2 * Windfarm PowerMac thermal control.
3 * Control loops for RackMack3,1 (Xserve G5)
5 * Copyright (C) 2012 Benjamin Herrenschmidt, IBM Corp.
7 * Use and redistribute under the terms of the GNU GPL v2.
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>
19 #include "windfarm_pid.h"
20 #include "windfarm_mpu.h"
28 #define DBG(args...) printk(args)
30 #define DBG(args...) do { } while(0)
34 #define DBG_LOTS(args...) printk(args)
36 #define DBG_LOTS(args...) do { } while(0)
39 /* define this to force CPU overtemp to 60 degree, useful for testing
42 #undef HACKED_OVERTEMP
44 /* We currently only handle 2 chips */
46 #define NR_CPU_FANS 3 * NR_CHIPS
48 /* Controls and sensors */
49 static struct wf_sensor *sens_cpu_temp[NR_CHIPS];
50 static struct wf_sensor *sens_cpu_volts[NR_CHIPS];
51 static struct wf_sensor *sens_cpu_amps[NR_CHIPS];
52 static struct wf_sensor *backside_temp;
53 static struct wf_sensor *slots_temp;
54 static struct wf_sensor *dimms_temp;
56 static struct wf_control *cpu_fans[NR_CHIPS][3];
57 static struct wf_control *backside_fan;
58 static struct wf_control *slots_fan;
59 static struct wf_control *cpufreq_clamp;
61 /* We keep a temperature history for average calculation of 180s */
62 #define CPU_TEMP_HIST_SIZE 180
65 static const struct mpu_data *cpu_mpu_data[NR_CHIPS];
66 static struct wf_cpu_pid_state cpu_pid[NR_CHIPS];
67 static u32 cpu_thist[CPU_TEMP_HIST_SIZE];
68 static int cpu_thist_pt;
69 static s64 cpu_thist_total;
70 static s32 cpu_all_tmax = 100 << 16;
71 static struct wf_pid_state backside_pid;
72 static int backside_tick;
73 static struct wf_pid_state slots_pid;
74 static int slots_tick;
75 static int slots_speed;
76 static struct wf_pid_state dimms_pid;
77 static int dimms_output_clamp;
80 static bool have_all_controls;
81 static bool have_all_sensors;
84 static int failure_state;
85 #define FAILURE_SENSOR 1
87 #define FAILURE_PERM 4
88 #define FAILURE_LOW_OVERTEMP 8
89 #define FAILURE_HIGH_OVERTEMP 16
92 #define LOW_OVER_AVERAGE 0
93 #define LOW_OVER_IMMEDIATE (10 << 16)
94 #define LOW_OVER_CLEAR ((-10) << 16)
95 #define HIGH_OVER_IMMEDIATE (14 << 16)
96 #define HIGH_OVER_AVERAGE (10 << 16)
97 #define HIGH_OVER_IMMEDIATE (14 << 16)
100 static void cpu_max_all_fans(void)
104 /* We max all CPU fans in case of a sensor error. We also do the
105 * cpufreq clamping now, even if it's supposedly done later by the
106 * generic code anyway, we do it earlier here to react faster
109 wf_control_set_max(cpufreq_clamp);
110 for (i = 0; i < nr_chips; i++) {
112 wf_control_set_max(cpu_fans[i][0]);
114 wf_control_set_max(cpu_fans[i][1]);
116 wf_control_set_max(cpu_fans[i][2]);
120 static int cpu_check_overtemp(s32 temp)
124 static bool first = true;
126 /* First check for immediate overtemps */
127 if (temp >= (cpu_all_tmax + LOW_OVER_IMMEDIATE)) {
128 new_state |= FAILURE_LOW_OVERTEMP;
129 if ((failure_state & FAILURE_LOW_OVERTEMP) == 0)
130 printk(KERN_ERR "windfarm: Overtemp due to immediate CPU"
133 if (temp >= (cpu_all_tmax + HIGH_OVER_IMMEDIATE)) {
134 new_state |= FAILURE_HIGH_OVERTEMP;
135 if ((failure_state & FAILURE_HIGH_OVERTEMP) == 0)
136 printk(KERN_ERR "windfarm: Critical overtemp due to"
137 " immediate CPU temperature !\n");
141 * The first time around, initialize the array with the first
142 * temperature reading
148 for (i = 0; i < CPU_TEMP_HIST_SIZE; i++) {
150 cpu_thist_total += temp;
156 * We calculate a history of max temperatures and use that for the
157 * overtemp management
159 t_old = cpu_thist[cpu_thist_pt];
160 cpu_thist[cpu_thist_pt] = temp;
161 cpu_thist_pt = (cpu_thist_pt + 1) % CPU_TEMP_HIST_SIZE;
162 cpu_thist_total -= t_old;
163 cpu_thist_total += temp;
164 t_avg = cpu_thist_total / CPU_TEMP_HIST_SIZE;
166 DBG_LOTS(" t_avg = %d.%03d (out: %d.%03d, in: %d.%03d)\n",
167 FIX32TOPRINT(t_avg), FIX32TOPRINT(t_old), FIX32TOPRINT(temp));
169 /* Now check for average overtemps */
170 if (t_avg >= (cpu_all_tmax + LOW_OVER_AVERAGE)) {
171 new_state |= FAILURE_LOW_OVERTEMP;
172 if ((failure_state & FAILURE_LOW_OVERTEMP) == 0)
173 printk(KERN_ERR "windfarm: Overtemp due to average CPU"
176 if (t_avg >= (cpu_all_tmax + HIGH_OVER_AVERAGE)) {
177 new_state |= FAILURE_HIGH_OVERTEMP;
178 if ((failure_state & FAILURE_HIGH_OVERTEMP) == 0)
179 printk(KERN_ERR "windfarm: Critical overtemp due to"
180 " average CPU temperature !\n");
183 /* Now handle overtemp conditions. We don't currently use the windfarm
184 * overtemp handling core as it's not fully suited to the needs of those
185 * new machine. This will be fixed later.
188 /* High overtemp -> immediate shutdown */
189 if (new_state & FAILURE_HIGH_OVERTEMP)
191 if ((failure_state & new_state) != new_state)
193 failure_state |= new_state;
194 } else if ((failure_state & FAILURE_LOW_OVERTEMP) &&
195 (temp < (cpu_all_tmax + LOW_OVER_CLEAR))) {
196 printk(KERN_ERR "windfarm: Overtemp condition cleared !\n");
197 failure_state &= ~FAILURE_LOW_OVERTEMP;
200 return failure_state & (FAILURE_LOW_OVERTEMP | FAILURE_HIGH_OVERTEMP);
203 static int read_one_cpu_vals(int cpu, s32 *temp, s32 *power)
205 s32 dtemp, volts, amps;
208 /* Get diode temperature */
209 rc = wf_sensor_get(sens_cpu_temp[cpu], &dtemp);
211 DBG(" CPU%d: temp reading error !\n", cpu);
214 DBG_LOTS(" CPU%d: temp = %d.%03d\n", cpu, FIX32TOPRINT((dtemp)));
218 rc = wf_sensor_get(sens_cpu_volts[cpu], &volts);
220 DBG(" CPU%d, volts reading error !\n", cpu);
223 DBG_LOTS(" CPU%d: volts = %d.%03d\n", cpu, FIX32TOPRINT((volts)));
226 rc = wf_sensor_get(sens_cpu_amps[cpu], &s);
228 DBG(" CPU%d, current reading error !\n", cpu);
231 DBG_LOTS(" CPU%d: amps = %d.%03d\n", cpu, FIX32TOPRINT((amps)));
233 /* Calculate power */
235 /* Scale voltage and current raw sensor values according to fixed scales
236 * obtained in Darwin and calculate power from I and V
238 *power = (((u64)volts) * ((u64)amps)) >> 16;
240 DBG_LOTS(" CPU%d: power = %d.%03d\n", cpu, FIX32TOPRINT((*power)));
246 static void cpu_fans_tick(void)
249 s32 speed, temp, power, t_max = 0;
251 DBG_LOTS("* cpu fans_tick_split()\n");
253 for (cpu = 0; cpu < nr_chips; ++cpu) {
254 struct wf_cpu_pid_state *sp = &cpu_pid[cpu];
256 /* Read current speed */
257 wf_control_get(cpu_fans[cpu][0], &sp->target);
259 err = read_one_cpu_vals(cpu, &temp, &power);
261 failure_state |= FAILURE_SENSOR;
266 /* Keep track of highest temp */
267 t_max = max(t_max, temp);
269 /* Handle possible overtemps */
270 if (cpu_check_overtemp(t_max))
274 wf_cpu_pid_run(sp, power, temp);
276 DBG_LOTS(" CPU%d: target = %d RPM\n", cpu, sp->target);
278 /* Apply DIMMs clamp */
279 speed = max(sp->target, dimms_output_clamp);
281 /* Apply result to all cpu fans */
282 for (i = 0; i < 3; i++) {
283 err = wf_control_set(cpu_fans[cpu][i], speed);
285 pr_warning("wf_rm31: Fan %s reports error %d\n",
286 cpu_fans[cpu][i]->name, err);
287 failure_state |= FAILURE_FAN;
293 /* Implementation... */
294 static int cpu_setup_pid(int cpu)
296 struct wf_cpu_pid_param pid;
297 const struct mpu_data *mpu = cpu_mpu_data[cpu];
298 s32 tmax, ttarget, ptarget;
299 int fmin, fmax, hsize;
301 /* Get PID params from the appropriate MPU EEPROM */
302 tmax = mpu->tmax << 16;
303 ttarget = mpu->ttarget << 16;
304 ptarget = ((s32)(mpu->pmaxh - mpu->padjmax)) << 16;
306 DBG("wf_72: CPU%d ttarget = %d.%03d, tmax = %d.%03d\n",
307 cpu, FIX32TOPRINT(ttarget), FIX32TOPRINT(tmax));
309 /* We keep a global tmax for overtemp calculations */
310 if (tmax < cpu_all_tmax)
313 /* Set PID min/max by using the rear fan min/max */
314 fmin = wf_control_get_min(cpu_fans[cpu][0]);
315 fmax = wf_control_get_max(cpu_fans[cpu][0]);
316 DBG("wf_72: CPU%d max RPM range = [%d..%d]\n", cpu, fmin, fmax);
319 hsize = min_t(int, mpu->tguardband, WF_PID_MAX_HISTORY);
320 DBG("wf_72: CPU%d history size = %d\n", cpu, hsize);
322 /* Initialize PID loop */
323 pid.interval = 1; /* seconds */
324 pid.history_len = hsize;
325 pid.gd = mpu->pid_gd;
326 pid.gp = mpu->pid_gp;
327 pid.gr = mpu->pid_gr;
329 pid.ttarget = ttarget;
330 pid.pmaxadj = ptarget;
334 wf_cpu_pid_init(&cpu_pid[cpu], &pid);
335 cpu_pid[cpu].target = 4000;
340 /* Backside/U3 fan */
341 static struct wf_pid_param backside_param = {
353 /* DIMMs temperature (clamp the backside fan) */
354 static struct wf_pid_param dimms_param = {
366 static void backside_fan_tick(void)
369 int speed, dspeed, fan_min;
372 if (!backside_fan || !backside_temp || !dimms_temp || !backside_tick)
374 if (--backside_tick > 0)
376 backside_tick = backside_pid.param.interval;
378 DBG_LOTS("* backside fans tick\n");
380 /* Update fan speed from actual fans */
381 err = wf_control_get(backside_fan, &speed);
383 backside_pid.target = speed;
385 err = wf_sensor_get(backside_temp, &temp);
387 printk(KERN_WARNING "windfarm: U3 temp sensor error %d\n",
389 failure_state |= FAILURE_SENSOR;
390 wf_control_set_max(backside_fan);
393 speed = wf_pid_run(&backside_pid, temp);
395 DBG_LOTS("backside PID temp=%d.%.3d speed=%d\n",
396 FIX32TOPRINT(temp), speed);
398 err = wf_sensor_get(dimms_temp, &dtemp);
400 printk(KERN_WARNING "windfarm: DIMMs temp sensor error %d\n",
402 failure_state |= FAILURE_SENSOR;
403 wf_control_set_max(backside_fan);
406 dspeed = wf_pid_run(&dimms_pid, dtemp);
407 dimms_output_clamp = dspeed;
409 fan_min = (dspeed * 100) / 14000;
410 fan_min = max(fan_min, backside_param.min);
411 speed = max(speed, fan_min);
413 err = wf_control_set(backside_fan, speed);
415 printk(KERN_WARNING "windfarm: backside fan error %d\n", err);
416 failure_state |= FAILURE_FAN;
420 static void backside_setup_pid(void)
422 /* first time initialize things */
423 s32 fmin = wf_control_get_min(backside_fan);
424 s32 fmax = wf_control_get_max(backside_fan);
425 struct wf_pid_param param;
427 param = backside_param;
428 param.min = max(param.min, fmin);
429 param.max = min(param.max, fmax);
430 wf_pid_init(&backside_pid, ¶m);
433 wf_pid_init(&dimms_pid, ¶m);
437 pr_info("wf_rm31: Backside control loop started.\n");
441 static const struct wf_pid_param slots_param = {
453 static void slots_fan_tick(void)
459 if (!slots_fan || !slots_temp || !slots_tick)
461 if (--slots_tick > 0)
463 slots_tick = slots_pid.param.interval;
465 DBG_LOTS("* slots fans tick\n");
467 err = wf_sensor_get(slots_temp, &temp);
469 pr_warning("wf_rm31: slots temp sensor error %d\n", err);
470 failure_state |= FAILURE_SENSOR;
471 wf_control_set_max(slots_fan);
474 speed = wf_pid_run(&slots_pid, temp);
476 DBG_LOTS("slots PID temp=%d.%.3d speed=%d\n",
477 FIX32TOPRINT(temp), speed);
480 err = wf_control_set(slots_fan, speed);
482 printk(KERN_WARNING "windfarm: slots bay fan error %d\n", err);
483 failure_state |= FAILURE_FAN;
487 static void slots_setup_pid(void)
489 /* first time initialize things */
490 s32 fmin = wf_control_get_min(slots_fan);
491 s32 fmax = wf_control_get_max(slots_fan);
492 struct wf_pid_param param = slots_param;
494 param.min = max(param.min, fmin);
495 param.max = min(param.max, fmax);
496 wf_pid_init(&slots_pid, ¶m);
499 pr_info("wf_rm31: Slots control loop started.\n");
502 static void set_fail_state(void)
507 wf_control_set_max(backside_fan);
509 wf_control_set_max(slots_fan);
512 static void rm31_tick(void)
518 printk(KERN_INFO "windfarm: CPUs control loops started.\n");
519 for (i = 0; i < nr_chips; ++i) {
520 if (cpu_setup_pid(i) < 0) {
521 failure_state = FAILURE_PERM;
526 DBG_LOTS("cpu_all_tmax=%d.%03d\n", FIX32TOPRINT(cpu_all_tmax));
528 backside_setup_pid();
531 #ifdef HACKED_OVERTEMP
532 cpu_all_tmax = 60 << 16;
536 /* Permanent failure, bail out */
537 if (failure_state & FAILURE_PERM)
541 * Clear all failure bits except low overtemp which will be eventually
542 * cleared by the control loop itself
544 last_failure = failure_state;
545 failure_state &= FAILURE_LOW_OVERTEMP;
549 /* We do CPUs last because they can be clamped high by
554 DBG_LOTS(" last_failure: 0x%x, failure_state: %x\n",
555 last_failure, failure_state);
557 /* Check for failures. Any failure causes cpufreq clamping */
558 if (failure_state && last_failure == 0 && cpufreq_clamp)
559 wf_control_set_max(cpufreq_clamp);
560 if (failure_state == 0 && last_failure && cpufreq_clamp)
561 wf_control_set_min(cpufreq_clamp);
563 /* That's it for now, we might want to deal with other failures
564 * differently in the future though
568 static void rm31_new_control(struct wf_control *ct)
572 if (!strcmp(ct->name, "cpu-fan-a-0"))
574 else if (!strcmp(ct->name, "cpu-fan-b-0"))
576 else if (!strcmp(ct->name, "cpu-fan-c-0"))
578 else if (!strcmp(ct->name, "cpu-fan-a-1"))
580 else if (!strcmp(ct->name, "cpu-fan-b-1"))
582 else if (!strcmp(ct->name, "cpu-fan-c-1"))
584 else if (!strcmp(ct->name, "backside-fan"))
586 else if (!strcmp(ct->name, "slots-fan"))
588 else if (!strcmp(ct->name, "cpufreq-clamp"))
602 have_all_controls = all_controls;
606 static void rm31_new_sensor(struct wf_sensor *sr)
610 if (!strcmp(sr->name, "cpu-diode-temp-0"))
611 sens_cpu_temp[0] = sr;
612 else if (!strcmp(sr->name, "cpu-diode-temp-1"))
613 sens_cpu_temp[1] = sr;
614 else if (!strcmp(sr->name, "cpu-voltage-0"))
615 sens_cpu_volts[0] = sr;
616 else if (!strcmp(sr->name, "cpu-voltage-1"))
617 sens_cpu_volts[1] = sr;
618 else if (!strcmp(sr->name, "cpu-current-0"))
619 sens_cpu_amps[0] = sr;
620 else if (!strcmp(sr->name, "cpu-current-1"))
621 sens_cpu_amps[1] = sr;
622 else if (!strcmp(sr->name, "backside-temp"))
624 else if (!strcmp(sr->name, "slots-temp"))
626 else if (!strcmp(sr->name, "dimms-temp"))
642 have_all_sensors = all_sensors;
645 static int rm31_wf_notify(struct notifier_block *self,
646 unsigned long event, void *data)
649 case WF_EVENT_NEW_SENSOR:
650 rm31_new_sensor(data);
652 case WF_EVENT_NEW_CONTROL:
653 rm31_new_control(data);
656 if (have_all_controls && have_all_sensors)
662 static struct notifier_block rm31_events = {
663 .notifier_call = rm31_wf_notify,
666 static int wf_rm31_probe(struct platform_device *dev)
668 wf_register_client(&rm31_events);
672 static int wf_rm31_remove(struct platform_device *dev)
674 wf_unregister_client(&rm31_events);
676 /* should release all sensors and controls */
680 static struct platform_driver wf_rm31_driver = {
681 .probe = wf_rm31_probe,
682 .remove = wf_rm31_remove,
685 .owner = THIS_MODULE,
689 static int __init wf_rm31_init(void)
691 struct device_node *cpu;
694 if (!of_machine_is_compatible("RackMac3,1"))
697 /* Count the number of CPU cores */
699 for_each_node_by_type(cpu, "cpu")
701 if (nr_chips > NR_CHIPS)
704 pr_info("windfarm: Initializing for desktop G5 with %d chips\n",
707 /* Get MPU data for each CPU */
708 for (i = 0; i < nr_chips; i++) {
709 cpu_mpu_data[i] = wf_get_mpu(i);
710 if (!cpu_mpu_data[i]) {
711 pr_err("wf_rm31: Failed to find MPU data for CPU %d\n", i);
717 request_module("windfarm_fcu_controls");
718 request_module("windfarm_lm75_sensor");
719 request_module("windfarm_lm87_sensor");
720 request_module("windfarm_ad7417_sensor");
721 request_module("windfarm_max6690_sensor");
722 request_module("windfarm_cpufreq_clamp");
725 platform_driver_register(&wf_rm31_driver);
729 static void __exit wf_rm31_exit(void)
731 platform_driver_unregister(&wf_rm31_driver);
734 module_init(wf_rm31_init);
735 module_exit(wf_rm31_exit);
737 MODULE_AUTHOR("Benjamin Herrenschmidt <benh@kernel.crashing.org>");
738 MODULE_DESCRIPTION("Thermal control for Xserve G5");
739 MODULE_LICENSE("GPL");
740 MODULE_ALIAS("platform:windfarm");