Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Windfarm PowerMac thermal control.
4 * Control loops for PowerMac7,2 and 7,3
5 *
6 * Copyright (C) 2012 Benjamin Herrenschmidt, IBM Corp.
7 */
8#include <linux/types.h>
9#include <linux/errno.h>
10#include <linux/kernel.h>
11#include <linux/device.h>
12#include <linux/platform_device.h>
13#include <linux/reboot.h>
14#include <asm/prom.h>
15#include <asm/smu.h>
16
17#include "windfarm.h"
18#include "windfarm_pid.h"
19#include "windfarm_mpu.h"
20
21#define VERSION "1.0"
22
23#undef DEBUG
24#undef LOTSA_DEBUG
25
26#ifdef DEBUG
27#define DBG(args...) printk(args)
28#else
29#define DBG(args...) do { } while(0)
30#endif
31
32#ifdef LOTSA_DEBUG
33#define DBG_LOTS(args...) printk(args)
34#else
35#define DBG_LOTS(args...) do { } while(0)
36#endif
37
38/* define this to force CPU overtemp to 60 degree, useful for testing
39 * the overtemp code
40 */
41#undef HACKED_OVERTEMP
42
43/* We currently only handle 2 chips */
44#define NR_CHIPS 2
45#define NR_CPU_FANS 3 * NR_CHIPS
46
47/* Controls and sensors */
48static struct wf_sensor *sens_cpu_temp[NR_CHIPS];
49static struct wf_sensor *sens_cpu_volts[NR_CHIPS];
50static struct wf_sensor *sens_cpu_amps[NR_CHIPS];
51static struct wf_sensor *backside_temp;
52static struct wf_sensor *drives_temp;
53
54static struct wf_control *cpu_front_fans[NR_CHIPS];
55static struct wf_control *cpu_rear_fans[NR_CHIPS];
56static struct wf_control *cpu_pumps[NR_CHIPS];
57static struct wf_control *backside_fan;
58static struct wf_control *drives_fan;
59static struct wf_control *slots_fan;
60static struct wf_control *cpufreq_clamp;
61
62/* We keep a temperature history for average calculation of 180s */
63#define CPU_TEMP_HIST_SIZE 180
64
65/* Fixed speed for slot fan */
66#define SLOTS_FAN_DEFAULT_PWM 40
67
68/* Scale value for CPU intake fans */
69#define CPU_INTAKE_SCALE 0x0000f852
70
71/* PID loop state */
72static const struct mpu_data *cpu_mpu_data[NR_CHIPS];
73static struct wf_cpu_pid_state cpu_pid[NR_CHIPS];
74static bool cpu_pid_combined;
75static u32 cpu_thist[CPU_TEMP_HIST_SIZE];
76static int cpu_thist_pt;
77static s64 cpu_thist_total;
78static s32 cpu_all_tmax = 100 << 16;
79static struct wf_pid_state backside_pid;
80static int backside_tick;
81static struct wf_pid_state drives_pid;
82static int drives_tick;
83
84static int nr_chips;
85static bool have_all_controls;
86static bool have_all_sensors;
87static bool started;
88
89static int failure_state;
90#define FAILURE_SENSOR 1
91#define FAILURE_FAN 2
92#define FAILURE_PERM 4
93#define FAILURE_LOW_OVERTEMP 8
94#define FAILURE_HIGH_OVERTEMP 16
95
96/* Overtemp values */
97#define LOW_OVER_AVERAGE 0
98#define LOW_OVER_IMMEDIATE (10 << 16)
99#define LOW_OVER_CLEAR ((-10) << 16)
100#define HIGH_OVER_IMMEDIATE (14 << 16)
101#define HIGH_OVER_AVERAGE (10 << 16)
102#define HIGH_OVER_IMMEDIATE (14 << 16)
103
104
105static void cpu_max_all_fans(void)
106{
107 int i;
108
109 /* We max all CPU fans in case of a sensor error. We also do the
110 * cpufreq clamping now, even if it's supposedly done later by the
111 * generic code anyway, we do it earlier here to react faster
112 */
113 if (cpufreq_clamp)
114 wf_control_set_max(cpufreq_clamp);
115 for (i = 0; i < nr_chips; i++) {
116 if (cpu_front_fans[i])
117 wf_control_set_max(cpu_front_fans[i]);
118 if (cpu_rear_fans[i])
119 wf_control_set_max(cpu_rear_fans[i]);
120 if (cpu_pumps[i])
121 wf_control_set_max(cpu_pumps[i]);
122 }
123}
124
125static int cpu_check_overtemp(s32 temp)
126{
127 int new_state = 0;
128 s32 t_avg, t_old;
129 static bool first = true;
130
131 /* First check for immediate overtemps */
132 if (temp >= (cpu_all_tmax + LOW_OVER_IMMEDIATE)) {
133 new_state |= FAILURE_LOW_OVERTEMP;
134 if ((failure_state & FAILURE_LOW_OVERTEMP) == 0)
135 printk(KERN_ERR "windfarm: Overtemp due to immediate CPU"
136 " temperature !\n");
137 }
138 if (temp >= (cpu_all_tmax + HIGH_OVER_IMMEDIATE)) {
139 new_state |= FAILURE_HIGH_OVERTEMP;
140 if ((failure_state & FAILURE_HIGH_OVERTEMP) == 0)
141 printk(KERN_ERR "windfarm: Critical overtemp due to"
142 " immediate CPU temperature !\n");
143 }
144
145 /*
146 * The first time around, initialize the array with the first
147 * temperature reading
148 */
149 if (first) {
150 int i;
151
152 cpu_thist_total = 0;
153 for (i = 0; i < CPU_TEMP_HIST_SIZE; i++) {
154 cpu_thist[i] = temp;
155 cpu_thist_total += temp;
156 }
157 first = false;
158 }
159
160 /*
161 * We calculate a history of max temperatures and use that for the
162 * overtemp management
163 */
164 t_old = cpu_thist[cpu_thist_pt];
165 cpu_thist[cpu_thist_pt] = temp;
166 cpu_thist_pt = (cpu_thist_pt + 1) % CPU_TEMP_HIST_SIZE;
167 cpu_thist_total -= t_old;
168 cpu_thist_total += temp;
169 t_avg = cpu_thist_total / CPU_TEMP_HIST_SIZE;
170
171 DBG_LOTS(" t_avg = %d.%03d (out: %d.%03d, in: %d.%03d)\n",
172 FIX32TOPRINT(t_avg), FIX32TOPRINT(t_old), FIX32TOPRINT(temp));
173
174 /* Now check for average overtemps */
175 if (t_avg >= (cpu_all_tmax + LOW_OVER_AVERAGE)) {
176 new_state |= FAILURE_LOW_OVERTEMP;
177 if ((failure_state & FAILURE_LOW_OVERTEMP) == 0)
178 printk(KERN_ERR "windfarm: Overtemp due to average CPU"
179 " temperature !\n");
180 }
181 if (t_avg >= (cpu_all_tmax + HIGH_OVER_AVERAGE)) {
182 new_state |= FAILURE_HIGH_OVERTEMP;
183 if ((failure_state & FAILURE_HIGH_OVERTEMP) == 0)
184 printk(KERN_ERR "windfarm: Critical overtemp due to"
185 " average CPU temperature !\n");
186 }
187
188 /* Now handle overtemp conditions. We don't currently use the windfarm
189 * overtemp handling core as it's not fully suited to the needs of those
190 * new machine. This will be fixed later.
191 */
192 if (new_state) {
193 /* High overtemp -> immediate shutdown */
194 if (new_state & FAILURE_HIGH_OVERTEMP)
195 machine_power_off();
196 if ((failure_state & new_state) != new_state)
197 cpu_max_all_fans();
198 failure_state |= new_state;
199 } else if ((failure_state & FAILURE_LOW_OVERTEMP) &&
200 (temp < (cpu_all_tmax + LOW_OVER_CLEAR))) {
201 printk(KERN_ERR "windfarm: Overtemp condition cleared !\n");
202 failure_state &= ~FAILURE_LOW_OVERTEMP;
203 }
204
205 return failure_state & (FAILURE_LOW_OVERTEMP | FAILURE_HIGH_OVERTEMP);
206}
207
208static int read_one_cpu_vals(int cpu, s32 *temp, s32 *power)
209{
210 s32 dtemp, volts, amps;
211 int rc;
212
213 /* Get diode temperature */
214 rc = wf_sensor_get(sens_cpu_temp[cpu], &dtemp);
215 if (rc) {
216 DBG(" CPU%d: temp reading error !\n", cpu);
217 return -EIO;
218 }
219 DBG_LOTS(" CPU%d: temp = %d.%03d\n", cpu, FIX32TOPRINT((dtemp)));
220 *temp = dtemp;
221
222 /* Get voltage */
223 rc = wf_sensor_get(sens_cpu_volts[cpu], &volts);
224 if (rc) {
225 DBG(" CPU%d, volts reading error !\n", cpu);
226 return -EIO;
227 }
228 DBG_LOTS(" CPU%d: volts = %d.%03d\n", cpu, FIX32TOPRINT((volts)));
229
230 /* Get current */
231 rc = wf_sensor_get(sens_cpu_amps[cpu], &s);
232 if (rc) {
233 DBG(" CPU%d, current reading error !\n", cpu);
234 return -EIO;
235 }
236 DBG_LOTS(" CPU%d: amps = %d.%03d\n", cpu, FIX32TOPRINT((amps)));
237
238 /* Calculate power */
239
240 /* Scale voltage and current raw sensor values according to fixed scales
241 * obtained in Darwin and calculate power from I and V
242 */
243 *power = (((u64)volts) * ((u64)amps)) >> 16;
244
245 DBG_LOTS(" CPU%d: power = %d.%03d\n", cpu, FIX32TOPRINT((*power)));
246
247 return 0;
248
249}
250
251static void cpu_fans_tick_split(void)
252{
253 int err, cpu;
254 s32 intake, temp, power, t_max = 0;
255
256 DBG_LOTS("* cpu fans_tick_split()\n");
257
258 for (cpu = 0; cpu < nr_chips; ++cpu) {
259 struct wf_cpu_pid_state *sp = &cpu_pid[cpu];
260
261 /* Read current speed */
262 wf_control_get(cpu_rear_fans[cpu], &sp->target);
263
264 DBG_LOTS(" CPU%d: cur_target = %d RPM\n", cpu, sp->target);
265
266 err = read_one_cpu_vals(cpu, &temp, &power);
267 if (err) {
268 failure_state |= FAILURE_SENSOR;
269 cpu_max_all_fans();
270 return;
271 }
272
273 /* Keep track of highest temp */
274 t_max = max(t_max, temp);
275
276 /* Handle possible overtemps */
277 if (cpu_check_overtemp(t_max))
278 return;
279
280 /* Run PID */
281 wf_cpu_pid_run(sp, power, temp);
282
283 DBG_LOTS(" CPU%d: target = %d RPM\n", cpu, sp->target);
284
285 /* Apply result directly to exhaust fan */
286 err = wf_control_set(cpu_rear_fans[cpu], sp->target);
287 if (err) {
288 pr_warning("wf_pm72: Fan %s reports error %d\n",
289 cpu_rear_fans[cpu]->name, err);
290 failure_state |= FAILURE_FAN;
291 break;
292 }
293
294 /* Scale result for intake fan */
295 intake = (sp->target * CPU_INTAKE_SCALE) >> 16;
296 DBG_LOTS(" CPU%d: intake = %d RPM\n", cpu, intake);
297 err = wf_control_set(cpu_front_fans[cpu], intake);
298 if (err) {
299 pr_warning("wf_pm72: Fan %s reports error %d\n",
300 cpu_front_fans[cpu]->name, err);
301 failure_state |= FAILURE_FAN;
302 break;
303 }
304 }
305}
306
307static void cpu_fans_tick_combined(void)
308{
309 s32 temp0, power0, temp1, power1, t_max = 0;
310 s32 temp, power, intake, pump;
311 struct wf_control *pump0, *pump1;
312 struct wf_cpu_pid_state *sp = &cpu_pid[0];
313 int err, cpu;
314
315 DBG_LOTS("* cpu fans_tick_combined()\n");
316
317 /* Read current speed from cpu 0 */
318 wf_control_get(cpu_rear_fans[0], &sp->target);
319
320 DBG_LOTS(" CPUs: cur_target = %d RPM\n", sp->target);
321
322 /* Read values for both CPUs */
323 err = read_one_cpu_vals(0, &temp0, &power0);
324 if (err) {
325 failure_state |= FAILURE_SENSOR;
326 cpu_max_all_fans();
327 return;
328 }
329 err = read_one_cpu_vals(1, &temp1, &power1);
330 if (err) {
331 failure_state |= FAILURE_SENSOR;
332 cpu_max_all_fans();
333 return;
334 }
335
336 /* Keep track of highest temp */
337 t_max = max(t_max, max(temp0, temp1));
338
339 /* Handle possible overtemps */
340 if (cpu_check_overtemp(t_max))
341 return;
342
343 /* Use the max temp & power of both */
344 temp = max(temp0, temp1);
345 power = max(power0, power1);
346
347 /* Run PID */
348 wf_cpu_pid_run(sp, power, temp);
349
350 /* Scale result for intake fan */
351 intake = (sp->target * CPU_INTAKE_SCALE) >> 16;
352
353 /* Same deal with pump speed */
354 pump0 = cpu_pumps[0];
355 pump1 = cpu_pumps[1];
356 if (!pump0) {
357 pump0 = pump1;
358 pump1 = NULL;
359 }
360 pump = (sp->target * wf_control_get_max(pump0)) /
361 cpu_mpu_data[0]->rmaxn_exhaust_fan;
362
363 DBG_LOTS(" CPUs: target = %d RPM\n", sp->target);
364 DBG_LOTS(" CPUs: intake = %d RPM\n", intake);
365 DBG_LOTS(" CPUs: pump = %d RPM\n", pump);
366
367 for (cpu = 0; cpu < nr_chips; cpu++) {
368 err = wf_control_set(cpu_rear_fans[cpu], sp->target);
369 if (err) {
370 pr_warning("wf_pm72: Fan %s reports error %d\n",
371 cpu_rear_fans[cpu]->name, err);
372 failure_state |= FAILURE_FAN;
373 }
374 err = wf_control_set(cpu_front_fans[cpu], intake);
375 if (err) {
376 pr_warning("wf_pm72: Fan %s reports error %d\n",
377 cpu_front_fans[cpu]->name, err);
378 failure_state |= FAILURE_FAN;
379 }
380 err = 0;
381 if (cpu_pumps[cpu])
382 err = wf_control_set(cpu_pumps[cpu], pump);
383 if (err) {
384 pr_warning("wf_pm72: Pump %s reports error %d\n",
385 cpu_pumps[cpu]->name, err);
386 failure_state |= FAILURE_FAN;
387 }
388 }
389}
390
391/* Implementation... */
392static int cpu_setup_pid(int cpu)
393{
394 struct wf_cpu_pid_param pid;
395 const struct mpu_data *mpu = cpu_mpu_data[cpu];
396 s32 tmax, ttarget, ptarget;
397 int fmin, fmax, hsize;
398
399 /* Get PID params from the appropriate MPU EEPROM */
400 tmax = mpu->tmax << 16;
401 ttarget = mpu->ttarget << 16;
402 ptarget = ((s32)(mpu->pmaxh - mpu->padjmax)) << 16;
403
404 DBG("wf_72: CPU%d ttarget = %d.%03d, tmax = %d.%03d\n",
405 cpu, FIX32TOPRINT(ttarget), FIX32TOPRINT(tmax));
406
407 /* We keep a global tmax for overtemp calculations */
408 if (tmax < cpu_all_tmax)
409 cpu_all_tmax = tmax;
410
411 /* Set PID min/max by using the rear fan min/max */
412 fmin = wf_control_get_min(cpu_rear_fans[cpu]);
413 fmax = wf_control_get_max(cpu_rear_fans[cpu]);
414 DBG("wf_72: CPU%d max RPM range = [%d..%d]\n", cpu, fmin, fmax);
415
416 /* History size */
417 hsize = min_t(int, mpu->tguardband, WF_PID_MAX_HISTORY);
418 DBG("wf_72: CPU%d history size = %d\n", cpu, hsize);
419
420 /* Initialize PID loop */
421 pid.interval = 1; /* seconds */
422 pid.history_len = hsize;
423 pid.gd = mpu->pid_gd;
424 pid.gp = mpu->pid_gp;
425 pid.gr = mpu->pid_gr;
426 pid.tmax = tmax;
427 pid.ttarget = ttarget;
428 pid.pmaxadj = ptarget;
429 pid.min = fmin;
430 pid.max = fmax;
431
432 wf_cpu_pid_init(&cpu_pid[cpu], &pid);
433 cpu_pid[cpu].target = 1000;
434
435 return 0;
436}
437
438/* Backside/U3 fan */
439static struct wf_pid_param backside_u3_param = {
440 .interval = 5,
441 .history_len = 2,
442 .gd = 40 << 20,
443 .gp = 5 << 20,
444 .gr = 0,
445 .itarget = 65 << 16,
446 .additive = 1,
447 .min = 20,
448 .max = 100,
449};
450
451static struct wf_pid_param backside_u3h_param = {
452 .interval = 5,
453 .history_len = 2,
454 .gd = 20 << 20,
455 .gp = 5 << 20,
456 .gr = 0,
457 .itarget = 75 << 16,
458 .additive = 1,
459 .min = 20,
460 .max = 100,
461};
462
463static void backside_fan_tick(void)
464{
465 s32 temp;
466 int speed;
467 int err;
468
469 if (!backside_fan || !backside_temp || !backside_tick)
470 return;
471 if (--backside_tick > 0)
472 return;
473 backside_tick = backside_pid.param.interval;
474
475 DBG_LOTS("* backside fans tick\n");
476
477 /* Update fan speed from actual fans */
478 err = wf_control_get(backside_fan, &speed);
479 if (!err)
480 backside_pid.target = speed;
481
482 err = wf_sensor_get(backside_temp, &temp);
483 if (err) {
484 printk(KERN_WARNING "windfarm: U4 temp sensor error %d\n",
485 err);
486 failure_state |= FAILURE_SENSOR;
487 wf_control_set_max(backside_fan);
488 return;
489 }
490 speed = wf_pid_run(&backside_pid, temp);
491
492 DBG_LOTS("backside PID temp=%d.%.3d speed=%d\n",
493 FIX32TOPRINT(temp), speed);
494
495 err = wf_control_set(backside_fan, speed);
496 if (err) {
497 printk(KERN_WARNING "windfarm: backside fan error %d\n", err);
498 failure_state |= FAILURE_FAN;
499 }
500}
501
502static void backside_setup_pid(void)
503{
504 /* first time initialize things */
505 s32 fmin = wf_control_get_min(backside_fan);
506 s32 fmax = wf_control_get_max(backside_fan);
507 struct wf_pid_param param;
508 struct device_node *u3;
509 int u3h = 1; /* conservative by default */
510
511 u3 = of_find_node_by_path("/u3@0,f8000000");
512 if (u3 != NULL) {
513 const u32 *vers = of_get_property(u3, "device-rev", NULL);
514 if (vers)
515 if (((*vers) & 0x3f) < 0x34)
516 u3h = 0;
517 of_node_put(u3);
518 }
519
520 param = u3h ? backside_u3h_param : backside_u3_param;
521
522 param.min = max(param.min, fmin);
523 param.max = min(param.max, fmax);
524 wf_pid_init(&backside_pid, ¶m);
525 backside_tick = 1;
526
527 pr_info("wf_pm72: Backside control loop started.\n");
528}
529
530/* Drive bay fan */
531static const struct wf_pid_param drives_param = {
532 .interval = 5,
533 .history_len = 2,
534 .gd = 30 << 20,
535 .gp = 5 << 20,
536 .gr = 0,
537 .itarget = 40 << 16,
538 .additive = 1,
539 .min = 300,
540 .max = 4000,
541};
542
543static void drives_fan_tick(void)
544{
545 s32 temp;
546 int speed;
547 int err;
548
549 if (!drives_fan || !drives_temp || !drives_tick)
550 return;
551 if (--drives_tick > 0)
552 return;
553 drives_tick = drives_pid.param.interval;
554
555 DBG_LOTS("* drives fans tick\n");
556
557 /* Update fan speed from actual fans */
558 err = wf_control_get(drives_fan, &speed);
559 if (!err)
560 drives_pid.target = speed;
561
562 err = wf_sensor_get(drives_temp, &temp);
563 if (err) {
564 pr_warning("wf_pm72: drive bay temp sensor error %d\n", err);
565 failure_state |= FAILURE_SENSOR;
566 wf_control_set_max(drives_fan);
567 return;
568 }
569 speed = wf_pid_run(&drives_pid, temp);
570
571 DBG_LOTS("drives PID temp=%d.%.3d speed=%d\n",
572 FIX32TOPRINT(temp), speed);
573
574 err = wf_control_set(drives_fan, speed);
575 if (err) {
576 printk(KERN_WARNING "windfarm: drive bay fan error %d\n", err);
577 failure_state |= FAILURE_FAN;
578 }
579}
580
581static void drives_setup_pid(void)
582{
583 /* first time initialize things */
584 s32 fmin = wf_control_get_min(drives_fan);
585 s32 fmax = wf_control_get_max(drives_fan);
586 struct wf_pid_param param = drives_param;
587
588 param.min = max(param.min, fmin);
589 param.max = min(param.max, fmax);
590 wf_pid_init(&drives_pid, ¶m);
591 drives_tick = 1;
592
593 pr_info("wf_pm72: Drive bay control loop started.\n");
594}
595
596static void set_fail_state(void)
597{
598 cpu_max_all_fans();
599
600 if (backside_fan)
601 wf_control_set_max(backside_fan);
602 if (slots_fan)
603 wf_control_set_max(slots_fan);
604 if (drives_fan)
605 wf_control_set_max(drives_fan);
606}
607
608static void pm72_tick(void)
609{
610 int i, last_failure;
611
612 if (!started) {
613 started = true;
614 printk(KERN_INFO "windfarm: CPUs control loops started.\n");
615 for (i = 0; i < nr_chips; ++i) {
616 if (cpu_setup_pid(i) < 0) {
617 failure_state = FAILURE_PERM;
618 set_fail_state();
619 break;
620 }
621 }
622 DBG_LOTS("cpu_all_tmax=%d.%03d\n", FIX32TOPRINT(cpu_all_tmax));
623
624 backside_setup_pid();
625 drives_setup_pid();
626
627 /*
628 * We don't have the right stuff to drive the PCI fan
629 * so we fix it to a default value
630 */
631 wf_control_set(slots_fan, SLOTS_FAN_DEFAULT_PWM);
632
633#ifdef HACKED_OVERTEMP
634 cpu_all_tmax = 60 << 16;
635#endif
636 }
637
638 /* Permanent failure, bail out */
639 if (failure_state & FAILURE_PERM)
640 return;
641
642 /*
643 * Clear all failure bits except low overtemp which will be eventually
644 * cleared by the control loop itself
645 */
646 last_failure = failure_state;
647 failure_state &= FAILURE_LOW_OVERTEMP;
648 if (cpu_pid_combined)
649 cpu_fans_tick_combined();
650 else
651 cpu_fans_tick_split();
652 backside_fan_tick();
653 drives_fan_tick();
654
655 DBG_LOTS(" last_failure: 0x%x, failure_state: %x\n",
656 last_failure, failure_state);
657
658 /* Check for failures. Any failure causes cpufreq clamping */
659 if (failure_state && last_failure == 0 && cpufreq_clamp)
660 wf_control_set_max(cpufreq_clamp);
661 if (failure_state == 0 && last_failure && cpufreq_clamp)
662 wf_control_set_min(cpufreq_clamp);
663
664 /* That's it for now, we might want to deal with other failures
665 * differently in the future though
666 */
667}
668
669static void pm72_new_control(struct wf_control *ct)
670{
671 bool all_controls;
672 bool had_pump = cpu_pumps[0] || cpu_pumps[1];
673
674 if (!strcmp(ct->name, "cpu-front-fan-0"))
675 cpu_front_fans[0] = ct;
676 else if (!strcmp(ct->name, "cpu-front-fan-1"))
677 cpu_front_fans[1] = ct;
678 else if (!strcmp(ct->name, "cpu-rear-fan-0"))
679 cpu_rear_fans[0] = ct;
680 else if (!strcmp(ct->name, "cpu-rear-fan-1"))
681 cpu_rear_fans[1] = ct;
682 else if (!strcmp(ct->name, "cpu-pump-0"))
683 cpu_pumps[0] = ct;
684 else if (!strcmp(ct->name, "cpu-pump-1"))
685 cpu_pumps[1] = ct;
686 else if (!strcmp(ct->name, "backside-fan"))
687 backside_fan = ct;
688 else if (!strcmp(ct->name, "slots-fan"))
689 slots_fan = ct;
690 else if (!strcmp(ct->name, "drive-bay-fan"))
691 drives_fan = ct;
692 else if (!strcmp(ct->name, "cpufreq-clamp"))
693 cpufreq_clamp = ct;
694
695 all_controls =
696 cpu_front_fans[0] &&
697 cpu_rear_fans[0] &&
698 backside_fan &&
699 slots_fan &&
700 drives_fan;
701 if (nr_chips > 1)
702 all_controls &=
703 cpu_front_fans[1] &&
704 cpu_rear_fans[1];
705 have_all_controls = all_controls;
706
707 if ((cpu_pumps[0] || cpu_pumps[1]) && !had_pump) {
708 pr_info("wf_pm72: Liquid cooling pump(s) detected,"
709 " using new algorithm !\n");
710 cpu_pid_combined = true;
711 }
712}
713
714
715static void pm72_new_sensor(struct wf_sensor *sr)
716{
717 bool all_sensors;
718
719 if (!strcmp(sr->name, "cpu-diode-temp-0"))
720 sens_cpu_temp[0] = sr;
721 else if (!strcmp(sr->name, "cpu-diode-temp-1"))
722 sens_cpu_temp[1] = sr;
723 else if (!strcmp(sr->name, "cpu-voltage-0"))
724 sens_cpu_volts[0] = sr;
725 else if (!strcmp(sr->name, "cpu-voltage-1"))
726 sens_cpu_volts[1] = sr;
727 else if (!strcmp(sr->name, "cpu-current-0"))
728 sens_cpu_amps[0] = sr;
729 else if (!strcmp(sr->name, "cpu-current-1"))
730 sens_cpu_amps[1] = sr;
731 else if (!strcmp(sr->name, "backside-temp"))
732 backside_temp = sr;
733 else if (!strcmp(sr->name, "hd-temp"))
734 drives_temp = sr;
735
736 all_sensors =
737 sens_cpu_temp[0] &&
738 sens_cpu_volts[0] &&
739 sens_cpu_amps[0] &&
740 backside_temp &&
741 drives_temp;
742 if (nr_chips > 1)
743 all_sensors &=
744 sens_cpu_temp[1] &&
745 sens_cpu_volts[1] &&
746 sens_cpu_amps[1];
747
748 have_all_sensors = all_sensors;
749}
750
751static int pm72_wf_notify(struct notifier_block *self,
752 unsigned long event, void *data)
753{
754 switch (event) {
755 case WF_EVENT_NEW_SENSOR:
756 pm72_new_sensor(data);
757 break;
758 case WF_EVENT_NEW_CONTROL:
759 pm72_new_control(data);
760 break;
761 case WF_EVENT_TICK:
762 if (have_all_controls && have_all_sensors)
763 pm72_tick();
764 }
765 return 0;
766}
767
768static struct notifier_block pm72_events = {
769 .notifier_call = pm72_wf_notify,
770};
771
772static int wf_pm72_probe(struct platform_device *dev)
773{
774 wf_register_client(&pm72_events);
775 return 0;
776}
777
778static int wf_pm72_remove(struct platform_device *dev)
779{
780 wf_unregister_client(&pm72_events);
781
782 /* should release all sensors and controls */
783 return 0;
784}
785
786static struct platform_driver wf_pm72_driver = {
787 .probe = wf_pm72_probe,
788 .remove = wf_pm72_remove,
789 .driver = {
790 .name = "windfarm",
791 },
792};
793
794static int __init wf_pm72_init(void)
795{
796 struct device_node *cpu;
797 int i;
798
799 if (!of_machine_is_compatible("PowerMac7,2") &&
800 !of_machine_is_compatible("PowerMac7,3"))
801 return -ENODEV;
802
803 /* Count the number of CPU cores */
804 nr_chips = 0;
805 for_each_node_by_type(cpu, "cpu")
806 ++nr_chips;
807 if (nr_chips > NR_CHIPS)
808 nr_chips = NR_CHIPS;
809
810 pr_info("windfarm: Initializing for desktop G5 with %d chips\n",
811 nr_chips);
812
813 /* Get MPU data for each CPU */
814 for (i = 0; i < nr_chips; i++) {
815 cpu_mpu_data[i] = wf_get_mpu(i);
816 if (!cpu_mpu_data[i]) {
817 pr_err("wf_pm72: Failed to find MPU data for CPU %d\n", i);
818 return -ENXIO;
819 }
820 }
821
822#ifdef MODULE
823 request_module("windfarm_fcu_controls");
824 request_module("windfarm_lm75_sensor");
825 request_module("windfarm_ad7417_sensor");
826 request_module("windfarm_max6690_sensor");
827 request_module("windfarm_cpufreq_clamp");
828#endif /* MODULE */
829
830 platform_driver_register(&wf_pm72_driver);
831 return 0;
832}
833
834static void __exit wf_pm72_exit(void)
835{
836 platform_driver_unregister(&wf_pm72_driver);
837}
838
839module_init(wf_pm72_init);
840module_exit(wf_pm72_exit);
841
842MODULE_AUTHOR("Benjamin Herrenschmidt <benh@kernel.crashing.org>");
843MODULE_DESCRIPTION("Thermal control for AGP PowerMac G5s");
844MODULE_LICENSE("GPL");
845MODULE_ALIAS("platform:windfarm");
1/*
2 * Windfarm PowerMac thermal control.
3 * Control loops for PowerMac7,2 and 7,3
4 *
5 * Copyright (C) 2012 Benjamin Herrenschmidt, IBM Corp.
6 *
7 * Use and redistribute under the terms of the GNU GPL v2.
8 */
9#include <linux/types.h>
10#include <linux/errno.h>
11#include <linux/kernel.h>
12#include <linux/device.h>
13#include <linux/platform_device.h>
14#include <linux/reboot.h>
15#include <asm/prom.h>
16#include <asm/smu.h>
17
18#include "windfarm.h"
19#include "windfarm_pid.h"
20#include "windfarm_mpu.h"
21
22#define VERSION "1.0"
23
24#undef DEBUG
25#undef LOTSA_DEBUG
26
27#ifdef DEBUG
28#define DBG(args...) printk(args)
29#else
30#define DBG(args...) do { } while(0)
31#endif
32
33#ifdef LOTSA_DEBUG
34#define DBG_LOTS(args...) printk(args)
35#else
36#define DBG_LOTS(args...) do { } while(0)
37#endif
38
39/* define this to force CPU overtemp to 60 degree, useful for testing
40 * the overtemp code
41 */
42#undef HACKED_OVERTEMP
43
44/* We currently only handle 2 chips */
45#define NR_CHIPS 2
46#define NR_CPU_FANS 3 * NR_CHIPS
47
48/* Controls and sensors */
49static struct wf_sensor *sens_cpu_temp[NR_CHIPS];
50static struct wf_sensor *sens_cpu_volts[NR_CHIPS];
51static struct wf_sensor *sens_cpu_amps[NR_CHIPS];
52static struct wf_sensor *backside_temp;
53static struct wf_sensor *drives_temp;
54
55static struct wf_control *cpu_front_fans[NR_CHIPS];
56static struct wf_control *cpu_rear_fans[NR_CHIPS];
57static struct wf_control *cpu_pumps[NR_CHIPS];
58static struct wf_control *backside_fan;
59static struct wf_control *drives_fan;
60static struct wf_control *slots_fan;
61static struct wf_control *cpufreq_clamp;
62
63/* We keep a temperature history for average calculation of 180s */
64#define CPU_TEMP_HIST_SIZE 180
65
66/* Fixed speed for slot fan */
67#define SLOTS_FAN_DEFAULT_PWM 40
68
69/* Scale value for CPU intake fans */
70#define CPU_INTAKE_SCALE 0x0000f852
71
72/* PID loop state */
73static const struct mpu_data *cpu_mpu_data[NR_CHIPS];
74static struct wf_cpu_pid_state cpu_pid[NR_CHIPS];
75static bool cpu_pid_combined;
76static u32 cpu_thist[CPU_TEMP_HIST_SIZE];
77static int cpu_thist_pt;
78static s64 cpu_thist_total;
79static s32 cpu_all_tmax = 100 << 16;
80static struct wf_pid_state backside_pid;
81static int backside_tick;
82static struct wf_pid_state drives_pid;
83static int drives_tick;
84
85static int nr_chips;
86static bool have_all_controls;
87static bool have_all_sensors;
88static bool started;
89
90static int failure_state;
91#define FAILURE_SENSOR 1
92#define FAILURE_FAN 2
93#define FAILURE_PERM 4
94#define FAILURE_LOW_OVERTEMP 8
95#define FAILURE_HIGH_OVERTEMP 16
96
97/* Overtemp values */
98#define LOW_OVER_AVERAGE 0
99#define LOW_OVER_IMMEDIATE (10 << 16)
100#define LOW_OVER_CLEAR ((-10) << 16)
101#define HIGH_OVER_IMMEDIATE (14 << 16)
102#define HIGH_OVER_AVERAGE (10 << 16)
103#define HIGH_OVER_IMMEDIATE (14 << 16)
104
105
106static void cpu_max_all_fans(void)
107{
108 int i;
109
110 /* We max all CPU fans in case of a sensor error. We also do the
111 * cpufreq clamping now, even if it's supposedly done later by the
112 * generic code anyway, we do it earlier here to react faster
113 */
114 if (cpufreq_clamp)
115 wf_control_set_max(cpufreq_clamp);
116 for (i = 0; i < nr_chips; i++) {
117 if (cpu_front_fans[i])
118 wf_control_set_max(cpu_front_fans[i]);
119 if (cpu_rear_fans[i])
120 wf_control_set_max(cpu_rear_fans[i]);
121 if (cpu_pumps[i])
122 wf_control_set_max(cpu_pumps[i]);
123 }
124}
125
126static int cpu_check_overtemp(s32 temp)
127{
128 int new_state = 0;
129 s32 t_avg, t_old;
130 static bool first = true;
131
132 /* First check for immediate overtemps */
133 if (temp >= (cpu_all_tmax + LOW_OVER_IMMEDIATE)) {
134 new_state |= FAILURE_LOW_OVERTEMP;
135 if ((failure_state & FAILURE_LOW_OVERTEMP) == 0)
136 printk(KERN_ERR "windfarm: Overtemp due to immediate CPU"
137 " temperature !\n");
138 }
139 if (temp >= (cpu_all_tmax + HIGH_OVER_IMMEDIATE)) {
140 new_state |= FAILURE_HIGH_OVERTEMP;
141 if ((failure_state & FAILURE_HIGH_OVERTEMP) == 0)
142 printk(KERN_ERR "windfarm: Critical overtemp due to"
143 " immediate CPU temperature !\n");
144 }
145
146 /*
147 * The first time around, initialize the array with the first
148 * temperature reading
149 */
150 if (first) {
151 int i;
152
153 cpu_thist_total = 0;
154 for (i = 0; i < CPU_TEMP_HIST_SIZE; i++) {
155 cpu_thist[i] = temp;
156 cpu_thist_total += temp;
157 }
158 first = false;
159 }
160
161 /*
162 * We calculate a history of max temperatures and use that for the
163 * overtemp management
164 */
165 t_old = cpu_thist[cpu_thist_pt];
166 cpu_thist[cpu_thist_pt] = temp;
167 cpu_thist_pt = (cpu_thist_pt + 1) % CPU_TEMP_HIST_SIZE;
168 cpu_thist_total -= t_old;
169 cpu_thist_total += temp;
170 t_avg = cpu_thist_total / CPU_TEMP_HIST_SIZE;
171
172 DBG_LOTS(" t_avg = %d.%03d (out: %d.%03d, in: %d.%03d)\n",
173 FIX32TOPRINT(t_avg), FIX32TOPRINT(t_old), FIX32TOPRINT(temp));
174
175 /* Now check for average overtemps */
176 if (t_avg >= (cpu_all_tmax + LOW_OVER_AVERAGE)) {
177 new_state |= FAILURE_LOW_OVERTEMP;
178 if ((failure_state & FAILURE_LOW_OVERTEMP) == 0)
179 printk(KERN_ERR "windfarm: Overtemp due to average CPU"
180 " temperature !\n");
181 }
182 if (t_avg >= (cpu_all_tmax + HIGH_OVER_AVERAGE)) {
183 new_state |= FAILURE_HIGH_OVERTEMP;
184 if ((failure_state & FAILURE_HIGH_OVERTEMP) == 0)
185 printk(KERN_ERR "windfarm: Critical overtemp due to"
186 " average CPU temperature !\n");
187 }
188
189 /* Now handle overtemp conditions. We don't currently use the windfarm
190 * overtemp handling core as it's not fully suited to the needs of those
191 * new machine. This will be fixed later.
192 */
193 if (new_state) {
194 /* High overtemp -> immediate shutdown */
195 if (new_state & FAILURE_HIGH_OVERTEMP)
196 machine_power_off();
197 if ((failure_state & new_state) != new_state)
198 cpu_max_all_fans();
199 failure_state |= new_state;
200 } else if ((failure_state & FAILURE_LOW_OVERTEMP) &&
201 (temp < (cpu_all_tmax + LOW_OVER_CLEAR))) {
202 printk(KERN_ERR "windfarm: Overtemp condition cleared !\n");
203 failure_state &= ~FAILURE_LOW_OVERTEMP;
204 }
205
206 return failure_state & (FAILURE_LOW_OVERTEMP | FAILURE_HIGH_OVERTEMP);
207}
208
209static int read_one_cpu_vals(int cpu, s32 *temp, s32 *power)
210{
211 s32 dtemp, volts, amps;
212 int rc;
213
214 /* Get diode temperature */
215 rc = wf_sensor_get(sens_cpu_temp[cpu], &dtemp);
216 if (rc) {
217 DBG(" CPU%d: temp reading error !\n", cpu);
218 return -EIO;
219 }
220 DBG_LOTS(" CPU%d: temp = %d.%03d\n", cpu, FIX32TOPRINT((dtemp)));
221 *temp = dtemp;
222
223 /* Get voltage */
224 rc = wf_sensor_get(sens_cpu_volts[cpu], &volts);
225 if (rc) {
226 DBG(" CPU%d, volts reading error !\n", cpu);
227 return -EIO;
228 }
229 DBG_LOTS(" CPU%d: volts = %d.%03d\n", cpu, FIX32TOPRINT((volts)));
230
231 /* Get current */
232 rc = wf_sensor_get(sens_cpu_amps[cpu], &s);
233 if (rc) {
234 DBG(" CPU%d, current reading error !\n", cpu);
235 return -EIO;
236 }
237 DBG_LOTS(" CPU%d: amps = %d.%03d\n", cpu, FIX32TOPRINT((amps)));
238
239 /* Calculate power */
240
241 /* Scale voltage and current raw sensor values according to fixed scales
242 * obtained in Darwin and calculate power from I and V
243 */
244 *power = (((u64)volts) * ((u64)amps)) >> 16;
245
246 DBG_LOTS(" CPU%d: power = %d.%03d\n", cpu, FIX32TOPRINT((*power)));
247
248 return 0;
249
250}
251
252static void cpu_fans_tick_split(void)
253{
254 int err, cpu;
255 s32 intake, temp, power, t_max = 0;
256
257 DBG_LOTS("* cpu fans_tick_split()\n");
258
259 for (cpu = 0; cpu < nr_chips; ++cpu) {
260 struct wf_cpu_pid_state *sp = &cpu_pid[cpu];
261
262 /* Read current speed */
263 wf_control_get(cpu_rear_fans[cpu], &sp->target);
264
265 DBG_LOTS(" CPU%d: cur_target = %d RPM\n", cpu, sp->target);
266
267 err = read_one_cpu_vals(cpu, &temp, &power);
268 if (err) {
269 failure_state |= FAILURE_SENSOR;
270 cpu_max_all_fans();
271 return;
272 }
273
274 /* Keep track of highest temp */
275 t_max = max(t_max, temp);
276
277 /* Handle possible overtemps */
278 if (cpu_check_overtemp(t_max))
279 return;
280
281 /* Run PID */
282 wf_cpu_pid_run(sp, power, temp);
283
284 DBG_LOTS(" CPU%d: target = %d RPM\n", cpu, sp->target);
285
286 /* Apply result directly to exhaust fan */
287 err = wf_control_set(cpu_rear_fans[cpu], sp->target);
288 if (err) {
289 pr_warning("wf_pm72: Fan %s reports error %d\n",
290 cpu_rear_fans[cpu]->name, err);
291 failure_state |= FAILURE_FAN;
292 break;
293 }
294
295 /* Scale result for intake fan */
296 intake = (sp->target * CPU_INTAKE_SCALE) >> 16;
297 DBG_LOTS(" CPU%d: intake = %d RPM\n", cpu, intake);
298 err = wf_control_set(cpu_front_fans[cpu], intake);
299 if (err) {
300 pr_warning("wf_pm72: Fan %s reports error %d\n",
301 cpu_front_fans[cpu]->name, err);
302 failure_state |= FAILURE_FAN;
303 break;
304 }
305 }
306}
307
308static void cpu_fans_tick_combined(void)
309{
310 s32 temp0, power0, temp1, power1, t_max = 0;
311 s32 temp, power, intake, pump;
312 struct wf_control *pump0, *pump1;
313 struct wf_cpu_pid_state *sp = &cpu_pid[0];
314 int err, cpu;
315
316 DBG_LOTS("* cpu fans_tick_combined()\n");
317
318 /* Read current speed from cpu 0 */
319 wf_control_get(cpu_rear_fans[0], &sp->target);
320
321 DBG_LOTS(" CPUs: cur_target = %d RPM\n", sp->target);
322
323 /* Read values for both CPUs */
324 err = read_one_cpu_vals(0, &temp0, &power0);
325 if (err) {
326 failure_state |= FAILURE_SENSOR;
327 cpu_max_all_fans();
328 return;
329 }
330 err = read_one_cpu_vals(1, &temp1, &power1);
331 if (err) {
332 failure_state |= FAILURE_SENSOR;
333 cpu_max_all_fans();
334 return;
335 }
336
337 /* Keep track of highest temp */
338 t_max = max(t_max, max(temp0, temp1));
339
340 /* Handle possible overtemps */
341 if (cpu_check_overtemp(t_max))
342 return;
343
344 /* Use the max temp & power of both */
345 temp = max(temp0, temp1);
346 power = max(power0, power1);
347
348 /* Run PID */
349 wf_cpu_pid_run(sp, power, temp);
350
351 /* Scale result for intake fan */
352 intake = (sp->target * CPU_INTAKE_SCALE) >> 16;
353
354 /* Same deal with pump speed */
355 pump0 = cpu_pumps[0];
356 pump1 = cpu_pumps[1];
357 if (!pump0) {
358 pump0 = pump1;
359 pump1 = NULL;
360 }
361 pump = (sp->target * wf_control_get_max(pump0)) /
362 cpu_mpu_data[0]->rmaxn_exhaust_fan;
363
364 DBG_LOTS(" CPUs: target = %d RPM\n", sp->target);
365 DBG_LOTS(" CPUs: intake = %d RPM\n", intake);
366 DBG_LOTS(" CPUs: pump = %d RPM\n", pump);
367
368 for (cpu = 0; cpu < nr_chips; cpu++) {
369 err = wf_control_set(cpu_rear_fans[cpu], sp->target);
370 if (err) {
371 pr_warning("wf_pm72: Fan %s reports error %d\n",
372 cpu_rear_fans[cpu]->name, err);
373 failure_state |= FAILURE_FAN;
374 }
375 err = wf_control_set(cpu_front_fans[cpu], intake);
376 if (err) {
377 pr_warning("wf_pm72: Fan %s reports error %d\n",
378 cpu_front_fans[cpu]->name, err);
379 failure_state |= FAILURE_FAN;
380 }
381 err = 0;
382 if (cpu_pumps[cpu])
383 err = wf_control_set(cpu_pumps[cpu], pump);
384 if (err) {
385 pr_warning("wf_pm72: Pump %s reports error %d\n",
386 cpu_pumps[cpu]->name, err);
387 failure_state |= FAILURE_FAN;
388 }
389 }
390}
391
392/* Implementation... */
393static int cpu_setup_pid(int cpu)
394{
395 struct wf_cpu_pid_param pid;
396 const struct mpu_data *mpu = cpu_mpu_data[cpu];
397 s32 tmax, ttarget, ptarget;
398 int fmin, fmax, hsize;
399
400 /* Get PID params from the appropriate MPU EEPROM */
401 tmax = mpu->tmax << 16;
402 ttarget = mpu->ttarget << 16;
403 ptarget = ((s32)(mpu->pmaxh - mpu->padjmax)) << 16;
404
405 DBG("wf_72: CPU%d ttarget = %d.%03d, tmax = %d.%03d\n",
406 cpu, FIX32TOPRINT(ttarget), FIX32TOPRINT(tmax));
407
408 /* We keep a global tmax for overtemp calculations */
409 if (tmax < cpu_all_tmax)
410 cpu_all_tmax = tmax;
411
412 /* Set PID min/max by using the rear fan min/max */
413 fmin = wf_control_get_min(cpu_rear_fans[cpu]);
414 fmax = wf_control_get_max(cpu_rear_fans[cpu]);
415 DBG("wf_72: CPU%d max RPM range = [%d..%d]\n", cpu, fmin, fmax);
416
417 /* History size */
418 hsize = min_t(int, mpu->tguardband, WF_PID_MAX_HISTORY);
419 DBG("wf_72: CPU%d history size = %d\n", cpu, hsize);
420
421 /* Initialize PID loop */
422 pid.interval = 1; /* seconds */
423 pid.history_len = hsize;
424 pid.gd = mpu->pid_gd;
425 pid.gp = mpu->pid_gp;
426 pid.gr = mpu->pid_gr;
427 pid.tmax = tmax;
428 pid.ttarget = ttarget;
429 pid.pmaxadj = ptarget;
430 pid.min = fmin;
431 pid.max = fmax;
432
433 wf_cpu_pid_init(&cpu_pid[cpu], &pid);
434 cpu_pid[cpu].target = 1000;
435
436 return 0;
437}
438
439/* Backside/U3 fan */
440static struct wf_pid_param backside_u3_param = {
441 .interval = 5,
442 .history_len = 2,
443 .gd = 40 << 20,
444 .gp = 5 << 20,
445 .gr = 0,
446 .itarget = 65 << 16,
447 .additive = 1,
448 .min = 20,
449 .max = 100,
450};
451
452static struct wf_pid_param backside_u3h_param = {
453 .interval = 5,
454 .history_len = 2,
455 .gd = 20 << 20,
456 .gp = 5 << 20,
457 .gr = 0,
458 .itarget = 75 << 16,
459 .additive = 1,
460 .min = 20,
461 .max = 100,
462};
463
464static void backside_fan_tick(void)
465{
466 s32 temp;
467 int speed;
468 int err;
469
470 if (!backside_fan || !backside_temp || !backside_tick)
471 return;
472 if (--backside_tick > 0)
473 return;
474 backside_tick = backside_pid.param.interval;
475
476 DBG_LOTS("* backside fans tick\n");
477
478 /* Update fan speed from actual fans */
479 err = wf_control_get(backside_fan, &speed);
480 if (!err)
481 backside_pid.target = speed;
482
483 err = wf_sensor_get(backside_temp, &temp);
484 if (err) {
485 printk(KERN_WARNING "windfarm: U4 temp sensor error %d\n",
486 err);
487 failure_state |= FAILURE_SENSOR;
488 wf_control_set_max(backside_fan);
489 return;
490 }
491 speed = wf_pid_run(&backside_pid, temp);
492
493 DBG_LOTS("backside PID temp=%d.%.3d speed=%d\n",
494 FIX32TOPRINT(temp), speed);
495
496 err = wf_control_set(backside_fan, speed);
497 if (err) {
498 printk(KERN_WARNING "windfarm: backside fan error %d\n", err);
499 failure_state |= FAILURE_FAN;
500 }
501}
502
503static void backside_setup_pid(void)
504{
505 /* first time initialize things */
506 s32 fmin = wf_control_get_min(backside_fan);
507 s32 fmax = wf_control_get_max(backside_fan);
508 struct wf_pid_param param;
509 struct device_node *u3;
510 int u3h = 1; /* conservative by default */
511
512 u3 = of_find_node_by_path("/u3@0,f8000000");
513 if (u3 != NULL) {
514 const u32 *vers = of_get_property(u3, "device-rev", NULL);
515 if (vers)
516 if (((*vers) & 0x3f) < 0x34)
517 u3h = 0;
518 of_node_put(u3);
519 }
520
521 param = u3h ? backside_u3h_param : backside_u3_param;
522
523 param.min = max(param.min, fmin);
524 param.max = min(param.max, fmax);
525 wf_pid_init(&backside_pid, ¶m);
526 backside_tick = 1;
527
528 pr_info("wf_pm72: Backside control loop started.\n");
529}
530
531/* Drive bay fan */
532static const struct wf_pid_param drives_param = {
533 .interval = 5,
534 .history_len = 2,
535 .gd = 30 << 20,
536 .gp = 5 << 20,
537 .gr = 0,
538 .itarget = 40 << 16,
539 .additive = 1,
540 .min = 300,
541 .max = 4000,
542};
543
544static void drives_fan_tick(void)
545{
546 s32 temp;
547 int speed;
548 int err;
549
550 if (!drives_fan || !drives_temp || !drives_tick)
551 return;
552 if (--drives_tick > 0)
553 return;
554 drives_tick = drives_pid.param.interval;
555
556 DBG_LOTS("* drives fans tick\n");
557
558 /* Update fan speed from actual fans */
559 err = wf_control_get(drives_fan, &speed);
560 if (!err)
561 drives_pid.target = speed;
562
563 err = wf_sensor_get(drives_temp, &temp);
564 if (err) {
565 pr_warning("wf_pm72: drive bay temp sensor error %d\n", err);
566 failure_state |= FAILURE_SENSOR;
567 wf_control_set_max(drives_fan);
568 return;
569 }
570 speed = wf_pid_run(&drives_pid, temp);
571
572 DBG_LOTS("drives PID temp=%d.%.3d speed=%d\n",
573 FIX32TOPRINT(temp), speed);
574
575 err = wf_control_set(drives_fan, speed);
576 if (err) {
577 printk(KERN_WARNING "windfarm: drive bay fan error %d\n", err);
578 failure_state |= FAILURE_FAN;
579 }
580}
581
582static void drives_setup_pid(void)
583{
584 /* first time initialize things */
585 s32 fmin = wf_control_get_min(drives_fan);
586 s32 fmax = wf_control_get_max(drives_fan);
587 struct wf_pid_param param = drives_param;
588
589 param.min = max(param.min, fmin);
590 param.max = min(param.max, fmax);
591 wf_pid_init(&drives_pid, ¶m);
592 drives_tick = 1;
593
594 pr_info("wf_pm72: Drive bay control loop started.\n");
595}
596
597static void set_fail_state(void)
598{
599 cpu_max_all_fans();
600
601 if (backside_fan)
602 wf_control_set_max(backside_fan);
603 if (slots_fan)
604 wf_control_set_max(slots_fan);
605 if (drives_fan)
606 wf_control_set_max(drives_fan);
607}
608
609static void pm72_tick(void)
610{
611 int i, last_failure;
612
613 if (!started) {
614 started = 1;
615 printk(KERN_INFO "windfarm: CPUs control loops started.\n");
616 for (i = 0; i < nr_chips; ++i) {
617 if (cpu_setup_pid(i) < 0) {
618 failure_state = FAILURE_PERM;
619 set_fail_state();
620 break;
621 }
622 }
623 DBG_LOTS("cpu_all_tmax=%d.%03d\n", FIX32TOPRINT(cpu_all_tmax));
624
625 backside_setup_pid();
626 drives_setup_pid();
627
628 /*
629 * We don't have the right stuff to drive the PCI fan
630 * so we fix it to a default value
631 */
632 wf_control_set(slots_fan, SLOTS_FAN_DEFAULT_PWM);
633
634#ifdef HACKED_OVERTEMP
635 cpu_all_tmax = 60 << 16;
636#endif
637 }
638
639 /* Permanent failure, bail out */
640 if (failure_state & FAILURE_PERM)
641 return;
642
643 /*
644 * Clear all failure bits except low overtemp which will be eventually
645 * cleared by the control loop itself
646 */
647 last_failure = failure_state;
648 failure_state &= FAILURE_LOW_OVERTEMP;
649 if (cpu_pid_combined)
650 cpu_fans_tick_combined();
651 else
652 cpu_fans_tick_split();
653 backside_fan_tick();
654 drives_fan_tick();
655
656 DBG_LOTS(" last_failure: 0x%x, failure_state: %x\n",
657 last_failure, failure_state);
658
659 /* Check for failures. Any failure causes cpufreq clamping */
660 if (failure_state && last_failure == 0 && cpufreq_clamp)
661 wf_control_set_max(cpufreq_clamp);
662 if (failure_state == 0 && last_failure && cpufreq_clamp)
663 wf_control_set_min(cpufreq_clamp);
664
665 /* That's it for now, we might want to deal with other failures
666 * differently in the future though
667 */
668}
669
670static void pm72_new_control(struct wf_control *ct)
671{
672 bool all_controls;
673 bool had_pump = cpu_pumps[0] || cpu_pumps[1];
674
675 if (!strcmp(ct->name, "cpu-front-fan-0"))
676 cpu_front_fans[0] = ct;
677 else if (!strcmp(ct->name, "cpu-front-fan-1"))
678 cpu_front_fans[1] = ct;
679 else if (!strcmp(ct->name, "cpu-rear-fan-0"))
680 cpu_rear_fans[0] = ct;
681 else if (!strcmp(ct->name, "cpu-rear-fan-1"))
682 cpu_rear_fans[1] = ct;
683 else if (!strcmp(ct->name, "cpu-pump-0"))
684 cpu_pumps[0] = ct;
685 else if (!strcmp(ct->name, "cpu-pump-1"))
686 cpu_pumps[1] = ct;
687 else if (!strcmp(ct->name, "backside-fan"))
688 backside_fan = ct;
689 else if (!strcmp(ct->name, "slots-fan"))
690 slots_fan = ct;
691 else if (!strcmp(ct->name, "drive-bay-fan"))
692 drives_fan = ct;
693 else if (!strcmp(ct->name, "cpufreq-clamp"))
694 cpufreq_clamp = ct;
695
696 all_controls =
697 cpu_front_fans[0] &&
698 cpu_rear_fans[0] &&
699 backside_fan &&
700 slots_fan &&
701 drives_fan;
702 if (nr_chips > 1)
703 all_controls &=
704 cpu_front_fans[1] &&
705 cpu_rear_fans[1];
706 have_all_controls = all_controls;
707
708 if ((cpu_pumps[0] || cpu_pumps[1]) && !had_pump) {
709 pr_info("wf_pm72: Liquid cooling pump(s) detected,"
710 " using new algorithm !\n");
711 cpu_pid_combined = true;
712 }
713}
714
715
716static void pm72_new_sensor(struct wf_sensor *sr)
717{
718 bool all_sensors;
719
720 if (!strcmp(sr->name, "cpu-diode-temp-0"))
721 sens_cpu_temp[0] = sr;
722 else if (!strcmp(sr->name, "cpu-diode-temp-1"))
723 sens_cpu_temp[1] = sr;
724 else if (!strcmp(sr->name, "cpu-voltage-0"))
725 sens_cpu_volts[0] = sr;
726 else if (!strcmp(sr->name, "cpu-voltage-1"))
727 sens_cpu_volts[1] = sr;
728 else if (!strcmp(sr->name, "cpu-current-0"))
729 sens_cpu_amps[0] = sr;
730 else if (!strcmp(sr->name, "cpu-current-1"))
731 sens_cpu_amps[1] = sr;
732 else if (!strcmp(sr->name, "backside-temp"))
733 backside_temp = sr;
734 else if (!strcmp(sr->name, "hd-temp"))
735 drives_temp = sr;
736
737 all_sensors =
738 sens_cpu_temp[0] &&
739 sens_cpu_volts[0] &&
740 sens_cpu_amps[0] &&
741 backside_temp &&
742 drives_temp;
743 if (nr_chips > 1)
744 all_sensors &=
745 sens_cpu_temp[1] &&
746 sens_cpu_volts[1] &&
747 sens_cpu_amps[1];
748
749 have_all_sensors = all_sensors;
750}
751
752static int pm72_wf_notify(struct notifier_block *self,
753 unsigned long event, void *data)
754{
755 switch (event) {
756 case WF_EVENT_NEW_SENSOR:
757 pm72_new_sensor(data);
758 break;
759 case WF_EVENT_NEW_CONTROL:
760 pm72_new_control(data);
761 break;
762 case WF_EVENT_TICK:
763 if (have_all_controls && have_all_sensors)
764 pm72_tick();
765 }
766 return 0;
767}
768
769static struct notifier_block pm72_events = {
770 .notifier_call = pm72_wf_notify,
771};
772
773static int wf_pm72_probe(struct platform_device *dev)
774{
775 wf_register_client(&pm72_events);
776 return 0;
777}
778
779static int wf_pm72_remove(struct platform_device *dev)
780{
781 wf_unregister_client(&pm72_events);
782
783 /* should release all sensors and controls */
784 return 0;
785}
786
787static struct platform_driver wf_pm72_driver = {
788 .probe = wf_pm72_probe,
789 .remove = wf_pm72_remove,
790 .driver = {
791 .name = "windfarm",
792 .owner = THIS_MODULE,
793 },
794};
795
796static int __init wf_pm72_init(void)
797{
798 struct device_node *cpu;
799 int i;
800
801 if (!of_machine_is_compatible("PowerMac7,2") &&
802 !of_machine_is_compatible("PowerMac7,3"))
803 return -ENODEV;
804
805 /* Count the number of CPU cores */
806 nr_chips = 0;
807 for_each_node_by_type(cpu, "cpu")
808 ++nr_chips;
809 if (nr_chips > NR_CHIPS)
810 nr_chips = NR_CHIPS;
811
812 pr_info("windfarm: Initializing for desktop G5 with %d chips\n",
813 nr_chips);
814
815 /* Get MPU data for each CPU */
816 for (i = 0; i < nr_chips; i++) {
817 cpu_mpu_data[i] = wf_get_mpu(i);
818 if (!cpu_mpu_data[i]) {
819 pr_err("wf_pm72: Failed to find MPU data for CPU %d\n", i);
820 return -ENXIO;
821 }
822 }
823
824#ifdef MODULE
825 request_module("windfarm_fcu_controls");
826 request_module("windfarm_lm75_sensor");
827 request_module("windfarm_ad7417_sensor");
828 request_module("windfarm_max6690_sensor");
829 request_module("windfarm_cpufreq_clamp");
830#endif /* MODULE */
831
832 platform_driver_register(&wf_pm72_driver);
833 return 0;
834}
835
836static void __exit wf_pm72_exit(void)
837{
838 platform_driver_unregister(&wf_pm72_driver);
839}
840
841module_init(wf_pm72_init);
842module_exit(wf_pm72_exit);
843
844MODULE_AUTHOR("Benjamin Herrenschmidt <benh@kernel.crashing.org>");
845MODULE_DESCRIPTION("Thermal control for AGP PowerMac G5s");
846MODULE_LICENSE("GPL");
847MODULE_ALIAS("platform:windfarm");