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/wait.h>
121 #include <linux/reboot.h>
122 #include <linux/kmod.h>
123 #include <linux/i2c.h>
124 #include <linux/kthread.h>
125 #include <linux/mutex.h>
126 #include <linux/of_device.h>
127 #include <linux/of_platform.h>
128 #include <asm/prom.h>
129 #include <asm/machdep.h>
131 #include <asm/system.h>
132 #include <asm/sections.h>
133 #include <asm/macio.h>
135 #include "therm_pm72.h"
137 #define VERSION "1.3"
142 #define DBG(args...) printk(args)
144 #define DBG(args...) do { } while(0)
152 static struct of_device * of_dev;
153 static struct i2c_adapter * u3_0;
154 static struct i2c_adapter * u3_1;
155 static struct i2c_adapter * k2;
156 static struct i2c_client * fcu;
157 static struct cpu_pid_state cpu_state[2];
158 static struct basckside_pid_params backside_params;
159 static struct backside_pid_state backside_state;
160 static struct drives_pid_state drives_state;
161 static struct dimm_pid_state dimms_state;
162 static struct slots_pid_state slots_state;
164 static int cpu_count;
165 static int cpu_pid_type;
166 static struct task_struct *ctrl_task;
167 static struct completion ctrl_complete;
168 static int critical_state;
170 static s32 dimm_output_clamp;
171 static int fcu_rpm_shift;
172 static int fcu_tickle_ticks;
173 static DEFINE_MUTEX(driver_lock);
176 * We have 3 types of CPU PID control. One is "split" old style control
177 * for intake & exhaust fans, the other is "combined" control for both
178 * CPUs that also deals with the pumps when present. To be "compatible"
179 * with OS X at this point, we only use "COMBINED" on the machines that
180 * are identified as having the pumps (though that identification is at
181 * least dodgy). Ultimately, we could probably switch completely to this
182 * algorithm provided we hack it to deal with the UP case
184 #define CPU_PID_TYPE_SPLIT 0
185 #define CPU_PID_TYPE_COMBINED 1
186 #define CPU_PID_TYPE_RACKMAC 2
189 * This table describes all fans in the FCU. The "id" and "type" values
190 * are defaults valid for all earlier machines. Newer machines will
191 * eventually override the table content based on the device-tree
195 char* loc; /* location code */
196 int type; /* 0 = rpm, 1 = pwm, 2 = pump */
197 int id; /* id or -1 */
200 #define FCU_FAN_RPM 0
201 #define FCU_FAN_PWM 1
203 #define FCU_FAN_ABSENT_ID -1
205 #define FCU_FAN_COUNT ARRAY_SIZE(fcu_fans)
207 struct fcu_fan_table fcu_fans[] = {
208 [BACKSIDE_FAN_PWM_INDEX] = {
209 .loc = "BACKSIDE,SYS CTRLR FAN",
211 .id = BACKSIDE_FAN_PWM_DEFAULT_ID,
213 [DRIVES_FAN_RPM_INDEX] = {
216 .id = DRIVES_FAN_RPM_DEFAULT_ID,
218 [SLOTS_FAN_PWM_INDEX] = {
219 .loc = "SLOT,PCI FAN",
221 .id = SLOTS_FAN_PWM_DEFAULT_ID,
223 [CPUA_INTAKE_FAN_RPM_INDEX] = {
224 .loc = "CPU A INTAKE",
226 .id = CPUA_INTAKE_FAN_RPM_DEFAULT_ID,
228 [CPUA_EXHAUST_FAN_RPM_INDEX] = {
229 .loc = "CPU A EXHAUST",
231 .id = CPUA_EXHAUST_FAN_RPM_DEFAULT_ID,
233 [CPUB_INTAKE_FAN_RPM_INDEX] = {
234 .loc = "CPU B INTAKE",
236 .id = CPUB_INTAKE_FAN_RPM_DEFAULT_ID,
238 [CPUB_EXHAUST_FAN_RPM_INDEX] = {
239 .loc = "CPU B EXHAUST",
241 .id = CPUB_EXHAUST_FAN_RPM_DEFAULT_ID,
243 /* pumps aren't present by default, have to be looked up in the
246 [CPUA_PUMP_RPM_INDEX] = {
249 .id = FCU_FAN_ABSENT_ID,
251 [CPUB_PUMP_RPM_INDEX] = {
254 .id = FCU_FAN_ABSENT_ID,
257 [CPU_A1_FAN_RPM_INDEX] = {
260 .id = FCU_FAN_ABSENT_ID,
262 [CPU_A2_FAN_RPM_INDEX] = {
265 .id = FCU_FAN_ABSENT_ID,
267 [CPU_A3_FAN_RPM_INDEX] = {
270 .id = FCU_FAN_ABSENT_ID,
272 [CPU_B1_FAN_RPM_INDEX] = {
275 .id = FCU_FAN_ABSENT_ID,
277 [CPU_B2_FAN_RPM_INDEX] = {
280 .id = FCU_FAN_ABSENT_ID,
282 [CPU_B3_FAN_RPM_INDEX] = {
285 .id = FCU_FAN_ABSENT_ID,
290 * Utility function to create an i2c_client structure and
291 * attach it to one of u3 adapters
293 static struct i2c_client *attach_i2c_chip(int id, const char *name)
295 struct i2c_client *clt;
296 struct i2c_adapter *adap;
297 struct i2c_board_info info;
308 memset(&info, 0, sizeof(struct i2c_board_info));
309 info.addr = (id >> 1) & 0x7f;
310 strlcpy(info.type, "therm_pm72", I2C_NAME_SIZE);
311 clt = i2c_new_device(adap, &info);
313 printk(KERN_ERR "therm_pm72: Failed to attach to i2c ID 0x%x\n", id);
318 * Let i2c-core delete that device on driver removal.
319 * This is safe because i2c-core holds the core_lock mutex for us.
321 list_add_tail(&clt->detected, &clt->driver->clients);
326 * Here are the i2c chip access wrappers
329 static void initialize_adc(struct cpu_pid_state *state)
334 /* Read ADC the configuration register and cache it. We
335 * also make sure Config2 contains proper values, I've seen
336 * cases where we got stale grabage in there, thus preventing
337 * proper reading of conv. values
343 i2c_master_send(state->monitor, buf, 2);
345 /* Read & cache Config1 */
347 rc = i2c_master_send(state->monitor, buf, 1);
349 rc = i2c_master_recv(state->monitor, buf, 1);
351 state->adc_config = buf[0];
352 DBG("ADC config reg: %02x\n", state->adc_config);
353 /* Disable shutdown mode */
354 state->adc_config &= 0xfe;
356 buf[1] = state->adc_config;
357 rc = i2c_master_send(state->monitor, buf, 2);
361 printk(KERN_ERR "therm_pm72: Error reading ADC config"
365 static int read_smon_adc(struct cpu_pid_state *state, int chan)
367 int rc, data, tries = 0;
373 buf[1] = (state->adc_config & 0x1f) | (chan << 5);
374 rc = i2c_master_send(state->monitor, buf, 2);
377 /* Wait for convertion */
379 /* Switch to data register */
381 rc = i2c_master_send(state->monitor, buf, 1);
385 rc = i2c_master_recv(state->monitor, buf, 2);
388 data = ((u16)buf[0]) << 8 | (u16)buf[1];
391 DBG("Error reading ADC, retrying...\n");
393 printk(KERN_ERR "therm_pm72: Error reading ADC !\n");
400 static int read_lm87_reg(struct i2c_client * chip, int reg)
408 rc = i2c_master_send(chip, &buf, 1);
411 rc = i2c_master_recv(chip, &buf, 1);
416 DBG("Error reading LM87, retrying...\n");
418 printk(KERN_ERR "therm_pm72: Error reading LM87 !\n");
425 static int fan_read_reg(int reg, unsigned char *buf, int nb)
432 nw = i2c_master_send(fcu, buf, 1);
433 if (nw > 0 || (nw < 0 && nw != -EIO) || tries >= 100)
439 printk(KERN_ERR "Failure writing address to FCU: %d", nw);
444 nr = i2c_master_recv(fcu, buf, nb);
445 if (nr > 0 || (nr < 0 && nr != ENODEV) || tries >= 100)
451 printk(KERN_ERR "Failure reading data from FCU: %d", nw);
455 static int fan_write_reg(int reg, const unsigned char *ptr, int nb)
458 unsigned char buf[16];
461 memcpy(buf+1, ptr, nb);
465 nw = i2c_master_send(fcu, buf, nb);
466 if (nw > 0 || (nw < 0 && nw != EIO) || tries >= 100)
472 printk(KERN_ERR "Failure writing to FCU: %d", nw);
476 static int start_fcu(void)
478 unsigned char buf = 0xff;
481 rc = fan_write_reg(0xe, &buf, 1);
484 rc = fan_write_reg(0x2e, &buf, 1);
487 rc = fan_read_reg(0, &buf, 1);
490 fcu_rpm_shift = (buf == 1) ? 2 : 3;
491 printk(KERN_DEBUG "FCU Initialized, RPM fan shift is %d\n",
497 static int set_rpm_fan(int fan_index, int rpm)
499 unsigned char buf[2];
500 int rc, id, min, max;
502 if (fcu_fans[fan_index].type != FCU_FAN_RPM)
504 id = fcu_fans[fan_index].id;
505 if (id == FCU_FAN_ABSENT_ID)
508 min = 2400 >> fcu_rpm_shift;
509 max = 56000 >> fcu_rpm_shift;
515 buf[0] = rpm >> (8 - fcu_rpm_shift);
516 buf[1] = rpm << fcu_rpm_shift;
517 rc = fan_write_reg(0x10 + (id * 2), buf, 2);
523 static int get_rpm_fan(int fan_index, int programmed)
525 unsigned char failure;
526 unsigned char active;
527 unsigned char buf[2];
528 int rc, id, reg_base;
530 if (fcu_fans[fan_index].type != FCU_FAN_RPM)
532 id = fcu_fans[fan_index].id;
533 if (id == FCU_FAN_ABSENT_ID)
536 rc = fan_read_reg(0xb, &failure, 1);
539 if ((failure & (1 << id)) != 0)
541 rc = fan_read_reg(0xd, &active, 1);
544 if ((active & (1 << id)) == 0)
547 /* Programmed value or real current speed */
548 reg_base = programmed ? 0x10 : 0x11;
549 rc = fan_read_reg(reg_base + (id * 2), buf, 2);
553 return (buf[0] << (8 - fcu_rpm_shift)) | buf[1] >> fcu_rpm_shift;
556 static int set_pwm_fan(int fan_index, int pwm)
558 unsigned char buf[2];
561 if (fcu_fans[fan_index].type != FCU_FAN_PWM)
563 id = fcu_fans[fan_index].id;
564 if (id == FCU_FAN_ABSENT_ID)
571 pwm = (pwm * 2559) / 1000;
573 rc = fan_write_reg(0x30 + (id * 2), buf, 1);
579 static int get_pwm_fan(int fan_index)
581 unsigned char failure;
582 unsigned char active;
583 unsigned char buf[2];
586 if (fcu_fans[fan_index].type != FCU_FAN_PWM)
588 id = fcu_fans[fan_index].id;
589 if (id == FCU_FAN_ABSENT_ID)
592 rc = fan_read_reg(0x2b, &failure, 1);
595 if ((failure & (1 << id)) != 0)
597 rc = fan_read_reg(0x2d, &active, 1);
600 if ((active & (1 << id)) == 0)
603 /* Programmed value or real current speed */
604 rc = fan_read_reg(0x30 + (id * 2), buf, 1);
608 return (buf[0] * 1000) / 2559;
611 static void tickle_fcu(void)
615 pwm = get_pwm_fan(SLOTS_FAN_PWM_INDEX);
617 DBG("FCU Tickle, slots fan is: %d\n", pwm);
622 pwm = SLOTS_FAN_DEFAULT_PWM;
623 } else if (pwm < SLOTS_PID_OUTPUT_MIN)
624 pwm = SLOTS_PID_OUTPUT_MIN;
626 /* That is hopefully enough to make the FCU happy */
627 set_pwm_fan(SLOTS_FAN_PWM_INDEX, pwm);
632 * Utility routine to read the CPU calibration EEPROM data
633 * from the device-tree
635 static int read_eeprom(int cpu, struct mpu_data *out)
637 struct device_node *np;
642 /* prom.c routine for finding a node by path is a bit brain dead
643 * and requires exact @xxx unit numbers. This is a bit ugly but
644 * will work for these machines
646 sprintf(nodename, "/u3@0,f8000000/i2c@f8001000/cpuid@a%d", cpu ? 2 : 0);
647 np = of_find_node_by_path(nodename);
649 printk(KERN_ERR "therm_pm72: Failed to retrieve cpuid node from device-tree\n");
652 data = of_get_property(np, "cpuid", &len);
654 printk(KERN_ERR "therm_pm72: Failed to retrieve cpuid property from device-tree\n");
658 memcpy(out, data, sizeof(struct mpu_data));
664 static void fetch_cpu_pumps_minmax(void)
666 struct cpu_pid_state *state0 = &cpu_state[0];
667 struct cpu_pid_state *state1 = &cpu_state[1];
668 u16 pump_min = 0, pump_max = 0xffff;
671 /* Try to fetch pumps min/max infos from eeprom */
673 memcpy(&tmp, &state0->mpu.processor_part_num, 8);
674 if (tmp[0] != 0xffff && tmp[1] != 0xffff) {
675 pump_min = max(pump_min, tmp[0]);
676 pump_max = min(pump_max, tmp[1]);
678 if (tmp[2] != 0xffff && tmp[3] != 0xffff) {
679 pump_min = max(pump_min, tmp[2]);
680 pump_max = min(pump_max, tmp[3]);
683 /* Double check the values, this _IS_ needed as the EEPROM on
684 * some dual 2.5Ghz G5s seem, at least, to have both min & max
685 * same to the same value ... (grrrr)
687 if (pump_min == pump_max || pump_min == 0 || pump_max == 0xffff) {
688 pump_min = CPU_PUMP_OUTPUT_MIN;
689 pump_max = CPU_PUMP_OUTPUT_MAX;
692 state0->pump_min = state1->pump_min = pump_min;
693 state0->pump_max = state1->pump_max = pump_max;
697 * Now, unfortunately, sysfs doesn't give us a nice void * we could
698 * pass around to the attribute functions, so we don't really have
699 * choice but implement a bunch of them...
701 * That sucks a bit, we take the lock because FIX32TOPRINT evaluates
702 * the input twice... I accept patches :)
704 #define BUILD_SHOW_FUNC_FIX(name, data) \
705 static ssize_t show_##name(struct device *dev, struct device_attribute *attr, char *buf) \
708 mutex_lock(&driver_lock); \
709 r = sprintf(buf, "%d.%03d", FIX32TOPRINT(data)); \
710 mutex_unlock(&driver_lock); \
713 #define BUILD_SHOW_FUNC_INT(name, data) \
714 static ssize_t show_##name(struct device *dev, struct device_attribute *attr, char *buf) \
716 return sprintf(buf, "%d", data); \
719 BUILD_SHOW_FUNC_FIX(cpu0_temperature, cpu_state[0].last_temp)
720 BUILD_SHOW_FUNC_FIX(cpu0_voltage, cpu_state[0].voltage)
721 BUILD_SHOW_FUNC_FIX(cpu0_current, cpu_state[0].current_a)
722 BUILD_SHOW_FUNC_INT(cpu0_exhaust_fan_rpm, cpu_state[0].rpm)
723 BUILD_SHOW_FUNC_INT(cpu0_intake_fan_rpm, cpu_state[0].intake_rpm)
725 BUILD_SHOW_FUNC_FIX(cpu1_temperature, cpu_state[1].last_temp)
726 BUILD_SHOW_FUNC_FIX(cpu1_voltage, cpu_state[1].voltage)
727 BUILD_SHOW_FUNC_FIX(cpu1_current, cpu_state[1].current_a)
728 BUILD_SHOW_FUNC_INT(cpu1_exhaust_fan_rpm, cpu_state[1].rpm)
729 BUILD_SHOW_FUNC_INT(cpu1_intake_fan_rpm, cpu_state[1].intake_rpm)
731 BUILD_SHOW_FUNC_FIX(backside_temperature, backside_state.last_temp)
732 BUILD_SHOW_FUNC_INT(backside_fan_pwm, backside_state.pwm)
734 BUILD_SHOW_FUNC_FIX(drives_temperature, drives_state.last_temp)
735 BUILD_SHOW_FUNC_INT(drives_fan_rpm, drives_state.rpm)
737 BUILD_SHOW_FUNC_FIX(slots_temperature, slots_state.last_temp)
738 BUILD_SHOW_FUNC_INT(slots_fan_pwm, slots_state.pwm)
740 BUILD_SHOW_FUNC_FIX(dimms_temperature, dimms_state.last_temp)
742 static DEVICE_ATTR(cpu0_temperature,S_IRUGO,show_cpu0_temperature,NULL);
743 static DEVICE_ATTR(cpu0_voltage,S_IRUGO,show_cpu0_voltage,NULL);
744 static DEVICE_ATTR(cpu0_current,S_IRUGO,show_cpu0_current,NULL);
745 static DEVICE_ATTR(cpu0_exhaust_fan_rpm,S_IRUGO,show_cpu0_exhaust_fan_rpm,NULL);
746 static DEVICE_ATTR(cpu0_intake_fan_rpm,S_IRUGO,show_cpu0_intake_fan_rpm,NULL);
748 static DEVICE_ATTR(cpu1_temperature,S_IRUGO,show_cpu1_temperature,NULL);
749 static DEVICE_ATTR(cpu1_voltage,S_IRUGO,show_cpu1_voltage,NULL);
750 static DEVICE_ATTR(cpu1_current,S_IRUGO,show_cpu1_current,NULL);
751 static DEVICE_ATTR(cpu1_exhaust_fan_rpm,S_IRUGO,show_cpu1_exhaust_fan_rpm,NULL);
752 static DEVICE_ATTR(cpu1_intake_fan_rpm,S_IRUGO,show_cpu1_intake_fan_rpm,NULL);
754 static DEVICE_ATTR(backside_temperature,S_IRUGO,show_backside_temperature,NULL);
755 static DEVICE_ATTR(backside_fan_pwm,S_IRUGO,show_backside_fan_pwm,NULL);
757 static DEVICE_ATTR(drives_temperature,S_IRUGO,show_drives_temperature,NULL);
758 static DEVICE_ATTR(drives_fan_rpm,S_IRUGO,show_drives_fan_rpm,NULL);
760 static DEVICE_ATTR(slots_temperature,S_IRUGO,show_slots_temperature,NULL);
761 static DEVICE_ATTR(slots_fan_pwm,S_IRUGO,show_slots_fan_pwm,NULL);
763 static DEVICE_ATTR(dimms_temperature,S_IRUGO,show_dimms_temperature,NULL);
766 * CPUs fans control loop
769 static int do_read_one_cpu_values(struct cpu_pid_state *state, s32 *temp, s32 *power)
771 s32 ltemp, volts, amps;
774 /* Default (in case of error) */
775 *temp = state->cur_temp;
776 *power = state->cur_power;
778 if (cpu_pid_type == CPU_PID_TYPE_RACKMAC)
779 index = (state->index == 0) ?
780 CPU_A1_FAN_RPM_INDEX : CPU_B1_FAN_RPM_INDEX;
782 index = (state->index == 0) ?
783 CPUA_EXHAUST_FAN_RPM_INDEX : CPUB_EXHAUST_FAN_RPM_INDEX;
785 /* Read current fan status */
786 rc = get_rpm_fan(index, !RPM_PID_USE_ACTUAL_SPEED);
788 /* XXX What do we do now ? Nothing for now, keep old value, but
789 * return error upstream
791 DBG(" cpu %d, fan reading error !\n", state->index);
794 DBG(" cpu %d, exhaust RPM: %d\n", state->index, state->rpm);
797 /* Get some sensor readings and scale it */
798 ltemp = read_smon_adc(state, 1);
800 /* XXX What do we do now ? */
804 DBG(" cpu %d, temp reading error !\n", state->index);
806 /* Fixup temperature according to diode calibration
808 DBG(" cpu %d, temp raw: %04x, m_diode: %04x, b_diode: %04x\n",
810 ltemp, state->mpu.mdiode, state->mpu.bdiode);
811 *temp = ((s32)ltemp * (s32)state->mpu.mdiode + ((s32)state->mpu.bdiode << 12)) >> 2;
812 state->last_temp = *temp;
813 DBG(" temp: %d.%03d\n", FIX32TOPRINT((*temp)));
817 * Read voltage & current and calculate power
819 volts = read_smon_adc(state, 3);
820 amps = read_smon_adc(state, 4);
822 /* Scale voltage and current raw sensor values according to fixed scales
823 * obtained in Darwin and calculate power from I and V
825 volts *= ADC_CPU_VOLTAGE_SCALE;
826 amps *= ADC_CPU_CURRENT_SCALE;
827 *power = (((u64)volts) * ((u64)amps)) >> 16;
828 state->voltage = volts;
829 state->current_a = amps;
830 state->last_power = *power;
832 DBG(" cpu %d, current: %d.%03d, voltage: %d.%03d, power: %d.%03d W\n",
833 state->index, FIX32TOPRINT(state->current_a),
834 FIX32TOPRINT(state->voltage), FIX32TOPRINT(*power));
839 static void do_cpu_pid(struct cpu_pid_state *state, s32 temp, s32 power)
841 s32 power_target, integral, derivative, proportional, adj_in_target, sval;
842 s64 integ_p, deriv_p, prop_p, sum;
845 /* Calculate power target value (could be done once for all)
846 * and convert to a 16.16 fp number
848 power_target = ((u32)(state->mpu.pmaxh - state->mpu.padjmax)) << 16;
849 DBG(" power target: %d.%03d, error: %d.%03d\n",
850 FIX32TOPRINT(power_target), FIX32TOPRINT(power_target - power));
852 /* Store temperature and power in history array */
853 state->cur_temp = (state->cur_temp + 1) % CPU_TEMP_HISTORY_SIZE;
854 state->temp_history[state->cur_temp] = temp;
855 state->cur_power = (state->cur_power + 1) % state->count_power;
856 state->power_history[state->cur_power] = power;
857 state->error_history[state->cur_power] = power_target - power;
859 /* If first loop, fill the history table */
861 for (i = 0; i < (state->count_power - 1); i++) {
862 state->cur_power = (state->cur_power + 1) % state->count_power;
863 state->power_history[state->cur_power] = power;
864 state->error_history[state->cur_power] = power_target - power;
866 for (i = 0; i < (CPU_TEMP_HISTORY_SIZE - 1); i++) {
867 state->cur_temp = (state->cur_temp + 1) % CPU_TEMP_HISTORY_SIZE;
868 state->temp_history[state->cur_temp] = temp;
873 /* Calculate the integral term normally based on the "power" values */
876 for (i = 0; i < state->count_power; i++)
877 integral += state->error_history[i];
878 integral *= CPU_PID_INTERVAL;
879 DBG(" integral: %08x\n", integral);
881 /* Calculate the adjusted input (sense value).
884 * so the result is 28.36
886 * input target is mpu.ttarget, input max is mpu.tmax
888 integ_p = ((s64)state->mpu.pid_gr) * (s64)integral;
889 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
890 sval = (state->mpu.tmax << 16) - ((integ_p >> 20) & 0xffffffff);
891 adj_in_target = (state->mpu.ttarget << 16);
892 if (adj_in_target > sval)
893 adj_in_target = sval;
894 DBG(" adj_in_target: %d.%03d, ttarget: %d\n", FIX32TOPRINT(adj_in_target),
897 /* Calculate the derivative term */
898 derivative = state->temp_history[state->cur_temp] -
899 state->temp_history[(state->cur_temp + CPU_TEMP_HISTORY_SIZE - 1)
900 % CPU_TEMP_HISTORY_SIZE];
901 derivative /= CPU_PID_INTERVAL;
902 deriv_p = ((s64)state->mpu.pid_gd) * (s64)derivative;
903 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
906 /* Calculate the proportional term */
907 proportional = temp - adj_in_target;
908 prop_p = ((s64)state->mpu.pid_gp) * (s64)proportional;
909 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
915 DBG(" sum: %d\n", (int)sum);
916 state->rpm += (s32)sum;
919 static void do_monitor_cpu_combined(void)
921 struct cpu_pid_state *state0 = &cpu_state[0];
922 struct cpu_pid_state *state1 = &cpu_state[1];
923 s32 temp0, power0, temp1, power1;
924 s32 temp_combi, power_combi;
925 int rc, intake, pump;
927 rc = do_read_one_cpu_values(state0, &temp0, &power0);
929 /* XXX What do we do now ? */
931 state1->overtemp = 0;
932 rc = do_read_one_cpu_values(state1, &temp1, &power1);
934 /* XXX What do we do now ? */
936 if (state1->overtemp)
939 temp_combi = max(temp0, temp1);
940 power_combi = max(power0, power1);
942 /* Check tmax, increment overtemp if we are there. At tmax+8, we go
943 * full blown immediately and try to trigger a shutdown
945 if (temp_combi >= ((state0->mpu.tmax + 8) << 16)) {
946 printk(KERN_WARNING "Warning ! Temperature way above maximum (%d) !\n",
948 state0->overtemp += CPU_MAX_OVERTEMP / 4;
949 } else if (temp_combi > (state0->mpu.tmax << 16))
952 state0->overtemp = 0;
953 if (state0->overtemp >= CPU_MAX_OVERTEMP)
955 if (state0->overtemp > 0) {
956 state0->rpm = state0->mpu.rmaxn_exhaust_fan;
957 state0->intake_rpm = intake = state0->mpu.rmaxn_intake_fan;
958 pump = state0->pump_max;
963 do_cpu_pid(state0, temp_combi, power_combi);
966 state0->rpm = max(state0->rpm, (int)state0->mpu.rminn_exhaust_fan);
967 state0->rpm = min(state0->rpm, (int)state0->mpu.rmaxn_exhaust_fan);
969 /* Calculate intake fan speed */
970 intake = (state0->rpm * CPU_INTAKE_SCALE) >> 16;
971 intake = max(intake, (int)state0->mpu.rminn_intake_fan);
972 intake = min(intake, (int)state0->mpu.rmaxn_intake_fan);
973 state0->intake_rpm = intake;
975 /* Calculate pump speed */
976 pump = (state0->rpm * state0->pump_max) /
977 state0->mpu.rmaxn_exhaust_fan;
978 pump = min(pump, state0->pump_max);
979 pump = max(pump, state0->pump_min);
982 /* We copy values from state 0 to state 1 for /sysfs */
983 state1->rpm = state0->rpm;
984 state1->intake_rpm = state0->intake_rpm;
986 DBG("** CPU %d RPM: %d Ex, %d, Pump: %d, In, overtemp: %d\n",
987 state1->index, (int)state1->rpm, intake, pump, state1->overtemp);
989 /* We should check for errors, shouldn't we ? But then, what
990 * do we do once the error occurs ? For FCU notified fan
991 * failures (-EFAULT) we probably want to notify userland
994 set_rpm_fan(CPUA_INTAKE_FAN_RPM_INDEX, intake);
995 set_rpm_fan(CPUA_EXHAUST_FAN_RPM_INDEX, state0->rpm);
996 set_rpm_fan(CPUB_INTAKE_FAN_RPM_INDEX, intake);
997 set_rpm_fan(CPUB_EXHAUST_FAN_RPM_INDEX, state0->rpm);
999 if (fcu_fans[CPUA_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID)
1000 set_rpm_fan(CPUA_PUMP_RPM_INDEX, pump);
1001 if (fcu_fans[CPUB_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID)
1002 set_rpm_fan(CPUB_PUMP_RPM_INDEX, pump);
1005 static void do_monitor_cpu_split(struct cpu_pid_state *state)
1010 /* Read current fan status */
1011 rc = do_read_one_cpu_values(state, &temp, &power);
1013 /* XXX What do we do now ? */
1016 /* Check tmax, increment overtemp if we are there. At tmax+8, we go
1017 * full blown immediately and try to trigger a shutdown
1019 if (temp >= ((state->mpu.tmax + 8) << 16)) {
1020 printk(KERN_WARNING "Warning ! CPU %d temperature way above maximum"
1022 state->index, temp >> 16);
1023 state->overtemp += CPU_MAX_OVERTEMP / 4;
1024 } else if (temp > (state->mpu.tmax << 16))
1027 state->overtemp = 0;
1028 if (state->overtemp >= CPU_MAX_OVERTEMP)
1030 if (state->overtemp > 0) {
1031 state->rpm = state->mpu.rmaxn_exhaust_fan;
1032 state->intake_rpm = intake = state->mpu.rmaxn_intake_fan;
1037 do_cpu_pid(state, temp, power);
1040 state->rpm = max(state->rpm, (int)state->mpu.rminn_exhaust_fan);
1041 state->rpm = min(state->rpm, (int)state->mpu.rmaxn_exhaust_fan);
1043 /* Calculate intake fan */
1044 intake = (state->rpm * CPU_INTAKE_SCALE) >> 16;
1045 intake = max(intake, (int)state->mpu.rminn_intake_fan);
1046 intake = min(intake, (int)state->mpu.rmaxn_intake_fan);
1047 state->intake_rpm = intake;
1050 DBG("** CPU %d RPM: %d Ex, %d In, overtemp: %d\n",
1051 state->index, (int)state->rpm, intake, state->overtemp);
1053 /* We should check for errors, shouldn't we ? But then, what
1054 * do we do once the error occurs ? For FCU notified fan
1055 * failures (-EFAULT) we probably want to notify userland
1058 if (state->index == 0) {
1059 set_rpm_fan(CPUA_INTAKE_FAN_RPM_INDEX, intake);
1060 set_rpm_fan(CPUA_EXHAUST_FAN_RPM_INDEX, state->rpm);
1062 set_rpm_fan(CPUB_INTAKE_FAN_RPM_INDEX, intake);
1063 set_rpm_fan(CPUB_EXHAUST_FAN_RPM_INDEX, state->rpm);
1067 static void do_monitor_cpu_rack(struct cpu_pid_state *state)
1069 s32 temp, power, fan_min;
1072 /* Read current fan status */
1073 rc = do_read_one_cpu_values(state, &temp, &power);
1075 /* XXX What do we do now ? */
1078 /* Check tmax, increment overtemp if we are there. At tmax+8, we go
1079 * full blown immediately and try to trigger a shutdown
1081 if (temp >= ((state->mpu.tmax + 8) << 16)) {
1082 printk(KERN_WARNING "Warning ! CPU %d temperature way above maximum"
1084 state->index, temp >> 16);
1085 state->overtemp = CPU_MAX_OVERTEMP / 4;
1086 } else if (temp > (state->mpu.tmax << 16))
1089 state->overtemp = 0;
1090 if (state->overtemp >= CPU_MAX_OVERTEMP)
1092 if (state->overtemp > 0) {
1093 state->rpm = state->intake_rpm = state->mpu.rmaxn_intake_fan;
1098 do_cpu_pid(state, temp, power);
1100 /* Check clamp from dimms */
1101 fan_min = dimm_output_clamp;
1102 fan_min = max(fan_min, (int)state->mpu.rminn_intake_fan);
1104 DBG(" CPU min mpu = %d, min dimm = %d\n",
1105 state->mpu.rminn_intake_fan, dimm_output_clamp);
1107 state->rpm = max(state->rpm, (int)fan_min);
1108 state->rpm = min(state->rpm, (int)state->mpu.rmaxn_intake_fan);
1109 state->intake_rpm = state->rpm;
1112 DBG("** CPU %d RPM: %d overtemp: %d\n",
1113 state->index, (int)state->rpm, state->overtemp);
1115 /* We should check for errors, shouldn't we ? But then, what
1116 * do we do once the error occurs ? For FCU notified fan
1117 * failures (-EFAULT) we probably want to notify userland
1120 if (state->index == 0) {
1121 set_rpm_fan(CPU_A1_FAN_RPM_INDEX, state->rpm);
1122 set_rpm_fan(CPU_A2_FAN_RPM_INDEX, state->rpm);
1123 set_rpm_fan(CPU_A3_FAN_RPM_INDEX, state->rpm);
1125 set_rpm_fan(CPU_B1_FAN_RPM_INDEX, state->rpm);
1126 set_rpm_fan(CPU_B2_FAN_RPM_INDEX, state->rpm);
1127 set_rpm_fan(CPU_B3_FAN_RPM_INDEX, state->rpm);
1132 * Initialize the state structure for one CPU control loop
1134 static int init_cpu_state(struct cpu_pid_state *state, int index)
1138 state->index = index;
1140 state->rpm = (cpu_pid_type == CPU_PID_TYPE_RACKMAC) ? 4000 : 1000;
1141 state->overtemp = 0;
1142 state->adc_config = 0x00;
1146 state->monitor = attach_i2c_chip(SUPPLY_MONITOR_ID, "CPU0_monitor");
1147 else if (index == 1)
1148 state->monitor = attach_i2c_chip(SUPPLY_MONITORB_ID, "CPU1_monitor");
1149 if (state->monitor == NULL)
1152 if (read_eeprom(index, &state->mpu))
1155 state->count_power = state->mpu.tguardband;
1156 if (state->count_power > CPU_POWER_HISTORY_SIZE) {
1157 printk(KERN_WARNING "Warning ! too many power history slots\n");
1158 state->count_power = CPU_POWER_HISTORY_SIZE;
1160 DBG("CPU %d Using %d power history entries\n", index, state->count_power);
1163 err = device_create_file(&of_dev->dev, &dev_attr_cpu0_temperature);
1164 err |= device_create_file(&of_dev->dev, &dev_attr_cpu0_voltage);
1165 err |= device_create_file(&of_dev->dev, &dev_attr_cpu0_current);
1166 err |= device_create_file(&of_dev->dev, &dev_attr_cpu0_exhaust_fan_rpm);
1167 err |= device_create_file(&of_dev->dev, &dev_attr_cpu0_intake_fan_rpm);
1169 err = device_create_file(&of_dev->dev, &dev_attr_cpu1_temperature);
1170 err |= device_create_file(&of_dev->dev, &dev_attr_cpu1_voltage);
1171 err |= device_create_file(&of_dev->dev, &dev_attr_cpu1_current);
1172 err |= device_create_file(&of_dev->dev, &dev_attr_cpu1_exhaust_fan_rpm);
1173 err |= device_create_file(&of_dev->dev, &dev_attr_cpu1_intake_fan_rpm);
1176 printk(KERN_WARNING "Failed to create some of the atribute"
1177 "files for CPU %d\n", index);
1181 state->monitor = NULL;
1187 * Dispose of the state data for one CPU control loop
1189 static void dispose_cpu_state(struct cpu_pid_state *state)
1191 if (state->monitor == NULL)
1194 if (state->index == 0) {
1195 device_remove_file(&of_dev->dev, &dev_attr_cpu0_temperature);
1196 device_remove_file(&of_dev->dev, &dev_attr_cpu0_voltage);
1197 device_remove_file(&of_dev->dev, &dev_attr_cpu0_current);
1198 device_remove_file(&of_dev->dev, &dev_attr_cpu0_exhaust_fan_rpm);
1199 device_remove_file(&of_dev->dev, &dev_attr_cpu0_intake_fan_rpm);
1201 device_remove_file(&of_dev->dev, &dev_attr_cpu1_temperature);
1202 device_remove_file(&of_dev->dev, &dev_attr_cpu1_voltage);
1203 device_remove_file(&of_dev->dev, &dev_attr_cpu1_current);
1204 device_remove_file(&of_dev->dev, &dev_attr_cpu1_exhaust_fan_rpm);
1205 device_remove_file(&of_dev->dev, &dev_attr_cpu1_intake_fan_rpm);
1208 state->monitor = NULL;
1212 * Motherboard backside & U3 heatsink fan control loop
1214 static void do_monitor_backside(struct backside_pid_state *state)
1216 s32 temp, integral, derivative, fan_min;
1217 s64 integ_p, deriv_p, prop_p, sum;
1220 if (--state->ticks != 0)
1222 state->ticks = backside_params.interval;
1226 /* Check fan status */
1227 rc = get_pwm_fan(BACKSIDE_FAN_PWM_INDEX);
1229 printk(KERN_WARNING "Error %d reading backside fan !\n", rc);
1230 /* XXX What do we do now ? */
1233 DBG(" current pwm: %d\n", state->pwm);
1235 /* Get some sensor readings */
1236 temp = i2c_smbus_read_byte_data(state->monitor, MAX6690_EXT_TEMP) << 16;
1237 state->last_temp = temp;
1238 DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1239 FIX32TOPRINT(backside_params.input_target));
1241 /* Store temperature and error in history array */
1242 state->cur_sample = (state->cur_sample + 1) % BACKSIDE_PID_HISTORY_SIZE;
1243 state->sample_history[state->cur_sample] = temp;
1244 state->error_history[state->cur_sample] = temp - backside_params.input_target;
1246 /* If first loop, fill the history table */
1248 for (i = 0; i < (BACKSIDE_PID_HISTORY_SIZE - 1); i++) {
1249 state->cur_sample = (state->cur_sample + 1) %
1250 BACKSIDE_PID_HISTORY_SIZE;
1251 state->sample_history[state->cur_sample] = temp;
1252 state->error_history[state->cur_sample] =
1253 temp - backside_params.input_target;
1258 /* Calculate the integral term */
1261 for (i = 0; i < BACKSIDE_PID_HISTORY_SIZE; i++)
1262 integral += state->error_history[i];
1263 integral *= backside_params.interval;
1264 DBG(" integral: %08x\n", integral);
1265 integ_p = ((s64)backside_params.G_r) * (s64)integral;
1266 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1269 /* Calculate the derivative term */
1270 derivative = state->error_history[state->cur_sample] -
1271 state->error_history[(state->cur_sample + BACKSIDE_PID_HISTORY_SIZE - 1)
1272 % BACKSIDE_PID_HISTORY_SIZE];
1273 derivative /= backside_params.interval;
1274 deriv_p = ((s64)backside_params.G_d) * (s64)derivative;
1275 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1278 /* Calculate the proportional term */
1279 prop_p = ((s64)backside_params.G_p) * (s64)(state->error_history[state->cur_sample]);
1280 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1286 DBG(" sum: %d\n", (int)sum);
1287 if (backside_params.additive)
1288 state->pwm += (s32)sum;
1292 /* Check for clamp */
1293 fan_min = (dimm_output_clamp * 100) / 14000;
1294 fan_min = max(fan_min, backside_params.output_min);
1296 state->pwm = max(state->pwm, fan_min);
1297 state->pwm = min(state->pwm, backside_params.output_max);
1299 DBG("** BACKSIDE PWM: %d\n", (int)state->pwm);
1300 set_pwm_fan(BACKSIDE_FAN_PWM_INDEX, state->pwm);
1304 * Initialize the state structure for the backside fan control loop
1306 static int init_backside_state(struct backside_pid_state *state)
1308 struct device_node *u3;
1309 int u3h = 1; /* conservative by default */
1313 * There are different PID params for machines with U3 and machines
1314 * with U3H, pick the right ones now
1316 u3 = of_find_node_by_path("/u3@0,f8000000");
1318 const u32 *vers = of_get_property(u3, "device-rev", NULL);
1320 if (((*vers) & 0x3f) < 0x34)
1326 backside_params.G_d = BACKSIDE_PID_RACK_G_d;
1327 backside_params.input_target = BACKSIDE_PID_RACK_INPUT_TARGET;
1328 backside_params.output_min = BACKSIDE_PID_U3H_OUTPUT_MIN;
1329 backside_params.interval = BACKSIDE_PID_RACK_INTERVAL;
1330 backside_params.G_p = BACKSIDE_PID_RACK_G_p;
1331 backside_params.G_r = BACKSIDE_PID_G_r;
1332 backside_params.output_max = BACKSIDE_PID_OUTPUT_MAX;
1333 backside_params.additive = 0;
1335 backside_params.G_d = BACKSIDE_PID_U3H_G_d;
1336 backside_params.input_target = BACKSIDE_PID_U3H_INPUT_TARGET;
1337 backside_params.output_min = BACKSIDE_PID_U3H_OUTPUT_MIN;
1338 backside_params.interval = BACKSIDE_PID_INTERVAL;
1339 backside_params.G_p = BACKSIDE_PID_G_p;
1340 backside_params.G_r = BACKSIDE_PID_G_r;
1341 backside_params.output_max = BACKSIDE_PID_OUTPUT_MAX;
1342 backside_params.additive = 1;
1344 backside_params.G_d = BACKSIDE_PID_U3_G_d;
1345 backside_params.input_target = BACKSIDE_PID_U3_INPUT_TARGET;
1346 backside_params.output_min = BACKSIDE_PID_U3_OUTPUT_MIN;
1347 backside_params.interval = BACKSIDE_PID_INTERVAL;
1348 backside_params.G_p = BACKSIDE_PID_G_p;
1349 backside_params.G_r = BACKSIDE_PID_G_r;
1350 backside_params.output_max = BACKSIDE_PID_OUTPUT_MAX;
1351 backside_params.additive = 1;
1358 state->monitor = attach_i2c_chip(BACKSIDE_MAX_ID, "backside_temp");
1359 if (state->monitor == NULL)
1362 err = device_create_file(&of_dev->dev, &dev_attr_backside_temperature);
1363 err |= device_create_file(&of_dev->dev, &dev_attr_backside_fan_pwm);
1365 printk(KERN_WARNING "Failed to create attribute file(s)"
1366 " for backside fan\n");
1372 * Dispose of the state data for the backside control loop
1374 static void dispose_backside_state(struct backside_pid_state *state)
1376 if (state->monitor == NULL)
1379 device_remove_file(&of_dev->dev, &dev_attr_backside_temperature);
1380 device_remove_file(&of_dev->dev, &dev_attr_backside_fan_pwm);
1382 state->monitor = NULL;
1386 * Drives bay fan control loop
1388 static void do_monitor_drives(struct drives_pid_state *state)
1390 s32 temp, integral, derivative;
1391 s64 integ_p, deriv_p, prop_p, sum;
1394 if (--state->ticks != 0)
1396 state->ticks = DRIVES_PID_INTERVAL;
1400 /* Check fan status */
1401 rc = get_rpm_fan(DRIVES_FAN_RPM_INDEX, !RPM_PID_USE_ACTUAL_SPEED);
1403 printk(KERN_WARNING "Error %d reading drives fan !\n", rc);
1404 /* XXX What do we do now ? */
1407 DBG(" current rpm: %d\n", state->rpm);
1409 /* Get some sensor readings */
1410 temp = le16_to_cpu(i2c_smbus_read_word_data(state->monitor,
1412 state->last_temp = temp;
1413 DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1414 FIX32TOPRINT(DRIVES_PID_INPUT_TARGET));
1416 /* Store temperature and error in history array */
1417 state->cur_sample = (state->cur_sample + 1) % DRIVES_PID_HISTORY_SIZE;
1418 state->sample_history[state->cur_sample] = temp;
1419 state->error_history[state->cur_sample] = temp - DRIVES_PID_INPUT_TARGET;
1421 /* If first loop, fill the history table */
1423 for (i = 0; i < (DRIVES_PID_HISTORY_SIZE - 1); i++) {
1424 state->cur_sample = (state->cur_sample + 1) %
1425 DRIVES_PID_HISTORY_SIZE;
1426 state->sample_history[state->cur_sample] = temp;
1427 state->error_history[state->cur_sample] =
1428 temp - DRIVES_PID_INPUT_TARGET;
1433 /* Calculate the integral term */
1436 for (i = 0; i < DRIVES_PID_HISTORY_SIZE; i++)
1437 integral += state->error_history[i];
1438 integral *= DRIVES_PID_INTERVAL;
1439 DBG(" integral: %08x\n", integral);
1440 integ_p = ((s64)DRIVES_PID_G_r) * (s64)integral;
1441 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1444 /* Calculate the derivative term */
1445 derivative = state->error_history[state->cur_sample] -
1446 state->error_history[(state->cur_sample + DRIVES_PID_HISTORY_SIZE - 1)
1447 % DRIVES_PID_HISTORY_SIZE];
1448 derivative /= DRIVES_PID_INTERVAL;
1449 deriv_p = ((s64)DRIVES_PID_G_d) * (s64)derivative;
1450 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1453 /* Calculate the proportional term */
1454 prop_p = ((s64)DRIVES_PID_G_p) * (s64)(state->error_history[state->cur_sample]);
1455 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1461 DBG(" sum: %d\n", (int)sum);
1462 state->rpm += (s32)sum;
1464 state->rpm = max(state->rpm, DRIVES_PID_OUTPUT_MIN);
1465 state->rpm = min(state->rpm, DRIVES_PID_OUTPUT_MAX);
1467 DBG("** DRIVES RPM: %d\n", (int)state->rpm);
1468 set_rpm_fan(DRIVES_FAN_RPM_INDEX, state->rpm);
1472 * Initialize the state structure for the drives bay fan control loop
1474 static int init_drives_state(struct drives_pid_state *state)
1482 state->monitor = attach_i2c_chip(DRIVES_DALLAS_ID, "drives_temp");
1483 if (state->monitor == NULL)
1486 err = device_create_file(&of_dev->dev, &dev_attr_drives_temperature);
1487 err |= device_create_file(&of_dev->dev, &dev_attr_drives_fan_rpm);
1489 printk(KERN_WARNING "Failed to create attribute file(s)"
1490 " for drives bay fan\n");
1496 * Dispose of the state data for the drives control loop
1498 static void dispose_drives_state(struct drives_pid_state *state)
1500 if (state->monitor == NULL)
1503 device_remove_file(&of_dev->dev, &dev_attr_drives_temperature);
1504 device_remove_file(&of_dev->dev, &dev_attr_drives_fan_rpm);
1506 state->monitor = NULL;
1510 * DIMMs temp control loop
1512 static void do_monitor_dimms(struct dimm_pid_state *state)
1514 s32 temp, integral, derivative, fan_min;
1515 s64 integ_p, deriv_p, prop_p, sum;
1518 if (--state->ticks != 0)
1520 state->ticks = DIMM_PID_INTERVAL;
1524 DBG(" current value: %d\n", state->output);
1526 temp = read_lm87_reg(state->monitor, LM87_INT_TEMP);
1530 state->last_temp = temp;
1531 DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1532 FIX32TOPRINT(DIMM_PID_INPUT_TARGET));
1534 /* Store temperature and error in history array */
1535 state->cur_sample = (state->cur_sample + 1) % DIMM_PID_HISTORY_SIZE;
1536 state->sample_history[state->cur_sample] = temp;
1537 state->error_history[state->cur_sample] = temp - DIMM_PID_INPUT_TARGET;
1539 /* If first loop, fill the history table */
1541 for (i = 0; i < (DIMM_PID_HISTORY_SIZE - 1); i++) {
1542 state->cur_sample = (state->cur_sample + 1) %
1543 DIMM_PID_HISTORY_SIZE;
1544 state->sample_history[state->cur_sample] = temp;
1545 state->error_history[state->cur_sample] =
1546 temp - DIMM_PID_INPUT_TARGET;
1551 /* Calculate the integral term */
1554 for (i = 0; i < DIMM_PID_HISTORY_SIZE; i++)
1555 integral += state->error_history[i];
1556 integral *= DIMM_PID_INTERVAL;
1557 DBG(" integral: %08x\n", integral);
1558 integ_p = ((s64)DIMM_PID_G_r) * (s64)integral;
1559 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1562 /* Calculate the derivative term */
1563 derivative = state->error_history[state->cur_sample] -
1564 state->error_history[(state->cur_sample + DIMM_PID_HISTORY_SIZE - 1)
1565 % DIMM_PID_HISTORY_SIZE];
1566 derivative /= DIMM_PID_INTERVAL;
1567 deriv_p = ((s64)DIMM_PID_G_d) * (s64)derivative;
1568 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1571 /* Calculate the proportional term */
1572 prop_p = ((s64)DIMM_PID_G_p) * (s64)(state->error_history[state->cur_sample]);
1573 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1579 DBG(" sum: %d\n", (int)sum);
1580 state->output = (s32)sum;
1581 state->output = max(state->output, DIMM_PID_OUTPUT_MIN);
1582 state->output = min(state->output, DIMM_PID_OUTPUT_MAX);
1583 dimm_output_clamp = state->output;
1585 DBG("** DIMM clamp value: %d\n", (int)state->output);
1587 /* Backside PID is only every 5 seconds, force backside fan clamping now */
1588 fan_min = (dimm_output_clamp * 100) / 14000;
1589 fan_min = max(fan_min, backside_params.output_min);
1590 if (backside_state.pwm < fan_min) {
1591 backside_state.pwm = fan_min;
1592 DBG(" -> applying clamp to backside fan now: %d !\n", fan_min);
1593 set_pwm_fan(BACKSIDE_FAN_PWM_INDEX, fan_min);
1598 * Initialize the state structure for the DIMM temp control loop
1600 static int init_dimms_state(struct dimm_pid_state *state)
1604 state->output = 4000;
1606 state->monitor = attach_i2c_chip(XSERVE_DIMMS_LM87, "dimms_temp");
1607 if (state->monitor == NULL)
1610 if (device_create_file(&of_dev->dev, &dev_attr_dimms_temperature))
1611 printk(KERN_WARNING "Failed to create attribute file"
1612 " for DIMM temperature\n");
1618 * Dispose of the state data for the DIMM control loop
1620 static void dispose_dimms_state(struct dimm_pid_state *state)
1622 if (state->monitor == NULL)
1625 device_remove_file(&of_dev->dev, &dev_attr_dimms_temperature);
1627 state->monitor = NULL;
1631 * Slots fan control loop
1633 static void do_monitor_slots(struct slots_pid_state *state)
1635 s32 temp, integral, derivative;
1636 s64 integ_p, deriv_p, prop_p, sum;
1639 if (--state->ticks != 0)
1641 state->ticks = SLOTS_PID_INTERVAL;
1645 /* Check fan status */
1646 rc = get_pwm_fan(SLOTS_FAN_PWM_INDEX);
1648 printk(KERN_WARNING "Error %d reading slots fan !\n", rc);
1649 /* XXX What do we do now ? */
1652 DBG(" current pwm: %d\n", state->pwm);
1654 /* Get some sensor readings */
1655 temp = le16_to_cpu(i2c_smbus_read_word_data(state->monitor,
1657 state->last_temp = temp;
1658 DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1659 FIX32TOPRINT(SLOTS_PID_INPUT_TARGET));
1661 /* Store temperature and error in history array */
1662 state->cur_sample = (state->cur_sample + 1) % SLOTS_PID_HISTORY_SIZE;
1663 state->sample_history[state->cur_sample] = temp;
1664 state->error_history[state->cur_sample] = temp - SLOTS_PID_INPUT_TARGET;
1666 /* If first loop, fill the history table */
1668 for (i = 0; i < (SLOTS_PID_HISTORY_SIZE - 1); i++) {
1669 state->cur_sample = (state->cur_sample + 1) %
1670 SLOTS_PID_HISTORY_SIZE;
1671 state->sample_history[state->cur_sample] = temp;
1672 state->error_history[state->cur_sample] =
1673 temp - SLOTS_PID_INPUT_TARGET;
1678 /* Calculate the integral term */
1681 for (i = 0; i < SLOTS_PID_HISTORY_SIZE; i++)
1682 integral += state->error_history[i];
1683 integral *= SLOTS_PID_INTERVAL;
1684 DBG(" integral: %08x\n", integral);
1685 integ_p = ((s64)SLOTS_PID_G_r) * (s64)integral;
1686 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1689 /* Calculate the derivative term */
1690 derivative = state->error_history[state->cur_sample] -
1691 state->error_history[(state->cur_sample + SLOTS_PID_HISTORY_SIZE - 1)
1692 % SLOTS_PID_HISTORY_SIZE];
1693 derivative /= SLOTS_PID_INTERVAL;
1694 deriv_p = ((s64)SLOTS_PID_G_d) * (s64)derivative;
1695 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1698 /* Calculate the proportional term */
1699 prop_p = ((s64)SLOTS_PID_G_p) * (s64)(state->error_history[state->cur_sample]);
1700 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1706 DBG(" sum: %d\n", (int)sum);
1707 state->pwm = (s32)sum;
1709 state->pwm = max(state->pwm, SLOTS_PID_OUTPUT_MIN);
1710 state->pwm = min(state->pwm, SLOTS_PID_OUTPUT_MAX);
1712 DBG("** DRIVES PWM: %d\n", (int)state->pwm);
1713 set_pwm_fan(SLOTS_FAN_PWM_INDEX, state->pwm);
1717 * Initialize the state structure for the slots bay fan control loop
1719 static int init_slots_state(struct slots_pid_state *state)
1727 state->monitor = attach_i2c_chip(XSERVE_SLOTS_LM75, "slots_temp");
1728 if (state->monitor == NULL)
1731 err = device_create_file(&of_dev->dev, &dev_attr_slots_temperature);
1732 err |= device_create_file(&of_dev->dev, &dev_attr_slots_fan_pwm);
1734 printk(KERN_WARNING "Failed to create attribute file(s)"
1735 " for slots bay fan\n");
1741 * Dispose of the state data for the slots control loop
1743 static void dispose_slots_state(struct slots_pid_state *state)
1745 if (state->monitor == NULL)
1748 device_remove_file(&of_dev->dev, &dev_attr_slots_temperature);
1749 device_remove_file(&of_dev->dev, &dev_attr_slots_fan_pwm);
1751 state->monitor = NULL;
1755 static int call_critical_overtemp(void)
1757 char *argv[] = { critical_overtemp_path, NULL };
1758 static char *envp[] = { "HOME=/",
1760 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
1763 return call_usermodehelper(critical_overtemp_path,
1764 argv, envp, UMH_WAIT_EXEC);
1769 * Here's the kernel thread that calls the various control loops
1771 static int main_control_loop(void *x)
1773 DBG("main_control_loop started\n");
1775 mutex_lock(&driver_lock);
1777 if (start_fcu() < 0) {
1778 printk(KERN_ERR "kfand: failed to start FCU\n");
1779 mutex_unlock(&driver_lock);
1783 /* Set the PCI fan once for now on non-RackMac */
1785 set_pwm_fan(SLOTS_FAN_PWM_INDEX, SLOTS_FAN_DEFAULT_PWM);
1787 /* Initialize ADCs */
1788 initialize_adc(&cpu_state[0]);
1789 if (cpu_state[1].monitor != NULL)
1790 initialize_adc(&cpu_state[1]);
1792 fcu_tickle_ticks = FCU_TICKLE_TICKS;
1794 mutex_unlock(&driver_lock);
1796 while (state == state_attached) {
1797 unsigned long elapsed, start;
1801 mutex_lock(&driver_lock);
1803 /* Tickle the FCU just in case */
1804 if (--fcu_tickle_ticks < 0) {
1805 fcu_tickle_ticks = FCU_TICKLE_TICKS;
1809 /* First, we always calculate the new DIMMs state on an Xserve */
1811 do_monitor_dimms(&dimms_state);
1813 /* Then, the CPUs */
1814 if (cpu_pid_type == CPU_PID_TYPE_COMBINED)
1815 do_monitor_cpu_combined();
1816 else if (cpu_pid_type == CPU_PID_TYPE_RACKMAC) {
1817 do_monitor_cpu_rack(&cpu_state[0]);
1818 if (cpu_state[1].monitor != NULL)
1819 do_monitor_cpu_rack(&cpu_state[1]);
1820 // better deal with UP
1822 do_monitor_cpu_split(&cpu_state[0]);
1823 if (cpu_state[1].monitor != NULL)
1824 do_monitor_cpu_split(&cpu_state[1]);
1825 // better deal with UP
1827 /* Then, the rest */
1828 do_monitor_backside(&backside_state);
1830 do_monitor_slots(&slots_state);
1832 do_monitor_drives(&drives_state);
1833 mutex_unlock(&driver_lock);
1835 if (critical_state == 1) {
1836 printk(KERN_WARNING "Temperature control detected a critical condition\n");
1837 printk(KERN_WARNING "Attempting to shut down...\n");
1838 if (call_critical_overtemp()) {
1839 printk(KERN_WARNING "Can't call %s, power off now!\n",
1840 critical_overtemp_path);
1841 machine_power_off();
1844 if (critical_state > 0)
1846 if (critical_state > MAX_CRITICAL_STATE) {
1847 printk(KERN_WARNING "Shutdown timed out, power off now !\n");
1848 machine_power_off();
1851 // FIXME: Deal with signals
1852 elapsed = jiffies - start;
1854 schedule_timeout_interruptible(HZ - elapsed);
1858 DBG("main_control_loop ended\n");
1861 complete_and_exit(&ctrl_complete, 0);
1865 * Dispose the control loops when tearing down
1867 static void dispose_control_loops(void)
1869 dispose_cpu_state(&cpu_state[0]);
1870 dispose_cpu_state(&cpu_state[1]);
1871 dispose_backside_state(&backside_state);
1872 dispose_drives_state(&drives_state);
1873 dispose_slots_state(&slots_state);
1874 dispose_dimms_state(&dimms_state);
1878 * Create the control loops. U3-0 i2c bus is up, so we can now
1879 * get to the various sensors
1881 static int create_control_loops(void)
1883 struct device_node *np;
1885 /* Count CPUs from the device-tree, we don't care how many are
1886 * actually used by Linux
1889 for (np = NULL; NULL != (np = of_find_node_by_type(np, "cpu"));)
1892 DBG("counted %d CPUs in the device-tree\n", cpu_count);
1894 /* Decide the type of PID algorithm to use based on the presence of
1895 * the pumps, though that may not be the best way, that is good enough
1899 cpu_pid_type = CPU_PID_TYPE_RACKMAC;
1900 else if (machine_is_compatible("PowerMac7,3")
1902 && fcu_fans[CPUA_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID
1903 && fcu_fans[CPUB_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID) {
1904 printk(KERN_INFO "Liquid cooling pumps detected, using new algorithm !\n");
1905 cpu_pid_type = CPU_PID_TYPE_COMBINED;
1907 cpu_pid_type = CPU_PID_TYPE_SPLIT;
1909 /* Create control loops for everything. If any fail, everything
1912 if (init_cpu_state(&cpu_state[0], 0))
1914 if (cpu_pid_type == CPU_PID_TYPE_COMBINED)
1915 fetch_cpu_pumps_minmax();
1917 if (cpu_count > 1 && init_cpu_state(&cpu_state[1], 1))
1919 if (init_backside_state(&backside_state))
1921 if (rackmac && init_dimms_state(&dimms_state))
1923 if (rackmac && init_slots_state(&slots_state))
1925 if (!rackmac && init_drives_state(&drives_state))
1928 DBG("all control loops up !\n");
1933 DBG("failure creating control loops, disposing\n");
1935 dispose_control_loops();
1941 * Start the control loops after everything is up, that is create
1942 * the thread that will make them run
1944 static void start_control_loops(void)
1946 init_completion(&ctrl_complete);
1948 ctrl_task = kthread_run(main_control_loop, NULL, "kfand");
1952 * Stop the control loops when tearing down
1954 static void stop_control_loops(void)
1957 wait_for_completion(&ctrl_complete);
1961 * Attach to the i2c FCU after detecting U3-1 bus
1963 static int attach_fcu(void)
1965 fcu = attach_i2c_chip(FAN_CTRLER_ID, "fcu");
1969 DBG("FCU attached\n");
1975 * Detach from the i2c FCU when tearing down
1977 static void detach_fcu(void)
1983 * Attach to the i2c controller. We probe the various chips based
1984 * on the device-tree nodes and build everything for the driver to
1985 * run, we then kick the driver monitoring thread
1987 static int therm_pm72_attach(struct i2c_adapter *adapter)
1989 mutex_lock(&driver_lock);
1992 if (state == state_detached)
1993 state = state_attaching;
1994 if (state != state_attaching) {
1995 mutex_unlock(&driver_lock);
1999 /* Check if we are looking for one of these */
2000 if (u3_0 == NULL && !strcmp(adapter->name, "u3 0")) {
2002 DBG("found U3-0\n");
2004 if (create_control_loops())
2006 } else if (u3_1 == NULL && !strcmp(adapter->name, "u3 1")) {
2008 DBG("found U3-1, attaching FCU\n");
2011 } else if (k2 == NULL && !strcmp(adapter->name, "mac-io 0")) {
2014 if (u3_0 && rackmac)
2015 if (create_control_loops())
2018 /* We got all we need, start control loops */
2019 if (u3_0 != NULL && u3_1 != NULL && (k2 || !rackmac)) {
2020 DBG("everything up, starting control loops\n");
2021 state = state_attached;
2022 start_control_loops();
2024 mutex_unlock(&driver_lock);
2029 static int therm_pm72_probe(struct i2c_client *client,
2030 const struct i2c_device_id *id)
2032 /* Always succeed, the real work was done in therm_pm72_attach() */
2037 * Called when any of the devices which participates into thermal management
2040 static int therm_pm72_remove(struct i2c_client *client)
2042 struct i2c_adapter *adapter = client->adapter;
2044 mutex_lock(&driver_lock);
2046 if (state != state_detached)
2047 state = state_detaching;
2049 /* Stop control loops if any */
2050 DBG("stopping control loops\n");
2051 mutex_unlock(&driver_lock);
2052 stop_control_loops();
2053 mutex_lock(&driver_lock);
2055 if (u3_0 != NULL && !strcmp(adapter->name, "u3 0")) {
2056 DBG("lost U3-0, disposing control loops\n");
2057 dispose_control_loops();
2061 if (u3_1 != NULL && !strcmp(adapter->name, "u3 1")) {
2062 DBG("lost U3-1, detaching FCU\n");
2066 if (u3_0 == NULL && u3_1 == NULL)
2067 state = state_detached;
2069 mutex_unlock(&driver_lock);
2075 * i2c_driver structure to attach to the host i2c controller
2078 static const struct i2c_device_id therm_pm72_id[] = {
2080 * Fake device name, thermal management is done by several
2081 * chips but we don't need to differentiate between them at
2084 { "therm_pm72", 0 },
2088 static struct i2c_driver therm_pm72_driver = {
2090 .name = "therm_pm72",
2092 .attach_adapter = therm_pm72_attach,
2093 .probe = therm_pm72_probe,
2094 .remove = therm_pm72_remove,
2095 .id_table = therm_pm72_id,
2098 static int fan_check_loc_match(const char *loc, int fan)
2103 strlcpy(tmp, fcu_fans[fan].loc, 64);
2110 if (strcmp(loc, c) == 0)
2119 static void fcu_lookup_fans(struct device_node *fcu_node)
2121 struct device_node *np = NULL;
2124 /* The table is filled by default with values that are suitable
2125 * for the old machines without device-tree informations. We scan
2126 * the device-tree and override those values with whatever is
2130 DBG("Looking up FCU controls in device-tree...\n");
2132 while ((np = of_get_next_child(fcu_node, np)) != NULL) {
2137 DBG(" control: %s, type: %s\n", np->name, np->type);
2139 /* Detect control type */
2140 if (!strcmp(np->type, "fan-rpm-control") ||
2141 !strcmp(np->type, "fan-rpm"))
2143 if (!strcmp(np->type, "fan-pwm-control") ||
2144 !strcmp(np->type, "fan-pwm"))
2146 /* Only care about fans for now */
2150 /* Lookup for a matching location */
2151 loc = of_get_property(np, "location", NULL);
2152 reg = of_get_property(np, "reg", NULL);
2153 if (loc == NULL || reg == NULL)
2155 DBG(" matching location: %s, reg: 0x%08x\n", loc, *reg);
2157 for (i = 0; i < FCU_FAN_COUNT; i++) {
2160 if (!fan_check_loc_match(loc, i))
2162 DBG(" location match, index: %d\n", i);
2163 fcu_fans[i].id = FCU_FAN_ABSENT_ID;
2164 if (type != fcu_fans[i].type) {
2165 printk(KERN_WARNING "therm_pm72: Fan type mismatch "
2166 "in device-tree for %s\n", np->full_name);
2169 if (type == FCU_FAN_RPM)
2170 fan_id = ((*reg) - 0x10) / 2;
2172 fan_id = ((*reg) - 0x30) / 2;
2174 printk(KERN_WARNING "therm_pm72: Can't parse "
2175 "fan ID in device-tree for %s\n", np->full_name);
2178 DBG(" fan id -> %d, type -> %d\n", fan_id, type);
2179 fcu_fans[i].id = fan_id;
2183 /* Now dump the array */
2184 printk(KERN_INFO "Detected fan controls:\n");
2185 for (i = 0; i < FCU_FAN_COUNT; i++) {
2186 if (fcu_fans[i].id == FCU_FAN_ABSENT_ID)
2188 printk(KERN_INFO " %d: %s fan, id %d, location: %s\n", i,
2189 fcu_fans[i].type == FCU_FAN_RPM ? "RPM" : "PWM",
2190 fcu_fans[i].id, fcu_fans[i].loc);
2194 static int fcu_of_probe(struct of_device* dev, const struct of_device_id *match)
2196 state = state_detached;
2198 /* Lookup the fans in the device tree */
2199 fcu_lookup_fans(dev->node);
2201 /* Add the driver */
2202 return i2c_add_driver(&therm_pm72_driver);
2205 static int fcu_of_remove(struct of_device* dev)
2207 i2c_del_driver(&therm_pm72_driver);
2212 static struct of_device_id fcu_match[] =
2220 static struct of_platform_driver fcu_of_platform_driver =
2222 .name = "temperature",
2223 .match_table = fcu_match,
2224 .probe = fcu_of_probe,
2225 .remove = fcu_of_remove
2229 * Check machine type, attach to i2c controller
2231 static int __init therm_pm72_init(void)
2233 struct device_node *np;
2235 rackmac = machine_is_compatible("RackMac3,1");
2237 if (!machine_is_compatible("PowerMac7,2") &&
2238 !machine_is_compatible("PowerMac7,3") &&
2242 printk(KERN_INFO "PowerMac G5 Thermal control driver %s\n", VERSION);
2244 np = of_find_node_by_type(NULL, "fcu");
2246 /* Some machines have strangely broken device-tree */
2247 np = of_find_node_by_path("/u3@0,f8000000/i2c@f8001000/fan@15e");
2249 printk(KERN_ERR "Can't find FCU in device-tree !\n");
2253 of_dev = of_platform_device_create(np, "temperature", NULL);
2254 if (of_dev == NULL) {
2255 printk(KERN_ERR "Can't register FCU platform device !\n");
2259 of_register_platform_driver(&fcu_of_platform_driver);
2264 static void __exit therm_pm72_exit(void)
2266 of_unregister_platform_driver(&fcu_of_platform_driver);
2269 of_device_unregister(of_dev);
2272 module_init(therm_pm72_init);
2273 module_exit(therm_pm72_exit);
2275 MODULE_AUTHOR("Benjamin Herrenschmidt <benh@kernel.crashing.org>");
2276 MODULE_DESCRIPTION("Driver for Apple's PowerMac G5 thermal control");
2277 MODULE_LICENSE("GPL");