per context */
80 if (v3d->active_perfmon) {
81 fprintf(stderr,
82 "Another query is already active; "
83 "finish it before starting a new one\n");
84 return false;
85 }
86
87 assert(pquery->perfmon);
88
89 /* Reset the counters by destroying the previously allocated perfmon */
90 if (pquery->perfmon->kperfmon_id)
91 kperfmon_destroy(v3d, pquery->perfmon);
92
93 for (i = 0; i < pquery->num_queries; i++)
94 createreq.counters[i] = pquery->perfmon->counters[i];
95
96 createreq.ncounters = pquery->num_queries;
97 ret = v3d_ioctl(v3d->fd, DRM_IOCTL_V3D_PERFMON_CREATE, &createreq);
98 if (ret != 0)
99 return false;
100
101 pquery->perfmon->kperfmon_id = createreq.id;
102 pquery->perfmon->job_submitted = false;
103 v3d_fence_unreference(&pquery->perfmon->last_job_fence);
104
105 /* Ensure all pending jobs are flushed before activating the
106 * perfmon
107 */
108 v3d_flush((struct pipe_context *)v3d);
109 v3d->active_perfmon = pquery->perfmon;
110
111 return true;
112 }
113
114 static bool
v3d_end_query_perfcnt(struct v3d_context * v3d,struct v3d_query * query)115 v3d_end_query_perfcnt(struct v3d_context *v3d, struct v3d_query *query)
116 {
117 struct v3d_query_perfcnt *pquery = (struct v3d_query_perfcnt *)query;
118
119 assert(pquery->perfmon);
120
121 if (v3d->active_perfmon != pquery->perfmon) {
122 fprintf(stderr, "This query is not active\n");
123 return false;
124 }
125
126 /* Ensure all pending jobs are flushed before deactivating the
127 * perfmon
128 */
129 v3d_flush((struct pipe_context *)v3d);
130
131 /* Get a copy of latest submitted job's fence to wait for its
132 * completion
133 */
134 if (v3d->active_perfmon->job_submitted) {
135 int fd = -1;
136 drmSyncobjExportSyncFile(v3d->fd, v3d->out_sync, &fd);
137 if (fd == -1) {
138 fprintf(stderr, "export failed\n");
139 v3d->active_perfmon->last_job_fence = NULL;
140 } else {
141 v3d->active_perfmon->last_job_fence =
142 v3d_fence_create(v3d, fd);
143 }
144 }
145
146 v3d->active_perfmon = NULL;
147
148 return true;
149 }
150
151 static bool
v3d_get_query_result_perfcnt(struct v3d_context * v3d,struct v3d_query * query,bool wait,union pipe_query_result * vresult)152 v3d_get_query_result_perfcnt(struct v3d_context *v3d, struct v3d_query *query,
153 bool wait, union pipe_query_result *vresult)
154 {
155 struct v3d_query_perfcnt *pquery = (struct v3d_query_perfcnt *)query;
156 struct drm_v3d_perfmon_get_values req = { 0 };
157 int i, ret;
158
159 assert(pquery->perfmon);
160
161 if (pquery->perfmon->job_submitted) {
162 if (!v3d_fence_wait(v3d->screen,
163 pquery->perfmon->last_job_fence,
164 wait ? OS_TIMEOUT_INFINITE : 0))
165 return false;
166
167 req.id = pquery->perfmon->kperfmon_id;
168 req.values_ptr = (uintptr_t)pquery->perfmon->values;
169 ret = v3d_ioctl(v3d->fd, DRM_IOCTL_V3D_PERFMON_GET_VALUES, &req);
170 if (ret != 0) {
171 fprintf(stderr, "Can't request perfmon counters values\n");
172 return false;
173 }
174 }
175
176 for (i = 0; i < pquery->num_queries; i++)
177 vresult->batch[i].u64 = pquery->perfmon->values[i];
178
179 return true;
180 }
181
182 static const struct v3d_query_funcs perfcnt_query_funcs = {
183 .destroy_query = v3d_destroy_query_perfcnt,
184 .begin_query = v3d_begin_query_perfcnt,
185 .end_query = v3d_end_query_perfcnt,
186 .get_query_result = v3d_get_query_result_perfcnt,
187 };
188
189 struct pipe_query *
v3d_create_batch_query_pipe(struct v3d_context * v3d,unsigned num_queries,unsigned * query_types)190 v3d_create_batch_query_pipe(struct v3d_context *v3d, unsigned num_queries,
191 unsigned *query_types)
192 {
193 struct v3d_screen *screen = v3d->screen;
194 struct v3d_query_perfcnt *pquery = NULL;
195 struct v3d_query *query;
196 struct v3d_perfmon_state *perfmon = NULL;
197 unsigned max_perfcnt = screen->perfcnt->max_perfcnt;
198 int i;
199
200 /* Validate queries */
201 for (i = 0; i < num_queries; i++) {
202 if (query_types[i] < PIPE_QUERY_DRIVER_SPECIFIC ||
203 query_types[i] >= PIPE_QUERY_DRIVER_SPECIFIC + max_perfcnt) {
204 fprintf(stderr, "Invalid query type\n");
205 return NULL;
206 }
207 }
208
209 pquery = calloc(1, sizeof(*pquery));
210 if (!pquery)
211 return NULL;
212
213 perfmon = calloc(1, sizeof(*perfmon));
214 if (!perfmon) {
215 free(pquery);
216 return NULL;
217 }
218
219 for (i = 0; i < num_queries; i++)
220 perfmon->counters[i] = query_types[i] - PIPE_QUERY_DRIVER_SPECIFIC;
221
222 pquery->perfmon = perfmon;
223 pquery->num_queries = num_queries;
224
225 query = &pquery->base;
226 query->funcs = &perfcnt_query_funcs;
227
228 /* Note that struct pipe_query isn't actually defined anywhere. */
229 return (struct pipe_query *)query;
230 }
231