From: Hans Zhang <hans.zhang@cixtech.com>
Add driver for the CIX Sky1 SoC PCIe Gen4 16 GT/s controller based
on the Cadence PCIe core.
Supports MSI/MSI-x via GICv3, Single Virtual Channel, Single Function.
Signed-off-by: Hans Zhang <hans.zhang@cixtech.com>
---
drivers/pci/controller/cadence/Kconfig | 13 +
drivers/pci/controller/cadence/Makefile | 1 +
drivers/pci/controller/cadence/pci-sky1.c | 294 ++++++++++++++++++++++
3 files changed, 308 insertions(+)
create mode 100644 drivers/pci/controller/cadence/pci-sky1.c
diff --git a/drivers/pci/controller/cadence/Kconfig b/drivers/pci/controller/cadence/Kconfig
index 117677a23d68..1716b7300fa1 100644
--- a/drivers/pci/controller/cadence/Kconfig
+++ b/drivers/pci/controller/cadence/Kconfig
@@ -67,4 +67,17 @@ config PCI_J721E_EP
Say Y here if you want to support the TI J721E PCIe platform
controller in endpoint mode. TI J721E PCIe controller uses Cadence PCIe
core.
+
+config PCI_SKY1_HOST
+ tristate "CIX SKY1 PCIe controller (host mode)"
+ depends on OF
+ select PCIE_CADENCE_HOST
+ select PCI_ECAM
+ help
+ Say Y here if you want to support the CIX SKY1 PCIe platform
+ controller in host mode. CIX SKY1 PCIe controller uses Cadence HPA(High
+ Performance Architecture IP[Second generation of cadence PCIe IP])
+
+ This driver requires Cadence PCIe core infrastructure (PCIE_CADENCE_HOST)
+ and hardware platform adaptation layer to function.
endmenu
diff --git a/drivers/pci/controller/cadence/Makefile b/drivers/pci/controller/cadence/Makefile
index de4ddae7aca4..40d7c6e98b4d 100644
--- a/drivers/pci/controller/cadence/Makefile
+++ b/drivers/pci/controller/cadence/Makefile
@@ -8,3 +8,4 @@ obj-$(CONFIG_PCIE_CADENCE_HOST) += pcie-cadence-host-mod.o
obj-$(CONFIG_PCIE_CADENCE_EP) += pcie-cadence-ep-mod.o
obj-$(CONFIG_PCIE_CADENCE_PLAT) += pcie-cadence-plat.o
obj-$(CONFIG_PCI_J721E) += pci-j721e.o
+obj-$(CONFIG_PCI_SKY1_HOST) += pci-sky1.o
diff --git a/drivers/pci/controller/cadence/pci-sky1.c b/drivers/pci/controller/cadence/pci-sky1.c
new file mode 100644
index 000000000000..36267d0bb5b6
--- /dev/null
+++ b/drivers/pci/controller/cadence/pci-sky1.c
@@ -0,0 +1,294 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * PCIe controller driver for CIX's sky1 SoCs
+ *
+ * Author: Hans Zhang <hans.zhang@cixtech.com>
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/of_device.h>
+#include <linux/pci.h>
+#include <linux/pci-ecam.h>
+#include <linux/pci_ids.h>
+
+#include "pcie-cadence.h"
+#include "pcie-cadence-host-common.h"
+
+#define STRAP_REG(n) ((n) * 0x04)
+#define STATUS_REG(n) ((n) * 0x04)
+
+#define RCSU_STRAP_REG 0x300
+#define RCSU_STATUS_REG 0x400
+
+#define SKY1_IP_REG_BANK_OFFSET 0x1000
+#define SKY1_IP_CFG_CTRL_REG_BANK_OFFSET 0x4c00
+#define SKY1_IP_AXI_MASTER_COMMON_OFFSET 0xf000
+#define SKY1_AXI_SLAVE_OFFSET 0x9000
+#define SKY1_AXI_MASTER_OFFSET 0xb000
+#define SKY1_AXI_HLS_REGISTERS_OFFSET 0xc000
+#define SKY1_AXI_RAS_REGISTERS_OFFSET 0xe000
+#define SKY1_DTI_REGISTERS_OFFSET 0xd000
+
+#define IP_REG_I_DBG_STS_0 0x420
+
+#define LINK_TRAINING_ENABLE BIT(0)
+#define LINK_COMPLETE BIT(0)
+
+enum cix_soc_type {
+ CIX_SKY1,
+};
+
+struct sky1_pcie_data {
+ struct cdns_plat_pcie_of_data reg_off;
+ enum cix_soc_type soc_type;
+};
+
+struct sky1_pcie {
+ struct device *dev;
+ const struct sky1_pcie_data *data;
+ struct cdns_pcie *cdns_pcie;
+ struct cdns_pcie_rc *cdns_pcie_rc;
+
+ struct resource *cfg_res;
+ struct resource *msg_res;
+ struct pci_config_window *cfg;
+ void __iomem *rcsu_base;
+ void __iomem *strap_base;
+ void __iomem *status_base;
+ void __iomem *reg_base;
+ void __iomem *cfg_base;
+ void __iomem *msg_base;
+};
+
+static void sky1_pcie_clear_and_set_dword(void __iomem *addr, u32 clear,
+ u32 set)
+{
+ u32 val;
+
+ val = readl(addr);
+ val &= ~clear;
+ val |= set;
+ writel(val, addr);
+}
+
+static void sky1_pcie_init_bases(struct sky1_pcie *pcie)
+{
+ pcie->strap_base = pcie->rcsu_base + RCSU_STRAP_REG;
+ pcie->status_base = pcie->rcsu_base + RCSU_STATUS_REG;
+}
+
+static int sky1_pcie_parse_mem(struct sky1_pcie *pcie)
+{
+ struct device *dev = pcie->dev;
+ struct platform_device *pdev = to_platform_device(dev);
+ struct resource *res;
+ void __iomem *base;
+ int ret = 0;
+
+ base = devm_platform_ioremap_resource_byname(pdev, "reg");
+ if (IS_ERR(base)) {
+ dev_err(dev, "Parse \"reg\" resource err\n");
+ return PTR_ERR(base);
+ }
+ pcie->reg_base = base;
+
+ res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "cfg");
+ if (!res) {
+ dev_err(dev, "Parse \"cfg\" resource err\n");
+ return -ENXIO;
+ }
+ pcie->cfg_res = res;
+
+ res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "rcsu");
+ if (!res) {
+ dev_err(dev, "Parse \"rcsu\" resource err\n");
+ return -ENXIO;
+ }
+ pcie->rcsu_base = devm_ioremap(dev, res->start, resource_size(res));
+ if (!pcie->rcsu_base) {
+ dev_err(dev, "ioremap failed for resource %pR\n", res);
+ return -ENOMEM;
+ }
+
+ res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "msg");
+ if (!res) {
+ dev_err(dev, "Parse \"msg\" resource err\n");
+ return -ENXIO;
+ }
+ pcie->msg_res = res;
+ pcie->msg_base = devm_ioremap(dev, res->start, resource_size(res));
+ if (!pcie->msg_base) {
+ dev_err(dev, "ioremap failed for resource %pR\n", res);
+ return -ENOMEM;
+ }
+
+ return ret;
+}
+
+static int sky1_pcie_parse_property(struct platform_device *pdev,
+ struct sky1_pcie *pcie)
+{
+ int ret = 0;
+
+ ret = sky1_pcie_parse_mem(pcie);
+ if (ret < 0)
+ return ret;
+
+ sky1_pcie_init_bases(pcie);
+
+ return ret;
+}
+
+static int sky1_pcie_start_link(struct cdns_pcie *cdns_pcie)
+{
+ struct sky1_pcie *pcie = dev_get_drvdata(cdns_pcie->dev);
+
+ sky1_pcie_clear_and_set_dword(pcie->strap_base + STRAP_REG(1),
+ 0, LINK_TRAINING_ENABLE);
+
+ return 0;
+}
+
+static void sky1_pcie_stop_link(struct cdns_pcie *cdns_pcie)
+{
+ struct sky1_pcie *pcie = dev_get_drvdata(cdns_pcie->dev);
+
+ sky1_pcie_clear_and_set_dword(pcie->strap_base + STRAP_REG(1),
+ LINK_TRAINING_ENABLE, 0);
+}
+
+
+static bool sky1_pcie_link_up(struct cdns_pcie *cdns_pcie)
+{
+ u32 val;
+
+ val = cdns_pcie_hpa_readl(cdns_pcie, REG_BANK_IP_REG,
+ IP_REG_I_DBG_STS_0);
+ return val & LINK_COMPLETE;
+}
+
+static const struct cdns_pcie_ops sky1_pcie_ops = {
+ .start_link = sky1_pcie_start_link,
+ .stop_link = sky1_pcie_stop_link,
+ .link_up = sky1_pcie_link_up,
+};
+
+static int sky1_pcie_probe(struct platform_device *pdev)
+{
+ const struct sky1_pcie_data *data;
+ struct device *dev = &pdev->dev;
+ struct pci_host_bridge *bridge;
+ struct cdns_pcie *cdns_pcie;
+ struct resource_entry *bus;
+ struct cdns_pcie_rc *rc;
+ struct sky1_pcie *pcie;
+ int ret;
+
+ pcie = devm_kzalloc(dev, sizeof(*pcie), GFP_KERNEL);
+ if (!pcie)
+ return -ENOMEM;
+
+ data = of_device_get_match_data(dev);
+ if (!data)
+ return -EINVAL;
+
+ pcie->data = data;
+ pcie->dev = dev;
+ dev_set_drvdata(dev, pcie);
+
+ bridge = devm_pci_alloc_host_bridge(dev, sizeof(*rc));
+ if (!bridge)
+ return -ENOMEM;
+
+ bus = resource_list_first_type(&bridge->windows, IORESOURCE_BUS);
+ if (!bus)
+ return -ENODEV;
+
+ ret = sky1_pcie_parse_property(pdev, pcie);
+ if (ret < 0)
+ return -ENXIO;
+
+ pcie->cfg = pci_ecam_create(dev, pcie->cfg_res, bus->res,
+ &pci_generic_ecam_ops);
+ if (IS_ERR(pcie->cfg))
+ return PTR_ERR(pcie->cfg);
+
+ bridge->ops = (struct pci_ops *)&pci_generic_ecam_ops.pci_ops;
+ rc = pci_host_bridge_priv(bridge);
+ rc->ecam_support_flag = 1;
+ rc->cfg_base = pcie->cfg->win;
+ rc->cfg_res = &pcie->cfg->res;
+
+ cdns_pcie = &rc->pcie;
+ cdns_pcie->dev = dev;
+ cdns_pcie->ops = &sky1_pcie_ops;
+ cdns_pcie->reg_base = pcie->reg_base;
+ cdns_pcie->msg_res = pcie->msg_res;
+ cdns_pcie->cdns_pcie_reg_offsets = &data->reg_off;
+ cdns_pcie->is_rc = data->reg_off.is_rc;
+
+ pcie->cdns_pcie = cdns_pcie;
+ pcie->cdns_pcie_rc = rc;
+ pcie->cfg_base = rc->cfg_base;
+ bridge->sysdata = pcie->cfg;
+
+ if (data->soc_type == CIX_SKY1) {
+ rc->vendor_id = PCI_VENDOR_ID_CIX;
+ rc->device_id = PCI_DEVICE_ID_CIX_SKY1;
+ rc->no_inbound_flag = 1;
+ }
+
+ ret = cdns_pcie_hpa_host_setup(rc);
+ if (ret < 0) {
+ pci_ecam_free(pcie->cfg);
+ return ret;
+ }
+
+ return 0;
+}
+
+static const struct sky1_pcie_data sky1_pcie_rc_data = {
+ .reg_off = {
+ .is_rc = true,
+ .ip_reg_bank_offset = SKY1_IP_REG_BANK_OFFSET,
+ .ip_cfg_ctrl_reg_offset = SKY1_IP_CFG_CTRL_REG_BANK_OFFSET,
+ .axi_mstr_common_offset = SKY1_IP_AXI_MASTER_COMMON_OFFSET,
+ .axi_slave_offset = SKY1_AXI_SLAVE_OFFSET,
+ .axi_master_offset = SKY1_AXI_MASTER_OFFSET,
+ .axi_hls_offset = SKY1_AXI_HLS_REGISTERS_OFFSET,
+ .axi_ras_offset = SKY1_AXI_RAS_REGISTERS_OFFSET,
+ .axi_dti_offset = SKY1_DTI_REGISTERS_OFFSET,
+ },
+ .soc_type = CIX_SKY1,
+};
+
+static const struct of_device_id of_sky1_pcie_match[] = {
+ {
+ .compatible = "cix,sky1-pcie-host",
+ .data = &sky1_pcie_rc_data,
+ },
+ {},
+};
+
+static void sky1_pcie_remove(struct platform_device *pdev)
+{
+ struct sky1_pcie *pcie = platform_get_drvdata(pdev);
+
+ pci_ecam_free(pcie->cfg);
+}
+
+static struct platform_driver sky1_pcie_driver = {
+ .probe = sky1_pcie_probe,
+ .remove = sky1_pcie_remove,
+ .driver = {
+ .name = "sky1-pcie",
+ .of_match_table = of_sky1_pcie_match,
+ },
+};
+module_platform_driver(sky1_pcie_driver);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("PCIe controller driver for CIX's sky1 SoCs");
+MODULE_AUTHOR("Hans Zhang <hans.zhang@cixtech.com>");
--
2.49.0On Wed, Aug 13, 2025 at 12:23:28PM +0800, hans.zhang@cixtech.com wrote:
> From: Hans Zhang <hans.zhang@cixtech.com>
>
> Add driver for the CIX Sky1 SoC PCIe Gen4 16 GT/s controller based
> on the Cadence PCIe core.
>
> Supports MSI/MSI-x via GICv3, Single Virtual Channel, Single Function.
> +config PCI_SKY1_HOST
> + tristate "CIX SKY1 PCIe controller (host mode)"
> + depends on OF
> + select PCIE_CADENCE_HOST
> + select PCI_ECAM
> + help
> + Say Y here if you want to support the CIX SKY1 PCIe platform
> + controller in host mode. CIX SKY1 PCIe controller uses Cadence HPA(High
> + Performance Architecture IP[Second generation of cadence PCIe IP])
> +
> + This driver requires Cadence PCIe core infrastructure (PCIE_CADENCE_HOST)
> + and hardware platform adaptation layer to function.
Reorder so this menu entry appears in alphabetical order by vendor.
Add space in "HPA(..".
Add space in "IP[Second ..."
s/cadence PCIe IP/Cadence PCIe IP/
Rewrap so help text fits in 80 columns when displayed by menuconfig.
> +++ b/drivers/pci/controller/cadence/Makefile
> +++ b/drivers/pci/controller/cadence/pci-sky1.c
> @@ -0,0 +1,294 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * PCIe controller driver for CIX's sky1 SoCs
> + *
> + * Author: Hans Zhang <hans.zhang@cixtech.com>
> + */
The typical comment style, thank you :)
> +#include <linux/kernel.h>
> +#include <linux/module.h>
> +#include <linux/of.h>
> +#include <linux/of_device.h>
> +#include <linux/pci.h>
> +#include <linux/pci-ecam.h>
> +#include <linux/pci_ids.h>
> +
> +#include "pcie-cadence.h"
> +#include "pcie-cadence-host-common.h"
> +
> +#define STRAP_REG(n) ((n) * 0x04)
> +#define STATUS_REG(n) ((n) * 0x04)
> +
> +#define RCSU_STRAP_REG 0x300
> +#define RCSU_STATUS_REG 0x400
> +
> +#define SKY1_IP_REG_BANK_OFFSET 0x1000
> +#define SKY1_IP_CFG_CTRL_REG_BANK_OFFSET 0x4c00
> +#define SKY1_IP_AXI_MASTER_COMMON_OFFSET 0xf000
> +#define SKY1_AXI_SLAVE_OFFSET 0x9000
> +#define SKY1_AXI_MASTER_OFFSET 0xb000
> +#define SKY1_AXI_HLS_REGISTERS_OFFSET 0xc000
> +#define SKY1_AXI_RAS_REGISTERS_OFFSET 0xe000
> +#define SKY1_DTI_REGISTERS_OFFSET 0xd000
Indent the values to line up.
Consider dropping "_OFFSET".
> +#define IP_REG_I_DBG_STS_0 0x420
> +
> +#define LINK_TRAINING_ENABLE BIT(0)
> +#define LINK_COMPLETE BIT(0)
Define next to the offset of the register that contains these.
Consider naming them to relate to the register.
> +enum cix_soc_type {
> + CIX_SKY1,
> +};
> +
> +struct sky1_pcie_data {
> + struct cdns_plat_pcie_of_data reg_off;
Weird name for this member. cdns_plat_pcie_of_data contains more than
just offsets.
> + enum cix_soc_type soc_type;
> +};
> +
> +struct sky1_pcie {
> + struct device *dev;
> + const struct sky1_pcie_data *data;
> + struct cdns_pcie *cdns_pcie;
Typically first in the driver struct. Also cdns_pcie already contains
a struct device *; do you need another?
> + struct cdns_pcie_rc *cdns_pcie_rc;
> +
> + struct resource *cfg_res;
> + struct resource *msg_res;
> + struct pci_config_window *cfg;
> + void __iomem *rcsu_base;
> + void __iomem *strap_base;
> + void __iomem *status_base;
> + void __iomem *reg_base;
> + void __iomem *cfg_base;
> + void __iomem *msg_base;
> +};
> +
> +static void sky1_pcie_clear_and_set_dword(void __iomem *addr, u32 clear,
> + u32 set)
> +{
> + u32 val;
> +
> + val = readl(addr);
> + val &= ~clear;
> + val |= set;
> + writel(val, addr);
Nothing specific to sky1 here. Find an existing interface or copy the
style of other drivers. Surely other drivers do something similar,
but "git grep clear_and_set_dword drivers/pci/controller/" finds
nothing.
> +static int sky1_pcie_parse_mem(struct sky1_pcie *pcie)
> +{
> + struct device *dev = pcie->dev;
> + struct platform_device *pdev = to_platform_device(dev);
> + struct resource *res;
> + void __iomem *base;
> + int ret = 0;
> +
> + base = devm_platform_ioremap_resource_byname(pdev, "reg");
> + if (IS_ERR(base)) {
> + dev_err(dev, "Parse \"reg\" resource err\n");
> + return PTR_ERR(base);
> + }
> + pcie->reg_base = base;
> +
> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "cfg");
> + if (!res) {
> + dev_err(dev, "Parse \"cfg\" resource err\n");
> + return -ENXIO;
> + }
> + pcie->cfg_res = res;
> +
> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "rcsu");
> + if (!res) {
> + dev_err(dev, "Parse \"rcsu\" resource err\n");
> + return -ENXIO;
> + }
> + pcie->rcsu_base = devm_ioremap(dev, res->start, resource_size(res));
> + if (!pcie->rcsu_base) {
> + dev_err(dev, "ioremap failed for resource %pR\n", res);
> + return -ENOMEM;
> + }
> +
> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "msg");
> + if (!res) {
> + dev_err(dev, "Parse \"msg\" resource err\n");
> + return -ENXIO;
> + }
> + pcie->msg_res = res;
> + pcie->msg_base = devm_ioremap(dev, res->start, resource_size(res));
> + if (!pcie->msg_base) {
> + dev_err(dev, "ioremap failed for resource %pR\n", res);
> + return -ENOMEM;
> + }
> +
> + return ret;
"ret" is never used; remove it and just "return 0" here.
> +}
> +
> +static int sky1_pcie_parse_property(struct platform_device *pdev,
> + struct sky1_pcie *pcie)
Find a similar function in other drivers and copy the name.
"git grep parse_property drivers/pci/controller/" finds nothing.
> +{
> + int ret = 0;
> +
> + ret = sky1_pcie_parse_mem(pcie);
> + if (ret < 0)
> + return ret;
> +
> + sky1_pcie_init_bases(pcie);
> +
> + return ret;
Drop "ret" and return 0 here.
> +}
> +
> +static int sky1_pcie_start_link(struct cdns_pcie *cdns_pcie)
> +{
> + struct sky1_pcie *pcie = dev_get_drvdata(cdns_pcie->dev);
> +
> + sky1_pcie_clear_and_set_dword(pcie->strap_base + STRAP_REG(1),
> + 0, LINK_TRAINING_ENABLE);
> +
> + return 0;
> +}
> +
> +static void sky1_pcie_stop_link(struct cdns_pcie *cdns_pcie)
> +{
> + struct sky1_pcie *pcie = dev_get_drvdata(cdns_pcie->dev);
> +
> + sky1_pcie_clear_and_set_dword(pcie->strap_base + STRAP_REG(1),
> + LINK_TRAINING_ENABLE, 0);
> +}
> +
> +
> +static bool sky1_pcie_link_up(struct cdns_pcie *cdns_pcie)
> +{
> + u32 val;
> +
> + val = cdns_pcie_hpa_readl(cdns_pcie, REG_BANK_IP_REG,
> + IP_REG_I_DBG_STS_0);
> + return val & LINK_COMPLETE;
> +}
> +
> +static const struct cdns_pcie_ops sky1_pcie_ops = {
> + .start_link = sky1_pcie_start_link,
> + .stop_link = sky1_pcie_stop_link,
> + .link_up = sky1_pcie_link_up,
> +};
> +
> +static int sky1_pcie_probe(struct platform_device *pdev)
> +{
> + const struct sky1_pcie_data *data;
> + struct device *dev = &pdev->dev;
> + struct pci_host_bridge *bridge;
> + struct cdns_pcie *cdns_pcie;
> + struct resource_entry *bus;
> + struct cdns_pcie_rc *rc;
> + struct sky1_pcie *pcie;
> + int ret;
> +
> + pcie = devm_kzalloc(dev, sizeof(*pcie), GFP_KERNEL);
> + if (!pcie)
> + return -ENOMEM;
> +
> + data = of_device_get_match_data(dev);
> + if (!data)
> + return -EINVAL;
> +
> + pcie->data = data;
> + pcie->dev = dev;
> + dev_set_drvdata(dev, pcie);
> +
> + bridge = devm_pci_alloc_host_bridge(dev, sizeof(*rc));
> + if (!bridge)
> + return -ENOMEM;
> +
> + bus = resource_list_first_type(&bridge->windows, IORESOURCE_BUS);
> + if (!bus)
> + return -ENODEV;
Maybe put this near the code that uses it instead of sticking the
unrelated sky1_pcie_parse_property() in the middle?
> + ret = sky1_pcie_parse_property(pdev, pcie);
> + if (ret < 0)
> + return -ENXIO;
> +
> + pcie->cfg = pci_ecam_create(dev, pcie->cfg_res, bus->res,
> + &pci_generic_ecam_ops);
> + if (IS_ERR(pcie->cfg))
> + return PTR_ERR(pcie->cfg);
> +
> + bridge->ops = (struct pci_ops *)&pci_generic_ecam_ops.pci_ops;
> + rc = pci_host_bridge_priv(bridge);
> + rc->ecam_support_flag = 1;
> + rc->cfg_base = pcie->cfg->win;
> + rc->cfg_res = &pcie->cfg->res;
> +
> + cdns_pcie = &rc->pcie;
> + cdns_pcie->dev = dev;
> + cdns_pcie->ops = &sky1_pcie_ops;
> + cdns_pcie->reg_base = pcie->reg_base;
> + cdns_pcie->msg_res = pcie->msg_res;
> + cdns_pcie->cdns_pcie_reg_offsets = &data->reg_off;
> + cdns_pcie->is_rc = data->reg_off.is_rc;
> +
> + pcie->cdns_pcie = cdns_pcie;
> + pcie->cdns_pcie_rc = rc;
> + pcie->cfg_base = rc->cfg_base;
> + bridge->sysdata = pcie->cfg;
> +
> + if (data->soc_type == CIX_SKY1) {
> + rc->vendor_id = PCI_VENDOR_ID_CIX;
> + rc->device_id = PCI_DEVICE_ID_CIX_SKY1;
> + rc->no_inbound_flag = 1;
> + }
> +
> + ret = cdns_pcie_hpa_host_setup(rc);
> + if (ret < 0) {
> + pci_ecam_free(pcie->cfg);
> + return ret;
> + }
> +
> + return 0;
> +}
> +
> +static const struct sky1_pcie_data sky1_pcie_rc_data = {
> + .reg_off = {
> + .is_rc = true,
> + .ip_reg_bank_offset = SKY1_IP_REG_BANK_OFFSET,
> + .ip_cfg_ctrl_reg_offset = SKY1_IP_CFG_CTRL_REG_BANK_OFFSET,
> + .axi_mstr_common_offset = SKY1_IP_AXI_MASTER_COMMON_OFFSET,
> + .axi_slave_offset = SKY1_AXI_SLAVE_OFFSET,
> + .axi_master_offset = SKY1_AXI_MASTER_OFFSET,
> + .axi_hls_offset = SKY1_AXI_HLS_REGISTERS_OFFSET,
> + .axi_ras_offset = SKY1_AXI_RAS_REGISTERS_OFFSET,
> + .axi_dti_offset = SKY1_DTI_REGISTERS_OFFSET,
> + },
> + .soc_type = CIX_SKY1,
> +};
> +
> +static const struct of_device_id of_sky1_pcie_match[] = {
> + {
> + .compatible = "cix,sky1-pcie-host",
> + .data = &sky1_pcie_rc_data,
> + },
> + {},
> +};
> +
> +static void sky1_pcie_remove(struct platform_device *pdev)
> +{
> + struct sky1_pcie *pcie = platform_get_drvdata(pdev);
> +
> + pci_ecam_free(pcie->cfg);
> +}
> +
> +static struct platform_driver sky1_pcie_driver = {
> + .probe = sky1_pcie_probe,
> + .remove = sky1_pcie_remove,
> + .driver = {
> + .name = "sky1-pcie",
> + .of_match_table = of_sky1_pcie_match,
> + },
> +};
> +module_platform_driver(sky1_pcie_driver);
> +
> +MODULE_LICENSE("GPL");
> +MODULE_DESCRIPTION("PCIe controller driver for CIX's sky1 SoCs");
> +MODULE_AUTHOR("Hans Zhang <hans.zhang@cixtech.com>");
> --
> 2.49.0
>
On 2025/8/15 06:46, Bjorn Helgaas wrote:
> EXTERNAL EMAIL
>
> On Wed, Aug 13, 2025 at 12:23:28PM +0800, hans.zhang@cixtech.com wrote:
>> From: Hans Zhang <hans.zhang@cixtech.com>
>>
>> Add driver for the CIX Sky1 SoC PCIe Gen4 16 GT/s controller based
>> on the Cadence PCIe core.
>>
>> Supports MSI/MSI-x via GICv3, Single Virtual Channel, Single Function.
>
>> +config PCI_SKY1_HOST
>> + tristate "CIX SKY1 PCIe controller (host mode)"
>> + depends on OF
>> + select PCIE_CADENCE_HOST
>> + select PCI_ECAM
>> + help
>> + Say Y here if you want to support the CIX SKY1 PCIe platform
>> + controller in host mode. CIX SKY1 PCIe controller uses Cadence HPA(High
>> + Performance Architecture IP[Second generation of cadence PCIe IP])
>> +
>> + This driver requires Cadence PCIe core infrastructure (PCIE_CADENCE_HOST)
>> + and hardware platform adaptation layer to function.
>
> Reorder so this menu entry appears in alphabetical order by vendor.
>
> Add space in "HPA(..".
>
> Add space in "IP[Second ..."
>
> s/cadence PCIe IP/Cadence PCIe IP/
>
> Rewrap so help text fits in 80 columns when displayed by menuconfig.
Dear Bjorn,
Thank you very much for your reply. Will change.
>
>> +++ b/drivers/pci/controller/cadence/Makefile
>> +++ b/drivers/pci/controller/cadence/pci-sky1.c
>> @@ -0,0 +1,294 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +/*
>> + * PCIe controller driver for CIX's sky1 SoCs
>> + *
>> + * Author: Hans Zhang <hans.zhang@cixtech.com>
>> + */
>
> The typical comment style, thank you :)
>
>> +#include <linux/kernel.h>
>> +#include <linux/module.h>
>> +#include <linux/of.h>
>> +#include <linux/of_device.h>
>> +#include <linux/pci.h>
>> +#include <linux/pci-ecam.h>
>> +#include <linux/pci_ids.h>
>> +
>> +#include "pcie-cadence.h"
>> +#include "pcie-cadence-host-common.h"
>> +
>> +#define STRAP_REG(n) ((n) * 0x04)
>> +#define STATUS_REG(n) ((n) * 0x04)
>> +
>> +#define RCSU_STRAP_REG 0x300
>> +#define RCSU_STATUS_REG 0x400
>> +
>> +#define SKY1_IP_REG_BANK_OFFSET 0x1000
>> +#define SKY1_IP_CFG_CTRL_REG_BANK_OFFSET 0x4c00
>> +#define SKY1_IP_AXI_MASTER_COMMON_OFFSET 0xf000
>> +#define SKY1_AXI_SLAVE_OFFSET 0x9000
>> +#define SKY1_AXI_MASTER_OFFSET 0xb000
>> +#define SKY1_AXI_HLS_REGISTERS_OFFSET 0xc000
>> +#define SKY1_AXI_RAS_REGISTERS_OFFSET 0xe000
>> +#define SKY1_DTI_REGISTERS_OFFSET 0xd000
>
> Indent the values to line up.
>
> Consider dropping "_OFFSET".
Will change.
>
>> +#define IP_REG_I_DBG_STS_0 0x420
>> +
>> +#define LINK_TRAINING_ENABLE BIT(0)
>> +#define LINK_COMPLETE BIT(0)
>
> Define next to the offset of the register that contains these.
> Consider naming them to relate to the register.
>
All of our registers are RCSU (Remote CIX System Unit registers.), and
they are named like this: I moved it below this.
#define STRAP_REG(n) ((n) * 0x04)
#define STATUS_REG(n) ((n) * 0x04)
>> +enum cix_soc_type {
>> + CIX_SKY1,
>> +};
>> +
>> +struct sky1_pcie_data {
>> + struct cdns_plat_pcie_of_data reg_off;
>
> Weird name for this member. cdns_plat_pcie_of_data contains more than
> just offsets.
We are considering deleting cdns_plat_pcie_of_data. Krzysztof also
provided a review opinion, intending to directly assign the register
offset value in the probe.
>
>> + enum cix_soc_type soc_type;
>> +};
>> +
>> +struct sky1_pcie {
>> + struct device *dev;
>> + const struct sky1_pcie_data *data;
>> + struct cdns_pcie *cdns_pcie;
>
> Typically first in the driver struct. Also cdns_pcie already contains
> a struct device *; do you need another?
I checked and it's indeed possible to delete it by using "dev" in cdns_pcie.
>
>> + struct cdns_pcie_rc *cdns_pcie_rc;
>> +
>> + struct resource *cfg_res;
>> + struct resource *msg_res;
>> + struct pci_config_window *cfg;
>> + void __iomem *rcsu_base;
>> + void __iomem *strap_base;
>> + void __iomem *status_base;
>> + void __iomem *reg_base;
>> + void __iomem *cfg_base;
>> + void __iomem *msg_base;
>> +};
>> +
>> +static void sky1_pcie_clear_and_set_dword(void __iomem *addr, u32 clear,
>> + u32 set)
>> +{
>> + u32 val;
>> +
>> + val = readl(addr);
>> + val &= ~clear;
>> + val |= set;
>> + writel(val, addr);
>
> Nothing specific to sky1 here. Find an existing interface or copy the
> style of other drivers. Surely other drivers do something similar,
> but "git grep clear_and_set_dword drivers/pci/controller/" finds
> nothing.
My original intention was for the future upstream, as there are many
repetitive read and write registers. Then I'll delete it first.
>
>> +static int sky1_pcie_parse_mem(struct sky1_pcie *pcie)
>> +{
>> + struct device *dev = pcie->dev;
>> + struct platform_device *pdev = to_platform_device(dev);
>> + struct resource *res;
>> + void __iomem *base;
>> + int ret = 0;
>> +
>> + base = devm_platform_ioremap_resource_byname(pdev, "reg");
>> + if (IS_ERR(base)) {
>> + dev_err(dev, "Parse \"reg\" resource err\n");
>> + return PTR_ERR(base);
>> + }
>> + pcie->reg_base = base;
>> +
>> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "cfg");
>> + if (!res) {
>> + dev_err(dev, "Parse \"cfg\" resource err\n");
>> + return -ENXIO;
>> + }
>> + pcie->cfg_res = res;
>> +
>> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "rcsu");
>> + if (!res) {
>> + dev_err(dev, "Parse \"rcsu\" resource err\n");
>> + return -ENXIO;
>> + }
>> + pcie->rcsu_base = devm_ioremap(dev, res->start, resource_size(res));
>> + if (!pcie->rcsu_base) {
>> + dev_err(dev, "ioremap failed for resource %pR\n", res);
>> + return -ENOMEM;
>> + }
>> +
>> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "msg");
>> + if (!res) {
>> + dev_err(dev, "Parse \"msg\" resource err\n");
>> + return -ENXIO;
>> + }
>> + pcie->msg_res = res;
>> + pcie->msg_base = devm_ioremap(dev, res->start, resource_size(res));
>> + if (!pcie->msg_base) {
>> + dev_err(dev, "ioremap failed for resource %pR\n", res);
>> + return -ENOMEM;
>> + }
>> +
>> + return ret;
>
> "ret" is never used; remove it and just "return 0" here.
Will change.
>
>> +}
>> +
>> +static int sky1_pcie_parse_property(struct platform_device *pdev,
>> + struct sky1_pcie *pcie)
>
> Find a similar function in other drivers and copy the name.
> "git grep parse_property drivers/pci/controller/" finds nothing.
Will change.
>
>> +{
>> + int ret = 0;
>> +
>> + ret = sky1_pcie_parse_mem(pcie);
>> + if (ret < 0)
>> + return ret;
>> +
>> + sky1_pcie_init_bases(pcie);
>> +
>> + return ret;
>
> Drop "ret" and return 0 here.
Will change.
>
>> +}
>> +
>> +static int sky1_pcie_start_link(struct cdns_pcie *cdns_pcie)
>> +{
>> + struct sky1_pcie *pcie = dev_get_drvdata(cdns_pcie->dev);
>> +
>> + sky1_pcie_clear_and_set_dword(pcie->strap_base + STRAP_REG(1),
>> + 0, LINK_TRAINING_ENABLE);
>> +
>> + return 0;
>> +}
>> +
>> +static void sky1_pcie_stop_link(struct cdns_pcie *cdns_pcie)
>> +{
>> + struct sky1_pcie *pcie = dev_get_drvdata(cdns_pcie->dev);
>> +
>> + sky1_pcie_clear_and_set_dword(pcie->strap_base + STRAP_REG(1),
>> + LINK_TRAINING_ENABLE, 0);
>> +}
>> +
>> +
>> +static bool sky1_pcie_link_up(struct cdns_pcie *cdns_pcie)
>> +{
>> + u32 val;
>> +
>> + val = cdns_pcie_hpa_readl(cdns_pcie, REG_BANK_IP_REG,
>> + IP_REG_I_DBG_STS_0);
>> + return val & LINK_COMPLETE;
>> +}
>> +
>> +static const struct cdns_pcie_ops sky1_pcie_ops = {
>> + .start_link = sky1_pcie_start_link,
>> + .stop_link = sky1_pcie_stop_link,
>> + .link_up = sky1_pcie_link_up,
>> +};
>> +
>> +static int sky1_pcie_probe(struct platform_device *pdev)
>> +{
>> + const struct sky1_pcie_data *data;
>> + struct device *dev = &pdev->dev;
>> + struct pci_host_bridge *bridge;
>> + struct cdns_pcie *cdns_pcie;
>> + struct resource_entry *bus;
>> + struct cdns_pcie_rc *rc;
>> + struct sky1_pcie *pcie;
>> + int ret;
>> +
>> + pcie = devm_kzalloc(dev, sizeof(*pcie), GFP_KERNEL);
>> + if (!pcie)
>> + return -ENOMEM;
>> +
>> + data = of_device_get_match_data(dev);
>> + if (!data)
>> + return -EINVAL;
>> +
>> + pcie->data = data;
>> + pcie->dev = dev;
>> + dev_set_drvdata(dev, pcie);
>> +
>> + bridge = devm_pci_alloc_host_bridge(dev, sizeof(*rc));
>> + if (!bridge)
>> + return -ENOMEM;
>> +
>> + bus = resource_list_first_type(&bridge->windows, IORESOURCE_BUS);
>> + if (!bus)
>> + return -ENODEV;
>
> Maybe put this near the code that uses it instead of sticking the
> unrelated sky1_pcie_parse_property() in the middle?
Will change.
>
>> + ret = sky1_pcie_parse_property(pdev, pcie);
>> + if (ret < 0)
>> + return -ENXIO;
>> +
>> + pcie->cfg = pci_ecam_create(dev, pcie->cfg_res, bus->res,
>> + &pci_generic_ecam_ops);
>> + if (IS_ERR(pcie->cfg))
>> + return PTR_ERR(pcie->cfg);
>> +
>> + bridge->ops = (struct pci_ops *)&pci_generic_ecam_ops.pci_ops;
>> + rc = pci_host_bridge_priv(bridge);
>> + rc->ecam_support_flag = 1;
>> + rc->cfg_base = pcie->cfg->win;
>> + rc->cfg_res = &pcie->cfg->res;
>> +
>> + cdns_pcie = &rc->pcie;
>> + cdns_pcie->dev = dev;
>> + cdns_pcie->ops = &sky1_pcie_ops;
>> + cdns_pcie->reg_base = pcie->reg_base;
>> + cdns_pcie->msg_res = pcie->msg_res;
>> + cdns_pcie->cdns_pcie_reg_offsets = &data->reg_off;
>> + cdns_pcie->is_rc = data->reg_off.is_rc;
>> +
>> + pcie->cdns_pcie = cdns_pcie;
>> + pcie->cdns_pcie_rc = rc;
>> + pcie->cfg_base = rc->cfg_base;
>> + bridge->sysdata = pcie->cfg;
>> +
>> + if (data->soc_type == CIX_SKY1) {
>> + rc->vendor_id = PCI_VENDOR_ID_CIX;
>> + rc->device_id = PCI_DEVICE_ID_CIX_SKY1;
>> + rc->no_inbound_flag = 1;
>> + }
>> +
>> + ret = cdns_pcie_hpa_host_setup(rc);
>> + if (ret < 0) {
>> + pci_ecam_free(pcie->cfg);
>> + return ret;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static const struct sky1_pcie_data sky1_pcie_rc_data = {
>> + .reg_off = {
>> + .is_rc = true,
>> + .ip_reg_bank_offset = SKY1_IP_REG_BANK_OFFSET,
>> + .ip_cfg_ctrl_reg_offset = SKY1_IP_CFG_CTRL_REG_BANK_OFFSET,
>> + .axi_mstr_common_offset = SKY1_IP_AXI_MASTER_COMMON_OFFSET,
>> + .axi_slave_offset = SKY1_AXI_SLAVE_OFFSET,
>> + .axi_master_offset = SKY1_AXI_MASTER_OFFSET,
>> + .axi_hls_offset = SKY1_AXI_HLS_REGISTERS_OFFSET,
>> + .axi_ras_offset = SKY1_AXI_RAS_REGISTERS_OFFSET,
>> + .axi_dti_offset = SKY1_DTI_REGISTERS_OFFSET,
>> + },
>> + .soc_type = CIX_SKY1,
>> +};
>> +
>> +static const struct of_device_id of_sky1_pcie_match[] = {
>> + {
>> + .compatible = "cix,sky1-pcie-host",
>> + .data = &sky1_pcie_rc_data,
>> + },
>> + {},
>> +};
>> +
>> +static void sky1_pcie_remove(struct platform_device *pdev)
>> +{
>> + struct sky1_pcie *pcie = platform_get_drvdata(pdev);
>> +
>> + pci_ecam_free(pcie->cfg);
>> +}
>> +
>> +static struct platform_driver sky1_pcie_driver = {
>> + .probe = sky1_pcie_probe,
>> + .remove = sky1_pcie_remove,
>> + .driver = {
>> + .name = "sky1-pcie",
>> + .of_match_table = of_sky1_pcie_match,
>> + },
>> +};
>> +module_platform_driver(sky1_pcie_driver);
>> +
>> +MODULE_LICENSE("GPL");
>> +MODULE_DESCRIPTION("PCIe controller driver for CIX's sky1 SoCs");
>> +MODULE_AUTHOR("Hans Zhang <hans.zhang@cixtech.com>");
>> --
>> 2.49.0
>>
Best regards,
Hans
On 13/08/2025 06:23, hans.zhang@cixtech.com wrote:
> +static int sky1_pcie_parse_mem(struct sky1_pcie *pcie)
> +{
> + struct device *dev = pcie->dev;
> + struct platform_device *pdev = to_platform_device(dev);
> + struct resource *res;
> + void __iomem *base;
> + int ret = 0;
> +
> + base = devm_platform_ioremap_resource_byname(pdev, "reg");
> + if (IS_ERR(base)) {
> + dev_err(dev, "Parse \"reg\" resource err\n");
Syntax is return dev_err_probe, and without \" so grepping works
correctly (see coding style).
> + return PTR_ERR(base);
> + }
> + pcie->reg_base = base;
> +
> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "cfg");
> + if (!res) {
> + dev_err(dev, "Parse \"cfg\" resource err\n");
> + return -ENXIO;
> + }
> + pcie->cfg_res = res;
> +
> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "rcsu");
> + if (!res) {
> + dev_err(dev, "Parse \"rcsu\" resource err\n");
> + return -ENXIO;
> + }
> + pcie->rcsu_base = devm_ioremap(dev, res->start, resource_size(res));
Why aren't you using wrapper over get_resource and ioremap? Isn't
devm_platform_ioremap_resource_byname exactly what you want?
And if argument from previous versions was - you need to backport it for
ancient 3.10 kernel - then it would be a no go.
> + if (!pcie->rcsu_base) {
> + dev_err(dev, "ioremap failed for resource %pR\n", res);
> + return -ENOMEM;
> + }
> +
> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "msg");
> + if (!res) {
> + dev_err(dev, "Parse \"msg\" resource err\n");
> + return -ENXIO;
> + }
> + pcie->msg_res = res;
> + pcie->msg_base = devm_ioremap(dev, res->start, resource_size(res));
> + if (!pcie->msg_base) {
> + dev_err(dev, "ioremap failed for resource %pR\n", res);
> + return -ENOMEM;
> + }
> +
> + return ret;
> +}
> +
> +static int sky1_pcie_parse_property(struct platform_device *pdev,
> + struct sky1_pcie *pcie)
> +{
> + int ret = 0;
> +
> + ret = sky1_pcie_parse_mem(pcie);
> + if (ret < 0)
> + return ret;
> +
> + sky1_pcie_init_bases(pcie);
> +
> + return ret;
> +}
> +
> +static int sky1_pcie_start_link(struct cdns_pcie *cdns_pcie)
> +{
> + struct sky1_pcie *pcie = dev_get_drvdata(cdns_pcie->dev);
> +
> + sky1_pcie_clear_and_set_dword(pcie->strap_base + STRAP_REG(1),
> + 0, LINK_TRAINING_ENABLE);
> +
> + return 0;
> +}
> +
> +static void sky1_pcie_stop_link(struct cdns_pcie *cdns_pcie)
> +{
> + struct sky1_pcie *pcie = dev_get_drvdata(cdns_pcie->dev);
> +
> + sky1_pcie_clear_and_set_dword(pcie->strap_base + STRAP_REG(1),
> + LINK_TRAINING_ENABLE, 0);
> +}
> +
> +
> +static bool sky1_pcie_link_up(struct cdns_pcie *cdns_pcie)
> +{
> + u32 val;
> +
> + val = cdns_pcie_hpa_readl(cdns_pcie, REG_BANK_IP_REG,
> + IP_REG_I_DBG_STS_0);
> + return val & LINK_COMPLETE;
> +}
> +
> +static const struct cdns_pcie_ops sky1_pcie_ops = {
> + .start_link = sky1_pcie_start_link,
> + .stop_link = sky1_pcie_stop_link,
> + .link_up = sky1_pcie_link_up,
> +};
> +
> +static int sky1_pcie_probe(struct platform_device *pdev)
> +{
> + const struct sky1_pcie_data *data;
> + struct device *dev = &pdev->dev;
> + struct pci_host_bridge *bridge;
> + struct cdns_pcie *cdns_pcie;
> + struct resource_entry *bus;
> + struct cdns_pcie_rc *rc;
> + struct sky1_pcie *pcie;
> + int ret;
> +
> + pcie = devm_kzalloc(dev, sizeof(*pcie), GFP_KERNEL);
> + if (!pcie)
> + return -ENOMEM;
> +
> + data = of_device_get_match_data(dev);
> + if (!data)
> + return -EINVAL;
> +
> + pcie->data = data;
> + pcie->dev = dev;
> + dev_set_drvdata(dev, pcie);
> +
> + bridge = devm_pci_alloc_host_bridge(dev, sizeof(*rc));
> + if (!bridge)
> + return -ENOMEM;
> +
> + bus = resource_list_first_type(&bridge->windows, IORESOURCE_BUS);
> + if (!bus)
> + return -ENODEV;
> +
> + ret = sky1_pcie_parse_property(pdev, pcie);
> + if (ret < 0)
> + return -ENXIO;
> +
> + pcie->cfg = pci_ecam_create(dev, pcie->cfg_res, bus->res,
> + &pci_generic_ecam_ops);
> + if (IS_ERR(pcie->cfg))
> + return PTR_ERR(pcie->cfg);
> +
> + bridge->ops = (struct pci_ops *)&pci_generic_ecam_ops.pci_ops;
> + rc = pci_host_bridge_priv(bridge);
> + rc->ecam_support_flag = 1;
> + rc->cfg_base = pcie->cfg->win;
> + rc->cfg_res = &pcie->cfg->res;
> +
> + cdns_pcie = &rc->pcie;
> + cdns_pcie->dev = dev;
> + cdns_pcie->ops = &sky1_pcie_ops;
> + cdns_pcie->reg_base = pcie->reg_base;
> + cdns_pcie->msg_res = pcie->msg_res;
> + cdns_pcie->cdns_pcie_reg_offsets = &data->reg_off;
> + cdns_pcie->is_rc = data->reg_off.is_rc;
> +
> + pcie->cdns_pcie = cdns_pcie;
> + pcie->cdns_pcie_rc = rc;
> + pcie->cfg_base = rc->cfg_base;
> + bridge->sysdata = pcie->cfg;
> +
> + if (data->soc_type == CIX_SKY1) {
Dead code or rather if (true) code. Don't do it, it's more difficult to
read.
> + rc->vendor_id = PCI_VENDOR_ID_CIX;
> + rc->device_id = PCI_DEVICE_ID_CIX_SKY1;
> + rc->no_inbound_flag = 1;
> + }
> +
> + ret = cdns_pcie_hpa_host_setup(rc);
> + if (ret < 0) {
> + pci_ecam_free(pcie->cfg);
> + return ret;
> + }
> +
> + return 0;
> +}
> +
> +static const struct sky1_pcie_data sky1_pcie_rc_data = {
> + .reg_off = {
> + .is_rc = true,
> + .ip_reg_bank_offset = SKY1_IP_REG_BANK_OFFSET,
> + .ip_cfg_ctrl_reg_offset = SKY1_IP_CFG_CTRL_REG_BANK_OFFSET,
> + .axi_mstr_common_offset = SKY1_IP_AXI_MASTER_COMMON_OFFSET,
> + .axi_slave_offset = SKY1_AXI_SLAVE_OFFSET,
> + .axi_master_offset = SKY1_AXI_MASTER_OFFSET,
> + .axi_hls_offset = SKY1_AXI_HLS_REGISTERS_OFFSET,
> + .axi_ras_offset = SKY1_AXI_RAS_REGISTERS_OFFSET,
> + .axi_dti_offset = SKY1_DTI_REGISTERS_OFFSET,
> + },
> + .soc_type = CIX_SKY1,
You have only one device variant, so this entire pcie_data feels redundant.
> +};
> +
> +static const struct of_device_id of_sky1_pcie_match[] = {
> + {
> + .compatible = "cix,sky1-pcie-host",
> + .data = &sky1_pcie_rc_data,
> + },
> + {},
Best regards,
Krzysztof
On 2025/8/14 03:16, Krzysztof Kozlowski wrote:
> EXTERNAL EMAIL
>
> On 13/08/2025 06:23, hans.zhang@cixtech.com wrote:
>> +static int sky1_pcie_parse_mem(struct sky1_pcie *pcie)
>> +{
>> + struct device *dev = pcie->dev;
>> + struct platform_device *pdev = to_platform_device(dev);
>> + struct resource *res;
>> + void __iomem *base;
>> + int ret = 0;
>> +
>> + base = devm_platform_ioremap_resource_byname(pdev, "reg");
>> + if (IS_ERR(base)) {
>> + dev_err(dev, "Parse \"reg\" resource err\n");
>
> Syntax is return dev_err_probe, and without \" so grepping works
> correctly (see coding style).
Dear Krzysztof,
Thank you very much for your reply. Will change.
>
>> + return PTR_ERR(base);
>> + }
>> + pcie->reg_base = base;
>> +
>> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "cfg");
>> + if (!res) {
>> + dev_err(dev, "Parse \"cfg\" resource err\n");
>> + return -ENXIO;
>> + }
>> + pcie->cfg_res = res;
>> +
>> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "rcsu");
>> + if (!res) {
>> + dev_err(dev, "Parse \"rcsu\" resource err\n");
>> + return -ENXIO;
>> + }
>> + pcie->rcsu_base = devm_ioremap(dev, res->start, resource_size(res));
>
> Why aren't you using wrapper over get_resource and ioremap? Isn't
> devm_platform_ioremap_resource_byname exactly what you want?
>
Thank you for the reminder. Will change.
> And if argument from previous versions was - you need to backport it for
> ancient 3.10 kernel - then it would be a no go.
Ok, will change.
>
>> + if (!pcie->rcsu_base) {
>> + dev_err(dev, "ioremap failed for resource %pR\n", res);
>> + return -ENOMEM;
>> + }
>> +
>> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "msg");
>> + if (!res) {
>> + dev_err(dev, "Parse \"msg\" resource err\n");
>> + return -ENXIO;
>> + }
>> + pcie->msg_res = res;
>> + pcie->msg_base = devm_ioremap(dev, res->start, resource_size(res));
>> + if (!pcie->msg_base) {
>> + dev_err(dev, "ioremap failed for resource %pR\n", res);
>> + return -ENOMEM;
>> + }
>> +
>> + return ret;
>> +}
>> +
>> +static int sky1_pcie_parse_property(struct platform_device *pdev,
>> + struct sky1_pcie *pcie)
>> +{
>> + int ret = 0;
>> +
>> + ret = sky1_pcie_parse_mem(pcie);
>> + if (ret < 0)
>> + return ret;
>> +
>> + sky1_pcie_init_bases(pcie);
>> +
>> + return ret;
>> +}
>> +
>> +static int sky1_pcie_start_link(struct cdns_pcie *cdns_pcie)
>> +{
>> + struct sky1_pcie *pcie = dev_get_drvdata(cdns_pcie->dev);
>> +
>> + sky1_pcie_clear_and_set_dword(pcie->strap_base + STRAP_REG(1),
>> + 0, LINK_TRAINING_ENABLE);
>> +
>> + return 0;
>> +}
>> +
>> +static void sky1_pcie_stop_link(struct cdns_pcie *cdns_pcie)
>> +{
>> + struct sky1_pcie *pcie = dev_get_drvdata(cdns_pcie->dev);
>> +
>> + sky1_pcie_clear_and_set_dword(pcie->strap_base + STRAP_REG(1),
>> + LINK_TRAINING_ENABLE, 0);
>> +}
>> +
>> +
>> +static bool sky1_pcie_link_up(struct cdns_pcie *cdns_pcie)
>> +{
>> + u32 val;
>> +
>> + val = cdns_pcie_hpa_readl(cdns_pcie, REG_BANK_IP_REG,
>> + IP_REG_I_DBG_STS_0);
>> + return val & LINK_COMPLETE;
>> +}
>> +
>> +static const struct cdns_pcie_ops sky1_pcie_ops = {
>> + .start_link = sky1_pcie_start_link,
>> + .stop_link = sky1_pcie_stop_link,
>> + .link_up = sky1_pcie_link_up,
>> +};
>> +
>> +static int sky1_pcie_probe(struct platform_device *pdev)
>> +{
>> + const struct sky1_pcie_data *data;
>> + struct device *dev = &pdev->dev;
>> + struct pci_host_bridge *bridge;
>> + struct cdns_pcie *cdns_pcie;
>> + struct resource_entry *bus;
>> + struct cdns_pcie_rc *rc;
>> + struct sky1_pcie *pcie;
>> + int ret;
>> +
>> + pcie = devm_kzalloc(dev, sizeof(*pcie), GFP_KERNEL);
>> + if (!pcie)
>> + return -ENOMEM;
>> +
>> + data = of_device_get_match_data(dev);
>> + if (!data)
>> + return -EINVAL;
>> +
>> + pcie->data = data;
>> + pcie->dev = dev;
>> + dev_set_drvdata(dev, pcie);
>> +
>> + bridge = devm_pci_alloc_host_bridge(dev, sizeof(*rc));
>> + if (!bridge)
>> + return -ENOMEM;
>> +
>> + bus = resource_list_first_type(&bridge->windows, IORESOURCE_BUS);
>> + if (!bus)
>> + return -ENODEV;
>> +
>> + ret = sky1_pcie_parse_property(pdev, pcie);
>> + if (ret < 0)
>> + return -ENXIO;
>> +
>> + pcie->cfg = pci_ecam_create(dev, pcie->cfg_res, bus->res,
>> + &pci_generic_ecam_ops);
>> + if (IS_ERR(pcie->cfg))
>> + return PTR_ERR(pcie->cfg);
>> +
>> + bridge->ops = (struct pci_ops *)&pci_generic_ecam_ops.pci_ops;
>> + rc = pci_host_bridge_priv(bridge);
>> + rc->ecam_support_flag = 1;
>> + rc->cfg_base = pcie->cfg->win;
>> + rc->cfg_res = &pcie->cfg->res;
>> +
>> + cdns_pcie = &rc->pcie;
>> + cdns_pcie->dev = dev;
>> + cdns_pcie->ops = &sky1_pcie_ops;
>> + cdns_pcie->reg_base = pcie->reg_base;
>> + cdns_pcie->msg_res = pcie->msg_res;
>> + cdns_pcie->cdns_pcie_reg_offsets = &data->reg_off;
>> + cdns_pcie->is_rc = data->reg_off.is_rc;
>> +
>> + pcie->cdns_pcie = cdns_pcie;
>> + pcie->cdns_pcie_rc = rc;
>> + pcie->cfg_base = rc->cfg_base;
>> + bridge->sysdata = pcie->cfg;
>> +
>> + if (data->soc_type == CIX_SKY1) {
>
>
> Dead code or rather if (true) code. Don't do it, it's more difficult to
> read.
Will remove.
>
>> + rc->vendor_id = PCI_VENDOR_ID_CIX;
>> + rc->device_id = PCI_DEVICE_ID_CIX_SKY1;
>> + rc->no_inbound_flag = 1;
>> + }
>> +
>> + ret = cdns_pcie_hpa_host_setup(rc);
>> + if (ret < 0) {
>> + pci_ecam_free(pcie->cfg);
>> + return ret;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> +static const struct sky1_pcie_data sky1_pcie_rc_data = {
>> + .reg_off = {
>> + .is_rc = true,
>> + .ip_reg_bank_offset = SKY1_IP_REG_BANK_OFFSET,
>> + .ip_cfg_ctrl_reg_offset = SKY1_IP_CFG_CTRL_REG_BANK_OFFSET,
>> + .axi_mstr_common_offset = SKY1_IP_AXI_MASTER_COMMON_OFFSET,
>> + .axi_slave_offset = SKY1_AXI_SLAVE_OFFSET,
>> + .axi_master_offset = SKY1_AXI_MASTER_OFFSET,
>> + .axi_hls_offset = SKY1_AXI_HLS_REGISTERS_OFFSET,
>> + .axi_ras_offset = SKY1_AXI_RAS_REGISTERS_OFFSET,
>> + .axi_dti_offset = SKY1_DTI_REGISTERS_OFFSET,
>> + },
>> + .soc_type = CIX_SKY1,
>
> You have only one device variant, so this entire pcie_data feels redundant.
>
Will remove.
>> +};
>> +
>> +static const struct of_device_id of_sky1_pcie_match[] = {
>> + {
>> + .compatible = "cix,sky1-pcie-host",
>> + .data = &sky1_pcie_rc_data,
>> + },
>> + {},
>
Best regards,
Hans
© 2016 - 2026 Red Hat, Inc.