summaryrefslogtreecommitdiffstats
path: root/drivers/macintosh/therm_pm72.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 15:20:36 -0700
committerLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 15:20:36 -0700
commit1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch)
tree0bba044c4ce775e45a88a51686b5d9f90697ea9d /drivers/macintosh/therm_pm72.c
Linux-2.6.12-rc2
Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
Diffstat (limited to 'drivers/macintosh/therm_pm72.c')
-rw-r--r--drivers/macintosh/therm_pm72.c2080
1 files changed, 2080 insertions, 0 deletions
diff --git a/drivers/macintosh/therm_pm72.c b/drivers/macintosh/therm_pm72.c
new file mode 100644
index 000000000000..82336a5a5474
--- /dev/null
+++ b/drivers/macintosh/therm_pm72.c
@@ -0,0 +1,2080 @@
+/*
+ * Device driver for the thermostats & fan controller of the
+ * Apple G5 "PowerMac7,2" desktop machines.
+ *
+ * (c) Copyright IBM Corp. 2003-2004
+ *
+ * Maintained by: Benjamin Herrenschmidt
+ * <benh@kernel.crashing.org>
+ *
+ *
+ * The algorithm used is the PID control algorithm, used the same
+ * way the published Darwin code does, using the same values that
+ * are present in the Darwin 7.0 snapshot property lists.
+ *
+ * As far as the CPUs control loops are concerned, I use the
+ * calibration & PID constants provided by the EEPROM,
+ * I do _not_ embed any value from the property lists, as the ones
+ * provided by Darwin 7.0 seem to always have an older version that
+ * what I've seen on the actual computers.
+ * It would be interesting to verify that though. Darwin has a
+ * version code of 1.0.0d11 for all control loops it seems, while
+ * so far, the machines EEPROMs contain a dataset versioned 1.0.0f
+ *
+ * Darwin doesn't provide source to all parts, some missing
+ * bits like the AppleFCU driver or the actual scale of some
+ * of the values returned by sensors had to be "guessed" some
+ * way... or based on what Open Firmware does.
+ *
+ * I didn't yet figure out how to get the slots power consumption
+ * out of the FCU, so that part has not been implemented yet and
+ * the slots fan is set to a fixed 50% PWM, hoping this value is
+ * safe enough ...
+ *
+ * Note: I have observed strange oscillations of the CPU control
+ * loop on a dual G5 here. When idle, the CPU exhaust fan tend to
+ * oscillates slowly (over several minutes) between the minimum
+ * of 300RPMs and approx. 1000 RPMs. I don't know what is causing
+ * this, it could be some incorrect constant or an error in the
+ * way I ported the algorithm, or it could be just normal. I
+ * don't have full understanding on the way Apple tweaked the PID
+ * algorithm for the CPU control, it is definitely not a standard
+ * implementation...
+ *
+ * TODO: - Check MPU structure version/signature
+ * - Add things like /sbin/overtemp for non-critical
+ * overtemp conditions so userland can take some policy
+ * decisions, like slewing down CPUs
+ * - Deal with fan and i2c failures in a better way
+ * - Maybe do a generic PID based on params used for
+ * U3 and Drives ? Definitely need to factor code a bit
+ * bettter... also make sensor detection more robust using
+ * the device-tree to probe for them
+ * - Figure out how to get the slots consumption and set the
+ * slots fan accordingly
+ *
+ * History:
+ *
+ * Nov. 13, 2003 : 0.5
+ * - First release
+ *
+ * Nov. 14, 2003 : 0.6
+ * - Read fan speed from FCU, low level fan routines now deal
+ * with errors & check fan status, though higher level don't
+ * do much.
+ * - Move a bunch of definitions to .h file
+ *
+ * Nov. 18, 2003 : 0.7
+ * - Fix build on ppc64 kernel
+ * - Move back statics definitions to .c file
+ * - Avoid calling schedule_timeout with a negative number
+ *
+ * Dec. 18, 2003 : 0.8
+ * - Fix typo when reading back fan speed on 2 CPU machines
+ *
+ * Mar. 11, 2004 : 0.9
+ * - Rework code accessing the ADC chips, make it more robust and
+ * closer to the chip spec. Also make sure it is configured properly,
+ * I've seen yet unexplained cases where on startup, I would have stale
+ * values in the configuration register
+ * - Switch back to use of target fan speed for PID, thus lowering
+ * pressure on i2c
+ *
+ * Oct. 20, 2004 : 1.1
+ * - Add device-tree lookup for fan IDs, should detect liquid cooling
+ * pumps when present
+ * - Enable driver for PowerMac7,3 machines
+ * - Split the U3/Backside cooling on U3 & U3H versions as Darwin does
+ * - Add new CPU cooling algorithm for machines with liquid cooling
+ * - Workaround for some PowerMac7,3 with empty "fan" node in the devtree
+ * - Fix a signed/unsigned compare issue in some PID loops
+ *
+ * Mar. 10, 2005 : 1.2
+ * - Add basic support for Xserve G5
+ * - Retreive pumps min/max from EEPROM image in device-tree (broken)
+ * - Use min/max macros here or there
+ * - Latest darwin updated U3H min fan speed to 20% PWM
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/types.h>
+#include <linux/module.h>
+#include <linux/errno.h>
+#include <linux/kernel.h>
+#include <linux/delay.h>
+#include <linux/sched.h>
+#include <linux/i2c.h>
+#include <linux/slab.h>
+#include <linux/init.h>
+#include <linux/spinlock.h>
+#include <linux/smp_lock.h>
+#include <linux/wait.h>
+#include <linux/reboot.h>
+#include <linux/kmod.h>
+#include <linux/i2c.h>
+#include <linux/i2c-dev.h>
+#include <asm/prom.h>
+#include <asm/machdep.h>
+#include <asm/io.h>
+#include <asm/system.h>
+#include <asm/sections.h>
+#include <asm/of_device.h>
+
+#include "therm_pm72.h"
+
+#define VERSION "1.2b2"
+
+#undef DEBUG
+
+#ifdef DEBUG
+#define DBG(args...) printk(args)
+#else
+#define DBG(args...) do { } while(0)
+#endif
+
+
+/*
+ * Driver statics
+ */
+
+static struct of_device * of_dev;
+static struct i2c_adapter * u3_0;
+static struct i2c_adapter * u3_1;
+static struct i2c_adapter * k2;
+static struct i2c_client * fcu;
+static struct cpu_pid_state cpu_state[2];
+static struct basckside_pid_params backside_params;
+static struct backside_pid_state backside_state;
+static struct drives_pid_state drives_state;
+static struct dimm_pid_state dimms_state;
+static int state;
+static int cpu_count;
+static int cpu_pid_type;
+static pid_t ctrl_task;
+static struct completion ctrl_complete;
+static int critical_state;
+static int rackmac;
+static s32 dimm_output_clamp;
+
+static DECLARE_MUTEX(driver_lock);
+
+/*
+ * We have 3 types of CPU PID control. One is "split" old style control
+ * for intake & exhaust fans, the other is "combined" control for both
+ * CPUs that also deals with the pumps when present. To be "compatible"
+ * with OS X at this point, we only use "COMBINED" on the machines that
+ * are identified as having the pumps (though that identification is at
+ * least dodgy). Ultimately, we could probably switch completely to this
+ * algorithm provided we hack it to deal with the UP case
+ */
+#define CPU_PID_TYPE_SPLIT 0
+#define CPU_PID_TYPE_COMBINED 1
+#define CPU_PID_TYPE_RACKMAC 2
+
+/*
+ * This table describes all fans in the FCU. The "id" and "type" values
+ * are defaults valid for all earlier machines. Newer machines will
+ * eventually override the table content based on the device-tree
+ */
+struct fcu_fan_table
+{
+ char* loc; /* location code */
+ int type; /* 0 = rpm, 1 = pwm, 2 = pump */
+ int id; /* id or -1 */
+};
+
+#define FCU_FAN_RPM 0
+#define FCU_FAN_PWM 1
+
+#define FCU_FAN_ABSENT_ID -1
+
+#define FCU_FAN_COUNT ARRAY_SIZE(fcu_fans)
+
+struct fcu_fan_table fcu_fans[] = {
+ [BACKSIDE_FAN_PWM_INDEX] = {
+ .loc = "BACKSIDE,SYS CTRLR FAN",
+ .type = FCU_FAN_PWM,
+ .id = BACKSIDE_FAN_PWM_DEFAULT_ID,
+ },
+ [DRIVES_FAN_RPM_INDEX] = {
+ .loc = "DRIVE BAY",
+ .type = FCU_FAN_RPM,
+ .id = DRIVES_FAN_RPM_DEFAULT_ID,
+ },
+ [SLOTS_FAN_PWM_INDEX] = {
+ .loc = "SLOT,PCI FAN",
+ .type = FCU_FAN_PWM,
+ .id = SLOTS_FAN_PWM_DEFAULT_ID,
+ },
+ [CPUA_INTAKE_FAN_RPM_INDEX] = {
+ .loc = "CPU A INTAKE",
+ .type = FCU_FAN_RPM,
+ .id = CPUA_INTAKE_FAN_RPM_DEFAULT_ID,
+ },
+ [CPUA_EXHAUST_FAN_RPM_INDEX] = {
+ .loc = "CPU A EXHAUST",
+ .type = FCU_FAN_RPM,
+ .id = CPUA_EXHAUST_FAN_RPM_DEFAULT_ID,
+ },
+ [CPUB_INTAKE_FAN_RPM_INDEX] = {
+ .loc = "CPU B INTAKE",
+ .type = FCU_FAN_RPM,
+ .id = CPUB_INTAKE_FAN_RPM_DEFAULT_ID,
+ },
+ [CPUB_EXHAUST_FAN_RPM_INDEX] = {
+ .loc = "CPU B EXHAUST",
+ .type = FCU_FAN_RPM,
+ .id = CPUB_EXHAUST_FAN_RPM_DEFAULT_ID,
+ },
+ /* pumps aren't present by default, have to be looked up in the
+ * device-tree
+ */
+ [CPUA_PUMP_RPM_INDEX] = {
+ .loc = "CPU A PUMP",
+ .type = FCU_FAN_RPM,
+ .id = FCU_FAN_ABSENT_ID,
+ },
+ [CPUB_PUMP_RPM_INDEX] = {
+ .loc = "CPU B PUMP",
+ .type = FCU_FAN_RPM,
+ .id = FCU_FAN_ABSENT_ID,
+ },
+ /* Xserve fans */
+ [CPU_A1_FAN_RPM_INDEX] = {
+ .loc = "CPU A 1",
+ .type = FCU_FAN_RPM,
+ .id = FCU_FAN_ABSENT_ID,
+ },
+ [CPU_A2_FAN_RPM_INDEX] = {
+ .loc = "CPU A 2",
+ .type = FCU_FAN_RPM,
+ .id = FCU_FAN_ABSENT_ID,
+ },
+ [CPU_A3_FAN_RPM_INDEX] = {
+ .loc = "CPU A 3",
+ .type = FCU_FAN_RPM,
+ .id = FCU_FAN_ABSENT_ID,
+ },
+ [CPU_B1_FAN_RPM_INDEX] = {
+ .loc = "CPU B 1",
+ .type = FCU_FAN_RPM,
+ .id = FCU_FAN_ABSENT_ID,
+ },
+ [CPU_B2_FAN_RPM_INDEX] = {
+ .loc = "CPU B 2",
+ .type = FCU_FAN_RPM,
+ .id = FCU_FAN_ABSENT_ID,
+ },
+ [CPU_B3_FAN_RPM_INDEX] = {
+ .loc = "CPU B 3",
+ .type = FCU_FAN_RPM,
+ .id = FCU_FAN_ABSENT_ID,
+ },
+};
+
+/*
+ * i2c_driver structure to attach to the host i2c controller
+ */
+
+static int therm_pm72_attach(struct i2c_adapter *adapter);
+static int therm_pm72_detach(struct i2c_adapter *adapter);
+
+static struct i2c_driver therm_pm72_driver =
+{
+ .owner = THIS_MODULE,
+ .name = "therm_pm72",
+ .flags = I2C_DF_NOTIFY,
+ .attach_adapter = therm_pm72_attach,
+ .detach_adapter = therm_pm72_detach,
+};
+
+/*
+ * Utility function to create an i2c_client structure and
+ * attach it to one of u3 adapters
+ */
+static struct i2c_client *attach_i2c_chip(int id, const char *name)
+{
+ struct i2c_client *clt;
+ struct i2c_adapter *adap;
+
+ if (id & 0x200)
+ adap = k2;
+ else if (id & 0x100)
+ adap = u3_1;
+ else
+ adap = u3_0;
+ if (adap == NULL)
+ return NULL;
+
+ clt = kmalloc(sizeof(struct i2c_client), GFP_KERNEL);
+ if (clt == NULL)
+ return NULL;
+ memset(clt, 0, sizeof(struct i2c_client));
+
+ clt->addr = (id >> 1) & 0x7f;
+ clt->adapter = adap;
+ clt->driver = &therm_pm72_driver;
+ strncpy(clt->name, name, I2C_NAME_SIZE-1);
+
+ if (i2c_attach_client(clt)) {
+ printk(KERN_ERR "therm_pm72: Failed to attach to i2c ID 0x%x\n", id);
+ kfree(clt);
+ return NULL;
+ }
+ return clt;
+}
+
+/*
+ * Utility function to get rid of the i2c_client structure
+ * (will also detach from the adapter hopepfully)
+ */
+static void detach_i2c_chip(struct i2c_client *clt)
+{
+ i2c_detach_client(clt);
+ kfree(clt);
+}
+
+/*
+ * Here are the i2c chip access wrappers
+ */
+
+static void initialize_adc(struct cpu_pid_state *state)
+{
+ int rc;
+ u8 buf[2];
+
+ /* Read ADC the configuration register and cache it. We
+ * also make sure Config2 contains proper values, I've seen
+ * cases where we got stale grabage in there, thus preventing
+ * proper reading of conv. values
+ */
+
+ /* Clear Config2 */
+ buf[0] = 5;
+ buf[1] = 0;
+ i2c_master_send(state->monitor, buf, 2);
+
+ /* Read & cache Config1 */
+ buf[0] = 1;
+ rc = i2c_master_send(state->monitor, buf, 1);
+ if (rc > 0) {
+ rc = i2c_master_recv(state->monitor, buf, 1);
+ if (rc > 0) {
+ state->adc_config = buf[0];
+ DBG("ADC config reg: %02x\n", state->adc_config);
+ /* Disable shutdown mode */
+ state->adc_config &= 0xfe;
+ buf[0] = 1;
+ buf[1] = state->adc_config;
+ rc = i2c_master_send(state->monitor, buf, 2);
+ }
+ }
+ if (rc <= 0)
+ printk(KERN_ERR "therm_pm72: Error reading ADC config"
+ " register !\n");
+}
+
+static int read_smon_adc(struct cpu_pid_state *state, int chan)
+{
+ int rc, data, tries = 0;
+ u8 buf[2];
+
+ for (;;) {
+ /* Set channel */
+ buf[0] = 1;
+ buf[1] = (state->adc_config & 0x1f) | (chan << 5);
+ rc = i2c_master_send(state->monitor, buf, 2);
+ if (rc <= 0)
+ goto error;
+ /* Wait for convertion */
+ msleep(1);
+ /* Switch to data register */
+ buf[0] = 4;
+ rc = i2c_master_send(state->monitor, buf, 1);
+ if (rc <= 0)
+ goto error;
+ /* Read result */
+ rc = i2c_master_recv(state->monitor, buf, 2);
+ if (rc < 0)
+ goto error;
+ data = ((u16)buf[0]) << 8 | (u16)buf[1];
+ return data >> 6;
+ error:
+ DBG("Error reading ADC, retrying...\n");
+ if (++tries > 10) {
+ printk(KERN_ERR "therm_pm72: Error reading ADC !\n");
+ return -1;
+ }
+ msleep(10);
+ }
+}
+
+static int read_lm87_reg(struct i2c_client * chip, int reg)
+{
+ int rc, tries = 0;
+ u8 buf;
+
+ for (;;) {
+ /* Set address */
+ buf = (u8)reg;
+ rc = i2c_master_send(chip, &buf, 1);
+ if (rc <= 0)
+ goto error;
+ rc = i2c_master_recv(chip, &buf, 1);
+ if (rc <= 0)
+ goto error;
+ return (int)buf;
+ error:
+ DBG("Error reading LM87, retrying...\n");
+ if (++tries > 10) {
+ printk(KERN_ERR "therm_pm72: Error reading LM87 !\n");
+ return -1;
+ }
+ msleep(10);
+ }
+}
+
+static int fan_read_reg(int reg, unsigned char *buf, int nb)
+{
+ int tries, nr, nw;
+
+ buf[0] = reg;
+ tries = 0;
+ for (;;) {
+ nw = i2c_master_send(fcu, buf, 1);
+ if (nw > 0 || (nw < 0 && nw != -EIO) || tries >= 100)
+ break;
+ msleep(10);
+ ++tries;
+ }
+ if (nw <= 0) {
+ printk(KERN_ERR "Failure writing address to FCU: %d", nw);
+ return -EIO;
+ }
+ tries = 0;
+ for (;;) {
+ nr = i2c_master_recv(fcu, buf, nb);
+ if (nr > 0 || (nr < 0 && nr != ENODEV) || tries >= 100)
+ break;
+ msleep(10);
+ ++tries;
+ }
+ if (nr <= 0)
+ printk(KERN_ERR "Failure reading data from FCU: %d", nw);
+ return nr;
+}
+
+static int fan_write_reg(int reg, const unsigned char *ptr, int nb)
+{
+ int tries, nw;
+ unsigned char buf[16];
+
+ buf[0] = reg;
+ memcpy(buf+1, ptr, nb);
+ ++nb;
+ tries = 0;
+ for (;;) {
+ nw = i2c_master_send(fcu, buf, nb);
+ if (nw > 0 || (nw < 0 && nw != EIO) || tries >= 100)
+ break;
+ msleep(10);
+ ++tries;
+ }
+ if (nw < 0)
+ printk(KERN_ERR "Failure writing to FCU: %d", nw);
+ return nw;
+}
+
+static int start_fcu(void)
+{
+ unsigned char buf = 0xff;
+ int rc;
+
+ rc = fan_write_reg(0xe, &buf, 1);
+ if (rc < 0)
+ return -EIO;
+ rc = fan_write_reg(0x2e, &buf, 1);
+ if (rc < 0)
+ return -EIO;
+ return 0;
+}
+
+static int set_rpm_fan(int fan_index, int rpm)
+{
+ unsigned char buf[2];
+ int rc, id;
+
+ if (fcu_fans[fan_index].type != FCU_FAN_RPM)
+ return -EINVAL;
+ id = fcu_fans[fan_index].id;
+ if (id == FCU_FAN_ABSENT_ID)
+ return -EINVAL;
+
+ if (rpm < 300)
+ rpm = 300;
+ else if (rpm > 8191)
+ rpm = 8191;
+ buf[0] = rpm >> 5;
+ buf[1] = rpm << 3;
+ rc = fan_write_reg(0x10 + (id * 2), buf, 2);
+ if (rc < 0)
+ return -EIO;
+ return 0;
+}
+
+static int get_rpm_fan(int fan_index, int programmed)
+{
+ unsigned char failure;
+ unsigned char active;
+ unsigned char buf[2];
+ int rc, id, reg_base;
+
+ if (fcu_fans[fan_index].type != FCU_FAN_RPM)
+ return -EINVAL;
+ id = fcu_fans[fan_index].id;
+ if (id == FCU_FAN_ABSENT_ID)
+ return -EINVAL;
+
+ rc = fan_read_reg(0xb, &failure, 1);
+ if (rc != 1)
+ return -EIO;
+ if ((failure & (1 << id)) != 0)
+ return -EFAULT;
+ rc = fan_read_reg(0xd, &active, 1);
+ if (rc != 1)
+ return -EIO;
+ if ((active & (1 << id)) == 0)
+ return -ENXIO;
+
+ /* Programmed value or real current speed */
+ reg_base = programmed ? 0x10 : 0x11;
+ rc = fan_read_reg(reg_base + (id * 2), buf, 2);
+ if (rc != 2)
+ return -EIO;
+
+ return (buf[0] << 5) | buf[1] >> 3;
+}
+
+static int set_pwm_fan(int fan_index, int pwm)
+{
+ unsigned char buf[2];
+ int rc, id;
+
+ if (fcu_fans[fan_index].type != FCU_FAN_PWM)
+ return -EINVAL;
+ id = fcu_fans[fan_index].id;
+ if (id == FCU_FAN_ABSENT_ID)
+ return -EINVAL;
+
+ if (pwm < 10)
+ pwm = 10;
+ else if (pwm > 100)
+ pwm = 100;
+ pwm = (pwm * 2559) / 1000;
+ buf[0] = pwm;
+ rc = fan_write_reg(0x30 + (id * 2), buf, 1);
+ if (rc < 0)
+ return rc;
+ return 0;
+}
+
+static int get_pwm_fan(int fan_index)
+{
+ unsigned char failure;
+ unsigned char active;
+ unsigned char buf[2];
+ int rc, id;
+
+ if (fcu_fans[fan_index].type != FCU_FAN_PWM)
+ return -EINVAL;
+ id = fcu_fans[fan_index].id;
+ if (id == FCU_FAN_ABSENT_ID)
+ return -EINVAL;
+
+ rc = fan_read_reg(0x2b, &failure, 1);
+ if (rc != 1)
+ return -EIO;
+ if ((failure & (1 << id)) != 0)
+ return -EFAULT;
+ rc = fan_read_reg(0x2d, &active, 1);
+ if (rc != 1)
+ return -EIO;
+ if ((active & (1 << id)) == 0)
+ return -ENXIO;
+
+ /* Programmed value or real current speed */
+ rc = fan_read_reg(0x30 + (id * 2), buf, 1);
+ if (rc != 1)
+ return -EIO;
+
+ return (buf[0] * 1000) / 2559;
+}
+
+/*
+ * Utility routine to read the CPU calibration EEPROM data
+ * from the device-tree
+ */
+static int read_eeprom(int cpu, struct mpu_data *out)
+{
+ struct device_node *np;
+ char nodename[64];
+ u8 *data;
+ int len;
+
+ /* prom.c routine for finding a node by path is a bit brain dead
+ * and requires exact @xxx unit numbers. This is a bit ugly but
+ * will work for these machines
+ */
+ sprintf(nodename, "/u3@0,f8000000/i2c@f8001000/cpuid@a%d", cpu ? 2 : 0);
+ np = of_find_node_by_path(nodename);
+ if (np == NULL) {
+ printk(KERN_ERR "therm_pm72: Failed to retreive cpuid node from device-tree\n");
+ return -ENODEV;
+ }
+ data = (u8 *)get_property(np, "cpuid", &len);
+ if (data == NULL) {
+ printk(KERN_ERR "therm_pm72: Failed to retreive cpuid property from device-tree\n");
+ of_node_put(np);
+ return -ENODEV;
+ }
+ memcpy(out, data, sizeof(struct mpu_data));
+ of_node_put(np);
+
+ return 0;
+}
+
+static void fetch_cpu_pumps_minmax(void)
+{
+ struct cpu_pid_state *state0 = &cpu_state[0];
+ struct cpu_pid_state *state1 = &cpu_state[1];
+ u16 pump_min = 0, pump_max = 0xffff;
+ u16 tmp[4];
+
+ /* Try to fetch pumps min/max infos from eeprom */
+
+ memcpy(&tmp, &state0->mpu.processor_part_num, 8);
+ if (tmp[0] != 0xffff && tmp[1] != 0xffff) {
+ pump_min = max(pump_min, tmp[0]);
+ pump_max = min(pump_max, tmp[1]);
+ }
+ if (tmp[2] != 0xffff && tmp[3] != 0xffff) {
+ pump_min = max(pump_min, tmp[2]);
+ pump_max = min(pump_max, tmp[3]);
+ }
+
+ /* Double check the values, this _IS_ needed as the EEPROM on
+ * some dual 2.5Ghz G5s seem, at least, to have both min & max
+ * same to the same value ... (grrrr)
+ */
+ if (pump_min == pump_max || pump_min == 0 || pump_max == 0xffff) {
+ pump_min = CPU_PUMP_OUTPUT_MIN;
+ pump_max = CPU_PUMP_OUTPUT_MAX;
+ }
+
+ state0->pump_min = state1->pump_min = pump_min;
+ state0->pump_max = state1->pump_max = pump_max;
+}
+
+/*
+ * Now, unfortunately, sysfs doesn't give us a nice void * we could
+ * pass around to the attribute functions, so we don't really have
+ * choice but implement a bunch of them...
+ *
+ * That sucks a bit, we take the lock because FIX32TOPRINT evaluates
+ * the input twice... I accept patches :)
+ */
+#define BUILD_SHOW_FUNC_FIX(name, data) \
+static ssize_t show_##name(struct device *dev, char *buf) \
+{ \
+ ssize_t r; \
+ down(&driver_lock); \
+ r = sprintf(buf, "%d.%03d", FIX32TOPRINT(data)); \
+ up(&driver_lock); \
+ return r; \
+}
+#define BUILD_SHOW_FUNC_INT(name, data) \
+static ssize_t show_##name(struct device *dev, char *buf) \
+{ \
+ return sprintf(buf, "%d", data); \
+}
+
+BUILD_SHOW_FUNC_FIX(cpu0_temperature, cpu_state[0].last_temp)
+BUILD_SHOW_FUNC_FIX(cpu0_voltage, cpu_state[0].voltage)
+BUILD_SHOW_FUNC_FIX(cpu0_current, cpu_state[0].current_a)
+BUILD_SHOW_FUNC_INT(cpu0_exhaust_fan_rpm, cpu_state[0].rpm)
+BUILD_SHOW_FUNC_INT(cpu0_intake_fan_rpm, cpu_state[0].intake_rpm)
+
+BUILD_SHOW_FUNC_FIX(cpu1_temperature, cpu_state[1].last_temp)
+BUILD_SHOW_FUNC_FIX(cpu1_voltage, cpu_state[1].voltage)
+BUILD_SHOW_FUNC_FIX(cpu1_current, cpu_state[1].current_a)
+BUILD_SHOW_FUNC_INT(cpu1_exhaust_fan_rpm, cpu_state[1].rpm)
+BUILD_SHOW_FUNC_INT(cpu1_intake_fan_rpm, cpu_state[1].intake_rpm)
+
+BUILD_SHOW_FUNC_FIX(backside_temperature, backside_state.last_temp)
+BUILD_SHOW_FUNC_INT(backside_fan_pwm, backside_state.pwm)
+
+BUILD_SHOW_FUNC_FIX(drives_temperature, drives_state.last_temp)
+BUILD_SHOW_FUNC_INT(drives_fan_rpm, drives_state.rpm)
+
+BUILD_SHOW_FUNC_FIX(dimms_temperature, dimms_state.last_temp)
+
+static DEVICE_ATTR(cpu0_temperature,S_IRUGO,show_cpu0_temperature,NULL);
+static DEVICE_ATTR(cpu0_voltage,S_IRUGO,show_cpu0_voltage,NULL);
+static DEVICE_ATTR(cpu0_current,S_IRUGO,show_cpu0_current,NULL);
+static DEVICE_ATTR(cpu0_exhaust_fan_rpm,S_IRUGO,show_cpu0_exhaust_fan_rpm,NULL);
+static DEVICE_ATTR(cpu0_intake_fan_rpm,S_IRUGO,show_cpu0_intake_fan_rpm,NULL);
+
+static DEVICE_ATTR(cpu1_temperature,S_IRUGO,show_cpu1_temperature,NULL);
+static DEVICE_ATTR(cpu1_voltage,S_IRUGO,show_cpu1_voltage,NULL);
+static DEVICE_ATTR(cpu1_current,S_IRUGO,show_cpu1_current,NULL);
+static DEVICE_ATTR(cpu1_exhaust_fan_rpm,S_IRUGO,show_cpu1_exhaust_fan_rpm,NULL);
+static DEVICE_ATTR(cpu1_intake_fan_rpm,S_IRUGO,show_cpu1_intake_fan_rpm,NULL);
+
+static DEVICE_ATTR(backside_temperature,S_IRUGO,show_backside_temperature,NULL);
+static DEVICE_ATTR(backside_fan_pwm,S_IRUGO,show_backside_fan_pwm,NULL);
+
+static DEVICE_ATTR(drives_temperature,S_IRUGO,show_drives_temperature,NULL);
+static DEVICE_ATTR(drives_fan_rpm,S_IRUGO,show_drives_fan_rpm,NULL);
+
+static DEVICE_ATTR(dimms_temperature,S_IRUGO,show_dimms_temperature,NULL);
+
+/*
+ * CPUs fans control loop
+ */
+
+static int do_read_one_cpu_values(struct cpu_pid_state *state, s32 *temp, s32 *power)
+{
+ s32 ltemp, volts, amps;
+ int index, rc = 0;
+
+ /* Default (in case of error) */
+ *temp = state->cur_temp;
+ *power = state->cur_power;
+
+ if (cpu_pid_type == CPU_PID_TYPE_RACKMAC)
+ index = (state->index == 0) ?
+ CPU_A1_FAN_RPM_INDEX : CPU_B1_FAN_RPM_INDEX;
+ else
+ index = (state->index == 0) ?
+ CPUA_EXHAUST_FAN_RPM_INDEX : CPUB_EXHAUST_FAN_RPM_INDEX;
+
+ /* Read current fan status */
+ rc = get_rpm_fan(index, !RPM_PID_USE_ACTUAL_SPEED);
+ if (rc < 0) {
+ /* XXX What do we do now ? Nothing for now, keep old value, but
+ * return error upstream
+ */
+ DBG(" cpu %d, fan reading error !\n", state->index);
+ } else {
+ state->rpm = rc;
+ DBG(" cpu %d, exhaust RPM: %d\n", state->index, state->rpm);
+ }
+
+ /* Get some sensor readings and scale it */
+ ltemp = read_smon_adc(state, 1);
+ if (ltemp == -1) {
+ /* XXX What do we do now ? */
+ state->overtemp++;
+ if (rc == 0)
+ rc = -EIO;
+ DBG(" cpu %d, temp reading error !\n", state->index);
+ } else {
+ /* Fixup temperature according to diode calibration
+ */
+ DBG(" cpu %d, temp raw: %04x, m_diode: %04x, b_diode: %04x\n",
+ state->index,
+ ltemp, state->mpu.mdiode, state->mpu.bdiode);
+ *temp = ((s32)ltemp * (s32)state->mpu.mdiode + ((s32)state->mpu.bdiode << 12)) >> 2;
+ state->last_temp = *temp;
+ DBG(" temp: %d.%03d\n", FIX32TOPRINT((*temp)));
+ }
+
+ /*
+ * Read voltage & current and calculate power
+ */
+ volts = read_smon_adc(state, 3);
+ amps = read_smon_adc(state, 4);
+
+ /* Scale voltage and current raw sensor values according to fixed scales
+ * obtained in Darwin and calculate power from I and V
+ */
+ volts *= ADC_CPU_VOLTAGE_SCALE;
+ amps *= ADC_CPU_CURRENT_SCALE;
+ *power = (((u64)volts) * ((u64)amps)) >> 16;
+ state->voltage = volts;
+ state->current_a = amps;
+ state->last_power = *power;
+
+ DBG(" cpu %d, current: %d.%03d, voltage: %d.%03d, power: %d.%03d W\n",
+ state->index, FIX32TOPRINT(state->current_a),
+ FIX32TOPRINT(state->voltage), FIX32TOPRINT(*power));
+
+ return 0;
+}
+
+static void do_cpu_pid(struct cpu_pid_state *state, s32 temp, s32 power)
+{
+ s32 power_target, integral, derivative, proportional, adj_in_target, sval;
+ s64 integ_p, deriv_p, prop_p, sum;
+ int i;
+
+ /* Calculate power target value (could be done once for all)
+ * and convert to a 16.16 fp number
+ */
+ power_target = ((u32)(state->mpu.pmaxh - state->mpu.padjmax)) << 16;
+ DBG(" power target: %d.%03d, error: %d.%03d\n",
+ FIX32TOPRINT(power_target), FIX32TOPRINT(power_target - power));
+
+ /* Store temperature and power in history array */
+ state->cur_temp = (state->cur_temp + 1) % CPU_TEMP_HISTORY_SIZE;
+ state->temp_history[state->cur_temp] = temp;
+ state->cur_power = (state->cur_power + 1) % state->count_power;
+ state->power_history[state->cur_power] = power;
+ state->error_history[state->cur_power] = power_target - power;
+
+ /* If first loop, fill the history table */
+ if (state->first) {
+ for (i = 0; i < (state->count_power - 1); i++) {
+ state->cur_power = (state->cur_power + 1) % state->count_power;
+ state->power_history[state->cur_power] = power;
+ state->error_history[state->cur_power] = power_target - power;
+ }
+ for (i = 0; i < (CPU_TEMP_HISTORY_SIZE - 1); i++) {
+ state->cur_temp = (state->cur_temp + 1) % CPU_TEMP_HISTORY_SIZE;
+ state->temp_history[state->cur_temp] = temp;
+ }
+ state->first = 0;
+ }
+
+ /* Calculate the integral term normally based on the "power" values */
+ sum = 0;
+ integral = 0;
+ for (i = 0; i < state->count_power; i++)
+ integral += state->error_history[i];
+ integral *= CPU_PID_INTERVAL;
+ DBG(" integral: %08x\n", integral);
+
+ /* Calculate the adjusted input (sense value).
+ * G_r is 12.20
+ * integ is 16.16
+ * so the result is 28.36
+ *
+ * input target is mpu.ttarget, input max is mpu.tmax
+ */
+ integ_p = ((s64)state->mpu.pid_gr) * (s64)integral;
+ DBG(" integ_p: %d\n", (int)(integ_p >> 36));
+ sval = (state->mpu.tmax << 16) - ((integ_p >> 20) & 0xffffffff);
+ adj_in_target = (state->mpu.ttarget << 16);
+ if (adj_in_target > sval)
+ adj_in_target = sval;
+ DBG(" adj_in_target: %d.%03d, ttarget: %d\n", FIX32TOPRINT(adj_in_target),
+ state->mpu.ttarget);
+
+ /* Calculate the derivative term */
+ derivative = state->temp_history[state->cur_temp] -
+ state->temp_history[(state->cur_temp + CPU_TEMP_HISTORY_SIZE - 1)
+ % CPU_TEMP_HISTORY_SIZE];
+ derivative /= CPU_PID_INTERVAL;
+ deriv_p = ((s64)state->mpu.pid_gd) * (s64)derivative;
+ DBG(" deriv_p: %d\n", (int)(deriv_p >> 36));
+ sum += deriv_p;
+
+ /* Calculate the proportional term */
+ proportional = temp - adj_in_target;
+ prop_p = ((s64)state->mpu.pid_gp) * (s64)proportional;
+ DBG(" prop_p: %d\n", (int)(prop_p >> 36));
+ sum += prop_p;
+
+ /* Scale sum */
+ sum >>= 36;
+
+ DBG(" sum: %d\n", (int)sum);
+ state->rpm += (s32)sum;
+}
+
+static void do_monitor_cpu_combined(void)
+{
+ struct cpu_pid_state *state0 = &cpu_state[0];
+ struct cpu_pid_state *state1 = &cpu_state[1];
+ s32 temp0, power0, temp1, power1;
+ s32 temp_combi, power_combi;
+ int rc, intake, pump;
+
+ rc = do_read_one_cpu_values(state0, &temp0, &power0);
+ if (rc < 0) {
+ /* XXX What do we do now ? */
+ }
+ state1->overtemp = 0;
+ rc = do_read_one_cpu_values(state1, &temp1, &power1);
+ if (rc < 0) {
+ /* XXX What do we do now ? */
+ }
+ if (state1->overtemp)
+ state0->overtemp++;
+
+ temp_combi = max(temp0, temp1);
+ power_combi = max(power0, power1);
+
+ /* Check tmax, increment overtemp if we are there. At tmax+8, we go
+ * full blown immediately and try to trigger a shutdown
+ */
+ if (temp_combi >= ((state0->mpu.tmax + 8) << 16)) {
+ printk(KERN_WARNING "Warning ! Temperature way above maximum (%d) !\n",
+ temp_combi >> 16);
+ state0->overtemp = CPU_MAX_OVERTEMP;
+ } else if (temp_combi > (state0->mpu.tmax << 16))
+ state0->overtemp++;
+ else
+ state0->overtemp = 0;
+ if (state0->overtemp >= CPU_MAX_OVERTEMP)
+ critical_state = 1;
+ if (state0->overtemp > 0) {
+ state0->rpm = state0->mpu.rmaxn_exhaust_fan;
+ state0->intake_rpm = intake = state0->mpu.rmaxn_intake_fan;
+ pump = state0->pump_min;
+ goto do_set_fans;
+ }
+
+ /* Do the PID */
+ do_cpu_pid(state0, temp_combi, power_combi);
+
+ /* Range check */
+ state0->rpm = max(state0->rpm, (int)state0->mpu.rminn_exhaust_fan);
+ state0->rpm = min(state0->rpm, (int)state0->mpu.rmaxn_exhaust_fan);
+
+ /* Calculate intake fan speed */
+ intake = (state0->rpm * CPU_INTAKE_SCALE) >> 16;
+ intake = max(intake, (int)state0->mpu.rminn_intake_fan);
+ intake = min(intake, (int)state0->mpu.rmaxn_intake_fan);
+ state0->intake_rpm = intake;
+
+ /* Calculate pump speed */
+ pump = (state0->rpm * state0->pump_max) /
+ state0->mpu.rmaxn_exhaust_fan;
+ pump = min(pump, state0->pump_max);
+ pump = max(pump, state0->pump_min);
+
+ do_set_fans:
+ /* We copy values from state 0 to state 1 for /sysfs */
+ state1->rpm = state0->rpm;
+ state1->intake_rpm = state0->intake_rpm;
+
+ DBG("** CPU %d RPM: %d Ex, %d, Pump: %d, In, overtemp: %d\n",
+ state1->index, (int)state1->rpm, intake, pump, state1->overtemp);
+
+ /* We should check for errors, shouldn't we ? But then, what
+ * do we do once the error occurs ? For FCU notified fan
+ * failures (-EFAULT) we probably want to notify userland
+ * some way...
+ */
+ set_rpm_fan(CPUA_INTAKE_FAN_RPM_INDEX, intake);
+ set_rpm_fan(CPUA_EXHAUST_FAN_RPM_INDEX, state0->rpm);
+ set_rpm_fan(CPUB_INTAKE_FAN_RPM_INDEX, intake);
+ set_rpm_fan(CPUB_EXHAUST_FAN_RPM_INDEX, state0->rpm);
+
+ if (fcu_fans[CPUA_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID)
+ set_rpm_fan(CPUA_PUMP_RPM_INDEX, pump);
+ if (fcu_fans[CPUB_PUMP_RPM_INDEX].id != FCU_FAN_ABSENT_ID)
+ set_rpm_fan(CPUB_PUMP_RPM_INDEX, pump);
+}
+
+static void do_monitor_cpu_split(struct cpu_pid_state *state)
+{
+ s32 temp, power;
+ int rc, intake;
+
+ /* Read current fan status */
+ rc = do_read_one_cpu_values(state, &temp, &power);
+ if (rc < 0) {
+ /* XXX What do we do now ? */
+ }
+
+ /* Check tmax, increment overtemp if we are there. At tmax+8, we go
+ * full blown immediately and try to trigger a shutdown
+ */
+ if (temp >= ((state->mpu.tmax + 8) << 16)) {
+ printk(KERN_WARNING "Warning ! CPU %d temperature way above maximum"
+ " (%d) !\n",
+ state->index, temp >> 16);
+ state->overtemp = CPU_MAX_OVERTEMP;
+ } else if (temp > (state->mpu.tmax << 16))
+ state->overtemp++;
+ else
+ state->overtemp = 0;
+ if (state->overtemp >= CPU_MAX_OVERTEMP)
+ critical_state = 1;
+ if (state->overtemp > 0) {
+ state->rpm = state->mpu.rmaxn_exhaust_fan;
+ state->intake_rpm = intake = state->mpu.rmaxn_intake_fan;
+ goto do_set_fans;
+ }
+
+ /* Do the PID */
+ do_cpu_pid(state, temp, power);
+
+ /* Range check */
+ state->rpm = max(state->rpm, (int)state->mpu.rminn_exhaust_fan);
+ state->rpm = min(state->rpm, (int)state->mpu.rmaxn_exhaust_fan);
+
+ /* Calculate intake fan */
+ intake = (state->rpm * CPU_INTAKE_SCALE) >> 16;
+ intake = max(intake, (int)state->mpu.rminn_intake_fan);
+ intake = min(intake, (int)state->mpu.rmaxn_intake_fan);
+ state->intake_rpm = intake;
+
+ do_set_fans:
+ DBG("** CPU %d RPM: %d Ex, %d In, overtemp: %d\n",
+ state->index, (int)state->rpm, intake, state->overtemp);
+
+ /* We should check for errors, shouldn't we ? But then, what
+ * do we do once the error occurs ? For FCU notified fan
+ * failures (-EFAULT) we probably want to notify userland
+ * some way...
+ */
+ if (state->index == 0) {
+ set_rpm_fan(CPUA_INTAKE_FAN_RPM_INDEX, intake);
+ set_rpm_fan(CPUA_EXHAUST_FAN_RPM_INDEX, state->rpm);
+ } else {
+ set_rpm_fan(CPUB_INTAKE_FAN_RPM_INDEX, intake);
+ set_rpm_fan(CPUB_EXHAUST_FAN_RPM_INDEX, state->rpm);
+ }
+}
+
+static void do_monitor_cpu_rack(struct cpu_pid_state *state)
+{
+ s32 temp, power, fan_min;
+ int rc;
+
+ /* Read current fan status */
+ rc = do_read_one_cpu_values(state, &temp, &power);
+ if (rc < 0) {
+ /* XXX What do we do now ? */
+ }
+
+ /* Check tmax, increment overtemp if we are there. At tmax+8, we go
+ * full blown immediately and try to trigger a shutdown
+ */
+ if (temp >= ((state->mpu.tmax + 8) << 16)) {
+ printk(KERN_WARNING "Warning ! CPU %d temperature way above maximum"
+ " (%d) !\n",
+ state->index, temp >> 16);
+ state->overtemp = CPU_MAX_OVERTEMP;
+ } else if (temp > (state->mpu.tmax << 16))
+ state->overtemp++;
+ else
+ state->overtemp = 0;
+ if (state->overtemp >= CPU_MAX_OVERTEMP)
+ critical_state = 1;
+ if (state->overtemp > 0) {
+ state->rpm = state->intake_rpm = state->mpu.rmaxn_intake_fan;
+ goto do_set_fans;
+ }
+
+ /* Do the PID */
+ do_cpu_pid(state, temp, power);
+
+ /* Check clamp from dimms */
+ fan_min = dimm_output_clamp;
+ fan_min = max(fan_min, (int)state->mpu.rminn_intake_fan);
+
+ state->rpm = max(state->rpm, (int)fan_min);
+ state->rpm = min(state->rpm, (int)state->mpu.rmaxn_intake_fan);
+ state->intake_rpm = state->rpm;
+
+ do_set_fans:
+ DBG("** CPU %d RPM: %d overtemp: %d\n",
+ state->index, (int)state->rpm, state->overtemp);
+
+ /* We should check for errors, shouldn't we ? But then, what
+ * do we do once the error occurs ? For FCU notified fan
+ * failures (-EFAULT) we probably want to notify userland
+ * some way...
+ */
+ if (state->index == 0) {
+ set_rpm_fan(CPU_A1_FAN_RPM_INDEX, state->rpm);
+ set_rpm_fan(CPU_A2_FAN_RPM_INDEX, state->rpm);
+ set_rpm_fan(CPU_A3_FAN_RPM_INDEX, state->rpm);
+ } else {
+ set_rpm_fan(CPU_B1_FAN_RPM_INDEX, state->rpm);
+ set_rpm_fan(CPU_B2_FAN_RPM_INDEX, state->rpm);
+ set_rpm_fan(CPU_B3_FAN_RPM_INDEX, state->rpm);
+ }
+}
+
+/*
+ * Initialize the state structure for one CPU control loop
+ */
+static int init_cpu_state(struct cpu_pid_state *state, int index)
+{
+ state->index = index;
+ state->first = 1;
+ state->rpm = (cpu_pid_type == CPU_PID_TYPE_RACKMAC) ? 4000 : 1000;
+ sta