Skip to contentSkip to conversation

pinctrl: qcom: Clear status bit on irq_unmask

PATCH In review Tengfei Fan 4 participants 8 messages May 31, 2019 – Jun 17, 2019 1 message not in the archive

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 pinctrl: qcom: Clear status bit on irq_unmask

The gpio interrupt status bit is getting set after the irq is disabled and causing an immediate interrupt after enablling the irq, so clear status bit on irq_unmask.

---
 drivers/pinctrl/qcom/pinctrl-msm.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/pinctrl/qcom/pinctrl-msm.c +7 -0
index ee81198..7283c50 100644
@@ -740,6 +740,7 @@ static void msm_gpio_irq_mask(struct irq_data *d)
 static void msm_gpio_irq_unmask(struct irq_data *d)
 {
 	struct gpio_chip *gc = irq_data_get_irq_chip_data(d);
+	uint32_t irqtype = irqd_get_trigger_type(d);
 	struct msm_pinctrl *pctrl = gpiochip_get_data(gc);
 	const struct msm_pingroup *g;
 	unsigned long flags;
@@ -749,6 +750,12 @@ static void msm_gpio_irq_unmask(struct irq_data *d)
 
 	raw_spin_lock_irqsave(&pctrl->lock, flags);
 
+	if (irqtype & (IRQF_TRIGGER_HIGH | IRQF_TRIGGER_LOW)) {
+		val = readl_relaxed(pctrl->regs + g->intr_status_reg);
+		val &= ~BIT(g->intr_status_bit);
+		 writel_relaxed(val, pctrl->regs + g->intr_status_reg);
+	}
+
 	val = msm_readl_intr_cfg(pctrl, g);
 	val |= BIT(g->intr_raw_status_bit);
 	val |= BIT(g->intr_enable_bit);
Signature · 2 lines
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
#1 lore raw
Quoted 5 lines · Tengfei Fan

The gpio interrupt status bit is getting set after the irq is disabled and causing an immediate interrupt after enablling the irq, so clear status bit on irq_unmask.

This looks pretty serious, can one of the Qcom maintainers ACK this?

Should it be sent to fixes and even stable?

Fixes tag?

Yours,
Linus Walleij

#2 lore raw

Quoting Linus Walleij (2019-06-07 14:08:10)

Quoted 6 lines · Tengfei Fan
Quoted 5 lines

The gpio interrupt status bit is getting set after the irq is disabled and causing an immediate interrupt after enablling the irq, so clear status bit on irq_unmask.

Quoted 9 lines
Quoted 1 line

This looks pretty serious, can one of the Qcom maintainers ACK this?

Should it be sent to fixes and even stable?

Fixes tag?

How is the interrupt status bit getting set after the irq is disabled? It looks like this is a level type interrupt? I thought that after commit b55326dc969e ("pinctrl: msm: Really mask level interrupts to prevent latching") this wouldn't be a problem. Am I wrong, or is qcom just clearing out patches on drivers and this is the last one that needs to be upstreamed?

#3 lore raw
Quoted 17 lines · Stephen Boyd

Quoting Linus Walleij (2019-06-07 14:08:10)

Quoted 6 lines · Tengfei Fan
Quoted 5 lines

The gpio interrupt status bit is getting set after the irq is disabled and causing an immediate interrupt after enablling the irq, so clear status bit on irq_unmask.

Quoted 9 lines
Quoted 1 line

This looks pretty serious, can one of the Qcom maintainers ACK this?

Should it be sent to fixes and even stable?

Fixes tag?
Quoted 6 lines

How is the interrupt status bit getting set after the irq is disabled? It looks like this is a level type interrupt? I thought that after commit b55326dc969e ("pinctrl: msm: Really mask level interrupts to prevent latching") this wouldn't be a problem. Am I wrong, or is qcom just clearing out patches on drivers and this is the last one that needs to be upstreamed?

Your patch(commit b55326dc969e) can cover our issue, and my patch is no longer needed. Your patch isn't included in our code, so I submitted this patch.

#4 lore raw

Quoting [email protected] (2019-06-11 03:41:26)

Quoted 27 lines · Stephen Boyd
Quoted 18 lines

Quoting Linus Walleij (2019-06-07 14:08:10)

Quoted 7 lines
Quoted 4 lines · Tengfei Fan

The gpio interrupt status bit is getting set after the irq is disabled and causing an immediate interrupt after enablling the irq, so clear status bit on irq_unmask.

Quoted 9 lines
Quoted 1 line

This looks pretty serious, can one of the Qcom maintainers ACK this?

Should it be sent to fixes and even stable?

Fixes tag?
Quoted 7 lines

How is the interrupt status bit getting set after the irq is disabled? It looks like this is a level type interrupt? I thought that after commit b55326dc969e ("pinctrl: msm: Really mask level interrupts to prevent latching") this wouldn't be a problem. Am I wrong, or is qcom just clearing out patches on drivers and this is the last one that needs
to be upstreamed?

Quoted 3 lines

Your patch(commit b55326dc969e) can cover our issue, and my patch is no longer needed.
Your patch isn't included in our code, so I submitted this patch.

Alright cool. Sounds like this patch can be dropped then and you can pick up the patch from upstream into your vendor kernel.

A message that is not in the archive (sent off-list or not yet received): [email protected]

#6 lore raw
Quoted 23 lines

Quoting [email protected] (2019-06-11 03:41:26)

Quoted 21 lines · Stephen Boyd
Quoted 2 lines

Quoting Linus Walleij (2019-06-07 14:08:10)

Quoted 1 line

On Fri, May 31, 2019 at 8:52 AM Tengfei Fan

Quoted 10 lines · <[email protected]> >> wrote:
Quoted 10 lines
Quoted 6 lines
Quoted 1 line
Quoted 1 line

The gpio interrupt status bit is getting set after the

irq is disabled and causing an immediate interrupt after enablling the irq, so clear status bit on irq_unmask.

Quoted 1 line

This looks pretty serious, can one of the Qcom maintainers ACK

this?

Quoted 3 lines
Quoted 3 lines

Should it be sent to fixes and even stable?
Fixes: tag?
How is the interrupt status bit getting set after the irq is

disabled?

Quoted 5 lines

It looks like this is a level type interrupt? I thought that after commit b55326dc969e ("pinctrl: msm: Really mask level interrupts to prevent latching") this wouldn't be a problem. Am I wrong, or is qcom just clearing out patches on drivers and this is the last one that needs to be upstreamed?

Quoted 6 lines
Quoted 2 lines

Your patch(commit b55326dc969e) can cover our issue, and my patch is no longer needed. Your patch isn't included in our code, so I submitted this patch.

Alright cool. Sounds like this patch can be dropped then and you can pick up the patch from upstream into your vendor kernel.

Hi Stephen, there is one use case with is not covered by commit b55326dc969e (

"pinctrl: msm: Really mask level interrupts to prevent latching"). That happens when

gpio line is toggled between i/o mode and interrupt mode :

1. GPIO is configured as irq line. Peripheral raises interrupt.

2. IRQ handler runs and disables the irq line (through wq work).

3. GPIO is configured for input and and data is received from the peripheral.

4. Now, when GPIO is re-enabled as irq, we see spurious irq, and there isn't

any data received on the gpio line, when it is read back after configuring as input.

This can happen for both edge and level interrupts.

Patch https://lkml.org/lkml/2019/6/17/226 tries to cover this use case. Can you please

provide your comments?

Thanks

Neeraj

Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
member of the Code Aurora Forum, hosted by The Linux Foundation
#7 lore raw
Quoted 6 lines · Neeraj Upadhyay

Hi Stephen, there is one use case with is not covered by commit b55326dc969e (

"pinctrl: msm: Really mask level interrupts to prevent latching"). That happens when

Quoted 8 lines

gpio line is toggled between i/o mode and interrupt mode :

1. GPIO is configured as irq line. Peripheral raises interrupt.

2. IRQ handler runs and disables the irq line (through wq work).

3. GPIO is configured for input and and data is received from the peripheral.

There is no distinction between using a GPIO line as input and using it for IRQ. All input GPIOs can be used for IRQs, if the hardware supports it (has an irqchip).

Quoted 5 lines

4. Now, when GPIO is re-enabled as irq, we see spurious irq, and there isn't

any data received on the gpio line, when it is read back after configuring as input.

That's an interesting usecase. Hans Verkuil reworked the GPIO irq support very elegantly exactly to support this type of usecase (irq switch on and off dynamically), where he was even switching the line into output mode between the IRQ trains. (one-wire transcactions for CEC).

Quoted 2 lines

Patch https://lkml.org/lkml/2019/6/17/226 tries to cover this use case. Can you please provide your comments?

What this patch does is clear all pending IRQs at irq unmask. This is usually safe, unless there may be cases where you *want* to catch any pending IRQs. I guess normally you don't so it should be safe?

The corner case is when you start some transaction or whatever that gets ACKed by an IRQ and you actually get the IRQ back before you had time to execute the code enabling the IRQ.

That would be racy and bad code, as you should clearly enable the IRQ first, then start the transaction. So I think this patch is safe.

But let's see what Bjorn says.

Yours,
Linus Walleij

#8 lore raw

Thanks for the review, Linus.

Quoted 12 lines · Linus Walleij
Quoted 10 lines · Neeraj Upadhyay

Hi Stephen, there is one use case with is not covered by commit b55326dc969e (

"pinctrl: msm: Really mask level interrupts to prevent latching"). That happens when

gpio line is toggled between i/o mode and interrupt mode :

1. GPIO is configured as irq line. Peripheral raises interrupt.

Quoted 7 lines
Quoted 4 lines

2. IRQ handler runs and disables the irq line (through wq work).

3. GPIO is configured for input and and data is received from the peripheral.

There is no distinction between using a GPIO line as input and using it for IRQ. All input GPIOs can be used for IRQs, if the hardware supports it (has an irqchip).

Ok

Quoted 12 lines
Quoted 6 lines

4. Now, when GPIO is re-enabled as irq, we see spurious irq, and there isn't

any data received on the gpio line, when it is read back after configuring as input.

That's an interesting usecase. Hans Verkuil reworked the GPIO irq support very elegantly exactly to support this type of usecase (irq switch on and off dynamically), where he was even switching the line into output mode between the IRQ trains. (one-wire transcactions for CEC).

Quoted 6 lines
Quoted 2 lines

Patch https://lkml.org/lkml/2019/6/17/226 tries to cover this use case. Can you please provide your comments?

What this patch does is clear all pending IRQs at irq unmask. This is usually safe, unless there may be cases where you *want* to catch any pending IRQs. I guess normally you don't so it should be safe?

We want to clear pending status at irq enable. Afaik, no,

we don't normally track these pending irqs. So, should be

fine here.

Quoted 14 lines

The corner case is when you start some transaction or whatever that gets ACKed by an IRQ and you actually get the IRQ back before you had time to execute the code enabling the IRQ.

That would be racy and bad code, as you should clearly enable the IRQ first, then start the transaction. So I think this patch is safe.

But let's see what Bjorn says.

Yours,
Linus Walleij

Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
member of the Code Aurora Forum, hosted by The Linux Foundation