Re: [PATCH v4 01/10] irqchip: Add driver for Loongson I/O Local Interrupt Controller

From: Marc Zyngier <maz@kernel.org>
To: Jiaxun Yang <jiaxun.yang@flygoat.com>
Cc: linux-mips@vger.kernel.org, Thomas Gleixner <tglx@linutronix.de>,
	Jason Cooper <jason@lakedaemon.net>,
	Rob Herring <robh+dt@kernel.org>,
	Mark Rutland <mark.rutland@arm.com>,
	Ralf Baechle <ralf@linux-mips.org>,
	Paul Burton <paulburton@kernel.org>,
	Huacai Chen <chenhc@lemote.com>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Allison Randal <allison@lohutok.net>,
	linux-kernel@vger.kernel.org, devicetree@vger.kernel.org
Subject: Re: [PATCH v4 01/10] irqchip: Add driver for Loongson I/O Local Interrupt Controller
Date: Sat, 07 Mar 2020 14:50:03 +0000	[thread overview]
Message-ID: <864kv06wys.wl-maz@kernel.org> (raw)
In-Reply-To: <20200221050942.507775-2-jiaxun.yang@flygoat.com>

On Fri, 21 Feb 2020 05:09:16 +0000,
Jiaxun Yang <jiaxun.yang@flygoat.com> wrote:
> 
> This controller appeared on Loongson family of chips as the primary
> package interrupt source.
> 
> Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
> ---
>  drivers/irqchip/Kconfig                |   9 +
>  drivers/irqchip/Makefile               |   1 +
>  drivers/irqchip/irq-loongson-liointc.c | 338 +++++++++++++++++++++++++
>  3 files changed, 348 insertions(+)
>  create mode 100644 drivers/irqchip/irq-loongson-liointc.c
> 
> diff --git a/drivers/irqchip/Kconfig b/drivers/irqchip/Kconfig
> index 6d397732138d..c609eaa319d2 100644
> --- a/drivers/irqchip/Kconfig
> +++ b/drivers/irqchip/Kconfig
> @@ -513,4 +513,13 @@ config EXYNOS_IRQ_COMBINER
>  	  Say yes here to add support for the IRQ combiner devices embedded
>  	  in Samsung Exynos chips.
>  
> +config LOONGSON_LIOINTC
> +	bool "Loongson Local I/O Interrupt Controller"
> +	depends on MACH_LOONGSON64
> +	default y
> +	select IRQ_DOMAIN
> +	select GENERIC_IRQ_CHIP
> +	help
> +	  Support for the Loongson Local I/O Interrupt Controller.
> +
>  endmenu
> diff --git a/drivers/irqchip/Makefile b/drivers/irqchip/Makefile
> index eae0d78cbf22..5e7678efdfe6 100644
> --- a/drivers/irqchip/Makefile
> +++ b/drivers/irqchip/Makefile
> @@ -105,3 +105,4 @@ obj-$(CONFIG_MADERA_IRQ)		+= irq-madera.o
>  obj-$(CONFIG_LS1X_IRQ)			+= irq-ls1x.o
>  obj-$(CONFIG_TI_SCI_INTR_IRQCHIP)	+= irq-ti-sci-intr.o
>  obj-$(CONFIG_TI_SCI_INTA_IRQCHIP)	+= irq-ti-sci-inta.o
> +obj-$(CONFIG_LOONGSON_LIOINTC)		+= irq-loongson-liointc.o
> diff --git a/drivers/irqchip/irq-loongson-liointc.c b/drivers/irqchip/irq-loongson-liointc.c
> new file mode 100644
> index 000000000000..d7efab9f4ee7
> --- /dev/null
> +++ b/drivers/irqchip/irq-loongson-liointc.c
> @@ -0,0 +1,338 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + *  Copyright (C) 2020, Jiaxun Yang <jiaxun.yang@flygoat.com>
> + *  Loongson Local IO Interrupt Controller support
> + */
> +
> +#include <linux/errno.h>
> +#include <linux/init.h>
> +#include <linux/types.h>
> +#include <linux/interrupt.h>
> +#include <linux/ioport.h>
> +#include <linux/irqchip.h>
> +#include <linux/of_address.h>
> +#include <linux/of_irq.h>
> +#include <linux/io.h>
> +#include <linux/smp.h>
> +#include <linux/irqchip/chained_irq.h>
> +
> +#include <boot_param.h>
> +
> +#define LIOINTC_CHIP_IRQ	32
> +#define LIOINTC_NUM_PARENT 4
> +
> +#define LIOINTC_REG_INTx_MAP(x)	(x * 0x1)

Wow. I guess the 0x prefix changes everything. Or did you intend to
have something else here?

> +#define LIOINTC_INTC_CHIP_START	0x20
> +
> +#define LIOINTC_REG_INTC_STATUS	(LIOINTC_INTC_CHIP_START + 0x20)
> +#define LIOINTC_REG_INTC_EN_STATUS	(LIOINTC_INTC_CHIP_START + 0x04)
> +#define LIOINTC_REG_INTC_ENABLE	(LIOINTC_INTC_CHIP_START + 0x08)
> +#define LIOINTC_REG_INTC_DISABLE	(LIOINTC_INTC_CHIP_START + 0x0c)
> +#define LIOINTC_REG_INTC_POL	(LIOINTC_INTC_CHIP_START + 0x10)
> +#define LIOINTC_REG_INTC_EDGE	(LIOINTC_INTC_CHIP_START + 0x14)
> +
> +#define BUGGY_LPC_IRQ	10

Why BUGGY? Yes, there is a bug in the controller, but the interrupt
does exist, right?

> +
> +#define LIOINTC_SHIFT_INTx	4
> +
> +struct liointc_handler_data {
> +	struct liointc_priv *priv;
> +	u32 parent_int_map;
> +};
> +
> +struct liointc_priv {
> +	void __iomem *base;
> +	struct irq_chip_generic *gc;
> +	u8 map_cache[LIOINTC_CHIP_IRQ];
> +	struct liointc_handler_data handler[LIOINTC_NUM_PARENT];
> +	u8 possible_parent_mask;
> +	bool have_lpc_irq_bug;

For the sake of making this readable, consider reformating this
structure like this:

	void __iomem			*base;
	struct irq_chip_generic		*gc;
	u8 				map_cache[LIOINTC_CHIP_IRQ];
	struct liointc_handler_data	handler[LIOINTC_NUM_PARENT];
	u8				possible_parent_mask;
	bool				have_lpc_irq_bug;

which shows that there is certainly room for improvement in the layout
(it'd be a good idea to group together the u8 fields, for example).

> +};
> +
> +static void liointc_chained_handle_irq(struct irq_desc *desc)
> +{
> +	struct liointc_handler_data *handler = irq_desc_get_handler_data(desc);
> +	struct irq_chip *chip = irq_desc_get_chip(desc);
> +	struct irq_chip_generic *gc = handler->priv->gc;
> +	u32 pending;
> +
> +	chained_irq_enter(chip, desc);
> +
> +	pending = readl(gc->reg_base + LIOINTC_REG_INTC_STATUS);
> +
> +	if (!pending) {
> +		/* Always blame LPC IRQ if we have that Bug with LPC interrupt enabled */
> +		if (handler->priv->have_lpc_irq_bug &&
> +			(handler->parent_int_map & ~gc->mask_cache & BIT(BUGGY_LPC_IRQ)))
> +			generic_handle_irq(irq_find_mapping(gc->domain, BUGGY_LPC_IRQ));

This seems overly convoluted. Why not write something like:

			pending = BIT(LPC_IRQ);

instead of having two calls to generic_handle_irq() and co?

Also for the sake of making the patch more easily readable, consider
splitting the errata workarounds into a separate patch. It is much
easier to first work out the inner working of the driver without any
wart, and only then add the bad stuff on top.

> +		else
> +			spurious_interrupt();
> +	}
> +
> +	while (pending) {
> +		int bit = __ffs(pending);
> +
> +		generic_handle_irq(irq_find_mapping(gc->domain, bit));
> +		pending &= ~BIT(bit);
> +	}
> +
> +	chained_irq_exit(chip, desc);
> +}
> +
> +static void map_cache_set_core(struct liointc_priv *priv, int irq, int core)
> +{
> +	priv->map_cache[irq] &= ~GENMASK(3, 0);
> +	priv->map_cache[irq] |= BIT(core);

This is only called once, and nothing ever touches bits [3:0] until
then. So the first line is useless, and the second one is better moved
to the calling site. It is also always the same value, known at boot
time...

> +}
> +
> +static void write_map_cache(struct liointc_priv *priv, int irq)
> +{
> +	writeb(priv->map_cache[irq],
> +		priv->base + LIOINTC_REG_INTx_MAP(irq));
> +}
> +
> +static void liointc_set_bit(struct irq_chip_generic *gc,
> +				unsigned int offset,
> +				u32 mask, bool set)
> +{
> +	if (set)
> +		writel(readl(gc->reg_base + offset) | mask,
> +				gc->reg_base + offset);
> +	else
> +		writel(readl(gc->reg_base + offset) & ~mask,
> +				gc->reg_base + offset);
> +}
> +
> +static int liointc_set_type(struct irq_data *data, unsigned int type)
> +{
> +	struct irq_chip_generic *gc = irq_data_get_irq_chip_data(data);
> +	u32 mask = data->mask;
> +	unsigned long flags;
> +
> +	irq_gc_lock_irqsave(gc, flags);
> +	switch (type) {
> +	case IRQ_TYPE_LEVEL_HIGH:
> +		liointc_set_bit(gc, LIOINTC_REG_INTC_EDGE, mask, false);
> +		liointc_set_bit(gc, LIOINTC_REG_INTC_POL, mask, true);
> +		break;
> +	case IRQ_TYPE_LEVEL_LOW:
> +		liointc_set_bit(gc, LIOINTC_REG_INTC_EDGE, mask, false);
> +		liointc_set_bit(gc, LIOINTC_REG_INTC_POL, mask, false);
> +		break;
> +	case IRQ_TYPE_EDGE_RISING:
> +		liointc_set_bit(gc, LIOINTC_REG_INTC_EDGE, mask, true);
> +		liointc_set_bit(gc, LIOINTC_REG_INTC_POL, mask, true);
> +		break;
> +	case IRQ_TYPE_EDGE_FALLING:
> +		liointc_set_bit(gc, LIOINTC_REG_INTC_EDGE, mask, true);
> +		liointc_set_bit(gc, LIOINTC_REG_INTC_POL, mask, false);
> +		break;
> +	default:
> +		return -EINVAL;
> +	}
> +	irq_gc_unlock_irqrestore(gc, flags);
> +
> +	irqd_set_trigger_type(data, type);
> +	return 0;
> +}
> +
> +static int liointc_set_affinity(struct irq_data *idata,
> +				const cpumask_t *cpu_mask, bool force)
> +{
> +	return -ENAVAIL;

-EINVAL is the canonical return code for unimplemented callbacks.

> +}
> +
> +static void liointc_resume(struct irq_chip_generic *gc)
> +{
> +	struct liointc_priv *priv = gc->private;
> +	unsigned long flags;
> +	int i;
> +
> +	irq_gc_lock_irqsave(gc, flags);
> +	/* Revert map cache */
> +	for (i = 0; i < LIOINTC_CHIP_IRQ; i++)
> +		write_map_cache(priv, i);
> +
> +	/* Revert mask cache again */
> +	writel(gc->mask_cache, gc->reg_base + LIOINTC_REG_INTC_DISABLE);

Shouldn't you *disable everything* before restoring anything at all?
I'd expect something like

writel(0xffffffff, gc->reg_base + LIOINTC_REG_INTC_DISABLE);

to be the first thing you do in this function. Otherwise, you could
end-up getting spurious interrupts as you have already started
restoring stuff.

> +	writel(~gc->mask_cache, gc->reg_base + LIOINTC_REG_INTC_ENABLE);
> +	irq_gc_unlock_irqrestore(gc, flags);
> +}
> +
> +static void validate_parent_mask(struct liointc_priv *priv, u32 of_parent_int_map[])
> +{
> +	u32 proceed_mask = 0x0, duplicated_mask = 0x0;
> +	int i;
> +	int fallback_parent = __ffs(priv->possible_parent_mask);
> +
> +	for (i = 0; i < LIOINTC_NUM_PARENT; i++) {
> +		/* Try if the parent is avilable */
> +		if (!(priv->possible_parent_mask & BIT(i)))
> +			continue;
> +
> +		priv->handler[i].parent_int_map = of_parent_int_map[i];
> +
> +		/* Detect if the IRQ have previously proceed */
> +		duplicated_mask |= (priv->handler[i].parent_int_map & proceed_mask);
> +		proceed_mask |= priv->handler[i].parent_int_map;
> +	}
> +
> +	/* Fallback IRQs with no map bit set */
> +	while (~proceed_mask) {
> +		int bit = __ffs(~proceed_mask);
> +
> +		pr_warn("loongson-liointc: Found homeless IRQ %d, map to INT%d\n",
> +			bit, fallback_parent);
> +		priv->handler[fallback_parent].parent_int_map |= BIT(bit);
> +		proceed_mask |= BIT(bit);
> +	}
> +
> +	/* Fallback IRQs with multiple map bit set */
> +	while (duplicated_mask) {
> +		int bit = __ffs(duplicated_mask);
> +
> +		pr_warn("loongson-liointc: IRQ %d have multiple parents, map to INT%d\n",
> +			bit, fallback_parent);
> +		/* Clear the bit in all parent bits */
> +		for (i = 0; i < LIOINTC_NUM_PARENT; i++)
> +			priv->handler[i].parent_int_map &= ~BIT(bit);
> +
> +		priv->handler[fallback_parent].parent_int_map |= BIT(bit);
> +		duplicated_mask &= ~BIT(bit);
> +	}
> +
> +	/* Generate parent INT part of map Cache */
> +	for (i = 0; i < LIOINTC_NUM_PARENT; i++) {
> +		u32 pending = priv->handler[i].parent_int_map;
> +
> +		while (pending) {
> +			int bit = __ffs(pending);
> +
> +			priv->map_cache[bit] = BIT(i) << LIOINTC_SHIFT_INTx;
> +			pending &= ~BIT(bit);
> +		}
> +	}

Most of this function seems to be a validation tool for badly written
DT. Please do that at DT compilation time if you want, but not in the
kernel. At the worse, check for overlap/missing interrupts and fail,
but don't let people rely on default behaviours.

> +}
> +
> +static const char *parent_names[] = {"int0", "int1", "int2", "int3"};
> +
> +int __init liointc_of_init(struct device_node *node,
> +				struct device_node *parent)
> +{
> +	struct irq_chip_generic *gc;
> +	struct irq_domain *domain;
> +	struct irq_chip_type *ct;
> +	struct liointc_priv *priv;
> +	u32 of_parent_int_map[LIOINTC_NUM_PARENT];
> +	int parent_irq[LIOINTC_NUM_PARENT];
> +	int core = loongson_sysconf.boot_cpu_id;

Spurious variable. Just use the boot cpu id where needed (there is
only one instance).

> +	int i, err = 0;
> +	int sz;
> +
> +	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
> +	if (!priv)
> +		return -ENOMEM;
> +
> +	priv->base = of_iomap(node, 0);
> +	if (!priv->base) {
> +		err = -ENODEV;
> +		goto out_free_priv;
> +	}
> +
> +	if (of_device_is_compatible(node, "loongson,liointc-1.0"))
> +		priv->have_lpc_irq_bug = true;
> +
> +	for (i = 0; i < LIOINTC_NUM_PARENT; i++) {
> +		parent_irq[i] = of_irq_get_byname(node, parent_names[i]);
> +		if (parent_irq[i] >= 0)

0 isn't a valid interrupt.

> +			priv->possible_parent_mask |= BIT(i);

And once you got rid of the above DT validation, this field should go
away.

> +	}
> +
> +
> +	if (!priv->possible_parent_mask) {
> +		pr_err("loongson-liointc: No parent\n");
> +		err = -ENOMEM;
> +		goto out_iounmap;
> +	}
> +
> +	sz = of_property_read_variable_u32_array(node, "loongson,parent_int_map",
> +						&of_parent_int_map[0], LIOINTC_NUM_PARENT,
> +						LIOINTC_NUM_PARENT);
> +	if (sz < 4) {
> +		pr_err("loongson-liointc: No parent_int_map\n");
> +		err = -ENODEV;

That's probably a -EINVAL again. The device is there, the data is bogus.

> +		goto out_iounmap;
> +	}
> +
> +	/* Setup IRQ domain */
> +	domain = irq_domain_add_linear(node, 32,
> +					&irq_generic_chip_ops, priv);
> +	if (!domain) {
> +		pr_err("loongson-liointc: cannot add IRQ domain\n");
> +		err = -ENOMEM;
> +		goto out_iounmap;
> +	}
> +
> +	err = irq_alloc_domain_generic_chips(domain, 32, 1,
> +					node->full_name, handle_level_irq,
> +					IRQ_NOPROBE, 0, 0);
> +	if (err) {
> +		pr_err("loongson-liointc: unable to register IRQ domain\n");
> +		err = -ENOMEM;

You already have err set to the right value.

> +		goto out_free_domain;
> +	}
> +
> +
> +	/* Disable all IRQs */
> +	writel(0xffffffff, priv->base + LIOINTC_REG_INTC_DISABLE);
> +	/* Set to level triggered */
> +	writel(0x0, priv->base + LIOINTC_REG_INTC_EDGE);
> +
> +	validate_parent_mask(priv, &of_parent_int_map[0]);
> +
> +	for (i = 0; i < LIOINTC_CHIP_IRQ; i++) {
> +		map_cache_set_core(priv, i, core);
> +		write_map_cache(priv, i);
> +	}
> +
> +	gc = irq_get_domain_generic_chip(domain, 0);
> +	gc->private = priv;
> +	gc->reg_base = priv->base;

If gc->base is already the VA for the irqchip, why do you need to
cache it in the private structure as well?

> +	gc->domain = domain;
> +	gc->resume = liointc_resume;
> +
> +	ct = gc->chip_types;
> +	ct->regs.enable = LIOINTC_REG_INTC_ENABLE;
> +	ct->regs.disable = LIOINTC_REG_INTC_DISABLE;
> +	ct->chip.irq_unmask = irq_gc_unmask_enable_reg;
> +	ct->chip.irq_mask = irq_gc_mask_disable_reg;
> +	ct->chip.irq_mask_ack = irq_gc_mask_disable_reg;
> +	ct->chip.irq_set_type = liointc_set_type;
> +	ct->chip.irq_set_affinity = liointc_set_affinity;
> +
> +	gc->mask_cache = 0xffffffff;
> +	priv->gc = gc;
> +
> +	for (i = 0; i < LIOINTC_NUM_PARENT; i++) {
> +		if (parent_irq[i] < 0)

0 is an invalid interrupt number.

> +			continue;
> +
> +		priv->handler[i].priv = priv;
> +		irq_set_chained_handler_and_data(parent_irq[i],
> +				liointc_chained_handle_irq, &priv->handler[i]);
> +	}
> +
> +	return 0;
> +
> +out_free_domain:
> +	irq_domain_remove(domain);
> +out_iounmap:
> +	iounmap(priv->base);
> +out_free_priv:
> +	kfree(priv);
> +
> +	return err;
> +}
> +
> +IRQCHIP_DECLARE(loongson_liointc_1_0, "loongson,liointc-1.0", liointc_of_init);
> +IRQCHIP_DECLARE(loongson_liointc_1_0a, "loongson,liointc-1.0a", liointc_of_init);

Thanks,

	M.

-- 
Jazz is not dead, it just smells funny.