drm/nouveau/therm: implement support for temperature alarms

For now, we only boost the fan speed to the maximum and auto-mode
when hitting the FAN_BOOST threshold and halt the computer when it
reaches the shutdown temperature. The downclock and critical thresholds
do nothing.

On nv43:50 and nva3+, temperature is polled because of the limited hardware.
I'll improve the nva3+ situation by implementing alarm management in PDAEMON
whenever I can but polling once every second shouldn't be such a problem.

v2 (Ben Skeggs):
- rebased

v3: fixed false-detections and threshold reprogrammation handling on nv50:nvc0

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drivers/gpu/drm/nouveau/core/os.h

@@ -16,6 +16,7 @@
 #include <linux/vmalloc.h>
 #include <linux/acpi.h>
 #include <linux/dmi.h>
+#include <linux/reboot.h>
 
 #include <asm/unaligned.h>
 

drivers/gpu/drm/nouveau/core/subdev/mc/nv98.c

@@ -35,6 +35,7 @@ nv98_mc_intr[] = {
 	{ 0x00001000, NVDEV_ENGINE_GR },
 	{ 0x00004000, NVDEV_ENGINE_CRYPT },	/* NV84:NVA3 */
 	{ 0x00008000, NVDEV_ENGINE_BSP },
+	{ 0x00080000, NVDEV_SUBDEV_THERM },	/* NVA3:NVC0 */
 	{ 0x00100000, NVDEV_SUBDEV_TIMER },
 	{ 0x00200000, NVDEV_SUBDEV_GPIO },
 	{ 0x00400000, NVDEV_ENGINE_COPY0 },	/* NVA3-     */

drivers/gpu/drm/nouveau/core/subdev/therm/base.c

@@ -123,7 +123,7 @@ nouveau_therm_alarm(struct nouveau_alarm *alarm)
 	nouveau_therm_update(&priv->base, -1);
 }
 
-static int
+int
 nouveau_therm_mode(struct nouveau_therm *therm, int mode)
 {
 	struct nouveau_therm_priv *priv = (void *)therm;
@@ -212,27 +212,35 @@ nouveau_therm_attr_set(struct nouveau_therm *therm,
 		return nouveau_therm_mode(therm, value);
 	case NOUVEAU_THERM_ATTR_THRS_FAN_BOOST:
 		priv->bios_sensor.thrs_fan_boost.temp = value;
+		priv->sensor.program_alarms(therm);
 		return 0;
 	case NOUVEAU_THERM_ATTR_THRS_FAN_BOOST_HYST:
 		priv->bios_sensor.thrs_fan_boost.hysteresis = value;
+		priv->sensor.program_alarms(therm);
 		return 0;
 	case NOUVEAU_THERM_ATTR_THRS_DOWN_CLK:
 		priv->bios_sensor.thrs_down_clock.temp = value;
+		priv->sensor.program_alarms(therm);
 		return 0;
 	case NOUVEAU_THERM_ATTR_THRS_DOWN_CLK_HYST:
 		priv->bios_sensor.thrs_down_clock.hysteresis = value;
+		priv->sensor.program_alarms(therm);
 		return 0;
 	case NOUVEAU_THERM_ATTR_THRS_CRITICAL:
 		priv->bios_sensor.thrs_critical.temp = value;
+		priv->sensor.program_alarms(therm);
 		return 0;
 	case NOUVEAU_THERM_ATTR_THRS_CRITICAL_HYST:
 		priv->bios_sensor.thrs_critical.hysteresis = value;
+		priv->sensor.program_alarms(therm);
 		return 0;
 	case NOUVEAU_THERM_ATTR_THRS_SHUTDOWN:
 		priv->bios_sensor.thrs_shutdown.temp = value;
+		priv->sensor.program_alarms(therm);
 		return 0;
 	case NOUVEAU_THERM_ATTR_THRS_SHUTDOWN_HYST:
 		priv->bios_sensor.thrs_shutdown.hysteresis = value;
+		priv->sensor.program_alarms(therm);
 		return 0;
 	}
 
@@ -253,6 +261,7 @@ _nouveau_therm_init(struct nouveau_object *object)
 	if (priv->fan->percent >= 0)
 		therm->fan_set(therm, priv->fan->percent);
 
+	priv->sensor.program_alarms(therm);
 	return 0;
 }
 

drivers/gpu/drm/nouveau/core/subdev/therm/nv40.c

@@ -38,6 +38,7 @@ nv40_sensor_setup(struct nouveau_therm *therm)
 	if (device->chipset >= 0x46) {
 		nv_mask(therm, 0x15b8, 0x80000000, 0);
 		nv_wr32(therm, 0x15b0, 0x80003fff);
+		mdelay(10); /* wait for the temperature to stabilize */
 		return nv_rd32(therm, 0x15b4) & 0x3fff;
 	} else {
 		nv_wr32(therm, 0x15b0, 0xff);
@@ -135,6 +136,20 @@ nv40_fan_pwm_set(struct nouveau_therm *therm, int line, u32 divs, u32 duty)
 	return 0;
 }
 
+static void
+nv40_therm_intr(struct nouveau_subdev *subdev)
+{
+	struct nouveau_therm *therm = nouveau_therm(subdev);
+	uint32_t stat = nv_rd32(therm, 0x1100);
+
+	/* traitement */
+
+	/* ack all IRQs */
+	nv_wr32(therm, 0x1100, 0x70000);
+
+	nv_error(therm, "THERM received an IRQ: stat = %x\n", stat);
+}
+
 static int
 nv40_therm_ctor(struct nouveau_object *parent,
 		struct nouveau_object *engine,
@@ -153,6 +168,8 @@ nv40_therm_ctor(struct nouveau_object *parent,
 	priv->base.base.pwm_get = nv40_fan_pwm_get;
 	priv->base.base.pwm_set = nv40_fan_pwm_set;
 	priv->base.base.temp_get = nv40_temp_get;
+	priv->base.sensor.program_alarms = nouveau_therm_program_alarms_polling;
+	nv_subdev(priv)->intr = nv40_therm_intr;
 	return nouveau_therm_preinit(&priv->base.base);
 }
 

drivers/gpu/drm/nouveau/core/subdev/therm/nv50.c

@@ -124,6 +124,141 @@ nv50_temp_get(struct nouveau_therm *therm)
 	return nv_rd32(therm, 0x20400);
 }
 
+static void
+nv50_therm_program_alarms(struct nouveau_therm *therm)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+	struct nvbios_therm_sensor *sensor = &priv->bios_sensor;
+	unsigned long flags;
+
+	spin_lock_irqsave(&priv->sensor.alarm_program_lock, flags);
+
+	/* enable RISING and FALLING IRQs for shutdown, THRS 0, 1, 2 and 4 */
+	nv_wr32(therm, 0x20000, 0x000003ff);
+
+	/* shutdown: The computer should be shutdown when reached */
+	nv_wr32(therm, 0x20484, sensor->thrs_shutdown.hysteresis);
+	nv_wr32(therm, 0x20480, sensor->thrs_shutdown.temp);
+
+	/* THRS_1 : fan boost*/
+	nv_wr32(therm, 0x204c4, sensor->thrs_fan_boost.temp);
+
+	/* THRS_2 : critical */
+	nv_wr32(therm, 0x204c0, sensor->thrs_critical.temp);
+
+	/* THRS_4 : down clock */
+	nv_wr32(therm, 0x20414, sensor->thrs_down_clock.temp);
+	spin_unlock_irqrestore(&priv->sensor.alarm_program_lock, flags);
+
+	nv_info(therm,
+		"Programmed thresholds [ %d(%d), %d(%d), %d(%d), %d(%d) ]\n",
+		sensor->thrs_fan_boost.temp, sensor->thrs_fan_boost.hysteresis,
+		sensor->thrs_down_clock.temp,
+		sensor->thrs_down_clock.hysteresis,
+		sensor->thrs_critical.temp, sensor->thrs_critical.hysteresis,
+		sensor->thrs_shutdown.temp, sensor->thrs_shutdown.hysteresis);
+
+}
+
+/* must be called with alarm_program_lock taken ! */
+static void
+nv50_therm_threshold_hyst_emulation(struct nouveau_therm *therm,
+				   uint32_t thrs_reg, u8 status_bit,
+				   const struct nvbios_therm_threshold *thrs,
+				   enum nouveau_therm_thrs thrs_name)
+{
+	enum nouveau_therm_thrs_direction direction;
+	enum nouveau_therm_thrs_state prev_state, new_state;
+	int temp, cur;
+
+	prev_state = nouveau_therm_sensor_get_threshold_state(therm, thrs_name);
+	temp = nv_rd32(therm, thrs_reg);
+
+	/* program the next threshold */
+	if (temp == thrs->temp) {
+		nv_wr32(therm, thrs_reg, thrs->temp - thrs->hysteresis);
+		new_state = NOUVEAU_THERM_THRS_HIGHER;
+	} else {
+		nv_wr32(therm, thrs_reg, thrs->temp);
+		new_state = NOUVEAU_THERM_THRS_LOWER;
+	}
+
+	/* fix the state (in case someone reprogrammed the alarms) */
+	cur = therm->temp_get(therm);
+	if (new_state == NOUVEAU_THERM_THRS_LOWER && cur > thrs->temp)
+		new_state = NOUVEAU_THERM_THRS_HIGHER;
+	else if (new_state == NOUVEAU_THERM_THRS_HIGHER &&
+		cur < thrs->temp - thrs->hysteresis)
+		new_state = NOUVEAU_THERM_THRS_LOWER;
+	nouveau_therm_sensor_set_threshold_state(therm, thrs_name, new_state);
+
+	/* find the direction */
+	if (prev_state < new_state)
+		direction = NOUVEAU_THERM_THRS_RISING;
+	else if (prev_state > new_state)
+		direction = NOUVEAU_THERM_THRS_FALLING;
+	else
+		return;
+
+	/* advertise a change in direction */
+	nouveau_therm_sensor_event(therm, thrs_name, direction);
+}
+
+static void
+nv50_therm_intr(struct nouveau_subdev *subdev)
+{
+	struct nouveau_therm *therm = nouveau_therm(subdev);
+	struct nouveau_therm_priv *priv = (void *)therm;
+	struct nvbios_therm_sensor *sensor = &priv->bios_sensor;
+	unsigned long flags;
+	uint32_t intr;
+
+	spin_lock_irqsave(&priv->sensor.alarm_program_lock, flags);
+
+	intr = nv_rd32(therm, 0x20100);
+
+	/* THRS_4: downclock */
+	if (intr & 0x002) {
+		nv50_therm_threshold_hyst_emulation(therm, 0x20414, 24,
+						  &sensor->thrs_down_clock,
+						  NOUVEAU_THERM_THRS_DOWNCLOCK);
+		intr &= ~0x002;
+	}
+
+	/* shutdown */
+	if (intr & 0x004) {
+		nv50_therm_threshold_hyst_emulation(therm, 0x20480, 20,
+						   &sensor->thrs_shutdown,
+						   NOUVEAU_THERM_THRS_SHUTDOWN);
+		intr &= ~0x004;
+	}
+
+	/* THRS_1 : fan boost */
+	if (intr & 0x008) {
+		nv50_therm_threshold_hyst_emulation(therm, 0x204c4, 21,
+						   &sensor->thrs_fan_boost,
+						   NOUVEAU_THERM_THRS_FANBOOST);
+		intr &= ~0x008;
+	}
+
+	/* THRS_2 : critical */
+	if (intr & 0x010) {
+		nv50_therm_threshold_hyst_emulation(therm, 0x204c0, 22,
+						   &sensor->thrs_critical,
+						   NOUVEAU_THERM_THRS_CRITICAL);
+		intr &= ~0x010;
+	}
+
+	if (intr)
+		nv_error(therm, "unhandled intr 0x%08x\n", intr);
+
+	/* ACK everything */
+	nv_wr32(therm, 0x20100, 0xffffffff);
+	nv_wr32(therm, 0x1100, 0x10000); /* PBUS */
+
+	spin_unlock_irqrestore(&priv->sensor.alarm_program_lock, flags);
+}
+
 static int
 nv50_therm_ctor(struct nouveau_object *parent,
 		struct nouveau_object *engine,
@@ -143,6 +278,24 @@ nv50_therm_ctor(struct nouveau_object *parent,
 	priv->base.base.pwm_set = nv50_fan_pwm_set;
 	priv->base.base.pwm_clock = nv50_fan_pwm_clock;
 	priv->base.base.temp_get = nv50_temp_get;
+	priv->base.sensor.program_alarms = nv50_therm_program_alarms;
+	spin_lock_init(&priv->base.sensor.alarm_program_lock);
+	nv_subdev(priv)->intr = nv50_therm_intr;
+
+	/* init the thresholds */
+	nouveau_therm_sensor_set_threshold_state(&priv->base.base,
+						 NOUVEAU_THERM_THRS_SHUTDOWN,
+						 NOUVEAU_THERM_THRS_LOWER);
+	nouveau_therm_sensor_set_threshold_state(&priv->base.base,
+						 NOUVEAU_THERM_THRS_FANBOOST,
+						 NOUVEAU_THERM_THRS_LOWER);
+	nouveau_therm_sensor_set_threshold_state(&priv->base.base,
+						 NOUVEAU_THERM_THRS_CRITICAL,
+						 NOUVEAU_THERM_THRS_LOWER);
+	nouveau_therm_sensor_set_threshold_state(&priv->base.base,
+						 NOUVEAU_THERM_THRS_DOWNCLOCK,
+						 NOUVEAU_THERM_THRS_LOWER);
+
 	return nouveau_therm_preinit(&priv->base.base);
 }
 

drivers/gpu/drm/nouveau/core/subdev/therm/nva3.c

@@ -83,6 +83,7 @@ nva3_therm_ctor(struct nouveau_object *parent,
 	priv->base.base.pwm_clock = nv50_fan_pwm_clock;
 	priv->base.base.temp_get = nv50_temp_get;
 	priv->base.base.fan_sense = nva3_therm_fan_sense;
+	priv->base.sensor.program_alarms = nouveau_therm_program_alarms_polling;
 	return nouveau_therm_preinit(&priv->base.base);
 }
 

drivers/gpu/drm/nouveau/core/subdev/therm/nvd0.c

@@ -137,6 +137,7 @@ nvd0_therm_ctor(struct nouveau_object *parent,
 	priv->base.base.pwm_clock = nvd0_fan_pwm_clock;
 	priv->base.base.temp_get = nv50_temp_get;
 	priv->base.base.fan_sense = nva3_therm_fan_sense;
+	priv->base.sensor.program_alarms = nouveau_therm_program_alarms_polling;
 	return nouveau_therm_preinit(&priv->base.base);
 }
 

drivers/gpu/drm/nouveau/core/subdev/therm/priv.h

@@ -50,6 +50,24 @@ struct nouveau_fan {
 	struct dcb_gpio_func tach;
 };
 
+enum nouveau_therm_thrs_direction {
+	NOUVEAU_THERM_THRS_FALLING = 0,
+	NOUVEAU_THERM_THRS_RISING = 1
+};
+
+enum nouveau_therm_thrs_state {
+	NOUVEAU_THERM_THRS_LOWER = 0,
+	NOUVEAU_THERM_THRS_HIGHER = 1
+};
+
+enum nouveau_therm_thrs {
+	NOUVEAU_THERM_THRS_FANBOOST = 0,
+	NOUVEAU_THERM_THRS_DOWNCLOCK = 1,
+	NOUVEAU_THERM_THRS_CRITICAL = 2,
+	NOUVEAU_THERM_THRS_SHUTDOWN = 3,
+	NOUVEAU_THERM_THRS_NR
+};
+
 struct nouveau_therm_priv {
 	struct nouveau_therm base;
 
@@ -65,10 +83,25 @@ struct nouveau_therm_priv {
 	/* fan priv */
 	struct nouveau_fan *fan;
 
+	/* alarms priv */
+	struct {
+		spinlock_t alarm_program_lock;
+		struct nouveau_alarm therm_poll_alarm;
+		enum nouveau_therm_thrs_state alarm_state[NOUVEAU_THERM_THRS_NR];
+		void (*program_alarms)(struct nouveau_therm *);
+	} sensor;
+
+	/* what should be done if the card overheats */
+	struct {
+		void (*downclock)(struct nouveau_therm *, bool active);
+		void (*pause)(struct nouveau_therm *, bool active);
+	} emergency;
+
 	/* ic */
 	struct i2c_client *ic;
 };
 
+int nouveau_therm_mode(struct nouveau_therm *therm, int mode);
 int nouveau_therm_attr_get(struct nouveau_therm *therm,
 		       enum nouveau_therm_attr_type type);
 int nouveau_therm_attr_set(struct nouveau_therm *therm,
@@ -88,6 +121,17 @@ int nouveau_therm_fan_sense(struct nouveau_therm *therm);
 
 int nouveau_therm_preinit(struct nouveau_therm *);
 
+void nouveau_therm_sensor_set_threshold_state(struct nouveau_therm *therm,
+					     enum nouveau_therm_thrs thrs,
+					     enum nouveau_therm_thrs_state st);
+enum nouveau_therm_thrs_state
+nouveau_therm_sensor_get_threshold_state(struct nouveau_therm *therm,
+					 enum nouveau_therm_thrs thrs);
+void nouveau_therm_sensor_event(struct nouveau_therm *therm,
+			        enum nouveau_therm_thrs thrs,
+			        enum nouveau_therm_thrs_direction dir);
+void nouveau_therm_program_alarms_polling(struct nouveau_therm *therm);
+
 int nv50_fan_pwm_ctrl(struct nouveau_therm *, int, bool);
 int nv50_fan_pwm_get(struct nouveau_therm *, int, u32 *, u32 *);
 int nv50_fan_pwm_set(struct nouveau_therm *, int, u32, u32);

drivers/gpu/drm/nouveau/core/subdev/therm/temp.c

@@ -65,12 +65,150 @@ nouveau_therm_temp_safety_checks(struct nouveau_therm *therm)
 		priv->bios_sensor.offset_den = 1;
 }
 
+/* must be called with alarm_program_lock taken ! */
+void nouveau_therm_sensor_set_threshold_state(struct nouveau_therm *therm,
+					     enum nouveau_therm_thrs thrs,
+					     enum nouveau_therm_thrs_state st)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+	priv->sensor.alarm_state[thrs] = st;
+}
+
+/* must be called with alarm_program_lock taken ! */
+enum nouveau_therm_thrs_state
+nouveau_therm_sensor_get_threshold_state(struct nouveau_therm *therm,
+					 enum nouveau_therm_thrs thrs)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+	return priv->sensor.alarm_state[thrs];
+}
+
+void nouveau_therm_sensor_event(struct nouveau_therm *therm,
+			        enum nouveau_therm_thrs thrs,
+			        enum nouveau_therm_thrs_direction dir)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+	bool active;
+	const char *thresolds[] = {
+		"fanboost", "downclock", "critical", "shutdown"
+	};
+	uint8_t temperature = therm->temp_get(therm);
+
+	if (thrs < 0 || thrs > 3)
+		return;
+
+	if (dir == NOUVEAU_THERM_THRS_FALLING)
+		nv_info(therm, "temperature (%u C) went bellow the '%s' threshold\n",
+			temperature, thresolds[thrs]);
+	else
+		nv_info(therm, "temperature (%u C) hit the '%s' threshold\n",
+			temperature, thresolds[thrs]);
+
+	active = (dir == NOUVEAU_THERM_THRS_RISING);
+	switch (thrs) {
+	case NOUVEAU_THERM_THRS_FANBOOST:
+		nouveau_therm_fan_set(therm, true, 100);
+		nouveau_therm_mode(therm, FAN_CONTROL_AUTO);
+		break;
+	case NOUVEAU_THERM_THRS_DOWNCLOCK:
+		if (priv->emergency.downclock)
+			priv->emergency.downclock(therm, active);
+		break;
+	case NOUVEAU_THERM_THRS_CRITICAL:
+		if (priv->emergency.pause)
+			priv->emergency.pause(therm, active);
+		break;
+	case NOUVEAU_THERM_THRS_SHUTDOWN:
+		orderly_poweroff(true);
+		break;
+	case NOUVEAU_THERM_THRS_NR:
+		break;
+	}
+
+}
+
+/* must be called with alarm_program_lock taken ! */
+static void
+nouveau_therm_threshold_hyst_polling(struct nouveau_therm *therm,
+				   const struct nvbios_therm_threshold *thrs,
+				   enum nouveau_therm_thrs thrs_name)
+{
+	enum nouveau_therm_thrs_direction direction;
+	enum nouveau_therm_thrs_state prev_state, new_state;
+	int temp = therm->temp_get(therm);
+
+	prev_state = nouveau_therm_sensor_get_threshold_state(therm, thrs_name);
+
+	if (temp >= thrs->temp && prev_state == NOUVEAU_THERM_THRS_LOWER) {
+		direction = NOUVEAU_THERM_THRS_RISING;
+		new_state = NOUVEAU_THERM_THRS_HIGHER;
+	} else if (temp <= thrs->temp - thrs->hysteresis &&
+			prev_state == NOUVEAU_THERM_THRS_HIGHER) {
+		direction = NOUVEAU_THERM_THRS_FALLING;
+		new_state = NOUVEAU_THERM_THRS_LOWER;
+	} else
+		return; /* nothing to do */
+
+	nouveau_therm_sensor_set_threshold_state(therm, thrs_name, new_state);
+	nouveau_therm_sensor_event(therm, thrs_name, direction);
+}
+
+static void
+alarm_timer_callback(struct nouveau_alarm *alarm)
+{
+	struct nouveau_therm_priv *priv =
+	container_of(alarm, struct nouveau_therm_priv, sensor.therm_poll_alarm);
+	struct nvbios_therm_sensor *sensor = &priv->bios_sensor;
+	struct nouveau_timer *ptimer = nouveau_timer(priv);
+	struct nouveau_therm *therm = &priv->base;
+	unsigned long flags;
+
+	spin_lock_irqsave(&priv->sensor.alarm_program_lock, flags);
+
+	nouveau_therm_threshold_hyst_polling(therm, &sensor->thrs_fan_boost,
+					     NOUVEAU_THERM_THRS_FANBOOST);
+
+	nouveau_therm_threshold_hyst_polling(therm, &sensor->thrs_down_clock,
+					     NOUVEAU_THERM_THRS_DOWNCLOCK);
+
+	nouveau_therm_threshold_hyst_polling(therm, &sensor->thrs_critical,
+					     NOUVEAU_THERM_THRS_CRITICAL);
+
+	nouveau_therm_threshold_hyst_polling(therm, &sensor->thrs_shutdown,
+					     NOUVEAU_THERM_THRS_SHUTDOWN);
+
+	/* schedule the next poll in one second */
+	if (list_empty(&alarm->head))
+		ptimer->alarm(ptimer, 1000 * 1000 * 1000, alarm);
+
+	spin_unlock_irqrestore(&priv->sensor.alarm_program_lock, flags);
+}
+
+void
+nouveau_therm_program_alarms_polling(struct nouveau_therm *therm)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+	struct nvbios_therm_sensor *sensor = &priv->bios_sensor;
+
+	nv_info(therm,
+		"programmed thresholds [ %d(%d), %d(%d), %d(%d), %d(%d) ]\n",
+		sensor->thrs_fan_boost.temp, sensor->thrs_fan_boost.hysteresis,
+		sensor->thrs_down_clock.temp,
+		sensor->thrs_down_clock.hysteresis,
+		sensor->thrs_critical.temp, sensor->thrs_critical.hysteresis,
+		sensor->thrs_shutdown.temp, sensor->thrs_shutdown.hysteresis);
+
+	alarm_timer_callback(&priv->sensor.therm_poll_alarm);
+}
+
 int
 nouveau_therm_sensor_ctor(struct nouveau_therm *therm)
 {
 	struct nouveau_therm_priv *priv = (void *)therm;
 	struct nouveau_bios *bios = nouveau_bios(therm);
 
+	nouveau_alarm_init(&priv->sensor.therm_poll_alarm, alarm_timer_callback);
+
 	nouveau_therm_temp_set_defaults(therm);
 	if (nvbios_therm_sensor_parse(bios, NVBIOS_THERM_DOMAIN_CORE,
 				      &priv->bios_sensor))