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>
132 #include <asm/of_platform.h>
134 #include "therm_pm72.h"
136 #define VERSION "1.3"
141 #define DBG(args...) printk(args)
143 #define DBG(args...) do { } while(0)
151 static struct of_device * of_dev;
152 static struct i2c_adapter * u3_0;
153 static struct i2c_adapter * u3_1;
154 static struct i2c_adapter * k2;
155 static struct i2c_client * fcu;
156 static struct cpu_pid_state cpu_state[2];
157 static struct basckside_pid_params backside_params;
158 static struct backside_pid_state backside_state;
159 static struct drives_pid_state drives_state;
160 static struct dimm_pid_state dimms_state;
161 static struct slots_pid_state slots_state;
163 static int cpu_count;
164 static int cpu_pid_type;
165 static pid_t ctrl_task;
166 static struct completion ctrl_complete;
167 static int critical_state;
169 static s32 dimm_output_clamp;
170 static int fcu_rpm_shift;
171 static int fcu_tickle_ticks;
172 static DECLARE_MUTEX(driver_lock);
175 * We have 3 types of CPU PID control. One is "split" old style control
176 * for intake & exhaust fans, the other is "combined" control for both
177 * CPUs that also deals with the pumps when present. To be "compatible"
178 * with OS X at this point, we only use "COMBINED" on the machines that
179 * are identified as having the pumps (though that identification is at
180 * least dodgy). Ultimately, we could probably switch completely to this
181 * algorithm provided we hack it to deal with the UP case
183 #define CPU_PID_TYPE_SPLIT 0
184 #define CPU_PID_TYPE_COMBINED 1
185 #define CPU_PID_TYPE_RACKMAC 2
188 * This table describes all fans in the FCU. The "id" and "type" values
189 * are defaults valid for all earlier machines. Newer machines will
190 * eventually override the table content based on the device-tree
194 char* loc; /* location code */
195 int type; /* 0 = rpm, 1 = pwm, 2 = pump */
196 int id; /* id or -1 */
199 #define FCU_FAN_RPM 0
200 #define FCU_FAN_PWM 1
202 #define FCU_FAN_ABSENT_ID -1
204 #define FCU_FAN_COUNT ARRAY_SIZE(fcu_fans)
206 struct fcu_fan_table fcu_fans[] = {
207 [BACKSIDE_FAN_PWM_INDEX] = {
208 .loc = "BACKSIDE,SYS CTRLR FAN",
210 .id = BACKSIDE_FAN_PWM_DEFAULT_ID,
212 [DRIVES_FAN_RPM_INDEX] = {
215 .id = DRIVES_FAN_RPM_DEFAULT_ID,
217 [SLOTS_FAN_PWM_INDEX] = {
218 .loc = "SLOT,PCI FAN",
220 .id = SLOTS_FAN_PWM_DEFAULT_ID,
222 [CPUA_INTAKE_FAN_RPM_INDEX] = {
223 .loc = "CPU A INTAKE",
225 .id = CPUA_INTAKE_FAN_RPM_DEFAULT_ID,
227 [CPUA_EXHAUST_FAN_RPM_INDEX] = {
228 .loc = "CPU A EXHAUST",
230 .id = CPUA_EXHAUST_FAN_RPM_DEFAULT_ID,
232 [CPUB_INTAKE_FAN_RPM_INDEX] = {
233 .loc = "CPU B INTAKE",
235 .id = CPUB_INTAKE_FAN_RPM_DEFAULT_ID,
237 [CPUB_EXHAUST_FAN_RPM_INDEX] = {
238 .loc = "CPU B EXHAUST",
240 .id = CPUB_EXHAUST_FAN_RPM_DEFAULT_ID,
242 /* pumps aren't present by default, have to be looked up in the
245 [CPUA_PUMP_RPM_INDEX] = {
248 .id = FCU_FAN_ABSENT_ID,
250 [CPUB_PUMP_RPM_INDEX] = {
253 .id = FCU_FAN_ABSENT_ID,
256 [CPU_A1_FAN_RPM_INDEX] = {
259 .id = FCU_FAN_ABSENT_ID,
261 [CPU_A2_FAN_RPM_INDEX] = {
264 .id = FCU_FAN_ABSENT_ID,
266 [CPU_A3_FAN_RPM_INDEX] = {
269 .id = FCU_FAN_ABSENT_ID,
271 [CPU_B1_FAN_RPM_INDEX] = {
274 .id = FCU_FAN_ABSENT_ID,
276 [CPU_B2_FAN_RPM_INDEX] = {
279 .id = FCU_FAN_ABSENT_ID,
281 [CPU_B3_FAN_RPM_INDEX] = {
284 .id = FCU_FAN_ABSENT_ID,
289 * i2c_driver structure to attach to the host i2c controller
292 static int therm_pm72_attach(struct i2c_adapter *adapter);
293 static int therm_pm72_detach(struct i2c_adapter *adapter);
295 static struct i2c_driver therm_pm72_driver =
298 .name = "therm_pm72",
300 .attach_adapter = therm_pm72_attach,
301 .detach_adapter = therm_pm72_detach,
305 * Utility function to create an i2c_client structure and
306 * attach it to one of u3 adapters
308 static struct i2c_client *attach_i2c_chip(int id, const char *name)
310 struct i2c_client *clt;
311 struct i2c_adapter *adap;
322 clt = kmalloc(sizeof(struct i2c_client), GFP_KERNEL);
325 memset(clt, 0, sizeof(struct i2c_client));
327 clt->addr = (id >> 1) & 0x7f;
329 clt->driver = &therm_pm72_driver;
330 strncpy(clt->name, name, I2C_NAME_SIZE-1);
332 if (i2c_attach_client(clt)) {
333 printk(KERN_ERR "therm_pm72: Failed to attach to i2c ID 0x%x\n", id);
341 * Utility function to get rid of the i2c_client structure
342 * (will also detach from the adapter hopepfully)
344 static void detach_i2c_chip(struct i2c_client *clt)
346 i2c_detach_client(clt);
351 * Here are the i2c chip access wrappers
354 static void initialize_adc(struct cpu_pid_state *state)
359 /* Read ADC the configuration register and cache it. We
360 * also make sure Config2 contains proper values, I've seen
361 * cases where we got stale grabage in there, thus preventing
362 * proper reading of conv. values
368 i2c_master_send(state->monitor, buf, 2);
370 /* Read & cache Config1 */
372 rc = i2c_master_send(state->monitor, buf, 1);
374 rc = i2c_master_recv(state->monitor, buf, 1);
376 state->adc_config = buf[0];
377 DBG("ADC config reg: %02x\n", state->adc_config);
378 /* Disable shutdown mode */
379 state->adc_config &= 0xfe;
381 buf[1] = state->adc_config;
382 rc = i2c_master_send(state->monitor, buf, 2);
386 printk(KERN_ERR "therm_pm72: Error reading ADC config"
390 static int read_smon_adc(struct cpu_pid_state *state, int chan)
392 int rc, data, tries = 0;
398 buf[1] = (state->adc_config & 0x1f) | (chan << 5);
399 rc = i2c_master_send(state->monitor, buf, 2);
402 /* Wait for convertion */
404 /* Switch to data register */
406 rc = i2c_master_send(state->monitor, buf, 1);
410 rc = i2c_master_recv(state->monitor, buf, 2);
413 data = ((u16)buf[0]) << 8 | (u16)buf[1];
416 DBG("Error reading ADC, retrying...\n");
418 printk(KERN_ERR "therm_pm72: Error reading ADC !\n");
425 static int read_lm87_reg(struct i2c_client * chip, int reg)
433 rc = i2c_master_send(chip, &buf, 1);
436 rc = i2c_master_recv(chip, &buf, 1);
441 DBG("Error reading LM87, retrying...\n");
443 printk(KERN_ERR "therm_pm72: Error reading LM87 !\n");
450 static int fan_read_reg(int reg, unsigned char *buf, int nb)
457 nw = i2c_master_send(fcu, buf, 1);
458 if (nw > 0 || (nw < 0 && nw != -EIO) || tries >= 100)
464 printk(KERN_ERR "Failure writing address to FCU: %d", nw);
469 nr = i2c_master_recv(fcu, buf, nb);
470 if (nr > 0 || (nr < 0 && nr != ENODEV) || tries >= 100)
476 printk(KERN_ERR "Failure reading data from FCU: %d", nw);
480 static int fan_write_reg(int reg, const unsigned char *ptr, int nb)
483 unsigned char buf[16];
486 memcpy(buf+1, ptr, nb);
490 nw = i2c_master_send(fcu, buf, nb);
491 if (nw > 0 || (nw < 0 && nw != EIO) || tries >= 100)
497 printk(KERN_ERR "Failure writing to FCU: %d", nw);
501 static int start_fcu(void)
503 unsigned char buf = 0xff;
506 rc = fan_write_reg(0xe, &buf, 1);
509 rc = fan_write_reg(0x2e, &buf, 1);
512 rc = fan_read_reg(0, &buf, 1);
515 fcu_rpm_shift = (buf == 1) ? 2 : 3;
516 printk(KERN_DEBUG "FCU Initialized, RPM fan shift is %d\n",
522 static int set_rpm_fan(int fan_index, int rpm)
524 unsigned char buf[2];
525 int rc, id, min, max;
527 if (fcu_fans[fan_index].type != FCU_FAN_RPM)
529 id = fcu_fans[fan_index].id;
530 if (id == FCU_FAN_ABSENT_ID)
533 min = 2400 >> fcu_rpm_shift;
534 max = 56000 >> fcu_rpm_shift;
540 buf[0] = rpm >> (8 - fcu_rpm_shift);
541 buf[1] = rpm << fcu_rpm_shift;
542 rc = fan_write_reg(0x10 + (id * 2), buf, 2);
548 static int get_rpm_fan(int fan_index, int programmed)
550 unsigned char failure;
551 unsigned char active;
552 unsigned char buf[2];
553 int rc, id, reg_base;
555 if (fcu_fans[fan_index].type != FCU_FAN_RPM)
557 id = fcu_fans[fan_index].id;
558 if (id == FCU_FAN_ABSENT_ID)
561 rc = fan_read_reg(0xb, &failure, 1);
564 if ((failure & (1 << id)) != 0)
566 rc = fan_read_reg(0xd, &active, 1);
569 if ((active & (1 << id)) == 0)
572 /* Programmed value or real current speed */
573 reg_base = programmed ? 0x10 : 0x11;
574 rc = fan_read_reg(reg_base + (id * 2), buf, 2);
578 return (buf[0] << (8 - fcu_rpm_shift)) | buf[1] >> fcu_rpm_shift;
581 static int set_pwm_fan(int fan_index, int pwm)
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)
596 pwm = (pwm * 2559) / 1000;
598 rc = fan_write_reg(0x30 + (id * 2), buf, 1);
604 static int get_pwm_fan(int fan_index)
606 unsigned char failure;
607 unsigned char active;
608 unsigned char buf[2];
611 if (fcu_fans[fan_index].type != FCU_FAN_PWM)
613 id = fcu_fans[fan_index].id;
614 if (id == FCU_FAN_ABSENT_ID)
617 rc = fan_read_reg(0x2b, &failure, 1);
620 if ((failure & (1 << id)) != 0)
622 rc = fan_read_reg(0x2d, &active, 1);
625 if ((active & (1 << id)) == 0)
628 /* Programmed value or real current speed */
629 rc = fan_read_reg(0x30 + (id * 2), buf, 1);
633 return (buf[0] * 1000) / 2559;
636 static void tickle_fcu(void)
640 pwm = get_pwm_fan(SLOTS_FAN_PWM_INDEX);
642 DBG("FCU Tickle, slots fan is: %d\n", pwm);
647 pwm = SLOTS_FAN_DEFAULT_PWM;
648 } else if (pwm < SLOTS_PID_OUTPUT_MIN)
649 pwm = SLOTS_PID_OUTPUT_MIN;
651 /* That is hopefully enough to make the FCU happy */
652 set_pwm_fan(SLOTS_FAN_PWM_INDEX, pwm);
657 * Utility routine to read the CPU calibration EEPROM data
658 * from the device-tree
660 static int read_eeprom(int cpu, struct mpu_data *out)
662 struct device_node *np;
667 /* prom.c routine for finding a node by path is a bit brain dead
668 * and requires exact @xxx unit numbers. This is a bit ugly but
669 * will work for these machines
671 sprintf(nodename, "/u3@0,f8000000/i2c@f8001000/cpuid@a%d", cpu ? 2 : 0);
672 np = of_find_node_by_path(nodename);
674 printk(KERN_ERR "therm_pm72: Failed to retrieve cpuid node from device-tree\n");
677 data = of_get_property(np, "cpuid", &len);
679 printk(KERN_ERR "therm_pm72: Failed to retrieve cpuid property from device-tree\n");
683 memcpy(out, data, sizeof(struct mpu_data));
689 static void fetch_cpu_pumps_minmax(void)
691 struct cpu_pid_state *state0 = &cpu_state[0];
692 struct cpu_pid_state *state1 = &cpu_state[1];
693 u16 pump_min = 0, pump_max = 0xffff;
696 /* Try to fetch pumps min/max infos from eeprom */
698 memcpy(&tmp, &state0->mpu.processor_part_num, 8);
699 if (tmp[0] != 0xffff && tmp[1] != 0xffff) {
700 pump_min = max(pump_min, tmp[0]);
701 pump_max = min(pump_max, tmp[1]);
703 if (tmp[2] != 0xffff && tmp[3] != 0xffff) {
704 pump_min = max(pump_min, tmp[2]);
705 pump_max = min(pump_max, tmp[3]);
708 /* Double check the values, this _IS_ needed as the EEPROM on
709 * some dual 2.5Ghz G5s seem, at least, to have both min & max
710 * same to the same value ... (grrrr)
712 if (pump_min == pump_max || pump_min == 0 || pump_max == 0xffff) {
713 pump_min = CPU_PUMP_OUTPUT_MIN;
714 pump_max = CPU_PUMP_OUTPUT_MAX;
717 state0->pump_min = state1->pump_min = pump_min;
718 state0->pump_max = state1->pump_max = pump_max;
722 * Now, unfortunately, sysfs doesn't give us a nice void * we could
723 * pass around to the attribute functions, so we don't really have
724 * choice but implement a bunch of them...
726 * That sucks a bit, we take the lock because FIX32TOPRINT evaluates
727 * the input twice... I accept patches :)
729 #define BUILD_SHOW_FUNC_FIX(name, data) \
730 static ssize_t show_##name(struct device *dev, struct device_attribute *attr, char *buf) \
733 down(&driver_lock); \
734 r = sprintf(buf, "%d.%03d", FIX32TOPRINT(data)); \
738 #define BUILD_SHOW_FUNC_INT(name, data) \
739 static ssize_t show_##name(struct device *dev, struct device_attribute *attr, char *buf) \
741 return sprintf(buf, "%d", data); \
744 BUILD_SHOW_FUNC_FIX(cpu0_temperature, cpu_state[0].last_temp)
745 BUILD_SHOW_FUNC_FIX(cpu0_voltage, cpu_state[0].voltage)
746 BUILD_SHOW_FUNC_FIX(cpu0_current, cpu_state[0].current_a)
747 BUILD_SHOW_FUNC_INT(cpu0_exhaust_fan_rpm, cpu_state[0].rpm)
748 BUILD_SHOW_FUNC_INT(cpu0_intake_fan_rpm, cpu_state[0].intake_rpm)
750 BUILD_SHOW_FUNC_FIX(cpu1_temperature, cpu_state[1].last_temp)
751 BUILD_SHOW_FUNC_FIX(cpu1_voltage, cpu_state[1].voltage)
752 BUILD_SHOW_FUNC_FIX(cpu1_current, cpu_state[1].current_a)
753 BUILD_SHOW_FUNC_INT(cpu1_exhaust_fan_rpm, cpu_state[1].rpm)
754 BUILD_SHOW_FUNC_INT(cpu1_intake_fan_rpm, cpu_state[1].intake_rpm)
756 BUILD_SHOW_FUNC_FIX(backside_temperature, backside_state.last_temp)
757 BUILD_SHOW_FUNC_INT(backside_fan_pwm, backside_state.pwm)
759 BUILD_SHOW_FUNC_FIX(drives_temperature, drives_state.last_temp)
760 BUILD_SHOW_FUNC_INT(drives_fan_rpm, drives_state.rpm)
762 BUILD_SHOW_FUNC_FIX(slots_temperature, slots_state.last_temp)
763 BUILD_SHOW_FUNC_INT(slots_fan_pwm, slots_state.pwm)
765 BUILD_SHOW_FUNC_FIX(dimms_temperature, dimms_state.last_temp)
767 static DEVICE_ATTR(cpu0_temperature,S_IRUGO,show_cpu0_temperature,NULL);
768 static DEVICE_ATTR(cpu0_voltage,S_IRUGO,show_cpu0_voltage,NULL);
769 static DEVICE_ATTR(cpu0_current,S_IRUGO,show_cpu0_current,NULL);
770 static DEVICE_ATTR(cpu0_exhaust_fan_rpm,S_IRUGO,show_cpu0_exhaust_fan_rpm,NULL);
771 static DEVICE_ATTR(cpu0_intake_fan_rpm,S_IRUGO,show_cpu0_intake_fan_rpm,NULL);
773 static DEVICE_ATTR(cpu1_temperature,S_IRUGO,show_cpu1_temperature,NULL);
774 static DEVICE_ATTR(cpu1_voltage,S_IRUGO,show_cpu1_voltage,NULL);
775 static DEVICE_ATTR(cpu1_current,S_IRUGO,show_cpu1_current,NULL);
776 static DEVICE_ATTR(cpu1_exhaust_fan_rpm,S_IRUGO,show_cpu1_exhaust_fan_rpm,NULL);
777 static DEVICE_ATTR(cpu1_intake_fan_rpm,S_IRUGO,show_cpu1_intake_fan_rpm,NULL);
779 static DEVICE_ATTR(backside_temperature,S_IRUGO,show_backside_temperature,NULL);
780 static DEVICE_ATTR(backside_fan_pwm,S_IRUGO,show_backside_fan_pwm,NULL);
782 static DEVICE_ATTR(drives_temperature,S_IRUGO,show_drives_temperature,NULL);
783 static DEVICE_ATTR(drives_fan_rpm,S_IRUGO,show_drives_fan_rpm,NULL);
785 static DEVICE_ATTR(slots_temperature,S_IRUGO,show_slots_temperature,NULL);
786 static DEVICE_ATTR(slots_fan_pwm,S_IRUGO,show_slots_fan_pwm,NULL);
788 static DEVICE_ATTR(dimms_temperature,S_IRUGO,show_dimms_temperature,NULL);
791 * CPUs fans control loop
794 static int do_read_one_cpu_values(struct cpu_pid_state *state, s32 *temp, s32 *power)
796 s32 ltemp, volts, amps;
799 /* Default (in case of error) */
800 *temp = state->cur_temp;
801 *power = state->cur_power;
803 if (cpu_pid_type == CPU_PID_TYPE_RACKMAC)
804 index = (state->index == 0) ?
805 CPU_A1_FAN_RPM_INDEX : CPU_B1_FAN_RPM_INDEX;
807 index = (state->index == 0) ?
808 CPUA_EXHAUST_FAN_RPM_INDEX : CPUB_EXHAUST_FAN_RPM_INDEX;
810 /* Read current fan status */
811 rc = get_rpm_fan(index, !RPM_PID_USE_ACTUAL_SPEED);
813 /* XXX What do we do now ? Nothing for now, keep old value, but
814 * return error upstream
816 DBG(" cpu %d, fan reading error !\n", state->index);
819 DBG(" cpu %d, exhaust RPM: %d\n", state->index, state->rpm);
822 /* Get some sensor readings and scale it */
823 ltemp = read_smon_adc(state, 1);
825 /* XXX What do we do now ? */
829 DBG(" cpu %d, temp reading error !\n", state->index);
831 /* Fixup temperature according to diode calibration
833 DBG(" cpu %d, temp raw: %04x, m_diode: %04x, b_diode: %04x\n",
835 ltemp, state->mpu.mdiode, state->mpu.bdiode);
836 *temp = ((s32)ltemp * (s32)state->mpu.mdiode + ((s32)state->mpu.bdiode << 12)) >> 2;
837 state->last_temp = *temp;
838 DBG(" temp: %d.%03d\n", FIX32TOPRINT((*temp)));
842 * Read voltage & current and calculate power
844 volts = read_smon_adc(state, 3);
845 amps = read_smon_adc(state, 4);
847 /* Scale voltage and current raw sensor values according to fixed scales
848 * obtained in Darwin and calculate power from I and V
850 volts *= ADC_CPU_VOLTAGE_SCALE;
851 amps *= ADC_CPU_CURRENT_SCALE;
852 *power = (((u64)volts) * ((u64)amps)) >> 16;
853 state->voltage = volts;
854 state->current_a = amps;
855 state->last_power = *power;
857 DBG(" cpu %d, current: %d.%03d, voltage: %d.%03d, power: %d.%03d W\n",
858 state->index, FIX32TOPRINT(state->current_a),
859 FIX32TOPRINT(state->voltage), FIX32TOPRINT(*power));
864 static void do_cpu_pid(struct cpu_pid_state *state, s32 temp, s32 power)
866 s32 power_target, integral, derivative, proportional, adj_in_target, sval;
867 s64 integ_p, deriv_p, prop_p, sum;
870 /* Calculate power target value (could be done once for all)
871 * and convert to a 16.16 fp number
873 power_target = ((u32)(state->mpu.pmaxh - state->mpu.padjmax)) << 16;
874 DBG(" power target: %d.%03d, error: %d.%03d\n",
875 FIX32TOPRINT(power_target), FIX32TOPRINT(power_target - power));
877 /* Store temperature and power in history array */
878 state->cur_temp = (state->cur_temp + 1) % CPU_TEMP_HISTORY_SIZE;
879 state->temp_history[state->cur_temp] = temp;
880 state->cur_power = (state->cur_power + 1) % state->count_power;
881 state->power_history[state->cur_power] = power;
882 state->error_history[state->cur_power] = power_target - power;
884 /* If first loop, fill the history table */
886 for (i = 0; i < (state->count_power - 1); i++) {
887 state->cur_power = (state->cur_power + 1) % state->count_power;
888 state->power_history[state->cur_power] = power;
889 state->error_history[state->cur_power] = power_target - power;
891 for (i = 0; i < (CPU_TEMP_HISTORY_SIZE - 1); i++) {
892 state->cur_temp = (state->cur_temp + 1) % CPU_TEMP_HISTORY_SIZE;
893 state->temp_history[state->cur_temp] = temp;
898 /* Calculate the integral term normally based on the "power" values */
901 for (i = 0; i < state->count_power; i++)
902 integral += state->error_history[i];
903 integral *= CPU_PID_INTERVAL;
904 DBG(" integral: %08x\n", integral);
906 /* Calculate the adjusted input (sense value).
909 * so the result is 28.36
911 * input target is mpu.ttarget, input max is mpu.tmax
913 integ_p = ((s64)state->mpu.pid_gr) * (s64)integral;
914 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
915 sval = (state->mpu.tmax << 16) - ((integ_p >> 20) & 0xffffffff);
916 adj_in_target = (state->mpu.ttarget << 16);
917 if (adj_in_target > sval)
918 adj_in_target = sval;
919 DBG(" adj_in_target: %d.%03d, ttarget: %d\n", FIX32TOPRINT(adj_in_target),
922 /* Calculate the derivative term */
923 derivative = state->temp_history[state->cur_temp] -
924 state->temp_history[(state->cur_temp + CPU_TEMP_HISTORY_SIZE - 1)
925 % CPU_TEMP_HISTORY_SIZE];
926 derivative /= CPU_PID_INTERVAL;
927 deriv_p = ((s64)state->mpu.pid_gd) * (s64)derivative;
928 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
931 /* Calculate the proportional term */
932 proportional = temp - adj_in_target;
933 prop_p = ((s64)state->mpu.pid_gp) * (s64)proportional;
934 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
940 DBG(" sum: %d\n", (int)sum);
941 state->rpm += (s32)sum;
944 static void do_monitor_cpu_combined(void)
946 struct cpu_pid_state *state0 = &cpu_state[0];
947 struct cpu_pid_state *state1 = &cpu_state[1];
948 s32 temp0, power0, temp1, power1;
949 s32 temp_combi, power_combi;
950 int rc, intake, pump;
952 rc = do_read_one_cpu_values(state0, &temp0, &power0);
954 /* XXX What do we do now ? */
956 state1->overtemp = 0;
957 rc = do_read_one_cpu_values(state1, &temp1, &power1);
959 /* XXX What do we do now ? */
961 if (state1->overtemp)
964 temp_combi = max(temp0, temp1);
965 power_combi = max(power0, power1);
967 /* Check tmax, increment overtemp if we are there. At tmax+8, we go
968 * full blown immediately and try to trigger a shutdown
970 if (temp_combi >= ((state0->mpu.tmax + 8) << 16)) {
971 printk(KERN_WARNING "Warning ! Temperature way above maximum (%d) !\n",
973 state0->overtemp += CPU_MAX_OVERTEMP / 4;
974 } else if (temp_combi > (state0->mpu.tmax << 16))
977 state0->overtemp = 0;
978 if (state0->overtemp >= CPU_MAX_OVERTEMP)
980 if (state0->overtemp > 0) {
981 state0->rpm = state0->mpu.rmaxn_exhaust_fan;
982 state0->intake_rpm = intake = state0->mpu.rmaxn_intake_fan;
983 pump = state0->pump_max;
988 do_cpu_pid(state0, temp_combi, power_combi);
991 state0->rpm = max(state0->rpm, (int)state0->mpu.rminn_exhaust_fan);
992 state0->rpm = min(state0->rpm, (int)state0->mpu.rmaxn_exhaust_fan);
994 /* Calculate intake fan speed */
995 intake = (state0->rpm * CPU_INTAKE_SCALE) >> 16;
996 intake = max(intake, (int)state0->mpu.rminn_intake_fan);
997 intake = min(intake, (int)state0->mpu.rmaxn_intake_fan);
998 state0->intake_rpm = intake;
1000 /* Calculate pump speed */
1001 pump = (state0->rpm * state0->pump_max) /
1002 state0->mpu.rmaxn_exhaust_fan;
1003 pump = min(pump, state0->pump_max);
1004 pump = max(pump, state0->pump_min);
1007 /* We copy values from state 0 to state 1 for /sysfs */
1008 state1->rpm = state0->rpm;
1009 state1->intake_rpm = state0->intake_rpm;
1011 DBG("** CPU %d RPM: %d Ex, %d, Pump: %d, In, overtemp: %d\n",
1012 state1->index, (int)state1->rpm, intake, pump, state1->overtemp);
1014 /* We should check for errors, shouldn't we ? But then, what
1015 * do we do once the error occurs ? For FCU notified fan
1016 * failures (-EFAULT) we probably want to notify userland
1019 set_rpm_fan(CPUA_INTAKE_FAN_RPM_INDEX, intake);
1020 set_rpm_fan(CPUA_EXHAUST_FAN_RPM_INDEX, state0->rpm);
1021 set_rpm_fan(CPUB_INTAKE_FAN_RPM_INDEX, intake);
1022 set_rpm_fan(CPUB_EXHAUST_FAN_RPM_INDEX, state0->rpm);
1024 if (fcu_fans[CPUA_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID)
1025 set_rpm_fan(CPUA_PUMP_RPM_INDEX, pump);
1026 if (fcu_fans[CPUB_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID)
1027 set_rpm_fan(CPUB_PUMP_RPM_INDEX, pump);
1030 static void do_monitor_cpu_split(struct cpu_pid_state *state)
1035 /* Read current fan status */
1036 rc = do_read_one_cpu_values(state, &temp, &power);
1038 /* XXX What do we do now ? */
1041 /* Check tmax, increment overtemp if we are there. At tmax+8, we go
1042 * full blown immediately and try to trigger a shutdown
1044 if (temp >= ((state->mpu.tmax + 8) << 16)) {
1045 printk(KERN_WARNING "Warning ! CPU %d temperature way above maximum"
1047 state->index, temp >> 16);
1048 state->overtemp += CPU_MAX_OVERTEMP / 4;
1049 } else if (temp > (state->mpu.tmax << 16))
1052 state->overtemp = 0;
1053 if (state->overtemp >= CPU_MAX_OVERTEMP)
1055 if (state->overtemp > 0) {
1056 state->rpm = state->mpu.rmaxn_exhaust_fan;
1057 state->intake_rpm = intake = state->mpu.rmaxn_intake_fan;
1062 do_cpu_pid(state, temp, power);
1065 state->rpm = max(state->rpm, (int)state->mpu.rminn_exhaust_fan);
1066 state->rpm = min(state->rpm, (int)state->mpu.rmaxn_exhaust_fan);
1068 /* Calculate intake fan */
1069 intake = (state->rpm * CPU_INTAKE_SCALE) >> 16;
1070 intake = max(intake, (int)state->mpu.rminn_intake_fan);
1071 intake = min(intake, (int)state->mpu.rmaxn_intake_fan);
1072 state->intake_rpm = intake;
1075 DBG("** CPU %d RPM: %d Ex, %d In, overtemp: %d\n",
1076 state->index, (int)state->rpm, intake, state->overtemp);
1078 /* We should check for errors, shouldn't we ? But then, what
1079 * do we do once the error occurs ? For FCU notified fan
1080 * failures (-EFAULT) we probably want to notify userland
1083 if (state->index == 0) {
1084 set_rpm_fan(CPUA_INTAKE_FAN_RPM_INDEX, intake);
1085 set_rpm_fan(CPUA_EXHAUST_FAN_RPM_INDEX, state->rpm);
1087 set_rpm_fan(CPUB_INTAKE_FAN_RPM_INDEX, intake);
1088 set_rpm_fan(CPUB_EXHAUST_FAN_RPM_INDEX, state->rpm);
1092 static void do_monitor_cpu_rack(struct cpu_pid_state *state)
1094 s32 temp, power, fan_min;
1097 /* Read current fan status */
1098 rc = do_read_one_cpu_values(state, &temp, &power);
1100 /* XXX What do we do now ? */
1103 /* Check tmax, increment overtemp if we are there. At tmax+8, we go
1104 * full blown immediately and try to trigger a shutdown
1106 if (temp >= ((state->mpu.tmax + 8) << 16)) {
1107 printk(KERN_WARNING "Warning ! CPU %d temperature way above maximum"
1109 state->index, temp >> 16);
1110 state->overtemp = CPU_MAX_OVERTEMP / 4;
1111 } else if (temp > (state->mpu.tmax << 16))
1114 state->overtemp = 0;
1115 if (state->overtemp >= CPU_MAX_OVERTEMP)
1117 if (state->overtemp > 0) {
1118 state->rpm = state->intake_rpm = state->mpu.rmaxn_intake_fan;
1123 do_cpu_pid(state, temp, power);
1125 /* Check clamp from dimms */
1126 fan_min = dimm_output_clamp;
1127 fan_min = max(fan_min, (int)state->mpu.rminn_intake_fan);
1129 DBG(" CPU min mpu = %d, min dimm = %d\n",
1130 state->mpu.rminn_intake_fan, dimm_output_clamp);
1132 state->rpm = max(state->rpm, (int)fan_min);
1133 state->rpm = min(state->rpm, (int)state->mpu.rmaxn_intake_fan);
1134 state->intake_rpm = state->rpm;
1137 DBG("** CPU %d RPM: %d overtemp: %d\n",
1138 state->index, (int)state->rpm, state->overtemp);
1140 /* We should check for errors, shouldn't we ? But then, what
1141 * do we do once the error occurs ? For FCU notified fan
1142 * failures (-EFAULT) we probably want to notify userland
1145 if (state->index == 0) {
1146 set_rpm_fan(CPU_A1_FAN_RPM_INDEX, state->rpm);
1147 set_rpm_fan(CPU_A2_FAN_RPM_INDEX, state->rpm);
1148 set_rpm_fan(CPU_A3_FAN_RPM_INDEX, state->rpm);
1150 set_rpm_fan(CPU_B1_FAN_RPM_INDEX, state->rpm);
1151 set_rpm_fan(CPU_B2_FAN_RPM_INDEX, state->rpm);
1152 set_rpm_fan(CPU_B3_FAN_RPM_INDEX, state->rpm);
1157 * Initialize the state structure for one CPU control loop
1159 static int init_cpu_state(struct cpu_pid_state *state, int index)
1161 state->index = index;
1163 state->rpm = (cpu_pid_type == CPU_PID_TYPE_RACKMAC) ? 4000 : 1000;
1164 state->overtemp = 0;
1165 state->adc_config = 0x00;
1169 state->monitor = attach_i2c_chip(SUPPLY_MONITOR_ID, "CPU0_monitor");
1170 else if (index == 1)
1171 state->monitor = attach_i2c_chip(SUPPLY_MONITORB_ID, "CPU1_monitor");
1172 if (state->monitor == NULL)
1175 if (read_eeprom(index, &state->mpu))
1178 state->count_power = state->mpu.tguardband;
1179 if (state->count_power > CPU_POWER_HISTORY_SIZE) {
1180 printk(KERN_WARNING "Warning ! too many power history slots\n");
1181 state->count_power = CPU_POWER_HISTORY_SIZE;
1183 DBG("CPU %d Using %d power history entries\n", index, state->count_power);
1186 device_create_file(&of_dev->dev, &dev_attr_cpu0_temperature);
1187 device_create_file(&of_dev->dev, &dev_attr_cpu0_voltage);
1188 device_create_file(&of_dev->dev, &dev_attr_cpu0_current);
1189 device_create_file(&of_dev->dev, &dev_attr_cpu0_exhaust_fan_rpm);
1190 device_create_file(&of_dev->dev, &dev_attr_cpu0_intake_fan_rpm);
1192 device_create_file(&of_dev->dev, &dev_attr_cpu1_temperature);
1193 device_create_file(&of_dev->dev, &dev_attr_cpu1_voltage);
1194 device_create_file(&of_dev->dev, &dev_attr_cpu1_current);
1195 device_create_file(&of_dev->dev, &dev_attr_cpu1_exhaust_fan_rpm);
1196 device_create_file(&of_dev->dev, &dev_attr_cpu1_intake_fan_rpm);
1202 detach_i2c_chip(state->monitor);
1203 state->monitor = NULL;
1209 * Dispose of the state data for one CPU control loop
1211 static void dispose_cpu_state(struct cpu_pid_state *state)
1213 if (state->monitor == NULL)
1216 if (state->index == 0) {
1217 device_remove_file(&of_dev->dev, &dev_attr_cpu0_temperature);
1218 device_remove_file(&of_dev->dev, &dev_attr_cpu0_voltage);
1219 device_remove_file(&of_dev->dev, &dev_attr_cpu0_current);
1220 device_remove_file(&of_dev->dev, &dev_attr_cpu0_exhaust_fan_rpm);
1221 device_remove_file(&of_dev->dev, &dev_attr_cpu0_intake_fan_rpm);
1223 device_remove_file(&of_dev->dev, &dev_attr_cpu1_temperature);
1224 device_remove_file(&of_dev->dev, &dev_attr_cpu1_voltage);
1225 device_remove_file(&of_dev->dev, &dev_attr_cpu1_current);
1226 device_remove_file(&of_dev->dev, &dev_attr_cpu1_exhaust_fan_rpm);
1227 device_remove_file(&of_dev->dev, &dev_attr_cpu1_intake_fan_rpm);
1230 detach_i2c_chip(state->monitor);
1231 state->monitor = NULL;
1235 * Motherboard backside & U3 heatsink fan control loop
1237 static void do_monitor_backside(struct backside_pid_state *state)
1239 s32 temp, integral, derivative, fan_min;
1240 s64 integ_p, deriv_p, prop_p, sum;
1243 if (--state->ticks != 0)
1245 state->ticks = backside_params.interval;
1249 /* Check fan status */
1250 rc = get_pwm_fan(BACKSIDE_FAN_PWM_INDEX);
1252 printk(KERN_WARNING "Error %d reading backside fan !\n", rc);
1253 /* XXX What do we do now ? */
1256 DBG(" current pwm: %d\n", state->pwm);
1258 /* Get some sensor readings */
1259 temp = i2c_smbus_read_byte_data(state->monitor, MAX6690_EXT_TEMP) << 16;
1260 state->last_temp = temp;
1261 DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1262 FIX32TOPRINT(backside_params.input_target));
1264 /* Store temperature and error in history array */
1265 state->cur_sample = (state->cur_sample + 1) % BACKSIDE_PID_HISTORY_SIZE;
1266 state->sample_history[state->cur_sample] = temp;
1267 state->error_history[state->cur_sample] = temp - backside_params.input_target;
1269 /* If first loop, fill the history table */
1271 for (i = 0; i < (BACKSIDE_PID_HISTORY_SIZE - 1); i++) {
1272 state->cur_sample = (state->cur_sample + 1) %
1273 BACKSIDE_PID_HISTORY_SIZE;
1274 state->sample_history[state->cur_sample] = temp;
1275 state->error_history[state->cur_sample] =
1276 temp - backside_params.input_target;
1281 /* Calculate the integral term */
1284 for (i = 0; i < BACKSIDE_PID_HISTORY_SIZE; i++)
1285 integral += state->error_history[i];
1286 integral *= backside_params.interval;
1287 DBG(" integral: %08x\n", integral);
1288 integ_p = ((s64)backside_params.G_r) * (s64)integral;
1289 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1292 /* Calculate the derivative term */
1293 derivative = state->error_history[state->cur_sample] -
1294 state->error_history[(state->cur_sample + BACKSIDE_PID_HISTORY_SIZE - 1)
1295 % BACKSIDE_PID_HISTORY_SIZE];
1296 derivative /= backside_params.interval;
1297 deriv_p = ((s64)backside_params.G_d) * (s64)derivative;
1298 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1301 /* Calculate the proportional term */
1302 prop_p = ((s64)backside_params.G_p) * (s64)(state->error_history[state->cur_sample]);
1303 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1309 DBG(" sum: %d\n", (int)sum);
1310 if (backside_params.additive)
1311 state->pwm += (s32)sum;
1315 /* Check for clamp */
1316 fan_min = (dimm_output_clamp * 100) / 14000;
1317 fan_min = max(fan_min, backside_params.output_min);
1319 state->pwm = max(state->pwm, fan_min);
1320 state->pwm = min(state->pwm, backside_params.output_max);
1322 DBG("** BACKSIDE PWM: %d\n", (int)state->pwm);
1323 set_pwm_fan(BACKSIDE_FAN_PWM_INDEX, state->pwm);
1327 * Initialize the state structure for the backside fan control loop
1329 static int init_backside_state(struct backside_pid_state *state)
1331 struct device_node *u3;
1332 int u3h = 1; /* conservative by default */
1335 * There are different PID params for machines with U3 and machines
1336 * with U3H, pick the right ones now
1338 u3 = of_find_node_by_path("/u3@0,f8000000");
1340 const u32 *vers = of_get_property(u3, "device-rev", NULL);
1342 if (((*vers) & 0x3f) < 0x34)
1348 backside_params.G_d = BACKSIDE_PID_RACK_G_d;
1349 backside_params.input_target = BACKSIDE_PID_RACK_INPUT_TARGET;
1350 backside_params.output_min = BACKSIDE_PID_U3H_OUTPUT_MIN;
1351 backside_params.interval = BACKSIDE_PID_RACK_INTERVAL;
1352 backside_params.G_p = BACKSIDE_PID_RACK_G_p;
1353 backside_params.G_r = BACKSIDE_PID_G_r;
1354 backside_params.output_max = BACKSIDE_PID_OUTPUT_MAX;
1355 backside_params.additive = 0;
1357 backside_params.G_d = BACKSIDE_PID_U3H_G_d;
1358 backside_params.input_target = BACKSIDE_PID_U3H_INPUT_TARGET;
1359 backside_params.output_min = BACKSIDE_PID_U3H_OUTPUT_MIN;
1360 backside_params.interval = BACKSIDE_PID_INTERVAL;
1361 backside_params.G_p = BACKSIDE_PID_G_p;
1362 backside_params.G_r = BACKSIDE_PID_G_r;
1363 backside_params.output_max = BACKSIDE_PID_OUTPUT_MAX;
1364 backside_params.additive = 1;
1366 backside_params.G_d = BACKSIDE_PID_U3_G_d;
1367 backside_params.input_target = BACKSIDE_PID_U3_INPUT_TARGET;
1368 backside_params.output_min = BACKSIDE_PID_U3_OUTPUT_MIN;
1369 backside_params.interval = BACKSIDE_PID_INTERVAL;
1370 backside_params.G_p = BACKSIDE_PID_G_p;
1371 backside_params.G_r = BACKSIDE_PID_G_r;
1372 backside_params.output_max = BACKSIDE_PID_OUTPUT_MAX;
1373 backside_params.additive = 1;
1380 state->monitor = attach_i2c_chip(BACKSIDE_MAX_ID, "backside_temp");
1381 if (state->monitor == NULL)
1384 device_create_file(&of_dev->dev, &dev_attr_backside_temperature);
1385 device_create_file(&of_dev->dev, &dev_attr_backside_fan_pwm);
1391 * Dispose of the state data for the backside control loop
1393 static void dispose_backside_state(struct backside_pid_state *state)
1395 if (state->monitor == NULL)
1398 device_remove_file(&of_dev->dev, &dev_attr_backside_temperature);
1399 device_remove_file(&of_dev->dev, &dev_attr_backside_fan_pwm);
1401 detach_i2c_chip(state->monitor);
1402 state->monitor = NULL;
1406 * Drives bay fan control loop
1408 static void do_monitor_drives(struct drives_pid_state *state)
1410 s32 temp, integral, derivative;
1411 s64 integ_p, deriv_p, prop_p, sum;
1414 if (--state->ticks != 0)
1416 state->ticks = DRIVES_PID_INTERVAL;
1420 /* Check fan status */
1421 rc = get_rpm_fan(DRIVES_FAN_RPM_INDEX, !RPM_PID_USE_ACTUAL_SPEED);
1423 printk(KERN_WARNING "Error %d reading drives fan !\n", rc);
1424 /* XXX What do we do now ? */
1427 DBG(" current rpm: %d\n", state->rpm);
1429 /* Get some sensor readings */
1430 temp = le16_to_cpu(i2c_smbus_read_word_data(state->monitor,
1432 state->last_temp = temp;
1433 DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1434 FIX32TOPRINT(DRIVES_PID_INPUT_TARGET));
1436 /* Store temperature and error in history array */
1437 state->cur_sample = (state->cur_sample + 1) % DRIVES_PID_HISTORY_SIZE;
1438 state->sample_history[state->cur_sample] = temp;
1439 state->error_history[state->cur_sample] = temp - DRIVES_PID_INPUT_TARGET;
1441 /* If first loop, fill the history table */
1443 for (i = 0; i < (DRIVES_PID_HISTORY_SIZE - 1); i++) {
1444 state->cur_sample = (state->cur_sample + 1) %
1445 DRIVES_PID_HISTORY_SIZE;
1446 state->sample_history[state->cur_sample] = temp;
1447 state->error_history[state->cur_sample] =
1448 temp - DRIVES_PID_INPUT_TARGET;
1453 /* Calculate the integral term */
1456 for (i = 0; i < DRIVES_PID_HISTORY_SIZE; i++)
1457 integral += state->error_history[i];
1458 integral *= DRIVES_PID_INTERVAL;
1459 DBG(" integral: %08x\n", integral);
1460 integ_p = ((s64)DRIVES_PID_G_r) * (s64)integral;
1461 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1464 /* Calculate the derivative term */
1465 derivative = state->error_history[state->cur_sample] -
1466 state->error_history[(state->cur_sample + DRIVES_PID_HISTORY_SIZE - 1)
1467 % DRIVES_PID_HISTORY_SIZE];
1468 derivative /= DRIVES_PID_INTERVAL;
1469 deriv_p = ((s64)DRIVES_PID_G_d) * (s64)derivative;
1470 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1473 /* Calculate the proportional term */
1474 prop_p = ((s64)DRIVES_PID_G_p) * (s64)(state->error_history[state->cur_sample]);
1475 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1481 DBG(" sum: %d\n", (int)sum);
1482 state->rpm += (s32)sum;
1484 state->rpm = max(state->rpm, DRIVES_PID_OUTPUT_MIN);
1485 state->rpm = min(state->rpm, DRIVES_PID_OUTPUT_MAX);
1487 DBG("** DRIVES RPM: %d\n", (int)state->rpm);
1488 set_rpm_fan(DRIVES_FAN_RPM_INDEX, state->rpm);
1492 * Initialize the state structure for the drives bay fan control loop
1494 static int init_drives_state(struct drives_pid_state *state)
1500 state->monitor = attach_i2c_chip(DRIVES_DALLAS_ID, "drives_temp");
1501 if (state->monitor == NULL)
1504 device_create_file(&of_dev->dev, &dev_attr_drives_temperature);
1505 device_create_file(&of_dev->dev, &dev_attr_drives_fan_rpm);
1511 * Dispose of the state data for the drives control loop
1513 static void dispose_drives_state(struct drives_pid_state *state)
1515 if (state->monitor == NULL)
1518 device_remove_file(&of_dev->dev, &dev_attr_drives_temperature);
1519 device_remove_file(&of_dev->dev, &dev_attr_drives_fan_rpm);
1521 detach_i2c_chip(state->monitor);
1522 state->monitor = NULL;
1526 * DIMMs temp control loop
1528 static void do_monitor_dimms(struct dimm_pid_state *state)
1530 s32 temp, integral, derivative, fan_min;
1531 s64 integ_p, deriv_p, prop_p, sum;
1534 if (--state->ticks != 0)
1536 state->ticks = DIMM_PID_INTERVAL;
1540 DBG(" current value: %d\n", state->output);
1542 temp = read_lm87_reg(state->monitor, LM87_INT_TEMP);
1546 state->last_temp = temp;
1547 DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1548 FIX32TOPRINT(DIMM_PID_INPUT_TARGET));
1550 /* Store temperature and error in history array */
1551 state->cur_sample = (state->cur_sample + 1) % DIMM_PID_HISTORY_SIZE;
1552 state->sample_history[state->cur_sample] = temp;
1553 state->error_history[state->cur_sample] = temp - DIMM_PID_INPUT_TARGET;
1555 /* If first loop, fill the history table */
1557 for (i = 0; i < (DIMM_PID_HISTORY_SIZE - 1); i++) {
1558 state->cur_sample = (state->cur_sample + 1) %
1559 DIMM_PID_HISTORY_SIZE;
1560 state->sample_history[state->cur_sample] = temp;
1561 state->error_history[state->cur_sample] =
1562 temp - DIMM_PID_INPUT_TARGET;
1567 /* Calculate the integral term */
1570 for (i = 0; i < DIMM_PID_HISTORY_SIZE; i++)
1571 integral += state->error_history[i];
1572 integral *= DIMM_PID_INTERVAL;
1573 DBG(" integral: %08x\n", integral);
1574 integ_p = ((s64)DIMM_PID_G_r) * (s64)integral;
1575 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1578 /* Calculate the derivative term */
1579 derivative = state->error_history[state->cur_sample] -
1580 state->error_history[(state->cur_sample + DIMM_PID_HISTORY_SIZE - 1)
1581 % DIMM_PID_HISTORY_SIZE];
1582 derivative /= DIMM_PID_INTERVAL;
1583 deriv_p = ((s64)DIMM_PID_G_d) * (s64)derivative;
1584 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1587 /* Calculate the proportional term */
1588 prop_p = ((s64)DIMM_PID_G_p) * (s64)(state->error_history[state->cur_sample]);
1589 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1595 DBG(" sum: %d\n", (int)sum);
1596 state->output = (s32)sum;
1597 state->output = max(state->output, DIMM_PID_OUTPUT_MIN);
1598 state->output = min(state->output, DIMM_PID_OUTPUT_MAX);
1599 dimm_output_clamp = state->output;
1601 DBG("** DIMM clamp value: %d\n", (int)state->output);
1603 /* Backside PID is only every 5 seconds, force backside fan clamping now */
1604 fan_min = (dimm_output_clamp * 100) / 14000;
1605 fan_min = max(fan_min, backside_params.output_min);
1606 if (backside_state.pwm < fan_min) {
1607 backside_state.pwm = fan_min;
1608 DBG(" -> applying clamp to backside fan now: %d !\n", fan_min);
1609 set_pwm_fan(BACKSIDE_FAN_PWM_INDEX, fan_min);
1614 * Initialize the state structure for the DIMM temp control loop
1616 static int init_dimms_state(struct dimm_pid_state *state)
1620 state->output = 4000;
1622 state->monitor = attach_i2c_chip(XSERVE_DIMMS_LM87, "dimms_temp");
1623 if (state->monitor == NULL)
1626 device_create_file(&of_dev->dev, &dev_attr_dimms_temperature);
1632 * Dispose of the state data for the DIMM control loop
1634 static void dispose_dimms_state(struct dimm_pid_state *state)
1636 if (state->monitor == NULL)
1639 device_remove_file(&of_dev->dev, &dev_attr_dimms_temperature);
1641 detach_i2c_chip(state->monitor);
1642 state->monitor = NULL;
1646 * Slots fan control loop
1648 static void do_monitor_slots(struct slots_pid_state *state)
1650 s32 temp, integral, derivative;
1651 s64 integ_p, deriv_p, prop_p, sum;
1654 if (--state->ticks != 0)
1656 state->ticks = SLOTS_PID_INTERVAL;
1660 /* Check fan status */
1661 rc = get_pwm_fan(SLOTS_FAN_PWM_INDEX);
1663 printk(KERN_WARNING "Error %d reading slots fan !\n", rc);
1664 /* XXX What do we do now ? */
1667 DBG(" current pwm: %d\n", state->pwm);
1669 /* Get some sensor readings */
1670 temp = le16_to_cpu(i2c_smbus_read_word_data(state->monitor,
1672 state->last_temp = temp;
1673 DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1674 FIX32TOPRINT(SLOTS_PID_INPUT_TARGET));
1676 /* Store temperature and error in history array */
1677 state->cur_sample = (state->cur_sample + 1) % SLOTS_PID_HISTORY_SIZE;
1678 state->sample_history[state->cur_sample] = temp;
1679 state->error_history[state->cur_sample] = temp - SLOTS_PID_INPUT_TARGET;
1681 /* If first loop, fill the history table */
1683 for (i = 0; i < (SLOTS_PID_HISTORY_SIZE - 1); i++) {
1684 state->cur_sample = (state->cur_sample + 1) %
1685 SLOTS_PID_HISTORY_SIZE;
1686 state->sample_history[state->cur_sample] = temp;
1687 state->error_history[state->cur_sample] =
1688 temp - SLOTS_PID_INPUT_TARGET;
1693 /* Calculate the integral term */
1696 for (i = 0; i < SLOTS_PID_HISTORY_SIZE; i++)
1697 integral += state->error_history[i];
1698 integral *= SLOTS_PID_INTERVAL;
1699 DBG(" integral: %08x\n", integral);
1700 integ_p = ((s64)SLOTS_PID_G_r) * (s64)integral;
1701 DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1704 /* Calculate the derivative term */
1705 derivative = state->error_history[state->cur_sample] -
1706 state->error_history[(state->cur_sample + SLOTS_PID_HISTORY_SIZE - 1)
1707 % SLOTS_PID_HISTORY_SIZE];
1708 derivative /= SLOTS_PID_INTERVAL;
1709 deriv_p = ((s64)SLOTS_PID_G_d) * (s64)derivative;
1710 DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1713 /* Calculate the proportional term */
1714 prop_p = ((s64)SLOTS_PID_G_p) * (s64)(state->error_history[state->cur_sample]);
1715 DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1721 DBG(" sum: %d\n", (int)sum);
1722 state->pwm = (s32)sum;
1724 state->pwm = max(state->pwm, SLOTS_PID_OUTPUT_MIN);
1725 state->pwm = min(state->pwm, SLOTS_PID_OUTPUT_MAX);
1727 DBG("** DRIVES PWM: %d\n", (int)state->pwm);
1728 set_pwm_fan(SLOTS_FAN_PWM_INDEX, state->pwm);
1732 * Initialize the state structure for the slots bay fan control loop
1734 static int init_slots_state(struct slots_pid_state *state)
1740 state->monitor = attach_i2c_chip(XSERVE_SLOTS_LM75, "slots_temp");
1741 if (state->monitor == NULL)
1744 device_create_file(&of_dev->dev, &dev_attr_slots_temperature);
1745 device_create_file(&of_dev->dev, &dev_attr_slots_fan_pwm);
1751 * Dispose of the state data for the slots control loop
1753 static void dispose_slots_state(struct slots_pid_state *state)
1755 if (state->monitor == NULL)
1758 device_remove_file(&of_dev->dev, &dev_attr_slots_temperature);
1759 device_remove_file(&of_dev->dev, &dev_attr_slots_fan_pwm);
1761 detach_i2c_chip(state->monitor);
1762 state->monitor = NULL;
1766 static int call_critical_overtemp(void)
1768 char *argv[] = { critical_overtemp_path, NULL };
1769 static char *envp[] = { "HOME=/",
1771 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
1774 return call_usermodehelper(critical_overtemp_path, argv, envp, 0);
1779 * Here's the kernel thread that calls the various control loops
1781 static int main_control_loop(void *x)
1785 DBG("main_control_loop started\n");
1789 if (start_fcu() < 0) {
1790 printk(KERN_ERR "kfand: failed to start FCU\n");
1795 /* Set the PCI fan once for now on non-RackMac */
1797 set_pwm_fan(SLOTS_FAN_PWM_INDEX, SLOTS_FAN_DEFAULT_PWM);
1799 /* Initialize ADCs */
1800 initialize_adc(&cpu_state[0]);
1801 if (cpu_state[1].monitor != NULL)
1802 initialize_adc(&cpu_state[1]);
1804 fcu_tickle_ticks = FCU_TICKLE_TICKS;
1808 while (state == state_attached) {
1809 unsigned long elapsed, start;
1815 /* Tickle the FCU just in case */
1816 if (--fcu_tickle_ticks < 0) {
1817 fcu_tickle_ticks = FCU_TICKLE_TICKS;
1821 /* First, we always calculate the new DIMMs state on an Xserve */
1823 do_monitor_dimms(&dimms_state);
1825 /* Then, the CPUs */
1826 if (cpu_pid_type == CPU_PID_TYPE_COMBINED)
1827 do_monitor_cpu_combined();
1828 else if (cpu_pid_type == CPU_PID_TYPE_RACKMAC) {
1829 do_monitor_cpu_rack(&cpu_state[0]);
1830 if (cpu_state[1].monitor != NULL)
1831 do_monitor_cpu_rack(&cpu_state[1]);
1832 // better deal with UP
1834 do_monitor_cpu_split(&cpu_state[0]);
1835 if (cpu_state[1].monitor != NULL)
1836 do_monitor_cpu_split(&cpu_state[1]);
1837 // better deal with UP
1839 /* Then, the rest */
1840 do_monitor_backside(&backside_state);
1842 do_monitor_slots(&slots_state);
1844 do_monitor_drives(&drives_state);
1847 if (critical_state == 1) {
1848 printk(KERN_WARNING "Temperature control detected a critical condition\n");
1849 printk(KERN_WARNING "Attempting to shut down...\n");
1850 if (call_critical_overtemp()) {
1851 printk(KERN_WARNING "Can't call %s, power off now!\n",
1852 critical_overtemp_path);
1853 machine_power_off();
1856 if (critical_state > 0)
1858 if (critical_state > MAX_CRITICAL_STATE) {
1859 printk(KERN_WARNING "Shutdown timed out, power off now !\n");
1860 machine_power_off();
1863 // FIXME: Deal with signals
1864 elapsed = jiffies - start;
1866 schedule_timeout_interruptible(HZ - elapsed);
1870 DBG("main_control_loop ended\n");
1873 complete_and_exit(&ctrl_complete, 0);
1877 * Dispose the control loops when tearing down
1879 static void dispose_control_loops(void)
1881 dispose_cpu_state(&cpu_state[0]);
1882 dispose_cpu_state(&cpu_state[1]);
1883 dispose_backside_state(&backside_state);
1884 dispose_drives_state(&drives_state);
1885 dispose_slots_state(&slots_state);
1886 dispose_dimms_state(&dimms_state);
1890 * Create the control loops. U3-0 i2c bus is up, so we can now
1891 * get to the various sensors
1893 static int create_control_loops(void)
1895 struct device_node *np;
1897 /* Count CPUs from the device-tree, we don't care how many are
1898 * actually used by Linux
1901 for (np = NULL; NULL != (np = of_find_node_by_type(np, "cpu"));)
1904 DBG("counted %d CPUs in the device-tree\n", cpu_count);
1906 /* Decide the type of PID algorithm to use based on the presence of
1907 * the pumps, though that may not be the best way, that is good enough
1911 cpu_pid_type = CPU_PID_TYPE_RACKMAC;
1912 else if (machine_is_compatible("PowerMac7,3")
1914 && fcu_fans[CPUA_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID
1915 && fcu_fans[CPUB_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID) {
1916 printk(KERN_INFO "Liquid cooling pumps detected, using new algorithm !\n");
1917 cpu_pid_type = CPU_PID_TYPE_COMBINED;
1919 cpu_pid_type = CPU_PID_TYPE_SPLIT;
1921 /* Create control loops for everything. If any fail, everything
1924 if (init_cpu_state(&cpu_state[0], 0))
1926 if (cpu_pid_type == CPU_PID_TYPE_COMBINED)
1927 fetch_cpu_pumps_minmax();
1929 if (cpu_count > 1 && init_cpu_state(&cpu_state[1], 1))
1931 if (init_backside_state(&backside_state))
1933 if (rackmac && init_dimms_state(&dimms_state))
1935 if (rackmac && init_slots_state(&slots_state))
1937 if (!rackmac && init_drives_state(&drives_state))
1940 DBG("all control loops up !\n");
1945 DBG("failure creating control loops, disposing\n");
1947 dispose_control_loops();
1953 * Start the control loops after everything is up, that is create
1954 * the thread that will make them run
1956 static void start_control_loops(void)
1958 init_completion(&ctrl_complete);
1960 ctrl_task = kernel_thread(main_control_loop, NULL, SIGCHLD | CLONE_KERNEL);
1964 * Stop the control loops when tearing down
1966 static void stop_control_loops(void)
1969 wait_for_completion(&ctrl_complete);
1973 * Attach to the i2c FCU after detecting U3-1 bus
1975 static int attach_fcu(void)
1977 fcu = attach_i2c_chip(FAN_CTRLER_ID, "fcu");
1981 DBG("FCU attached\n");
1987 * Detach from the i2c FCU when tearing down
1989 static void detach_fcu(void)
1992 detach_i2c_chip(fcu);
1997 * Attach to the i2c controller. We probe the various chips based
1998 * on the device-tree nodes and build everything for the driver to
1999 * run, we then kick the driver monitoring thread
2001 static int therm_pm72_attach(struct i2c_adapter *adapter)
2006 if (state == state_detached)
2007 state = state_attaching;
2008 if (state != state_attaching) {
2013 /* Check if we are looking for one of these */
2014 if (u3_0 == NULL && !strcmp(adapter->name, "u3 0")) {
2016 DBG("found U3-0\n");
2018 if (create_control_loops())
2020 } else if (u3_1 == NULL && !strcmp(adapter->name, "u3 1")) {
2022 DBG("found U3-1, attaching FCU\n");
2025 } else if (k2 == NULL && !strcmp(adapter->name, "mac-io 0")) {
2028 if (u3_0 && rackmac)
2029 if (create_control_loops())
2032 /* We got all we need, start control loops */
2033 if (u3_0 != NULL && u3_1 != NULL && (k2 || !rackmac)) {
2034 DBG("everything up, starting control loops\n");
2035 state = state_attached;
2036 start_control_loops();
2044 * Called on every adapter when the driver or the i2c controller
2047 static int therm_pm72_detach(struct i2c_adapter *adapter)
2051 if (state != state_detached)
2052 state = state_detaching;
2054 /* Stop control loops if any */
2055 DBG("stopping control loops\n");
2057 stop_control_loops();
2060 if (u3_0 != NULL && !strcmp(adapter->name, "u3 0")) {
2061 DBG("lost U3-0, disposing control loops\n");
2062 dispose_control_loops();
2066 if (u3_1 != NULL && !strcmp(adapter->name, "u3 1")) {
2067 DBG("lost U3-1, detaching FCU\n");
2071 if (u3_0 == NULL && u3_1 == NULL)
2072 state = state_detached;
2079 static int fan_check_loc_match(const char *loc, int fan)
2084 strlcpy(tmp, fcu_fans[fan].loc, 64);
2091 if (strcmp(loc, c) == 0)
2100 static void fcu_lookup_fans(struct device_node *fcu_node)
2102 struct device_node *np = NULL;
2105 /* The table is filled by default with values that are suitable
2106 * for the old machines without device-tree informations. We scan
2107 * the device-tree and override those values with whatever is
2111 DBG("Looking up FCU controls in device-tree...\n");
2113 while ((np = of_get_next_child(fcu_node, np)) != NULL) {
2118 DBG(" control: %s, type: %s\n", np->name, np->type);
2120 /* Detect control type */
2121 if (!strcmp(np->type, "fan-rpm-control") ||
2122 !strcmp(np->type, "fan-rpm"))
2124 if (!strcmp(np->type, "fan-pwm-control") ||
2125 !strcmp(np->type, "fan-pwm"))
2127 /* Only care about fans for now */
2131 /* Lookup for a matching location */
2132 loc = of_get_property(np, "location", NULL);
2133 reg = of_get_property(np, "reg", NULL);
2134 if (loc == NULL || reg == NULL)
2136 DBG(" matching location: %s, reg: 0x%08x\n", loc, *reg);
2138 for (i = 0; i < FCU_FAN_COUNT; i++) {
2141 if (!fan_check_loc_match(loc, i))
2143 DBG(" location match, index: %d\n", i);
2144 fcu_fans[i].id = FCU_FAN_ABSENT_ID;
2145 if (type != fcu_fans[i].type) {
2146 printk(KERN_WARNING "therm_pm72: Fan type mismatch "
2147 "in device-tree for %s\n", np->full_name);
2150 if (type == FCU_FAN_RPM)
2151 fan_id = ((*reg) - 0x10) / 2;
2153 fan_id = ((*reg) - 0x30) / 2;
2155 printk(KERN_WARNING "therm_pm72: Can't parse "
2156 "fan ID in device-tree for %s\n", np->full_name);
2159 DBG(" fan id -> %d, type -> %d\n", fan_id, type);
2160 fcu_fans[i].id = fan_id;
2164 /* Now dump the array */
2165 printk(KERN_INFO "Detected fan controls:\n");
2166 for (i = 0; i < FCU_FAN_COUNT; i++) {
2167 if (fcu_fans[i].id == FCU_FAN_ABSENT_ID)
2169 printk(KERN_INFO " %d: %s fan, id %d, location: %s\n", i,
2170 fcu_fans[i].type == FCU_FAN_RPM ? "RPM" : "PWM",
2171 fcu_fans[i].id, fcu_fans[i].loc);
2175 static int fcu_of_probe(struct of_device* dev, const struct of_device_id *match)
2177 state = state_detached;
2179 /* Lookup the fans in the device tree */
2180 fcu_lookup_fans(dev->node);
2182 /* Add the driver */
2183 return i2c_add_driver(&therm_pm72_driver);
2186 static int fcu_of_remove(struct of_device* dev)
2188 i2c_del_driver(&therm_pm72_driver);
2193 static struct of_device_id fcu_match[] =
2201 static struct of_platform_driver fcu_of_platform_driver =
2203 .name = "temperature",
2204 .match_table = fcu_match,
2205 .probe = fcu_of_probe,
2206 .remove = fcu_of_remove
2210 * Check machine type, attach to i2c controller
2212 static int __init therm_pm72_init(void)
2214 struct device_node *np;
2216 rackmac = machine_is_compatible("RackMac3,1");
2218 if (!machine_is_compatible("PowerMac7,2") &&
2219 !machine_is_compatible("PowerMac7,3") &&
2223 printk(KERN_INFO "PowerMac G5 Thermal control driver %s\n", VERSION);
2225 np = of_find_node_by_type(NULL, "fcu");
2227 /* Some machines have strangely broken device-tree */
2228 np = of_find_node_by_path("/u3@0,f8000000/i2c@f8001000/fan@15e");
2230 printk(KERN_ERR "Can't find FCU in device-tree !\n");
2234 of_dev = of_platform_device_create(np, "temperature", NULL);
2235 if (of_dev == NULL) {
2236 printk(KERN_ERR "Can't register FCU platform device !\n");
2240 of_register_platform_driver(&fcu_of_platform_driver);
2245 static void __exit therm_pm72_exit(void)
2247 of_unregister_platform_driver(&fcu_of_platform_driver);
2250 of_device_unregister(of_dev);
2253 module_init(therm_pm72_init);
2254 module_exit(therm_pm72_exit);
2256 MODULE_AUTHOR("Benjamin Herrenschmidt <benh@kernel.crashing.org>");
2257 MODULE_DESCRIPTION("Driver for Apple's PowerMac G5 thermal control");
2258 MODULE_LICENSE("GPL");