diff mbox

[v2,4/5] PCI: iproc: Add iProc PCIe MSI support

Message ID 1448406294-732-5-git-send-email-rjui@broadcom.com (mailing list archive)
State New, archived
Delegated to: Bjorn Helgaas
Headers show

Commit Message

Ray Jui Nov. 24, 2015, 11:04 p.m. UTC
This patch adds PCIe MSI support for both PAXB and PAXC interfaces on
all iProc based platforms. The patch follows the latest trend in the
kernel to use MSI domain based implementation

This iProc event queue based MSI support should not be used with newer
platforms with integrated MSI support in the GIC (e.g., giv2m or
gicv3-its)

Signed-off-by: Ray Jui <rjui@broadcom.com>
Reviewed-by: Anup Patel <anup.patel@broadcom.com>
Reviewed-by: Vikram Prakash <vikramp@broadcom.com>
Reviewed-by: Scott Branden <sbranden@broadcom.com>
---
 drivers/pci/host/Kconfig          |   9 +
 drivers/pci/host/Makefile         |   1 +
 drivers/pci/host/pcie-iproc-msi.c | 662 ++++++++++++++++++++++++++++++++++++++
 drivers/pci/host/pcie-iproc.c     |  26 ++
 drivers/pci/host/pcie-iproc.h     |  21 +-
 5 files changed, 717 insertions(+), 2 deletions(-)
 create mode 100644 drivers/pci/host/pcie-iproc-msi.c

Comments

Marc Zyngier Nov. 25, 2015, 5:36 p.m. UTC | #1
On Tue, 24 Nov 2015 15:04:53 -0800
Ray Jui <rjui@broadcom.com> wrote:

> This patch adds PCIe MSI support for both PAXB and PAXC interfaces on
> all iProc based platforms. The patch follows the latest trend in the
> kernel to use MSI domain based implementation
> 

That's a pretty useless comment. The general trend in the kernel is to
use the most appropriate infrastructure.

> This iProc event queue based MSI support should not be used with newer
> platforms with integrated MSI support in the GIC (e.g., giv2m or
> gicv3-its)
> 

I'd be more interested in some documentation explaining how the HW
works, how the various data structures are updated, and when.

> Signed-off-by: Ray Jui <rjui@broadcom.com>
> Reviewed-by: Anup Patel <anup.patel@broadcom.com>
> Reviewed-by: Vikram Prakash <vikramp@broadcom.com>
> Reviewed-by: Scott Branden <sbranden@broadcom.com>
> ---
>  drivers/pci/host/Kconfig          |   9 +
>  drivers/pci/host/Makefile         |   1 +
>  drivers/pci/host/pcie-iproc-msi.c | 662 ++++++++++++++++++++++++++++++++++++++
>  drivers/pci/host/pcie-iproc.c     |  26 ++
>  drivers/pci/host/pcie-iproc.h     |  21 +-
>  5 files changed, 717 insertions(+), 2 deletions(-)
>  create mode 100644 drivers/pci/host/pcie-iproc-msi.c
> 
> diff --git a/drivers/pci/host/Kconfig b/drivers/pci/host/Kconfig
> index f131ba9..972e906 100644
> --- a/drivers/pci/host/Kconfig
> +++ b/drivers/pci/host/Kconfig
> @@ -126,6 +126,15 @@ config PCIE_IPROC
>  	  iProc family of SoCs. An appropriate bus interface driver also needs
>  	  to be enabled
>  
> +config PCIE_IPROC_MSI
> +	bool "Broadcom iProc PCIe MSI support"
> +	depends on ARCH_BCM_IPROC && PCI_MSI
> +	select PCI_MSI_IRQ_DOMAIN
> +	default ARCH_BCM_IPROC
> +	help
> +	  Say Y here if you want to enable MSI support for Broadcom's iProc
> +	  PCIe controller
> +
>  config PCIE_IPROC_PLATFORM
>  	tristate "Broadcom iProc PCIe platform bus driver"
>  	depends on ARCH_BCM_IPROC || (ARM && COMPILE_TEST)
> diff --git a/drivers/pci/host/Makefile b/drivers/pci/host/Makefile
> index 9d4d3c6..0e4e95e 100644
> --- a/drivers/pci/host/Makefile
> +++ b/drivers/pci/host/Makefile
> @@ -15,6 +15,7 @@ obj-$(CONFIG_PCI_XGENE_MSI) += pci-xgene-msi.o
>  obj-$(CONFIG_PCI_LAYERSCAPE) += pci-layerscape.o
>  obj-$(CONFIG_PCI_VERSATILE) += pci-versatile.o
>  obj-$(CONFIG_PCIE_IPROC) += pcie-iproc.o
> +obj-$(CONFIG_PCIE_IPROC_MSI) += pcie-iproc-msi.o
>  obj-$(CONFIG_PCIE_IPROC_PLATFORM) += pcie-iproc-platform.o
>  obj-$(CONFIG_PCIE_IPROC_BCMA) += pcie-iproc-bcma.o
>  obj-$(CONFIG_PCIE_ALTERA) += pcie-altera.o
> diff --git a/drivers/pci/host/pcie-iproc-msi.c b/drivers/pci/host/pcie-iproc-msi.c
> new file mode 100644
> index 0000000..afc54c2
> --- /dev/null
> +++ b/drivers/pci/host/pcie-iproc-msi.c
> @@ -0,0 +1,662 @@
> +/*
> + * Copyright (C) 2015 Broadcom Corporation
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License as
> + * published by the Free Software Foundation version 2.
> + *
> + * This program is distributed "as is" WITHOUT ANY WARRANTY of any
> + * kind, whether express or implied; without even the implied warranty
> + * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> + * GNU General Public License for more details.
> + */
> +
> +#include <linux/interrupt.h>
> +#include <linux/irqchip/chained_irq.h>
> +#include <linux/irqdomain.h>
> +#include <linux/msi.h>
> +#include <linux/of_irq.h>
> +#include <linux/of_pci.h>
> +#include <linux/pci.h>
> +
> +#include "pcie-iproc.h"
> +
> +#define IPROC_MSI_INTR_EN_SHIFT        11
> +#define IPROC_MSI_INTR_EN              BIT(IPROC_MSI_INTR_EN_SHIFT)
> +#define IPROC_MSI_INT_N_EVENT_SHIFT    1
> +#define IPROC_MSI_INT_N_EVENT          BIT(IPROC_MSI_INT_N_EVENT_SHIFT)
> +#define IPROC_MSI_EQ_EN_SHIFT          0
> +#define IPROC_MSI_EQ_EN                BIT(IPROC_MSI_EQ_EN_SHIFT)
> +
> +#define IPROC_MSI_EQ_MASK              0x3f
> +
> +/* max number of GIC interrupts */
> +#define NR_HW_IRQS                     6
> +
> +/* number of entries in each event queue */
> +#define EQ_LEN                         64
> +
> +/* size of each event queue memory region */
> +#define EQ_MEM_REGION_SIZE             SZ_4K
> +
> +/* size of each MSI message memory region */
> +#define MSI_MEM_REGION_SIZE            SZ_4K
> +
> +enum iproc_msi_reg {
> +	IPROC_MSI_EQ_PAGE = 0,
> +	IPROC_MSI_EQ_PAGE_UPPER,
> +	IPROC_MSI_PAGE,
> +	IPROC_MSI_PAGE_UPPER,
> +	IPROC_MSI_CTRL,
> +	IPROC_MSI_EQ_HEAD,
> +	IPROC_MSI_EQ_TAIL,
> +	IPROC_MSI_INTS_EN,
> +	IPROC_MSI_REG_SIZE,
> +};
> +
> +struct iproc_msi;
> +
> +/**
> + * iProc MSI group
> + *
> + * One MSI group is allocated per GIC interrupt, serviced by one iProc MSI
> + * event queue
> + *
> + * @msi: pointer to iProc MSI data
> + * @gic_irq: GIC interrupt
> + * @eq: Event queue number
> + */
> +struct iproc_msi_grp {
> +	struct iproc_msi *msi;
> +	int gic_irq;
> +	unsigned int eq;
> +};
> +
> +/**
> + * iProc event queue based MSI
> + *
> + * Only meant to be used on platforms without MSI support integrated into the
> + * GIC
> + *
> + * @pcie: pointer to iProc PCIe data
> + * @reg_offsets: MSI register offsets
> + * @grps: MSI groups
> + * @nr_irqs: number of total interrupts connected to GIC
> + * @nr_cpus: number of toal CPUs
> + * @has_inten_reg: indicates the MSI interrupt enable register needs to be
> + * set explicitly (required for some legacy platforms)
> + * @bitmap: MSI vector bitmap
> + * @bitmap_lock: lock to protect access to the MSI bitmap
> + * @nr_msi_vecs: total number of MSI vectors
> + * @inner_domain: inner IRQ domain
> + * @msi_domain: MSI IRQ domain
> + * @nr_eq_region: required number of 4K aligned memory region for MSI event
> + * queues
> + * @nr_msi_region: required number of 4K aligned memory region for MSI posted
> + * writes
> + * @eq_cpu: pointer to allocated memory region for MSI event queues
> + * @eq_dma: DMA address of MSI event queues
> + * @msi_cpu: pointer to allocated memory region for MSI posted writes
> + * @msi_dma: DMA address of MSI posted writes
> + */
> +struct iproc_msi {
> +	struct iproc_pcie *pcie;
> +	const u16 (*reg_offsets)[IPROC_MSI_REG_SIZE];
> +	struct iproc_msi_grp *grps;
> +	int nr_irqs;
> +	int nr_cpus;
> +	bool has_inten_reg;
> +	unsigned long *bitmap;
> +	struct mutex bitmap_lock;
> +	unsigned int nr_msi_vecs;
> +	struct irq_domain *inner_domain;
> +	struct irq_domain *msi_domain;
> +	unsigned int nr_eq_region;
> +	unsigned int nr_msi_region;
> +	void *eq_cpu;
> +	dma_addr_t eq_dma;
> +	void *msi_cpu;
> +	dma_addr_t msi_dma;
> +};
> +
> +static const u16 iproc_msi_reg_paxb[NR_HW_IRQS][IPROC_MSI_REG_SIZE] = {
> +	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x210, 0x250, 0x254, 0x208 },
> +	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x214, 0x258, 0x25c, 0x208 },
> +	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x218, 0x260, 0x264, 0x208 },
> +	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x21c, 0x268, 0x26c, 0x208 },
> +	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x220, 0x270, 0x274, 0x208 },
> +	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x224, 0x278, 0x27c, 0x208 },
> +};
> +
> +static const u16 iproc_msi_reg_paxc[NR_HW_IRQS][IPROC_MSI_REG_SIZE] = {
> +	{ 0xc00, 0xc04, 0xc08, 0xc0c, 0xc40, 0xc50, 0xc60 },
> +	{ 0xc10, 0xc14, 0xc18, 0xc1c, 0xc44, 0xc54, 0xc64 },
> +	{ 0xc20, 0xc24, 0xc28, 0xc2c, 0xc48, 0xc58, 0xc68 },
> +	{ 0xc30, 0xc34, 0xc38, 0xc3c, 0xc4c, 0xc5c, 0xc6c },
> +};
> +
> +static inline u32 iproc_msi_read_reg(struct iproc_msi *msi,
> +				     enum iproc_msi_reg reg,
> +				     unsigned int eq)
> +{
> +	struct iproc_pcie *pcie = msi->pcie;
> +
> +	return readl_relaxed(pcie->base + msi->reg_offsets[eq][reg]);
> +}
> +
> +static inline void iproc_msi_write_reg(struct iproc_msi *msi,
> +				       enum iproc_msi_reg reg,
> +				       int eq, u32 val)
> +{
> +	struct iproc_pcie *pcie = msi->pcie;
> +
> +	writel_relaxed(val, pcie->base + msi->reg_offsets[eq][reg]);
> +}
> +
> +static inline bool iproc_msi_has_mult_regions(struct iproc_msi *msi)
> +{
> +	return ((msi->nr_msi_region > 1) ? true : false);

	return msi->nr_msi_region > 1;

> +}
> +
> +static inline bool iproc_eq_has_mult_regions(struct iproc_msi *msi)
> +{
> +	return ((msi->nr_eq_region > 1) ? true : false);

	return msi->nr_eq_region > 1;

> +}
> +
> +static struct irq_chip iproc_msi_irq_chip = {
> +	.name = "iProc-MSI",
> +};
> +
> +static struct msi_domain_info iproc_msi_domain_info = {
> +	.flags = MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
> +		MSI_FLAG_PCI_MSIX,
> +	.chip = &iproc_msi_irq_chip,
> +};
> +
> +/*
> + * In iProc PCIe core, each MSI group is serviced by a GIC interrupt and a
> + * dedicated event queue. Each MSI group can support up to 64 MSI vectors
> + *
> + * The number of MSI groups varies between different iProc SoCs. The total
> + * number of CPU cores also varies. To support MSI IRQ affinity, we
> + * distribute GIC interrupts across all available CPUs. MSI vector is moved
> + * from one GIC interrupt to another to steer to the target CPU
> + *
> + * Assuming:
> + * - the number of MSI groups is M
> + * - the number of CPU cores is N
> + * - M is always a multiple of N

How do you enforce that last condition?

> + *
> + * Total number of raw MSI vectors = M * 64
> + * Total number of supported MSI vectors = (M * 64) / N
> + */
> +static inline int hwirq_to_cpu(struct iproc_msi *msi, unsigned long hwirq)
> +{
> +	return (hwirq % msi->nr_cpus);
> +}
> +
> +static inline unsigned long hwirq_to_canonical_hwirq(struct iproc_msi *msi,
> +						     unsigned long hwirq)
> +{
> +	return (hwirq - hwirq_to_cpu(msi, hwirq));
> +}
> +
> +static int iproc_msi_irq_set_affinity(struct irq_data *data,
> +				      const struct cpumask *mask, bool force)
> +{
> +	struct iproc_msi *msi = irq_data_get_irq_chip_data(data);
> +	int target_cpu = cpumask_first(mask);
> +	int curr_cpu;
> +
> +	curr_cpu = hwirq_to_cpu(msi, data->hwirq);
> +	if (curr_cpu == target_cpu)
> +		return IRQ_SET_MASK_OK_DONE;
> +
> +	/* steer MSI to the target CPU */
> +	data->hwirq = hwirq_to_canonical_hwirq(msi, data->hwirq) + target_cpu;
> +
> +	return IRQ_SET_MASK_OK;
> +}
> +
> +static inline u32 hwirq_to_group(struct iproc_msi *msi, unsigned long hwirq)
> +{
> +	return (hwirq % msi->nr_irqs);
> +}
> +
> +static void iproc_msi_irq_compose_msi_msg(struct irq_data *data,
> +					  struct msi_msg *msg)
> +{
> +	struct iproc_msi *msi = irq_data_get_irq_chip_data(data);
> +	dma_addr_t addr;
> +	unsigned int mul;
> +
> +	if (iproc_msi_has_mult_regions(msi))
> +		mul = MSI_MEM_REGION_SIZE;
> +	else
> +		mul = sizeof(u32);

Since this is the only use of that function, why don't you have it to
directly return the right multiplier?

> +
> +	addr = msi->msi_dma + hwirq_to_group(msi, data->hwirq) * mul;
> +	msg->address_lo = lower_32_bits(addr);
> +	msg->address_hi = upper_32_bits(addr);
> +	msg->data = data->hwirq;
> +}
> +
> +static struct irq_chip iproc_msi_bottom_irq_chip = {
> +	.name = "MSI",
> +	.irq_set_affinity = iproc_msi_irq_set_affinity,
> +	.irq_compose_msi_msg = iproc_msi_irq_compose_msi_msg,
> +};
> +
> +static int iproc_msi_irq_domain_alloc(struct irq_domain *domain,
> +				      unsigned int virq, unsigned int nr_irqs,
> +				      void *args)
> +{
> +	struct iproc_msi *msi = domain->host_data;
> +	int hwirq;
> +
> +	mutex_lock(&msi->bitmap_lock);
> +
> +	/* allocate 'nr_cpus' number of MSI vectors each time */
> +	hwirq = bitmap_find_next_zero_area(msi->bitmap, msi->nr_msi_vecs, 0,
> +					   msi->nr_cpus, 0);
> +	if (hwirq < msi->nr_msi_vecs)
> +		bitmap_set(msi->bitmap, hwirq, msi->nr_cpus);
> +	else
> +		return -ENOSPC;

Deadlock, here we come...

> +
> +	mutex_unlock(&msi->bitmap_lock);
> +
> +	irq_domain_set_info(domain, virq, hwirq, &iproc_msi_bottom_irq_chip,
> +			    domain->host_data, handle_simple_irq, NULL, NULL);
> +
> +	return 0;
> +}
> +
> +static void iproc_msi_irq_domain_free(struct irq_domain *domain,
> +				      unsigned int virq, unsigned int nr_irqs)
> +{
> +	struct irq_data *data = irq_domain_get_irq_data(domain, virq);
> +	struct iproc_msi *msi = irq_data_get_irq_chip_data(data);
> +	unsigned int hwirq;
> +
> +	mutex_lock(&msi->bitmap_lock);
> +
> +	hwirq = hwirq_to_canonical_hwirq(msi, data->hwirq);
> +	bitmap_clear(msi->bitmap, hwirq, msi->nr_cpus);
> +
> +	mutex_unlock(&msi->bitmap_lock);
> +
> +	irq_domain_free_irqs_parent(domain, virq, nr_irqs);
> +}
> +
> +static const struct irq_domain_ops msi_domain_ops = {
> +	.alloc = iproc_msi_irq_domain_alloc,
> +	.free = iproc_msi_irq_domain_free,
> +};
> +
> +static inline u32 decode_msi_hwirq(struct iproc_msi *msi, u32 eq, u32 head)
> +{
> +	u32 *msg, hwirq;
> +	unsigned int offs;
> +
> +	if (iproc_eq_has_mult_regions(msi))
> +		offs = eq * EQ_MEM_REGION_SIZE;
> +	else
> +		offs = eq * EQ_LEN * sizeof(u32);

Same here.

> +
> +	offs += head * sizeof(u32);
> +	msg = (u32 *)(msi->eq_cpu + offs);

If that's the only place where you dereference msi->eq_cpu, why doesn't
it have the right type?

> +	hwirq = *msg & IPROC_MSI_EQ_MASK;
> +
> +	/*
> +	 * Since we have multiple hwirq mapped to a single MSI vector,
> +	 * now we need to derive the hwirq at CPU0. It can then be used to
> +	 * mapped back to virq
> +	 */
> +	return hwirq_to_canonical_hwirq(msi, hwirq);
> +}
> +
> +static void iproc_msi_handler(struct irq_desc *desc)
> +{
> +	struct irq_chip *chip = irq_desc_get_chip(desc);
> +	struct iproc_msi_grp *grp;
> +	struct iproc_msi *msi;
> +	struct iproc_pcie *pcie;
> +	u32 eq, head, tail, nr_events;
> +	unsigned long hwirq;
> +	int virq;
> +
> +	chained_irq_enter(chip, desc);
> +
> +	grp = irq_desc_get_handler_data(desc);
> +	msi = grp->msi;
> +	pcie = msi->pcie;
> +	eq = grp->eq;
> +
> +	/*
> +	 * iProc MSI event queue is tracked by head and tail pointers. Head
> +	 * pointer indicates the next entry to be processed by SW in the
> +	 * queue. Entries between head and tail pointers contain valid MSI
> +	 * data to be processed
> +	 */
> +	head = iproc_msi_read_reg(msi, IPROC_MSI_EQ_HEAD,
> +				  eq) & IPROC_MSI_EQ_MASK;
> +	do {
> +		tail = iproc_msi_read_reg(msi, IPROC_MSI_EQ_TAIL,
> +					  eq) & IPROC_MSI_EQ_MASK;
> +
> +		/*
> +		 * Figure out total number of events (MSI data) to be
> +		 * processed
> +		 */
> +		nr_events = (tail < head) ?
> +			(EQ_LEN - (head - tail)) : (tail - head);
> +		if (!nr_events)
> +			break;
> +
> +		/* process all outstanding events */
> +		while (nr_events--) {
> +			hwirq = decode_msi_hwirq(msi, eq, head);
> +			virq = irq_find_mapping(msi->inner_domain, hwirq);
> +			generic_handle_irq(virq);
> +
> +			head++;
> +			head %= EQ_LEN;
> +			iproc_msi_write_reg(msi, IPROC_MSI_EQ_HEAD, eq, head);
> +		}

Wouldn't it be better to process all nr_events and only update head
once?

> +
> +		/*
> +		 * Now go read the tail pointer again to see if there are new
> +		 * oustanding events that came in during the above window
> +		 */
> +	} while (true);
> +
> +	chained_irq_exit(chip, desc);
> +}
> +
> +static void iproc_msi_enable(struct iproc_msi *msi)
> +{
> +	int i, eq;
> +	u32 val;
> +
> +	/* program memory region for each event queue */
> +	for (i = 0; i < msi->nr_eq_region; i++) {
> +		dma_addr_t addr = msi->eq_dma + (i * EQ_MEM_REGION_SIZE);
> +
> +		iproc_msi_write_reg(msi, IPROC_MSI_EQ_PAGE, i,
> +				    lower_32_bits(addr));
> +		iproc_msi_write_reg(msi, IPROC_MSI_EQ_PAGE_UPPER, i,
> +				    upper_32_bits(addr));
> +	}
> +
> +	/* program memory region for MSI posted writes */
> +	for (i = 0; i < msi->nr_msi_region; i++) {
> +		dma_addr_t addr = msi->msi_dma + (i * MSI_MEM_REGION_SIZE);
> +
> +		iproc_msi_write_reg(msi, IPROC_MSI_PAGE, i,
> +				    lower_32_bits(addr));
> +		iproc_msi_write_reg(msi, IPROC_MSI_PAGE_UPPER, i,
> +				    upper_32_bits(addr));
> +	}
> +
> +	for (eq = 0; eq < msi->nr_irqs; eq++) {
> +		/* enable MSI event queue */
> +		val = IPROC_MSI_INTR_EN | IPROC_MSI_INT_N_EVENT |
> +			IPROC_MSI_EQ_EN;
> +		iproc_msi_write_reg(msi, IPROC_MSI_CTRL, eq, val);
> +
> +		/*
> +		 * Some legacy platforms require the MSI interrupt enable
> +		 * register to be set explicitly
> +		 */
> +		if (msi->has_inten_reg) {
> +			val = iproc_msi_read_reg(msi, IPROC_MSI_INTS_EN, eq);
> +			val |= BIT(eq);
> +			iproc_msi_write_reg(msi, IPROC_MSI_INTS_EN, eq, val);
> +		}
> +	}
> +}
> +
> +static void iproc_msi_disable(struct iproc_msi *msi)
> +{
> +	u32 eq, val;
> +
> +	for (eq = 0; eq < msi->nr_irqs; eq++) {
> +		if (msi->has_inten_reg) {
> +			val = iproc_msi_read_reg(msi, IPROC_MSI_INTS_EN, eq);
> +			val &= ~BIT(eq);
> +			iproc_msi_write_reg(msi, IPROC_MSI_INTS_EN, eq, val);
> +		}
> +
> +		val = iproc_msi_read_reg(msi, IPROC_MSI_CTRL, eq);
> +		val &= ~(IPROC_MSI_INTR_EN | IPROC_MSI_INT_N_EVENT |
> +			 IPROC_MSI_EQ_EN);
> +		iproc_msi_write_reg(msi, IPROC_MSI_CTRL, eq, val);
> +	}
> +}
> +
> +static int iproc_msi_alloc_domains(struct device_node *node,
> +				   struct iproc_msi *msi)
> +{
> +	msi->inner_domain = irq_domain_add_linear(NULL, msi->nr_msi_vecs,
> +						  &msi_domain_ops, msi);
> +	if (!msi->inner_domain)
> +		return -ENOMEM;
> +
> +	msi->msi_domain = pci_msi_create_irq_domain(of_node_to_fwnode(node),
> +						    &iproc_msi_domain_info,
> +						    msi->inner_domain);
> +	if (!msi->msi_domain) {
> +		irq_domain_remove(msi->inner_domain);
> +		return -ENOMEM;
> +	}
> +
> +	return 0;
> +}
> +
> +static void iproc_msi_free_domains(struct iproc_msi *msi)
> +{
> +	if (msi->msi_domain)
> +		irq_domain_remove(msi->msi_domain);
> +
> +	if (msi->inner_domain)
> +		irq_domain_remove(msi->inner_domain);
> +}
> +
> +static int iproc_msi_irq_setup(struct iproc_msi *msi, unsigned int cpu)
> +{
> +	int i, ret;
> +	cpumask_var_t mask;
> +	struct iproc_pcie *pcie = msi->pcie;
> +
> +	for (i = cpu; i < msi->nr_irqs; i += msi->nr_cpus) {
> +		irq_set_chained_handler_and_data(msi->grps[i].gic_irq,
> +						 iproc_msi_handler,
> +						 &msi->grps[i]);
> +		/* dedicate GIC interrupt to each CPU core */
> +		if (alloc_cpumask_var(&mask, GFP_KERNEL)) {
> +			cpumask_clear(mask);
> +			cpumask_set_cpu(cpu, mask);
> +			ret = irq_set_affinity(msi->grps[i].gic_irq, mask);
> +			if (ret)
> +				dev_err(pcie->dev,
> +					"failed to set affinity for IRQ%d\n",
> +					msi->grps[i].gic_irq);
> +			free_cpumask_var(mask);
> +		} else {
> +			dev_err(pcie->dev, "failed to alloc CPU mask\n");
> +			ret = -EINVAL;
> +		}
> +
> +		if (ret) {
> +			irq_set_chained_handler_and_data(msi->grps[i].gic_irq,
> +							 NULL, NULL);
> +			return ret;

What happens to interrupts you've already configured? I'd expect a full
rollback.

> +		}
> +	}
> +
> +	return 0;
> +}
> +
> +static void iproc_msi_irq_free(struct iproc_msi *msi, unsigned int cpu)
> +{
> +	int i;
> +
> +	for (i = cpu; i < msi->nr_irqs; i += msi->nr_cpus) {
> +		irq_set_chained_handler_and_data(msi->grps[i].gic_irq,
> +						 NULL, NULL);
> +	}
> +}
> +
> +int iproc_msi_init(struct iproc_pcie *pcie, struct device_node *node)
> +{
> +	struct iproc_msi *msi;
> +	int i, ret;
> +	unsigned int cpu;
> +
> +	if (!of_device_is_compatible(node, "brcm,iproc-msi"))
> +		return -ENODEV;
> +
> +	if (!of_find_property(node, "msi-controller", NULL))
> +		return -ENODEV;
> +
> +	if (pcie->msi)
> +		return -EBUSY;
> +
> +	msi = devm_kzalloc(pcie->dev, sizeof(*msi), GFP_KERNEL);
> +	if (!msi)
> +		return -ENOMEM;
> +
> +	msi->pcie = pcie;
> +	pcie->msi = msi;
> +	mutex_init(&msi->bitmap_lock);
> +	msi->nr_cpus = num_possible_cpus();
> +
> +	switch (pcie->type) {
> +	case IPROC_PCIE_PAXB:
> +		msi->reg_offsets = iproc_msi_reg_paxb;
> +		break;
> +	case IPROC_PCIE_PAXC:
> +		msi->reg_offsets = iproc_msi_reg_paxc;
> +		break;
> +	default:
> +		dev_err(pcie->dev, "incompatible iProc PCIe interface\n");
> +		return -EINVAL;
> +	}
> +
> +	if (of_find_property(node, "brcm,pcie-msi-inten", NULL))
> +		msi->has_inten_reg = true;
> +
> +	ret = of_property_read_u32(node, "brcm,num-eq-region",
> +				   &msi->nr_eq_region);
> +	if (ret || !msi->nr_eq_region)
> +		msi->nr_eq_region = 1;
> +
> +	ret = of_property_read_u32(node, "brcm,num-msi-msg-region",
> +				   &msi->nr_msi_region);
> +	if (ret || !msi->nr_msi_region)
> +		msi->nr_msi_region = 1;
> +
> +	msi->nr_irqs = of_irq_count(node);
> +	if (!msi->nr_irqs) {
> +		dev_err(pcie->dev, "found no MSI GIC interrupt\n");
> +		return -ENODEV;
> +	}
> +	if (msi->nr_irqs > NR_HW_IRQS) {
> +		dev_warn(pcie->dev, "too many MSI GIC interrupts defined %d\n",
> +			 msi->nr_irqs);
> +		msi->nr_irqs = NR_HW_IRQS;
> +	}
> +
> +	msi->nr_msi_vecs = msi->nr_irqs * EQ_LEN;
> +	msi->bitmap = devm_kcalloc(pcie->dev, BITS_TO_LONGS(msi->nr_msi_vecs),
> +				   sizeof(*msi->bitmap), GFP_KERNEL);
> +	if (!msi->bitmap)
> +		return -ENOMEM;
> +
> +	msi->grps = devm_kcalloc(pcie->dev, msi->nr_irqs, sizeof(*msi->grps),
> +				 GFP_KERNEL);
> +	if (!msi->grps)
> +		return -ENOMEM;
> +
> +	for (i = 0; i < msi->nr_irqs; i++) {
> +		unsigned int irq = irq_of_parse_and_map(node, i);
> +
> +		if (!irq) {
> +			dev_err(pcie->dev, "unable to parse/map interrupt\n");
> +			return -ENODEV;
> +		}
> +		msi->grps[i].gic_irq = irq;
> +		msi->grps[i].msi = msi;
> +		msi->grps[i].eq = i;
> +	}
> +
> +	/* reserve memory for MSI event queue */
> +	msi->eq_cpu = dma_alloc_coherent(pcie->dev,
> +					 msi->nr_eq_region * EQ_MEM_REGION_SIZE,
> +					 &msi->eq_dma, GFP_KERNEL);

Do you need to zero that memory? Or is the HW happy with whatever will
be there?

> +	if (!msi->eq_cpu)
> +		return -ENOMEM;
> +
> +	/* reserve memory for MSI posted writes */
> +	msi->msi_cpu = dma_alloc_coherent(pcie->dev,
> +					  msi->nr_msi_region * MSI_MEM_REGION_SIZE,
> +					  &msi->msi_dma, GFP_KERNEL);

Same here. Also, what is the exact purpose of that memory? You have a
coherent mapping with the CPU, but you never read from it. So what's
the point?

> +	if (!msi->msi_cpu) {
> +		ret = -ENOMEM;
> +		goto free_eq_dma;
> +	}
> +
> +	ret = iproc_msi_alloc_domains(node, msi);
> +	if (ret) {
> +		dev_err(pcie->dev, "failed to create MSI domains\n");
> +		goto free_msi_dma;
> +	}
> +
> +	for_each_online_cpu(cpu) {
> +		ret = iproc_msi_irq_setup(msi, cpu);
> +		if (ret)
> +			goto free_msi_irq;
> +	}
> +
> +	iproc_msi_enable(msi);
> +
> +	return 0;
> +
> +free_msi_irq:
> +	for_each_online_cpu(cpu)
> +		iproc_msi_irq_free(msi, cpu);
> +	iproc_msi_free_domains(msi);
> +
> +free_msi_dma:
> +	dma_free_coherent(pcie->dev, msi->nr_msi_region * MSI_MEM_REGION_SIZE,
> +			  msi->msi_cpu, msi->msi_dma);
> +free_eq_dma:
> +	dma_free_coherent(pcie->dev, msi->nr_eq_region * EQ_MEM_REGION_SIZE,
> +			  msi->eq_cpu, msi->eq_dma);
> +	pcie->msi = NULL;
> +	return ret;
> +}
> +EXPORT_SYMBOL(iproc_msi_init);

[...]

Thanks,

	M.
Ray Jui Nov. 26, 2015, 1:52 a.m. UTC | #2
Hi Marc,

On 11/25/2015 9:36 AM, Marc Zyngier wrote:
> On Tue, 24 Nov 2015 15:04:53 -0800
> Ray Jui <rjui@broadcom.com> wrote:
>
>> This patch adds PCIe MSI support for both PAXB and PAXC interfaces on
>> all iProc based platforms. The patch follows the latest trend in the
>> kernel to use MSI domain based implementation
>>
>
> That's a pretty useless comment. The general trend in the kernel is to
> use the most appropriate infrastructure.
>
>> This iProc event queue based MSI support should not be used with newer
>> platforms with integrated MSI support in the GIC (e.g., giv2m or
>> gicv3-its)
>>
>
> I'd be more interested in some documentation explaining how the HW
> works, how the various data structures are updated, and when.
>

Okay will try my best to describe my understanding of iProc MSI in the 
commit message of my next iteration of patches.

>> Signed-off-by: Ray Jui <rjui@broadcom.com>
>> Reviewed-by: Anup Patel <anup.patel@broadcom.com>
>> Reviewed-by: Vikram Prakash <vikramp@broadcom.com>
>> Reviewed-by: Scott Branden <sbranden@broadcom.com>
>> ---
>>   drivers/pci/host/Kconfig          |   9 +
>>   drivers/pci/host/Makefile         |   1 +
>>   drivers/pci/host/pcie-iproc-msi.c | 662 ++++++++++++++++++++++++++++++++++++++
>>   drivers/pci/host/pcie-iproc.c     |  26 ++
>>   drivers/pci/host/pcie-iproc.h     |  21 +-
>>   5 files changed, 717 insertions(+), 2 deletions(-)
>>   create mode 100644 drivers/pci/host/pcie-iproc-msi.c
>>
>> diff --git a/drivers/pci/host/Kconfig b/drivers/pci/host/Kconfig
>> index f131ba9..972e906 100644
>> --- a/drivers/pci/host/Kconfig
>> +++ b/drivers/pci/host/Kconfig
>> @@ -126,6 +126,15 @@ config PCIE_IPROC
>>   	  iProc family of SoCs. An appropriate bus interface driver also needs
>>   	  to be enabled
>>
>> +config PCIE_IPROC_MSI
>> +	bool "Broadcom iProc PCIe MSI support"
>> +	depends on ARCH_BCM_IPROC && PCI_MSI
>> +	select PCI_MSI_IRQ_DOMAIN
>> +	default ARCH_BCM_IPROC
>> +	help
>> +	  Say Y here if you want to enable MSI support for Broadcom's iProc
>> +	  PCIe controller
>> +
>>   config PCIE_IPROC_PLATFORM
>>   	tristate "Broadcom iProc PCIe platform bus driver"
>>   	depends on ARCH_BCM_IPROC || (ARM && COMPILE_TEST)
>> diff --git a/drivers/pci/host/Makefile b/drivers/pci/host/Makefile
>> index 9d4d3c6..0e4e95e 100644
>> --- a/drivers/pci/host/Makefile
>> +++ b/drivers/pci/host/Makefile
>> @@ -15,6 +15,7 @@ obj-$(CONFIG_PCI_XGENE_MSI) += pci-xgene-msi.o
>>   obj-$(CONFIG_PCI_LAYERSCAPE) += pci-layerscape.o
>>   obj-$(CONFIG_PCI_VERSATILE) += pci-versatile.o
>>   obj-$(CONFIG_PCIE_IPROC) += pcie-iproc.o
>> +obj-$(CONFIG_PCIE_IPROC_MSI) += pcie-iproc-msi.o
>>   obj-$(CONFIG_PCIE_IPROC_PLATFORM) += pcie-iproc-platform.o
>>   obj-$(CONFIG_PCIE_IPROC_BCMA) += pcie-iproc-bcma.o
>>   obj-$(CONFIG_PCIE_ALTERA) += pcie-altera.o
>> diff --git a/drivers/pci/host/pcie-iproc-msi.c b/drivers/pci/host/pcie-iproc-msi.c
>> new file mode 100644
>> index 0000000..afc54c2
>> --- /dev/null
>> +++ b/drivers/pci/host/pcie-iproc-msi.c
>> @@ -0,0 +1,662 @@
>> +/*
>> + * Copyright (C) 2015 Broadcom Corporation
>> + *
>> + * This program is free software; you can redistribute it and/or
>> + * modify it under the terms of the GNU General Public License as
>> + * published by the Free Software Foundation version 2.
>> + *
>> + * This program is distributed "as is" WITHOUT ANY WARRANTY of any
>> + * kind, whether express or implied; without even the implied warranty
>> + * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
>> + * GNU General Public License for more details.
>> + */
>> +
>> +#include <linux/interrupt.h>
>> +#include <linux/irqchip/chained_irq.h>
>> +#include <linux/irqdomain.h>
>> +#include <linux/msi.h>
>> +#include <linux/of_irq.h>
>> +#include <linux/of_pci.h>
>> +#include <linux/pci.h>
>> +
>> +#include "pcie-iproc.h"
>> +
>> +#define IPROC_MSI_INTR_EN_SHIFT        11
>> +#define IPROC_MSI_INTR_EN              BIT(IPROC_MSI_INTR_EN_SHIFT)
>> +#define IPROC_MSI_INT_N_EVENT_SHIFT    1
>> +#define IPROC_MSI_INT_N_EVENT          BIT(IPROC_MSI_INT_N_EVENT_SHIFT)
>> +#define IPROC_MSI_EQ_EN_SHIFT          0
>> +#define IPROC_MSI_EQ_EN                BIT(IPROC_MSI_EQ_EN_SHIFT)
>> +
>> +#define IPROC_MSI_EQ_MASK              0x3f
>> +
>> +/* max number of GIC interrupts */
>> +#define NR_HW_IRQS                     6
>> +
>> +/* number of entries in each event queue */
>> +#define EQ_LEN                         64
>> +
>> +/* size of each event queue memory region */
>> +#define EQ_MEM_REGION_SIZE             SZ_4K
>> +
>> +/* size of each MSI message memory region */
>> +#define MSI_MEM_REGION_SIZE            SZ_4K
>> +
>> +enum iproc_msi_reg {
>> +	IPROC_MSI_EQ_PAGE = 0,
>> +	IPROC_MSI_EQ_PAGE_UPPER,
>> +	IPROC_MSI_PAGE,
>> +	IPROC_MSI_PAGE_UPPER,
>> +	IPROC_MSI_CTRL,
>> +	IPROC_MSI_EQ_HEAD,
>> +	IPROC_MSI_EQ_TAIL,
>> +	IPROC_MSI_INTS_EN,
>> +	IPROC_MSI_REG_SIZE,
>> +};
>> +
>> +struct iproc_msi;
>> +
>> +/**
>> + * iProc MSI group
>> + *
>> + * One MSI group is allocated per GIC interrupt, serviced by one iProc MSI
>> + * event queue
>> + *
>> + * @msi: pointer to iProc MSI data
>> + * @gic_irq: GIC interrupt
>> + * @eq: Event queue number
>> + */
>> +struct iproc_msi_grp {
>> +	struct iproc_msi *msi;
>> +	int gic_irq;
>> +	unsigned int eq;
>> +};
>> +
>> +/**
>> + * iProc event queue based MSI
>> + *
>> + * Only meant to be used on platforms without MSI support integrated into the
>> + * GIC
>> + *
>> + * @pcie: pointer to iProc PCIe data
>> + * @reg_offsets: MSI register offsets
>> + * @grps: MSI groups
>> + * @nr_irqs: number of total interrupts connected to GIC
>> + * @nr_cpus: number of toal CPUs
>> + * @has_inten_reg: indicates the MSI interrupt enable register needs to be
>> + * set explicitly (required for some legacy platforms)
>> + * @bitmap: MSI vector bitmap
>> + * @bitmap_lock: lock to protect access to the MSI bitmap
>> + * @nr_msi_vecs: total number of MSI vectors
>> + * @inner_domain: inner IRQ domain
>> + * @msi_domain: MSI IRQ domain
>> + * @nr_eq_region: required number of 4K aligned memory region for MSI event
>> + * queues
>> + * @nr_msi_region: required number of 4K aligned memory region for MSI posted
>> + * writes
>> + * @eq_cpu: pointer to allocated memory region for MSI event queues
>> + * @eq_dma: DMA address of MSI event queues
>> + * @msi_cpu: pointer to allocated memory region for MSI posted writes
>> + * @msi_dma: DMA address of MSI posted writes
>> + */
>> +struct iproc_msi {
>> +	struct iproc_pcie *pcie;
>> +	const u16 (*reg_offsets)[IPROC_MSI_REG_SIZE];
>> +	struct iproc_msi_grp *grps;
>> +	int nr_irqs;
>> +	int nr_cpus;
>> +	bool has_inten_reg;
>> +	unsigned long *bitmap;
>> +	struct mutex bitmap_lock;
>> +	unsigned int nr_msi_vecs;
>> +	struct irq_domain *inner_domain;
>> +	struct irq_domain *msi_domain;
>> +	unsigned int nr_eq_region;
>> +	unsigned int nr_msi_region;
>> +	void *eq_cpu;
>> +	dma_addr_t eq_dma;
>> +	void *msi_cpu;
>> +	dma_addr_t msi_dma;
>> +};
>> +
>> +static const u16 iproc_msi_reg_paxb[NR_HW_IRQS][IPROC_MSI_REG_SIZE] = {
>> +	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x210, 0x250, 0x254, 0x208 },
>> +	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x214, 0x258, 0x25c, 0x208 },
>> +	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x218, 0x260, 0x264, 0x208 },
>> +	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x21c, 0x268, 0x26c, 0x208 },
>> +	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x220, 0x270, 0x274, 0x208 },
>> +	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x224, 0x278, 0x27c, 0x208 },
>> +};
>> +
>> +static const u16 iproc_msi_reg_paxc[NR_HW_IRQS][IPROC_MSI_REG_SIZE] = {
>> +	{ 0xc00, 0xc04, 0xc08, 0xc0c, 0xc40, 0xc50, 0xc60 },
>> +	{ 0xc10, 0xc14, 0xc18, 0xc1c, 0xc44, 0xc54, 0xc64 },
>> +	{ 0xc20, 0xc24, 0xc28, 0xc2c, 0xc48, 0xc58, 0xc68 },
>> +	{ 0xc30, 0xc34, 0xc38, 0xc3c, 0xc4c, 0xc5c, 0xc6c },
>> +};
>> +
>> +static inline u32 iproc_msi_read_reg(struct iproc_msi *msi,
>> +				     enum iproc_msi_reg reg,
>> +				     unsigned int eq)
>> +{
>> +	struct iproc_pcie *pcie = msi->pcie;
>> +
>> +	return readl_relaxed(pcie->base + msi->reg_offsets[eq][reg]);
>> +}
>> +
>> +static inline void iproc_msi_write_reg(struct iproc_msi *msi,
>> +				       enum iproc_msi_reg reg,
>> +				       int eq, u32 val)
>> +{
>> +	struct iproc_pcie *pcie = msi->pcie;
>> +
>> +	writel_relaxed(val, pcie->base + msi->reg_offsets[eq][reg]);
>> +}
>> +
>> +static inline bool iproc_msi_has_mult_regions(struct iproc_msi *msi)
>> +{
>> +	return ((msi->nr_msi_region > 1) ? true : false);
>
> 	return msi->nr_msi_region > 1;
>

Will change this function to return the multiplier according to your 
comment below.

>> +}
>> +
>> +static inline bool iproc_eq_has_mult_regions(struct iproc_msi *msi)
>> +{
>> +	return ((msi->nr_eq_region > 1) ? true : false);
>
> 	return msi->nr_eq_region > 1;
>

Will change this function to return the offsets directly.

>> +}
>> +
>> +static struct irq_chip iproc_msi_irq_chip = {
>> +	.name = "iProc-MSI",
>> +};
>> +
>> +static struct msi_domain_info iproc_msi_domain_info = {
>> +	.flags = MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
>> +		MSI_FLAG_PCI_MSIX,
>> +	.chip = &iproc_msi_irq_chip,
>> +};
>> +
>> +/*
>> + * In iProc PCIe core, each MSI group is serviced by a GIC interrupt and a
>> + * dedicated event queue. Each MSI group can support up to 64 MSI vectors
>> + *
>> + * The number of MSI groups varies between different iProc SoCs. The total
>> + * number of CPU cores also varies. To support MSI IRQ affinity, we
>> + * distribute GIC interrupts across all available CPUs. MSI vector is moved
>> + * from one GIC interrupt to another to steer to the target CPU
>> + *
>> + * Assuming:
>> + * - the number of MSI groups is M
>> + * - the number of CPU cores is N
>> + * - M is always a multiple of N
>
> How do you enforce that last condition?
>

Okay I need to enforce this as soon as the number of GIC interrupts is 
determined from DT in the init routine. I will add code to enforce this 
condition.

>> + *
>> + * Total number of raw MSI vectors = M * 64
>> + * Total number of supported MSI vectors = (M * 64) / N
>> + */
>> +static inline int hwirq_to_cpu(struct iproc_msi *msi, unsigned long hwirq)
>> +{
>> +	return (hwirq % msi->nr_cpus);
>> +}
>> +
>> +static inline unsigned long hwirq_to_canonical_hwirq(struct iproc_msi *msi,
>> +						     unsigned long hwirq)
>> +{
>> +	return (hwirq - hwirq_to_cpu(msi, hwirq));
>> +}
>> +
>> +static int iproc_msi_irq_set_affinity(struct irq_data *data,
>> +				      const struct cpumask *mask, bool force)
>> +{
>> +	struct iproc_msi *msi = irq_data_get_irq_chip_data(data);
>> +	int target_cpu = cpumask_first(mask);
>> +	int curr_cpu;
>> +
>> +	curr_cpu = hwirq_to_cpu(msi, data->hwirq);
>> +	if (curr_cpu == target_cpu)
>> +		return IRQ_SET_MASK_OK_DONE;
>> +
>> +	/* steer MSI to the target CPU */
>> +	data->hwirq = hwirq_to_canonical_hwirq(msi, data->hwirq) + target_cpu;
>> +
>> +	return IRQ_SET_MASK_OK;
>> +}
>> +
>> +static inline u32 hwirq_to_group(struct iproc_msi *msi, unsigned long hwirq)
>> +{
>> +	return (hwirq % msi->nr_irqs);
>> +}
>> +
>> +static void iproc_msi_irq_compose_msi_msg(struct irq_data *data,
>> +					  struct msi_msg *msg)
>> +{
>> +	struct iproc_msi *msi = irq_data_get_irq_chip_data(data);
>> +	dma_addr_t addr;
>> +	unsigned int mul;
>> +
>> +	if (iproc_msi_has_mult_regions(msi))
>> +		mul = MSI_MEM_REGION_SIZE;
>> +	else
>> +		mul = sizeof(u32);
>
> Since this is the only use of that function, why don't you have it to
> directly return the right multiplier?
>

True, that will make the above code simpler just one line. Will do!

>> +
>> +	addr = msi->msi_dma + hwirq_to_group(msi, data->hwirq) * mul;
>> +	msg->address_lo = lower_32_bits(addr);
>> +	msg->address_hi = upper_32_bits(addr);
>> +	msg->data = data->hwirq;
>> +}
>> +
>> +static struct irq_chip iproc_msi_bottom_irq_chip = {
>> +	.name = "MSI",
>> +	.irq_set_affinity = iproc_msi_irq_set_affinity,
>> +	.irq_compose_msi_msg = iproc_msi_irq_compose_msi_msg,
>> +};
>> +
>> +static int iproc_msi_irq_domain_alloc(struct irq_domain *domain,
>> +				      unsigned int virq, unsigned int nr_irqs,
>> +				      void *args)
>> +{
>> +	struct iproc_msi *msi = domain->host_data;
>> +	int hwirq;
>> +
>> +	mutex_lock(&msi->bitmap_lock);
>> +
>> +	/* allocate 'nr_cpus' number of MSI vectors each time */
>> +	hwirq = bitmap_find_next_zero_area(msi->bitmap, msi->nr_msi_vecs, 0,
>> +					   msi->nr_cpus, 0);
>> +	if (hwirq < msi->nr_msi_vecs)
>> +		bitmap_set(msi->bitmap, hwirq, msi->nr_cpus);
>> +	else
>> +		return -ENOSPC;
>
> Deadlock, here we come...
>

Damn it. My bad. Will fix.

>> +
>> +	mutex_unlock(&msi->bitmap_lock);
>> +
>> +	irq_domain_set_info(domain, virq, hwirq, &iproc_msi_bottom_irq_chip,
>> +			    domain->host_data, handle_simple_irq, NULL, NULL);
>> +
>> +	return 0;
>> +}
>> +
>> +static void iproc_msi_irq_domain_free(struct irq_domain *domain,
>> +				      unsigned int virq, unsigned int nr_irqs)
>> +{
>> +	struct irq_data *data = irq_domain_get_irq_data(domain, virq);
>> +	struct iproc_msi *msi = irq_data_get_irq_chip_data(data);
>> +	unsigned int hwirq;
>> +
>> +	mutex_lock(&msi->bitmap_lock);
>> +
>> +	hwirq = hwirq_to_canonical_hwirq(msi, data->hwirq);
>> +	bitmap_clear(msi->bitmap, hwirq, msi->nr_cpus);
>> +
>> +	mutex_unlock(&msi->bitmap_lock);
>> +
>> +	irq_domain_free_irqs_parent(domain, virq, nr_irqs);
>> +}
>> +
>> +static const struct irq_domain_ops msi_domain_ops = {
>> +	.alloc = iproc_msi_irq_domain_alloc,
>> +	.free = iproc_msi_irq_domain_free,
>> +};
>> +
>> +static inline u32 decode_msi_hwirq(struct iproc_msi *msi, u32 eq, u32 head)
>> +{
>> +	u32 *msg, hwirq;
>> +	unsigned int offs;
>> +
>> +	if (iproc_eq_has_mult_regions(msi))
>> +		offs = eq * EQ_MEM_REGION_SIZE;
>> +	else
>> +		offs = eq * EQ_LEN * sizeof(u32);
>
> Same here.
>

Okay. Will change the code. Thanks.

>> +
>> +	offs += head * sizeof(u32);
>> +	msg = (u32 *)(msi->eq_cpu + offs);
>
> If that's the only place where you dereference msi->eq_cpu, why doesn't
> it have the right type?
>

Okay this really caught me, :) I actually changed msi->eq_cpu to type 
'u32 *' and found things all of a sudden stopped working. Then I 
realized it's a lot easier to let msi->eq_cpu stay as 'void *', so we 
can keep the arithmetic to calcualte 'offs' as it is. If I change it to 
'u32 *', then I would need to divide 'offs' by sizeof(u32) to get right 
address of the pointer.

>> +	hwirq = *msg & IPROC_MSI_EQ_MASK;
>> +
>> +	/*
>> +	 * Since we have multiple hwirq mapped to a single MSI vector,
>> +	 * now we need to derive the hwirq at CPU0. It can then be used to
>> +	 * mapped back to virq
>> +	 */
>> +	return hwirq_to_canonical_hwirq(msi, hwirq);
>> +}
>> +
>> +static void iproc_msi_handler(struct irq_desc *desc)
>> +{
>> +	struct irq_chip *chip = irq_desc_get_chip(desc);
>> +	struct iproc_msi_grp *grp;
>> +	struct iproc_msi *msi;
>> +	struct iproc_pcie *pcie;
>> +	u32 eq, head, tail, nr_events;
>> +	unsigned long hwirq;
>> +	int virq;
>> +
>> +	chained_irq_enter(chip, desc);
>> +
>> +	grp = irq_desc_get_handler_data(desc);
>> +	msi = grp->msi;
>> +	pcie = msi->pcie;
>> +	eq = grp->eq;
>> +
>> +	/*
>> +	 * iProc MSI event queue is tracked by head and tail pointers. Head
>> +	 * pointer indicates the next entry to be processed by SW in the
>> +	 * queue. Entries between head and tail pointers contain valid MSI
>> +	 * data to be processed
>> +	 */
>> +	head = iproc_msi_read_reg(msi, IPROC_MSI_EQ_HEAD,
>> +				  eq) & IPROC_MSI_EQ_MASK;
>> +	do {
>> +		tail = iproc_msi_read_reg(msi, IPROC_MSI_EQ_TAIL,
>> +					  eq) & IPROC_MSI_EQ_MASK;
>> +
>> +		/*
>> +		 * Figure out total number of events (MSI data) to be
>> +		 * processed
>> +		 */
>> +		nr_events = (tail < head) ?
>> +			(EQ_LEN - (head - tail)) : (tail - head);
>> +		if (!nr_events)
>> +			break;
>> +
>> +		/* process all outstanding events */
>> +		while (nr_events--) {
>> +			hwirq = decode_msi_hwirq(msi, eq, head);
>> +			virq = irq_find_mapping(msi->inner_domain, hwirq);
>> +			generic_handle_irq(virq);
>> +
>> +			head++;
>> +			head %= EQ_LEN;
>> +			iproc_msi_write_reg(msi, IPROC_MSI_EQ_HEAD, eq, head);
>> +		}
>
> Wouldn't it be better to process all nr_events and only update head
> once?
>

You are right. That will help to get rid of the overhead of repeatedly 
writing to the event queue head register. Will make the change.

>> +
>> +		/*
>> +		 * Now go read the tail pointer again to see if there are new
>> +		 * oustanding events that came in during the above window
>> +		 */
>> +	} while (true);
>> +
>> +	chained_irq_exit(chip, desc);
>> +}
>> +
>> +static void iproc_msi_enable(struct iproc_msi *msi)
>> +{
>> +	int i, eq;
>> +	u32 val;
>> +
>> +	/* program memory region for each event queue */
>> +	for (i = 0; i < msi->nr_eq_region; i++) {
>> +		dma_addr_t addr = msi->eq_dma + (i * EQ_MEM_REGION_SIZE);
>> +
>> +		iproc_msi_write_reg(msi, IPROC_MSI_EQ_PAGE, i,
>> +				    lower_32_bits(addr));
>> +		iproc_msi_write_reg(msi, IPROC_MSI_EQ_PAGE_UPPER, i,
>> +				    upper_32_bits(addr));
>> +	}
>> +
>> +	/* program memory region for MSI posted writes */
>> +	for (i = 0; i < msi->nr_msi_region; i++) {
>> +		dma_addr_t addr = msi->msi_dma + (i * MSI_MEM_REGION_SIZE);
>> +
>> +		iproc_msi_write_reg(msi, IPROC_MSI_PAGE, i,
>> +				    lower_32_bits(addr));
>> +		iproc_msi_write_reg(msi, IPROC_MSI_PAGE_UPPER, i,
>> +				    upper_32_bits(addr));
>> +	}
>> +
>> +	for (eq = 0; eq < msi->nr_irqs; eq++) {
>> +		/* enable MSI event queue */
>> +		val = IPROC_MSI_INTR_EN | IPROC_MSI_INT_N_EVENT |
>> +			IPROC_MSI_EQ_EN;
>> +		iproc_msi_write_reg(msi, IPROC_MSI_CTRL, eq, val);
>> +
>> +		/*
>> +		 * Some legacy platforms require the MSI interrupt enable
>> +		 * register to be set explicitly
>> +		 */
>> +		if (msi->has_inten_reg) {
>> +			val = iproc_msi_read_reg(msi, IPROC_MSI_INTS_EN, eq);
>> +			val |= BIT(eq);
>> +			iproc_msi_write_reg(msi, IPROC_MSI_INTS_EN, eq, val);
>> +		}
>> +	}
>> +}
>> +
>> +static void iproc_msi_disable(struct iproc_msi *msi)
>> +{
>> +	u32 eq, val;
>> +
>> +	for (eq = 0; eq < msi->nr_irqs; eq++) {
>> +		if (msi->has_inten_reg) {
>> +			val = iproc_msi_read_reg(msi, IPROC_MSI_INTS_EN, eq);
>> +			val &= ~BIT(eq);
>> +			iproc_msi_write_reg(msi, IPROC_MSI_INTS_EN, eq, val);
>> +		}
>> +
>> +		val = iproc_msi_read_reg(msi, IPROC_MSI_CTRL, eq);
>> +		val &= ~(IPROC_MSI_INTR_EN | IPROC_MSI_INT_N_EVENT |
>> +			 IPROC_MSI_EQ_EN);
>> +		iproc_msi_write_reg(msi, IPROC_MSI_CTRL, eq, val);
>> +	}
>> +}
>> +
>> +static int iproc_msi_alloc_domains(struct device_node *node,
>> +				   struct iproc_msi *msi)
>> +{
>> +	msi->inner_domain = irq_domain_add_linear(NULL, msi->nr_msi_vecs,
>> +						  &msi_domain_ops, msi);
>> +	if (!msi->inner_domain)
>> +		return -ENOMEM;
>> +
>> +	msi->msi_domain = pci_msi_create_irq_domain(of_node_to_fwnode(node),
>> +						    &iproc_msi_domain_info,
>> +						    msi->inner_domain);
>> +	if (!msi->msi_domain) {
>> +		irq_domain_remove(msi->inner_domain);
>> +		return -ENOMEM;
>> +	}
>> +
>> +	return 0;
>> +}
>> +
>> +static void iproc_msi_free_domains(struct iproc_msi *msi)
>> +{
>> +	if (msi->msi_domain)
>> +		irq_domain_remove(msi->msi_domain);
>> +
>> +	if (msi->inner_domain)
>> +		irq_domain_remove(msi->inner_domain);
>> +}
>> +
>> +static int iproc_msi_irq_setup(struct iproc_msi *msi, unsigned int cpu)
>> +{
>> +	int i, ret;
>> +	cpumask_var_t mask;
>> +	struct iproc_pcie *pcie = msi->pcie;
>> +
>> +	for (i = cpu; i < msi->nr_irqs; i += msi->nr_cpus) {
>> +		irq_set_chained_handler_and_data(msi->grps[i].gic_irq,
>> +						 iproc_msi_handler,
>> +						 &msi->grps[i]);
>> +		/* dedicate GIC interrupt to each CPU core */
>> +		if (alloc_cpumask_var(&mask, GFP_KERNEL)) {
>> +			cpumask_clear(mask);
>> +			cpumask_set_cpu(cpu, mask);
>> +			ret = irq_set_affinity(msi->grps[i].gic_irq, mask);
>> +			if (ret)
>> +				dev_err(pcie->dev,
>> +					"failed to set affinity for IRQ%d\n",
>> +					msi->grps[i].gic_irq);
>> +			free_cpumask_var(mask);
>> +		} else {
>> +			dev_err(pcie->dev, "failed to alloc CPU mask\n");
>> +			ret = -EINVAL;
>> +		}
>> +
>> +		if (ret) {
>> +			irq_set_chained_handler_and_data(msi->grps[i].gic_irq,
>> +							 NULL, NULL);
>> +			return ret;
>
> What happens to interrupts you've already configured? I'd expect a full
> rollback.
>

Yeah in the error case, need to roll back all previously configured irq 
by the time we exit this function. Will fix!

>> +		}
>> +	}
>> +
>> +	return 0;
>> +}
>> +
>> +static void iproc_msi_irq_free(struct iproc_msi *msi, unsigned int cpu)
>> +{
>> +	int i;
>> +
>> +	for (i = cpu; i < msi->nr_irqs; i += msi->nr_cpus) {
>> +		irq_set_chained_handler_and_data(msi->grps[i].gic_irq,
>> +						 NULL, NULL);
>> +	}
>> +}
>> +
>> +int iproc_msi_init(struct iproc_pcie *pcie, struct device_node *node)
>> +{
>> +	struct iproc_msi *msi;
>> +	int i, ret;
>> +	unsigned int cpu;
>> +
>> +	if (!of_device_is_compatible(node, "brcm,iproc-msi"))
>> +		return -ENODEV;
>> +
>> +	if (!of_find_property(node, "msi-controller", NULL))
>> +		return -ENODEV;
>> +
>> +	if (pcie->msi)
>> +		return -EBUSY;
>> +
>> +	msi = devm_kzalloc(pcie->dev, sizeof(*msi), GFP_KERNEL);
>> +	if (!msi)
>> +		return -ENOMEM;
>> +
>> +	msi->pcie = pcie;
>> +	pcie->msi = msi;
>> +	mutex_init(&msi->bitmap_lock);
>> +	msi->nr_cpus = num_possible_cpus();
>> +
>> +	switch (pcie->type) {
>> +	case IPROC_PCIE_PAXB:
>> +		msi->reg_offsets = iproc_msi_reg_paxb;
>> +		break;
>> +	case IPROC_PCIE_PAXC:
>> +		msi->reg_offsets = iproc_msi_reg_paxc;
>> +		break;
>> +	default:
>> +		dev_err(pcie->dev, "incompatible iProc PCIe interface\n");
>> +		return -EINVAL;
>> +	}
>> +
>> +	if (of_find_property(node, "brcm,pcie-msi-inten", NULL))
>> +		msi->has_inten_reg = true;
>> +
>> +	ret = of_property_read_u32(node, "brcm,num-eq-region",
>> +				   &msi->nr_eq_region);
>> +	if (ret || !msi->nr_eq_region)
>> +		msi->nr_eq_region = 1;
>> +
>> +	ret = of_property_read_u32(node, "brcm,num-msi-msg-region",
>> +				   &msi->nr_msi_region);
>> +	if (ret || !msi->nr_msi_region)
>> +		msi->nr_msi_region = 1;
>> +
>> +	msi->nr_irqs = of_irq_count(node);
>> +	if (!msi->nr_irqs) {
>> +		dev_err(pcie->dev, "found no MSI GIC interrupt\n");
>> +		return -ENODEV;
>> +	}
>> +	if (msi->nr_irqs > NR_HW_IRQS) {
>> +		dev_warn(pcie->dev, "too many MSI GIC interrupts defined %d\n",
>> +			 msi->nr_irqs);
>> +		msi->nr_irqs = NR_HW_IRQS;
>> +	}
>> +
>> +	msi->nr_msi_vecs = msi->nr_irqs * EQ_LEN;
>> +	msi->bitmap = devm_kcalloc(pcie->dev, BITS_TO_LONGS(msi->nr_msi_vecs),
>> +				   sizeof(*msi->bitmap), GFP_KERNEL);
>> +	if (!msi->bitmap)
>> +		return -ENOMEM;
>> +
>> +	msi->grps = devm_kcalloc(pcie->dev, msi->nr_irqs, sizeof(*msi->grps),
>> +				 GFP_KERNEL);
>> +	if (!msi->grps)
>> +		return -ENOMEM;
>> +
>> +	for (i = 0; i < msi->nr_irqs; i++) {
>> +		unsigned int irq = irq_of_parse_and_map(node, i);
>> +
>> +		if (!irq) {
>> +			dev_err(pcie->dev, "unable to parse/map interrupt\n");
>> +			return -ENODEV;
>> +		}
>> +		msi->grps[i].gic_irq = irq;
>> +		msi->grps[i].msi = msi;
>> +		msi->grps[i].eq = i;
>> +	}
>> +
>> +	/* reserve memory for MSI event queue */
>> +	msi->eq_cpu = dma_alloc_coherent(pcie->dev,
>> +					 msi->nr_eq_region * EQ_MEM_REGION_SIZE,
>> +					 &msi->eq_dma, GFP_KERNEL);
>
> Do you need to zero that memory? Or is the HW happy with whatever will
> be there?
>

It seems to work fine. But you are right, it's much safer to zero the 
memory since the PCIe controller could be using the upper 16-bit of each 
word-sized entry in the event queue.

>> +	if (!msi->eq_cpu)
>> +		return -ENOMEM;
>> +
>> +	/* reserve memory for MSI posted writes */
>> +	msi->msi_cpu = dma_alloc_coherent(pcie->dev,
>> +					  msi->nr_msi_region * MSI_MEM_REGION_SIZE,
>> +					  &msi->msi_dma, GFP_KERNEL);
>
> Same here. Also, what is the exact purpose of that memory? You have a
> coherent mapping with the CPU, but you never read from it. So what's
> the point?
>

Yeah I guess I can change this back to kmalloc since coherent memory is 
a scarce resource, and the CPU does not need to access the memory, so 
there's no cache issue.

I know I have not answered the first part of your question. Let me do 
some experiments first and I'll get back to you on that, :)

>> +	if (!msi->msi_cpu) {
>> +		ret = -ENOMEM;
>> +		goto free_eq_dma;
>> +	}
>> +
>> +	ret = iproc_msi_alloc_domains(node, msi);
>> +	if (ret) {
>> +		dev_err(pcie->dev, "failed to create MSI domains\n");
>> +		goto free_msi_dma;
>> +	}
>> +
>> +	for_each_online_cpu(cpu) {
>> +		ret = iproc_msi_irq_setup(msi, cpu);
>> +		if (ret)
>> +			goto free_msi_irq;
>> +	}
>> +
>> +	iproc_msi_enable(msi);
>> +
>> +	return 0;
>> +
>> +free_msi_irq:
>> +	for_each_online_cpu(cpu)
>> +		iproc_msi_irq_free(msi, cpu);
>> +	iproc_msi_free_domains(msi);
>> +
>> +free_msi_dma:
>> +	dma_free_coherent(pcie->dev, msi->nr_msi_region * MSI_MEM_REGION_SIZE,
>> +			  msi->msi_cpu, msi->msi_dma);
>> +free_eq_dma:
>> +	dma_free_coherent(pcie->dev, msi->nr_eq_region * EQ_MEM_REGION_SIZE,
>> +			  msi->eq_cpu, msi->eq_dma);
>> +	pcie->msi = NULL;
>> +	return ret;
>> +}
>> +EXPORT_SYMBOL(iproc_msi_init);
>
> [...]
>
> Thanks,
>
> 	M.
>

Thanks, Marc!

Ray
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Ray Jui Nov. 26, 2015, 5:14 a.m. UTC | #3
Hi Marc,

On 11/25/2015 5:52 PM, Ray Jui wrote:
> Hi Marc,
>
[...]
>>> +    /* reserve memory for MSI posted writes */
>>> +    msi->msi_cpu = dma_alloc_coherent(pcie->dev,
>>> +                      msi->nr_msi_region * MSI_MEM_REGION_SIZE,
>>> +                      &msi->msi_dma, GFP_KERNEL);
>>
>> Same here. Also, what is the exact purpose of that memory? You have a
>> coherent mapping with the CPU, but you never read from it. So what's
>> the point?
>>
>
> Yeah I guess I can change this back to kmalloc since coherent memory is
> a scarce resource, and the CPU does not need to access the memory, so
> there's no cache issue.
>
> I know I have not answered the first part of your question. Let me do
> some experiments first and I'll get back to you on that, :)
>

I did some experiment with the msi_dma here. It looks like it can be any 
address as long as it's 4K aligned (i.e., can be from the device address 
range instead of the RAM address range). The MSI message data actually 
goes to the memory allocated for the event queue (makes sense...), and 
never made it to the MSI page memory allocated here. Our arch doc is 
just confusing....:(

I saw your comment on the other email thread with Xilinx MSI. I'll set 
the address to the base address of the iProc PCIe controller (which is 
always 4K aligned).

[...]
>>
>> Thanks,
>>
>>     M.
>>
>
> Thanks, Marc!
>
> Ray

Thanks again!

Ray
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/drivers/pci/host/Kconfig b/drivers/pci/host/Kconfig
index f131ba9..972e906 100644
--- a/drivers/pci/host/Kconfig
+++ b/drivers/pci/host/Kconfig
@@ -126,6 +126,15 @@  config PCIE_IPROC
 	  iProc family of SoCs. An appropriate bus interface driver also needs
 	  to be enabled
 
+config PCIE_IPROC_MSI
+	bool "Broadcom iProc PCIe MSI support"
+	depends on ARCH_BCM_IPROC && PCI_MSI
+	select PCI_MSI_IRQ_DOMAIN
+	default ARCH_BCM_IPROC
+	help
+	  Say Y here if you want to enable MSI support for Broadcom's iProc
+	  PCIe controller
+
 config PCIE_IPROC_PLATFORM
 	tristate "Broadcom iProc PCIe platform bus driver"
 	depends on ARCH_BCM_IPROC || (ARM && COMPILE_TEST)
diff --git a/drivers/pci/host/Makefile b/drivers/pci/host/Makefile
index 9d4d3c6..0e4e95e 100644
--- a/drivers/pci/host/Makefile
+++ b/drivers/pci/host/Makefile
@@ -15,6 +15,7 @@  obj-$(CONFIG_PCI_XGENE_MSI) += pci-xgene-msi.o
 obj-$(CONFIG_PCI_LAYERSCAPE) += pci-layerscape.o
 obj-$(CONFIG_PCI_VERSATILE) += pci-versatile.o
 obj-$(CONFIG_PCIE_IPROC) += pcie-iproc.o
+obj-$(CONFIG_PCIE_IPROC_MSI) += pcie-iproc-msi.o
 obj-$(CONFIG_PCIE_IPROC_PLATFORM) += pcie-iproc-platform.o
 obj-$(CONFIG_PCIE_IPROC_BCMA) += pcie-iproc-bcma.o
 obj-$(CONFIG_PCIE_ALTERA) += pcie-altera.o
diff --git a/drivers/pci/host/pcie-iproc-msi.c b/drivers/pci/host/pcie-iproc-msi.c
new file mode 100644
index 0000000..afc54c2
--- /dev/null
+++ b/drivers/pci/host/pcie-iproc-msi.c
@@ -0,0 +1,662 @@ 
+/*
+ * Copyright (C) 2015 Broadcom Corporation
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation version 2.
+ *
+ * This program is distributed "as is" WITHOUT ANY WARRANTY of any
+ * kind, whether express or implied; without even the implied warranty
+ * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#include <linux/interrupt.h>
+#include <linux/irqchip/chained_irq.h>
+#include <linux/irqdomain.h>
+#include <linux/msi.h>
+#include <linux/of_irq.h>
+#include <linux/of_pci.h>
+#include <linux/pci.h>
+
+#include "pcie-iproc.h"
+
+#define IPROC_MSI_INTR_EN_SHIFT        11
+#define IPROC_MSI_INTR_EN              BIT(IPROC_MSI_INTR_EN_SHIFT)
+#define IPROC_MSI_INT_N_EVENT_SHIFT    1
+#define IPROC_MSI_INT_N_EVENT          BIT(IPROC_MSI_INT_N_EVENT_SHIFT)
+#define IPROC_MSI_EQ_EN_SHIFT          0
+#define IPROC_MSI_EQ_EN                BIT(IPROC_MSI_EQ_EN_SHIFT)
+
+#define IPROC_MSI_EQ_MASK              0x3f
+
+/* max number of GIC interrupts */
+#define NR_HW_IRQS                     6
+
+/* number of entries in each event queue */
+#define EQ_LEN                         64
+
+/* size of each event queue memory region */
+#define EQ_MEM_REGION_SIZE             SZ_4K
+
+/* size of each MSI message memory region */
+#define MSI_MEM_REGION_SIZE            SZ_4K
+
+enum iproc_msi_reg {
+	IPROC_MSI_EQ_PAGE = 0,
+	IPROC_MSI_EQ_PAGE_UPPER,
+	IPROC_MSI_PAGE,
+	IPROC_MSI_PAGE_UPPER,
+	IPROC_MSI_CTRL,
+	IPROC_MSI_EQ_HEAD,
+	IPROC_MSI_EQ_TAIL,
+	IPROC_MSI_INTS_EN,
+	IPROC_MSI_REG_SIZE,
+};
+
+struct iproc_msi;
+
+/**
+ * iProc MSI group
+ *
+ * One MSI group is allocated per GIC interrupt, serviced by one iProc MSI
+ * event queue
+ *
+ * @msi: pointer to iProc MSI data
+ * @gic_irq: GIC interrupt
+ * @eq: Event queue number
+ */
+struct iproc_msi_grp {
+	struct iproc_msi *msi;
+	int gic_irq;
+	unsigned int eq;
+};
+
+/**
+ * iProc event queue based MSI
+ *
+ * Only meant to be used on platforms without MSI support integrated into the
+ * GIC
+ *
+ * @pcie: pointer to iProc PCIe data
+ * @reg_offsets: MSI register offsets
+ * @grps: MSI groups
+ * @nr_irqs: number of total interrupts connected to GIC
+ * @nr_cpus: number of toal CPUs
+ * @has_inten_reg: indicates the MSI interrupt enable register needs to be
+ * set explicitly (required for some legacy platforms)
+ * @bitmap: MSI vector bitmap
+ * @bitmap_lock: lock to protect access to the MSI bitmap
+ * @nr_msi_vecs: total number of MSI vectors
+ * @inner_domain: inner IRQ domain
+ * @msi_domain: MSI IRQ domain
+ * @nr_eq_region: required number of 4K aligned memory region for MSI event
+ * queues
+ * @nr_msi_region: required number of 4K aligned memory region for MSI posted
+ * writes
+ * @eq_cpu: pointer to allocated memory region for MSI event queues
+ * @eq_dma: DMA address of MSI event queues
+ * @msi_cpu: pointer to allocated memory region for MSI posted writes
+ * @msi_dma: DMA address of MSI posted writes
+ */
+struct iproc_msi {
+	struct iproc_pcie *pcie;
+	const u16 (*reg_offsets)[IPROC_MSI_REG_SIZE];
+	struct iproc_msi_grp *grps;
+	int nr_irqs;
+	int nr_cpus;
+	bool has_inten_reg;
+	unsigned long *bitmap;
+	struct mutex bitmap_lock;
+	unsigned int nr_msi_vecs;
+	struct irq_domain *inner_domain;
+	struct irq_domain *msi_domain;
+	unsigned int nr_eq_region;
+	unsigned int nr_msi_region;
+	void *eq_cpu;
+	dma_addr_t eq_dma;
+	void *msi_cpu;
+	dma_addr_t msi_dma;
+};
+
+static const u16 iproc_msi_reg_paxb[NR_HW_IRQS][IPROC_MSI_REG_SIZE] = {
+	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x210, 0x250, 0x254, 0x208 },
+	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x214, 0x258, 0x25c, 0x208 },
+	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x218, 0x260, 0x264, 0x208 },
+	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x21c, 0x268, 0x26c, 0x208 },
+	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x220, 0x270, 0x274, 0x208 },
+	{ 0x200, 0x2c0, 0x204, 0x2c4, 0x224, 0x278, 0x27c, 0x208 },
+};
+
+static const u16 iproc_msi_reg_paxc[NR_HW_IRQS][IPROC_MSI_REG_SIZE] = {
+	{ 0xc00, 0xc04, 0xc08, 0xc0c, 0xc40, 0xc50, 0xc60 },
+	{ 0xc10, 0xc14, 0xc18, 0xc1c, 0xc44, 0xc54, 0xc64 },
+	{ 0xc20, 0xc24, 0xc28, 0xc2c, 0xc48, 0xc58, 0xc68 },
+	{ 0xc30, 0xc34, 0xc38, 0xc3c, 0xc4c, 0xc5c, 0xc6c },
+};
+
+static inline u32 iproc_msi_read_reg(struct iproc_msi *msi,
+				     enum iproc_msi_reg reg,
+				     unsigned int eq)
+{
+	struct iproc_pcie *pcie = msi->pcie;
+
+	return readl_relaxed(pcie->base + msi->reg_offsets[eq][reg]);
+}
+
+static inline void iproc_msi_write_reg(struct iproc_msi *msi,
+				       enum iproc_msi_reg reg,
+				       int eq, u32 val)
+{
+	struct iproc_pcie *pcie = msi->pcie;
+
+	writel_relaxed(val, pcie->base + msi->reg_offsets[eq][reg]);
+}
+
+static inline bool iproc_msi_has_mult_regions(struct iproc_msi *msi)
+{
+	return ((msi->nr_msi_region > 1) ? true : false);
+}
+
+static inline bool iproc_eq_has_mult_regions(struct iproc_msi *msi)
+{
+	return ((msi->nr_eq_region > 1) ? true : false);
+}
+
+static struct irq_chip iproc_msi_irq_chip = {
+	.name = "iProc-MSI",
+};
+
+static struct msi_domain_info iproc_msi_domain_info = {
+	.flags = MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
+		MSI_FLAG_PCI_MSIX,
+	.chip = &iproc_msi_irq_chip,
+};
+
+/*
+ * In iProc PCIe core, each MSI group is serviced by a GIC interrupt and a
+ * dedicated event queue. Each MSI group can support up to 64 MSI vectors
+ *
+ * The number of MSI groups varies between different iProc SoCs. The total
+ * number of CPU cores also varies. To support MSI IRQ affinity, we
+ * distribute GIC interrupts across all available CPUs. MSI vector is moved
+ * from one GIC interrupt to another to steer to the target CPU
+ *
+ * Assuming:
+ * - the number of MSI groups is M
+ * - the number of CPU cores is N
+ * - M is always a multiple of N
+ *
+ * Total number of raw MSI vectors = M * 64
+ * Total number of supported MSI vectors = (M * 64) / N
+ */
+static inline int hwirq_to_cpu(struct iproc_msi *msi, unsigned long hwirq)
+{
+	return (hwirq % msi->nr_cpus);
+}
+
+static inline unsigned long hwirq_to_canonical_hwirq(struct iproc_msi *msi,
+						     unsigned long hwirq)
+{
+	return (hwirq - hwirq_to_cpu(msi, hwirq));
+}
+
+static int iproc_msi_irq_set_affinity(struct irq_data *data,
+				      const struct cpumask *mask, bool force)
+{
+	struct iproc_msi *msi = irq_data_get_irq_chip_data(data);
+	int target_cpu = cpumask_first(mask);
+	int curr_cpu;
+
+	curr_cpu = hwirq_to_cpu(msi, data->hwirq);
+	if (curr_cpu == target_cpu)
+		return IRQ_SET_MASK_OK_DONE;
+
+	/* steer MSI to the target CPU */
+	data->hwirq = hwirq_to_canonical_hwirq(msi, data->hwirq) + target_cpu;
+
+	return IRQ_SET_MASK_OK;
+}
+
+static inline u32 hwirq_to_group(struct iproc_msi *msi, unsigned long hwirq)
+{
+	return (hwirq % msi->nr_irqs);
+}
+
+static void iproc_msi_irq_compose_msi_msg(struct irq_data *data,
+					  struct msi_msg *msg)
+{
+	struct iproc_msi *msi = irq_data_get_irq_chip_data(data);
+	dma_addr_t addr;
+	unsigned int mul;
+
+	if (iproc_msi_has_mult_regions(msi))
+		mul = MSI_MEM_REGION_SIZE;
+	else
+		mul = sizeof(u32);
+
+	addr = msi->msi_dma + hwirq_to_group(msi, data->hwirq) * mul;
+	msg->address_lo = lower_32_bits(addr);
+	msg->address_hi = upper_32_bits(addr);
+	msg->data = data->hwirq;
+}
+
+static struct irq_chip iproc_msi_bottom_irq_chip = {
+	.name = "MSI",
+	.irq_set_affinity = iproc_msi_irq_set_affinity,
+	.irq_compose_msi_msg = iproc_msi_irq_compose_msi_msg,
+};
+
+static int iproc_msi_irq_domain_alloc(struct irq_domain *domain,
+				      unsigned int virq, unsigned int nr_irqs,
+				      void *args)
+{
+	struct iproc_msi *msi = domain->host_data;
+	int hwirq;
+
+	mutex_lock(&msi->bitmap_lock);
+
+	/* allocate 'nr_cpus' number of MSI vectors each time */
+	hwirq = bitmap_find_next_zero_area(msi->bitmap, msi->nr_msi_vecs, 0,
+					   msi->nr_cpus, 0);
+	if (hwirq < msi->nr_msi_vecs)
+		bitmap_set(msi->bitmap, hwirq, msi->nr_cpus);
+	else
+		return -ENOSPC;
+
+	mutex_unlock(&msi->bitmap_lock);
+
+	irq_domain_set_info(domain, virq, hwirq, &iproc_msi_bottom_irq_chip,
+			    domain->host_data, handle_simple_irq, NULL, NULL);
+
+	return 0;
+}
+
+static void iproc_msi_irq_domain_free(struct irq_domain *domain,
+				      unsigned int virq, unsigned int nr_irqs)
+{
+	struct irq_data *data = irq_domain_get_irq_data(domain, virq);
+	struct iproc_msi *msi = irq_data_get_irq_chip_data(data);
+	unsigned int hwirq;
+
+	mutex_lock(&msi->bitmap_lock);
+
+	hwirq = hwirq_to_canonical_hwirq(msi, data->hwirq);
+	bitmap_clear(msi->bitmap, hwirq, msi->nr_cpus);
+
+	mutex_unlock(&msi->bitmap_lock);
+
+	irq_domain_free_irqs_parent(domain, virq, nr_irqs);
+}
+
+static const struct irq_domain_ops msi_domain_ops = {
+	.alloc = iproc_msi_irq_domain_alloc,
+	.free = iproc_msi_irq_domain_free,
+};
+
+static inline u32 decode_msi_hwirq(struct iproc_msi *msi, u32 eq, u32 head)
+{
+	u32 *msg, hwirq;
+	unsigned int offs;
+
+	if (iproc_eq_has_mult_regions(msi))
+		offs = eq * EQ_MEM_REGION_SIZE;
+	else
+		offs = eq * EQ_LEN * sizeof(u32);
+
+	offs += head * sizeof(u32);
+	msg = (u32 *)(msi->eq_cpu + offs);
+	hwirq = *msg & IPROC_MSI_EQ_MASK;
+
+	/*
+	 * Since we have multiple hwirq mapped to a single MSI vector,
+	 * now we need to derive the hwirq at CPU0. It can then be used to
+	 * mapped back to virq
+	 */
+	return hwirq_to_canonical_hwirq(msi, hwirq);
+}
+
+static void iproc_msi_handler(struct irq_desc *desc)
+{
+	struct irq_chip *chip = irq_desc_get_chip(desc);
+	struct iproc_msi_grp *grp;
+	struct iproc_msi *msi;
+	struct iproc_pcie *pcie;
+	u32 eq, head, tail, nr_events;
+	unsigned long hwirq;
+	int virq;
+
+	chained_irq_enter(chip, desc);
+
+	grp = irq_desc_get_handler_data(desc);
+	msi = grp->msi;
+	pcie = msi->pcie;
+	eq = grp->eq;
+
+	/*
+	 * iProc MSI event queue is tracked by head and tail pointers. Head
+	 * pointer indicates the next entry to be processed by SW in the
+	 * queue. Entries between head and tail pointers contain valid MSI
+	 * data to be processed
+	 */
+	head = iproc_msi_read_reg(msi, IPROC_MSI_EQ_HEAD,
+				  eq) & IPROC_MSI_EQ_MASK;
+	do {
+		tail = iproc_msi_read_reg(msi, IPROC_MSI_EQ_TAIL,
+					  eq) & IPROC_MSI_EQ_MASK;
+
+		/*
+		 * Figure out total number of events (MSI data) to be
+		 * processed
+		 */
+		nr_events = (tail < head) ?
+			(EQ_LEN - (head - tail)) : (tail - head);
+		if (!nr_events)
+			break;
+
+		/* process all outstanding events */
+		while (nr_events--) {
+			hwirq = decode_msi_hwirq(msi, eq, head);
+			virq = irq_find_mapping(msi->inner_domain, hwirq);
+			generic_handle_irq(virq);
+
+			head++;
+			head %= EQ_LEN;
+			iproc_msi_write_reg(msi, IPROC_MSI_EQ_HEAD, eq, head);
+		}
+
+		/*
+		 * Now go read the tail pointer again to see if there are new
+		 * oustanding events that came in during the above window
+		 */
+	} while (true);
+
+	chained_irq_exit(chip, desc);
+}
+
+static void iproc_msi_enable(struct iproc_msi *msi)
+{
+	int i, eq;
+	u32 val;
+
+	/* program memory region for each event queue */
+	for (i = 0; i < msi->nr_eq_region; i++) {
+		dma_addr_t addr = msi->eq_dma + (i * EQ_MEM_REGION_SIZE);
+
+		iproc_msi_write_reg(msi, IPROC_MSI_EQ_PAGE, i,
+				    lower_32_bits(addr));
+		iproc_msi_write_reg(msi, IPROC_MSI_EQ_PAGE_UPPER, i,
+				    upper_32_bits(addr));
+	}
+
+	/* program memory region for MSI posted writes */
+	for (i = 0; i < msi->nr_msi_region; i++) {
+		dma_addr_t addr = msi->msi_dma + (i * MSI_MEM_REGION_SIZE);
+
+		iproc_msi_write_reg(msi, IPROC_MSI_PAGE, i,
+				    lower_32_bits(addr));
+		iproc_msi_write_reg(msi, IPROC_MSI_PAGE_UPPER, i,
+				    upper_32_bits(addr));
+	}
+
+	for (eq = 0; eq < msi->nr_irqs; eq++) {
+		/* enable MSI event queue */
+		val = IPROC_MSI_INTR_EN | IPROC_MSI_INT_N_EVENT |
+			IPROC_MSI_EQ_EN;
+		iproc_msi_write_reg(msi, IPROC_MSI_CTRL, eq, val);
+
+		/*
+		 * Some legacy platforms require the MSI interrupt enable
+		 * register to be set explicitly
+		 */
+		if (msi->has_inten_reg) {
+			val = iproc_msi_read_reg(msi, IPROC_MSI_INTS_EN, eq);
+			val |= BIT(eq);
+			iproc_msi_write_reg(msi, IPROC_MSI_INTS_EN, eq, val);
+		}
+	}
+}
+
+static void iproc_msi_disable(struct iproc_msi *msi)
+{
+	u32 eq, val;
+
+	for (eq = 0; eq < msi->nr_irqs; eq++) {
+		if (msi->has_inten_reg) {
+			val = iproc_msi_read_reg(msi, IPROC_MSI_INTS_EN, eq);
+			val &= ~BIT(eq);
+			iproc_msi_write_reg(msi, IPROC_MSI_INTS_EN, eq, val);
+		}
+
+		val = iproc_msi_read_reg(msi, IPROC_MSI_CTRL, eq);
+		val &= ~(IPROC_MSI_INTR_EN | IPROC_MSI_INT_N_EVENT |
+			 IPROC_MSI_EQ_EN);
+		iproc_msi_write_reg(msi, IPROC_MSI_CTRL, eq, val);
+	}
+}
+
+static int iproc_msi_alloc_domains(struct device_node *node,
+				   struct iproc_msi *msi)
+{
+	msi->inner_domain = irq_domain_add_linear(NULL, msi->nr_msi_vecs,
+						  &msi_domain_ops, msi);
+	if (!msi->inner_domain)
+		return -ENOMEM;
+
+	msi->msi_domain = pci_msi_create_irq_domain(of_node_to_fwnode(node),
+						    &iproc_msi_domain_info,
+						    msi->inner_domain);
+	if (!msi->msi_domain) {
+		irq_domain_remove(msi->inner_domain);
+		return -ENOMEM;
+	}
+
+	return 0;
+}
+
+static void iproc_msi_free_domains(struct iproc_msi *msi)
+{
+	if (msi->msi_domain)
+		irq_domain_remove(msi->msi_domain);
+
+	if (msi->inner_domain)
+		irq_domain_remove(msi->inner_domain);
+}
+
+static int iproc_msi_irq_setup(struct iproc_msi *msi, unsigned int cpu)
+{
+	int i, ret;
+	cpumask_var_t mask;
+	struct iproc_pcie *pcie = msi->pcie;
+
+	for (i = cpu; i < msi->nr_irqs; i += msi->nr_cpus) {
+		irq_set_chained_handler_and_data(msi->grps[i].gic_irq,
+						 iproc_msi_handler,
+						 &msi->grps[i]);
+		/* dedicate GIC interrupt to each CPU core */
+		if (alloc_cpumask_var(&mask, GFP_KERNEL)) {
+			cpumask_clear(mask);
+			cpumask_set_cpu(cpu, mask);
+			ret = irq_set_affinity(msi->grps[i].gic_irq, mask);
+			if (ret)
+				dev_err(pcie->dev,
+					"failed to set affinity for IRQ%d\n",
+					msi->grps[i].gic_irq);
+			free_cpumask_var(mask);
+		} else {
+			dev_err(pcie->dev, "failed to alloc CPU mask\n");
+			ret = -EINVAL;
+		}
+
+		if (ret) {
+			irq_set_chained_handler_and_data(msi->grps[i].gic_irq,
+							 NULL, NULL);
+			return ret;
+		}
+	}
+
+	return 0;
+}
+
+static void iproc_msi_irq_free(struct iproc_msi *msi, unsigned int cpu)
+{
+	int i;
+
+	for (i = cpu; i < msi->nr_irqs; i += msi->nr_cpus) {
+		irq_set_chained_handler_and_data(msi->grps[i].gic_irq,
+						 NULL, NULL);
+	}
+}
+
+int iproc_msi_init(struct iproc_pcie *pcie, struct device_node *node)
+{
+	struct iproc_msi *msi;
+	int i, ret;
+	unsigned int cpu;
+
+	if (!of_device_is_compatible(node, "brcm,iproc-msi"))
+		return -ENODEV;
+
+	if (!of_find_property(node, "msi-controller", NULL))
+		return -ENODEV;
+
+	if (pcie->msi)
+		return -EBUSY;
+
+	msi = devm_kzalloc(pcie->dev, sizeof(*msi), GFP_KERNEL);
+	if (!msi)
+		return -ENOMEM;
+
+	msi->pcie = pcie;
+	pcie->msi = msi;
+	mutex_init(&msi->bitmap_lock);
+	msi->nr_cpus = num_possible_cpus();
+
+	switch (pcie->type) {
+	case IPROC_PCIE_PAXB:
+		msi->reg_offsets = iproc_msi_reg_paxb;
+		break;
+	case IPROC_PCIE_PAXC:
+		msi->reg_offsets = iproc_msi_reg_paxc;
+		break;
+	default:
+		dev_err(pcie->dev, "incompatible iProc PCIe interface\n");
+		return -EINVAL;
+	}
+
+	if (of_find_property(node, "brcm,pcie-msi-inten", NULL))
+		msi->has_inten_reg = true;
+
+	ret = of_property_read_u32(node, "brcm,num-eq-region",
+				   &msi->nr_eq_region);
+	if (ret || !msi->nr_eq_region)
+		msi->nr_eq_region = 1;
+
+	ret = of_property_read_u32(node, "brcm,num-msi-msg-region",
+				   &msi->nr_msi_region);
+	if (ret || !msi->nr_msi_region)
+		msi->nr_msi_region = 1;
+
+	msi->nr_irqs = of_irq_count(node);
+	if (!msi->nr_irqs) {
+		dev_err(pcie->dev, "found no MSI GIC interrupt\n");
+		return -ENODEV;
+	}
+	if (msi->nr_irqs > NR_HW_IRQS) {
+		dev_warn(pcie->dev, "too many MSI GIC interrupts defined %d\n",
+			 msi->nr_irqs);
+		msi->nr_irqs = NR_HW_IRQS;
+	}
+
+	msi->nr_msi_vecs = msi->nr_irqs * EQ_LEN;
+	msi->bitmap = devm_kcalloc(pcie->dev, BITS_TO_LONGS(msi->nr_msi_vecs),
+				   sizeof(*msi->bitmap), GFP_KERNEL);
+	if (!msi->bitmap)
+		return -ENOMEM;
+
+	msi->grps = devm_kcalloc(pcie->dev, msi->nr_irqs, sizeof(*msi->grps),
+				 GFP_KERNEL);
+	if (!msi->grps)
+		return -ENOMEM;
+
+	for (i = 0; i < msi->nr_irqs; i++) {
+		unsigned int irq = irq_of_parse_and_map(node, i);
+
+		if (!irq) {
+			dev_err(pcie->dev, "unable to parse/map interrupt\n");
+			return -ENODEV;
+		}
+		msi->grps[i].gic_irq = irq;
+		msi->grps[i].msi = msi;
+		msi->grps[i].eq = i;
+	}
+
+	/* reserve memory for MSI event queue */
+	msi->eq_cpu = dma_alloc_coherent(pcie->dev,
+					 msi->nr_eq_region * EQ_MEM_REGION_SIZE,
+					 &msi->eq_dma, GFP_KERNEL);
+	if (!msi->eq_cpu)
+		return -ENOMEM;
+
+	/* reserve memory for MSI posted writes */
+	msi->msi_cpu = dma_alloc_coherent(pcie->dev,
+					  msi->nr_msi_region * MSI_MEM_REGION_SIZE,
+					  &msi->msi_dma, GFP_KERNEL);
+	if (!msi->msi_cpu) {
+		ret = -ENOMEM;
+		goto free_eq_dma;
+	}
+
+	ret = iproc_msi_alloc_domains(node, msi);
+	if (ret) {
+		dev_err(pcie->dev, "failed to create MSI domains\n");
+		goto free_msi_dma;
+	}
+
+	for_each_online_cpu(cpu) {
+		ret = iproc_msi_irq_setup(msi, cpu);
+		if (ret)
+			goto free_msi_irq;
+	}
+
+	iproc_msi_enable(msi);
+
+	return 0;
+
+free_msi_irq:
+	for_each_online_cpu(cpu)
+		iproc_msi_irq_free(msi, cpu);
+	iproc_msi_free_domains(msi);
+
+free_msi_dma:
+	dma_free_coherent(pcie->dev, msi->nr_msi_region * MSI_MEM_REGION_SIZE,
+			  msi->msi_cpu, msi->msi_dma);
+free_eq_dma:
+	dma_free_coherent(pcie->dev, msi->nr_eq_region * EQ_MEM_REGION_SIZE,
+			  msi->eq_cpu, msi->eq_dma);
+	pcie->msi = NULL;
+	return ret;
+}
+EXPORT_SYMBOL(iproc_msi_init);
+
+void iproc_msi_exit(struct iproc_pcie *pcie)
+{
+	struct iproc_msi *msi = pcie->msi;
+	unsigned int cpu;
+
+	if (!msi)
+		return;
+
+	iproc_msi_disable(msi);
+
+	for_each_online_cpu(cpu)
+		iproc_msi_irq_free(msi, cpu);
+
+	iproc_msi_free_domains(msi);
+
+	dma_free_coherent(pcie->dev, msi->nr_msi_region * MSI_MEM_REGION_SIZE,
+			  msi->msi_cpu, msi->msi_dma);
+	dma_free_coherent(pcie->dev, msi->nr_eq_region * EQ_MEM_REGION_SIZE,
+			  msi->eq_cpu, msi->eq_dma);
+}
+EXPORT_SYMBOL(iproc_msi_exit);
diff --git a/drivers/pci/host/pcie-iproc.c b/drivers/pci/host/pcie-iproc.c
index 24d5b62..b7fda356 100644
--- a/drivers/pci/host/pcie-iproc.c
+++ b/drivers/pci/host/pcie-iproc.c
@@ -440,6 +440,26 @@  static int iproc_pcie_map_ranges(struct iproc_pcie *pcie,
 	return 0;
 }
 
+static int iproc_pcie_msi_enable(struct iproc_pcie *pcie)
+{
+	struct device_node *msi_node;
+
+	msi_node = of_parse_phandle(pcie->dev->of_node, "msi-parent", 0);
+	if (!msi_node)
+		return -ENODEV;
+
+	/*
+	 * If another MSI controller is being used, the call below should fail
+	 * but that is okay
+	 */
+	return iproc_msi_init(pcie, msi_node);
+}
+
+static void iproc_pcie_msi_disable(struct iproc_pcie *pcie)
+{
+	iproc_msi_exit(pcie);
+}
+
 int iproc_pcie_setup(struct iproc_pcie *pcie, struct list_head *res)
 {
 	int ret;
@@ -507,6 +527,10 @@  int iproc_pcie_setup(struct iproc_pcie *pcie, struct list_head *res)
 
 	iproc_pcie_enable(pcie);
 
+	if (IS_ENABLED(CONFIG_PCI_MSI))
+		if (iproc_pcie_msi_enable(pcie))
+			dev_info(pcie->dev, "not using iProc MSI\n");
+
 	pci_scan_child_bus(bus);
 	pci_assign_unassigned_bus_resources(bus);
 	pci_fixup_irqs(pci_common_swizzle, pcie->map_irq);
@@ -531,6 +555,8 @@  int iproc_pcie_remove(struct iproc_pcie *pcie)
 	pci_stop_root_bus(pcie->root_bus);
 	pci_remove_root_bus(pcie->root_bus);
 
+	iproc_pcie_msi_disable(pcie);
+
 	phy_power_off(pcie->phy);
 	phy_exit(pcie->phy);
 
diff --git a/drivers/pci/host/pcie-iproc.h b/drivers/pci/host/pcie-iproc.h
index 051b651..8b9bb86 100644
--- a/drivers/pci/host/pcie-iproc.h
+++ b/drivers/pci/host/pcie-iproc.h
@@ -41,6 +41,8 @@  struct iproc_pcie_ob {
 	resource_size_t window_size;
 };
 
+struct iproc_msi;
+
 /**
  * iProc PCIe device
  *
@@ -51,10 +53,10 @@  struct iproc_pcie_ob {
  * @sysdata: Per PCI controller data (ARM-specific)
  * @root_bus: pointer to root bus
  * @phy: optional PHY device that controls the Serdes
- * @irqs: interrupt IDs
  * @map_irq: function callback to map interrupts
- * @need_ob_cfg: indidates SW needs to configure the outbound mapping window
+ * @need_ob_cfg: indicates SW needs to configure the outbound mapping window
  * @ob: outbound mapping parameters
+ * @msi: MSI data
  */
 struct iproc_pcie {
 	struct device *dev;
@@ -69,9 +71,24 @@  struct iproc_pcie {
 	int (*map_irq)(const struct pci_dev *, u8, u8);
 	bool need_ob_cfg;
 	struct iproc_pcie_ob ob;
+	struct iproc_msi *msi;
 };
 
 int iproc_pcie_setup(struct iproc_pcie *pcie, struct list_head *res);
 int iproc_pcie_remove(struct iproc_pcie *pcie);
 
+#ifdef CONFIG_PCI_MSI
+int iproc_msi_init(struct iproc_pcie *pcie, struct device_node *node);
+void iproc_msi_exit(struct iproc_pcie *pcie);
+#else
+static inline int iproc_msi_init(struct iproc_pcie *pcie,
+				 struct device_node *node)
+{
+	return -ENODEV;
+}
+static void iproc_msi_exit(struct iproc_pcie *pcie)
+{
+}
+#endif
+
 #endif /* _PCIE_IPROC_H */