diff mbox series

[v2,2/3] pinctrl: cherryview: Do not allow the same interrupt line to be used by 2 pins

Message ID 20211118105650.207638-2-hdegoede@redhat.com
State Superseded
Headers show
Series None | expand

Commit Message

Hans de Goede Nov. 18, 2021, 10:56 a.m. UTC
It is impossible to use the same interrupt line for 2 pins, this will
result in the interrupts only being delivered to the IRQ handler for
the pin for which chv_gpio_irq_type() was called last.

The pinctrl-cherryview.c code relies on the BIOS to correctly setup the
interrupt line, but there is a BIOS bug on at least the Medion Akoya E1239T
and the GPD win models where both INT33FF:02 pin 8, used by the powerbutton
and INT33FF:02 pin 21 used as IRQ input for the accelerometer are mapped to
interrupt line 0.

This causes 2 problems:
1. The accelerometer IRQ does not work, since the power button is probed
later taking over the intr_lines[0] slot.

2. Since the accelerometer IRQ is not marked as wakeup, interrupt line 0
gets masked on suspend, causing the power button to not work to wake
the system from suspend.

Likewise on the Lenovo Yogabook, which has a touchscreen as keyboard
and the keyboard half of the tablet also has a Wacom digitizer, the BIOS
by default assigns the same interrupt line to the GPIOs used
for their interrupts.

Fix these problems by adding a check for this and assigning a new
interrupt line to the 2nd pin for which chv_gpio_irq_type() gets called.

With this fix in place the following 2 messages show up in dmesg on
the Medion Akoya E1239T and the GPD win:

 cherryview-pinctrl INT33FF:02: interrupt line 0 is used by both pin 21 and pin 8
 cherryview-pinctrl INT33FF:02: changing the interrupt line for pin 8 to 15

And the following gets logged on the Lenovo Yogabook:

 cherryview-pinctrl INT33FF:01: interrupt-line 0 is used by both pin 49 and pin 56
 cherryview-pinctrl INT33FF:01: changing the interrupt line for pin 56 to 7

Note commit 9747070c11d6 ("Input: axp20x-pek - always register interrupt
handlers") was added as a work around for the power button not being able
to wakeup the system. This relies on using the PMIC's connection to the
power button but that only works on systems with the AXP288 PMIC.
Once this fix has been merged that workaround can be removed.

Cc: Yauhen Kharuzhy <jekhor@gmail.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
---
Changes in v2:
- Reword commit + log messages a bit (as requested by Mika)
---
 drivers/pinctrl/intel/pinctrl-cherryview.c | 69 +++++++++++++++++++---
 1 file changed, 61 insertions(+), 8 deletions(-)

Comments

Andy Shevchenko Nov. 26, 2021, 6:13 p.m. UTC | #1
On Thu, Nov 18, 2021 at 01:14:53PM +0200, Mika Westerberg wrote:
> On Thu, Nov 18, 2021 at 11:56:49AM +0100, Hans de Goede wrote:
> > It is impossible to use the same interrupt line for 2 pins, this will
> > result in the interrupts only being delivered to the IRQ handler for
> > the pin for which chv_gpio_irq_type() was called last.
> > 
> > The pinctrl-cherryview.c code relies on the BIOS to correctly setup the
> > interrupt line, but there is a BIOS bug on at least the Medion Akoya E1239T
> > and the GPD win models where both INT33FF:02 pin 8, used by the powerbutton
> > and INT33FF:02 pin 21 used as IRQ input for the accelerometer are mapped to
> > interrupt line 0.
> > 
> > This causes 2 problems:
> > 1. The accelerometer IRQ does not work, since the power button is probed
> > later taking over the intr_lines[0] slot.
> > 
> > 2. Since the accelerometer IRQ is not marked as wakeup, interrupt line 0
> > gets masked on suspend, causing the power button to not work to wake
> > the system from suspend.
> > 
> > Likewise on the Lenovo Yogabook, which has a touchscreen as keyboard
> > and the keyboard half of the tablet also has a Wacom digitizer, the BIOS
> > by default assigns the same interrupt line to the GPIOs used
> > for their interrupts.
> > 
> > Fix these problems by adding a check for this and assigning a new
> > interrupt line to the 2nd pin for which chv_gpio_irq_type() gets called.
> > 
> > With this fix in place the following 2 messages show up in dmesg on
> > the Medion Akoya E1239T and the GPD win:
> > 
> >  cherryview-pinctrl INT33FF:02: interrupt line 0 is used by both pin 21 and pin 8
> >  cherryview-pinctrl INT33FF:02: changing the interrupt line for pin 8 to 15
> > 
> > And the following gets logged on the Lenovo Yogabook:
> > 
> >  cherryview-pinctrl INT33FF:01: interrupt-line 0 is used by both pin 49 and pin 56
> >  cherryview-pinctrl INT33FF:01: changing the interrupt line for pin 56 to 7
> > 
> > Note commit 9747070c11d6 ("Input: axp20x-pek - always register interrupt
> > handlers") was added as a work around for the power button not being able
> > to wakeup the system. This relies on using the PMIC's connection to the
> > power button but that only works on systems with the AXP288 PMIC.
> > Once this fix has been merged that workaround can be removed.
> > 
> > Cc: Yauhen Kharuzhy <jekhor@gmail.com>
> > Signed-off-by: Hans de Goede <hdegoede@redhat.com>
> 
> Acked-by: Mika Westerberg <mika.westerberg@linux.intel.com>

Pushed to my review and testing queue, thanks!
diff mbox series

Patch

diff --git a/drivers/pinctrl/intel/pinctrl-cherryview.c b/drivers/pinctrl/intel/pinctrl-cherryview.c
index 55ccdcecd94e..da68f8a849ab 100644
--- a/drivers/pinctrl/intel/pinctrl-cherryview.c
+++ b/drivers/pinctrl/intel/pinctrl-cherryview.c
@@ -1323,6 +1323,8 @@  static unsigned chv_gpio_irq_startup(struct irq_data *d)
 
 		if (cctx->intr_lines[intsel] == INTR_LINE_UNUSED) {
 			irq_set_handler_locked(d, handler);
+			dev_dbg(pctrl->dev, "using interrupt line %u for IRQ_TYPE_NONE IRQ on pin %u\n",
+				intsel, pin);
 			cctx->intr_lines[intsel] = pin;
 		}
 		raw_spin_unlock_irqrestore(&chv_lock, flags);
@@ -1332,17 +1334,73 @@  static unsigned chv_gpio_irq_startup(struct irq_data *d)
 	return 0;
 }
 
+static int chv_gpio_set_intr_line(struct intel_pinctrl *pctrl, unsigned int pin)
+{
+	struct intel_community_context *cctx = &pctrl->context.communities[0];
+	const struct intel_community *community = &pctrl->communities[0];
+	u32 value, intsel;
+	int i;
+
+	value = chv_readl(pctrl, pin, CHV_PADCTRL0);
+	intsel = (value & CHV_PADCTRL0_INTSEL_MASK) >> CHV_PADCTRL0_INTSEL_SHIFT;
+
+	if (cctx->intr_lines[intsel] == pin)
+		return 0;
+
+	if (cctx->intr_lines[intsel] == INTR_LINE_UNUSED) {
+		dev_dbg(pctrl->dev, "using interrupt line %u for pin %u\n", intsel, pin);
+		cctx->intr_lines[intsel] = pin;
+		return 0;
+	}
+
+	/*
+	 * The interrupt line selected by the BIOS is already in use by
+	 * another pin, this is a known BIOS bug found on several models.
+	 * But this may also be caused by Linux deciding to use a pin as
+	 * IRQ which was not expected to be used as such by the BIOS authors,
+	 * so log this at info level only.
+	 */
+	dev_info(pctrl->dev, "interrupt line %u is used by both pin %u and pin %u\n",
+		 intsel, cctx->intr_lines[intsel], pin);
+
+	if (chv_pad_locked(pctrl, pin))
+		return -EBUSY;
+
+	/*
+	 * The BIOS fills the interrupt lines from 0 counting up, start at
+	 * the other end to find a free interrupt line to workaround this.
+	 */
+	for (i = community->nirqs - 1; i >= 0; i--) {
+		if (cctx->intr_lines[i] == INTR_LINE_UNUSED)
+			break;
+	}
+	if (i < 0)
+		return -EBUSY;
+
+	dev_info(pctrl->dev, "changing the interrupt line for pin %u to %d\n", pin, i);
+
+	value = (value & ~CHV_PADCTRL0_INTSEL_MASK) | (i << CHV_PADCTRL0_INTSEL_SHIFT);
+	chv_writel(pctrl, pin, CHV_PADCTRL0, value);
+	cctx->intr_lines[i] = pin;
+
+	return 0;
+}
+
 static int chv_gpio_irq_type(struct irq_data *d, unsigned int type)
 {
 	struct gpio_chip *gc = irq_data_get_irq_chip_data(d);
 	struct intel_pinctrl *pctrl = gpiochip_get_data(gc);
-	struct intel_community_context *cctx = &pctrl->context.communities[0];
 	unsigned int pin = irqd_to_hwirq(d);
 	unsigned long flags;
 	u32 value;
+	int ret;
 
 	raw_spin_lock_irqsave(&chv_lock, flags);
 
+	ret = chv_gpio_set_intr_line(pctrl, pin);
+	if (ret)
+		goto out_unlock;
+
 	/*
 	 * Pins which can be used as shared interrupt are configured in
 	 * BIOS. Driver trusts BIOS configurations and assigns different
@@ -1377,20 +1435,15 @@  static int chv_gpio_irq_type(struct irq_data *d, unsigned int type)
 		chv_writel(pctrl, pin, CHV_PADCTRL1, value);
 	}
 
-	value = chv_readl(pctrl, pin, CHV_PADCTRL0);
-	value &= CHV_PADCTRL0_INTSEL_MASK;
-	value >>= CHV_PADCTRL0_INTSEL_SHIFT;
-
-	cctx->intr_lines[value] = pin;
-
 	if (type & IRQ_TYPE_EDGE_BOTH)
 		irq_set_handler_locked(d, handle_edge_irq);
 	else if (type & IRQ_TYPE_LEVEL_MASK)
 		irq_set_handler_locked(d, handle_level_irq);
 
+out_unlock:
 	raw_spin_unlock_irqrestore(&chv_lock, flags);
 
-	return 0;
+	return ret;
 }
 
 static void chv_gpio_irq_handler(struct irq_desc *desc)