[PATCH] irqchip/stm32-exti: avoid interrupts losing due to clearing pending bit by mistake

[Marc Zyngier]

On 2020-08-19 03:39, qiuguorui1 wrote:
> In the previous code, when the eoi handle of the exti clears the 
> pending
> bit of the current interrupt, it will first read the values of fpr and
> rpr, then logically OR the corresponding bit of the interrupt number,
> and finally write back to fpr and rpr.
> 
> We found through experiments that if two exti interrupts,
> we call them int1/int2, arrive almost at the same time. in our 
> scenario,
> the time difference is 30 microseconds, assuming int1 is triggered 
> first.
> 
> there will be an extreme scenario: both int's pending bit are set to 1,
> the irq handle of int1 is executed first, and eoi handle is then 
> executed,
> at this moment, all pending bits are cleared, but the int 2 has not
> finally been reported to the cpu yet, which eventually lost int2.
> 
> According to stm32's TRM description about rpr and fpr: Writing a 1 to 
> this
> bit will trigger a rising edge event on event x, Writing 0 has no
> effect.
> 
> Therefore, when clearing the pending bit, we only need to clear the
> pending bit of the irq.

Interesting findings!

> 
> Signed-off-by: qiuguorui1 <qiuguorui1 at huawei.com>

This definitely needs a Fixes: tag and a Cc: stable, as lost
interrupts are not fun at all.

> ---
>  drivers/irqchip/irq-stm32-exti.c | 18 ++++++++++++------
>  1 file changed, 12 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/irqchip/irq-stm32-exti.c 
> b/drivers/irqchip/irq-stm32-exti.c
> index 03a36be757d8..ee4faf5c90b8 100644
> --- a/drivers/irqchip/irq-stm32-exti.c
> +++ b/drivers/irqchip/irq-stm32-exti.c
> @@ -26,6 +26,11 @@
> 
>  #define HWSPNLCK_TIMEOUT	1000 /* usec */
> 
> +enum reg_ops {
> +	REG_WRITE_ONLY,
> +	REG_READ_WRITE
> +};
> +
>  struct stm32_exti_bank {
>  	u32 imr_ofst;
>  	u32 emr_ofst;
> @@ -416,13 +421,14 @@ static void stm32_irq_ack(struct irq_data *d)
>  	irq_gc_unlock(gc);
>  }
> 
> -static inline u32 stm32_exti_set_bit(struct irq_data *d, u32 reg)
> +static inline u32 stm32_exti_set_bit(struct irq_data *d, u32 reg,
> enum reg_ops op)
>  {
>  	struct stm32_exti_chip_data *chip_data = 
> irq_data_get_irq_chip_data(d);
>  	void __iomem *base = chip_data->host_data->base;
> -	u32 val;
> +	u32 val = 0;
> 
> -	val = readl_relaxed(base + reg);
> +	if (op == REG_READ_WRITE)
> +		val = readl_relaxed(base + reg);
>  	val |= BIT(d->hwirq % IRQS_PER_BANK);
>  	writel_relaxed(val, base + reg);
> 
> @@ -449,9 +455,9 @@ static void stm32_exti_h_eoi(struct irq_data *d)
> 
>  	raw_spin_lock(&chip_data->rlock);
> 
> -	stm32_exti_set_bit(d, stm32_bank->rpr_ofst);
> +	stm32_exti_set_bit(d, stm32_bank->rpr_ofst, REG_WRITE_ONLY);
>  	if (stm32_bank->fpr_ofst != UNDEF_REG)
> -		stm32_exti_set_bit(d, stm32_bank->fpr_ofst);
> +		stm32_exti_set_bit(d, stm32_bank->fpr_ofst, REG_WRITE_ONLY);
> 
>  	raw_spin_unlock(&chip_data->rlock);
> 
> @@ -478,7 +484,7 @@ static void stm32_exti_h_unmask(struct irq_data *d)
>  	const struct stm32_exti_bank *stm32_bank = chip_data->reg_bank;
> 
>  	raw_spin_lock(&chip_data->rlock);
> -	chip_data->mask_cache = stm32_exti_set_bit(d, stm32_bank->imr_ofst);
> +	chip_data->mask_cache = stm32_exti_set_bit(d, stm32_bank->imr_ofst,
> REG_READ_WRITE);
>  	raw_spin_unlock(&chip_data->rlock);
> 
>  	if (d->parent_data->chip)

I think this could be made much simpler by simply providing
an accessor that doesn't do a RMW. Something like this (untested):

diff --git a/drivers/irqchip/irq-stm32-exti.c 
b/drivers/irqchip/irq-stm32-exti.c
index 03a36be757d8..e35c5561a10d 100644
--- a/drivers/irqchip/irq-stm32-exti.c
+++ b/drivers/irqchip/irq-stm32-exti.c
@@ -416,6 +416,14 @@ static void stm32_irq_ack(struct irq_data *d)
  	irq_gc_unlock(gc);
  }

+static void stm32_exti_write_bit(struct irq_data *d, u32 reg)
+{
+	struct stm32_exti_chip_data *chip_data = 
irq_data_get_irq_chip_data(d);
+	void __iomem *base = chip_data->host_data->base;
+
+	writel_relaxed(BIT(d->hwirq % IRQS_PER_BANK), base + reg);
+}
+
  static inline u32 stm32_exti_set_bit(struct irq_data *d, u32 reg)
  {
  	struct stm32_exti_chip_data *chip_data = 
irq_data_get_irq_chip_data(d);
@@ -449,9 +457,9 @@ static void stm32_exti_h_eoi(struct irq_data *d)

  	raw_spin_lock(&chip_data->rlock);

-	stm32_exti_set_bit(d, stm32_bank->rpr_ofst);
+	stm32_exti_write_bit(d, stm32_bank->rpr_ofst);
  	if (stm32_bank->fpr_ofst != UNDEF_REG)
-		stm32_exti_set_bit(d, stm32_bank->fpr_ofst);
+		stm32_exti_write_bit(d, stm32_bank->fpr_ofst);

  	raw_spin_unlock(&chip_data->rlock);

Thanks,

         M.
-- 
Jazz is not dead. It just smells funny...