diff options
Diffstat (limited to 'drivers/gpu/nvgpu/gk20a/gk20a_scale.c')
-rw-r--r-- | drivers/gpu/nvgpu/gk20a/gk20a_scale.c | 358 |
1 files changed, 358 insertions, 0 deletions
diff --git a/drivers/gpu/nvgpu/gk20a/gk20a_scale.c b/drivers/gpu/nvgpu/gk20a/gk20a_scale.c new file mode 100644 index 00000000..d1fd71fe --- /dev/null +++ b/drivers/gpu/nvgpu/gk20a/gk20a_scale.c | |||
@@ -0,0 +1,358 @@ | |||
1 | /* | ||
2 | * gk20a clock scaling profile | ||
3 | * | ||
4 | * Copyright (c) 2013-2014, NVIDIA Corporation. All rights reserved. | ||
5 | * | ||
6 | * This program is free software; you can redistribute it and/or modify it | ||
7 | * under the terms and conditions of the GNU General Public License, | ||
8 | * version 2, as published by the Free Software Foundation. | ||
9 | * | ||
10 | * This program is distributed in the hope it will be useful, but WITHOUT | ||
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | ||
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | ||
13 | * more details. | ||
14 | * | ||
15 | * You should have received a copy of the GNU General Public License | ||
16 | * along with this program. If not, see <http://www.gnu.org/licenses/>. | ||
17 | */ | ||
18 | |||
19 | #include <linux/devfreq.h> | ||
20 | #include <linux/debugfs.h> | ||
21 | #include <linux/types.h> | ||
22 | #include <linux/clk.h> | ||
23 | #include <linux/export.h> | ||
24 | #include <linux/slab.h> | ||
25 | #include <linux/clk/tegra.h> | ||
26 | #include <linux/tegra-soc.h> | ||
27 | #include <linux/platform_data/tegra_edp.h> | ||
28 | #include <linux/pm_qos.h> | ||
29 | |||
30 | #include <governor.h> | ||
31 | |||
32 | #include "gk20a.h" | ||
33 | #include "pmu_gk20a.h" | ||
34 | #include "clk_gk20a.h" | ||
35 | #include "gk20a_scale.h" | ||
36 | |||
37 | static ssize_t gk20a_scale_load_show(struct device *dev, | ||
38 | struct device_attribute *attr, | ||
39 | char *buf) | ||
40 | { | ||
41 | struct platform_device *pdev = to_platform_device(dev); | ||
42 | struct gk20a *g = get_gk20a(pdev); | ||
43 | u32 busy_time; | ||
44 | ssize_t res; | ||
45 | |||
46 | if (!g->power_on) { | ||
47 | busy_time = 0; | ||
48 | } else { | ||
49 | gk20a_busy(g->dev); | ||
50 | gk20a_pmu_load_norm(g, &busy_time); | ||
51 | gk20a_idle(g->dev); | ||
52 | } | ||
53 | |||
54 | res = snprintf(buf, PAGE_SIZE, "%u\n", busy_time); | ||
55 | |||
56 | return res; | ||
57 | } | ||
58 | |||
59 | static DEVICE_ATTR(load, S_IRUGO, gk20a_scale_load_show, NULL); | ||
60 | |||
61 | /* | ||
62 | * gk20a_scale_qos_notify() | ||
63 | * | ||
64 | * This function is called when the minimum QoS requirement for the device | ||
65 | * has changed. The function calls postscaling callback if it is defined. | ||
66 | */ | ||
67 | |||
68 | static int gk20a_scale_qos_notify(struct notifier_block *nb, | ||
69 | unsigned long n, void *p) | ||
70 | { | ||
71 | struct gk20a_scale_profile *profile = | ||
72 | container_of(nb, struct gk20a_scale_profile, | ||
73 | qos_notify_block); | ||
74 | struct gk20a_platform *platform = platform_get_drvdata(profile->pdev); | ||
75 | struct gk20a *g = get_gk20a(profile->pdev); | ||
76 | unsigned long freq; | ||
77 | |||
78 | if (!platform->postscale) | ||
79 | return NOTIFY_OK; | ||
80 | |||
81 | /* get the frequency requirement. if devfreq is enabled, check if it | ||
82 | * has higher demand than qos */ | ||
83 | freq = gk20a_clk_round_rate(g, pm_qos_request(platform->qos_id)); | ||
84 | if (g->devfreq) | ||
85 | freq = max(g->devfreq->previous_freq, freq); | ||
86 | |||
87 | platform->postscale(profile->pdev, freq); | ||
88 | |||
89 | return NOTIFY_OK; | ||
90 | } | ||
91 | |||
92 | /* | ||
93 | * gk20a_scale_make_freq_table(profile) | ||
94 | * | ||
95 | * This function initialises the frequency table for the given device profile | ||
96 | */ | ||
97 | |||
98 | static int gk20a_scale_make_freq_table(struct gk20a_scale_profile *profile) | ||
99 | { | ||
100 | struct gk20a *g = get_gk20a(profile->pdev); | ||
101 | unsigned long *freqs; | ||
102 | int num_freqs, err; | ||
103 | |||
104 | /* make sure the clock is available */ | ||
105 | if (!gk20a_clk_get(g)) | ||
106 | return -ENOSYS; | ||
107 | |||
108 | /* get gpu dvfs table */ | ||
109 | err = tegra_dvfs_get_freqs(clk_get_parent(g->clk.tegra_clk), | ||
110 | &freqs, &num_freqs); | ||
111 | if (err) | ||
112 | return -ENOSYS; | ||
113 | |||
114 | profile->devfreq_profile.freq_table = (unsigned long *)freqs; | ||
115 | profile->devfreq_profile.max_state = num_freqs; | ||
116 | |||
117 | return 0; | ||
118 | } | ||
119 | |||
120 | /* | ||
121 | * gk20a_scale_target(dev, *freq, flags) | ||
122 | * | ||
123 | * This function scales the clock | ||
124 | */ | ||
125 | |||
126 | static int gk20a_scale_target(struct device *dev, unsigned long *freq, | ||
127 | u32 flags) | ||
128 | { | ||
129 | struct gk20a *g = get_gk20a(to_platform_device(dev)); | ||
130 | struct gk20a_platform *platform = dev_get_drvdata(dev); | ||
131 | struct gk20a_scale_profile *profile = g->scale_profile; | ||
132 | unsigned long rounded_rate = gk20a_clk_round_rate(g, *freq); | ||
133 | |||
134 | if (gk20a_clk_get_rate(g) == rounded_rate) { | ||
135 | *freq = rounded_rate; | ||
136 | return 0; | ||
137 | } | ||
138 | |||
139 | gk20a_clk_set_rate(g, rounded_rate); | ||
140 | if (platform->postscale) | ||
141 | platform->postscale(profile->pdev, rounded_rate); | ||
142 | *freq = gk20a_clk_get_rate(g); | ||
143 | |||
144 | return 0; | ||
145 | } | ||
146 | |||
147 | /* | ||
148 | * update_load_estimate_gpmu(profile) | ||
149 | * | ||
150 | * Update load estimate using gpmu. The gpmu value is normalised | ||
151 | * based on the time it was asked last time. | ||
152 | */ | ||
153 | |||
154 | static void update_load_estimate_gpmu(struct platform_device *pdev) | ||
155 | { | ||
156 | struct gk20a *g = get_gk20a(pdev); | ||
157 | struct gk20a_scale_profile *profile = g->scale_profile; | ||
158 | unsigned long dt; | ||
159 | u32 busy_time; | ||
160 | ktime_t t; | ||
161 | |||
162 | t = ktime_get(); | ||
163 | dt = ktime_us_delta(t, profile->last_event_time); | ||
164 | |||
165 | profile->dev_stat.total_time = dt; | ||
166 | profile->last_event_time = t; | ||
167 | gk20a_pmu_load_norm(g, &busy_time); | ||
168 | profile->dev_stat.busy_time = (busy_time * dt) / 1000; | ||
169 | } | ||
170 | |||
171 | /* | ||
172 | * gk20a_scale_suspend(pdev) | ||
173 | * | ||
174 | * This function informs devfreq of suspend | ||
175 | */ | ||
176 | |||
177 | void gk20a_scale_suspend(struct platform_device *pdev) | ||
178 | { | ||
179 | struct gk20a *g = get_gk20a(pdev); | ||
180 | struct devfreq *devfreq = g->devfreq; | ||
181 | |||
182 | if (!devfreq) | ||
183 | return; | ||
184 | |||
185 | devfreq_suspend_device(devfreq); | ||
186 | } | ||
187 | |||
188 | /* | ||
189 | * gk20a_scale_resume(pdev) | ||
190 | * | ||
191 | * This functions informs devfreq of resume | ||
192 | */ | ||
193 | |||
194 | void gk20a_scale_resume(struct platform_device *pdev) | ||
195 | { | ||
196 | struct gk20a *g = get_gk20a(pdev); | ||
197 | struct devfreq *devfreq = g->devfreq; | ||
198 | |||
199 | if (!devfreq) | ||
200 | return; | ||
201 | |||
202 | devfreq_resume_device(devfreq); | ||
203 | } | ||
204 | |||
205 | /* | ||
206 | * gk20a_scale_notify(pdev, busy) | ||
207 | * | ||
208 | * Calling this function informs that the device is idling (..or busy). This | ||
209 | * data is used to estimate the current load | ||
210 | */ | ||
211 | |||
212 | static void gk20a_scale_notify(struct platform_device *pdev, bool busy) | ||
213 | { | ||
214 | struct gk20a_platform *platform = platform_get_drvdata(pdev); | ||
215 | struct gk20a *g = get_gk20a(pdev); | ||
216 | struct gk20a_scale_profile *profile = g->scale_profile; | ||
217 | struct devfreq *devfreq = g->devfreq; | ||
218 | |||
219 | /* inform edp about new constraint */ | ||
220 | if (platform->prescale) | ||
221 | platform->prescale(pdev); | ||
222 | |||
223 | /* Is the device profile initialised? */ | ||
224 | if (!(profile && devfreq)) | ||
225 | return; | ||
226 | |||
227 | mutex_lock(&devfreq->lock); | ||
228 | profile->dev_stat.busy = busy; | ||
229 | update_devfreq(devfreq); | ||
230 | mutex_unlock(&devfreq->lock); | ||
231 | } | ||
232 | |||
233 | void gk20a_scale_notify_idle(struct platform_device *pdev) | ||
234 | { | ||
235 | gk20a_scale_notify(pdev, false); | ||
236 | |||
237 | } | ||
238 | |||
239 | void gk20a_scale_notify_busy(struct platform_device *pdev) | ||
240 | { | ||
241 | gk20a_scale_notify(pdev, true); | ||
242 | } | ||
243 | |||
244 | /* | ||
245 | * gk20a_scale_get_dev_status(dev, *stat) | ||
246 | * | ||
247 | * This function queries the current device status. | ||
248 | */ | ||
249 | |||
250 | static int gk20a_scale_get_dev_status(struct device *dev, | ||
251 | struct devfreq_dev_status *stat) | ||
252 | { | ||
253 | struct gk20a *g = get_gk20a(to_platform_device(dev)); | ||
254 | struct gk20a_scale_profile *profile = g->scale_profile; | ||
255 | |||
256 | /* Make sure there are correct values for the current frequency */ | ||
257 | profile->dev_stat.current_frequency = gk20a_clk_get_rate(g); | ||
258 | |||
259 | /* Update load estimate */ | ||
260 | update_load_estimate_gpmu(to_platform_device(dev)); | ||
261 | |||
262 | /* Copy the contents of the current device status */ | ||
263 | *stat = profile->dev_stat; | ||
264 | |||
265 | /* Finally, clear out the local values */ | ||
266 | profile->dev_stat.total_time = 0; | ||
267 | profile->dev_stat.busy_time = 0; | ||
268 | |||
269 | return 0; | ||
270 | } | ||
271 | |||
272 | /* | ||
273 | * gk20a_scale_init(pdev) | ||
274 | */ | ||
275 | |||
276 | void gk20a_scale_init(struct platform_device *pdev) | ||
277 | { | ||
278 | struct gk20a_platform *platform = platform_get_drvdata(pdev); | ||
279 | struct gk20a *g = platform->g; | ||
280 | struct gk20a_scale_profile *profile; | ||
281 | int err; | ||
282 | |||
283 | if (g->scale_profile) | ||
284 | return; | ||
285 | |||
286 | profile = kzalloc(sizeof(*profile), GFP_KERNEL); | ||
287 | |||
288 | profile->pdev = pdev; | ||
289 | profile->dev_stat.busy = false; | ||
290 | |||
291 | /* Create frequency table */ | ||
292 | err = gk20a_scale_make_freq_table(profile); | ||
293 | if (err || !profile->devfreq_profile.max_state) | ||
294 | goto err_get_freqs; | ||
295 | |||
296 | if (device_create_file(&pdev->dev, &dev_attr_load)) | ||
297 | goto err_create_sysfs_entry; | ||
298 | |||
299 | /* Store device profile so we can access it if devfreq governor | ||
300 | * init needs that */ | ||
301 | g->scale_profile = profile; | ||
302 | |||
303 | if (platform->devfreq_governor) { | ||
304 | struct devfreq *devfreq; | ||
305 | |||
306 | profile->devfreq_profile.initial_freq = | ||
307 | profile->devfreq_profile.freq_table[0]; | ||
308 | profile->devfreq_profile.target = gk20a_scale_target; | ||
309 | profile->devfreq_profile.get_dev_status = | ||
310 | gk20a_scale_get_dev_status; | ||
311 | |||
312 | devfreq = devfreq_add_device(&pdev->dev, | ||
313 | &profile->devfreq_profile, | ||
314 | platform->devfreq_governor, NULL); | ||
315 | |||
316 | if (IS_ERR(devfreq)) | ||
317 | devfreq = NULL; | ||
318 | |||
319 | g->devfreq = devfreq; | ||
320 | } | ||
321 | |||
322 | /* Should we register QoS callback for this device? */ | ||
323 | if (platform->qos_id < PM_QOS_NUM_CLASSES && | ||
324 | platform->qos_id != PM_QOS_RESERVED && | ||
325 | platform->postscale) { | ||
326 | profile->qos_notify_block.notifier_call = | ||
327 | &gk20a_scale_qos_notify; | ||
328 | pm_qos_add_notifier(platform->qos_id, | ||
329 | &profile->qos_notify_block); | ||
330 | } | ||
331 | |||
332 | return; | ||
333 | |||
334 | err_get_freqs: | ||
335 | device_remove_file(&pdev->dev, &dev_attr_load); | ||
336 | err_create_sysfs_entry: | ||
337 | kfree(g->scale_profile); | ||
338 | g->scale_profile = NULL; | ||
339 | } | ||
340 | |||
341 | /* | ||
342 | * gk20a_scale_hw_init(dev) | ||
343 | * | ||
344 | * Initialize hardware portion of the device | ||
345 | */ | ||
346 | |||
347 | void gk20a_scale_hw_init(struct platform_device *pdev) | ||
348 | { | ||
349 | struct gk20a_platform *platform = platform_get_drvdata(pdev); | ||
350 | struct gk20a_scale_profile *profile = platform->g->scale_profile; | ||
351 | |||
352 | /* make sure that scaling has bee initialised */ | ||
353 | if (!profile) | ||
354 | return; | ||
355 | |||
356 | profile->dev_stat.total_time = 0; | ||
357 | profile->last_event_time = ktime_get(); | ||
358 | } | ||