1 /*
2 * Copyright (c) 2014, 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 #define gk20a_pmu(p) container_of((p), struct gk20a_pmu, base.subdev)
23 #include "priv.h"
24
25 #include <subdev/clk.h>
26 #include <subdev/timer.h>
27 #include <subdev/volt.h>
28
29 #define BUSY_SLOT 0
30 #define CLK_SLOT 7
31
32 struct gk20a_pmu_dvfs_data {
33 int p_load_target;
34 int p_load_max;
35 int p_smooth;
36 unsigned int avg_load;
37 };
38
39 struct gk20a_pmu {
40 struct nvkm_pmu base;
41 struct nvkm_alarm alarm;
42 struct gk20a_pmu_dvfs_data *data;
43 };
44
45 struct gk20a_pmu_dvfs_dev_status {
46 unsigned long total;
47 unsigned long busy;
48 int cur_state;
49 };
50
51 static int
gk20a_pmu_dvfs_target(struct gk20a_pmu * pmu,int * state)52 gk20a_pmu_dvfs_target(struct gk20a_pmu *pmu, int *state)
53 {
54 struct nvkm_clk *clk = pmu->base.subdev.device->clk;
55
56 return nvkm_clk_astate(clk, *state, 0, false);
57 }
58
59 static int
gk20a_pmu_dvfs_get_cur_state(struct gk20a_pmu * pmu,int * state)60 gk20a_pmu_dvfs_get_cur_state(struct gk20a_pmu *pmu, int *state)
61 {
62 struct nvkm_clk *clk = pmu->base.subdev.device->clk;
63
64 *state = clk->pstate;
65 return 0;
66 }
67
68 static int
gk20a_pmu_dvfs_get_target_state(struct gk20a_pmu * pmu,int * state,int load)69 gk20a_pmu_dvfs_get_target_state(struct gk20a_pmu *pmu,
70 int *state, int load)
71 {
72 struct gk20a_pmu_dvfs_data *data = pmu->data;
73 struct nvkm_clk *clk = pmu->base.subdev.device->clk;
74 int cur_level, level;
75
76 /* For GK20A, the performance level is directly mapped to pstate */
77 level = cur_level = clk->pstate;
78
79 if (load > data->p_load_max) {
80 level = min(clk->state_nr - 1, level + (clk->state_nr / 3));
81 } else {
82 level += ((load - data->p_load_target) * 10 /
83 data->p_load_target) / 2;
84 level = max(0, level);
85 level = min(clk->state_nr - 1, level);
86 }
87
88 nvkm_trace(&pmu->base.subdev, "cur level = %d, new level = %d\n",
89 cur_level, level);
90
91 *state = level;
92
93 if (level == cur_level)
94 return 0;
95 else
96 return 1;
97 }
98
99 static int
gk20a_pmu_dvfs_get_dev_status(struct gk20a_pmu * pmu,struct gk20a_pmu_dvfs_dev_status * status)100 gk20a_pmu_dvfs_get_dev_status(struct gk20a_pmu *pmu,
101 struct gk20a_pmu_dvfs_dev_status *status)
102 {
103 struct nvkm_device *device = pmu->base.subdev.device;
104 status->busy = nvkm_rd32(device, 0x10a508 + (BUSY_SLOT * 0x10));
105 status->total= nvkm_rd32(device, 0x10a508 + (CLK_SLOT * 0x10));
106 return 0;
107 }
108
109 static void
gk20a_pmu_dvfs_reset_dev_status(struct gk20a_pmu * pmu)110 gk20a_pmu_dvfs_reset_dev_status(struct gk20a_pmu *pmu)
111 {
112 struct nvkm_device *device = pmu->base.subdev.device;
113 nvkm_wr32(device, 0x10a508 + (BUSY_SLOT * 0x10), 0x80000000);
114 nvkm_wr32(device, 0x10a508 + (CLK_SLOT * 0x10), 0x80000000);
115 }
116
117 static void
gk20a_pmu_dvfs_work(struct nvkm_alarm * alarm)118 gk20a_pmu_dvfs_work(struct nvkm_alarm *alarm)
119 {
120 struct gk20a_pmu *pmu =
121 container_of(alarm, struct gk20a_pmu, alarm);
122 struct gk20a_pmu_dvfs_data *data = pmu->data;
123 struct gk20a_pmu_dvfs_dev_status status;
124 struct nvkm_subdev *subdev = &pmu->base.subdev;
125 struct nvkm_device *device = subdev->device;
126 struct nvkm_clk *clk = device->clk;
127 struct nvkm_timer *tmr = device->timer;
128 struct nvkm_volt *volt = device->volt;
129 u32 utilization = 0;
130 int state, ret;
131
132 /*
133 * The PMU is initialized before CLK and VOLT, so we have to make sure the
134 * CLK and VOLT are ready here.
135 */
136 if (!clk || !volt)
137 goto resched;
138
139 ret = gk20a_pmu_dvfs_get_dev_status(pmu, &status);
140 if (ret) {
141 nvkm_warn(subdev, "failed to get device status\n");
142 goto resched;
143 }
144
145 if (status.total)
146 utilization = div_u64((u64)status.busy * 100, status.total);
147
148 data->avg_load = (data->p_smooth * data->avg_load) + utilization;
149 data->avg_load /= data->p_smooth + 1;
150 nvkm_trace(subdev, "utilization = %d %%, avg_load = %d %%\n",
151 utilization, data->avg_load);
152
153 ret = gk20a_pmu_dvfs_get_cur_state(pmu, &state);
154 if (ret) {
155 nvkm_warn(subdev, "failed to get current state\n");
156 goto resched;
157 }
158
159 if (gk20a_pmu_dvfs_get_target_state(pmu, &state, data->avg_load)) {
160 nvkm_trace(subdev, "set new state to %d\n", state);
161 gk20a_pmu_dvfs_target(pmu, &state);
162 }
163
164 resched:
165 gk20a_pmu_dvfs_reset_dev_status(pmu);
166 nvkm_timer_alarm(tmr, 100000000, alarm);
167 }
168
169 static int
gk20a_pmu_fini(struct nvkm_subdev * subdev,bool suspend)170 gk20a_pmu_fini(struct nvkm_subdev *subdev, bool suspend)
171 {
172 struct gk20a_pmu *pmu = gk20a_pmu(subdev);
173 nvkm_timer_alarm_cancel(subdev->device->timer, &pmu->alarm);
174 return 0;
175 }
176
177 static void *
gk20a_pmu_dtor(struct nvkm_subdev * subdev)178 gk20a_pmu_dtor(struct nvkm_subdev *subdev)
179 {
180 return gk20a_pmu(subdev);
181 }
182
183 static int
gk20a_pmu_init(struct nvkm_subdev * subdev)184 gk20a_pmu_init(struct nvkm_subdev *subdev)
185 {
186 struct gk20a_pmu *pmu = gk20a_pmu(subdev);
187 struct nvkm_device *device = pmu->base.subdev.device;
188
189 /* init pwr perf counter */
190 nvkm_wr32(device, 0x10a504 + (BUSY_SLOT * 0x10), 0x00200001);
191 nvkm_wr32(device, 0x10a50c + (BUSY_SLOT * 0x10), 0x00000002);
192 nvkm_wr32(device, 0x10a50c + (CLK_SLOT * 0x10), 0x00000003);
193
194 nvkm_timer_alarm(device->timer, 2000000000, &pmu->alarm);
195 return 0;
196 }
197
198 static struct gk20a_pmu_dvfs_data
199 gk20a_dvfs_data= {
200 .p_load_target = 70,
201 .p_load_max = 90,
202 .p_smooth = 1,
203 };
204
205 static const struct nvkm_subdev_func
206 gk20a_pmu = {
207 .init = gk20a_pmu_init,
208 .fini = gk20a_pmu_fini,
209 .dtor = gk20a_pmu_dtor,
210 };
211
212 int
gk20a_pmu_new(struct nvkm_device * device,int index,struct nvkm_pmu ** ppmu)213 gk20a_pmu_new(struct nvkm_device *device, int index, struct nvkm_pmu **ppmu)
214 {
215 static const struct nvkm_pmu_func func = {};
216 struct gk20a_pmu *pmu;
217
218 if (!(pmu = kzalloc(sizeof(*pmu), GFP_KERNEL)))
219 return -ENOMEM;
220 pmu->base.func = &func;
221 *ppmu = &pmu->base;
222
223 nvkm_subdev_ctor(&gk20a_pmu, device, index, 0, &pmu->base.subdev);
224 pmu->data = &gk20a_dvfs_data;
225 nvkm_alarm_init(&pmu->alarm, gk20a_pmu_dvfs_work);
226 return 0;
227 }
228