Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
therm_pm72.c
Go to the documentation of this file.
1 /*
2  * Device driver for the thermostats & fan controller of the
3  * Apple G5 "PowerMac7,2" desktop machines.
4  *
5  * (c) Copyright IBM Corp. 2003-2004
6  *
7  * Maintained by: Benjamin Herrenschmidt
9  *
10  *
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.
14  *
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
23  *
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.
28  *
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
32  * safe enough ...
33  *
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
42  * implementation...
43  *
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 slowing 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  * better... 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
55  *
56  * History:
57  *
58  * Nov. 13, 2003 : 0.5
59  * - First release
60  *
61  * Nov. 14, 2003 : 0.6
62  * - Read fan speed from FCU, low level fan routines now deal
63  * with errors & check fan status, though higher level don't
64  * do much.
65  * - Move a bunch of definitions to .h file
66  *
67  * Nov. 18, 2003 : 0.7
68  * - Fix build on ppc64 kernel
69  * - Move back statics definitions to .c file
70  * - Avoid calling schedule_timeout with a negative number
71  *
72  * Dec. 18, 2003 : 0.8
73  * - Fix typo when reading back fan speed on 2 CPU machines
74  *
75  * Mar. 11, 2004 : 0.9
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
81  * pressure on i2c
82  *
83  * Oct. 20, 2004 : 1.1
84  * - Add device-tree lookup for fan IDs, should detect liquid cooling
85  * pumps when present
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
91  *
92  * Mar. 10, 2005 : 1.2
93  * - Add basic support for Xserve G5
94  * - Retrieve 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
97  *
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
108  *
109  */
110 
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/init.h>
118 #include <linux/spinlock.h>
119 #include <linux/wait.h>
120 #include <linux/reboot.h>
121 #include <linux/kmod.h>
122 #include <linux/i2c.h>
123 #include <linux/kthread.h>
124 #include <linux/mutex.h>
125 #include <linux/of_device.h>
126 #include <linux/of_platform.h>
127 #include <asm/prom.h>
128 #include <asm/machdep.h>
129 #include <asm/io.h>
130 #include <asm/sections.h>
131 #include <asm/macio.h>
132 
133 #include "therm_pm72.h"
134 
135 #define VERSION "1.3"
136 
137 #undef DEBUG
138 
139 #ifdef DEBUG
140 #define DBG(args...) printk(args)
141 #else
142 #define DBG(args...) do { } while(0)
143 #endif
144 
145 
146 /*
147  * Driver statics
148  */
149 
150 static struct platform_device * of_dev;
151 static struct i2c_adapter * u3_0;
152 static struct i2c_adapter * u3_1;
153 static struct i2c_adapter * k2;
154 static struct i2c_client * fcu;
155 static struct cpu_pid_state processor_state[2];
156 static struct basckside_pid_params backside_params;
157 static struct backside_pid_state backside_state;
158 static struct drives_pid_state drives_state;
159 static struct dimm_pid_state dimms_state;
160 static struct slots_pid_state slots_state;
161 static int state;
162 static int cpu_count;
163 static int cpu_pid_type;
164 static struct task_struct *ctrl_task;
165 static struct completion ctrl_complete;
166 static int critical_state;
167 static int rackmac;
168 static s32 dimm_output_clamp;
169 static int fcu_rpm_shift;
170 static int fcu_tickle_ticks;
171 static DEFINE_MUTEX(driver_lock);
172 
173 /*
174  * We have 3 types of CPU PID control. One is "split" old style control
175  * for intake & exhaust fans, the other is "combined" control for both
176  * CPUs that also deals with the pumps when present. To be "compatible"
177  * with OS X at this point, we only use "COMBINED" on the machines that
178  * are identified as having the pumps (though that identification is at
179  * least dodgy). Ultimately, we could probably switch completely to this
180  * algorithm provided we hack it to deal with the UP case
181  */
182 #define CPU_PID_TYPE_SPLIT 0
183 #define CPU_PID_TYPE_COMBINED 1
184 #define CPU_PID_TYPE_RACKMAC 2
185 
186 /*
187  * This table describes all fans in the FCU. The "id" and "type" values
188  * are defaults valid for all earlier machines. Newer machines will
189  * eventually override the table content based on the device-tree
190  */
192 {
193  char* loc; /* location code */
194  int type; /* 0 = rpm, 1 = pwm, 2 = pump */
195  int id; /* id or -1 */
196 };
197 
198 #define FCU_FAN_RPM 0
199 #define FCU_FAN_PWM 1
200 
201 #define FCU_FAN_ABSENT_ID -1
202 
203 #define FCU_FAN_COUNT ARRAY_SIZE(fcu_fans)
204 
207  .loc = "BACKSIDE,SYS CTRLR FAN",
208  .type = FCU_FAN_PWM,
210  },
212  .loc = "DRIVE BAY",
213  .type = FCU_FAN_RPM,
215  },
216  [SLOTS_FAN_PWM_INDEX] = {
217  .loc = "SLOT,PCI FAN",
218  .type = FCU_FAN_PWM,
220  },
222  .loc = "CPU A INTAKE",
223  .type = FCU_FAN_RPM,
225  },
227  .loc = "CPU A EXHAUST",
228  .type = FCU_FAN_RPM,
230  },
232  .loc = "CPU B INTAKE",
233  .type = FCU_FAN_RPM,
235  },
237  .loc = "CPU B EXHAUST",
238  .type = FCU_FAN_RPM,
240  },
241  /* pumps aren't present by default, have to be looked up in the
242  * device-tree
243  */
244  [CPUA_PUMP_RPM_INDEX] = {
245  .loc = "CPU A PUMP",
246  .type = FCU_FAN_RPM,
247  .id = FCU_FAN_ABSENT_ID,
248  },
249  [CPUB_PUMP_RPM_INDEX] = {
250  .loc = "CPU B PUMP",
251  .type = FCU_FAN_RPM,
252  .id = FCU_FAN_ABSENT_ID,
253  },
254  /* Xserve fans */
256  .loc = "CPU A 1",
257  .type = FCU_FAN_RPM,
258  .id = FCU_FAN_ABSENT_ID,
259  },
261  .loc = "CPU A 2",
262  .type = FCU_FAN_RPM,
263  .id = FCU_FAN_ABSENT_ID,
264  },
266  .loc = "CPU A 3",
267  .type = FCU_FAN_RPM,
268  .id = FCU_FAN_ABSENT_ID,
269  },
271  .loc = "CPU B 1",
272  .type = FCU_FAN_RPM,
273  .id = FCU_FAN_ABSENT_ID,
274  },
276  .loc = "CPU B 2",
277  .type = FCU_FAN_RPM,
278  .id = FCU_FAN_ABSENT_ID,
279  },
281  .loc = "CPU B 3",
282  .type = FCU_FAN_RPM,
283  .id = FCU_FAN_ABSENT_ID,
284  },
285 };
286 
287 static struct i2c_driver therm_pm72_driver;
288 
289 /*
290  * Utility function to create an i2c_client structure and
291  * attach it to one of u3 adapters
292  */
293 static struct i2c_client *attach_i2c_chip(int id, const char *name)
294 {
295  struct i2c_client *clt;
296  struct i2c_adapter *adap;
297  struct i2c_board_info info;
298 
299  if (id & 0x200)
300  adap = k2;
301  else if (id & 0x100)
302  adap = u3_1;
303  else
304  adap = u3_0;
305  if (adap == NULL)
306  return NULL;
307 
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);
312  if (!clt) {
313  printk(KERN_ERR "therm_pm72: Failed to attach to i2c ID 0x%x\n", id);
314  return NULL;
315  }
316 
317  /*
318  * Let i2c-core delete that device on driver removal.
319  * This is safe because i2c-core holds the core_lock mutex for us.
320  */
321  list_add_tail(&clt->detected, &therm_pm72_driver.clients);
322  return clt;
323 }
324 
325 /*
326  * Here are the i2c chip access wrappers
327  */
328 
329 static void initialize_adc(struct cpu_pid_state *state)
330 {
331  int rc;
332  u8 buf[2];
333 
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
338  */
339 
340  /* Clear Config2 */
341  buf[0] = 5;
342  buf[1] = 0;
343  i2c_master_send(state->monitor, buf, 2);
344 
345  /* Read & cache Config1 */
346  buf[0] = 1;
347  rc = i2c_master_send(state->monitor, buf, 1);
348  if (rc > 0) {
349  rc = i2c_master_recv(state->monitor, buf, 1);
350  if (rc > 0) {
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;
355  buf[0] = 1;
356  buf[1] = state->adc_config;
357  rc = i2c_master_send(state->monitor, buf, 2);
358  }
359  }
360  if (rc <= 0)
361  printk(KERN_ERR "therm_pm72: Error reading ADC config"
362  " register !\n");
363 }
364 
365 static int read_smon_adc(struct cpu_pid_state *state, int chan)
366 {
367  int rc, data, tries = 0;
368  u8 buf[2];
369 
370  for (;;) {
371  /* Set channel */
372  buf[0] = 1;
373  buf[1] = (state->adc_config & 0x1f) | (chan << 5);
374  rc = i2c_master_send(state->monitor, buf, 2);
375  if (rc <= 0)
376  goto error;
377  /* Wait for conversion */
378  msleep(1);
379  /* Switch to data register */
380  buf[0] = 4;
381  rc = i2c_master_send(state->monitor, buf, 1);
382  if (rc <= 0)
383  goto error;
384  /* Read result */
385  rc = i2c_master_recv(state->monitor, buf, 2);
386  if (rc < 0)
387  goto error;
388  data = ((u16)buf[0]) << 8 | (u16)buf[1];
389  return data >> 6;
390  error:
391  DBG("Error reading ADC, retrying...\n");
392  if (++tries > 10) {
393  printk(KERN_ERR "therm_pm72: Error reading ADC !\n");
394  return -1;
395  }
396  msleep(10);
397  }
398 }
399 
400 static int read_lm87_reg(struct i2c_client * chip, int reg)
401 {
402  int rc, tries = 0;
403  u8 buf;
404 
405  for (;;) {
406  /* Set address */
407  buf = (u8)reg;
408  rc = i2c_master_send(chip, &buf, 1);
409  if (rc <= 0)
410  goto error;
411  rc = i2c_master_recv(chip, &buf, 1);
412  if (rc <= 0)
413  goto error;
414  return (int)buf;
415  error:
416  DBG("Error reading LM87, retrying...\n");
417  if (++tries > 10) {
418  printk(KERN_ERR "therm_pm72: Error reading LM87 !\n");
419  return -1;
420  }
421  msleep(10);
422  }
423 }
424 
425 static int fan_read_reg(int reg, unsigned char *buf, int nb)
426 {
427  int tries, nr, nw;
428 
429  buf[0] = reg;
430  tries = 0;
431  for (;;) {
432  nw = i2c_master_send(fcu, buf, 1);
433  if (nw > 0 || (nw < 0 && nw != -EIO) || tries >= 100)
434  break;
435  msleep(10);
436  ++tries;
437  }
438  if (nw <= 0) {
439  printk(KERN_ERR "Failure writing address to FCU: %d", nw);
440  return -EIO;
441  }
442  tries = 0;
443  for (;;) {
444  nr = i2c_master_recv(fcu, buf, nb);
445  if (nr > 0 || (nr < 0 && nr != -ENODEV) || tries >= 100)
446  break;
447  msleep(10);
448  ++tries;
449  }
450  if (nr <= 0)
451  printk(KERN_ERR "Failure reading data from FCU: %d", nw);
452  return nr;
453 }
454 
455 static int fan_write_reg(int reg, const unsigned char *ptr, int nb)
456 {
457  int tries, nw;
458  unsigned char buf[16];
459 
460  buf[0] = reg;
461  memcpy(buf+1, ptr, nb);
462  ++nb;
463  tries = 0;
464  for (;;) {
465  nw = i2c_master_send(fcu, buf, nb);
466  if (nw > 0 || (nw < 0 && nw != -EIO) || tries >= 100)
467  break;
468  msleep(10);
469  ++tries;
470  }
471  if (nw < 0)
472  printk(KERN_ERR "Failure writing to FCU: %d", nw);
473  return nw;
474 }
475 
476 static int start_fcu(void)
477 {
478  unsigned char buf = 0xff;
479  int rc;
480 
481  rc = fan_write_reg(0xe, &buf, 1);
482  if (rc < 0)
483  return -EIO;
484  rc = fan_write_reg(0x2e, &buf, 1);
485  if (rc < 0)
486  return -EIO;
487  rc = fan_read_reg(0, &buf, 1);
488  if (rc < 0)
489  return -EIO;
490  fcu_rpm_shift = (buf == 1) ? 2 : 3;
491  printk(KERN_DEBUG "FCU Initialized, RPM fan shift is %d\n",
492  fcu_rpm_shift);
493 
494  return 0;
495 }
496 
497 static int set_rpm_fan(int fan_index, int rpm)
498 {
499  unsigned char buf[2];
500  int rc, id, min, max;
501 
502  if (fcu_fans[fan_index].type != FCU_FAN_RPM)
503  return -EINVAL;
504  id = fcu_fans[fan_index].id;
505  if (id == FCU_FAN_ABSENT_ID)
506  return -EINVAL;
507 
508  min = 2400 >> fcu_rpm_shift;
509  max = 56000 >> fcu_rpm_shift;
510 
511  if (rpm < min)
512  rpm = min;
513  else if (rpm > max)
514  rpm = max;
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);
518  if (rc < 0)
519  return -EIO;
520  return 0;
521 }
522 
523 static int get_rpm_fan(int fan_index, int programmed)
524 {
525  unsigned char failure;
526  unsigned char active;
527  unsigned char buf[2];
528  int rc, id, reg_base;
529 
530  if (fcu_fans[fan_index].type != FCU_FAN_RPM)
531  return -EINVAL;
532  id = fcu_fans[fan_index].id;
533  if (id == FCU_FAN_ABSENT_ID)
534  return -EINVAL;
535 
536  rc = fan_read_reg(0xb, &failure, 1);
537  if (rc != 1)
538  return -EIO;
539  if ((failure & (1 << id)) != 0)
540  return -EFAULT;
541  rc = fan_read_reg(0xd, &active, 1);
542  if (rc != 1)
543  return -EIO;
544  if ((active & (1 << id)) == 0)
545  return -ENXIO;
546 
547  /* Programmed value or real current speed */
548  reg_base = programmed ? 0x10 : 0x11;
549  rc = fan_read_reg(reg_base + (id * 2), buf, 2);
550  if (rc != 2)
551  return -EIO;
552 
553  return (buf[0] << (8 - fcu_rpm_shift)) | buf[1] >> fcu_rpm_shift;
554 }
555 
556 static int set_pwm_fan(int fan_index, int pwm)
557 {
558  unsigned char buf[2];
559  int rc, id;
560 
561  if (fcu_fans[fan_index].type != FCU_FAN_PWM)
562  return -EINVAL;
563  id = fcu_fans[fan_index].id;
564  if (id == FCU_FAN_ABSENT_ID)
565  return -EINVAL;
566 
567  if (pwm < 10)
568  pwm = 10;
569  else if (pwm > 100)
570  pwm = 100;
571  pwm = (pwm * 2559) / 1000;
572  buf[0] = pwm;
573  rc = fan_write_reg(0x30 + (id * 2), buf, 1);
574  if (rc < 0)
575  return rc;
576  return 0;
577 }
578 
579 static int get_pwm_fan(int fan_index)
580 {
581  unsigned char failure;
582  unsigned char active;
583  unsigned char buf[2];
584  int rc, id;
585 
586  if (fcu_fans[fan_index].type != FCU_FAN_PWM)
587  return -EINVAL;
588  id = fcu_fans[fan_index].id;
589  if (id == FCU_FAN_ABSENT_ID)
590  return -EINVAL;
591 
592  rc = fan_read_reg(0x2b, &failure, 1);
593  if (rc != 1)
594  return -EIO;
595  if ((failure & (1 << id)) != 0)
596  return -EFAULT;
597  rc = fan_read_reg(0x2d, &active, 1);
598  if (rc != 1)
599  return -EIO;
600  if ((active & (1 << id)) == 0)
601  return -ENXIO;
602 
603  /* Programmed value or real current speed */
604  rc = fan_read_reg(0x30 + (id * 2), buf, 1);
605  if (rc != 1)
606  return -EIO;
607 
608  return (buf[0] * 1000) / 2559;
609 }
610 
611 static void tickle_fcu(void)
612 {
613  int pwm;
614 
615  pwm = get_pwm_fan(SLOTS_FAN_PWM_INDEX);
616 
617  DBG("FCU Tickle, slots fan is: %d\n", pwm);
618  if (pwm < 0)
619  pwm = 100;
620 
621  if (!rackmac) {
622  pwm = SLOTS_FAN_DEFAULT_PWM;
623  } else if (pwm < SLOTS_PID_OUTPUT_MIN)
624  pwm = SLOTS_PID_OUTPUT_MIN;
625 
626  /* That is hopefully enough to make the FCU happy */
627  set_pwm_fan(SLOTS_FAN_PWM_INDEX, pwm);
628 }
629 
630 
631 /*
632  * Utility routine to read the CPU calibration EEPROM data
633  * from the device-tree
634  */
635 static int read_eeprom(int cpu, struct mpu_data *out)
636 {
637  struct device_node *np;
638  char nodename[64];
639  const u8 *data;
640  int len;
641 
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
645  */
646  sprintf(nodename, "/u3@0,f8000000/i2c@f8001000/cpuid@a%d", cpu ? 2 : 0);
647  np = of_find_node_by_path(nodename);
648  if (np == NULL) {
649  printk(KERN_ERR "therm_pm72: Failed to retrieve cpuid node from device-tree\n");
650  return -ENODEV;
651  }
652  data = of_get_property(np, "cpuid", &len);
653  if (data == NULL) {
654  printk(KERN_ERR "therm_pm72: Failed to retrieve cpuid property from device-tree\n");
655  of_node_put(np);
656  return -ENODEV;
657  }
658  memcpy(out, data, sizeof(struct mpu_data));
659  of_node_put(np);
660 
661  return 0;
662 }
663 
664 static void fetch_cpu_pumps_minmax(void)
665 {
666  struct cpu_pid_state *state0 = &processor_state[0];
667  struct cpu_pid_state *state1 = &processor_state[1];
668  u16 pump_min = 0, pump_max = 0xffff;
669  u16 tmp[4];
670 
671  /* Try to fetch pumps min/max infos from eeprom */
672 
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]);
677  }
678  if (tmp[2] != 0xffff && tmp[3] != 0xffff) {
679  pump_min = max(pump_min, tmp[2]);
680  pump_max = min(pump_max, tmp[3]);
681  }
682 
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)
686  */
687  if (pump_min == pump_max || pump_min == 0 || pump_max == 0xffff) {
688  pump_min = CPU_PUMP_OUTPUT_MIN;
690  }
691 
692  state0->pump_min = state1->pump_min = pump_min;
693  state0->pump_max = state1->pump_max = pump_max;
694 }
695 
696 /*
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...
700  *
701  * That sucks a bit, we take the lock because FIX32TOPRINT evaluates
702  * the input twice... I accept patches :)
703  */
704 #define BUILD_SHOW_FUNC_FIX(name, data) \
705 static ssize_t show_##name(struct device *dev, struct device_attribute *attr, char *buf) \
706 { \
707  ssize_t r; \
708  mutex_lock(&driver_lock); \
709  r = sprintf(buf, "%d.%03d", FIX32TOPRINT(data)); \
710  mutex_unlock(&driver_lock); \
711  return r; \
712 }
713 #define BUILD_SHOW_FUNC_INT(name, data) \
714 static ssize_t show_##name(struct device *dev, struct device_attribute *attr, char *buf) \
715 { \
716  return sprintf(buf, "%d", data); \
717 }
718 
719 BUILD_SHOW_FUNC_FIX(cpu0_temperature, processor_state[0].last_temp)
720 BUILD_SHOW_FUNC_FIX(cpu0_voltage, processor_state[0].voltage)
721 BUILD_SHOW_FUNC_FIX(cpu0_current, processor_state[0].current_a)
722 BUILD_SHOW_FUNC_INT(cpu0_exhaust_fan_rpm, processor_state[0].rpm)
723 BUILD_SHOW_FUNC_INT(cpu0_intake_fan_rpm, processor_state[0].intake_rpm)
724 
725 BUILD_SHOW_FUNC_FIX(cpu1_temperature, processor_state[1].last_temp)
726 BUILD_SHOW_FUNC_FIX(cpu1_voltage, processor_state[1].voltage)
727 BUILD_SHOW_FUNC_FIX(cpu1_current, processor_state[1].current_a)
728 BUILD_SHOW_FUNC_INT(cpu1_exhaust_fan_rpm, processor_state[1].rpm)
729 BUILD_SHOW_FUNC_INT(cpu1_intake_fan_rpm, processor_state[1].intake_rpm)
730 
731 BUILD_SHOW_FUNC_FIX(backside_temperature, backside_state.last_temp)
732 BUILD_SHOW_FUNC_INT(backside_fan_pwm, backside_state.pwm)
733 
734 BUILD_SHOW_FUNC_FIX(drives_temperature, drives_state.last_temp)
735 BUILD_SHOW_FUNC_INT(drives_fan_rpm, drives_state.rpm)
736 
737 BUILD_SHOW_FUNC_FIX(slots_temperature, slots_state.last_temp)
738 BUILD_SHOW_FUNC_INT(slots_fan_pwm, slots_state.pwm)
739 
740 BUILD_SHOW_FUNC_FIX(dimms_temperature, dimms_state.last_temp)
741 
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);
747 
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);
753 
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);
756 
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);
759 
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);
762 
763 static DEVICE_ATTR(dimms_temperature,S_IRUGO,show_dimms_temperature,NULL);
764 
765 /*
766  * CPUs fans control loop
767  */
768 
769 static int do_read_one_cpu_values(struct cpu_pid_state *state, s32 *temp, s32 *power)
770 {
771  s32 ltemp, volts, amps;
772  int index, rc = 0;
773 
774  /* Default (in case of error) */
775  *temp = state->cur_temp;
776  *power = state->cur_power;
777 
778  if (cpu_pid_type == CPU_PID_TYPE_RACKMAC)
779  index = (state->index == 0) ?
781  else
782  index = (state->index == 0) ?
784 
785  /* Read current fan status */
786  rc = get_rpm_fan(index, !RPM_PID_USE_ACTUAL_SPEED);
787  if (rc < 0) {
788  /* XXX What do we do now ? Nothing for now, keep old value, but
789  * return error upstream
790  */
791  DBG(" cpu %d, fan reading error !\n", state->index);
792  } else {
793  state->rpm = rc;
794  DBG(" cpu %d, exhaust RPM: %d\n", state->index, state->rpm);
795  }
796 
797  /* Get some sensor readings and scale it */
798  ltemp = read_smon_adc(state, 1);
799  if (ltemp == -1) {
800  /* XXX What do we do now ? */
801  state->overtemp++;
802  if (rc == 0)
803  rc = -EIO;
804  DBG(" cpu %d, temp reading error !\n", state->index);
805  } else {
806  /* Fixup temperature according to diode calibration
807  */
808  DBG(" cpu %d, temp raw: %04x, m_diode: %04x, b_diode: %04x\n",
809  state->index,
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)));
814  }
815 
816  /*
817  * Read voltage & current and calculate power
818  */
819  volts = read_smon_adc(state, 3);
820  amps = read_smon_adc(state, 4);
821 
822  /* Scale voltage and current raw sensor values according to fixed scales
823  * obtained in Darwin and calculate power from I and V
824  */
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;
831 
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));
835 
836  return 0;
837 }
838 
839 static void do_cpu_pid(struct cpu_pid_state *state, s32 temp, s32 power)
840 {
841  s32 power_target, integral, derivative, proportional, adj_in_target, sval;
842  s64 integ_p, deriv_p, prop_p, sum;
843  int i;
844 
845  /* Calculate power target value (could be done once for all)
846  * and convert to a 16.16 fp number
847  */
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));
851 
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;
858 
859  /* If first loop, fill the history table */
860  if (state->first) {
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;
865  }
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;
869  }
870  state->first = 0;
871  }
872 
873  /* Calculate the integral term normally based on the "power" values */
874  sum = 0;
875  integral = 0;
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);
880 
881  /* Calculate the adjusted input (sense value).
882  * G_r is 12.20
883  * integ is 16.16
884  * so the result is 28.36
885  *
886  * input target is mpu.ttarget, input max is mpu.tmax
887  */
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),
895  state->mpu.ttarget);
896 
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)
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));
904  sum += deriv_p;
905 
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));
910  sum += prop_p;
911 
912  /* Scale sum */
913  sum >>= 36;
914 
915  DBG(" sum: %d\n", (int)sum);
916  state->rpm += (s32)sum;
917 }
918 
919 static void do_monitor_cpu_combined(void)
920 {
921  struct cpu_pid_state *state0 = &processor_state[0];
922  struct cpu_pid_state *state1 = &processor_state[1];
923  s32 temp0, power0, temp1, power1;
924  s32 temp_combi, power_combi;
925  int rc, intake, pump;
926 
927  rc = do_read_one_cpu_values(state0, &temp0, &power0);
928  if (rc < 0) {
929  /* XXX What do we do now ? */
930  }
931  state1->overtemp = 0;
932  rc = do_read_one_cpu_values(state1, &temp1, &power1);
933  if (rc < 0) {
934  /* XXX What do we do now ? */
935  }
936  if (state1->overtemp)
937  state0->overtemp++;
938 
939  temp_combi = max(temp0, temp1);
940  power_combi = max(power0, power1);
941 
942  /* Check tmax, increment overtemp if we are there. At tmax+8, we go
943  * full blown immediately and try to trigger a shutdown
944  */
945  if (temp_combi >= ((state0->mpu.tmax + 8) << 16)) {
946  printk(KERN_WARNING "Warning ! Temperature way above maximum (%d) !\n",
947  temp_combi >> 16);
948  state0->overtemp += CPU_MAX_OVERTEMP / 4;
949  } else if (temp_combi > (state0->mpu.tmax << 16)) {
950  state0->overtemp++;
951  printk(KERN_WARNING "Temperature %d above max %d. overtemp %d\n",
952  temp_combi >> 16, state0->mpu.tmax, state0->overtemp);
953  } else {
954  if (state0->overtemp)
955  printk(KERN_WARNING "Temperature back down to %d\n",
956  temp_combi >> 16);
957  state0->overtemp = 0;
958  }
959  if (state0->overtemp >= CPU_MAX_OVERTEMP)
960  critical_state = 1;
961  if (state0->overtemp > 0) {
962  state0->rpm = state0->mpu.rmaxn_exhaust_fan;
963  state0->intake_rpm = intake = state0->mpu.rmaxn_intake_fan;
964  pump = state0->pump_max;
965  goto do_set_fans;
966  }
967 
968  /* Do the PID */
969  do_cpu_pid(state0, temp_combi, power_combi);
970 
971  /* Range check */
972  state0->rpm = max(state0->rpm, (int)state0->mpu.rminn_exhaust_fan);
973  state0->rpm = min(state0->rpm, (int)state0->mpu.rmaxn_exhaust_fan);
974 
975  /* Calculate intake fan speed */
976  intake = (state0->rpm * CPU_INTAKE_SCALE) >> 16;
977  intake = max(intake, (int)state0->mpu.rminn_intake_fan);
978  intake = min(intake, (int)state0->mpu.rmaxn_intake_fan);
979  state0->intake_rpm = intake;
980 
981  /* Calculate pump speed */
982  pump = (state0->rpm * state0->pump_max) /
983  state0->mpu.rmaxn_exhaust_fan;
984  pump = min(pump, state0->pump_max);
985  pump = max(pump, state0->pump_min);
986 
987  do_set_fans:
988  /* We copy values from state 0 to state 1 for /sysfs */
989  state1->rpm = state0->rpm;
990  state1->intake_rpm = state0->intake_rpm;
991 
992  DBG("** CPU %d RPM: %d Ex, %d, Pump: %d, In, overtemp: %d\n",
993  state1->index, (int)state1->rpm, intake, pump, state1->overtemp);
994 
995  /* We should check for errors, shouldn't we ? But then, what
996  * do we do once the error occurs ? For FCU notified fan
997  * failures (-EFAULT) we probably want to notify userland
998  * some way...
999  */
1000  set_rpm_fan(CPUA_INTAKE_FAN_RPM_INDEX, intake);
1001  set_rpm_fan(CPUA_EXHAUST_FAN_RPM_INDEX, state0->rpm);
1002  set_rpm_fan(CPUB_INTAKE_FAN_RPM_INDEX, intake);
1003  set_rpm_fan(CPUB_EXHAUST_FAN_RPM_INDEX, state0->rpm);
1004 
1005  if (fcu_fans[CPUA_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID)
1006  set_rpm_fan(CPUA_PUMP_RPM_INDEX, pump);
1007  if (fcu_fans[CPUB_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID)
1008  set_rpm_fan(CPUB_PUMP_RPM_INDEX, pump);
1009 }
1010 
1011 static void do_monitor_cpu_split(struct cpu_pid_state *state)
1012 {
1013  s32 temp, power;
1014  int rc, intake;
1015 
1016  /* Read current fan status */
1017  rc = do_read_one_cpu_values(state, &temp, &power);
1018  if (rc < 0) {
1019  /* XXX What do we do now ? */
1020  }
1021 
1022  /* Check tmax, increment overtemp if we are there. At tmax+8, we go
1023  * full blown immediately and try to trigger a shutdown
1024  */
1025  if (temp >= ((state->mpu.tmax + 8) << 16)) {
1026  printk(KERN_WARNING "Warning ! CPU %d temperature way above maximum"
1027  " (%d) !\n",
1028  state->index, temp >> 16);
1029  state->overtemp += CPU_MAX_OVERTEMP / 4;
1030  } else if (temp > (state->mpu.tmax << 16)) {
1031  state->overtemp++;
1032  printk(KERN_WARNING "CPU %d temperature %d above max %d. overtemp %d\n",
1033  state->index, temp >> 16, state->mpu.tmax, state->overtemp);
1034  } else {
1035  if (state->overtemp)
1036  printk(KERN_WARNING "CPU %d temperature back down to %d\n",
1037  state->index, temp >> 16);
1038  state->overtemp = 0;
1039  }
1040  if (state->overtemp >= CPU_MAX_OVERTEMP)
1041  critical_state = 1;
1042  if (state->overtemp > 0) {
1043  state->rpm = state->mpu.rmaxn_exhaust_fan;
1044  state->intake_rpm = intake = state->mpu.rmaxn_intake_fan;
1045  goto do_set_fans;
1046  }
1047 
1048  /* Do the PID */
1049  do_cpu_pid(state, temp, power);
1050 
1051  /* Range check */
1052  state->rpm = max(state->rpm, (int)state->mpu.rminn_exhaust_fan);
1053  state->rpm = min(state->rpm, (int)state->mpu.rmaxn_exhaust_fan);
1054 
1055  /* Calculate intake fan */
1056  intake = (state->rpm * CPU_INTAKE_SCALE) >> 16;
1057  intake = max(intake, (int)state->mpu.rminn_intake_fan);
1058  intake = min(intake, (int)state->mpu.rmaxn_intake_fan);
1059  state->intake_rpm = intake;
1060 
1061  do_set_fans:
1062  DBG("** CPU %d RPM: %d Ex, %d In, overtemp: %d\n",
1063  state->index, (int)state->rpm, intake, state->overtemp);
1064 
1065  /* We should check for errors, shouldn't we ? But then, what
1066  * do we do once the error occurs ? For FCU notified fan
1067  * failures (-EFAULT) we probably want to notify userland
1068  * some way...
1069  */
1070  if (state->index == 0) {
1071  set_rpm_fan(CPUA_INTAKE_FAN_RPM_INDEX, intake);
1072  set_rpm_fan(CPUA_EXHAUST_FAN_RPM_INDEX, state->rpm);
1073  } else {
1074  set_rpm_fan(CPUB_INTAKE_FAN_RPM_INDEX, intake);
1075  set_rpm_fan(CPUB_EXHAUST_FAN_RPM_INDEX, state->rpm);
1076  }
1077 }
1078 
1079 static void do_monitor_cpu_rack(struct cpu_pid_state *state)
1080 {
1081  s32 temp, power, fan_min;
1082  int rc;
1083 
1084  /* Read current fan status */
1085  rc = do_read_one_cpu_values(state, &temp, &power);
1086  if (rc < 0) {
1087  /* XXX What do we do now ? */
1088  }
1089 
1090  /* Check tmax, increment overtemp if we are there. At tmax+8, we go
1091  * full blown immediately and try to trigger a shutdown
1092  */
1093  if (temp >= ((state->mpu.tmax + 8) << 16)) {
1094  printk(KERN_WARNING "Warning ! CPU %d temperature way above maximum"
1095  " (%d) !\n",
1096  state->index, temp >> 16);
1097  state->overtemp = CPU_MAX_OVERTEMP / 4;
1098  } else if (temp > (state->mpu.tmax << 16)) {
1099  state->overtemp++;
1100  printk(KERN_WARNING "CPU %d temperature %d above max %d. overtemp %d\n",
1101  state->index, temp >> 16, state->mpu.tmax, state->overtemp);
1102  } else {
1103  if (state->overtemp)
1104  printk(KERN_WARNING "CPU %d temperature back down to %d\n",
1105  state->index, temp >> 16);
1106  state->overtemp = 0;
1107  }
1108  if (state->overtemp >= CPU_MAX_OVERTEMP)
1109  critical_state = 1;
1110  if (state->overtemp > 0) {
1111  state->rpm = state->intake_rpm = state->mpu.rmaxn_intake_fan;
1112  goto do_set_fans;
1113  }
1114 
1115  /* Do the PID */
1116  do_cpu_pid(state, temp, power);
1117 
1118  /* Check clamp from dimms */
1119  fan_min = dimm_output_clamp;
1120  fan_min = max(fan_min, (int)state->mpu.rminn_intake_fan);
1121 
1122  DBG(" CPU min mpu = %d, min dimm = %d\n",
1123  state->mpu.rminn_intake_fan, dimm_output_clamp);
1124 
1125  state->rpm = max(state->rpm, (int)fan_min);
1126  state->rpm = min(state->rpm, (int)state->mpu.rmaxn_intake_fan);
1127  state->intake_rpm = state->rpm;
1128 
1129  do_set_fans:
1130  DBG("** CPU %d RPM: %d overtemp: %d\n",
1131  state->index, (int)state->rpm, state->overtemp);
1132 
1133  /* We should check for errors, shouldn't we ? But then, what
1134  * do we do once the error occurs ? For FCU notified fan
1135  * failures (-EFAULT) we probably want to notify userland
1136  * some way...
1137  */
1138  if (state->index == 0) {
1139  set_rpm_fan(CPU_A1_FAN_RPM_INDEX, state->rpm);
1140  set_rpm_fan(CPU_A2_FAN_RPM_INDEX, state->rpm);
1141  set_rpm_fan(CPU_A3_FAN_RPM_INDEX, state->rpm);
1142  } else {
1143  set_rpm_fan(CPU_B1_FAN_RPM_INDEX, state->rpm);
1144  set_rpm_fan(CPU_B2_FAN_RPM_INDEX, state->rpm);
1145  set_rpm_fan(CPU_B3_FAN_RPM_INDEX, state->rpm);
1146  }
1147 }
1148 
1149 /*
1150  * Initialize the state structure for one CPU control loop
1151  */
1152 static int init_processor_state(struct cpu_pid_state *state, int index)
1153 {
1154  int err;
1155 
1156  state->index = index;
1157  state->first = 1;
1158  state->rpm = (cpu_pid_type == CPU_PID_TYPE_RACKMAC) ? 4000 : 1000;
1159  state->overtemp = 0;
1160  state->adc_config = 0x00;
1161 
1162 
1163  if (index == 0)
1164  state->monitor = attach_i2c_chip(SUPPLY_MONITOR_ID, "CPU0_monitor");
1165  else if (index == 1)
1166  state->monitor = attach_i2c_chip(SUPPLY_MONITORB_ID, "CPU1_monitor");
1167  if (state->monitor == NULL)
1168  goto fail;
1169 
1170  if (read_eeprom(index, &state->mpu))
1171  goto fail;
1172 
1173  state->count_power = state->mpu.tguardband;
1174  if (state->count_power > CPU_POWER_HISTORY_SIZE) {
1175  printk(KERN_WARNING "Warning ! too many power history slots\n");
1177  }
1178  DBG("CPU %d Using %d power history entries\n", index, state->count_power);
1179 
1180  if (index == 0) {
1181  err = device_create_file(&of_dev->dev, &dev_attr_cpu0_temperature);
1182  err |= device_create_file(&of_dev->dev, &dev_attr_cpu0_voltage);
1183  err |= device_create_file(&of_dev->dev, &dev_attr_cpu0_current);
1184  err |= device_create_file(&of_dev->dev, &dev_attr_cpu0_exhaust_fan_rpm);
1185  err |= device_create_file(&of_dev->dev, &dev_attr_cpu0_intake_fan_rpm);
1186  } else {
1187  err = device_create_file(&of_dev->dev, &dev_attr_cpu1_temperature);
1188  err |= device_create_file(&of_dev->dev, &dev_attr_cpu1_voltage);
1189  err |= device_create_file(&of_dev->dev, &dev_attr_cpu1_current);
1190  err |= device_create_file(&of_dev->dev, &dev_attr_cpu1_exhaust_fan_rpm);
1191  err |= device_create_file(&of_dev->dev, &dev_attr_cpu1_intake_fan_rpm);
1192  }
1193  if (err)
1194  printk(KERN_WARNING "Failed to create some of the attribute"
1195  "files for CPU %d\n", index);
1196 
1197  return 0;
1198  fail:
1199  state->monitor = NULL;
1200 
1201  return -ENODEV;
1202 }
1203 
1204 /*
1205  * Dispose of the state data for one CPU control loop
1206  */
1207 static void dispose_processor_state(struct cpu_pid_state *state)
1208 {
1209  if (state->monitor == NULL)
1210  return;
1211 
1212  if (state->index == 0) {
1213  device_remove_file(&of_dev->dev, &dev_attr_cpu0_temperature);
1214  device_remove_file(&of_dev->dev, &dev_attr_cpu0_voltage);
1215  device_remove_file(&of_dev->dev, &dev_attr_cpu0_current);
1216  device_remove_file(&of_dev->dev, &dev_attr_cpu0_exhaust_fan_rpm);
1217  device_remove_file(&of_dev->dev, &dev_attr_cpu0_intake_fan_rpm);
1218  } else {
1219  device_remove_file(&of_dev->dev, &dev_attr_cpu1_temperature);
1220  device_remove_file(&of_dev->dev, &dev_attr_cpu1_voltage);
1221  device_remove_file(&of_dev->dev, &dev_attr_cpu1_current);
1222  device_remove_file(&of_dev->dev, &dev_attr_cpu1_exhaust_fan_rpm);
1223  device_remove_file(&of_dev->dev, &dev_attr_cpu1_intake_fan_rpm);
1224  }
1225 
1226  state->monitor = NULL;
1227 }
1228 
1229 /*
1230  * Motherboard backside & U3 heatsink fan control loop
1231  */
1232 static void do_monitor_backside(struct backside_pid_state *state)
1233 {
1234  s32 temp, integral, derivative, fan_min;
1235  s64 integ_p, deriv_p, prop_p, sum;
1236  int i, rc;
1237 
1238  if (--state->ticks != 0)
1239  return;
1240  state->ticks = backside_params.interval;
1241 
1242  DBG("backside:\n");
1243 
1244  /* Check fan status */
1245  rc = get_pwm_fan(BACKSIDE_FAN_PWM_INDEX);
1246  if (rc < 0) {
1247  printk(KERN_WARNING "Error %d reading backside fan !\n", rc);
1248  /* XXX What do we do now ? */
1249  } else
1250  state->pwm = rc;
1251  DBG(" current pwm: %d\n", state->pwm);
1252 
1253  /* Get some sensor readings */
1254  temp = i2c_smbus_read_byte_data(state->monitor, MAX6690_EXT_TEMP) << 16;
1255  state->last_temp = temp;
1256  DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1257  FIX32TOPRINT(backside_params.input_target));
1258 
1259  /* Store temperature and error in history array */
1260  state->cur_sample = (state->cur_sample + 1) % BACKSIDE_PID_HISTORY_SIZE;
1261  state->sample_history[state->cur_sample] = temp;
1262  state->error_history[state->cur_sample] = temp - backside_params.input_target;
1263 
1264  /* If first loop, fill the history table */
1265  if (state->first) {
1266  for (i = 0; i < (BACKSIDE_PID_HISTORY_SIZE - 1); i++) {
1267  state->cur_sample = (state->cur_sample + 1) %
1269  state->sample_history[state->cur_sample] = temp;
1270  state->error_history[state->cur_sample] =
1271  temp - backside_params.input_target;
1272  }
1273  state->first = 0;
1274  }
1275 
1276  /* Calculate the integral term */
1277  sum = 0;
1278  integral = 0;
1279  for (i = 0; i < BACKSIDE_PID_HISTORY_SIZE; i++)
1280  integral += state->error_history[i];
1281  integral *= backside_params.interval;
1282  DBG(" integral: %08x\n", integral);
1283  integ_p = ((s64)backside_params.G_r) * (s64)integral;
1284  DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1285  sum += integ_p;
1286 
1287  /* Calculate the derivative term */
1288  derivative = state->error_history[state->cur_sample] -
1289  state->error_history[(state->cur_sample + BACKSIDE_PID_HISTORY_SIZE - 1)
1290  % BACKSIDE_PID_HISTORY_SIZE];
1291  derivative /= backside_params.interval;
1292  deriv_p = ((s64)backside_params.G_d) * (s64)derivative;
1293  DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1294  sum += deriv_p;
1295 
1296  /* Calculate the proportional term */
1297  prop_p = ((s64)backside_params.G_p) * (s64)(state->error_history[state->cur_sample]);
1298  DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1299  sum += prop_p;
1300 
1301  /* Scale sum */
1302  sum >>= 36;
1303 
1304  DBG(" sum: %d\n", (int)sum);
1305  if (backside_params.additive)
1306  state->pwm += (s32)sum;
1307  else
1308  state->pwm = sum;
1309 
1310  /* Check for clamp */
1311  fan_min = (dimm_output_clamp * 100) / 14000;
1312  fan_min = max(fan_min, backside_params.output_min);
1313 
1314  state->pwm = max(state->pwm, fan_min);
1315  state->pwm = min(state->pwm, backside_params.output_max);
1316 
1317  DBG("** BACKSIDE PWM: %d\n", (int)state->pwm);
1318  set_pwm_fan(BACKSIDE_FAN_PWM_INDEX, state->pwm);
1319 }
1320 
1321 /*
1322  * Initialize the state structure for the backside fan control loop
1323  */
1324 static int init_backside_state(struct backside_pid_state *state)
1325 {
1326  struct device_node *u3;
1327  int u3h = 1; /* conservative by default */
1328  int err;
1329 
1330  /*
1331  * There are different PID params for machines with U3 and machines
1332  * with U3H, pick the right ones now
1333  */
1334  u3 = of_find_node_by_path("/u3@0,f8000000");
1335  if (u3 != NULL) {
1336  const u32 *vers = of_get_property(u3, "device-rev", NULL);
1337  if (vers)
1338  if (((*vers) & 0x3f) < 0x34)
1339  u3h = 0;
1340  of_node_put(u3);
1341  }
1342 
1343  if (rackmac) {
1344  backside_params.G_d = BACKSIDE_PID_RACK_G_d;
1345  backside_params.input_target = BACKSIDE_PID_RACK_INPUT_TARGET;
1346  backside_params.output_min = BACKSIDE_PID_U3H_OUTPUT_MIN;
1347  backside_params.interval = BACKSIDE_PID_RACK_INTERVAL;
1348  backside_params.G_p = BACKSIDE_PID_RACK_G_p;
1349  backside_params.G_r = BACKSIDE_PID_G_r;
1350  backside_params.output_max = BACKSIDE_PID_OUTPUT_MAX;
1351  backside_params.additive = 0;
1352  } else if (u3h) {
1353  backside_params.G_d = BACKSIDE_PID_U3H_G_d;
1354  backside_params.input_target = BACKSIDE_PID_U3H_INPUT_TARGET;
1355  backside_params.output_min = BACKSIDE_PID_U3H_OUTPUT_MIN;
1356  backside_params.interval = BACKSIDE_PID_INTERVAL;
1357  backside_params.G_p = BACKSIDE_PID_G_p;
1358  backside_params.G_r = BACKSIDE_PID_G_r;
1359  backside_params.output_max = BACKSIDE_PID_OUTPUT_MAX;
1360  backside_params.additive = 1;
1361  } else {
1362  backside_params.G_d = BACKSIDE_PID_U3_G_d;
1363  backside_params.input_target = BACKSIDE_PID_U3_INPUT_TARGET;
1364  backside_params.output_min = BACKSIDE_PID_U3_OUTPUT_MIN;
1365  backside_params.interval = BACKSIDE_PID_INTERVAL;
1366  backside_params.G_p = BACKSIDE_PID_G_p;
1367  backside_params.G_r = BACKSIDE_PID_G_r;
1368  backside_params.output_max = BACKSIDE_PID_OUTPUT_MAX;
1369  backside_params.additive = 1;
1370  }
1371 
1372  state->ticks = 1;
1373  state->first = 1;
1374  state->pwm = 50;
1375 
1376  state->monitor = attach_i2c_chip(BACKSIDE_MAX_ID, "backside_temp");
1377  if (state->monitor == NULL)
1378  return -ENODEV;
1379 
1380  err = device_create_file(&of_dev->dev, &dev_attr_backside_temperature);
1381  err |= device_create_file(&of_dev->dev, &dev_attr_backside_fan_pwm);
1382  if (err)
1383  printk(KERN_WARNING "Failed to create attribute file(s)"
1384  " for backside fan\n");
1385 
1386  return 0;
1387 }
1388 
1389 /*
1390  * Dispose of the state data for the backside control loop
1391  */
1392 static void dispose_backside_state(struct backside_pid_state *state)
1393 {
1394  if (state->monitor == NULL)
1395  return;
1396 
1397  device_remove_file(&of_dev->dev, &dev_attr_backside_temperature);
1398  device_remove_file(&of_dev->dev, &dev_attr_backside_fan_pwm);
1399 
1400  state->monitor = NULL;
1401 }
1402 
1403 /*
1404  * Drives bay fan control loop
1405  */
1406 static void do_monitor_drives(struct drives_pid_state *state)
1407 {
1408  s32 temp, integral, derivative;
1409  s64 integ_p, deriv_p, prop_p, sum;
1410  int i, rc;
1411 
1412  if (--state->ticks != 0)
1413  return;
1414  state->ticks = DRIVES_PID_INTERVAL;
1415 
1416  DBG("drives:\n");
1417 
1418  /* Check fan status */
1419  rc = get_rpm_fan(DRIVES_FAN_RPM_INDEX, !RPM_PID_USE_ACTUAL_SPEED);
1420  if (rc < 0) {
1421  printk(KERN_WARNING "Error %d reading drives fan !\n", rc);
1422  /* XXX What do we do now ? */
1423  } else
1424  state->rpm = rc;
1425  DBG(" current rpm: %d\n", state->rpm);
1426 
1427  /* Get some sensor readings */
1429  DS1775_TEMP)) << 8;
1430  state->last_temp = temp;
1431  DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1433 
1434  /* Store temperature and error in history array */
1435  state->cur_sample = (state->cur_sample + 1) % DRIVES_PID_HISTORY_SIZE;
1436  state->sample_history[state->cur_sample] = temp;
1437  state->error_history[state->cur_sample] = temp - DRIVES_PID_INPUT_TARGET;
1438 
1439  /* If first loop, fill the history table */
1440  if (state->first) {
1441  for (i = 0; i < (DRIVES_PID_HISTORY_SIZE - 1); i++) {
1442  state->cur_sample = (state->cur_sample + 1) %
1444  state->sample_history[state->cur_sample] = temp;
1445  state->error_history[state->cur_sample] =
1446  temp - DRIVES_PID_INPUT_TARGET;
1447  }
1448  state->first = 0;
1449  }
1450 
1451  /* Calculate the integral term */
1452  sum = 0;
1453  integral = 0;
1454  for (i = 0; i < DRIVES_PID_HISTORY_SIZE; i++)
1455  integral += state->error_history[i];
1456  integral *= DRIVES_PID_INTERVAL;
1457  DBG(" integral: %08x\n", integral);
1458  integ_p = ((s64)DRIVES_PID_G_r) * (s64)integral;
1459  DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1460  sum += integ_p;
1461 
1462  /* Calculate the derivative term */
1463  derivative = state->error_history[state->cur_sample] -
1464  state->error_history[(state->cur_sample + DRIVES_PID_HISTORY_SIZE - 1)
1465  % DRIVES_PID_HISTORY_SIZE];
1466  derivative /= DRIVES_PID_INTERVAL;
1467  deriv_p = ((s64)DRIVES_PID_G_d) * (s64)derivative;
1468  DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1469  sum += deriv_p;
1470 
1471  /* Calculate the proportional term */
1472  prop_p = ((s64)DRIVES_PID_G_p) * (s64)(state->error_history[state->cur_sample]);
1473  DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1474  sum += prop_p;
1475 
1476  /* Scale sum */
1477  sum >>= 36;
1478 
1479  DBG(" sum: %d\n", (int)sum);
1480  state->rpm += (s32)sum;
1481 
1482  state->rpm = max(state->rpm, DRIVES_PID_OUTPUT_MIN);
1483  state->rpm = min(state->rpm, DRIVES_PID_OUTPUT_MAX);
1484 
1485  DBG("** DRIVES RPM: %d\n", (int)state->rpm);
1486  set_rpm_fan(DRIVES_FAN_RPM_INDEX, state->rpm);
1487 }
1488 
1489 /*
1490  * Initialize the state structure for the drives bay fan control loop
1491  */
1492 static int init_drives_state(struct drives_pid_state *state)
1493 {
1494  int err;
1495 
1496  state->ticks = 1;
1497  state->first = 1;
1498  state->rpm = 1000;
1499 
1500  state->monitor = attach_i2c_chip(DRIVES_DALLAS_ID, "drives_temp");
1501  if (state->monitor == NULL)
1502  return -ENODEV;
1503 
1504  err = device_create_file(&of_dev->dev, &dev_attr_drives_temperature);
1505  err |= device_create_file(&of_dev->dev, &dev_attr_drives_fan_rpm);
1506  if (err)
1507  printk(KERN_WARNING "Failed to create attribute file(s)"
1508  " for drives bay fan\n");
1509 
1510  return 0;
1511 }
1512 
1513 /*
1514  * Dispose of the state data for the drives control loop
1515  */
1516 static void dispose_drives_state(struct drives_pid_state *state)
1517 {
1518  if (state->monitor == NULL)
1519  return;
1520 
1521  device_remove_file(&of_dev->dev, &dev_attr_drives_temperature);
1522  device_remove_file(&of_dev->dev, &dev_attr_drives_fan_rpm);
1523 
1524  state->monitor = NULL;
1525 }
1526 
1527 /*
1528  * DIMMs temp control loop
1529  */
1530 static void do_monitor_dimms(struct dimm_pid_state *state)
1531 {
1532  s32 temp, integral, derivative, fan_min;
1533  s64 integ_p, deriv_p, prop_p, sum;
1534  int i;
1535 
1536  if (--state->ticks != 0)
1537  return;
1538  state->ticks = DIMM_PID_INTERVAL;
1539 
1540  DBG("DIMM:\n");
1541 
1542  DBG(" current value: %d\n", state->output);
1543 
1544  temp = read_lm87_reg(state->monitor, LM87_INT_TEMP);
1545  if (temp < 0)
1546  return;
1547  temp <<= 16;
1548  state->last_temp = temp;
1549  DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1551 
1552  /* Store temperature and error in history array */
1553  state->cur_sample = (state->cur_sample + 1) % DIMM_PID_HISTORY_SIZE;
1554  state->sample_history[state->cur_sample] = temp;
1555  state->error_history[state->cur_sample] = temp - DIMM_PID_INPUT_TARGET;
1556 
1557  /* If first loop, fill the history table */
1558  if (state->first) {
1559  for (i = 0; i < (DIMM_PID_HISTORY_SIZE - 1); i++) {
1560  state->cur_sample = (state->cur_sample + 1) %
1562  state->sample_history[state->cur_sample] = temp;
1563  state->error_history[state->cur_sample] =
1564  temp - DIMM_PID_INPUT_TARGET;
1565  }
1566  state->first = 0;
1567  }
1568 
1569  /* Calculate the integral term */
1570  sum = 0;
1571  integral = 0;
1572  for (i = 0; i < DIMM_PID_HISTORY_SIZE; i++)
1573  integral += state->error_history[i];
1574  integral *= DIMM_PID_INTERVAL;
1575  DBG(" integral: %08x\n", integral);
1576  integ_p = ((s64)DIMM_PID_G_r) * (s64)integral;
1577  DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1578  sum += integ_p;
1579 
1580  /* Calculate the derivative term */
1581  derivative = state->error_history[state->cur_sample] -
1582  state->error_history[(state->cur_sample + DIMM_PID_HISTORY_SIZE - 1)
1583  % DIMM_PID_HISTORY_SIZE];
1584  derivative /= DIMM_PID_INTERVAL;
1585  deriv_p = ((s64)DIMM_PID_G_d) * (s64)derivative;
1586  DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1587  sum += deriv_p;
1588 
1589  /* Calculate the proportional term */
1590  prop_p = ((s64)DIMM_PID_G_p) * (s64)(state->error_history[state->cur_sample]);
1591  DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1592  sum += prop_p;
1593 
1594  /* Scale sum */
1595  sum >>= 36;
1596 
1597  DBG(" sum: %d\n", (int)sum);
1598  state->output = (s32)sum;
1599  state->output = max(state->output, DIMM_PID_OUTPUT_MIN);
1600  state->output = min(state->output, DIMM_PID_OUTPUT_MAX);
1601  dimm_output_clamp = state->output;
1602 
1603  DBG("** DIMM clamp value: %d\n", (int)state->output);
1604 
1605  /* Backside PID is only every 5 seconds, force backside fan clamping now */
1606  fan_min = (dimm_output_clamp * 100) / 14000;
1607  fan_min = max(fan_min, backside_params.output_min);
1608  if (backside_state.pwm < fan_min) {
1609  backside_state.pwm = fan_min;
1610  DBG(" -> applying clamp to backside fan now: %d !\n", fan_min);
1611  set_pwm_fan(BACKSIDE_FAN_PWM_INDEX, fan_min);
1612  }
1613 }
1614 
1615 /*
1616  * Initialize the state structure for the DIMM temp control loop
1617  */
1618 static int init_dimms_state(struct dimm_pid_state *state)
1619 {
1620  state->ticks = 1;
1621  state->first = 1;
1622  state->output = 4000;
1623 
1624  state->monitor = attach_i2c_chip(XSERVE_DIMMS_LM87, "dimms_temp");
1625  if (state->monitor == NULL)
1626  return -ENODEV;
1627 
1628  if (device_create_file(&of_dev->dev, &dev_attr_dimms_temperature))
1629  printk(KERN_WARNING "Failed to create attribute file"
1630  " for DIMM temperature\n");
1631 
1632  return 0;
1633 }
1634 
1635 /*
1636  * Dispose of the state data for the DIMM control loop
1637  */
1638 static void dispose_dimms_state(struct dimm_pid_state *state)
1639 {
1640  if (state->monitor == NULL)
1641  return;
1642 
1643  device_remove_file(&of_dev->dev, &dev_attr_dimms_temperature);
1644 
1645  state->monitor = NULL;
1646 }
1647 
1648 /*
1649  * Slots fan control loop
1650  */
1651 static void do_monitor_slots(struct slots_pid_state *state)
1652 {
1653  s32 temp, integral, derivative;
1654  s64 integ_p, deriv_p, prop_p, sum;
1655  int i, rc;
1656 
1657  if (--state->ticks != 0)
1658  return;
1659  state->ticks = SLOTS_PID_INTERVAL;
1660 
1661  DBG("slots:\n");
1662 
1663  /* Check fan status */
1664  rc = get_pwm_fan(SLOTS_FAN_PWM_INDEX);
1665  if (rc < 0) {
1666  printk(KERN_WARNING "Error %d reading slots fan !\n", rc);
1667  /* XXX What do we do now ? */
1668  } else
1669  state->pwm = rc;
1670  DBG(" current pwm: %d\n", state->pwm);
1671 
1672  /* Get some sensor readings */
1674  DS1775_TEMP)) << 8;
1675  state->last_temp = temp;
1676  DBG(" temp: %d.%03d, target: %d.%03d\n", FIX32TOPRINT(temp),
1678 
1679  /* Store temperature and error in history array */
1680  state->cur_sample = (state->cur_sample + 1) % SLOTS_PID_HISTORY_SIZE;
1681  state->sample_history[state->cur_sample] = temp;
1682  state->error_history[state->cur_sample] = temp - SLOTS_PID_INPUT_TARGET;
1683 
1684  /* If first loop, fill the history table */
1685  if (state->first) {
1686  for (i = 0; i < (SLOTS_PID_HISTORY_SIZE - 1); i++) {
1687  state->cur_sample = (state->cur_sample + 1) %
1689  state->sample_history[state->cur_sample] = temp;
1690  state->error_history[state->cur_sample] =
1691  temp - SLOTS_PID_INPUT_TARGET;
1692  }
1693  state->first = 0;
1694  }
1695 
1696  /* Calculate the integral term */
1697  sum = 0;
1698  integral = 0;
1699  for (i = 0; i < SLOTS_PID_HISTORY_SIZE; i++)
1700  integral += state->error_history[i];
1701  integral *= SLOTS_PID_INTERVAL;
1702  DBG(" integral: %08x\n", integral);
1703  integ_p = ((s64)SLOTS_PID_G_r) * (s64)integral;
1704  DBG(" integ_p: %d\n", (int)(integ_p >> 36));
1705  sum += integ_p;
1706 
1707  /* Calculate the derivative term */
1708  derivative = state->error_history[state->cur_sample] -
1709  state->error_history[(state->cur_sample + SLOTS_PID_HISTORY_SIZE - 1)
1710  % SLOTS_PID_HISTORY_SIZE];
1711  derivative /= SLOTS_PID_INTERVAL;
1712  deriv_p = ((s64)SLOTS_PID_G_d) * (s64)derivative;
1713  DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
1714  sum += deriv_p;
1715 
1716  /* Calculate the proportional term */
1717  prop_p = ((s64)SLOTS_PID_G_p) * (s64)(state->error_history[state->cur_sample]);
1718  DBG(" prop_p: %d\n", (int)(prop_p >> 36));
1719  sum += prop_p;
1720 
1721  /* Scale sum */
1722  sum >>= 36;
1723 
1724  DBG(" sum: %d\n", (int)sum);
1725  state->pwm = (s32)sum;
1726 
1727  state->pwm = max(state->pwm, SLOTS_PID_OUTPUT_MIN);
1728  state->pwm = min(state->pwm, SLOTS_PID_OUTPUT_MAX);
1729 
1730  DBG("** DRIVES PWM: %d\n", (int)state->pwm);
1731  set_pwm_fan(SLOTS_FAN_PWM_INDEX, state->pwm);
1732 }
1733 
1734 /*
1735  * Initialize the state structure for the slots bay fan control loop
1736  */
1737 static int init_slots_state(struct slots_pid_state *state)
1738 {
1739  int err;
1740 
1741  state->ticks = 1;
1742  state->first = 1;
1743  state->pwm = 50;
1744 
1745  state->monitor = attach_i2c_chip(XSERVE_SLOTS_LM75, "slots_temp");
1746  if (state->monitor == NULL)
1747  return -ENODEV;
1748 
1749  err = device_create_file(&of_dev->dev, &dev_attr_slots_temperature);
1750  err |= device_create_file(&of_dev->dev, &dev_attr_slots_fan_pwm);
1751  if (err)
1752  printk(KERN_WARNING "Failed to create attribute file(s)"
1753  " for slots bay fan\n");
1754 
1755  return 0;
1756 }
1757 
1758 /*
1759  * Dispose of the state data for the slots control loop
1760  */
1761 static void dispose_slots_state(struct slots_pid_state *state)
1762 {
1763  if (state->monitor == NULL)
1764  return;
1765 
1766  device_remove_file(&of_dev->dev, &dev_attr_slots_temperature);
1767  device_remove_file(&of_dev->dev, &dev_attr_slots_fan_pwm);
1768 
1769  state->monitor = NULL;
1770 }
1771 
1772 
1773 static int call_critical_overtemp(void)
1774 {
1775  char *argv[] = { critical_overtemp_path, NULL };
1776  static char *envp[] = { "HOME=/",
1777  "TERM=linux",
1778  "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
1779  NULL };
1780 
1781  return call_usermodehelper(critical_overtemp_path,
1782  argv, envp, UMH_WAIT_EXEC);
1783 }
1784 
1785 
1786 /*
1787  * Here's the kernel thread that calls the various control loops
1788  */
1789 static int main_control_loop(void *x)
1790 {
1791  DBG("main_control_loop started\n");
1792 
1793  mutex_lock(&driver_lock);
1794 
1795  if (start_fcu() < 0) {
1796  printk(KERN_ERR "kfand: failed to start FCU\n");
1797  mutex_unlock(&driver_lock);
1798  goto out;
1799  }
1800 
1801  /* Set the PCI fan once for now on non-RackMac */
1802  if (!rackmac)
1804 
1805  /* Initialize ADCs */
1806  initialize_adc(&processor_state[0]);
1807  if (processor_state[1].monitor != NULL)
1808  initialize_adc(&processor_state[1]);
1809 
1810  fcu_tickle_ticks = FCU_TICKLE_TICKS;
1811 
1812  mutex_unlock(&driver_lock);
1813 
1814  while (state == state_attached) {
1815  unsigned long elapsed, start;
1816 
1817  start = jiffies;
1818 
1819  mutex_lock(&driver_lock);
1820 
1821  /* Tickle the FCU just in case */
1822  if (--fcu_tickle_ticks < 0) {
1823  fcu_tickle_ticks = FCU_TICKLE_TICKS;
1824  tickle_fcu();
1825  }
1826 
1827  /* First, we always calculate the new DIMMs state on an Xserve */
1828  if (rackmac)
1829  do_monitor_dimms(&dimms_state);
1830 
1831  /* Then, the CPUs */
1832  if (cpu_pid_type == CPU_PID_TYPE_COMBINED)
1833  do_monitor_cpu_combined();
1834  else if (cpu_pid_type == CPU_PID_TYPE_RACKMAC) {
1835  do_monitor_cpu_rack(&processor_state[0]);
1836  if (processor_state[1].monitor != NULL)
1837  do_monitor_cpu_rack(&processor_state[1]);
1838  // better deal with UP
1839  } else {
1840  do_monitor_cpu_split(&processor_state[0]);
1841  if (processor_state[1].monitor != NULL)
1842  do_monitor_cpu_split(&processor_state[1]);
1843  // better deal with UP
1844  }
1845  /* Then, the rest */
1846  do_monitor_backside(&backside_state);
1847  if (rackmac)
1848  do_monitor_slots(&slots_state);
1849  else
1850  do_monitor_drives(&drives_state);
1851  mutex_unlock(&driver_lock);
1852 
1853  if (critical_state == 1) {
1854  printk(KERN_WARNING "Temperature control detected a critical condition\n");
1855  printk(KERN_WARNING "Attempting to shut down...\n");
1856  if (call_critical_overtemp()) {
1857  printk(KERN_WARNING "Can't call %s, power off now!\n",
1858  critical_overtemp_path);
1860  }
1861  }
1862  if (critical_state > 0)
1863  critical_state++;
1864  if (critical_state > MAX_CRITICAL_STATE) {
1865  printk(KERN_WARNING "Shutdown timed out, power off now !\n");
1867  }
1868 
1869  // FIXME: Deal with signals
1870  elapsed = jiffies - start;
1871  if (elapsed < HZ)
1873  }
1874 
1875  out:
1876  DBG("main_control_loop ended\n");
1877 
1878  ctrl_task = 0;
1879  complete_and_exit(&ctrl_complete, 0);
1880 }
1881 
1882 /*
1883  * Dispose the control loops when tearing down
1884  */
1885 static void dispose_control_loops(void)
1886 {
1887  dispose_processor_state(&processor_state[0]);
1888  dispose_processor_state(&processor_state[1]);
1889  dispose_backside_state(&backside_state);
1890  dispose_drives_state(&drives_state);
1891  dispose_slots_state(&slots_state);
1892  dispose_dimms_state(&dimms_state);
1893 }
1894 
1895 /*
1896  * Create the control loops. U3-0 i2c bus is up, so we can now
1897  * get to the various sensors
1898  */
1899 static int create_control_loops(void)
1900 {
1901  struct device_node *np;
1902 
1903  /* Count CPUs from the device-tree, we don't care how many are
1904  * actually used by Linux
1905  */
1906  cpu_count = 0;
1907  for (np = NULL; NULL != (np = of_find_node_by_type(np, "cpu"));)
1908  cpu_count++;
1909 
1910  DBG("counted %d CPUs in the device-tree\n", cpu_count);
1911 
1912  /* Decide the type of PID algorithm to use based on the presence of
1913  * the pumps, though that may not be the best way, that is good enough
1914  * for now
1915  */
1916  if (rackmac)
1917  cpu_pid_type = CPU_PID_TYPE_RACKMAC;
1918  else if (of_machine_is_compatible("PowerMac7,3")
1919  && (cpu_count > 1)
1920  && fcu_fans[CPUA_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID
1921  && fcu_fans[CPUB_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID) {
1922  printk(KERN_INFO "Liquid cooling pumps detected, using new algorithm !\n");
1923  cpu_pid_type = CPU_PID_TYPE_COMBINED;
1924  } else
1925  cpu_pid_type = CPU_PID_TYPE_SPLIT;
1926 
1927  /* Create control loops for everything. If any fail, everything
1928  * fails
1929  */
1930  if (init_processor_state(&processor_state[0], 0))
1931  goto fail;
1932  if (cpu_pid_type == CPU_PID_TYPE_COMBINED)
1933  fetch_cpu_pumps_minmax();
1934 
1935  if (cpu_count > 1 && init_processor_state(&processor_state[1], 1))
1936  goto fail;
1937  if (init_backside_state(&backside_state))
1938  goto fail;
1939  if (rackmac && init_dimms_state(&dimms_state))
1940  goto fail;
1941  if (rackmac && init_slots_state(&slots_state))
1942  goto fail;
1943  if (!rackmac && init_drives_state(&drives_state))
1944  goto fail;
1945 
1946  DBG("all control loops up !\n");
1947 
1948  return 0;
1949 
1950  fail:
1951  DBG("failure creating control loops, disposing\n");
1952 
1953  dispose_control_loops();
1954 
1955  return -ENODEV;
1956 }
1957 
1958 /*
1959  * Start the control loops after everything is up, that is create
1960  * the thread that will make them run
1961  */
1962 static void start_control_loops(void)
1963 {
1964  init_completion(&ctrl_complete);
1965 
1966  ctrl_task = kthread_run(main_control_loop, NULL, "kfand");
1967 }
1968 
1969 /*
1970  * Stop the control loops when tearing down
1971  */
1972 static void stop_control_loops(void)
1973 {
1974  if (ctrl_task)
1975  wait_for_completion(&ctrl_complete);
1976 }
1977 
1978 /*
1979  * Attach to the i2c FCU after detecting U3-1 bus
1980  */
1981 static int attach_fcu(void)
1982 {
1983  fcu = attach_i2c_chip(FAN_CTRLER_ID, "fcu");
1984  if (fcu == NULL)
1985  return -ENODEV;
1986 
1987  DBG("FCU attached\n");
1988 
1989  return 0;
1990 }
1991 
1992 /*
1993  * Detach from the i2c FCU when tearing down
1994  */
1995 static void detach_fcu(void)
1996 {
1997  fcu = NULL;
1998 }
1999 
2000 /*
2001  * Attach to the i2c controller. We probe the various chips based
2002  * on the device-tree nodes and build everything for the driver to
2003  * run, we then kick the driver monitoring thread
2004  */
2005 static int therm_pm72_attach(struct i2c_adapter *adapter)
2006 {
2007  mutex_lock(&driver_lock);
2008 
2009  /* Check state */
2010  if (state == state_detached)
2011  state = state_attaching;
2012  if (state != state_attaching) {
2013  mutex_unlock(&driver_lock);
2014  return 0;
2015  }
2016 
2017  /* Check if we are looking for one of these */
2018  if (u3_0 == NULL && !strcmp(adapter->name, "u3 0")) {
2019  u3_0 = adapter;
2020  DBG("found U3-0\n");
2021  if (k2 || !rackmac)
2022  if (create_control_loops())
2023  u3_0 = NULL;
2024  } else if (u3_1 == NULL && !strcmp(adapter->name, "u3 1")) {
2025  u3_1 = adapter;
2026  DBG("found U3-1, attaching FCU\n");
2027  if (attach_fcu())
2028  u3_1 = NULL;
2029  } else if (k2 == NULL && !strcmp(adapter->name, "mac-io 0")) {
2030  k2 = adapter;
2031  DBG("Found K2\n");
2032  if (u3_0 && rackmac)
2033  if (create_control_loops())
2034  k2 = NULL;
2035  }
2036  /* We got all we need, start control loops */
2037  if (u3_0 != NULL && u3_1 != NULL && (k2 || !rackmac)) {
2038  DBG("everything up, starting control loops\n");
2039  state = state_attached;
2040  start_control_loops();
2041  }
2042  mutex_unlock(&driver_lock);
2043 
2044  return 0;
2045 }
2046 
2047 static int therm_pm72_probe(struct i2c_client *client,
2048  const struct i2c_device_id *id)
2049 {
2050  /* Always succeed, the real work was done in therm_pm72_attach() */
2051  return 0;
2052 }
2053 
2054 /*
2055  * Called when any of the devices which participates into thermal management
2056  * is going away.
2057  */
2058 static int therm_pm72_remove(struct i2c_client *client)
2059 {
2060  struct i2c_adapter *adapter = client->adapter;
2061 
2062  mutex_lock(&driver_lock);
2063 
2064  if (state != state_detached)
2065  state = state_detaching;
2066 
2067  /* Stop control loops if any */
2068  DBG("stopping control loops\n");
2069  mutex_unlock(&driver_lock);
2070  stop_control_loops();
2071  mutex_lock(&driver_lock);
2072 
2073  if (u3_0 != NULL && !strcmp(adapter->name, "u3 0")) {
2074  DBG("lost U3-0, disposing control loops\n");
2075  dispose_control_loops();
2076  u3_0 = NULL;
2077  }
2078 
2079  if (u3_1 != NULL && !strcmp(adapter->name, "u3 1")) {
2080  DBG("lost U3-1, detaching FCU\n");
2081  detach_fcu();
2082  u3_1 = NULL;
2083  }
2084  if (u3_0 == NULL && u3_1 == NULL)
2085  state = state_detached;
2086 
2087  mutex_unlock(&driver_lock);
2088 
2089  return 0;
2090 }
2091 
2092 /*
2093  * i2c_driver structure to attach to the host i2c controller
2094  */
2095 
2096 static const struct i2c_device_id therm_pm72_id[] = {
2097  /*
2098  * Fake device name, thermal management is done by several
2099  * chips but we don't need to differentiate between them at
2100  * this point.
2101  */
2102  { "therm_pm72", 0 },
2103  { }
2104 };
2105 
2106 static struct i2c_driver therm_pm72_driver = {
2107  .driver = {
2108  .name = "therm_pm72",
2109  },
2110  .attach_adapter = therm_pm72_attach,
2111  .probe = therm_pm72_probe,
2112  .remove = therm_pm72_remove,
2113  .id_table = therm_pm72_id,
2114 };
2115 
2116 static int fan_check_loc_match(const char *loc, int fan)
2117 {
2118  char tmp[64];
2119  char *c, *e;
2120 
2121  strlcpy(tmp, fcu_fans[fan].loc, 64);
2122 
2123  c = tmp;
2124  for (;;) {
2125  e = strchr(c, ',');
2126  if (e)
2127  *e = 0;
2128  if (strcmp(loc, c) == 0)
2129  return 1;
2130  if (e == NULL)
2131  break;
2132  c = e + 1;
2133  }
2134  return 0;
2135 }
2136 
2137 static void fcu_lookup_fans(struct device_node *fcu_node)
2138 {
2139  struct device_node *np = NULL;
2140  int i;
2141 
2142  /* The table is filled by default with values that are suitable
2143  * for the old machines without device-tree informations. We scan
2144  * the device-tree and override those values with whatever is
2145  * there
2146  */
2147 
2148  DBG("Looking up FCU controls in device-tree...\n");
2149 
2150  while ((np = of_get_next_child(fcu_node, np)) != NULL) {
2151  int type = -1;
2152  const char *loc;
2153  const u32 *reg;
2154 
2155  DBG(" control: %s, type: %s\n", np->name, np->type);
2156 
2157  /* Detect control type */
2158  if (!strcmp(np->type, "fan-rpm-control") ||
2159  !strcmp(np->type, "fan-rpm"))
2160  type = FCU_FAN_RPM;
2161  if (!strcmp(np->type, "fan-pwm-control") ||
2162  !strcmp(np->type, "fan-pwm"))
2163  type = FCU_FAN_PWM;
2164  /* Only care about fans for now */
2165  if (type == -1)
2166  continue;
2167 
2168  /* Lookup for a matching location */
2169  loc = of_get_property(np, "location", NULL);
2170  reg = of_get_property(np, "reg", NULL);
2171  if (loc == NULL || reg == NULL)
2172  continue;
2173  DBG(" matching location: %s, reg: 0x%08x\n", loc, *reg);
2174 
2175  for (i = 0; i < FCU_FAN_COUNT; i++) {
2176  int fan_id;
2177 
2178  if (!fan_check_loc_match(loc, i))
2179  continue;
2180  DBG(" location match, index: %d\n", i);
2181  fcu_fans[i].id = FCU_FAN_ABSENT_ID;
2182  if (type != fcu_fans[i].type) {
2183  printk(KERN_WARNING "therm_pm72: Fan type mismatch "
2184  "in device-tree for %s\n", np->full_name);
2185  break;
2186  }
2187  if (type == FCU_FAN_RPM)
2188  fan_id = ((*reg) - 0x10) / 2;
2189  else
2190  fan_id = ((*reg) - 0x30) / 2;
2191  if (fan_id > 7) {
2192  printk(KERN_WARNING "therm_pm72: Can't parse "
2193  "fan ID in device-tree for %s\n", np->full_name);
2194  break;
2195  }
2196  DBG(" fan id -> %d, type -> %d\n", fan_id, type);
2197  fcu_fans[i].id = fan_id;
2198  }
2199  }
2200 
2201  /* Now dump the array */
2202  printk(KERN_INFO "Detected fan controls:\n");
2203  for (i = 0; i < FCU_FAN_COUNT; i++) {
2204  if (fcu_fans[i].id == FCU_FAN_ABSENT_ID)
2205  continue;
2206  printk(KERN_INFO " %d: %s fan, id %d, location: %s\n", i,
2207  fcu_fans[i].type == FCU_FAN_RPM ? "RPM" : "PWM",
2208  fcu_fans[i].id, fcu_fans[i].loc);
2209  }
2210 }
2211 
2212 static int fcu_of_probe(struct platform_device* dev)
2213 {
2214  state = state_detached;
2215  of_dev = dev;
2216 
2217  dev_info(&dev->dev, "PowerMac G5 Thermal control driver %s\n", VERSION);
2218 
2219  /* Lookup the fans in the device tree */
2220  fcu_lookup_fans(dev->dev.of_node);
2221 
2222  /* Add the driver */
2223  return i2c_add_driver(&therm_pm72_driver);
2224 }
2225 
2226 static int fcu_of_remove(struct platform_device* dev)
2227 {
2228  i2c_del_driver(&therm_pm72_driver);
2229 
2230  return 0;
2231 }
2232 
2233 static const struct of_device_id fcu_match[] =
2234 {
2235  {
2236  .type = "fcu",
2237  },
2238  {},
2239 };
2240 MODULE_DEVICE_TABLE(of, fcu_match);
2241 
2242 static struct platform_driver fcu_of_platform_driver =
2243 {
2244  .driver = {
2245  .name = "temperature",
2246  .owner = THIS_MODULE,
2247  .of_match_table = fcu_match,
2248  },
2249  .probe = fcu_of_probe,
2250  .remove = fcu_of_remove
2251 };
2252 
2253 /*
2254  * Check machine type, attach to i2c controller
2255  */
2256 static int __init therm_pm72_init(void)
2257 {
2258  rackmac = of_machine_is_compatible("RackMac3,1");
2259 
2260  if (!of_machine_is_compatible("PowerMac7,2") &&
2261  !of_machine_is_compatible("PowerMac7,3") &&
2262  !rackmac)
2263  return -ENODEV;
2264 
2265  return platform_driver_register(&fcu_of_platform_driver);
2266 }
2267 
2268 static void __exit therm_pm72_exit(void)
2269 {
2270  platform_driver_unregister(&fcu_of_platform_driver);
2271 }
2272 
2273 module_init(therm_pm72_init);
2274 module_exit(therm_pm72_exit);
2275 
2276 MODULE_AUTHOR("Benjamin Herrenschmidt <[email protected]>");
2277 MODULE_DESCRIPTION("Driver for Apple's PowerMac G5 thermal control");
2278 MODULE_LICENSE("GPL");
2279