Re: [PATCH v5 2/6] PCI: iproc: Add INTx support with better modeling

[Marc Zyngier <maz@kernel.org>]

On 2020-03-26 06:48, Srinath Mannam wrote:
> From: Ray Jui <ray.jui@broadcom.com>
> 
> Add PCIe legacy interrupt INTx support to the iProc PCIe driver by
> modeling it with its own IRQ domain. All 4 interrupts INTA, INTB, INTC,
> INTD share the same interrupt line connected to the GIC in the system,
> while the status of each INTx can be obtained through the INTX CSR
> register.
> 
> Signed-off-by: Ray Jui <ray.jui@broadcom.com>
> Signed-off-by: Srinath Mannam <srinath.mannam@broadcom.com>
> Reviewed-by: Andrew Murray <andrew.murray@arm.com>
> ---
>  drivers/pci/controller/pcie-iproc.c | 147 
> +++++++++++++++++++++++++++++++++++-
>  drivers/pci/controller/pcie-iproc.h |   8 ++
>  2 files changed, 153 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/pci/controller/pcie-iproc.c
> b/drivers/pci/controller/pcie-iproc.c
> index 0a468c7..62d8f43 100644
> --- a/drivers/pci/controller/pcie-iproc.c
> +++ b/drivers/pci/controller/pcie-iproc.c
> @@ -14,6 +14,7 @@
>  #include <linux/delay.h>
>  #include <linux/interrupt.h>
>  #include <linux/irqchip/arm-gic-v3.h>
> +#include <linux/irqchip/chained_irq.h>
>  #include <linux/platform_device.h>
>  #include <linux/of_address.h>
>  #include <linux/of_pci.h>
> @@ -270,6 +271,7 @@ enum iproc_pcie_reg {
> 
>  	/* enable INTx */
>  	IPROC_PCIE_INTX_EN,
> +	IPROC_PCIE_INTX_CSR,
> 
>  	/* outbound address mapping */
>  	IPROC_PCIE_OARR0,
> @@ -314,6 +316,7 @@ static const u16 iproc_pcie_reg_paxb_bcma[] = {
>  	[IPROC_PCIE_CFG_ADDR]		= 0x1f8,
>  	[IPROC_PCIE_CFG_DATA]		= 0x1fc,
>  	[IPROC_PCIE_INTX_EN]		= 0x330,
> +	[IPROC_PCIE_INTX_CSR]		= 0x334,
>  	[IPROC_PCIE_LINK_STATUS]	= 0xf0c,
>  };
> 
> @@ -325,6 +328,7 @@ static const u16 iproc_pcie_reg_paxb[] = {
>  	[IPROC_PCIE_CFG_ADDR]		= 0x1f8,
>  	[IPROC_PCIE_CFG_DATA]		= 0x1fc,
>  	[IPROC_PCIE_INTX_EN]		= 0x330,
> +	[IPROC_PCIE_INTX_CSR]		= 0x334,
>  	[IPROC_PCIE_OARR0]		= 0xd20,
>  	[IPROC_PCIE_OMAP0]		= 0xd40,
>  	[IPROC_PCIE_OARR1]		= 0xd28,
> @@ -341,6 +345,7 @@ static const u16 iproc_pcie_reg_paxb_v2[] = {
>  	[IPROC_PCIE_CFG_ADDR]		= 0x1f8,
>  	[IPROC_PCIE_CFG_DATA]		= 0x1fc,
>  	[IPROC_PCIE_INTX_EN]		= 0x330,
> +	[IPROC_PCIE_INTX_CSR]		= 0x334,
>  	[IPROC_PCIE_OARR0]		= 0xd20,
>  	[IPROC_PCIE_OMAP0]		= 0xd40,
>  	[IPROC_PCIE_OARR1]		= 0xd28,
> @@ -846,9 +851,142 @@ static int iproc_pcie_check_link(struct 
> iproc_pcie *pcie)
>  	return link_is_active ? 0 : -ENODEV;
>  }
> 
> -static void iproc_pcie_enable(struct iproc_pcie *pcie)
> +static void iproc_pcie_mask_irq(struct irq_data *d)
>  {
> +	struct iproc_pcie *pcie = irq_data_get_irq_chip_data(d);
> +	u32 val;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&pcie->intx_lock, flags);

As Arnd already noticed, this needs to be defined as  a raw spinlock.

> +	val =  iproc_pcie_read_reg(pcie, IPROC_PCIE_INTX_EN);
> +	val &= ~(BIT(irqd_to_hwirq(d)));
> +	iproc_pcie_write_reg(pcie, IPROC_PCIE_INTX_EN, val);
> +	spin_unlock_irqrestore(&pcie->intx_lock, flags);
> +}
> +
> +static void iproc_pcie_unmask_irq(struct irq_data *d)
> +{
> +	struct iproc_pcie *pcie = irq_data_get_irq_chip_data(d);
> +	u32 val;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&pcie->intx_lock, flags);
> +	val =  iproc_pcie_read_reg(pcie, IPROC_PCIE_INTX_EN);
> +	val |= (BIT(irqd_to_hwirq(d)));
> +	iproc_pcie_write_reg(pcie, IPROC_PCIE_INTX_EN, val);
> +	spin_unlock_irqrestore(&pcie->intx_lock, flags);
> +}
> +
> +static struct irq_chip iproc_pcie_irq_chip = {
> +	.name = "pcie-iproc-intc",
> +	.irq_enable = iproc_pcie_unmask_irq,
> +	.irq_disable = iproc_pcie_mask_irq,

No. Either the enable/disable callbacks are different from unmask/mask,
or they don't exist.

- A disabled interrupt is free to loose pending interrupts
- A masked interrupt doesn't loose interrupts.

I'm pretty sure yours is the latter.

> +	.irq_mask = iproc_pcie_mask_irq,
> +	.irq_unmask = iproc_pcie_unmask_irq,
> +};
> +
> +static int iproc_pcie_intx_map(struct irq_domain *domain, unsigned int 
> irq,
> +			       irq_hw_number_t hwirq)
> +{
> +	irq_set_chip_and_handler(irq, &iproc_pcie_irq_chip, 
> handle_level_irq);
> +	irq_set_chip_data(irq, domain->host_data);
> +
> +	return 0;
> +}
> +
> +static const struct irq_domain_ops intx_domain_ops = {
> +	.map = iproc_pcie_intx_map,
> +};
> +
> +static void iproc_pcie_isr(struct irq_desc *desc)
> +{
> +	struct irq_chip *chip = irq_desc_get_chip(desc);
> +	struct iproc_pcie *pcie;
> +	struct device *dev;
> +	unsigned long status;
> +	u32 bit, virq;
> +
> +	chained_irq_enter(chip, desc);
> +	pcie = irq_desc_get_handler_data(desc);
> +	dev = pcie->dev;
> +
> +	/* go through INTx A, B, C, D until all interrupts are handled */
> +	do {
> +		status = iproc_pcie_read_reg(pcie, IPROC_PCIE_INTX_CSR);
> +		for_each_set_bit(bit, &status, PCI_NUM_INTX) {
> +			virq = irq_find_mapping(pcie->irq_domain, bit);
> +			if (virq)
> +				generic_handle_irq(virq);
> +			else
> +				dev_err(dev, "unexpected INTx%u\n", bit);

I'd rather you avoid this kind of unlimited console screaming in 
interrupt
context. Either ratelimit it, or even better turn it into a debug 
statement.

> +		}
> +	} while ((status & SYS_RC_INTX_MASK) != 0);
> +
> +	chained_irq_exit(chip, desc);
> +}
> +
> +static int iproc_pcie_intx_enable(struct iproc_pcie *pcie)
> +{
> +	struct device *dev = pcie->dev;
> +	struct device_node *node;
> +	int ret;
> +
> +	/*
> +	 * BCMA devices do not map INTx the same way as platform devices. All
> +	 * BCMA needs below line to enable INTx
> +	 */
>  	iproc_pcie_write_reg(pcie, IPROC_PCIE_INTX_EN, SYS_RC_INTX_MASK);
> +
> +	node = of_get_compatible_child(dev->of_node, "brcm,iproc-intc");
> +	if (node)
> +		pcie->irq = of_irq_get(node, 0);
> +
> +	if (!node || pcie->irq <= 0)
> +		return 0;
> +
> +	spin_lock_init(&pcie->intx_lock);
> +
> +	/* set IRQ handler */
> +	irq_set_chained_handler_and_data(pcie->irq, iproc_pcie_isr, pcie);

Please do this last, once the domain is allocated (you leave a window 
where
a screaming interrupt can fire here).

> +
> +	/* add IRQ domain for INTx */
> +	pcie->irq_domain = irq_domain_add_linear(node, PCI_NUM_INTX,
> +						 &intx_domain_ops, pcie);
> +	if (!pcie->irq_domain) {
> +		dev_err(dev, "failed to add INTx IRQ domain\n");
> +		ret = -ENOMEM;
> +		goto err_rm_handler_data;
> +	}
> +
> +	return 0;
> +
> +err_rm_handler_data:
> +	of_node_put(node);
> +	irq_set_chained_handler_and_data(pcie->irq, NULL, NULL);
> +
> +	return ret;
> +}
> +
> +static void iproc_pcie_intx_disable(struct iproc_pcie *pcie)
> +{
> +	uint32_t offset, virq;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&pcie->intx_lock, flags);
> +	iproc_pcie_write_reg(pcie, IPROC_PCIE_INTX_EN, 0x0);
> +	spin_unlock_irqrestore(&pcie->intx_lock, flags);

What's the reason for this lock/unlock? What are you racing against?

> +
> +	if (pcie->irq <= 0)
> +		return;
> +
> +	for (offset = 0; offset < PCI_NUM_INTX; offset++) {
> +		virq = irq_find_mapping(pcie->irq_domain, offset);
> +		if (virq)
> +			irq_dispose_mapping(virq);
> +	}
> +
> +	irq_domain_remove(pcie->irq_domain);
> +	irq_set_chained_handler_and_data(pcie->irq, NULL, NULL);

The other way around: first you disconnect the interrupt, then you
free the data that the interrupt can make use of.

>  }
> 
>  static inline bool iproc_pcie_ob_is_valid(struct iproc_pcie *pcie,
> @@ -1518,7 +1656,11 @@ int iproc_pcie_setup(struct iproc_pcie *pcie,
> struct list_head *res)
>  		goto err_power_off_phy;
>  	}
> 
> -	iproc_pcie_enable(pcie);
> +	ret = iproc_pcie_intx_enable(pcie);
> +	if (ret) {
> +		dev_err(dev, "failed to enable INTx\n");
> +		goto err_power_off_phy;
> +	}
> 
>  	if (IS_ENABLED(CONFIG_PCI_MSI))
>  		if (iproc_pcie_msi_enable(pcie))
> @@ -1562,6 +1704,7 @@ int iproc_pcie_remove(struct iproc_pcie *pcie)
>  	pci_remove_root_bus(pcie->root_bus);
> 
>  	iproc_pcie_msi_disable(pcie);
> +	iproc_pcie_intx_disable(pcie);
> 
>  	phy_power_off(pcie->phy);
>  	phy_exit(pcie->phy);
> diff --git a/drivers/pci/controller/pcie-iproc.h
> b/drivers/pci/controller/pcie-iproc.h
> index 4f03ea5..787bfba 100644
> --- a/drivers/pci/controller/pcie-iproc.h
> +++ b/drivers/pci/controller/pcie-iproc.h
> @@ -74,9 +74,13 @@ struct iproc_msi;
>   * @ib: inbound mapping related parameters
>   * @ib_map: outbound mapping region related parameters
>   *
> + * @irq: interrupt line wired to the generic GIC for INTx
> + * @irq_domain: IRQ domain for INTx
> + *
>   * @need_msi_steer: indicates additional configuration of the iProc 
> PCIe
>   * controller is required to steer MSI writes to external interrupt 
> controller
>   * @msi: MSI data
> + * @intx_lock: spinlock to protect access to INTx related registers
>   */
>  struct iproc_pcie {
>  	struct device *dev;
> @@ -102,8 +106,12 @@ struct iproc_pcie {
>  	struct iproc_pcie_ib ib;
>  	const struct iproc_pcie_ib_map *ib_map;
> 
> +	int irq;
> +	struct irq_domain *irq_domain;
> +
>  	bool need_msi_steer;
>  	struct iproc_msi *msi;
> +	spinlock_t intx_lock;
>  };
> 
>  int iproc_pcie_setup(struct iproc_pcie *pcie, struct list_head *res);

Thanks,

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