Skip to contentSkip to conversation

charger: tps65090: Allow charger module to be used when no irq

PATCH RESEND Applied Doug Anderson 4 participants 7 messages Apr 23, 2014 – May 6, 2014

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH RESEND v3 2/5 charger: tps65090: Allow charger module to be used when no irq

On the ARM Chromebook tps65090 has two masters: the AP (the main processor running linux) and the EC (the embedded controller). The AP is allowed to mess with FETs but the EC is in charge of charge control.

The tps65090 interupt line is routed to both the AP and the EC, which can cause quite a headache. Having two people adjusting masks and acking interrupts is a recipe for disaster.

In the shipping kernel we had a hack to have the AP pay attention to the IRQ but not to ack it. It also wasn't supposed to configure the IRQ in any way. That hack allowed us to detect when the device was charging without messing with the EC's state.

The current tps65090 infrastructure makes the above difficult, and it was a bit of a hack to begin with. Rather than uglify the driver to support it, just extend the driver's existing notion of "no irq" to the charger. This makes the charger code poll every 2 seconds for AC detect, which is sufficient.

For proper functioning, requires (mfd: tps65090: Don't tell child devices we have an IRQ if we don't). If we don't have that patch we'll simply fail to probe on devices without an interrupt (just like we did before this patch).

---
Changes in v3: None
Changes in v2:
- Split noirq (polling mode) changes into MFD and charger

 drivers/power/tps65090-charger.c | 76 +++++++++++++++++++++++++++++++---------
 1 file changed, 59 insertions(+), 17 deletions(-)
drivers/power/tps65090-charger.c +59 -17
index 8fc9d6d..cc26c12 100644
@@ -17,9 +17,11 @@
  */
 #include <linux/delay.h>
 #include <linux/err.h>
+#include <linux/freezer.h>
 #include <linux/init.h>
 #include <linux/interrupt.h>
 #include <linux/kernel.h>
+#include <linux/kthread.h>
 #include <linux/module.h>
 #include <linux/of_device.h>
 #include <linux/platform_device.h>
@@ -43,11 +45,15 @@
 #define TPS65090_VACG		BIT(1)
 #define TPS65090_NOITERM	BIT(5)
 
+#define POLL_INTERVAL		(HZ * 2)	/* Used when no irq */
+
 struct tps65090_charger {
 	struct	device	*dev;
 	int	ac_online;
 	int	prev_ac_online;
 	int	irq;
+	struct task_struct	*poll_task;
+	bool			passive_mode;
 	struct power_supply	ac;
 	struct tps65090_platform_data *pdata;
 };
@@ -60,6 +66,9 @@ static int tps65090_low_chrg_current(struct tps65090_charger *charger)
 {
 	int ret;
 
+	if (charger->passive_mode)
+		return 0;
+
 	ret = tps65090_write(charger->dev->parent, TPS65090_REG_CG_CTRL5,
 			TPS65090_NOITERM);
 	if (ret < 0) {
@@ -75,6 +84,9 @@ static int tps65090_enable_charging(struct tps65090_charger *charger)
 	int ret;
 	uint8_t ctrl0 = 0;
 
+	if (charger->passive_mode)
+		return 0;
+
 	ret = tps65090_read(charger->dev->parent, TPS65090_REG_CG_CTRL0,
 			    &ctrl0);
 	if (ret < 0) {
@@ -98,6 +110,9 @@ static int tps65090_config_charger(struct tps65090_charger *charger)
 	uint8_t intrmask = 0;
 	int ret;
 
+	if (charger->passive_mode)
+		return 0;
+
 	if (charger->pdata->enable_low_current_chrg) {
 		ret = tps65090_low_chrg_current(charger);
 		if (ret < 0) {
@@ -175,10 +190,14 @@ static irqreturn_t tps65090_charger_isr(int irq, void *dev_id)
 	}
 
 	/* Clear interrupts. */
-	ret = tps65090_write(charger->dev->parent, TPS65090_REG_INTR_STS, 0x00);
-	if (ret < 0) {
-		dev_err(charger->dev, "%s(): Error in writing reg 0x%x\n",
+	if (!charger->passive_mode) {
+		ret = tps65090_write(charger->dev->parent,
+				     TPS65090_REG_INTR_STS, 0x00);
+		if (ret < 0) {
+			dev_err(charger->dev,
+				"%s(): Error in writing reg 0x%x\n",
 				__func__, TPS65090_REG_INTR_STS);
+		}
 	}
 
 	if (charger->prev_ac_online != charger->ac_online)
@@ -209,6 +228,18 @@ static struct tps65090_platform_data *
 
 }
 
+static int tps65090_charger_poll_task(void *data)
+{
+	set_freezable();
+
+	while (!kthread_should_stop()) {
+		schedule_timeout_interruptible(POLL_INTERVAL);
+		try_to_freeze();
+		tps65090_charger_isr(-1, data);
+	}
+	return 0;
+}
+
 static int tps65090_charger_probe(struct platform_device *pdev)
 {
 	struct tps65090_charger *cdata;
@@ -255,22 +286,10 @@ static int tps65090_charger_probe(struct platform_device *pdev)
 	}
 
 	irq = platform_get_irq(pdev, 0);
-	if (irq <= 0) {
-		dev_warn(&pdev->dev, "Unable to get charger irq = %d\n", irq);
-		ret = irq;
-		goto fail_unregister_supply;
-	}
-
+	if (irq < 0)
+		irq = NO_IRQ;
 	cdata->irq = irq;
 
-	ret = devm_request_threaded_irq(&pdev->dev, irq, NULL,
-		tps65090_charger_isr, 0, "tps65090-charger", cdata);
-	if (ret) {
-		dev_err(cdata->dev, "Unable to register irq %d err %d\n", irq,
-			ret);
-		goto fail_unregister_supply;
-	}
-
 	ret = tps65090_config_charger(cdata);
 	if (ret < 0) {
 		dev_err(&pdev->dev, "charger config failed, err %d\n", ret);
@@ -296,6 +315,27 @@ static int tps65090_charger_probe(struct platform_device *pdev)
 		power_supply_changed(&cdata->ac);
 	}
 
+	if (irq != NO_IRQ) {
+		ret = devm_request_threaded_irq(&pdev->dev, irq, NULL,
+			tps65090_charger_isr, 0, "tps65090-charger", cdata);
+		if (ret) {
+			dev_err(cdata->dev,
+				"Unable to register irq %d err %d\n", irq,
+				ret);
+			goto fail_unregister_supply;
+		}
+	} else {
+		cdata->poll_task = kthread_run(tps65090_charger_poll_task,
+					      cdata, "ktps65090charger");
+		cdata->passive_mode = true;
+		if (IS_ERR(cdata->poll_task)) {
+			ret = PTR_ERR(cdata->poll_task);
+			dev_err(cdata->dev,
+				"Unable to run kthread err %d\n", ret);
+			goto fail_unregister_supply;
+		}
+	}
+
 	return 0;
 
 fail_unregister_supply:
@@ -308,6 +348,8 @@ static int tps65090_charger_remove(struct platform_device *pdev)
 {
 	struct tps65090_charger *cdata = platform_get_drvdata(pdev);
 
+	if (cdata->irq == NO_IRQ)
+		kthread_stop(cdata->poll_task);
 	power_supply_unregister(&cdata->ac);
 
 	return 0;
Signature · 1 line
1.9.1.423.g4596e3a
#1 lore raw

PATCH RESEND v3 5/5 regulator: tps65090: Make FETs more reliable by adding retries

An issue was discovered with tps65090 where sometimes the FETs wouldn't actually turn on when requested (they would report overcurrent). The most problematic FET was the one used for the LCD backlight on the Samsung ARM Chromebook (FET1). Problems were especially prevalent when the device was plugged in to AC power (when the backlight voltage was higher).

Mitigate the problem by adding retries on the enables of the FETs, which works around the problem fairly effectively.

---
Changes in v3:
- Fixed kernel-doc notation for return

Changes in v2:
- Separated the overcurrent and retries changes into two patches.
- No longer open code fet_is_enabled().
- Fixed tps6090 typo.
- For loop => "while true".
- Removed a set of braces.

 drivers/regulator/tps65090-regulator.c | 155 +++++++++++++++++++++++++++++----
 1 file changed, 140 insertions(+), 15 deletions(-)
drivers/regulator/tps65090-regulator.c +140 -15
index ca04e9f..2057e2e 100644
@@ -17,6 +17,7 @@
  */
 
 #include <linux/module.h>
+#include <linux/delay.h>
 #include <linux/init.h>
 #include <linux/gpio.h>
 #include <linux/of_gpio.h>
@@ -28,7 +29,13 @@
 #include <linux/regulator/of_regulator.h>
 #include <linux/mfd/tps65090.h>
 
+#define MAX_CTRL_READ_TRIES	5
+#define MAX_FET_ENABLE_TRIES	1000
+
+#define CTRL_EN_BIT		0 /* Regulator enable bit, active high */
 #define CTRL_WT_BIT		2 /* Regulator wait time 0 bit */
+#define CTRL_PG_BIT		4 /* Regulator power good bit, 1=good */
+#define CTRL_TO_BIT		7 /* Regulator timeout bit, 1=wait */
 
 #define MAX_OVERCURRENT_WAIT	3 /* Overcurrent wait must be <= this */
 
@@ -80,40 +87,158 @@ static int tps65090_reg_set_overcurrent_wait(struct tps65090_regulator *ri,
 	return ret;
 }
 
-static struct regulator_ops tps65090_reg_contol_ops = {
+/**
+ * tps65090_try_enable_fet - Try to enable a FET
+ *
+ * @rdev:	Regulator device
+ *
+ * Return: 0 if ok, -ENOTRECOVERABLE if the FET power good bit did not get
+ * set, or some other -ve value if another error occurred (e.g. i2c error)
+ */
+static int tps65090_try_enable_fet(struct regulator_dev *rdev)
+{
+	unsigned int control;
+	int ret, i;
+
+	ret = regmap_update_bits(rdev->regmap, rdev->desc->enable_reg,
+				 rdev->desc->enable_mask,
+				 rdev->desc->enable_mask);
+	if (ret < 0) {
+		dev_err(&rdev->dev, "Error in updating reg %#x\n",
+			rdev->desc->enable_reg);
+		return ret;
+	}
+
+	for (i = 0; i < MAX_CTRL_READ_TRIES; i++) {
+		ret = regmap_read(rdev->regmap, rdev->desc->enable_reg,
+				  &control);
+		if (ret < 0)
+			return ret;
+
+		if (!(control & BIT(CTRL_TO_BIT)))
+			break;
+
+		usleep_range(1000, 1500);
+	}
+	if (!(control & BIT(CTRL_PG_BIT)))
+		return -ENOTRECOVERABLE;
+
+	return 0;
+}
+
+/**
+ * tps65090_fet_enable - Enable a FET, trying a few times if it fails
+ *
+ * Some versions of the tps65090 have issues when turning on the FETs.
+ * This function goes through several steps to ensure the best chance of the
+ * FET going on.  Specifically:
+ * - We'll make sure that we bump the "overcurrent wait" to the maximum, which
+ *   increases the chances that we'll turn on properly.
+ * - We'll retry turning the FET on multiple times (turning off in between).
+ *
+ * @rdev:	Regulator device
+ *
+ * Return: 0 if ok, non-zero if it fails.
+ */
+static int tps65090_fet_enable(struct regulator_dev *rdev)
+{
+	int ret, tries;
+
+	/*
+	 * Try enabling multiple times until we succeed since sometimes the
+	 * first try times out.
+	 */
+	tries = 0;
+	while (true) {
+		ret = tps65090_try_enable_fet(rdev);
+		if (!ret)
+			break;
+		if (ret != -ENOTRECOVERABLE || tries == MAX_FET_ENABLE_TRIES)
+			goto err;
+
+		/* Try turning the FET off (and then on again) */
+		ret = regmap_update_bits(rdev->regmap, rdev->desc->enable_reg,
+					 rdev->desc->enable_mask, 0);
+		if (ret)
+			goto err;
+
+		tries++;
+	}
+
+	if (tries)
+		dev_warn(&rdev->dev, "reg %#x enable ok after %d tries\n",
+			 rdev->desc->enable_reg, tries);
+
+	return 0;
+err:
+	dev_warn(&rdev->dev, "reg %#x enable failed\n", rdev->desc->enable_reg);
+	WARN_ON(1);
+
+	return ret;
+}
+
+static struct regulator_ops tps65090_reg_control_ops = {
 	.enable		= regulator_enable_regmap,
 	.disable	= regulator_disable_regmap,
 	.is_enabled	= regulator_is_enabled_regmap,
 };
 
+static struct regulator_ops tps65090_fet_control_ops = {
+	.enable		= tps65090_fet_enable,
+	.disable	= regulator_disable_regmap,
+	.is_enabled	= regulator_is_enabled_regmap,
+};
+
 static struct regulator_ops tps65090_ldo_ops = {
 };
 
-#define tps65090_REG_DESC(_id, _sname, _en_reg, _ops)	\
+#define tps65090_REG_DESC(_id, _sname, _en_reg, _en_bits, _ops)	\
 {							\
 	.name = "TPS65090_RAILS"#_id,			\
 	.supply_name = _sname,				\
 	.id = TPS65090_REGULATOR_##_id,			\
 	.ops = &_ops,					\
 	.enable_reg = _en_reg,				\
-	.enable_mask = BIT(0),				\
+	.enable_val = _en_bits,				\
+	.enable_mask = _en_bits,			\
 	.type = REGULATOR_VOLTAGE,			\
 	.owner = THIS_MODULE,				\
 }
 
 static struct regulator_desc tps65090_regulator_desc[] = {
-	tps65090_REG_DESC(DCDC1, "vsys1",   0x0C, tps65090_reg_contol_ops),
-	tps65090_REG_DESC(DCDC2, "vsys2",   0x0D, tps65090_reg_contol_ops),
-	tps65090_REG_DESC(DCDC3, "vsys3",   0x0E, tps65090_reg_contol_ops),
-	tps65090_REG_DESC(FET1,  "infet1",  0x0F, tps65090_reg_contol_ops),
-	tps65090_REG_DESC(FET2,  "infet2",  0x10, tps65090_reg_contol_ops),
-	tps65090_REG_DESC(FET3,  "infet3",  0x11, tps65090_reg_contol_ops),
-	tps65090_REG_DESC(FET4,  "infet4",  0x12, tps65090_reg_contol_ops),
-	tps65090_REG_DESC(FET5,  "infet5",  0x13, tps65090_reg_contol_ops),
-	tps65090_REG_DESC(FET6,  "infet6",  0x14, tps65090_reg_contol_ops),
-	tps65090_REG_DESC(FET7,  "infet7",  0x15, tps65090_reg_contol_ops),
-	tps65090_REG_DESC(LDO1,  "vsys-l1", 0,    tps65090_ldo_ops),
-	tps65090_REG_DESC(LDO2,  "vsys-l2", 0,    tps65090_ldo_ops),
+	tps65090_REG_DESC(DCDC1, "vsys1",   0x0C, BIT(CTRL_EN_BIT),
+			  tps65090_reg_control_ops),
+	tps65090_REG_DESC(DCDC2, "vsys2",   0x0D, BIT(CTRL_EN_BIT),
+			  tps65090_reg_control_ops),
+	tps65090_REG_DESC(DCDC3, "vsys3",   0x0E, BIT(CTRL_EN_BIT),
+			  tps65090_reg_control_ops),
+
+	tps65090_REG_DESC(FET1,  "infet1",  0x0F,
+			  BIT(CTRL_EN_BIT) | BIT(CTRL_PG_BIT),
+			  tps65090_fet_control_ops),
+	tps65090_REG_DESC(FET2,  "infet2",  0x10,
+			  BIT(CTRL_EN_BIT) | BIT(CTRL_PG_BIT),
+			  tps65090_fet_control_ops),
+	tps65090_REG_DESC(FET3,  "infet3",  0x11,
+			  BIT(CTRL_EN_BIT) | BIT(CTRL_PG_BIT),
+			  tps65090_fet_control_ops),
+	tps65090_REG_DESC(FET4,  "infet4",  0x12,
+			  BIT(CTRL_EN_BIT) | BIT(CTRL_PG_BIT),
+			  tps65090_fet_control_ops),
+	tps65090_REG_DESC(FET5,  "infet5",  0x13,
+			  BIT(CTRL_EN_BIT) | BIT(CTRL_PG_BIT),
+			  tps65090_fet_control_ops),
+	tps65090_REG_DESC(FET6,  "infet6",  0x14,
+			  BIT(CTRL_EN_BIT) | BIT(CTRL_PG_BIT),
+			  tps65090_fet_control_ops),
+	tps65090_REG_DESC(FET7,  "infet7",  0x15,
+			  BIT(CTRL_EN_BIT) | BIT(CTRL_PG_BIT),
+			  tps65090_fet_control_ops),
+
+	tps65090_REG_DESC(LDO1,  "vsys-l1", 0, 0,
+			  tps65090_ldo_ops),
+	tps65090_REG_DESC(LDO2,  "vsys-l2", 0, 0,
+			  tps65090_ldo_ops),
 };
 
 static inline bool is_dcdc(int id)
Signature · 1 line
1.9.1.423.g4596e3a
maintainer Applied
#2 lore raw

Re: [RESEND PATCH v3 5/5] regulator: tps65090: Make FETs more reliable by adding retries

Quoted 3 lines · Doug Anderson

An issue was discovered with tps65090 where sometimes the FETs wouldn't actually turn on when requested (they would report overcurrent). The most problematic FET was the one used for the LCD

Applied, thanks.

#3 lore raw
In reply to Doug Anderson

Anton,

Quoted 24 lines · Doug Anderson

On the ARM Chromebook tps65090 has two masters: the AP (the main processor running linux) and the EC (the embedded controller). The AP is allowed to mess with FETs but the EC is in charge of charge control.

The tps65090 interupt line is routed to both the AP and the EC, which can cause quite a headache. Having two people adjusting masks and acking interrupts is a recipe for disaster.

In the shipping kernel we had a hack to have the AP pay attention to the IRQ but not to ack it. It also wasn't supposed to configure the IRQ in any way. That hack allowed us to detect when the device was charging without messing with the EC's state.

The current tps65090 infrastructure makes the above difficult, and it was a bit of a hack to begin with. Rather than uglify the driver to support it, just extend the driver's existing notion of "no irq" to the charger. This makes the charger code poll every 2 seconds for AC detect, which is sufficient.

For proper functioning, requires (mfd: tps65090: Don't tell child devices we have an IRQ if we don't). If we don't have that patch we'll simply fail to probe on devices without an interrupt (just like we did before this patch).

Quoted 8 lines

---
Changes in v3: None
Changes in v2:
- Split noirq (polling mode) changes into MFD and charger

 drivers/power/tps65090-charger.c | 76 +++++++++++++++++++++++++++++++---------
 1 file changed, 59 insertions(+), 17 deletions(-)

All the rest of this series has been acked and applied. Do you have time to review this patch?

Thanks! :)

-Doug

#4 lore raw
Quoted 33 lines · Doug Anderson

Anton,

Quoted 30 lines · Doug Anderson

On the ARM Chromebook tps65090 has two masters: the AP (the main processor running linux) and the EC (the embedded controller). The AP is allowed to mess with FETs but the EC is in charge of charge control.

The tps65090 interupt line is routed to both the AP and the EC, which can cause quite a headache. Having two people adjusting masks and acking interrupts is a recipe for disaster.

In the shipping kernel we had a hack to have the AP pay attention to the IRQ but not to ack it. It also wasn't supposed to configure the IRQ in any way. That hack allowed us to detect when the device was charging without messing with the EC's state.

The current tps65090 infrastructure makes the above difficult, and it was a bit of a hack to begin with. Rather than uglify the driver to support it, just extend the driver's existing notion of "no irq" to the charger. This makes the charger code poll every 2 seconds for AC detect, which is sufficient.

For proper functioning, requires (mfd: tps65090: Don't tell child devices we have an IRQ if we don't). If we don't have that patch we'll simply fail to probe on devices without an interrupt (just like we did before this patch).

---
Changes in v3: None
Changes in v2:
- Split noirq (polling mode) changes into MFD and charger

Quoted 7 lines
Quoted 2 lines
 drivers/power/tps65090-charger.c | 76 +++++++++++++++++++++++++++++++---------
 1 file changed, 59 insertions(+), 17 deletions(-)

All the rest of this series has been acked and applied. Do you have time to review this patch?

Thanks! :)

FWIW, I've seen very little email traffic from Anton this year, he might have limited time for maintainership at the moment. Usually what we do in these cases is that we give the maintainer as much time as possible, but eventually if there's no action, and the patches look reasonable, we apply them with the maintainer on Cc: line in the signed-offs. Or we send them to/through akpm, which tends to be the catch-all for high-latency maintainers, etc.

Since this patch isn't holding anything up, let's give Anton as much time as possible to review, and revisit in ~2 weeks (so it has time to land somewhere in time for the merge window).

-Olof

#5 lore raw
Quoted 7 lines · Olof Johansson
Quoted 4 lines

All the rest of this series has been acked and applied. Do you have time to review this patch?

Thanks! :)

FWIW, I've seen very little email traffic from Anton this year, he might have limited time for maintainership at the moment.

Yup.

commit 573189354b7c97cd2256b87cf083ee435584594e

Author: Dmitry Eremin-Solenikov <[email protected]>
Date:   Tue Jan 21 03:33:09 2014 +0400

MAINTAINERS: Pick up power supply maintainership

Anton stated that he would have to abandon power supply maintainership due to the lack of time. By agreement with him and David, pick up power supply tree.

#6 lore raw

Anton,

Quoted 15 lines · Anton Vorontsov
Quoted 7 lines · Olof Johansson
Quoted 4 lines

All the rest of this series has been acked and applied. Do you have time to review this patch?

Thanks! :)

FWIW, I've seen very little email traffic from Anton this year, he might have limited time for maintainership at the moment.

Yup.

commit 573189354b7c97cd2256b87cf083ee435584594e

Author: Dmitry Eremin-Solenikov <[email protected]>
Date:   Tue Jan 21 03:33:09 2014 +0400
Quoted 8 lines

MAINTAINERS: Pick up power supply maintainership

Anton stated that he would have to abandon power supply maintainership due to the lack of time. By agreement with him and David, pick up power supply tree.

Doh, thanks for the pointer.

Dmitry: you were CCed on the patch. Do you still have it in your email somewhere? I'm happy to repost it with you in the "To" line if it's helpful.

Thanks! :)

-Doug