bbc_envctrl.c

linux-2.6.15.6

/* $Id: bbc_envctrl.c,v 1.4 2001/04/06 16:48:08 davem Exp $
 * bbc_envctrl.c: UltraSPARC-III environment control driver.
 *
 * Copyright (C) 2001 David S. Miller ([e-mail])
 */

#define __KERNEL_SYSCALLS__
static int errno;

#include <linux/kernel.h>
#include <linux/kthread.h>
#include <linux/sched.h>
#include <linux/slab.h>
#include <linux/delay.h>
#include <asm/oplib.h>
#include <asm/ebus.h>

#include "bbc_i2c.h"
#include "max1617.h"

#undef ENVCTRL_TRACE

/* WARNING: Making changes to this driver is very dangerous.
 *          If you misprogram the sensor chips they can
 *          cut the power on you instantly.
 */

/* Two temperature sensors exist in the SunBLADE-1000 enclosure.
 * Both are implemented using max1617 i2c devices.  Each max1617
 * monitors 2 temperatures, one for one of the cpu dies and the other
 * for the ambient temperature.
 *
 * The max1617 is capable of being programmed with power-off
 * temperature values, one low limit and one high limit.  These
 * can be controlled independently for the cpu or ambient temperature.
 * If a limit is violated, the power is simply shut off.  The frequency
 * with which the max1617 does temperature sampling can be controlled
 * as well.
 *
 * Three fans exist inside the machine, all three are controlled with
 * an i2c digital to analog converter.  There is a fan directed at the
 * two processor slots, another for the rest of the enclosure, and the
 * third is for the power supply.  The first two fans may be speed
 * controlled by changing the voltage fed to them.  The third fan may
 * only be completely off or on.  The third fan is meant to only be
 * disabled/enabled when entering/exiting the lowest power-saving
 * mode of the machine.
 *
 * An environmental control kernel thread periodically monitors all
 * temperature sensors.  Based upon the samples it will adjust the
 * fan speeds to try and keep the system within a certain temperature
 * range (the goal being to make the fans as quiet as possible without
 * allowing the system to get too hot).
 *
 * If the temperature begins to rise/fall outside of the acceptable
 * operating range, a periodic warning will be sent to the kernel log.
 * The fans will be put on full blast to attempt to deal with this
 * situation.  After exceeding the acceptable operating range by a
 * certain threshold, the kernel thread will shut down the system.
 * Here, the thread is attempting to shut the machine down cleanly
 * before the hardware based power-off event is triggered.
 */

/* These settings are in Celsius.  We use these defaults only
 * if we cannot interrogate the cpu-fru SEEPROM.
 */
struct temp_limits {
	s8 high_pwroff, high_shutdown, high_warn;
	s8 low_warn, low_shutdown, low_pwroff;
};

static struct temp_limits cpu_temp_limits[2] = {
	{ 100, 85, 80, 5, -5, -10 },
	{ 100, 85, 80, 5, -5, -10 },
};

static struct temp_limits amb_temp_limits[2] = {
	{ 65, 55, 40, 5, -5, -10 },
	{ 65, 55, 40, 5, -5, -10 },
};

enum fan_action { FAN_SLOWER, FAN_SAME, FAN_FASTER, FAN_FULLBLAST, FAN_STATE_MAX };

struct bbc_cpu_temperature {
	struct bbc_cpu_temperature	*next;

	struct bbc_i2c_client		*client;
	int				index;

	/* Current readings, and history. */
	s8				curr_cpu_temp;
	s8				curr_amb_temp;
	s8				prev_cpu_temp;
	s8				prev_amb_temp;
	s8				avg_cpu_temp;
	s8				avg_amb_temp;

	int				sample_tick;

	enum fan_action			fan_todo[2];
#define FAN_AMBIENT	0
#define FAN_CPU		1
};

struct bbc_cpu_temperature *all_bbc_temps;

struct bbc_fan_control {
	struct bbc_fan_control 	*next;

	struct bbc_i2c_client 	*client;
	int 			index;

	int			psupply_fan_on;
	int			cpu_fan_speed;
	int			system_fan_speed;
};

struct bbc_fan_control *all_bbc_fans;

#define CPU_FAN_REG	0xf0
#define SYS_FAN_REG	0xf2
#define PSUPPLY_FAN_REG	0xf4

#define FAN_SPEED_MIN	0x0c
#define FAN_SPEED_MAX	0x3f

#define PSUPPLY_FAN_ON	0x1f
#define PSUPPLY_FAN_OFF	0x00

static void set_fan_speeds(struct bbc_fan_control *fp)
{
	/* Put temperatures into range so we don't mis-program
	 * the hardware.
	 */
	if (fp->cpu_fan_speed < FAN_SPEED_MIN)
		fp->cpu_fan_speed = FAN_SPEED_MIN;
	if (fp->cpu_fan_speed > FAN_SPEED_MAX)
		fp->cpu_fan_speed = FAN_SPEED_MAX;
	if (fp->system_fan_speed < FAN_SPEED_MIN)
		fp->system_fan_speed = FAN_SPEED_MIN;
	if (fp->system_fan_speed > FAN_SPEED_MAX)
		fp->system_fan_speed = FAN_SPEED_MAX;
#ifdef ENVCTRL_TRACE
	printk("fan%d: Changed fan speed to cpu(%02x) sys(%02x)\n",
	       fp->index,
	       fp->cpu_fan_speed, fp->system_fan_speed);
#endif

	bbc_i2c_writeb(fp->client, fp->cpu_fan_speed, CPU_FAN_REG);
	bbc_i2c_writeb(fp->client, fp->system_fan_speed, SYS_FAN_REG);
	bbc_i2c_writeb(fp->client,
		       (fp->psupply_fan_on ?
			PSUPPLY_FAN_ON : PSUPPLY_FAN_OFF),
		       PSUPPLY_FAN_REG);
}

static void get_current_temps(struct bbc_cpu_temperature *tp)
{
	tp->prev_amb_temp = tp->curr_amb_temp;
	bbc_i2c_readb(tp->client,
		      (unsigned char *) &tp->curr_amb_temp,
		      MAX1617_AMB_TEMP);
	tp->prev_cpu_temp = tp->curr_cpu_temp;
	bbc_i2c_readb(tp->client,
		      (unsigned char *) &tp->curr_cpu_temp,
		      MAX1617_CPU_TEMP);
#ifdef ENVCTRL_TRACE
	printk("temp%d: cpu(%d C) amb(%d C)\n",
	       tp->index,
	       (int) tp->curr_cpu_temp, (int) tp->curr_amb_temp);
#endif
}


static void do_envctrl_shutdown(struct bbc_cpu_temperature *tp)
{
	static int shutting_down = 0;
	static char *envp[] = { "HOME=/", "TERM=linux", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
	char *argv[] = { "/sbin/shutdown", "-h", "now", NULL };
	char *type = "???";
	s8 val = -1;

	if (shutting_down != 0)
		return;

	if (tp->curr_amb_temp >= amb_temp_limits[tp->index].high_shutdown ||
	    tp->curr_amb_temp < amb_temp_limits[tp->index].low_shutdown) {
		type = "ambient";
		val = tp->curr_amb_temp;
	} else if (tp->curr_cpu_temp >= cpu_temp_limits[tp->index].high_shutdown ||
		   tp->curr_cpu_temp < cpu_temp_limits[tp->index].low_shutdown) {
		type = "CPU";
		val = tp->curr_cpu_temp;
	}

	printk(KERN_CRIT "temp%d: Outside of safe %s "
	       "operating temperature, %d C.\n",
	       tp->index, type, val);

	printk(KERN_CRIT "kenvctrld: Shutting down the system now.\n");

	shutting_down = 1;
	if (execve("/sbin/shutdown", argv, envp) < 0)
		printk(KERN_CRIT "envctrl: shutdown execution failed\n");
}

#define WARN_INTERVAL	(30 * HZ)

static void analyze_ambient_temp(struct bbc_cpu_temperature *tp, unsigned long *last_warn, int tick)
{
	int ret = 0;

	if (time_after(jiffies, (*last_warn + WARN_INTERVAL))) {
		if (tp->curr_amb_temp >=
		    amb_temp_limits[tp->index].high_warn) {
			printk(KERN_WARNING "temp%d: "
			       "Above safe ambient operating temperature, %d C.\n",
			       tp->index, (int) tp->curr_amb_temp);
			ret = 1;
		} else if (tp->curr_amb_temp <
			   amb_temp_limits[tp->index].low_warn) {
			printk(KERN_WARNING "temp%d: "
			       "Below safe ambient operating temperature, %d C.\n",
			       tp->index, (int) tp->curr_amb_temp);
			ret = 1;
		}
		if (ret)
			*last_warn = jiffies;
	} else if (tp->curr_amb_temp >= amb_temp_limits[tp->index].high_warn ||
		   tp->curr_amb_temp < amb_temp_limits[tp->index].low_warn)
		ret = 1;

	/* Now check the shutdown limits. */
	if (tp->curr_amb_temp >= amb_temp_limits[tp->index].high_shutdown ||
	    tp->curr_amb_temp < amb_temp_limits[tp->index].low_shutdown) {
		do_envctrl_shutdown(tp);
		ret = 1;
	}

	if (ret) {
		tp->fan_todo[FAN_AMBIENT] = FAN_FULLBLAST;
	} else if ((tick & (8 - 1)) == 0) {
		s8 amb_goal_hi = amb_temp_limits[tp->index].high_warn - 10;
		s8 amb_goal_lo;

		amb_goal_lo = amb_goal_hi - 3;

		/* We do not try to avoid 'too cold' events.  Basically we
		 * only try to deal with over-heating and fan noise reduction.
		 */
		if (tp->avg_amb_temp < amb_goal_hi) {
			if (tp->avg_amb_temp >= amb_goal_lo)
				tp->fan_todo[FAN_AMBIENT] = FAN_SAME;
			else
				tp->fan_todo[FAN_AMBIENT] = FAN_SLOWER;
		} else {
			tp->fan_todo[FAN_AMBIENT] = FAN_FASTER;
		}
	} else {
		tp->fan_todo[FAN_AMBIENT] = FAN_SAME;
	}
}

static void analyze_cpu_temp(struct bbc_cpu_temperature *tp, unsigned long *last_warn, int tick)
{
	int ret = 0;

	if (time_after(jiffies, (*last_warn + WARN_INTERVAL))) {
		if (tp->curr_cpu_temp >=
		    cpu_temp_limits[tp->index].high_warn) {
			printk(KERN_WARNING "temp%d: "
			       "Above safe CPU operating temperature, %d C.\n",
			       tp->index, (int) tp->curr_cpu_temp);
			ret = 1;
		} else if (tp->curr_cpu_temp <
			   cpu_temp_limits[tp->index].low_warn) {
			printk(KERN_WARNING "temp%d: "
			       "Below safe CPU operating temperature, %d C.\n",
			       tp->index, (int) tp->curr_cpu_temp);
			ret = 1;
		}
		if (ret)
			*last_warn = jiffies;
	} else if (tp->curr_cpu_temp >= cpu_temp_limits[tp->index].high_warn ||
		   tp->curr_cpu_temp < cpu_temp_limits[tp->index].low_warn)
		ret = 1;

	/* Now check the shutdown limits. */
	if (tp->curr_cpu_temp >= cpu_temp_limits[tp->index].high_shutdown ||
	    tp->curr_cpu_temp < cpu_temp_limits[tp->index].low_shutdown) {
		do_envctrl_shutdown(tp);
		ret = 1;
	}

	if (ret) {
		tp->fan_todo[FAN_CPU] = FAN_FULLBLAST;
	} else if ((tick & (8 - 1)) == 0) {
		s8 cpu_goal_hi = cpu_temp_limits[tp->index].high_warn - 10;
		s8 cpu_goal_lo;

		cpu_goal_lo = cpu_goal_hi - 3;

		/* We do not try to avoid 'too cold' events.  Basically we
		 * only try to deal with over-heating and fan noise reduction.
		 */
		if (tp->avg_cpu_temp < cpu_goal_hi) {
			if (tp->avg_cpu_temp >= cpu_goal_lo)
				tp->fan_todo[FAN_CPU] = FAN_SAME;
			else
				tp->fan_todo[FAN_CPU] = FAN_SLOWER;
		} else {
			tp->fan_todo[FAN_CPU] = FAN_FASTER;
		}
	} else {