Re: [PATCH v2 2/3] PCI: dwc: Add Qualcomm PCIe Endpoint controller driver

[Bjorn Andersson]

On Thu 03 Jun 05:38 CDT 2021, Manivannan Sadhasivam wrote:

> Add driver support for Qualcomm PCIe Endpoint controller driver based on
> the Designware core with added Qualcomm specific wrapper around the
> core. The driver support is very basic such that it supports only
> enumeration, PCIe read/write, and MSI. There is no ASPM and PM support
> for now but these will be added later.
> 
> The driver is capable of using the PERST# and WAKE# side-band GPIOs for
> operation and written on top of the DWC PCI framework.
> 
> Co-developed-by: Siddartha Mohanadoss <smohanad@xxxxxxxxxxxxxx>
> Signed-off-by: Siddartha Mohanadoss <smohanad@xxxxxxxxxxxxxx>
> [mani: restructured the driver and fixed several bugs for upstream]
> Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx>

Really nice to see this working!

> ---
>  drivers/pci/controller/dwc/Kconfig        |  10 +
>  drivers/pci/controller/dwc/Makefile       |   1 +
>  drivers/pci/controller/dwc/pcie-qcom-ep.c | 780 ++++++++++++++++++++++
>  3 files changed, 791 insertions(+)
>  create mode 100644 drivers/pci/controller/dwc/pcie-qcom-ep.c
> 
> diff --git a/drivers/pci/controller/dwc/Kconfig b/drivers/pci/controller/dwc/Kconfig
> index 423d35872ce4..32e735b1fd85 100644
> --- a/drivers/pci/controller/dwc/Kconfig
> +++ b/drivers/pci/controller/dwc/Kconfig
> @@ -180,6 +180,16 @@ config PCIE_QCOM
>  	  PCIe controller uses the DesignWare core plus Qualcomm-specific
>  	  hardware wrappers.
>  
> +config PCIE_QCOM_EP
> +	bool "Qualcomm PCIe controller - Endpoint mode"
> +	depends on OF && (ARCH_QCOM || COMPILE_TEST)
> +	depends on PCI_ENDPOINT
> +	select PCIE_DW_EP
> +	help
> +	  Say Y here to enable support for the PCIe controllers on Qualcomm SoCs
> +	  to work in endpoint mode. The PCIe controller uses the DesignWare core
> +	  plus Qualcomm-specific hardware wrappers.
> +
>  config PCIE_ARMADA_8K
>  	bool "Marvell Armada-8K PCIe controller"
>  	depends on ARCH_MVEBU || COMPILE_TEST
> diff --git a/drivers/pci/controller/dwc/Makefile b/drivers/pci/controller/dwc/Makefile
> index eca805c1a023..abb27642d46b 100644
> --- a/drivers/pci/controller/dwc/Makefile
> +++ b/drivers/pci/controller/dwc/Makefile
> @@ -12,6 +12,7 @@ obj-$(CONFIG_PCI_KEYSTONE) += pci-keystone.o
>  obj-$(CONFIG_PCI_LAYERSCAPE) += pci-layerscape.o
>  obj-$(CONFIG_PCI_LAYERSCAPE_EP) += pci-layerscape-ep.o
>  obj-$(CONFIG_PCIE_QCOM) += pcie-qcom.o
> +obj-$(CONFIG_PCIE_QCOM_EP) += pcie-qcom-ep.o
>  obj-$(CONFIG_PCIE_ARMADA_8K) += pcie-armada8k.o
>  obj-$(CONFIG_PCIE_ARTPEC6) += pcie-artpec6.o
>  obj-$(CONFIG_PCIE_INTEL_GW) += pcie-intel-gw.o
> diff --git a/drivers/pci/controller/dwc/pcie-qcom-ep.c b/drivers/pci/controller/dwc/pcie-qcom-ep.c
> new file mode 100644
> index 000000000000..b68511bacc2a
> --- /dev/null
> +++ b/drivers/pci/controller/dwc/pcie-qcom-ep.c
> @@ -0,0 +1,780 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Qualcomm PCIe Endpoint controller driver
> + *
> + * Copyright (c) 2020, The Linux Foundation. All rights reserved.
> + * Author: Siddartha Mohanadoss <smohanad@xxxxxxxxxxxxxx
> + *
> + * Copyright (c) 2021, Linaro Ltd.
> + * Author: Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx
> + */
> +
> +#include <linux/clk.h>
> +#include <linux/platform_device.h>
> +#include <linux/gpio/consumer.h>
> +#include <linux/reset.h>
> +#include <linux/delay.h>
> +#include <linux/phy/phy.h>
> +#include <linux/pm_domain.h>
> +
> +#include "pcie-designware.h"
> +
> +/* PARF registers */
> +#define PARF_SYS_CTRL				0x00
> +#define PARF_DB_CTRL				0x10
> +#define PARF_PM_CTRL				0x20
> +#define PARF_DEBUG_INT_EN			0x190
> +#define PARF_AXI_MSTR_RD_HALT_NO_WRITES		0x1a4
> +#define PARF_AXI_MSTR_WR_ADDR_HALT		0x1a8
> +#define PARF_Q2A_FLUSH				0x1aC
> +#define PARF_LTSSM				0x1b0
> +#define PARF_CFG_BITS				0x210
> +#define PARF_INT_ALL_STATUS			0x224
> +#define PARF_INT_ALL_CLEAR			0x228
> +#define PARF_INT_ALL_MASK			0x22c
> +#define PARF_SLV_ADDR_MSB_CTRL			0x2c0
> +#define PARF_DBI_BASE_ADDR			0x350
> +#define PARF_DBI_BASE_ADDR_HI			0x354
> +#define PARF_SLV_ADDR_SPACE_SIZE		0x358
> +#define PARF_SLV_ADDR_SPACE_SIZE_HI		0x35c
> +#define PARF_ATU_BASE_ADDR			0x634
> +#define PARF_ATU_BASE_ADDR_HI			0x638
> +#define PARF_SRIS_MODE				0x644
> +#define PARF_DEVICE_TYPE			0x1000
> +#define PARF_BDF_TO_SID_CFG			0x2c00
> +
> +/* ELBI registers */
> +#define ELBI_SYS_STTS				0x08
> +
> +/* DBI registers */
> +#define DBI_CAP_ID_NXT_PTR			0x40
> +#define DBI_CON_STATUS				0x44
> +#define DBI_DEVICE_CAPABILITIES			0x74
> +#define DBI_LINK_CAPABILITIES			0x7c
> +#define DBI_LINK_CONTROL2_LINK_STATUS2		0xa0
> +#define DBI_L1SUB_CAPABILITY			0x234
> +#define DBI_ACK_F_ASPM_CTRL			0x70c
> +#define DBI_GEN3_RELATED_OFF			0x890
> +#define DBI_AUX_CLK_FREQ			0xb40
> +
> +#define DBI_L0S_ACCPT_LATENCY_MASK		GENMASK(8, 6)
> +#define DBI_L1_ACCPT_LATENCY_MASK		GENMASK(11, 9)
> +#define DBI_L0S_EXIT_LATENCY_MASK		GENMASK(14, 12)
> +#define DBI_L1_EXIT_LATENCY_MASK		GENMASK(17, 15)
> +#define DBI_ACK_N_FTS_MASK			GENMASK(15, 8)
> +
> +/* TCSR registers */
> +#define TCSR_PCIE_PERST_EN			0x258
> +#define TCSR_PERST_SEPARATION_ENABLE		0x270
> +
> +#define XMLH_LINK_UP				0x400
> +#define CORE_RESET_TIME_US_MIN			1000
> +#define CORE_RESET_TIME_US_MAX			1005
> +#define WAKE_DELAY_US				2000 /* 2 ms */
> +
> +#define to_pcie_ep(x)				dev_get_drvdata((x)->dev)

Isn't this also container_of(x, struct qcom_pcie_ep, pci), but without
the need to assigning and use drvdata?

> +
> +enum qcom_pcie_ep_link_status {
> +	QCOM_PCIE_EP_LINK_DISABLED,
> +	QCOM_PCIE_EP_LINK_ENABLED,
> +	QCOM_PCIE_EP_LINK_UP,
> +	QCOM_PCIE_EP_LINK_DOWN,
> +};
> +
> +enum qcom_pcie_ep_irq {
> +	QCOM_PCIE_EP_INT_RESERVED,
> +	QCOM_PCIE_EP_INT_LINK_DOWN,
> +	QCOM_PCIE_EP_INT_BME,
> +	QCOM_PCIE_EP_INT_PM_TURNOFF,
> +	QCOM_PCIE_EP_INT_DEBUG,
> +	QCOM_PCIE_EP_INT_LTR,
> +	QCOM_PCIE_EP_INT_MHI_Q6,
> +	QCOM_PCIE_EP_INT_MHI_A7,
> +	QCOM_PCIE_EP_INT_DSTATE_CHANGE,
> +	QCOM_PCIE_EP_INT_L1SUB_TIMEOUT,
> +	QCOM_PCIE_EP_INT_MMIO_WRITE,
> +	QCOM_PCIE_EP_INT_CFG_WRITE,
> +	QCOM_PCIE_EP_INT_BRIDGE_FLUSH_N,
> +	QCOM_PCIE_EP_INT_LINK_UP,
> +	QCOM_PCIE_EP_INT_AER_LEGACY,
> +	QCOM_PCIE_EP_INT_PLS_ERR,
> +	QCOM_PCIE_EP_INT_PME_LEGACY,
> +	QCOM_PCIE_EP_INT_PLS_PME,
> +	QCOM_PCIE_EP_INT_MAX,

Afaict these aren't some logical sequence of numbers, they are bit
offsets in the interrupt registers. Using #define would make this
obvious.

> +};
> +
> +static struct clk_bulk_data qcom_pcie_ep_clks[] = {
> +	{ .id = "cfg" },
> +	{ .id = "aux" },
> +	{ .id = "bus_master" },
> +	{ .id = "bus_slave" },
> +	{ .id = "ref" },
> +	{ .id = "sleep" },
> +	{ .id = "slave_q2a" },
> +};
> +
> +struct qcom_pcie_ep {
> +	struct dw_pcie pci;
> +
> +	void __iomem *parf;
> +	void __iomem *elbi;
> +	void __iomem *tcsr;
> +
> +	struct reset_control *core_reset;
> +	struct gpio_desc *reset;
> +	struct gpio_desc *wake;
> +	struct phy *phy;
> +
> +	resource_size_t dbi_phys;
> +	resource_size_t atu_phys;
> +
> +	enum qcom_pcie_ep_link_status link_status;
> +	int global_irq;
> +	int perst_irq;
> +};
> +
> +static void qcom_pcie_ep_enable_ltssm(struct qcom_pcie_ep *pcie_ep)
> +{
> +	u32 reg;
> +
> +	reg = readl(pcie_ep->parf + PARF_LTSSM);
> +	reg |= BIT(8);
> +	writel_relaxed(reg, pcie_ep->parf + PARF_LTSSM);
> +}
> +
> +static int qcom_pcie_ep_core_reset(struct qcom_pcie_ep *pcie_ep)
> +{
> +	struct dw_pcie *pci = &pcie_ep->pci;
> +	struct device *dev = pci->dev;
> +	int ret = 0;

No need to initialize ret, when the first access is an assignment.

> +
> +	ret = reset_control_assert(pcie_ep->core_reset);
> +	if (ret) {
> +		dev_err(dev, "Cannot assert core reset\n");
> +		return ret;
> +	}
> +
> +	usleep_range(CORE_RESET_TIME_US_MIN, CORE_RESET_TIME_US_MAX);
> +
> +	ret = reset_control_deassert(pcie_ep->core_reset);
> +	if (ret) {
> +		dev_err(dev, "Cannot de-assert core reset\n");
> +		return ret;
> +	}
> +
> +	usleep_range(CORE_RESET_TIME_US_MIN, CORE_RESET_TIME_US_MAX);
> +
> +	return 0;
> +}
> +
> +/*
> + * Delatch PERST_EN and PERST_SEPARATION_ENABLE with TCSR to avoid
> + * device reset during host reboot and hibernation. The driver is
> + * expected to handle this situation.
> + */
> +static void qcom_pcie_ep_configure_tcsr(struct qcom_pcie_ep *pcie_ep)
> +{
> +	writel_relaxed(0x0, pcie_ep->tcsr + TCSR_PCIE_PERST_EN);

Please avoid _relaxed accessor unless there's a strong reason, and if so
document it.

> +	writel_relaxed(0x0, pcie_ep->tcsr + TCSR_PERST_SEPARATION_ENABLE);
> +}
> +
> +static int qcom_pcie_ep_enable_resources(struct qcom_pcie_ep *pcie_ep)
> +{
> +	struct dw_pcie *pci = &pcie_ep->pci;
> +	struct device *dev = pci->dev;
> +	int ret;
> +
> +	ret = clk_bulk_prepare_enable(ARRAY_SIZE(qcom_pcie_ep_clks),
> +				      qcom_pcie_ep_clks);
> +	if (ret) {
> +		dev_err(dev, "Failed to enable clocks: %d\n", ret);

clk_bulk_prepare_enable() will print an error indicating which of the
clocks it failed to prepare/enable, so it's better to skip this.

As such, you can simply do:

	return clk_bulk_prepare_enable();

> +		return ret;
> +	}
> +
> +	return 0;
> +}
> +
> +static void qcom_pcie_ep_disable_resources(struct qcom_pcie_ep *pcie_ep)
> +{
> +	clk_bulk_disable_unprepare(ARRAY_SIZE(qcom_pcie_ep_clks),
> +				   qcom_pcie_ep_clks);
> +}
> +
> +static void qcom_pcie_ep_configure_irq(struct qcom_pcie_ep *pcie_ep)
> +{
> +	u32 val;
> +
> +	writel_relaxed(0, pcie_ep->parf + PARF_INT_ALL_MASK);
> +	val = BIT(QCOM_PCIE_EP_INT_LINK_DOWN) |
> +		BIT(QCOM_PCIE_EP_INT_BME) |
> +		BIT(QCOM_PCIE_EP_INT_PM_TURNOFF) |
> +		BIT(QCOM_PCIE_EP_INT_DSTATE_CHANGE) |
> +		BIT(QCOM_PCIE_EP_INT_LINK_UP);
> +	writel_relaxed(val, pcie_ep->parf + PARF_INT_ALL_MASK);

Again, please avoid the _relaxed versions.

> +}
> +
> +static int qcom_pcie_ep_core_init(struct qcom_pcie_ep *pcie_ep)
> +{
> +	struct dw_pcie *pci = &pcie_ep->pci;
> +	u32 val;
> +
> +	/* Disable BDF to SID mapping */
> +	val = readl_relaxed(pcie_ep->parf + PARF_BDF_TO_SID_CFG);
> +	val |= BIT(0);
> +	writel_relaxed(val, pcie_ep->parf + PARF_BDF_TO_SID_CFG);
> +
> +	/* enable debug IRQ */
> +	writel_relaxed((BIT(3) | BIT(2) | BIT(1)),
> +		       pcie_ep->parf + PARF_DEBUG_INT_EN);
> +
> +	/* Configure PCIe to endpoint mode */
> +	writel_relaxed(0x0, pcie_ep->parf + PARF_DEVICE_TYPE);
> +
> +	/* Configure PCIe core to support 1GB aperture */
> +	writel_relaxed(0x40000000, pcie_ep->parf + PARF_SLV_ADDR_SPACE_SIZE);
> +
> +	/* Allow entering L1 state */
> +	val = readl_relaxed(pcie_ep->parf + PARF_PM_CTRL);
> +	val &= ~BIT(5);
> +	writel_relaxed(val, pcie_ep->parf + PARF_PM_CTRL);
> +
> +	/* Configure Slave, DBI and iATU base addresses */
> +	writel_relaxed(BIT(0), pcie_ep->parf + PARF_SLV_ADDR_MSB_CTRL);
> +	writel_relaxed(0x200, pcie_ep->parf + PARF_SLV_ADDR_SPACE_SIZE_HI);
> +	writel_relaxed(0x0, pcie_ep->parf + PARF_SLV_ADDR_SPACE_SIZE);
> +	writel_relaxed(0x100, pcie_ep->parf + PARF_DBI_BASE_ADDR_HI);
> +	writel_relaxed(pcie_ep->dbi_phys, pcie_ep->parf + PARF_DBI_BASE_ADDR);
> +	writel_relaxed(0x100, pcie_ep->parf + PARF_ATU_BASE_ADDR_HI);
> +	writel_relaxed(pcie_ep->atu_phys, pcie_ep->parf + PARF_ATU_BASE_ADDR);
> +
> +	/* Read halts write */
> +	writel_relaxed(0x0, pcie_ep->parf + PARF_AXI_MSTR_RD_HALT_NO_WRITES);
> +	/* Write after write halt */
> +	writel_relaxed(BIT(31), pcie_ep->parf + PARF_AXI_MSTR_WR_ADDR_HALT);
> +	/* Q2A flush disable */
> +	writel_relaxed(0, pcie_ep->parf + PARF_Q2A_FLUSH);
> +
> +	/* Disable the DBI Wakeup */
> +	writel_relaxed(BIT(11), pcie_ep->parf + PARF_SYS_CTRL);
> +	/* Disable the debouncers */
> +	writel_relaxed(0x73, pcie_ep->parf + PARF_DB_CTRL);
> +	/* Disable core clock CGC */
> +	writel_relaxed(BIT(6), pcie_ep->parf + PARF_SYS_CTRL);
> +	/* Set AUX power to be on */
> +	writel_relaxed(BIT(4), pcie_ep->parf + PARF_SYS_CTRL);
> +	/* Request to exit from L1SS for MSI and LTR MSG */
> +	writel_relaxed(BIT(1), pcie_ep->parf + PARF_CFG_BITS);
> +
> +	dw_pcie_dbi_ro_wr_en(pci);
> +
> +	/* Set the PMC Register - to support PME in D0/D3hot/D3cold */
> +	val = dw_pcie_readl_dbi(pci, DBI_CAP_ID_NXT_PTR);
> +	val |= BIT(31) | BIT(30) | BIT(27);
> +	dw_pcie_writel_dbi(pci, DBI_CAP_ID_NXT_PTR, val);
> +
> +	/* Set the Endpoint L0s Acceptable Latency to 1us (max) */
> +	val = dw_pcie_readl_dbi(pci, DBI_DEVICE_CAPABILITIES);
> +	val |= FIELD_PREP(DBI_L0S_ACCPT_LATENCY_MASK, 0x7);
> +	dw_pcie_writel_dbi(pci, DBI_DEVICE_CAPABILITIES, val);
> +
> +	/* Set the Endpoint L1 Acceptable Latency to 1us (max) */
> +	val = dw_pcie_readl_dbi(pci, DBI_DEVICE_CAPABILITIES);
> +	val |= FIELD_PREP(DBI_L1_ACCPT_LATENCY_MASK, 0x7);
> +	dw_pcie_writel_dbi(pci, DBI_DEVICE_CAPABILITIES, val);
> +
> +	/* Set the L0s Exit Latency to 2us-4us = 0x6 */
> +	val = dw_pcie_readl_dbi(pci, DBI_LINK_CAPABILITIES);
> +	val |= FIELD_PREP(DBI_L0S_EXIT_LATENCY_MASK, 0x6);
> +	dw_pcie_writel_dbi(pci, DBI_LINK_CAPABILITIES, val);
> +
> +	/* Set the L1 Exit Latency to be 32us-64 us = 0x6 */
> +	val = dw_pcie_readl_dbi(pci, DBI_LINK_CAPABILITIES);
> +	val |= FIELD_PREP(DBI_L1_EXIT_LATENCY_MASK, 0x6);
> +	dw_pcie_writel_dbi(pci, DBI_LINK_CAPABILITIES, val);
> +
> +	/* L1ss is supported */
> +	val = dw_pcie_readl_dbi(pci, DBI_L1SUB_CAPABILITY);
> +	val |= 0x1f;
> +	dw_pcie_writel_dbi(pci, DBI_L1SUB_CAPABILITY, val);
> +
> +	/* Enable Clock Power Management */
> +	val = dw_pcie_readl_dbi(pci, DBI_LINK_CAPABILITIES);
> +	val |= BIT(18);
> +	dw_pcie_writel_dbi(pci, DBI_LINK_CAPABILITIES, val);
> +
> +	dw_pcie_dbi_ro_wr_dis(pci);
> +
> +	/* Set FTS value to match the PHY setting */
> +	val = dw_pcie_readl_dbi(pci, DBI_ACK_F_ASPM_CTRL);
> +	val |= FIELD_PREP(DBI_ACK_N_FTS_MASK, 0x80);
> +	dw_pcie_writel_dbi(pci, DBI_ACK_F_ASPM_CTRL, val);
> +
> +	dw_pcie_writel_dbi(pci, DBI_AUX_CLK_FREQ, 0x14);
> +
> +	/* Prevent L1ss wakeup after 100ms */
> +	val = dw_pcie_readl_dbi(pci, DBI_GEN3_RELATED_OFF);
> +	val &= ~BIT(0);
> +	dw_pcie_writel_dbi(pci, DBI_GEN3_RELATED_OFF, val);
> +
> +	/* Disable SRIS_MODE */
> +	val = readl_relaxed(pcie_ep->parf + PARF_SRIS_MODE);
> +	val &= ~BIT(0);
> +	writel_relaxed(val, pcie_ep->parf + PARF_SRIS_MODE);
> +
> +	qcom_pcie_ep_configure_irq(pcie_ep);

Why is the irq part broken out to its own function?

> +
> +	return 0;
> +}
> +
> +static int qcom_pcie_confirm_linkup(struct dw_pcie *pci)
> +{
> +	struct qcom_pcie_ep *pcie_ep = to_pcie_ep(pci);
> +	u32 reg;
> +
> +	reg = readl_relaxed(pcie_ep->elbi + ELBI_SYS_STTS);
> +
> +	return reg & XMLH_LINK_UP;
> +}
> +
> +static int qcom_pcie_start_link(struct dw_pcie *pci)
> +{
> +	struct qcom_pcie_ep *pcie_ep = to_pcie_ep(pci);
> +
> +	enable_irq(pcie_ep->perst_irq);
> +
> +	return 0;
> +}
> +
> +static int qcom_pcie_establish_link(struct dw_pcie *pci)
> +{
> +	struct qcom_pcie_ep *pcie_ep = to_pcie_ep(pci);
> +	struct device *dev = pci->dev;
> +	int ret;
> +
> +	if (pcie_ep->link_status == QCOM_PCIE_EP_LINK_ENABLED) {
> +		dev_info(dev, "Link is already enabled\n");

Per qcom_pcie_ep_global_threaded_irq() it seems possible that the link
is enabled but link_status != QCOM_PCIE_EP_LINK_ENABLED.

I'm also not sure how you can end up here with link_status == ENABLED;
it starts off DISABLED and expect the first thing to happen is "perst"
asserting, which would end up here and after that the only way you get
here again is if perst is asserted again - without first being
deasserted. So perhaps this is part of your manual edge detector?

Perhaps it's not needed if you actually use edge detection on your
interrupt?

> +		return 0;
> +	}
> +
> +	/* Enable power and clocks */
> +	ret = qcom_pcie_ep_enable_resources(pcie_ep);
> +	if (ret) {
> +		dev_err(dev, "Failed to enable resources: %d\n", ret);
> +		return ret;
> +	}
> +
> +	/* Perform controller reset */
> +	ret = qcom_pcie_ep_core_reset(pcie_ep);
> +	if (ret) {
> +		dev_err(dev, "Failed to reset the core: %d\n", ret);
> +		goto err_disable_resources;
> +	}
> +
> +	/* Initialize PHY */
> +	ret = phy_init(pcie_ep->phy);
> +	if (ret) {
> +		dev_err(dev, "Failed to initialize PHY: %d\n", ret);
> +		goto err_disable_resources;
> +	}
> +
> +	/* Power ON PHY */
> +	ret = phy_power_on(pcie_ep->phy);
> +	if (ret) {
> +		dev_err(dev, "Failed to Power ON PHY: %d\n", ret);
> +		goto err_phy_exit;
> +	}
> +
> +	/* Assert WAKE# to RC to indicate device is ready */
> +	gpiod_set_value_cansleep(pcie_ep->wake, 0);

Assert sounds like "activate" or "turn on", so I would prefer that you
make this a logical 1 here and if the signal is inverted specify this
when you're requesting the gpio.

> +	usleep_range(WAKE_DELAY_US, WAKE_DELAY_US + 500);
> +	gpiod_set_value_cansleep(pcie_ep->wake, 1);
> +
> +	/* Configure TCSR */
> +	qcom_pcie_ep_configure_tcsr(pcie_ep);
> +
> +	/* Initialize the controller */
> +	ret = qcom_pcie_ep_core_init(pcie_ep);
> +	if (ret) {
> +		dev_err(dev, "Failed to init controller: %d\n", ret);
> +		goto err_phy_power_off;
> +	}
> +
> +	ret = dw_pcie_ep_init_complete(&pcie_ep->pci.ep);
> +	if (ret) {
> +		dev_err(dev, "Failed to complete initialization: %d\n", ret);
> +		goto err_phy_power_off;
> +	}
> +
> +	dw_pcie_ep_init_notify(&pcie_ep->pci.ep);
> +
> +	/* Enable LTSSM */
> +	qcom_pcie_ep_enable_ltssm(pcie_ep);
> +
> +	return 0;
> +/* TODO* assert reset? */
> +err_phy_power_off:
> +	phy_power_off(pcie_ep->phy);
> +err_phy_exit:
> +	phy_exit(pcie_ep->phy);
> +err_disable_resources:
> +	qcom_pcie_ep_disable_resources(pcie_ep);
> +
> +	return ret;
> +}
> +
> +static void qcom_pcie_disable_link(struct dw_pcie *pci)
> +{
> +	struct qcom_pcie_ep *pcie_ep = to_pcie_ep(pci);
> +	struct device *dev = pci->dev;
> +
> +	if (pcie_ep->link_status == QCOM_PCIE_EP_LINK_DISABLED) {
> +		dev_info(dev, "Link is already disabled\n");

Why can disable_link happen both from the interrupt handler directly and
from the dw_pcie_ops?

> +		return;
> +	}
> +
> +	phy_power_off(pcie_ep->phy);
> +	phy_exit(pcie_ep->phy);
> +	qcom_pcie_ep_disable_resources(pcie_ep);
> +	pcie_ep->link_status = QCOM_PCIE_EP_LINK_DISABLED;
> +}
> +
> +/* Common DWC controller ops */
> +static const struct dw_pcie_ops pci_ops = {
> +	.link_up = qcom_pcie_confirm_linkup,
> +	.start_link = qcom_pcie_start_link,
> +	.stop_link = qcom_pcie_disable_link,
> +};
> +
> +static int qcom_pcie_ep_get_io_resources(struct platform_device *pdev,
> +					 struct qcom_pcie_ep *pcie_ep)
> +{
> +	struct device *dev = &pdev->dev;
> +	struct dw_pcie *pci = &pcie_ep->pci;
> +	struct resource *res;
> +
> +	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "parf");
> +	pcie_ep->parf = devm_ioremap_resource(dev, res);

Better make this devm_platform_ioremap_resource_byname(), or someone
will send a patch the minute this code lands.

> +	if (IS_ERR(pcie_ep->parf))
> +		return PTR_ERR(pcie_ep->parf);
> +
> +	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "dbi");
> +	pci->dbi_base = devm_pci_remap_cfg_resource(dev, res);
> +	if (IS_ERR(pci->dbi_base))
> +		return PTR_ERR(pci->dbi_base);
> +	pci->dbi_base2 = pci->dbi_base;
> +	pcie_ep->dbi_phys = res->start;
> +
> +	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "elbi");
> +	pcie_ep->elbi = devm_pci_remap_cfg_resource(dev, res);
> +	if (IS_ERR(pcie_ep->elbi))
> +		return PTR_ERR(pcie_ep->elbi);
> +
> +	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "atu");
> +	pci->atu_base = devm_pci_remap_cfg_resource(dev, res);
> +	if (IS_ERR(pci->atu_base))
> +		return PTR_ERR(pci->atu_base);
> +	pcie_ep->atu_phys = res->start;
> +
> +	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "tcsr");
> +	pcie_ep->tcsr = devm_ioremap_resource(dev, res);
> +	if (IS_ERR(pcie_ep->tcsr))
> +		return PTR_ERR(pcie_ep->tcsr);
> +
> +	return 0;
> +}
> +
> +static int qcom_pcie_ep_get_resources(struct platform_device *pdev,
> +				      struct qcom_pcie_ep *pcie_ep)
> +{
> +	struct device *dev = &pdev->dev;
> +	int ret;
> +
> +	ret = qcom_pcie_ep_get_io_resources(pdev, pcie_ep);
> +	if (ret) {
> +		dev_err(&pdev->dev, "failed to get io resources %d\n", ret);
> +		return ret;
> +	}
> +
> +	/* Enable GDSC power domain */
> +	ret = dev_pm_domain_attach(dev, true);

If you only have a single power-domain specified, then it would already
be attached to "dev".

> +	if (ret)
> +		return ret;
> +
> +	ret = devm_clk_bulk_get(dev, ARRAY_SIZE(qcom_pcie_ep_clks),
> +				qcom_pcie_ep_clks);
> +	if (ret) {
> +		dev_err(dev, "Failed to get clocks: %d\n", ret);

Don't you already have a print in devm_clk_bulk_get()?

> +		return ret;
> +	}
> +
> +	pcie_ep->core_reset = devm_reset_control_get_exclusive(dev, "core");
> +	if (IS_ERR(pcie_ep->core_reset))
> +		return PTR_ERR(pcie_ep->core_reset);
> +
> +	pcie_ep->reset = devm_gpiod_get(dev, "reset", GPIOD_IN);
> +	if (IS_ERR(pcie_ep->reset))
> +		return PTR_ERR(pcie_ep->reset);
> +
> +	pcie_ep->wake = devm_gpiod_get_optional(dev, "wake", GPIOD_OUT_HIGH);
> +	if (IS_ERR(pcie_ep->wake))
> +		return PTR_ERR(pcie_ep->wake);
> +
> +	pcie_ep->phy = devm_phy_optional_get(&pdev->dev, "pciephy");
> +	if (IS_ERR(pcie_ep->phy))
> +		ret = PTR_ERR(pcie_ep->phy);
> +
> +	return ret;
> +}
> +
> +/* TODO: Notify clients about PCIe state change */
> +static irqreturn_t qcom_pcie_ep_global_threaded_irq(int irq, void *data)
> +{
> +	struct qcom_pcie_ep *pcie_ep = data;
> +	struct dw_pcie *pci = &pcie_ep->pci;
> +	struct device *dev = pci->dev;
> +	u32 status = readl(pcie_ep->parf + PARF_INT_ALL_STATUS);
> +	u32 mask = readl(pcie_ep->parf + PARF_INT_ALL_MASK);
> +	u32 dstate, event, val;
> +
> +	writel_relaxed(status, pcie_ep->parf + PARF_INT_ALL_CLEAR);
> +	status &= mask;
> +
> +	for (event = 1; event < QCOM_PCIE_EP_INT_MAX; event++) {
> +		if (status & BIT(event)) {

for_each_set_bit()

> +			switch (event) {
> +			case QCOM_PCIE_EP_INT_LINK_DOWN:
> +				dev_info(dev, "Received Linkdown event\n");
> +				pcie_ep->link_status = QCOM_PCIE_EP_LINK_DOWN;
> +				break;
> +			case QCOM_PCIE_EP_INT_BME:
> +				dev_info(dev, "Received BME event. Link is enabled!\n");
> +				pcie_ep->link_status = QCOM_PCIE_EP_LINK_ENABLED;
> +				break;
> +			case QCOM_PCIE_EP_INT_PM_TURNOFF:
> +				dev_info(dev, "Received PM Turn-off event! Entering L23\n");
> +				val = readl_relaxed(pcie_ep->parf + PARF_PM_CTRL);
> +				val |= BIT(2);
> +				writel_relaxed(val, pcie_ep->parf + PARF_PM_CTRL);
> +				break;
> +			case QCOM_PCIE_EP_INT_DSTATE_CHANGE:
> +				dstate = dw_pcie_readl_dbi(pci, DBI_CON_STATUS) & 0x3;
> +				dev_info(dev, "Received D%d state event\n", dstate);
> +				if (dstate == 3) {
> +					val = readl_relaxed(pcie_ep->parf + PARF_PM_CTRL);
> +					val |= BIT(1);
> +					writel_relaxed(val, pcie_ep->parf + PARF_PM_CTRL);
> +				}
> +				/* Handle D0 state change */
> +				break;
> +			case QCOM_PCIE_EP_INT_LINK_UP:
> +				dev_info(dev, "Received Linkup event. Enumeration complete!\n");
> +				dw_pcie_ep_linkup(&pci->ep);
> +				pcie_ep->link_status = QCOM_PCIE_EP_LINK_UP;
> +				break;
> +			default:
> +				dev_err(dev, "Received unknown event: %d\n", event);
> +				break;
> +			}
> +		}
> +	}
> +
> +	return IRQ_HANDLED;
> +}
> +
> +static irqreturn_t qcom_pcie_ep_perst_threaded_irq(int irq, void *data)
> +{
> +	struct qcom_pcie_ep *pcie_ep = data;
> +	struct dw_pcie *pci = &pcie_ep->pci;
> +	struct device *dev = pci->dev;
> +	u32 perst;
> +
> +	perst = gpiod_get_value(pcie_ep->reset);
> +
> +	if (perst) {
> +		/* Start link training */
> +		dev_info(dev, "PERST de-asserted by host. Starting link training!\n");
> +		qcom_pcie_establish_link(pci);
> +	} else {
> +		/* Shutdown the link if the link is already on */
> +		dev_info(dev, "PERST asserted by host. Shutting down the PCIe link!\n");
> +		qcom_pcie_disable_link(pci);
> +	}
> +
> +	/* Set trigger type based on the next expected value of perst gpio */
> +	irq_set_irq_type(gpiod_to_irq(pcie_ep->reset),
> +			 (perst ? IRQF_TRIGGER_LOW : IRQF_TRIGGER_HIGH));

Looks like you're manually implementing edge triggering, is there any
reason for that? EDGE_BOTH seems to do the same thing...

> +
> +	return IRQ_HANDLED;
> +}
> +
> +static int qcom_pcie_ep_enable_irq_resources(struct platform_device *pdev,
> +					     struct qcom_pcie_ep *pcie_ep)
> +{
> +	int irq, ret;
> +
> +	irq = platform_get_irq_byname(pdev, "global");
> +	if (irq < 0) {
> +		dev_err(&pdev->dev, "Failed to get Global IRQ\n");
> +		return irq;
> +	}
> +
> +	ret = devm_request_threaded_irq(&pdev->dev, irq, NULL,
> +					qcom_pcie_ep_global_threaded_irq,
> +					IRQF_TRIGGER_HIGH | IRQF_ONESHOT,

Leave out the trigger and rely on DT.

> +					"global_irq", pcie_ep);
> +	if (ret) {
> +		dev_err(&pdev->dev, "Failed to request Global IRQ\n");
> +		return ret;
> +	}
> +
> +	pcie_ep->perst_irq = gpiod_to_irq(pcie_ep->reset);
> +	irq_set_status_flags(pcie_ep->perst_irq, IRQ_NOAUTOEN);

Is the global interrupt needed for dw_pcie_ep_init()? Or could this
simply be done when things are ready to handle the interrupts?

> +	ret = devm_request_threaded_irq(&pdev->dev, pcie_ep->perst_irq, NULL,
> +					qcom_pcie_ep_perst_threaded_irq,
> +					IRQF_TRIGGER_HIGH | IRQF_ONESHOT,
> +					"perst_irq", pcie_ep);
> +	if (ret) {
> +		dev_err(&pdev->dev, "Failed to request PERST IRQ\n");
> +		return ret;
> +	}
> +
> +	return 0;
> +}
> +
> +static int qcom_pcie_ep_raise_irq(struct dw_pcie_ep *ep, u8 func_no,
> +				  enum pci_epc_irq_type type, u16 interrupt_num)
> +{
> +	struct dw_pcie *pci = to_dw_pcie_from_ep(ep);
> +
> +	switch (type) {
> +	case PCI_EPC_IRQ_LEGACY:
> +		return dw_pcie_ep_raise_legacy_irq(ep, func_no);
> +	case PCI_EPC_IRQ_MSI:
> +		return dw_pcie_ep_raise_msi_irq(ep, func_no, interrupt_num);
> +	default:
> +		dev_err(pci->dev, "Unknown IRQ type\n");
> +		return -EINVAL;
> +	}
> +}
> +
> +static const struct pci_epc_features qcom_pcie_epc_features = {
> +	.linkup_notifier = true,
> +	.core_init_notifier = true,
> +	.msi_capable = true,
> +	.msix_capable = false,
> +};
> +
> +static const struct pci_epc_features *
> +qcom_pcie_epc_get_features(struct dw_pcie_ep *pci_ep)
> +{
> +	return &qcom_pcie_epc_features;
> +}
> +
> +static void qcom_pcie_ep_init(struct dw_pcie_ep *ep)
> +{
> +	struct dw_pcie *pci = to_dw_pcie_from_ep(ep);
> +	enum pci_barno bar;
> +
> +	for (bar = BAR_0; bar <= BAR_5; bar++)
> +		dw_pcie_ep_reset_bar(pci, bar);
> +}
> +
> +static struct dw_pcie_ep_ops pci_ep_ops = {
> +	.ep_init = qcom_pcie_ep_init,
> +	.raise_irq = qcom_pcie_ep_raise_irq,
> +	.get_features = qcom_pcie_epc_get_features,
> +};
> +
> +static const struct of_device_id qcom_pcie_ep_match[] = {
> +	{ .compatible = "qcom,sdx55-pcie-ep", },
> +	{ }
> +};

Move this below the probe function.

> +
> +static int qcom_pcie_ep_probe(struct platform_device *pdev)
> +{
> +	struct device *dev = &pdev->dev;
> +	struct qcom_pcie_ep *pcie_ep;
> +	int ret;
> +
> +	pcie_ep = devm_kzalloc(dev, sizeof(*pcie_ep), GFP_KERNEL);
> +	if (!pcie_ep)
> +		return -ENOMEM;
> +
> +	pcie_ep->pci.dev = dev;
> +	pcie_ep->pci.ops = &pci_ops;
> +	pcie_ep->pci.ep.ops = &pci_ep_ops;
> +
> +	ret = qcom_pcie_ep_get_resources(pdev, pcie_ep);
> +	if (ret) {
> +		dev_err(dev, "Failed to get resources:%d\n", ret);

You printed an error in about half the error cases in
qcom_pcie_ep_get_resources(), it would be better to ensure you print a
specific error for each of the remaining cases there and "silently"
return here.

Same comment for below cases.

> +		return ret;
> +	}
> +
> +	/* Enable power and clocks */
> +	ret = qcom_pcie_ep_enable_resources(pcie_ep);
> +	if (ret) {
> +		dev_err(dev, "Failed to enable resources: %d\n", ret);
> +		return ret;
> +	}
> +
> +	/* Perform controller reset */
> +	ret = qcom_pcie_ep_core_reset(pcie_ep);
> +	if (ret) {
> +		dev_err(dev, "Failed to reset the core: %d\n", ret);
> +		goto err_disable_resources;
> +	}
> +
> +	/* Initialize PHY */
> +	ret = phy_init(pcie_ep->phy);
> +	if (ret) {
> +		dev_err(dev, "Failed to initialize PHY: %d\n", ret);
> +		goto err_disable_resources;
> +	}
> +
> +	/* PHY needs to be powered on for dw_pcie_ep_init() */
> +	ret = phy_power_on(pcie_ep->phy);
> +	if (ret) {
> +		dev_err(dev, "Failed to Power ON PHY: %d\n", ret);
> +		goto err_phy_exit;
> +	}
> +
> +	platform_set_drvdata(pdev, pcie_ep);
> +
> +	ret = qcom_pcie_ep_enable_irq_resources(pdev, pcie_ep);
> +	if (ret) {
> +		dev_err(dev, "Failed to get IRQ resources %d\n", ret);
> +		goto err_phy_power_off;
> +	}
> +
> +	ret = dw_pcie_ep_init(&pcie_ep->pci.ep);
> +	if (ret) {
> +		dev_err(dev, "Failed to initialize endpoint:%d\n", ret);
> +		goto err_phy_power_off;
> +	}
> +
> +	return 0;
> +
> +err_phy_power_off:

Your interrupts remains enabled until devres takes them down after
returning, so I suspect you need to disable them here as well - at least
before shutting down the clocks.

> +	phy_power_off(pcie_ep->phy);
> +err_phy_exit:
> +	phy_exit(pcie_ep->phy);
> +err_disable_resources:
> +	qcom_pcie_ep_disable_resources(pcie_ep);
> +
> +	return ret;
> +}
> +
> +static struct platform_driver qcom_pcie_ep_driver = {
> +	.probe	= qcom_pcie_ep_probe,
> +	.driver	= {
> +		.name		= "qcom-pcie-ep",

Skip the indentation of the '='.

> +		.suppress_bind_attrs = true,

Why do we suppress_bind_attrs?

Regards,
Bjorn

> +		.of_match_table	= qcom_pcie_ep_match,
> +	},
> +};
> +builtin_platform_driver(qcom_pcie_ep_driver);
> +
> +MODULE_AUTHOR("Siddartha Mohanadoss <smohanad@xxxxxxxxxxxxxx>");
> +MODULE_AUTHOR("Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx>");
> +MODULE_DESCRIPTION("Qualcomm PCIe Endpoint controller driver");
> +MODULE_LICENSE("GPL v2");
> -- 
> 2.25.1
>