2 * Device driver for the thermostats & fan controller of the
3 * Apple G5 "PowerMac7,2" desktop machines.
5 * (c) Copyright IBM Corp. 2003-2004
7 * Maintained by: Benjamin Herrenschmidt
8 * <benh@kernel.crashing.org>
11 * The algorithm used is the PID control algorithm, used the same
12 * way the published Darwin code does, using the same values that
13 * are present in the Darwin 7.0 snapshot property lists.
15 * As far as the CPUs control loops are concerned, I use the
16 * calibration & PID constants provided by the EEPROM,
17 * I do _not_ embed any value from the property lists, as the ones
18 * provided by Darwin 7.0 seem to always have an older version that
19 * what I've seen on the actual computers.
20 * It would be interesting to verify that though. Darwin has a
21 * version code of 1.0.0d11 for all control loops it seems, while
22 * so far, the machines EEPROMs contain a dataset versioned 1.0.0f
24 * Darwin doesn't provide source to all parts, some missing
25 * bits like the AppleFCU driver or the actual scale of some
26 * of the values returned by sensors had to be "guessed" some
27 * way... or based on what Open Firmware does.
29 * I didn't yet figure out how to get the slots power consumption
30 * out of the FCU, so that part has not been implemented yet and
31 * the slots fan is set to a fixed 50% PWM, hoping this value is
34 * Note: I have observed strange oscillations of the CPU control
35 * loop on a dual G5 here. When idle, the CPU exhaust fan tend to
36 * oscillates slowly (over several minutes) between the minimum
37 * of 300RPMs and approx. 1000 RPMs. I don't know what is causing
38 * this, it could be some incorrect constant or an error in the
39 * way I ported the algorithm, or it could be just normal. I
40 * don't have full understanding on the way Apple tweaked the PID
41 * algorithm for the CPU control, it is definitely not a standard
44 * TODO: - Check MPU structure version/signature
45 * - Add things like /sbin/overtemp for non-critical
46 * overtemp conditions so userland can take some policy
47 * decisions, like slewing down CPUs
48 * - Deal with fan and i2c failures in a better way
49 * - Maybe do a generic PID based on params used for
50 * U3 and Drives ? Definitely need to factor code a bit
51 * bettter... also make sensor detection more robust using
52 * the device-tree to probe for them
53 * - Figure out how to get the slots consumption and set the
54 * slots fan accordingly
62 * - Read fan speed from FCU, low level fan routines now deal
63 * with errors & check fan status, though higher level don't
65 * - Move a bunch of definitions to .h file
68 * - Fix build on ppc64 kernel
69 * - Move back statics definitions to .c file
70 * - Avoid calling schedule_timeout with a negative number
73 * - Fix typo when reading back fan speed on 2 CPU machines
76 * - Rework code accessing the ADC chips, make it more robust and
77 * closer to the chip spec. Also make sure it is configured properly,
78 * I've seen yet unexplained cases where on startup, I would have stale
79 * values in the configuration register
80 * - Switch back to use of target fan speed for PID, thus lowering
84 * - Add device-tree lookup for fan IDs, should detect liquid cooling
86 * - Enable driver for PowerMac7,3 machines
87 * - Split the U3/Backside cooling on U3 & U3H versions as Darwin does
88 * - Add new CPU cooling algorithm for machines with liquid cooling
89 * - Workaround for some PowerMac7,3 with empty "fan" node in the devtree
90 * - Fix a signed/unsigned compare issue in some PID loops
93 * - Add basic support for Xserve G5
94 * - Retreive pumps min/max from EEPROM image in device-tree (broken)
95 * - Use min/max macros here or there
96 * - Latest darwin updated U3H min fan speed to 20% PWM
98 * July. 06, 2006 : 1.3
99 * - Fix setting of RPM fans on Xserve G5 (they were going too fast)
100 * - Add missing slots fan control loop for Xserve G5
101 * - Lower fixed slots fan speed from 50% to 40% on desktop G5s. We
102 * still can't properly implement the control loop for these, so let's
103 * reduce the noise a little bit, it appears that 40% still gives us
104 * a pretty good air flow
105 * - Add code to "tickle" the FCU regulary so it doesn't think that
106 * we are gone while in fact, the machine just didn't need any fan
107 * speed change lately
111 #include <linux/types.h>
112 #include <linux/module.h>
113 #include <linux/errno.h>
114 #include <linux/kernel.h>
115 #include <linux/delay.h>
116 #include <linux/sched.h>
117 #include <linux/slab.h>
118 #include <linux/init.h>
119 #include <linux/spinlock.h>
120 #include <linux/smp_lock.h>
121 #include <linux/wait.h>
122 #include <linux/reboot.h>
123 #include <linux/kmod.h>
124 #include <linux/i2c.h>
125 #include <asm/prom.h>
126 #include <asm/machdep.h>
128 #include <asm/system.h>
129 #include <asm/sections.h>
130 #include <asm/of_device.h>
131 #include <asm/macio.h>
133 #include "therm_pm72.h"
135 #define VERSION "1.3"
140 #define DBG(args...) printk(args)
142 #define DBG(args...) do { } while(0)
150 static struct of_device * of_dev;
151 static struct i2c_adapter * u3_0;
152 static struct i2c_adapter * u3_1;
153 static struct i2c_adapter * k2;
154 static struct i2c_client * fcu;
155 static struct cpu_pid_state cpu_state[2];
156 static struct basckside_pid_params backside_params;
157 static struct backside_pid_state backside_state;
158 static struct drives_pid_state drives_state;
159 static struct dimm_pid_state dimms_state;
160 static struct slots_pid_state slots_state;
162 static int cpu_count;
163 static int cpu_pid_type;
164 static pid_t ctrl_task;
165 static struct completion ctrl_complete;
166 static int critical_state;
168 static s32 dimm_output_clamp;
169 static int fcu_rpm_shift;
170 static int fcu_tickle_ticks;
171 static DECLARE_MUTEX(driver_lock);
174 * We have 3 types of CPU PID control. One is "split" old style control
175 * for intake & exhaust fans, the other is "combined" control for both
176 * CPUs that also deals with the pumps when present. To be "compatible"
177 * with OS X at this point, we only use "COMBINED" on the machines that
178 * are identified as having the pumps (though that identification is at
179 * least dodgy). Ultimately, we could probably switch completely to this
180 * algorithm provided we hack it to deal with the UP case
182 #define CPU_PID_TYPE_SPLIT 0
183 #define CPU_PID_TYPE_COMBINED 1
184 #define CPU_PID_TYPE_RACKMAC 2
187 * This table describes all fans in the FCU. The "id" and "type" values
188 * are defaults valid for all earlier machines. Newer machines will
189 * eventually override the table content based on the device-tree
193 char* loc; /* location code */
194 int type; /* 0 = rpm, 1 = pwm, 2 = pump */
195 int id; /* id or -1 */
198 #define FCU_FAN_RPM 0
199 #define FCU_FAN_PWM 1
201 #define FCU_FAN_ABSENT_ID -1
203 #define FCU_FAN_COUNT ARRAY_SIZE(fcu_fans)
205 struct fcu_fan_table fcu_fans[] = {
206 [BACKSIDE_FAN_PWM_INDEX] = {
207 .loc = "BACKSIDE,SYS CTRLR FAN",
209 .id = BACKSIDE_FAN_PWM_DEFAULT_ID,
211 [DRIVES_FAN_RPM_INDEX] = {
214 .id = DRIVES_FAN_RPM_DEFAULT_ID,
216 [SLOTS_FAN_PWM_INDEX] = {
217 .loc = "SLOT,PCI FAN",
219 .id = SLOTS_FAN_PWM_DEFAULT_ID,
221 [CPUA_INTAKE_FAN_RPM_INDEX] = {
222 .loc = "CPU A INTAKE",
224 .id = CPUA_INTAKE_FAN_RPM_DEFAULT_ID,
226 [CPUA_EXHAUST_FAN_RPM_INDEX] = {
227 .loc = "CPU A EXHAUST",
229 .id = CPUA_EXHAUST_FAN_RPM_DEFAULT_ID,
231 [CPUB_INTAKE_FAN_RPM_INDEX] = {
232 .loc = "CPU B INTAKE",
234 .id = CPUB_INTAKE_FAN_RPM_DEFAULT_ID,
236 [CPUB_EXHAUST_FAN_RPM_INDEX] = {
237 .loc = "CPU B EXHAUST",
239 .id = CPUB_EXHAUST_FAN_RPM_DEFAULT_ID,
241 /* pumps aren't present by default, have to be looked up in the
244 [CPUA_PUMP_RPM_INDEX] = {
247 .id = FCU_FAN_ABSENT_ID,
249 [CPUB_PUMP_RPM_INDEX] = {
252 .id = FCU_FAN_ABSENT_ID,
255 [CPU_A1_FAN_RPM_INDEX] = {
258 .id = FCU_FAN_ABSENT_ID,
260 [CPU_A2_FAN_RPM_INDEX] = {
263 .id = FCU_FAN_ABSENT_ID,
265 [CPU_A3_FAN_RPM_INDEX] = {
268 .id = FCU_FAN_ABSENT_ID,
270 [CPU_B1_FAN_RPM_INDEX] = {
273 .id = FCU_FAN_ABSENT_ID,
275 [CPU_B2_FAN_RPM_INDEX] = {
278 .id = FCU_FAN_ABSENT_ID,
280 [CPU_B3_FAN_RPM_INDEX] = {
283 .id = FCU_FAN_ABSENT_ID,
288 * i2c_driver structure to attach to the host i2c controller
291 static int therm_pm72_attach(struct i2c_adapter *adapter);
292 static int therm_pm72_detach(struct i2c_adapter *adapter);
294 static struct i2c_driver therm_pm72_driver =
297 .name = "therm_pm72",
299 .attach_adapter = therm_pm72_attach,
300 .detach_adapter = therm_pm72_detach,
304 * Utility function to create an i2c_client structure and
305 * attach it to one of u3 adapters
307 static struct i2c_client *attach_i2c_chip(int id, const char *name)
309 struct i2c_client *clt;
310 struct i2c_adapter *adap;
321 clt = kmalloc(sizeof(struct i2c_client), GFP_KERNEL);
324 memset(clt, 0, sizeof(struct i2c_client));
326 clt->addr = (id >> 1) & 0x7f;
328 clt->driver = &therm_pm72_driver;
329 strncpy(clt->name, name, I2C_NAME_SIZE-1);
331 if (i2c_attach_client(clt)) {
332 printk(KERN_ERR "therm_pm72: Failed to attach to i2c ID 0x%x\n", id);
340 * Utility function to get rid of the i2c_client structure
341 * (will also detach from the adapter hopepfully)
343 static void detach_i2c_chip(struct i2c_client *clt)
345 i2c_detach_client(clt);
350 * Here are the i2c chip access wrappers
353 static void initialize_adc(struct cpu_pid_state *state)
358 /* Read ADC the configuration register and cache it. We
359 * also make sure Config2 contains proper values, I've seen
360 * cases where we got stale grabage in there, thus preventing
361 * proper reading of conv. values
367 i2c_master_send(state->monitor, buf, 2);
369 /* Read & cache Config1 */
371 rc = i2c_master_send(state->monitor, buf, 1);
373 rc = i2c_master_recv(state->monitor, buf, 1);
375 state->adc_config = buf[0];
376 DBG("ADC config reg: %02x\n", state->adc_config);
377 /* Disable shutdown mode */
378 state->adc_config &= 0xfe;
380 buf[1] = state->adc_config;
381 rc = i2c_master_send(state->monitor, buf, 2);
385 printk(KERN_ERR "therm_pm72: Error reading ADC config"
389 static int read_smon_adc(struct cpu_pid_state *state, int chan)
391 int rc, data, tries = 0;
397 buf[1] = (state->adc_config & 0x1f) | (chan << 5);
398 rc = i2c_master_send(state->monitor, buf, 2);
401 /* Wait for convertion */
403 /* Switch to data register */
405 rc = i2c_master_send(state->monitor, buf, 1);
409 rc = i2c_master_recv(state->monitor, buf, 2);
412 data = ((u16)buf[0]) << 8 | (u16)buf[1];
415 DBG("Error reading ADC, retrying...\n");
417 printk(KERN_ERR "therm_pm72: Error reading ADC !\n");
424 static int read_lm87_reg(struct i2c_client * chip, int reg)
432 rc = i2c_master_send(chip, &buf, 1);
435 rc = i2c_master_recv(chip, &buf, 1);
440 DBG("Error reading LM87, retrying...\n");
442 printk(KERN_ERR "therm_pm72: Error reading LM87 !\n");
449 static int fan_read_reg(int reg, unsigned char *buf, int nb)
456 nw = i2c_master_send(fcu, buf, 1);
457 if (nw > 0 || (nw < 0 && nw != -EIO) || tries >= 100)
463 printk(KERN_ERR "Failure writing address to FCU: %d", nw);
468 nr = i2c_master_recv(fcu, buf, nb);
469 if (nr > 0 || (nr < 0 && nr != ENODEV) || tries >= 100)
475 printk(KERN_ERR "Failure reading data from FCU: %d", nw);
479 static int fan_write_reg(int reg, const unsigned char *ptr, int nb)
482 unsigned char buf[16];
485 memcpy(buf+1, ptr, nb);
489 nw = i2c_master_send(fcu, buf, nb);
490 if (nw > 0 || (nw < 0 && nw != EIO) || tries >= 100)
496 printk(KERN_ERR "Failure writing to FCU: %d", nw);
500 static int start_fcu(void)
502 unsigned char buf = 0xff;
505 rc = fan_write_reg(0xe, &buf, 1);
508 rc = fan_write_reg(0x2e, &buf, 1);
511 rc = fan_read_reg(0, &buf, 1);
514 fcu_rpm_shift = (buf == 1) ? 2 : 3;
515 printk(KERN_DEBUG "FCU Initialized, RPM fan shift is %d\n",
521 static int set_rpm_fan(int fan_index, int rpm)
523 unsigned char buf[2];
524 int rc, id, min, max;
526 if (fcu_fans[fan_index].type != FCU_FAN_RPM)
528 id = fcu_fans[fan_index].id;
529 if (id == FCU_FAN_ABSENT_ID)
532 min = 2400 >> fcu_rpm_shift;
533 max = 56000 >> fcu_rpm_shift;
539 buf[0] = rpm >> (8 - fcu_rpm_shift);
540 buf[1] = rpm << fcu_rpm_shift;
541 rc = fan_write_reg(0x10 + (id * 2), buf, 2);
547 static int get_rpm_fan(int fan_index, int programmed)
549 unsigned char failure;
550 unsigned char active;
551 unsigned char buf[2];
552 int rc, id, reg_base;
554 if (fcu_fans[fan_index].type != FCU_FAN_RPM)
556 id = fcu_fans[fan_index].id;
557 if (id == FCU_FAN_ABSENT_ID)
560 rc = fan_read_reg(0xb, &failure, 1);
563 if ((failure & (1 << id)) != 0)
565 rc = fan_read_reg(0xd, &active, 1);
568 if ((active & (1 << id)) == 0)
571 /* Programmed value or real current speed */
572 reg_base = programmed ? 0x10 : 0x11;
573 rc = fan_read_reg(reg_base + (id * 2), buf, 2);
577 return (buf[0] << (8 - fcu_rpm_shift)) | buf[1] >> fcu_rpm_shift;
580 static int set_pwm_fan(int fan_index, int pwm)
582 unsigned char buf[2];
585 if (fcu_fans[fan_index].type != FCU_FAN_PWM)
587 id = fcu_fans[fan_index].id;
588 if (id == FCU_FAN_ABSENT_ID)
595 pwm = (pwm * 2559) / 1000;
597 rc = fan_write_reg(0x30 + (id * 2), buf, 1);
603 static int get_pwm_fan(int fan_index)
605 unsigned char failure;
606 unsigned char active;
607 unsigned char buf[2];
610 if (fcu_fans[fan_index].type != FCU_FAN_PWM)
612 id = fcu_fans[fan_index].id;
613 if (id == FCU_FAN_ABSENT_ID)
616 rc = fan_read_reg(0x2b, &failure, 1);
619 if ((failure & (1 << id)) != 0)
621 rc = fan_read_reg(0x2d, &active, 1);
624 if ((active & (1 << id)) == 0)
627 /* Programmed value or real current speed */
628 rc = fan_read_reg(0x30 + (id * 2), buf, 1);
632 return (buf[0] * 1000) / 2559;
635 static void tickle_fcu(void)
639 pwm = get_pwm_fan(SLOTS_FAN_PWM_INDEX);
641 DBG("FCU Tickle, slots fan is: %d\n", pwm);
646 pwm = SLOTS_FAN_DEFAULT_PWM;
647 } else if (pwm < SLOTS_PID_OUTPUT_MIN)
648 pwm = SLOTS_PID_OUTPUT_MIN;
650 /* That is hopefully enough to make the FCU happy */
651 set_pwm_fan(SLOTS_FAN_PWM_INDEX, pwm);
656 * Utility routine to read the CPU calibration EEPROM data
657 * from the device-tree
659 static int read_eeprom(int cpu, struct mpu_data *out)
661 struct device_node *np;
666 /* prom.c routine for finding a node by path is a bit brain dead
667 * and requires exact @xxx unit numbers. This is a bit ugly but
668 * will work for these machines
670 sprintf(nodename, "/u3@0,f8000000/i2c@f8001000/cpuid@a%d", cpu ? 2 : 0);
671 np = of_find_node_by_path(nodename);
673 printk(KERN_ERR "therm_pm72: Failed to retrieve cpuid node from device-tree\n");
676 data = get_property(np, "cpuid", &len);
678 printk(KERN_ERR "therm_pm72: Failed to retrieve cpuid property from device-tree\n");
682 memcpy(out, data, sizeof(struct mpu_data));
688 static void fetch_cpu_pumps_minmax(void)
690 struct cpu_pid_state *state0 = &cpu_state[0];
691 struct cpu_pid_state *state1 = &cpu_state[1];
692 u16 pump_min = 0, pump_max = 0xffff;
695 /* Try to fetch pumps min/max infos from eeprom */
697 memcpy(&tmp, &state0->mpu.processor_part_num, 8);
698 if (tmp[0] != 0xffff && tmp[1] != 0xffff) {
699 pump_min = max(pump_min, tmp[0]);
700 pump_max = min(pump_max, tmp[1]);
702 if (tmp[2] != 0xffff && tmp[3] != 0xffff) {
703 pump_min = max(pump_min, tmp[2]);
704 pump_max = min(pump_max, tmp[3]);
707 /* Double check the values, this _IS_ needed as the EEPROM on
708 * some dual 2.5Ghz G5s seem, at least, to have both min & max
709 * same to the same value ... (grrrr)
711 if (pump_min == pump_max || pump_min == 0 || pump_max == 0xffff) {
712 pump_min = CPU_PUMP_OUTPUT_MIN;
713 pump_max = CPU_PUMP_OUTPUT_MAX;
716 state0->pump_min = state1->pump_min = pump_min;
717 state0->pump_max = state1->pump_max = pump_max;
721 * Now, unfortunately, sysfs doesn't give us a nice void * we could
722 * pass around to the attribute functions, so we don't really have
723 * choice but implement a bunch of them...
725 * That sucks a bit, we take the lock because FIX32TOPRINT evaluates
726 * the input twice... I accept patches :)
728 #define BUILD_SHOW_FUNC_FIX(name, data) \
729 static ssize_t show_##name(struct device *dev, struct device_attribute *attr, char *buf) \
732 down(&driver_lock); \
733 r = sprintf(buf, "%d.%03d", FIX32TOPRINT(data)); \
737 #define BUILD_SHOW_FUNC_INT(name, data) \
738 static ssize_t show_##name(struct device *dev, struct device_attribute *attr, char *buf) \
740 return sprintf(buf, "%d", data); \
743 BUILD_SHOW_FUNC_FIX(cpu0_temperature, cpu_state[0].last_temp)
744 BUILD_SHOW_FUNC_FIX(cpu0_voltage, cpu_state[0].voltage)
745 BUILD_SHOW_FUNC_FIX(cpu0_current, cpu_state[0].current_a)
746 BUILD_SHOW_FUNC_INT(cpu0_exhaust_fan_rpm, cpu_state[0].rpm)
747 BUILD_SHOW_FUNC_INT(cpu0_intake_fan_rpm, cpu_state[0].intake_rpm)
749 BUILD_SHOW_FUNC_FIX(cpu1_temperature, cpu_state[1].last_temp)
750 BUILD_SHOW_FUNC_FIX(cpu1_voltage, cpu_state[1].voltage)
751 BUILD_SHOW_FUNC_FIX(cpu1_current, cpu_state[1].current_a)
752 BUILD_SHOW_FUNC_INT(cpu1_exhaust_fan_rpm, cpu_state[1].rpm)
753 BUILD_SHOW_FUNC_INT(cpu1_intake_fan_rpm, cpu_state[1].intake_rpm)
755 BUILD_SHOW_FUNC_FIX(backside_temperature, backside_state.last_temp)
756 BUILD_SHOW_FUNC_INT(backside_fan_pwm, backside_state.pwm)
758 BUILD_SHOW_FUNC_FIX(drives_temperature, drives_state.last_temp)
759 BUILD_SHOW_FUNC_INT(drives_fan_rpm, drives_state.rpm)
761 BUILD_SHOW_FUNC_FIX(slots_temperature, slots_state.last_temp)
762 BUILD_SHOW_FUNC_INT(slots_fan_pwm, slots_state.pwm)
764 BUILD_SHOW_FUNC_FIX(dimms_temperature, dimms_state.last_temp)
766 static DEVICE_ATTR(cpu0_temperature,S_IRUGO,show_cpu0_temperature,NULL);
767 static DEVICE_ATTR(cpu0_voltage,S_IRUGO,show_cpu0_voltage,NULL);
768 static DEVICE_ATTR(cpu0_current,S_IRUGO,show_cpu0_current,NULL);
769 static DEVICE_ATTR(cpu0_exhaust_fan_rpm,S_IRUGO,show_cpu0_exhaust_fan_rpm,NULL);
770 static DEVICE_ATTR(cpu0_intake_fan_rpm,S_IRUGO,show_cpu0_intake_fan_rpm,NULL);
772 static DEVICE_ATTR(cpu1_temperature,S_IRUGO,show_cpu1_temperature,NULL);
773 static DEVICE_ATTR(cpu1_voltage,S_IRUGO,show_cpu1_voltage,NULL);
774 static DEVICE_ATTR(cpu1_current,S_IRUGO,show_cpu1_current,NULL);
775 static DEVICE_ATTR(cpu1_exhaust_fan_rpm,S_IRUGO,show_cpu1_exhaust_fan_rpm,NULL);
776 static DEVICE_ATTR(cpu1_intake_fan_rpm,S_IRUGO,show_cpu1_intake_fan_rpm,NULL);
778 static DEVICE_ATTR(backside_temperature,S_IRUGO,show_backside_temperature,NULL);
779 static DEVICE_ATTR(backside_fan_pwm,S_IRUGO,show_backside_fan_pwm,NULL);
781 static DEVICE_ATTR(drives_temperature,S_IRUGO,show_drives_temperature,NULL);
782 static DEVICE_ATTR(drives_fan_rpm,S_IRUGO,show_drives_fan_rpm,NULL);
784 static DEVICE_ATTR(slots_temperature,S_IRUGO,show_slots_temperature,NULL);
785 static DEVICE_ATTR(slots_fan_pwm,S_IRUGO,show_slots_fan_pwm,NULL);
787 static DEVICE_ATTR(dimms_temperature,S_IRUGO,show_dimms_temperature,NULL);
790 * CPUs fans control loop
793 static int do_read_one_cpu_values(struct cpu_pid_state *state, s32 *temp, s32 *power)
795 s32 ltemp, volts, amps;
798 /* Default (in case of error) */
799 *temp = state->cur_temp;
800 *power = state->cur_power;
802 if (cpu_pid_type == CPU_PID_TYPE_RACKMAC)
803 index = (state->index == 0) ?
804 CPU_A1_FAN_RPM_INDEX : CPU_B1_FAN_RPM_INDEX;
806 index = (state->index == 0) ?
807 CPUA_EXHAUST_FAN_RPM_INDEX : CPUB_EXHAUST_FAN_RPM_INDEX;
809 /* Read current fan status */
810 rc = get_rpm_fan(index, !RPM_PID_USE_ACTUAL_SPEED);
812 /* XXX What do we do now ? Nothing for now, keep old value, but
813 * return error upstream
815 DBG(" cpu %d, fan reading error !\n", state->index);
818 DBG(" cpu %d, exhaust RPM: %d\n", state->index, state->rpm);
821 /* Get some sensor readings and scale it */
822 ltemp = read_smon_adc(state, 1);
824 /* XXX What do we do now ? */
828 DBG(" cpu %d, temp reading error !\n", state->index);
830 /* Fixup temperature according to diode calibration
832 DBG(" cpu %d, temp raw: %04x, m_diode: %04x, b_diode: %04x\n",
834 ltemp, state->mpu.mdiode, state->mpu.bdiode);
835 *temp = ((s32)ltemp * (s32)state->mpu.mdiode + ((s32)state->mpu.bdiode << 12)) >> 2;
836 state->last_temp = *temp;
837 DBG(" temp: %d.%03d\n", FIX32TOPRINT((*temp)));
841 * Read voltage & current and calculate power
843 volts = read_smon_adc(state, 3);
844 amps = read_smon_adc(state, 4);
846 /* Scale voltage and current raw sensor values according to fixed scales
847 * obtained in Darwin and calculate power from I and V
849 volts *= ADC_CPU_VOLTAGE_SCALE;
850 amps *= ADC_CPU_CURRENT_SCALE;
851 *power = (((u64)volts) * ((u64)amps)) >> 16;
852 state->voltage = volts;
853 state->current_a = amps;
854 state->last_power = *power;
856 DBG(" cpu %d, current: %d.%03d, voltage: %d.%03d, power: %d.%03d W\n",
857 state->index, FIX32TOPRINT(state->current_a),
858 FIX32TOPRINT(state->voltage), FIX32TOPRINT(*power));
863 static void do_cpu_pid(struct cpu_pid_state *state, s32 temp, s32 power)
865 s32 power_target, integral, derivative, proportional, adj_in_target, sval;
866 s64 integ_p, deriv_p, prop_p, sum;
869 /* Calculate power target value (could be done once for all)
870 * and convert to a 16.16 fp number
872 power_target = ((u32)(state->mpu.pmaxh - state->mpu.padjmax)) << 16;
873 DBG(" power target: %d.%03d, error: %d.%03d\n",
874 FIX32TOPRINT(power_target), FIX32TOPRINT(power_target - power));
876 /* Store temperature and power in history array */
877 state->cur_temp = (state->cur_temp + 1) % CPU_TEMP_HISTORY_SIZE;
878 state->temp_history[state->cur_temp] = temp;
879 state->cur_power = (state->cur_power + 1) % state->count_power;
880 state->power_history[state->cur_power] = power;
881 state->error_history[state->cur_power] = power_target - power;
883 /* If first loop, fill the history table */
885 for (i = 0; i < (state->count_power - 1); i++) {
886 state->cur_power = (state->cur_power + 1) % state->count_power;
887 state->power_history[state->cur_power] = power;
888 state->error_history[state->cur_power] = power_target - power;
890 for (i = 0; i < (CPU_TEMP_HISTORY_SIZE - 1); i++) {
891 state->cur_temp = (state->cur_temp + 1) % CPU_TEMP_HISTORY_SIZE;
892 state->temp_history[state->cur_temp] = temp;
897 /* Calculate the integral term normally based on the "power" values */
900 for (i = 0; i < state->count_power; i++)
901 integral += state->error_history[i];
902 integral *= CPU_PID_INTERVAL;
903 DBG(" integral: %08x\n", integral);
905 /* Calculate the adjusted input (sense value).
908 * so the result is 28.36
910 * input target is mpu.ttarget, input max is mpu.tmax
912 integ_p = ((s64)state->mpu.pid_gr) * (s64)integral;
913 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
914 sval = (state->mpu.tmax << 16) - ((integ_p >> 20) & 0xffffffff);
915 adj_in_target = (state->mpu.ttarget << 16);
916 if (adj_in_target > sval)
917 adj_in_target = sval;
918 DBG(" adj_in_target: %d.%03d, ttarget: %d\n", FIX32TOPRINT(adj_in_target),
921 /* Calculate the derivative term */
922 derivative = state->temp_history[state->cur_temp] -
923 state->temp_history[(state->cur_temp + CPU_TEMP_HISTORY_SIZE - 1)
924 % CPU_TEMP_HISTORY_SIZE];
925 derivative /= CPU_PID_INTERVAL;
926 deriv_p = ((s64)state->mpu.pid_gd) * (s64)derivative;
927 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
930 /* Calculate the proportional term */
931 proportional = temp - adj_in_target;
932 prop_p = ((s64)state->mpu.pid_gp) * (s64)proportional;
933 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
939 DBG(" sum: %d\n", (int)sum);
940 state->rpm += (s32)sum;
943 static void do_monitor_cpu_combined(void)
945 struct cpu_pid_state *state0 = &cpu_state[0];
946 struct cpu_pid_state *state1 = &cpu_state[1];
947 s32 temp0, power0, temp1, power1;
948 s32 temp_combi, power_combi;
949 int rc, intake, pump;
951 rc = do_read_one_cpu_values(state0, &temp0, &power0);
953 /* XXX What do we do now ? */
955 state1->overtemp = 0;
956 rc = do_read_one_cpu_values(state1, &temp1, &power1);
958 /* XXX What do we do now ? */
960 if (state1->overtemp)
963 temp_combi = max(temp0, temp1);
964 power_combi = max(power0, power1);
966 /* Check tmax, increment overtemp if we are there. At tmax+8, we go
967 * full blown immediately and try to trigger a shutdown
969 if (temp_combi >= ((state0->mpu.tmax + 8) << 16)) {
970 printk(KERN_WARNING "Warning ! Temperature way above maximum (%d) !\n",
972 state0->overtemp += CPU_MAX_OVERTEMP / 4;
973 } else if (temp_combi > (state0->mpu.tmax << 16))
976 state0->overtemp = 0;
977 if (state0->overtemp >= CPU_MAX_OVERTEMP)
979 if (state0->overtemp > 0) {
980 state0->rpm = state0->mpu.rmaxn_exhaust_fan;
981 state0->intake_rpm = intake = state0->mpu.rmaxn_intake_fan;
982 pump = state0->pump_max;
987 do_cpu_pid(state0, temp_combi, power_combi);
990 state0->rpm = max(state0->rpm, (int)state0->mpu.rminn_exhaust_fan);
991 state0->rpm = min(state0->rpm, (int)state0->mpu.rmaxn_exhaust_fan);
993 /* Calculate intake fan speed */
994 intake = (state0->rpm * CPU_INTAKE_SCALE) >> 16;
995 intake = max(intake, (int)state0->mpu.rminn_intake_fan);
996 intake = min(intake, (int)state0->mpu.rmaxn_intake_fan);
997 state0->intake_rpm = intake;
999 /* Calculate pump speed */
1000 pump = (state0->rpm * state0->pump_max) /
1001 state0->mpu.rmaxn_exhaust_fan;
1002 pump = min(pump, state0->pump_max);
1003 pump = max(pump, state0->pump_min);
1006 /* We copy values from state 0 to state 1 for /sysfs */
1007 state1->rpm = state0->rpm;
1008 state1->intake_rpm = state0->intake_rpm;
1010 DBG("** CPU %d RPM: %d Ex, %d, Pump: %d, In, overtemp: %d\n",
1011 state1->index, (int)state1->rpm, intake, pump, state1->overtemp);
1013 /* We should check for errors, shouldn't we ? But then, what
1014 * do we do once the error occurs ? For FCU notified fan
1015 * failures (-EFAULT) we probably want to notify userland
1018 set_rpm_fan(CPUA_INTAKE_FAN_RPM_INDEX, intake);
1019 set_rpm_fan(CPUA_EXHAUST_FAN_RPM_INDEX, state0->rpm);
1020 set_rpm_fan(CPUB_INTAKE_FAN_RPM_INDEX, intake);
1021 set_rpm_fan(CPUB_EXHAUST_FAN_RPM_INDEX, state0->rpm);
1023 if (fcu_fans[CPUA_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID)
1024 set_rpm_fan(CPUA_PUMP_RPM_INDEX, pump);
1025 if (fcu_fans[CPUB_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID)
1026 set_rpm_fan(CPUB_PUMP_RPM_INDEX, pump);
1029 static void do_monitor_cpu_split(struct cpu_pid_state *state)
1034 /* Read current fan status */
1035 rc = do_read_one_cpu_values(state, &temp, &power);
1037 /* XXX What do we do now ? */
1040 /* Check tmax, increment overtemp if we are there. At tmax+8, we go
1041 * full blown immediately and try to trigger a shutdown
1043 if (temp >= ((state->mpu.tmax + 8) << 16)) {
1044 printk(KERN_WARNING "Warning ! CPU %d temperature way above maximum"
1046 state->index, temp >> 16);
1047 state->overtemp += CPU_MAX_OVERTEMP / 4;
1048 } else if (temp > (state->mpu.tmax << 16))
1051 state->overtemp = 0;
1052 if (state->overtemp >= CPU_MAX_OVERTEMP)
1054 if (state->overtemp > 0) {
1055 state->rpm = state->mpu.rmaxn_exhaust_fan;
1056 state->intake_rpm = intake = state->mpu.rmaxn_intake_fan;
1061 do_cpu_pid(state, temp, power);
1064 state->rpm = max(state->rpm, (int)state->mpu.rminn_exhaust_fan);
1065 state->rpm = min(state->rpm, (int)state->mpu.rmaxn_exhaust_fan);
1067 /* Calculate intake fan */
1068 intake = (state->rpm * CPU_INTAKE_SCALE) >> 16;
1069 intake = max(intake, (int)state->mpu.rminn_intake_fan);
1070 intake = min(intake, (int)state->mpu.rmaxn_intake_fan);
1071 state->intake_rpm = intake;
1074 DBG("** CPU %d RPM: %d Ex, %d In, overtemp: %d\n",
1075 state->index, (int)state->rpm, intake, state->overtemp);
1077 /* We should check for errors, shouldn't we ? But then, what
1078 * do we do once the error occurs ? For FCU notified fan
1079 * failures (-EFAULT) we probably want to notify userland
1082 if (state->index == 0) {
1083 set_rpm_fan(CPUA_INTAKE_FAN_RPM_INDEX, intake);
1084 set_rpm_fan(CPUA_EXHAUST_FAN_RPM_INDEX, state->rpm);
1086 set_rpm_fan(CPUB_INTAKE_FAN_RPM_INDEX, intake);
1087 set_rpm_fan(CPUB_EXHAUST_FAN_RPM_INDEX, state->rpm);
1091 static void do_monitor_cpu_rack(struct cpu_pid_state *state)
1093 s32 temp, power, fan_min;
1096 /* Read current fan status */
1097 rc = do_read_one_cpu_values(state, &temp, &power);
1099 /* XXX What do we do now ? */
1102 /* Check tmax, increment overtemp if we are there. At tmax+8, we go
1103 * full blown immediately and try to trigger a shutdown
1105 if (temp >= ((state->mpu.tmax + 8) << 16)) {
1106 printk(KERN_WARNING "Warning ! CPU %d temperature way above maximum"
1108 state->index, temp >> 16);
1109 state->overtemp = CPU_MAX_OVERTEMP / 4;
1110 } else if (temp > (state->mpu.tmax << 16))
1113 state->overtemp = 0;
1114 if (state->overtemp >= CPU_MAX_OVERTEMP)
1116 if (state->overtemp > 0) {
1117 state->rpm = state->intake_rpm = state->mpu.rmaxn_intake_fan;
1122 do_cpu_pid(state, temp, power);
1124 /* Check clamp from dimms */
1125 fan_min = dimm_output_clamp;
1126 fan_min = max(fan_min, (int)state->mpu.rminn_intake_fan);
1128 DBG(" CPU min mpu = %d, min dimm = %d\n",
1129 state->mpu.rminn_intake_fan, dimm_output_clamp);
1131 state->rpm = max(state->rpm, (int)fan_min);
1132 state->rpm = min(state->rpm, (int)state->mpu.rmaxn_intake_fan);
1133 state->intake_rpm = state->rpm;
1136 DBG("** CPU %d RPM: %d overtemp: %d\n",
1137 state->index, (int)state->rpm, state->overtemp);
1139 /* We should check for errors, shouldn't we ? But then, what
1140 * do we do once the error occurs ? For FCU notified fan
1141 * failures (-EFAULT) we probably want to notify userland
1144 if (state->index == 0) {
1145 set_rpm_fan(CPU_A1_FAN_RPM_INDEX, state->rpm);
1146 set_rpm_fan(CPU_A2_FAN_RPM_INDEX, state->rpm);
1147 set_rpm_fan(CPU_A3_FAN_RPM_INDEX, state->rpm);
1149 set_rpm_fan(CPU_B1_FAN_RPM_INDEX, state->rpm);
1150 set_rpm_fan(CPU_B2_FAN_RPM_INDEX, state->rpm);
1151 set_rpm_fan(CPU_B3_FAN_RPM_INDEX, state->rpm);
1156 * Initialize the state structure for one CPU control loop
1158 static int init_cpu_state(struct cpu_pid_state *state, int index)
1160 state->index = index;
1162 state->rpm = (cpu_pid_type == CPU_PID_TYPE_RACKMAC) ? 4000 : 1000;
1163 state->overtemp = 0;
1164 state->adc_config = 0x00;
1168 state->monitor = attach_i2c_chip(SUPPLY_MONITOR_ID, "CPU0_monitor");
1169 else if (index == 1)
1170 state->monitor = attach_i2c_chip(SUPPLY_MONITORB_ID, "CPU1_monitor");
1171 if (state->monitor == NULL)
1174 if (read_eeprom(index, &state->mpu))
1177 state->count_power = state->mpu.tguardband;
1178 if (state->count_power > CPU_POWER_HISTORY_SIZE) {
1179 printk(KERN_WARNING "Warning ! too many power history slots\n");
1180 state->count_power = CPU_POWER_HISTORY_SIZE;
1182 DBG("CPU %d Using %d power history entries\n", index, state->count_power);
1185 device_create_file(&of_dev->dev, &dev_attr_cpu0_temperature);
1186 device_create_file(&of_dev->dev, &dev_attr_cpu0_voltage);
1187 device_create_file(&of_dev->dev, &dev_attr_cpu0_current);
1188 device_create_file(&of_dev->dev, &dev_attr_cpu0_exhaust_fan_rpm);
1189 device_create_file(&of_dev->dev, &dev_attr_cpu0_intake_fan_rpm);
1191 device_create_file(&of_dev->dev, &dev_attr_cpu1_temperature);
1192 device_create_file(&of_dev->dev, &dev_attr_cpu1_voltage);
1193 device_create_file(&of_dev->dev, &dev_attr_cpu1_current);
1194 device_create_file(&of_dev->dev, &dev_attr_cpu1_exhaust_fan_rpm);
1195 device_create_file(&of_dev->dev, &dev_attr_cpu1_intake_fan_rpm);
1201 detach_i2c_chip(state->monitor);
1202 state->monitor = NULL;
1208 * Dispose of the state data for one CPU control loop
1210 static void dispose_cpu_state(struct cpu_pid_state *state)
1212 if (state->monitor == NULL)
1215 if (state->index == 0) {
1216 device_remove_file(&of_dev->dev, &dev_attr_cpu0_temperature);
1217 device_remove_file(&of_dev->dev, &dev_attr_cpu0_voltage);
1218 device_remove_file(&of_dev->dev, &dev_attr_cpu0_current);
1219 device_remove_file(&of_dev->dev, &dev_attr_cpu0_exhaust_fan_rpm);
1220 device_remove_file(&of_dev->dev, &dev_attr_cpu0_intake_fan_rpm);
1222 device_remove_file(&of_dev->dev, &dev_attr_cpu1_temperature);
1223 device_remove_file(&of_dev->dev, &dev_attr_cpu1_voltage);
1224 device_remove_file(&of_dev->dev, &dev_attr_cpu1_current);
1225 device_remove_file(&of_dev->dev, &dev_attr_cpu1_exhaust_fan_rpm);
1226 device_remove_file(&of_dev->dev, &dev_attr_cpu1_intake_fan_rpm);
1229 detach_i2c_chip(state->monitor);
1230 state->monitor = NULL;
1234 * Motherboard backside & U3 heatsink fan control loop
1236 static void do_monitor_backside(struct backside_pid_state *state)
1238 s32 temp, integral, derivative, fan_min;
1239 s64 integ_p, deriv_p, prop_p, sum;
1242 if (--state->ticks != 0)
1244 state->ticks = backside_params.interval;
1248 /* Check fan status */
1249 rc = get_pwm_fan(BACKSIDE_FAN_PWM_INDEX);
1251 printk(KERN_WARNING "Error %d reading backside fan !\n", rc);
1252 /* XXX What do we do now ? */
1255 DBG(" current pwm: %d\n", state->pwm);
1257 /* Get some sensor readings */
1258 temp = i2c_smbus_read_byte_data(state->monitor, MAX6690_EXT_TEMP) << 16;
1259 state->last_temp = temp;
1260 DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1261 FIX32TOPRINT(backside_params.input_target));
1263 /* Store temperature and error in history array */
1264 state->cur_sample = (state->cur_sample + 1) % BACKSIDE_PID_HISTORY_SIZE;
1265 state->sample_history[state->cur_sample] = temp;
1266 state->error_history[state->cur_sample] = temp - backside_params.input_target;
1268 /* If first loop, fill the history table */
1270 for (i = 0; i < (BACKSIDE_PID_HISTORY_SIZE - 1); i++) {
1271 state->cur_sample = (state->cur_sample + 1) %
1272 BACKSIDE_PID_HISTORY_SIZE;
1273 state->sample_history[state->cur_sample] = temp;
1274 state->error_history[state->cur_sample] =
1275 temp - backside_params.input_target;
1280 /* Calculate the integral term */
1283 for (i = 0; i < BACKSIDE_PID_HISTORY_SIZE; i++)
1284 integral += state->error_history[i];
1285 integral *= backside_params.interval;
1286 DBG(" integral: %08x\n", integral);
1287 integ_p = ((s64)backside_params.G_r) * (s64)integral;
1288 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1291 /* Calculate the derivative term */
1292 derivative = state->error_history[state->cur_sample] -
1293 state->error_history[(state->cur_sample + BACKSIDE_PID_HISTORY_SIZE - 1)
1294 % BACKSIDE_PID_HISTORY_SIZE];
1295 derivative /= backside_params.interval;
1296 deriv_p = ((s64)backside_params.G_d) * (s64)derivative;
1297 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1300 /* Calculate the proportional term */
1301 prop_p = ((s64)backside_params.G_p) * (s64)(state->error_history[state->cur_sample]);
1302 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1308 DBG(" sum: %d\n", (int)sum);
1309 if (backside_params.additive)
1310 state->pwm += (s32)sum;
1314 /* Check for clamp */
1315 fan_min = (dimm_output_clamp * 100) / 14000;
1316 fan_min = max(fan_min, backside_params.output_min);
1318 state->pwm = max(state->pwm, fan_min);
1319 state->pwm = min(state->pwm, backside_params.output_max);
1321 DBG("** BACKSIDE PWM: %d\n", (int)state->pwm);
1322 set_pwm_fan(BACKSIDE_FAN_PWM_INDEX, state->pwm);
1326 * Initialize the state structure for the backside fan control loop
1328 static int init_backside_state(struct backside_pid_state *state)
1330 struct device_node *u3;
1331 int u3h = 1; /* conservative by default */
1334 * There are different PID params for machines with U3 and machines
1335 * with U3H, pick the right ones now
1337 u3 = of_find_node_by_path("/u3@0,f8000000");
1339 const u32 *vers = get_property(u3, "device-rev", NULL);
1341 if (((*vers) & 0x3f) < 0x34)
1347 backside_params.G_d = BACKSIDE_PID_RACK_G_d;
1348 backside_params.input_target = BACKSIDE_PID_RACK_INPUT_TARGET;
1349 backside_params.output_min = BACKSIDE_PID_U3H_OUTPUT_MIN;
1350 backside_params.interval = BACKSIDE_PID_RACK_INTERVAL;
1351 backside_params.G_p = BACKSIDE_PID_RACK_G_p;
1352 backside_params.G_r = BACKSIDE_PID_G_r;
1353 backside_params.output_max = BACKSIDE_PID_OUTPUT_MAX;
1354 backside_params.additive = 0;
1356 backside_params.G_d = BACKSIDE_PID_U3H_G_d;
1357 backside_params.input_target = BACKSIDE_PID_U3H_INPUT_TARGET;
1358 backside_params.output_min = BACKSIDE_PID_U3H_OUTPUT_MIN;
1359 backside_params.interval = BACKSIDE_PID_INTERVAL;
1360 backside_params.G_p = BACKSIDE_PID_G_p;
1361 backside_params.G_r = BACKSIDE_PID_G_r;
1362 backside_params.output_max = BACKSIDE_PID_OUTPUT_MAX;
1363 backside_params.additive = 1;
1365 backside_params.G_d = BACKSIDE_PID_U3_G_d;
1366 backside_params.input_target = BACKSIDE_PID_U3_INPUT_TARGET;
1367 backside_params.output_min = BACKSIDE_PID_U3_OUTPUT_MIN;
1368 backside_params.interval = BACKSIDE_PID_INTERVAL;
1369 backside_params.G_p = BACKSIDE_PID_G_p;
1370 backside_params.G_r = BACKSIDE_PID_G_r;
1371 backside_params.output_max = BACKSIDE_PID_OUTPUT_MAX;
1372 backside_params.additive = 1;
1379 state->monitor = attach_i2c_chip(BACKSIDE_MAX_ID, "backside_temp");
1380 if (state->monitor == NULL)
1383 device_create_file(&of_dev->dev, &dev_attr_backside_temperature);
1384 device_create_file(&of_dev->dev, &dev_attr_backside_fan_pwm);
1390 * Dispose of the state data for the backside control loop
1392 static void dispose_backside_state(struct backside_pid_state *state)
1394 if (state->monitor == NULL)
1397 device_remove_file(&of_dev->dev, &dev_attr_backside_temperature);
1398 device_remove_file(&of_dev->dev, &dev_attr_backside_fan_pwm);
1400 detach_i2c_chip(state->monitor);
1401 state->monitor = NULL;
1405 * Drives bay fan control loop
1407 static void do_monitor_drives(struct drives_pid_state *state)
1409 s32 temp, integral, derivative;
1410 s64 integ_p, deriv_p, prop_p, sum;
1413 if (--state->ticks != 0)
1415 state->ticks = DRIVES_PID_INTERVAL;
1419 /* Check fan status */
1420 rc = get_rpm_fan(DRIVES_FAN_RPM_INDEX, !RPM_PID_USE_ACTUAL_SPEED);
1422 printk(KERN_WARNING "Error %d reading drives fan !\n", rc);
1423 /* XXX What do we do now ? */
1426 DBG(" current rpm: %d\n", state->rpm);
1428 /* Get some sensor readings */
1429 temp = le16_to_cpu(i2c_smbus_read_word_data(state->monitor,
1431 state->last_temp = temp;
1432 DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1433 FIX32TOPRINT(DRIVES_PID_INPUT_TARGET));
1435 /* Store temperature and error in history array */
1436 state->cur_sample = (state->cur_sample + 1) % DRIVES_PID_HISTORY_SIZE;
1437 state->sample_history[state->cur_sample] = temp;
1438 state->error_history[state->cur_sample] = temp - DRIVES_PID_INPUT_TARGET;
1440 /* If first loop, fill the history table */
1442 for (i = 0; i < (DRIVES_PID_HISTORY_SIZE - 1); i++) {
1443 state->cur_sample = (state->cur_sample + 1) %
1444 DRIVES_PID_HISTORY_SIZE;
1445 state->sample_history[state->cur_sample] = temp;
1446 state->error_history[state->cur_sample] =
1447 temp - DRIVES_PID_INPUT_TARGET;
1452 /* Calculate the integral term */
1455 for (i = 0; i < DRIVES_PID_HISTORY_SIZE; i++)
1456 integral += state->error_history[i];
1457 integral *= DRIVES_PID_INTERVAL;
1458 DBG(" integral: %08x\n", integral);
1459 integ_p = ((s64)DRIVES_PID_G_r) * (s64)integral;
1460 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1463 /* Calculate the derivative term */
1464 derivative = state->error_history[state->cur_sample] -
1465 state->error_history[(state->cur_sample + DRIVES_PID_HISTORY_SIZE - 1)
1466 % DRIVES_PID_HISTORY_SIZE];
1467 derivative /= DRIVES_PID_INTERVAL;
1468 deriv_p = ((s64)DRIVES_PID_G_d) * (s64)derivative;
1469 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1472 /* Calculate the proportional term */
1473 prop_p = ((s64)DRIVES_PID_G_p) * (s64)(state->error_history[state->cur_sample]);
1474 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1480 DBG(" sum: %d\n", (int)sum);
1481 state->rpm += (s32)sum;
1483 state->rpm = max(state->rpm, DRIVES_PID_OUTPUT_MIN);
1484 state->rpm = min(state->rpm, DRIVES_PID_OUTPUT_MAX);
1486 DBG("** DRIVES RPM: %d\n", (int)state->rpm);
1487 set_rpm_fan(DRIVES_FAN_RPM_INDEX, state->rpm);
1491 * Initialize the state structure for the drives bay fan control loop
1493 static int init_drives_state(struct drives_pid_state *state)
1499 state->monitor = attach_i2c_chip(DRIVES_DALLAS_ID, "drives_temp");
1500 if (state->monitor == NULL)
1503 device_create_file(&of_dev->dev, &dev_attr_drives_temperature);
1504 device_create_file(&of_dev->dev, &dev_attr_drives_fan_rpm);
1510 * Dispose of the state data for the drives control loop
1512 static void dispose_drives_state(struct drives_pid_state *state)
1514 if (state->monitor == NULL)
1517 device_remove_file(&of_dev->dev, &dev_attr_drives_temperature);
1518 device_remove_file(&of_dev->dev, &dev_attr_drives_fan_rpm);
1520 detach_i2c_chip(state->monitor);
1521 state->monitor = NULL;
1525 * DIMMs temp control loop
1527 static void do_monitor_dimms(struct dimm_pid_state *state)
1529 s32 temp, integral, derivative, fan_min;
1530 s64 integ_p, deriv_p, prop_p, sum;
1533 if (--state->ticks != 0)
1535 state->ticks = DIMM_PID_INTERVAL;
1539 DBG(" current value: %d\n", state->output);
1541 temp = read_lm87_reg(state->monitor, LM87_INT_TEMP);
1545 state->last_temp = temp;
1546 DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1547 FIX32TOPRINT(DIMM_PID_INPUT_TARGET));
1549 /* Store temperature and error in history array */
1550 state->cur_sample = (state->cur_sample + 1) % DIMM_PID_HISTORY_SIZE;
1551 state->sample_history[state->cur_sample] = temp;
1552 state->error_history[state->cur_sample] = temp - DIMM_PID_INPUT_TARGET;
1554 /* If first loop, fill the history table */
1556 for (i = 0; i < (DIMM_PID_HISTORY_SIZE - 1); i++) {
1557 state->cur_sample = (state->cur_sample + 1) %
1558 DIMM_PID_HISTORY_SIZE;
1559 state->sample_history[state->cur_sample] = temp;
1560 state->error_history[state->cur_sample] =
1561 temp - DIMM_PID_INPUT_TARGET;
1566 /* Calculate the integral term */
1569 for (i = 0; i < DIMM_PID_HISTORY_SIZE; i++)
1570 integral += state->error_history[i];
1571 integral *= DIMM_PID_INTERVAL;
1572 DBG(" integral: %08x\n", integral);
1573 integ_p = ((s64)DIMM_PID_G_r) * (s64)integral;
1574 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1577 /* Calculate the derivative term */
1578 derivative = state->error_history[state->cur_sample] -
1579 state->error_history[(state->cur_sample + DIMM_PID_HISTORY_SIZE - 1)
1580 % DIMM_PID_HISTORY_SIZE];
1581 derivative /= DIMM_PID_INTERVAL;
1582 deriv_p = ((s64)DIMM_PID_G_d) * (s64)derivative;
1583 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1586 /* Calculate the proportional term */
1587 prop_p = ((s64)DIMM_PID_G_p) * (s64)(state->error_history[state->cur_sample]);
1588 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1594 DBG(" sum: %d\n", (int)sum);
1595 state->output = (s32)sum;
1596 state->output = max(state->output, DIMM_PID_OUTPUT_MIN);
1597 state->output = min(state->output, DIMM_PID_OUTPUT_MAX);
1598 dimm_output_clamp = state->output;
1600 DBG("** DIMM clamp value: %d\n", (int)state->output);
1602 /* Backside PID is only every 5 seconds, force backside fan clamping now */
1603 fan_min = (dimm_output_clamp * 100) / 14000;
1604 fan_min = max(fan_min, backside_params.output_min);
1605 if (backside_state.pwm < fan_min) {
1606 backside_state.pwm = fan_min;
1607 DBG(" -> applying clamp to backside fan now: %d !\n", fan_min);
1608 set_pwm_fan(BACKSIDE_FAN_PWM_INDEX, fan_min);
1613 * Initialize the state structure for the DIMM temp control loop
1615 static int init_dimms_state(struct dimm_pid_state *state)
1619 state->output = 4000;
1621 state->monitor = attach_i2c_chip(XSERVE_DIMMS_LM87, "dimms_temp");
1622 if (state->monitor == NULL)
1625 device_create_file(&of_dev->dev, &dev_attr_dimms_temperature);
1631 * Dispose of the state data for the DIMM control loop
1633 static void dispose_dimms_state(struct dimm_pid_state *state)
1635 if (state->monitor == NULL)
1638 device_remove_file(&of_dev->dev, &dev_attr_dimms_temperature);
1640 detach_i2c_chip(state->monitor);
1641 state->monitor = NULL;
1645 * Slots fan control loop
1647 static void do_monitor_slots(struct slots_pid_state *state)
1649 s32 temp, integral, derivative;
1650 s64 integ_p, deriv_p, prop_p, sum;
1653 if (--state->ticks != 0)
1655 state->ticks = SLOTS_PID_INTERVAL;
1659 /* Check fan status */
1660 rc = get_pwm_fan(SLOTS_FAN_PWM_INDEX);
1662 printk(KERN_WARNING "Error %d reading slots fan !\n", rc);
1663 /* XXX What do we do now ? */
1666 DBG(" current pwm: %d\n", state->pwm);
1668 /* Get some sensor readings */
1669 temp = le16_to_cpu(i2c_smbus_read_word_data(state->monitor,
1671 state->last_temp = temp;
1672 DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1673 FIX32TOPRINT(SLOTS_PID_INPUT_TARGET));
1675 /* Store temperature and error in history array */
1676 state->cur_sample = (state->cur_sample + 1) % SLOTS_PID_HISTORY_SIZE;
1677 state->sample_history[state->cur_sample] = temp;
1678 state->error_history[state->cur_sample] = temp - SLOTS_PID_INPUT_TARGET;
1680 /* If first loop, fill the history table */
1682 for (i = 0; i < (SLOTS_PID_HISTORY_SIZE - 1); i++) {
1683 state->cur_sample = (state->cur_sample + 1) %
1684 SLOTS_PID_HISTORY_SIZE;
1685 state->sample_history[state->cur_sample] = temp;
1686 state->error_history[state->cur_sample] =
1687 temp - SLOTS_PID_INPUT_TARGET;
1692 /* Calculate the integral term */
1695 for (i = 0; i < SLOTS_PID_HISTORY_SIZE; i++)
1696 integral += state->error_history[i];
1697 integral *= SLOTS_PID_INTERVAL;
1698 DBG(" integral: %08x\n", integral);
1699 integ_p = ((s64)SLOTS_PID_G_r) * (s64)integral;
1700 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1703 /* Calculate the derivative term */
1704 derivative = state->error_history[state->cur_sample] -
1705 state->error_history[(state->cur_sample + SLOTS_PID_HISTORY_SIZE - 1)
1706 % SLOTS_PID_HISTORY_SIZE];
1707 derivative /= SLOTS_PID_INTERVAL;
1708 deriv_p = ((s64)SLOTS_PID_G_d) * (s64)derivative;
1709 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1712 /* Calculate the proportional term */
1713 prop_p = ((s64)SLOTS_PID_G_p) * (s64)(state->error_history[state->cur_sample]);
1714 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1720 DBG(" sum: %d\n", (int)sum);
1721 state->pwm = (s32)sum;
1723 state->pwm = max(state->pwm, SLOTS_PID_OUTPUT_MIN);
1724 state->pwm = min(state->pwm, SLOTS_PID_OUTPUT_MAX);
1726 DBG("** DRIVES PWM: %d\n", (int)state->pwm);
1727 set_pwm_fan(SLOTS_FAN_PWM_INDEX, state->pwm);
1731 * Initialize the state structure for the slots bay fan control loop
1733 static int init_slots_state(struct slots_pid_state *state)
1739 state->monitor = attach_i2c_chip(XSERVE_SLOTS_LM75, "slots_temp");
1740 if (state->monitor == NULL)
1743 device_create_file(&of_dev->dev, &dev_attr_slots_temperature);
1744 device_create_file(&of_dev->dev, &dev_attr_slots_fan_pwm);
1750 * Dispose of the state data for the slots control loop
1752 static void dispose_slots_state(struct slots_pid_state *state)
1754 if (state->monitor == NULL)
1757 device_remove_file(&of_dev->dev, &dev_attr_slots_temperature);
1758 device_remove_file(&of_dev->dev, &dev_attr_slots_fan_pwm);
1760 detach_i2c_chip(state->monitor);
1761 state->monitor = NULL;
1765 static int call_critical_overtemp(void)
1767 char *argv[] = { critical_overtemp_path, NULL };
1768 static char *envp[] = { "HOME=/",
1770 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
1773 return call_usermodehelper(critical_overtemp_path, argv, envp, 0);
1778 * Here's the kernel thread that calls the various control loops
1780 static int main_control_loop(void *x)
1784 DBG("main_control_loop started\n");
1788 if (start_fcu() < 0) {
1789 printk(KERN_ERR "kfand: failed to start FCU\n");
1794 /* Set the PCI fan once for now on non-RackMac */
1796 set_pwm_fan(SLOTS_FAN_PWM_INDEX, SLOTS_FAN_DEFAULT_PWM);
1798 /* Initialize ADCs */
1799 initialize_adc(&cpu_state[0]);
1800 if (cpu_state[1].monitor != NULL)
1801 initialize_adc(&cpu_state[1]);
1803 fcu_tickle_ticks = FCU_TICKLE_TICKS;
1807 while (state == state_attached) {
1808 unsigned long elapsed, start;
1814 /* Tickle the FCU just in case */
1815 if (--fcu_tickle_ticks < 0) {
1816 fcu_tickle_ticks = FCU_TICKLE_TICKS;
1820 /* First, we always calculate the new DIMMs state on an Xserve */
1822 do_monitor_dimms(&dimms_state);
1824 /* Then, the CPUs */
1825 if (cpu_pid_type == CPU_PID_TYPE_COMBINED)
1826 do_monitor_cpu_combined();
1827 else if (cpu_pid_type == CPU_PID_TYPE_RACKMAC) {
1828 do_monitor_cpu_rack(&cpu_state[0]);
1829 if (cpu_state[1].monitor != NULL)
1830 do_monitor_cpu_rack(&cpu_state[1]);
1831 // better deal with UP
1833 do_monitor_cpu_split(&cpu_state[0]);
1834 if (cpu_state[1].monitor != NULL)
1835 do_monitor_cpu_split(&cpu_state[1]);
1836 // better deal with UP
1838 /* Then, the rest */
1839 do_monitor_backside(&backside_state);
1841 do_monitor_slots(&slots_state);
1843 do_monitor_drives(&drives_state);
1846 if (critical_state == 1) {
1847 printk(KERN_WARNING "Temperature control detected a critical condition\n");
1848 printk(KERN_WARNING "Attempting to shut down...\n");
1849 if (call_critical_overtemp()) {
1850 printk(KERN_WARNING "Can't call %s, power off now!\n",
1851 critical_overtemp_path);
1852 machine_power_off();
1855 if (critical_state > 0)
1857 if (critical_state > MAX_CRITICAL_STATE) {
1858 printk(KERN_WARNING "Shutdown timed out, power off now !\n");
1859 machine_power_off();
1862 // FIXME: Deal with signals
1863 elapsed = jiffies - start;
1865 schedule_timeout_interruptible(HZ - elapsed);
1869 DBG("main_control_loop ended\n");
1872 complete_and_exit(&ctrl_complete, 0);
1876 * Dispose the control loops when tearing down
1878 static void dispose_control_loops(void)
1880 dispose_cpu_state(&cpu_state[0]);
1881 dispose_cpu_state(&cpu_state[1]);
1882 dispose_backside_state(&backside_state);
1883 dispose_drives_state(&drives_state);
1884 dispose_slots_state(&slots_state);
1885 dispose_dimms_state(&dimms_state);
1889 * Create the control loops. U3-0 i2c bus is up, so we can now
1890 * get to the various sensors
1892 static int create_control_loops(void)
1894 struct device_node *np;
1896 /* Count CPUs from the device-tree, we don't care how many are
1897 * actually used by Linux
1900 for (np = NULL; NULL != (np = of_find_node_by_type(np, "cpu"));)
1903 DBG("counted %d CPUs in the device-tree\n", cpu_count);
1905 /* Decide the type of PID algorithm to use based on the presence of
1906 * the pumps, though that may not be the best way, that is good enough
1910 cpu_pid_type = CPU_PID_TYPE_RACKMAC;
1911 else if (machine_is_compatible("PowerMac7,3")
1913 && fcu_fans[CPUA_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID
1914 && fcu_fans[CPUB_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID) {
1915 printk(KERN_INFO "Liquid cooling pumps detected, using new algorithm !\n");
1916 cpu_pid_type = CPU_PID_TYPE_COMBINED;
1918 cpu_pid_type = CPU_PID_TYPE_SPLIT;
1920 /* Create control loops for everything. If any fail, everything
1923 if (init_cpu_state(&cpu_state[0], 0))
1925 if (cpu_pid_type == CPU_PID_TYPE_COMBINED)
1926 fetch_cpu_pumps_minmax();
1928 if (cpu_count > 1 && init_cpu_state(&cpu_state[1], 1))
1930 if (init_backside_state(&backside_state))
1932 if (rackmac && init_dimms_state(&dimms_state))
1934 if (rackmac && init_slots_state(&slots_state))
1936 if (!rackmac && init_drives_state(&drives_state))
1939 DBG("all control loops up !\n");
1944 DBG("failure creating control loops, disposing\n");
1946 dispose_control_loops();
1952 * Start the control loops after everything is up, that is create
1953 * the thread that will make them run
1955 static void start_control_loops(void)
1957 init_completion(&ctrl_complete);
1959 ctrl_task = kernel_thread(main_control_loop, NULL, SIGCHLD | CLONE_KERNEL);
1963 * Stop the control loops when tearing down
1965 static void stop_control_loops(void)
1968 wait_for_completion(&ctrl_complete);
1972 * Attach to the i2c FCU after detecting U3-1 bus
1974 static int attach_fcu(void)
1976 fcu = attach_i2c_chip(FAN_CTRLER_ID, "fcu");
1980 DBG("FCU attached\n");
1986 * Detach from the i2c FCU when tearing down
1988 static void detach_fcu(void)
1991 detach_i2c_chip(fcu);
1996 * Attach to the i2c controller. We probe the various chips based
1997 * on the device-tree nodes and build everything for the driver to
1998 * run, we then kick the driver monitoring thread
2000 static int therm_pm72_attach(struct i2c_adapter *adapter)
2005 if (state == state_detached)
2006 state = state_attaching;
2007 if (state != state_attaching) {
2012 /* Check if we are looking for one of these */
2013 if (u3_0 == NULL && !strcmp(adapter->name, "u3 0")) {
2015 DBG("found U3-0\n");
2017 if (create_control_loops())
2019 } else if (u3_1 == NULL && !strcmp(adapter->name, "u3 1")) {
2021 DBG("found U3-1, attaching FCU\n");
2024 } else if (k2 == NULL && !strcmp(adapter->name, "mac-io 0")) {
2027 if (u3_0 && rackmac)
2028 if (create_control_loops())
2031 /* We got all we need, start control loops */
2032 if (u3_0 != NULL && u3_1 != NULL && (k2 || !rackmac)) {
2033 DBG("everything up, starting control loops\n");
2034 state = state_attached;
2035 start_control_loops();
2043 * Called on every adapter when the driver or the i2c controller
2046 static int therm_pm72_detach(struct i2c_adapter *adapter)
2050 if (state != state_detached)
2051 state = state_detaching;
2053 /* Stop control loops if any */
2054 DBG("stopping control loops\n");
2056 stop_control_loops();
2059 if (u3_0 != NULL && !strcmp(adapter->name, "u3 0")) {
2060 DBG("lost U3-0, disposing control loops\n");
2061 dispose_control_loops();
2065 if (u3_1 != NULL && !strcmp(adapter->name, "u3 1")) {
2066 DBG("lost U3-1, detaching FCU\n");
2070 if (u3_0 == NULL && u3_1 == NULL)
2071 state = state_detached;
2078 static int fan_check_loc_match(const char *loc, int fan)
2083 strlcpy(tmp, fcu_fans[fan].loc, 64);
2090 if (strcmp(loc, c) == 0)
2099 static void fcu_lookup_fans(struct device_node *fcu_node)
2101 struct device_node *np = NULL;
2104 /* The table is filled by default with values that are suitable
2105 * for the old machines without device-tree informations. We scan
2106 * the device-tree and override those values with whatever is
2110 DBG("Looking up FCU controls in device-tree...\n");
2112 while ((np = of_get_next_child(fcu_node, np)) != NULL) {
2117 DBG(" control: %s, type: %s\n", np->name, np->type);
2119 /* Detect control type */
2120 if (!strcmp(np->type, "fan-rpm-control") ||
2121 !strcmp(np->type, "fan-rpm"))
2123 if (!strcmp(np->type, "fan-pwm-control") ||
2124 !strcmp(np->type, "fan-pwm"))
2126 /* Only care about fans for now */
2130 /* Lookup for a matching location */
2131 loc = get_property(np, "location", NULL);
2132 reg = get_property(np, "reg", NULL);
2133 if (loc == NULL || reg == NULL)
2135 DBG(" matching location: %s, reg: 0x%08x\n", loc, *reg);
2137 for (i = 0; i < FCU_FAN_COUNT; i++) {
2140 if (!fan_check_loc_match(loc, i))
2142 DBG(" location match, index: %d\n", i);
2143 fcu_fans[i].id = FCU_FAN_ABSENT_ID;
2144 if (type != fcu_fans[i].type) {
2145 printk(KERN_WARNING "therm_pm72: Fan type mismatch "
2146 "in device-tree for %s\n", np->full_name);
2149 if (type == FCU_FAN_RPM)
2150 fan_id = ((*reg) - 0x10) / 2;
2152 fan_id = ((*reg) - 0x30) / 2;
2154 printk(KERN_WARNING "therm_pm72: Can't parse "
2155 "fan ID in device-tree for %s\n", np->full_name);
2158 DBG(" fan id -> %d, type -> %d\n", fan_id, type);
2159 fcu_fans[i].id = fan_id;
2163 /* Now dump the array */
2164 printk(KERN_INFO "Detected fan controls:\n");
2165 for (i = 0; i < FCU_FAN_COUNT; i++) {
2166 if (fcu_fans[i].id == FCU_FAN_ABSENT_ID)
2168 printk(KERN_INFO " %d: %s fan, id %d, location: %s\n", i,
2169 fcu_fans[i].type == FCU_FAN_RPM ? "RPM" : "PWM",
2170 fcu_fans[i].id, fcu_fans[i].loc);
2174 static int fcu_of_probe(struct of_device* dev, const struct of_device_id *match)
2176 state = state_detached;
2178 /* Lookup the fans in the device tree */
2179 fcu_lookup_fans(dev->node);
2181 /* Add the driver */
2182 return i2c_add_driver(&therm_pm72_driver);
2185 static int fcu_of_remove(struct of_device* dev)
2187 i2c_del_driver(&therm_pm72_driver);
2192 static struct of_device_id fcu_match[] =
2200 static struct of_platform_driver fcu_of_platform_driver =
2202 .name = "temperature",
2203 .match_table = fcu_match,
2204 .probe = fcu_of_probe,
2205 .remove = fcu_of_remove
2209 * Check machine type, attach to i2c controller
2211 static int __init therm_pm72_init(void)
2213 struct device_node *np;
2215 rackmac = machine_is_compatible("RackMac3,1");
2217 if (!machine_is_compatible("PowerMac7,2") &&
2218 !machine_is_compatible("PowerMac7,3") &&
2222 printk(KERN_INFO "PowerMac G5 Thermal control driver %s\n", VERSION);
2224 np = of_find_node_by_type(NULL, "fcu");
2226 /* Some machines have strangely broken device-tree */
2227 np = of_find_node_by_path("/u3@0,f8000000/i2c@f8001000/fan@15e");
2229 printk(KERN_ERR "Can't find FCU in device-tree !\n");
2233 of_dev = of_platform_device_create(np, "temperature", NULL);
2234 if (of_dev == NULL) {
2235 printk(KERN_ERR "Can't register FCU platform device !\n");
2239 of_register_driver(&fcu_of_platform_driver);
2244 static void __exit therm_pm72_exit(void)
2246 of_unregister_driver(&fcu_of_platform_driver);
2249 of_device_unregister(of_dev);
2252 module_init(therm_pm72_init);
2253 module_exit(therm_pm72_exit);
2255 MODULE_AUTHOR("Benjamin Herrenschmidt <benh@kernel.crashing.org>");
2256 MODULE_DESCRIPTION("Driver for Apple's PowerMac G5 thermal control");
2257 MODULE_LICENSE("GPL");