On Wed, 24 Feb 2021 07:56:40 +0000,
Álvaro Fernández Rojas <noltari@xxxxxxxxx> wrote:
>
> This interrupt controller is present on bcm63xx SoCs in order to generate
> interrupts based on GPIO status changes.
>
> Signed-off-by: Álvaro Fernández Rojas <noltari@xxxxxxxxx>
> Signed-off-by: Jonas Gorski <jonas.gorski@xxxxxxxxx>
> Reviewed-by: Florian Fainelli <f.fainelli@xxxxxxxxx>
> ---
> v3: no changes.
> v2: no changes.
>
> drivers/irqchip/Kconfig | 4 +
> drivers/irqchip/Makefile | 1 +
> drivers/irqchip/irq-bcm6345-ext.c | 271 ++++++++++++++++++++++++++++++
> 3 files changed, 276 insertions(+)
> create mode 100644 drivers/irqchip/irq-bcm6345-ext.c
>
> diff --git a/drivers/irqchip/Kconfig b/drivers/irqchip/Kconfig
> index e74fa206240a..eaa101939a34 100644
> --- a/drivers/irqchip/Kconfig
> +++ b/drivers/irqchip/Kconfig
> @@ -113,6 +113,10 @@ config I8259
> bool
> select IRQ_DOMAIN
>
> +config BCM6345_EXT_IRQ
> + bool "BCM6345 External IRQ Controller"
> + select IRQ_DOMAIN
> +
> config BCM6345_L1_IRQ
> bool
> select GENERIC_IRQ_CHIP
> diff --git a/drivers/irqchip/Makefile b/drivers/irqchip/Makefile
> index c59b95a0532c..3cba65bc0aa5 100644
> --- a/drivers/irqchip/Makefile
> +++ b/drivers/irqchip/Makefile
> @@ -62,6 +62,7 @@ obj-$(CONFIG_XTENSA_MX) += irq-xtensa-mx.o
> obj-$(CONFIG_XILINX_INTC) += irq-xilinx-intc.o
> obj-$(CONFIG_IRQ_CROSSBAR) += irq-crossbar.o
> obj-$(CONFIG_SOC_VF610) += irq-vf610-mscm-ir.o
> +obj-$(CONFIG_BCM6345_EXT_IRQ) += irq-bcm6345-ext.o
> obj-$(CONFIG_BCM6345_L1_IRQ) += irq-bcm6345-l1.o
> obj-$(CONFIG_BCM7038_L1_IRQ) += irq-bcm7038-l1.o
> obj-$(CONFIG_BCM7120_L2_IRQ) += irq-bcm7120-l2.o
> diff --git a/drivers/irqchip/irq-bcm6345-ext.c b/drivers/irqchip/irq-bcm6345-ext.c
> new file mode 100644
> index 000000000000..5721ac8de295
> --- /dev/null
> +++ b/drivers/irqchip/irq-bcm6345-ext.c
> @@ -0,0 +1,271 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * Broadcom BCM6345 style external interrupt controller driver
> + *
> + * Copyright (C) 2021 Álvaro Fernández Rojas <noltari@xxxxxxxxx>
> + * Copyright (C) 2014 Jonas Gorski <jonas.gorski@xxxxxxxxx>
> + */
> +
> +#include <linux/ioport.h>
> +#include <linux/irq.h>
> +#include <linux/irqchip.h>
> +#include <linux/irqchip/chained_irq.h>
> +#include <linux/kernel.h>
> +#include <linux/of.h>
> +#include <linux/of_irq.h>
> +#include <linux/of_address.h>
> +#include <linux/slab.h>
> +#include <linux/spinlock.h>
> +
> +#define MAX_IRQS 4
> +
> +#define EXTIRQ_CFG_SENSE 0
> +#define EXTIRQ_CFG_STAT 1
> +#define EXTIRQ_CFG_CLEAR 2
> +#define EXTIRQ_CFG_MASK 3
> +#define EXTIRQ_CFG_BOTHEDGE 4
> +#define EXTIRQ_CFG_LEVELSENSE 5
> +
> +struct intc_data {
> + struct irq_chip chip;
> + struct irq_domain *domain;
> + raw_spinlock_t lock;
> +
> + int parent_irq[MAX_IRQS];
> + void __iomem *reg;
> + int shift;
> + unsigned int toggle_clear_on_ack:1;
Please use the bool type.
> +};
> +
> +static void bcm6345_ext_intc_irq_handle(struct irq_desc *desc)
> +{
> + struct intc_data *data = irq_desc_get_handler_data(desc);
> + struct irq_chip *chip = irq_desc_get_chip(desc);
> + unsigned int irq = irq_desc_get_irq(desc);
> + unsigned int idx;
> +
> + chained_irq_enter(chip, desc);
> +
> + for (idx = 0; idx < MAX_IRQS; idx++) {
> + if (data->parent_irq[idx] != irq)
> + continue;
> +
> + generic_handle_irq(irq_find_mapping(data->domain, idx));
One parent IRQ per input? Why isn't this a hierarchical interrupt
controller? Even *if* this really has to be a chained interrupt
controller, I'm sure there are better ways to identify the input then
this loop (offset from a base, for example).
> + }
> +
> + chained_irq_exit(chip, desc);
> +}
> +
> +static void bcm6345_ext_intc_irq_ack(struct irq_data *data)
> +{
> + struct intc_data *priv = data->domain->host_data;
> + irq_hw_number_t hwirq = irqd_to_hwirq(data);
> + u32 reg;
> +
> + raw_spin_lock(&priv->lock);
> + reg = __raw_readl(priv->reg);
> + __raw_writel(reg | (1 << (hwirq + EXTIRQ_CFG_CLEAR * priv->shift)),
> + priv->reg);
> + if (priv->toggle_clear_on_ack)
Under what condition do you need this?
> + __raw_writel(reg, priv->reg);
> + raw_spin_unlock(&priv->lock);
> +}
> +
> +static void bcm6345_ext_intc_irq_mask(struct irq_data *data)
> +{
> + struct intc_data *priv = data->domain->host_data;
> + irq_hw_number_t hwirq = irqd_to_hwirq(data);
> + u32 reg;
> +
> + raw_spin_lock(&priv->lock);
> + reg = __raw_readl(priv->reg);
> + reg &= ~(1 << (hwirq + EXTIRQ_CFG_MASK * priv->shift));
> + __raw_writel(reg, priv->reg);
> + raw_spin_unlock(&priv->lock);
> +}
> +
> +static void bcm6345_ext_intc_irq_unmask(struct irq_data *data)
> +{
> + struct intc_data *priv = data->domain->host_data;
> + irq_hw_number_t hwirq = irqd_to_hwirq(data);
> + u32 reg;
> +
> + raw_spin_lock(&priv->lock);
> + reg = __raw_readl(priv->reg);
> + reg |= 1 << (hwirq + EXTIRQ_CFG_MASK * priv->shift);
> + __raw_writel(reg, priv->reg);
> + raw_spin_unlock(&priv->lock);
> +}
> +
> +static int bcm6345_ext_intc_set_type(struct irq_data *data,
> + unsigned int flow_type)
> +{
> + struct intc_data *priv = data->domain->host_data;
> + irq_hw_number_t hwirq = irqd_to_hwirq(data);
> + bool levelsense = 0, sense = 0, bothedge = 0;
> + u32 reg;
> +
> + flow_type &= IRQ_TYPE_SENSE_MASK;
> +
> + if (flow_type == IRQ_TYPE_NONE)
You should never get NONE. If you have that value here, that's
probably a bug somewhere else.
> + flow_type = IRQ_TYPE_LEVEL_LOW;
> +
> + switch (flow_type) {
> + case IRQ_TYPE_EDGE_BOTH:
> + bothedge = 1;
> + break;
> +
> + case IRQ_TYPE_EDGE_RISING:
> + sense = 1;
> + break;
> +
> + case IRQ_TYPE_EDGE_FALLING:
> + break;
> +
> + case IRQ_TYPE_LEVEL_HIGH:
> + levelsense = 1;
> + sense = 1;
> + break;
> +
> + case IRQ_TYPE_LEVEL_LOW:
> + levelsense = 1;
> + break;
> +
> + default:
> + pr_err("bogus flow type combination given!\n");
> + return -EINVAL;
How can this happen?
> + }
> +
> + raw_spin_lock(&priv->lock);
> + reg = __raw_readl(priv->reg);
> +
> + if (levelsense)
> + reg |= 1 << (hwirq + EXTIRQ_CFG_LEVELSENSE * priv->shift);
> + else
> + reg &= ~(1 << (hwirq + EXTIRQ_CFG_LEVELSENSE * priv->shift));
> + if (sense)
> + reg |= 1 << (hwirq + EXTIRQ_CFG_SENSE * priv->shift);
> + else
> + reg &= ~(1 << (hwirq + EXTIRQ_CFG_SENSE * priv->shift));
> + if (bothedge)
> + reg |= 1 << (hwirq + EXTIRQ_CFG_BOTHEDGE * priv->shift);
> + else
> + reg &= ~(1 << (hwirq + EXTIRQ_CFG_BOTHEDGE * priv->shift));
So all these levelsense, sense and bothedge variables are already
contained in flow_type. Why the need to reinvent the wheel?
> +
> + __raw_writel(reg, priv->reg);
> + raw_spin_unlock(&priv->lock);
> +
> + irqd_set_trigger_type(data, flow_type);
> + if (flow_type & (IRQ_TYPE_LEVEL_LOW | IRQ_TYPE_LEVEL_HIGH))
> + irq_set_handler_locked(data, handle_level_irq);
> + else
> + irq_set_handler_locked(data, handle_edge_irq);
> +
> + return 0;
> +}
> +
> +static int bcm6345_ext_intc_map(struct irq_domain *d, unsigned int irq,
> + irq_hw_number_t hw)
> +{
> + struct intc_data *priv = d->host_data;
> +
> + irq_set_chip_and_handler(irq, &priv->chip, handle_level_irq);
> +
> + return 0;
> +}
> +
> +static const struct irq_domain_ops bcm6345_ext_domain_ops = {
> + .xlate = irq_domain_xlate_twocell,
> + .map = bcm6345_ext_intc_map,
> +};
> +
> +static int __init bcm6345_ext_intc_init(struct device_node *node,
> + int num_irqs, int *irqs,
> + void __iomem *reg, int shift,
> + bool toggle_clear_on_ack)
> +{
> + struct intc_data *data;
> + unsigned int i;
> +
> + data = kzalloc(sizeof(*data), GFP_KERNEL);
> + if (!data)
> + return -ENOMEM;
> +
> + raw_spin_lock_init(&data->lock);
> +
> + for (i = 0; i < num_irqs; i++) {
> + data->parent_irq[i] = irqs[i];
> +
> + irq_set_handler_data(irqs[i], data);
> + irq_set_chained_handler(irqs[i], bcm6345_ext_intc_irq_handle);
> + }
> +
> + data->reg = reg;
> + data->shift = shift;
> + data->toggle_clear_on_ack = toggle_clear_on_ack;
> +
> + data->chip.name = "bcm6345-ext-intc";
> + data->chip.irq_ack = bcm6345_ext_intc_irq_ack;
> + data->chip.irq_mask = bcm6345_ext_intc_irq_mask;
> + data->chip.irq_unmask = bcm6345_ext_intc_irq_unmask;
> + data->chip.irq_set_type = bcm6345_ext_intc_set_type;
> +
> + data->domain = irq_domain_add_simple(node, num_irqs, 0,
> + &bcm6345_ext_domain_ops, data);
Consider using irq_domain_add_linear(), given that you don't seem to
care about the first interrupt number.
> + if (!data->domain) {
> + kfree(data);
> + return -ENOMEM;
> + }
At this point, you have registered a bunch of chained handlers with
data structures that you have freed. What could possibly go wrong?
> +
> + return 0;
> +}
> +
> +static int __init bcm6345_ext_intc_of_init(struct device_node *node,
> + struct device_node *parent)
> +{
> + int num_irqs, ret = -EINVAL;
> + unsigned i;
> + void __iomem *base;
> + int irqs[MAX_IRQS] = { 0 };
> + u32 shift;
> + bool toggle_clear_on_ack = false;
> +
> + num_irqs = of_irq_count(node);
> +
> + if (!num_irqs || num_irqs > MAX_IRQS)
> + return -EINVAL;
> +
> + if (of_property_read_u32(node, "brcm,field-width", &shift))
> + shift = 4;
Why this default? Given that it is a new driver without any backward
compatibility requirement, either make the property mandatory or get
rid of it.
> +
> + /* On BCM6318 setting CLEAR seems to continuously mask interrupts */
> + if (of_device_is_compatible(node, "brcm,bcm6318-ext-intc"))
> + toggle_clear_on_ack = true;
Seems? What does the documentation says about this?
> +
> + for (i = 0; i < num_irqs; i++) {
> + irqs[i] = irq_of_parse_and_map(node, i);
> + if (!irqs[i])
> + return -ENOMEM;
> + }
> +
> + base = of_iomap(node, 0);
> + if (!base)
> + return -ENXIO;
> +
> + ret = bcm6345_ext_intc_init(node, num_irqs, irqs, base, shift,
> + toggle_clear_on_ack);
> + if (!ret)
> + return 0;
> +
> + iounmap(base);
> +
> + for (i = 0; i < num_irqs; i++)
> + irq_dispose_mapping(irqs[i]);
> +
> + return ret;
> +}
> +
> +IRQCHIP_DECLARE(bcm6318_ext_intc, "brcm,bcm6318-ext-intc",
> + bcm6345_ext_intc_of_init);
> +IRQCHIP_DECLARE(bcm6345_ext_intc, "brcm,bcm6345-ext-intc",
> + bcm6345_ext_intc_of_init);
> --
> 2.20.1
>
>
Thanks,
M.
--
Without deviation from the norm, progress is not possible.
