[PATCH v2 3/4] irqchip: Add BCM2835 AUX interrupt controller

[Marc Zyngier]

On 12/06/17 16:21, Phil Elwell wrote:
> On 12/06/2017 15:59, Marc Zyngier wrote:> On 12/06/17 15:25, Phil Elwell wrote:
>>> Devices in the BCM2835 AUX block share a common interrupt line, with a
>>> register indicating which devices have active IRQs. Expose this as a
>>> nested interrupt controller to avoid IRQ sharing problems (easily
>>> observed if UART1 and SPI1/2 are enabled simultaneously).
>>>
>>> Signed-off-by: Phil Elwell <phil at raspberrypi.org>
>>> ---
>>>  drivers/irqchip/Makefile          |   2 +-
>>>  drivers/irqchip/irq-bcm2835-aux.c | 155 ++++++++++++++++++++++++++++++++++++++
>>>  2 files changed, 156 insertions(+), 1 deletion(-)
>>>  create mode 100644 drivers/irqchip/irq-bcm2835-aux.c
>>>
>>> diff --git a/drivers/irqchip/Makefile b/drivers/irqchip/Makefile
>>> index b64c59b..cf01920 100644
>>> --- a/drivers/irqchip/Makefile
>>> +++ b/drivers/irqchip/Makefile
>>> @@ -3,7 +3,7 @@ obj-$(CONFIG_IRQCHIP)			+= irqchip.o
>>>  obj-$(CONFIG_ALPINE_MSI)		+= irq-alpine-msi.o
>>>  obj-$(CONFIG_ATH79)			+= irq-ath79-cpu.o
>>>  obj-$(CONFIG_ATH79)			+= irq-ath79-misc.o
>>> -obj-$(CONFIG_ARCH_BCM2835)		+= irq-bcm2835.o
>>> +obj-$(CONFIG_ARCH_BCM2835)		+= irq-bcm2835.o irq-bcm2835-aux.o
>>>  obj-$(CONFIG_ARCH_BCM2835)		+= irq-bcm2836.o
>>>  obj-$(CONFIG_ARCH_EXYNOS)		+= exynos-combiner.o
>>>  obj-$(CONFIG_FARADAY_FTINTC010)		+= irq-ftintc010.o
>>> diff --git a/drivers/irqchip/irq-bcm2835-aux.c b/drivers/irqchip/irq-bcm2835-aux.c
>>> new file mode 100644
>>> index 0000000..545f12e
>>> --- /dev/null
>>> +++ b/drivers/irqchip/irq-bcm2835-aux.c
>>> @@ -0,0 +1,155 @@
>>> +/*
>>> + * Copyright (C) 2017 Raspberry Pi (Trading) Ltd.
>>> + *
>>> + * This program is free software; you can redistribute it and/or modify it
>>> + * under the terms and conditions of the GNU General Public License,
>>> + * version 2, as published by the Free Software Foundation.
>>> + *
>>> + * This program is distributed in the hope it will be useful, but WITHOUT
>>> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
>>> + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
>>> + * more details.
>>> + *
>>> + * You should have received a copy of the GNU General Public License
>>> + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
>>> + */
>>> +
>>> +#include <linux/interrupt.h>
>>> +#include <linux/irqdomain.h>
>>> +#include <linux/module.h>
>>> +#include <linux/of_irq.h>
>>> +#include <linux/platform_device.h>
>>> +#include <dt-bindings/interrupt-controller/bcm2835-aux-intc.h>
>>> +
>>> +#define BCM2835_AUXIRQ		0x00
>>> +
>>> +#define BCM2835_AUX_IRQ_UART_MASK BIT(BCM2835_AUX_IRQ_UART)
>>> +#define BCM2835_AUX_IRQ_SPI1_MASK BIT(BCM2835_AUX_IRQ_SPI1)
>>> +#define BCM2835_AUX_IRQ_SPI2_MASK BIT(BCM2835_AUX_IRQ_SPI2)
>>> +
>>> +#define BCM2835_AUX_IRQ_ALL_MASK \
>>> +	(BCM2835_AUX_IRQ_UART_MASK | \
>>> +	 BCM2835_AUX_IRQ_SPI1_MASK | \
>>> +	 BCM2835_AUX_IRQ_SPI2_MASK)
>>> +
>>> +struct aux_irq_state {
>>> +	void __iomem      *status;
>>> +	struct irq_domain *domain;
>>> +};
>>> +
>>> +static struct aux_irq_state aux_irq __read_mostly;
>>> +
>>> +static irqreturn_t bcm2835_aux_irq_handler(int irq, void *dev_id)
>>> +{
>>> +	u32 stat = readl_relaxed(aux_irq.status);
>>> +
>>> +	if (stat & BCM2835_AUX_IRQ_UART_MASK)
>>> +		generic_handle_irq(irq_linear_revmap(aux_irq.domain,
>>> +						     BCM2835_AUX_IRQ_UART));
>>> +
>>> +	if (stat & BCM2835_AUX_IRQ_SPI1_MASK)
>>> +		generic_handle_irq(irq_linear_revmap(aux_irq.domain,
>>> +						     BCM2835_AUX_IRQ_SPI1));
>>> +
>>> +	if (stat & BCM2835_AUX_IRQ_SPI2_MASK)
>>> +		generic_handle_irq(irq_linear_revmap(aux_irq.domain,
>>> +						     BCM2835_AUX_IRQ_SPI2));
>>> +
>>> +	return (stat & BCM2835_AUX_IRQ_ALL_MASK) ? IRQ_HANDLED : IRQ_NONE;
>>
>> I could understand the use of a normal interrupt handler instead of a
>> chained handler, as the HW doesn't have any way of masking interrupts
>> (whoever designed this should be forced to fix each and every SoC with a
>> magnifier and a tiny drill) if it wasn't for this last line.
>>
>> Here, you're making sure that you always return IRQ_HANDLED if something
>> was pending, irrespective of whether it has been handled or not. How do
>> you recover when you have a screaming interrupt and no handler?
> 
> Does Linux not notice when one calls generic_handle_irq with the number of an
> interrupt without a handler?

It is not so much that the interrupt doesn't have a handler, but that
the device (or one of the devices) is in some sort of interrupt frenzy,
and the driver is not able to handle this interrupt.

In such a case, Linux tries to mask this interrupt, which in your case
does exactly nothing. At this point, the system is dead.

> 
>> Why don't you simply request the interrupt as a shared one, and check
>> for the state in the handlers themselves? This way, the kernel will be
>> able to recover from a screaming interrupt by disabling it.
> 
> I'm not sure quite how the problem arises - the AUX SPI driver uses IRQF_SHARED,
> and the AUX UART (8250 clone) driver sets UPF_SHARE_IRQ, but the end result
> is a lockup. Putting checking of the parent status bits into the drivers (one of
> which is a fairly generic 8250 driver) seems wrong.

Well, all the 8250 variants have some glue of some sort... And you
definitely should investigate what the issue is with this lock-up. You
don't even have to read this status register, BTH. The kernel will
happily iterate over the handlers for you.

> Adding this simple driver fixed the problem, and I think it better reflects the
> hardware modularity.

It'd certainly be better to investigate the actual source of the problem.

>>
>>> +}
>>> +
>>> +static int bcm2835_aux_irq_xlate(struct irq_domain *d,
>>> +				 struct device_node *ctrlr,
>>> +				 const u32 *intspec, unsigned int intsize,
>>> +				 unsigned long *out_hwirq,
>>> +				 unsigned int *out_type)
>>> +{
>>> +	if (WARN_ON(intsize != 1))
>>> +		return -EINVAL;
>>> +
>>> +	if (WARN_ON(intspec[0] >= BCM2835_AUX_IRQ_COUNT))
>>> +		return -EINVAL;
>>> +
>>> +	*out_hwirq = intspec[0];
>>> +	*out_type = IRQ_TYPE_NONE;

By the way, what is this IRQ_TYPE_NONE here? From what I can read, it
can only be IRQ_TYPE_LEVEL...

>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +/*
>>> + * The irq_mask and irq_unmask function pointers are used without
>>> + * validity checks, so they must not be NULL. Create a dummy function
>>> + * with the expected type for use as a no-op.
>>> + */
>>> +static void bcm2835_aux_irq_dummy(struct irq_data *data)
>>> +{
>>> +}
>>> +
>>> +static struct irq_chip bcm2835_aux_irq_chip = {
>>> +	.name = "bcm2835-aux_irq",
>>> +	.irq_mask = bcm2835_aux_irq_dummy,
>>> +	.irq_unmask = bcm2835_aux_irq_dummy,
>>> +};
>>> +
>>> +static const struct irq_domain_ops bcm2835_aux_irq_ops = {
>>> +	.xlate = bcm2835_aux_irq_xlate
>>> +};
>>> +
>>> +static int bcm2835_aux_irq_probe(struct platform_device *pdev)
>>> +{
>>> +	struct device *dev = &pdev->dev;
>>> +	struct device_node *node = dev->of_node;
>>> +	int parent_irq;
>>> +	struct resource *res;
>>> +	void __iomem *reg;
>>> +	int i;
>>> +
>>> +	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
>>> +	reg = devm_ioremap_resource(dev, res);
>>> +	if (IS_ERR(reg))
>>> +		return PTR_ERR(reg);
>>> +
>>> +	parent_irq = irq_of_parse_and_map(node, 0);
>>> +	if (!parent_irq)
>>> +		return -ENXIO;
>>> +
>>> +	aux_irq.status = reg + BCM2835_AUXIRQ;
>>> +	aux_irq.domain = irq_domain_add_linear(node,
>>> +					       BCM2835_AUX_IRQ_COUNT,
>>> +					       &bcm2835_aux_irq_ops,
>>> +					       NULL);
>>> +	if (!aux_irq.domain)
>>> +		return -ENXIO;
>>> +
>>> +	for (i = 0; i < BCM2835_AUX_IRQ_COUNT; i++) {
>>> +		unsigned int irq = irq_create_mapping(aux_irq.domain, i);
>>> +
>>> +		if (irq == 0)
>>> +			return -ENXIO;
>>> +
>>> +		irq_set_chip_and_handler(irq, &bcm2835_aux_irq_chip,
>>> +					 handle_level_irq);
>>> +	}
>>
>> My initial question notwithstanding, why do you need any of this? This
>> should be done at map time, and the irq_create_mapping() call should
>> entirely be driven from DT.
> 
> Can you explain this in more detail? I'm open to a better solution.

irq_create mapping is (indirectly) called by the core code when parsing
the DT (of_platform_populate), so it is fairly pointless here. As for
the irq_set_*() call, it should be in a .map callback on the irqdomain
ops, so that it is configured on a per-irq basis (there is plenty of
existing code in the drivers/irqchip for you to have a look).

In general, we don't instantiate interrupts in the irqchip itself. It is
the core code duty to do so.

Thanks,

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