1 /* bbc_envctrl.c: UltraSPARC-III environment control driver.
2 *
3 * Copyright (C) 2001, 2008 David S. Miller (davem@davemloft.net)
4 */
5
6 #include <linux/kthread.h>
7 #include <linux/delay.h>
8 #include <linux/kmod.h>
9 #include <linux/reboot.h>
10 #include <linux/of.h>
11 #include <linux/slab.h>
12 #include <linux/of_device.h>
13 #include <asm/oplib.h>
14
15 #include "bbc_i2c.h"
16 #include "max1617.h"
17
18 #undef ENVCTRL_TRACE
19
20 /* WARNING: Making changes to this driver is very dangerous.
21 * If you misprogram the sensor chips they can
22 * cut the power on you instantly.
23 */
24
25 /* Two temperature sensors exist in the SunBLADE-1000 enclosure.
26 * Both are implemented using max1617 i2c devices. Each max1617
27 * monitors 2 temperatures, one for one of the cpu dies and the other
28 * for the ambient temperature.
29 *
30 * The max1617 is capable of being programmed with power-off
31 * temperature values, one low limit and one high limit. These
32 * can be controlled independently for the cpu or ambient temperature.
33 * If a limit is violated, the power is simply shut off. The frequency
34 * with which the max1617 does temperature sampling can be controlled
35 * as well.
36 *
37 * Three fans exist inside the machine, all three are controlled with
38 * an i2c digital to analog converter. There is a fan directed at the
39 * two processor slots, another for the rest of the enclosure, and the
40 * third is for the power supply. The first two fans may be speed
41 * controlled by changing the voltage fed to them. The third fan may
42 * only be completely off or on. The third fan is meant to only be
43 * disabled/enabled when entering/exiting the lowest power-saving
44 * mode of the machine.
45 *
46 * An environmental control kernel thread periodically monitors all
47 * temperature sensors. Based upon the samples it will adjust the
48 * fan speeds to try and keep the system within a certain temperature
49 * range (the goal being to make the fans as quiet as possible without
50 * allowing the system to get too hot).
51 *
52 * If the temperature begins to rise/fall outside of the acceptable
53 * operating range, a periodic warning will be sent to the kernel log.
54 * The fans will be put on full blast to attempt to deal with this
55 * situation. After exceeding the acceptable operating range by a
56 * certain threshold, the kernel thread will shut down the system.
57 * Here, the thread is attempting to shut the machine down cleanly
58 * before the hardware based power-off event is triggered.
59 */
60
61 /* These settings are in Celsius. We use these defaults only
62 * if we cannot interrogate the cpu-fru SEEPROM.
63 */
64 struct temp_limits {
65 s8 high_pwroff, high_shutdown, high_warn;
66 s8 low_warn, low_shutdown, low_pwroff;
67 };
68
69 static struct temp_limits cpu_temp_limits[2] = {
70 { 100, 85, 80, 5, -5, -10 },
71 { 100, 85, 80, 5, -5, -10 },
72 };
73
74 static struct temp_limits amb_temp_limits[2] = {
75 { 65, 55, 40, 5, -5, -10 },
76 { 65, 55, 40, 5, -5, -10 },
77 };
78
79 static LIST_HEAD(all_temps);
80 static LIST_HEAD(all_fans);
81
82 #define CPU_FAN_REG 0xf0
83 #define SYS_FAN_REG 0xf2
84 #define PSUPPLY_FAN_REG 0xf4
85
86 #define FAN_SPEED_MIN 0x0c
87 #define FAN_SPEED_MAX 0x3f
88
89 #define PSUPPLY_FAN_ON 0x1f
90 #define PSUPPLY_FAN_OFF 0x00
91
set_fan_speeds(struct bbc_fan_control * fp)92 static void set_fan_speeds(struct bbc_fan_control *fp)
93 {
94 /* Put temperatures into range so we don't mis-program
95 * the hardware.
96 */
97 if (fp->cpu_fan_speed < FAN_SPEED_MIN)
98 fp->cpu_fan_speed = FAN_SPEED_MIN;
99 if (fp->cpu_fan_speed > FAN_SPEED_MAX)
100 fp->cpu_fan_speed = FAN_SPEED_MAX;
101 if (fp->system_fan_speed < FAN_SPEED_MIN)
102 fp->system_fan_speed = FAN_SPEED_MIN;
103 if (fp->system_fan_speed > FAN_SPEED_MAX)
104 fp->system_fan_speed = FAN_SPEED_MAX;
105 #ifdef ENVCTRL_TRACE
106 printk("fan%d: Changed fan speed to cpu(%02x) sys(%02x)\n",
107 fp->index,
108 fp->cpu_fan_speed, fp->system_fan_speed);
109 #endif
110
111 bbc_i2c_writeb(fp->client, fp->cpu_fan_speed, CPU_FAN_REG);
112 bbc_i2c_writeb(fp->client, fp->system_fan_speed, SYS_FAN_REG);
113 bbc_i2c_writeb(fp->client,
114 (fp->psupply_fan_on ?
115 PSUPPLY_FAN_ON : PSUPPLY_FAN_OFF),
116 PSUPPLY_FAN_REG);
117 }
118
get_current_temps(struct bbc_cpu_temperature * tp)119 static void get_current_temps(struct bbc_cpu_temperature *tp)
120 {
121 tp->prev_amb_temp = tp->curr_amb_temp;
122 bbc_i2c_readb(tp->client,
123 (unsigned char *) &tp->curr_amb_temp,
124 MAX1617_AMB_TEMP);
125 tp->prev_cpu_temp = tp->curr_cpu_temp;
126 bbc_i2c_readb(tp->client,
127 (unsigned char *) &tp->curr_cpu_temp,
128 MAX1617_CPU_TEMP);
129 #ifdef ENVCTRL_TRACE
130 printk("temp%d: cpu(%d C) amb(%d C)\n",
131 tp->index,
132 (int) tp->curr_cpu_temp, (int) tp->curr_amb_temp);
133 #endif
134 }
135
136
do_envctrl_shutdown(struct bbc_cpu_temperature * tp)137 static void do_envctrl_shutdown(struct bbc_cpu_temperature *tp)
138 {
139 static int shutting_down = 0;
140 char *type = "???";
141 s8 val = -1;
142
143 if (shutting_down != 0)
144 return;
145
146 if (tp->curr_amb_temp >= amb_temp_limits[tp->index].high_shutdown ||
147 tp->curr_amb_temp < amb_temp_limits[tp->index].low_shutdown) {
148 type = "ambient";
149 val = tp->curr_amb_temp;
150 } else if (tp->curr_cpu_temp >= cpu_temp_limits[tp->index].high_shutdown ||
151 tp->curr_cpu_temp < cpu_temp_limits[tp->index].low_shutdown) {
152 type = "CPU";
153 val = tp->curr_cpu_temp;
154 }
155
156 printk(KERN_CRIT "temp%d: Outside of safe %s "
157 "operating temperature, %d C.\n",
158 tp->index, type, val);
159
160 printk(KERN_CRIT "kenvctrld: Shutting down the system now.\n");
161
162 shutting_down = 1;
163 if (orderly_poweroff(true) < 0)
164 printk(KERN_CRIT "envctrl: shutdown execution failed\n");
165 }
166
167 #define WARN_INTERVAL (30 * HZ)
168
analyze_ambient_temp(struct bbc_cpu_temperature * tp,unsigned long * last_warn,int tick)169 static void analyze_ambient_temp(struct bbc_cpu_temperature *tp, unsigned long *last_warn, int tick)
170 {
171 int ret = 0;
172
173 if (time_after(jiffies, (*last_warn + WARN_INTERVAL))) {
174 if (tp->curr_amb_temp >=
175 amb_temp_limits[tp->index].high_warn) {
176 printk(KERN_WARNING "temp%d: "
177 "Above safe ambient operating temperature, %d C.\n",
178 tp->index, (int) tp->curr_amb_temp);
179 ret = 1;
180 } else if (tp->curr_amb_temp <
181 amb_temp_limits[tp->index].low_warn) {
182 printk(KERN_WARNING "temp%d: "
183 "Below safe ambient operating temperature, %d C.\n",
184 tp->index, (int) tp->curr_amb_temp);
185 ret = 1;
186 }
187 if (ret)
188 *last_warn = jiffies;
189 } else if (tp->curr_amb_temp >= amb_temp_limits[tp->index].high_warn ||
190 tp->curr_amb_temp < amb_temp_limits[tp->index].low_warn)
191 ret = 1;
192
193 /* Now check the shutdown limits. */
194 if (tp->curr_amb_temp >= amb_temp_limits[tp->index].high_shutdown ||
195 tp->curr_amb_temp < amb_temp_limits[tp->index].low_shutdown) {
196 do_envctrl_shutdown(tp);
197 ret = 1;
198 }
199
200 if (ret) {
201 tp->fan_todo[FAN_AMBIENT] = FAN_FULLBLAST;
202 } else if ((tick & (8 - 1)) == 0) {
203 s8 amb_goal_hi = amb_temp_limits[tp->index].high_warn - 10;
204 s8 amb_goal_lo;
205
206 amb_goal_lo = amb_goal_hi - 3;
207
208 /* We do not try to avoid 'too cold' events. Basically we
209 * only try to deal with over-heating and fan noise reduction.
210 */
211 if (tp->avg_amb_temp < amb_goal_hi) {
212 if (tp->avg_amb_temp >= amb_goal_lo)
213 tp->fan_todo[FAN_AMBIENT] = FAN_SAME;
214 else
215 tp->fan_todo[FAN_AMBIENT] = FAN_SLOWER;
216 } else {
217 tp->fan_todo[FAN_AMBIENT] = FAN_FASTER;
218 }
219 } else {
220 tp->fan_todo[FAN_AMBIENT] = FAN_SAME;
221 }
222 }
223
analyze_cpu_temp(struct bbc_cpu_temperature * tp,unsigned long * last_warn,int tick)224 static void analyze_cpu_temp(struct bbc_cpu_temperature *tp, unsigned long *last_warn, int tick)
225 {
226 int ret = 0;
227
228 if (time_after(jiffies, (*last_warn + WARN_INTERVAL))) {
229 if (tp->curr_cpu_temp >=
230 cpu_temp_limits[tp->index].high_warn) {
231 printk(KERN_WARNING "temp%d: "
232 "Above safe CPU operating temperature, %d C.\n",
233 tp->index, (int) tp->curr_cpu_temp);
234 ret = 1;
235 } else if (tp->curr_cpu_temp <
236 cpu_temp_limits[tp->index].low_warn) {
237 printk(KERN_WARNING "temp%d: "
238 "Below safe CPU operating temperature, %d C.\n",
239 tp->index, (int) tp->curr_cpu_temp);
240 ret = 1;
241 }
242 if (ret)
243 *last_warn = jiffies;
244 } else if (tp->curr_cpu_temp >= cpu_temp_limits[tp->index].high_warn ||
245 tp->curr_cpu_temp < cpu_temp_limits[tp->index].low_warn)
246 ret = 1;
247
248 /* Now check the shutdown limits. */
249 if (tp->curr_cpu_temp >= cpu_temp_limits[tp->index].high_shutdown ||
250 tp->curr_cpu_temp < cpu_temp_limits[tp->index].low_shutdown) {
251 do_envctrl_shutdown(tp);
252 ret = 1;
253 }
254
255 if (ret) {
256 tp->fan_todo[FAN_CPU] = FAN_FULLBLAST;
257 } else if ((tick & (8 - 1)) == 0) {
258 s8 cpu_goal_hi = cpu_temp_limits[tp->index].high_warn - 10;
259 s8 cpu_goal_lo;
260
261 cpu_goal_lo = cpu_goal_hi - 3;
262
263 /* We do not try to avoid 'too cold' events. Basically we
264 * only try to deal with over-heating and fan noise reduction.
265 */
266 if (tp->avg_cpu_temp < cpu_goal_hi) {
267 if (tp->avg_cpu_temp >= cpu_goal_lo)
268 tp->fan_todo[FAN_CPU] = FAN_SAME;
269 else
270 tp->fan_todo[FAN_CPU] = FAN_SLOWER;
271 } else {
272 tp->fan_todo[FAN_CPU] = FAN_FASTER;
273 }
274 } else {
275 tp->fan_todo[FAN_CPU] = FAN_SAME;
276 }
277 }
278
analyze_temps(struct bbc_cpu_temperature * tp,unsigned long * last_warn)279 static void analyze_temps(struct bbc_cpu_temperature *tp, unsigned long *last_warn)
280 {
281 tp->avg_amb_temp = (s8)((int)((int)tp->avg_amb_temp + (int)tp->curr_amb_temp) / 2);
282 tp->avg_cpu_temp = (s8)((int)((int)tp->avg_cpu_temp + (int)tp->curr_cpu_temp) / 2);
283
284 analyze_ambient_temp(tp, last_warn, tp->sample_tick);
285 analyze_cpu_temp(tp, last_warn, tp->sample_tick);
286
287 tp->sample_tick++;
288 }
289
prioritize_fan_action(int which_fan)290 static enum fan_action prioritize_fan_action(int which_fan)
291 {
292 struct bbc_cpu_temperature *tp;
293 enum fan_action decision = FAN_STATE_MAX;
294
295 /* Basically, prioritize what the temperature sensors
296 * recommend we do, and perform that action on all the
297 * fans.
298 */
299 list_for_each_entry(tp, &all_temps, glob_list) {
300 if (tp->fan_todo[which_fan] == FAN_FULLBLAST) {
301 decision = FAN_FULLBLAST;
302 break;
303 }
304 if (tp->fan_todo[which_fan] == FAN_SAME &&
305 decision != FAN_FASTER)
306 decision = FAN_SAME;
307 else if (tp->fan_todo[which_fan] == FAN_FASTER)
308 decision = FAN_FASTER;
309 else if (decision != FAN_FASTER &&
310 decision != FAN_SAME &&
311 tp->fan_todo[which_fan] == FAN_SLOWER)
312 decision = FAN_SLOWER;
313 }
314 if (decision == FAN_STATE_MAX)
315 decision = FAN_SAME;
316
317 return decision;
318 }
319
maybe_new_ambient_fan_speed(struct bbc_fan_control * fp)320 static int maybe_new_ambient_fan_speed(struct bbc_fan_control *fp)
321 {
322 enum fan_action decision = prioritize_fan_action(FAN_AMBIENT);
323 int ret;
324
325 if (decision == FAN_SAME)
326 return 0;
327
328 ret = 1;
329 if (decision == FAN_FULLBLAST) {
330 if (fp->system_fan_speed >= FAN_SPEED_MAX)
331 ret = 0;
332 else
333 fp->system_fan_speed = FAN_SPEED_MAX;
334 } else {
335 if (decision == FAN_FASTER) {
336 if (fp->system_fan_speed >= FAN_SPEED_MAX)
337 ret = 0;
338 else
339 fp->system_fan_speed += 2;
340 } else {
341 int orig_speed = fp->system_fan_speed;
342
343 if (orig_speed <= FAN_SPEED_MIN ||
344 orig_speed <= (fp->cpu_fan_speed - 3))
345 ret = 0;
346 else
347 fp->system_fan_speed -= 1;
348 }
349 }
350
351 return ret;
352 }
353
maybe_new_cpu_fan_speed(struct bbc_fan_control * fp)354 static int maybe_new_cpu_fan_speed(struct bbc_fan_control *fp)
355 {
356 enum fan_action decision = prioritize_fan_action(FAN_CPU);
357 int ret;
358
359 if (decision == FAN_SAME)
360 return 0;
361
362 ret = 1;
363 if (decision == FAN_FULLBLAST) {
364 if (fp->cpu_fan_speed >= FAN_SPEED_MAX)
365 ret = 0;
366 else
367 fp->cpu_fan_speed = FAN_SPEED_MAX;
368 } else {
369 if (decision == FAN_FASTER) {
370 if (fp->cpu_fan_speed >= FAN_SPEED_MAX)
371 ret = 0;
372 else {
373 fp->cpu_fan_speed += 2;
374 if (fp->system_fan_speed <
375 (fp->cpu_fan_speed - 3))
376 fp->system_fan_speed =
377 fp->cpu_fan_speed - 3;
378 }
379 } else {
380 if (fp->cpu_fan_speed <= FAN_SPEED_MIN)
381 ret = 0;
382 else
383 fp->cpu_fan_speed -= 1;
384 }
385 }
386
387 return ret;
388 }
389
maybe_new_fan_speeds(struct bbc_fan_control * fp)390 static void maybe_new_fan_speeds(struct bbc_fan_control *fp)
391 {
392 int new;
393
394 new = maybe_new_ambient_fan_speed(fp);
395 new |= maybe_new_cpu_fan_speed(fp);
396
397 if (new)
398 set_fan_speeds(fp);
399 }
400
fans_full_blast(void)401 static void fans_full_blast(void)
402 {
403 struct bbc_fan_control *fp;
404
405 /* Since we will not be monitoring things anymore, put
406 * the fans on full blast.
407 */
408 list_for_each_entry(fp, &all_fans, glob_list) {
409 fp->cpu_fan_speed = FAN_SPEED_MAX;
410 fp->system_fan_speed = FAN_SPEED_MAX;
411 fp->psupply_fan_on = 1;
412 set_fan_speeds(fp);
413 }
414 }
415
416 #define POLL_INTERVAL (5 * 1000)
417 static unsigned long last_warning_jiffies;
418 static struct task_struct *kenvctrld_task;
419
kenvctrld(void * __unused)420 static int kenvctrld(void *__unused)
421 {
422 printk(KERN_INFO "bbc_envctrl: kenvctrld starting...\n");
423 last_warning_jiffies = jiffies - WARN_INTERVAL;
424 for (;;) {
425 struct bbc_cpu_temperature *tp;
426 struct bbc_fan_control *fp;
427
428 msleep_interruptible(POLL_INTERVAL);
429 if (kthread_should_stop())
430 break;
431
432 list_for_each_entry(tp, &all_temps, glob_list) {
433 get_current_temps(tp);
434 analyze_temps(tp, &last_warning_jiffies);
435 }
436 list_for_each_entry(fp, &all_fans, glob_list)
437 maybe_new_fan_speeds(fp);
438 }
439 printk(KERN_INFO "bbc_envctrl: kenvctrld exiting...\n");
440
441 fans_full_blast();
442
443 return 0;
444 }
445
attach_one_temp(struct bbc_i2c_bus * bp,struct platform_device * op,int temp_idx)446 static void attach_one_temp(struct bbc_i2c_bus *bp, struct platform_device *op,
447 int temp_idx)
448 {
449 struct bbc_cpu_temperature *tp;
450
451 tp = kzalloc(sizeof(*tp), GFP_KERNEL);
452 if (!tp)
453 return;
454
455 tp->client = bbc_i2c_attach(bp, op);
456 if (!tp->client) {
457 kfree(tp);
458 return;
459 }
460
461
462 tp->index = temp_idx;
463
464 list_add(&tp->glob_list, &all_temps);
465 list_add(&tp->bp_list, &bp->temps);
466
467 /* Tell it to convert once every 5 seconds, clear all cfg
468 * bits.
469 */
470 bbc_i2c_writeb(tp->client, 0x00, MAX1617_WR_CFG_BYTE);
471 bbc_i2c_writeb(tp->client, 0x02, MAX1617_WR_CVRATE_BYTE);
472
473 /* Program the hard temperature limits into the chip. */
474 bbc_i2c_writeb(tp->client, amb_temp_limits[tp->index].high_pwroff,
475 MAX1617_WR_AMB_HIGHLIM);
476 bbc_i2c_writeb(tp->client, amb_temp_limits[tp->index].low_pwroff,
477 MAX1617_WR_AMB_LOWLIM);
478 bbc_i2c_writeb(tp->client, cpu_temp_limits[tp->index].high_pwroff,
479 MAX1617_WR_CPU_HIGHLIM);
480 bbc_i2c_writeb(tp->client, cpu_temp_limits[tp->index].low_pwroff,
481 MAX1617_WR_CPU_LOWLIM);
482
483 get_current_temps(tp);
484 tp->prev_cpu_temp = tp->avg_cpu_temp = tp->curr_cpu_temp;
485 tp->prev_amb_temp = tp->avg_amb_temp = tp->curr_amb_temp;
486
487 tp->fan_todo[FAN_AMBIENT] = FAN_SAME;
488 tp->fan_todo[FAN_CPU] = FAN_SAME;
489 }
490
attach_one_fan(struct bbc_i2c_bus * bp,struct platform_device * op,int fan_idx)491 static void attach_one_fan(struct bbc_i2c_bus *bp, struct platform_device *op,
492 int fan_idx)
493 {
494 struct bbc_fan_control *fp;
495
496 fp = kzalloc(sizeof(*fp), GFP_KERNEL);
497 if (!fp)
498 return;
499
500 fp->client = bbc_i2c_attach(bp, op);
501 if (!fp->client) {
502 kfree(fp);
503 return;
504 }
505
506 fp->index = fan_idx;
507
508 list_add(&fp->glob_list, &all_fans);
509 list_add(&fp->bp_list, &bp->fans);
510
511 /* The i2c device controlling the fans is write-only.
512 * So the only way to keep track of the current power
513 * level fed to the fans is via software. Choose half
514 * power for cpu/system and 'on' fo the powersupply fan
515 * and set it now.
516 */
517 fp->psupply_fan_on = 1;
518 fp->cpu_fan_speed = (FAN_SPEED_MAX - FAN_SPEED_MIN) / 2;
519 fp->cpu_fan_speed += FAN_SPEED_MIN;
520 fp->system_fan_speed = (FAN_SPEED_MAX - FAN_SPEED_MIN) / 2;
521 fp->system_fan_speed += FAN_SPEED_MIN;
522
523 set_fan_speeds(fp);
524 }
525
destroy_one_temp(struct bbc_cpu_temperature * tp)526 static void destroy_one_temp(struct bbc_cpu_temperature *tp)
527 {
528 bbc_i2c_detach(tp->client);
529 kfree(tp);
530 }
531
destroy_all_temps(struct bbc_i2c_bus * bp)532 static void destroy_all_temps(struct bbc_i2c_bus *bp)
533 {
534 struct bbc_cpu_temperature *tp, *tpos;
535
536 list_for_each_entry_safe(tp, tpos, &bp->temps, bp_list) {
537 list_del(&tp->bp_list);
538 list_del(&tp->glob_list);
539 destroy_one_temp(tp);
540 }
541 }
542
destroy_one_fan(struct bbc_fan_control * fp)543 static void destroy_one_fan(struct bbc_fan_control *fp)
544 {
545 bbc_i2c_detach(fp->client);
546 kfree(fp);
547 }
548
destroy_all_fans(struct bbc_i2c_bus * bp)549 static void destroy_all_fans(struct bbc_i2c_bus *bp)
550 {
551 struct bbc_fan_control *fp, *fpos;
552
553 list_for_each_entry_safe(fp, fpos, &bp->fans, bp_list) {
554 list_del(&fp->bp_list);
555 list_del(&fp->glob_list);
556 destroy_one_fan(fp);
557 }
558 }
559
bbc_envctrl_init(struct bbc_i2c_bus * bp)560 int bbc_envctrl_init(struct bbc_i2c_bus *bp)
561 {
562 struct platform_device *op;
563 int temp_index = 0;
564 int fan_index = 0;
565 int devidx = 0;
566
567 while ((op = bbc_i2c_getdev(bp, devidx++)) != NULL) {
568 if (!strcmp(op->dev.of_node->name, "temperature"))
569 attach_one_temp(bp, op, temp_index++);
570 if (!strcmp(op->dev.of_node->name, "fan-control"))
571 attach_one_fan(bp, op, fan_index++);
572 }
573 if (temp_index != 0 && fan_index != 0) {
574 kenvctrld_task = kthread_run(kenvctrld, NULL, "kenvctrld");
575 if (IS_ERR(kenvctrld_task)) {
576 int err = PTR_ERR(kenvctrld_task);
577
578 kenvctrld_task = NULL;
579 destroy_all_temps(bp);
580 destroy_all_fans(bp);
581 return err;
582 }
583 }
584
585 return 0;
586 }
587
bbc_envctrl_cleanup(struct bbc_i2c_bus * bp)588 void bbc_envctrl_cleanup(struct bbc_i2c_bus *bp)
589 {
590 if (kenvctrld_task)
591 kthread_stop(kenvctrld_task);
592
593 destroy_all_temps(bp);
594 destroy_all_fans(bp);
595 }
596