diff options
Diffstat (limited to 'drivers/gpu/nvgpu/common/pmu/pmu_perfmon.c')
-rw-r--r-- | drivers/gpu/nvgpu/common/pmu/pmu_perfmon.c | 293 |
1 files changed, 293 insertions, 0 deletions
diff --git a/drivers/gpu/nvgpu/common/pmu/pmu_perfmon.c b/drivers/gpu/nvgpu/common/pmu/pmu_perfmon.c new file mode 100644 index 00000000..2b952868 --- /dev/null +++ b/drivers/gpu/nvgpu/common/pmu/pmu_perfmon.c | |||
@@ -0,0 +1,293 @@ | |||
1 | /* | ||
2 | * Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved. | ||
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 shall be included in | ||
12 | * all copies or substantial portions of the Software. | ||
13 | * | ||
14 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | ||
15 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | ||
16 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | ||
17 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | ||
18 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | ||
19 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER | ||
20 | * DEALINGS IN THE SOFTWARE. | ||
21 | */ | ||
22 | |||
23 | #include <nvgpu/enabled.h> | ||
24 | #include <nvgpu/pmu.h> | ||
25 | #include <nvgpu/log.h> | ||
26 | #include <nvgpu/bug.h> | ||
27 | #include <nvgpu/pmuif/nvgpu_gpmu_cmdif.h> | ||
28 | |||
29 | #include "gk20a/gk20a.h" | ||
30 | |||
31 | #ifdef CONFIG_TEGRA_19x_GPU | ||
32 | #include "nvgpu_gpuid_t19x.h" | ||
33 | #endif | ||
34 | |||
35 | static u8 get_perfmon_id(struct nvgpu_pmu *pmu) | ||
36 | { | ||
37 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
38 | u32 ver = g->params.gpu_arch + g->params.gpu_impl; | ||
39 | u8 unit_id; | ||
40 | |||
41 | switch (ver) { | ||
42 | case GK20A_GPUID_GK20A: | ||
43 | case GK20A_GPUID_GM20B: | ||
44 | case GK20A_GPUID_GM20B_B: | ||
45 | unit_id = PMU_UNIT_PERFMON; | ||
46 | break; | ||
47 | case NVGPU_GPUID_GP10B: | ||
48 | case NVGPU_GPUID_GP104: | ||
49 | case NVGPU_GPUID_GP106: | ||
50 | unit_id = PMU_UNIT_PERFMON_T18X; | ||
51 | break; | ||
52 | #if defined(CONFIG_TEGRA_19x_GPU) | ||
53 | case TEGRA_19x_GPUID: | ||
54 | unit_id = PMU_UNIT_PERFMON_T18X; | ||
55 | break; | ||
56 | #endif | ||
57 | default: | ||
58 | unit_id = PMU_UNIT_INVALID; | ||
59 | nvgpu_err(g, "no support for %x", ver); | ||
60 | WARN_ON(1); | ||
61 | } | ||
62 | |||
63 | return unit_id; | ||
64 | } | ||
65 | |||
66 | int nvgpu_pmu_init_perfmon(struct nvgpu_pmu *pmu) | ||
67 | { | ||
68 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
69 | struct pmu_v *pv = &g->ops.pmu_ver; | ||
70 | struct pmu_cmd cmd; | ||
71 | struct pmu_payload payload; | ||
72 | u32 seq; | ||
73 | |||
74 | if (!nvgpu_is_enabled(g, NVGPU_PMU_PERFMON)) | ||
75 | return 0; | ||
76 | |||
77 | nvgpu_log_fn(g, " "); | ||
78 | |||
79 | pmu->perfmon_ready = 0; | ||
80 | |||
81 | gk20a_pmu_init_perfmon_counter(g); | ||
82 | |||
83 | if (!pmu->sample_buffer) | ||
84 | pmu->sample_buffer = nvgpu_alloc(&pmu->dmem, | ||
85 | 2 * sizeof(u16)); | ||
86 | if (!pmu->sample_buffer) { | ||
87 | nvgpu_err(g, "failed to allocate perfmon sample buffer"); | ||
88 | return -ENOMEM; | ||
89 | } | ||
90 | |||
91 | /* init PERFMON */ | ||
92 | memset(&cmd, 0, sizeof(struct pmu_cmd)); | ||
93 | |||
94 | cmd.hdr.unit_id = get_perfmon_id(pmu); | ||
95 | if (cmd.hdr.unit_id == PMU_UNIT_INVALID) { | ||
96 | nvgpu_err(g, "failed to get perfmon UNIT ID, command skipped"); | ||
97 | return -EINVAL; | ||
98 | } | ||
99 | |||
100 | cmd.hdr.size = PMU_CMD_HDR_SIZE + pv->get_pmu_perfmon_cmd_init_size(); | ||
101 | cmd.cmd.perfmon.cmd_type = PMU_PERFMON_CMD_ID_INIT; | ||
102 | /* buffer to save counter values for pmu perfmon */ | ||
103 | pv->perfmon_cmd_init_set_sample_buffer(&cmd.cmd.perfmon, | ||
104 | (u16)pmu->sample_buffer); | ||
105 | /* number of sample periods below lower threshold | ||
106 | * before pmu triggers perfmon decrease event | ||
107 | * TBD: = 15 | ||
108 | */ | ||
109 | pv->perfmon_cmd_init_set_dec_cnt(&cmd.cmd.perfmon, 15); | ||
110 | /* index of base counter, aka. always ticking counter */ | ||
111 | pv->perfmon_cmd_init_set_base_cnt_id(&cmd.cmd.perfmon, 6); | ||
112 | /* microseconds interval between pmu polls perf counters */ | ||
113 | pv->perfmon_cmd_init_set_samp_period_us(&cmd.cmd.perfmon, 16700); | ||
114 | /* number of perfmon counters | ||
115 | * counter #3 (GR and CE2) for gk20a | ||
116 | */ | ||
117 | pv->perfmon_cmd_init_set_num_cnt(&cmd.cmd.perfmon, 1); | ||
118 | /* moving average window for sample periods | ||
119 | * TBD: = 3000000 / sample_period_us = 17 | ||
120 | */ | ||
121 | pv->perfmon_cmd_init_set_mov_avg(&cmd.cmd.perfmon, 17); | ||
122 | |||
123 | memset(&payload, 0, sizeof(struct pmu_payload)); | ||
124 | payload.in.buf = pv->get_perfmon_cntr_ptr(pmu); | ||
125 | payload.in.size = pv->get_perfmon_cntr_sz(pmu); | ||
126 | payload.in.offset = pv->get_perfmon_cmd_init_offsetofvar(COUNTER_ALLOC); | ||
127 | |||
128 | nvgpu_pmu_dbg(g, "cmd post PMU_PERFMON_CMD_ID_INIT"); | ||
129 | nvgpu_pmu_cmd_post(g, &cmd, NULL, &payload, PMU_COMMAND_QUEUE_LPQ, | ||
130 | NULL, NULL, &seq, ~0); | ||
131 | |||
132 | return 0; | ||
133 | } | ||
134 | |||
135 | int nvgpu_pmu_perfmon_start_sampling(struct nvgpu_pmu *pmu) | ||
136 | { | ||
137 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
138 | struct pmu_v *pv = &g->ops.pmu_ver; | ||
139 | struct pmu_cmd cmd; | ||
140 | struct pmu_payload payload; | ||
141 | u32 seq; | ||
142 | |||
143 | if (!nvgpu_is_enabled(g, NVGPU_PMU_PERFMON)) | ||
144 | return 0; | ||
145 | |||
146 | /* PERFMON Start */ | ||
147 | memset(&cmd, 0, sizeof(struct pmu_cmd)); | ||
148 | cmd.hdr.unit_id = get_perfmon_id(pmu); | ||
149 | if (cmd.hdr.unit_id == PMU_UNIT_INVALID) { | ||
150 | nvgpu_err(g, "failed to get perfmon UNIT ID, command skipped"); | ||
151 | return -EINVAL; | ||
152 | } | ||
153 | cmd.hdr.size = PMU_CMD_HDR_SIZE + pv->get_pmu_perfmon_cmd_start_size(); | ||
154 | pv->perfmon_start_set_cmd_type(&cmd.cmd.perfmon, | ||
155 | PMU_PERFMON_CMD_ID_START); | ||
156 | pv->perfmon_start_set_group_id(&cmd.cmd.perfmon, | ||
157 | PMU_DOMAIN_GROUP_PSTATE); | ||
158 | pv->perfmon_start_set_state_id(&cmd.cmd.perfmon, | ||
159 | pmu->perfmon_state_id[PMU_DOMAIN_GROUP_PSTATE]); | ||
160 | |||
161 | pv->perfmon_start_set_flags(&cmd.cmd.perfmon, | ||
162 | PMU_PERFMON_FLAG_ENABLE_INCREASE | | ||
163 | PMU_PERFMON_FLAG_ENABLE_DECREASE | | ||
164 | PMU_PERFMON_FLAG_CLEAR_PREV); | ||
165 | |||
166 | memset(&payload, 0, sizeof(struct pmu_payload)); | ||
167 | |||
168 | /* TBD: PMU_PERFMON_PCT_TO_INC * 100 */ | ||
169 | pv->set_perfmon_cntr_ut(pmu, 3000); /* 30% */ | ||
170 | /* TBD: PMU_PERFMON_PCT_TO_DEC * 100 */ | ||
171 | pv->set_perfmon_cntr_lt(pmu, 1000); /* 10% */ | ||
172 | pv->set_perfmon_cntr_valid(pmu, true); | ||
173 | |||
174 | payload.in.buf = pv->get_perfmon_cntr_ptr(pmu); | ||
175 | payload.in.size = pv->get_perfmon_cntr_sz(pmu); | ||
176 | payload.in.offset = | ||
177 | pv->get_perfmon_cmd_start_offsetofvar(COUNTER_ALLOC); | ||
178 | |||
179 | nvgpu_pmu_dbg(g, "cmd post PMU_PERFMON_CMD_ID_START"); | ||
180 | nvgpu_pmu_cmd_post(g, &cmd, NULL, &payload, PMU_COMMAND_QUEUE_LPQ, | ||
181 | NULL, NULL, &seq, ~0); | ||
182 | |||
183 | return 0; | ||
184 | } | ||
185 | |||
186 | int nvgpu_pmu_perfmon_stop_sampling(struct nvgpu_pmu *pmu) | ||
187 | { | ||
188 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
189 | struct pmu_cmd cmd; | ||
190 | u32 seq; | ||
191 | |||
192 | if (!nvgpu_is_enabled(g, NVGPU_PMU_PERFMON)) | ||
193 | return 0; | ||
194 | |||
195 | /* PERFMON Stop */ | ||
196 | memset(&cmd, 0, sizeof(struct pmu_cmd)); | ||
197 | cmd.hdr.unit_id = get_perfmon_id(pmu); | ||
198 | if (cmd.hdr.unit_id == PMU_UNIT_INVALID) { | ||
199 | nvgpu_err(g, "failed to get perfmon UNIT ID, command skipped"); | ||
200 | return -EINVAL; | ||
201 | } | ||
202 | cmd.hdr.size = PMU_CMD_HDR_SIZE + sizeof(struct pmu_perfmon_cmd_stop); | ||
203 | cmd.cmd.perfmon.stop.cmd_type = PMU_PERFMON_CMD_ID_STOP; | ||
204 | |||
205 | nvgpu_pmu_dbg(g, "cmd post PMU_PERFMON_CMD_ID_STOP"); | ||
206 | nvgpu_pmu_cmd_post(g, &cmd, NULL, NULL, PMU_COMMAND_QUEUE_LPQ, | ||
207 | NULL, NULL, &seq, ~0); | ||
208 | return 0; | ||
209 | } | ||
210 | |||
211 | int nvgpu_pmu_load_norm(struct gk20a *g, u32 *load) | ||
212 | { | ||
213 | *load = g->pmu.load_shadow; | ||
214 | return 0; | ||
215 | } | ||
216 | |||
217 | int nvgpu_pmu_load_update(struct gk20a *g) | ||
218 | { | ||
219 | struct nvgpu_pmu *pmu = &g->pmu; | ||
220 | u16 load = 0; | ||
221 | |||
222 | if (!pmu->perfmon_ready) { | ||
223 | pmu->load_shadow = 0; | ||
224 | return 0; | ||
225 | } | ||
226 | |||
227 | nvgpu_flcn_copy_from_dmem(pmu->flcn, pmu->sample_buffer, | ||
228 | (u8 *)&load, 2, 0); | ||
229 | pmu->load_shadow = load / 10; | ||
230 | pmu->load_avg = (((9*pmu->load_avg) + pmu->load_shadow) / 10); | ||
231 | |||
232 | return 0; | ||
233 | } | ||
234 | |||
235 | void nvgpu_pmu_get_load_counters(struct gk20a *g, u32 *busy_cycles, | ||
236 | u32 *total_cycles) | ||
237 | { | ||
238 | if (!g->power_on || gk20a_busy(g)) { | ||
239 | *busy_cycles = 0; | ||
240 | *total_cycles = 0; | ||
241 | return; | ||
242 | } | ||
243 | |||
244 | *busy_cycles = gk20a_pmu_read_idle_counter(g, 1); | ||
245 | *total_cycles = gk20a_pmu_read_idle_counter(g, 2); | ||
246 | |||
247 | gk20a_idle(g); | ||
248 | } | ||
249 | |||
250 | void nvgpu_pmu_reset_load_counters(struct gk20a *g) | ||
251 | { | ||
252 | if (!g->power_on || gk20a_busy(g)) | ||
253 | return; | ||
254 | |||
255 | gk20a_pmu_reset_idle_counter(g, 2); | ||
256 | gk20a_pmu_reset_idle_counter(g, 1); | ||
257 | |||
258 | gk20a_idle(g); | ||
259 | } | ||
260 | |||
261 | int nvgpu_pmu_handle_perfmon_event(struct nvgpu_pmu *pmu, | ||
262 | struct pmu_perfmon_msg *msg) | ||
263 | { | ||
264 | struct gk20a *g = gk20a_from_pmu(pmu); | ||
265 | |||
266 | nvgpu_log_fn(g, " "); | ||
267 | |||
268 | switch (msg->msg_type) { | ||
269 | case PMU_PERFMON_MSG_ID_INCREASE_EVENT: | ||
270 | nvgpu_pmu_dbg(g, "perfmon increase event: "); | ||
271 | nvgpu_pmu_dbg(g, "state_id %d, ground_id %d, pct %d", | ||
272 | msg->gen.state_id, msg->gen.group_id, msg->gen.data); | ||
273 | (pmu->perfmon_events_cnt)++; | ||
274 | break; | ||
275 | case PMU_PERFMON_MSG_ID_DECREASE_EVENT: | ||
276 | nvgpu_pmu_dbg(g, "perfmon decrease event: "); | ||
277 | nvgpu_pmu_dbg(g, "state_id %d, ground_id %d, pct %d", | ||
278 | msg->gen.state_id, msg->gen.group_id, msg->gen.data); | ||
279 | (pmu->perfmon_events_cnt)++; | ||
280 | break; | ||
281 | case PMU_PERFMON_MSG_ID_INIT_EVENT: | ||
282 | pmu->perfmon_ready = 1; | ||
283 | nvgpu_pmu_dbg(g, "perfmon init event"); | ||
284 | break; | ||
285 | default: | ||
286 | break; | ||
287 | } | ||
288 | |||
289 | /* restart sampling */ | ||
290 | if (pmu->perfmon_sampling_enabled) | ||
291 | return nvgpu_pmu_perfmon_start_sampling(pmu); | ||
292 | return 0; | ||
293 | } | ||