Re: [PATCH v2 5/6] pci: dwc: pci-exynos: rework the driver to support Exynos5433 variant

From: Marek Szyprowski
Date: Tue Oct 27 2020 - 08:04:29 EST


Hi Rob,

On 26.10.2020 20:14, Rob Herring wrote:
> On Fri, Oct 23, 2020 at 2:58 AM Marek Szyprowski
> <m.szyprowski@xxxxxxxxxxx> wrote:
>> From: Jaehoon Chung <jh80.chung@xxxxxxxxxxx>
>>
>> Exynos5440 SoC support has been dropped since commit 8c83315da1cf ("ARM:
>> dts: exynos: Remove Exynos5440"). Rework this driver to support DWC PCIe
>> variant found in the Exynos5433 SoCs.
>>
>> The main difference in Exynos5433 variant is lack of the MSI support
>> (the MSI interrupt is not even routed to the CPU).
>>
>> Signed-off-by: Jaehoon Chung <jh80.chung@xxxxxxxxxxx>
>> [mszyprow: reworked the driver to support only Exynos5433 variant,
>> simplified code, rebased onto current kernel code, added
>> regulator support, converted to the regular platform driver,
>> removed MSI related code, rewrote commit message]
>> Signed-off-by: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>
>> Acked-by: Krzysztof Kozlowski <krzk@xxxxxxxxxx>
>> ---
>> drivers/pci/controller/dwc/Kconfig | 3 +-
>> drivers/pci/controller/dwc/pci-exynos.c | 358 ++++++++++--------------
>> drivers/pci/quirks.c | 1 +
>> 3 files changed, 145 insertions(+), 217 deletions(-)
>>
>> diff --git a/drivers/pci/controller/dwc/Kconfig b/drivers/pci/controller/dwc/Kconfig
>> index bc049865f8e0..ade07abd23c9 100644
>> --- a/drivers/pci/controller/dwc/Kconfig
>> +++ b/drivers/pci/controller/dwc/Kconfig
>> @@ -84,8 +84,7 @@ config PCIE_DW_PLAT_EP
>>
>> config PCI_EXYNOS
>> bool "Samsung Exynos PCIe controller"
>> - depends on SOC_EXYNOS5440 || COMPILE_TEST
>> - depends on PCI_MSI_IRQ_DOMAIN
>> + depends on ARCH_EXYNOS || COMPILE_TEST
>> select PCIE_DW_HOST
>>
>> config PCI_IMX6
>> diff --git a/drivers/pci/controller/dwc/pci-exynos.c b/drivers/pci/controller/dwc/pci-exynos.c
>> index 242683cde04a..58056fbdc2fa 100644
>> --- a/drivers/pci/controller/dwc/pci-exynos.c
>> +++ b/drivers/pci/controller/dwc/pci-exynos.c
>> @@ -2,26 +2,23 @@
>> /*
>> * PCIe host controller driver for Samsung Exynos SoCs
>> *
>> - * Copyright (C) 2013 Samsung Electronics Co., Ltd.
>> + * Copyright (C) 2013-2020 Samsung Electronics Co., Ltd.
>> * https://www.samsung.com
>> *
>> * Author: Jingoo Han <jg1.han@xxxxxxxxxxx>
>> + * Jaehoon Chung <jh80.chung@xxxxxxxxxxx>
>> */
>>
>> #include <linux/clk.h>
>> #include <linux/delay.h>
>> -#include <linux/gpio.h>
>> #include <linux/interrupt.h>
>> #include <linux/kernel.h>
>> #include <linux/init.h>
>> #include <linux/of_device.h>
>> -#include <linux/of_gpio.h>
>> #include <linux/pci.h>
>> #include <linux/platform_device.h>
>> #include <linux/phy/phy.h>
>> -#include <linux/resource.h>
>> -#include <linux/signal.h>
>> -#include <linux/types.h>
>> +#include <linux/regulator/consumer.h>
>>
>> #include "pcie-designware.h"
>>
>> @@ -37,102 +34,47 @@
>> #define PCIE_IRQ_SPECIAL 0x008
>> #define PCIE_IRQ_EN_PULSE 0x00c
>> #define PCIE_IRQ_EN_LEVEL 0x010
>> -#define IRQ_MSI_ENABLE BIT(2)
>> #define PCIE_IRQ_EN_SPECIAL 0x014
>> -#define PCIE_PWR_RESET 0x018
>> +#define PCIE_SW_WAKE 0x018
>> +#define PCIE_BUS_EN BIT(1)
>> #define PCIE_CORE_RESET 0x01c
>> #define PCIE_CORE_RESET_ENABLE BIT(0)
>> #define PCIE_STICKY_RESET 0x020
>> #define PCIE_NONSTICKY_RESET 0x024
>> #define PCIE_APP_INIT_RESET 0x028
>> #define PCIE_APP_LTSSM_ENABLE 0x02c
>> -#define PCIE_ELBI_RDLH_LINKUP 0x064
>> +#define PCIE_ELBI_RDLH_LINKUP 0x074
>> +#define PCIE_ELBI_XMLH_LINKUP BIT(4)
>> #define PCIE_ELBI_LTSSM_ENABLE 0x1
>> #define PCIE_ELBI_SLV_AWMISC 0x11c
>> #define PCIE_ELBI_SLV_ARMISC 0x120
>> #define PCIE_ELBI_SLV_DBI_ENABLE BIT(21)
>>
>> -struct exynos_pcie_mem_res {
>> - void __iomem *elbi_base; /* DT 0th resource: PCIe CTRL */
>> -};
>> -
>> -struct exynos_pcie_clk_res {
>> - struct clk *clk;
>> - struct clk *bus_clk;
>> -};
>> +/* DBI register */
>> +#define PCIE_MISC_CONTROL_1_OFF 0x8BC
>> +#define DBI_RO_WR_EN BIT(0)
> Standard DWC port logic register. The core already handles this
> mostly. And provides a function to it where it doesn't. Looking at
> your use, I think you can drop the access.
>
>> ...
>> @@ -243,19 +168,25 @@ static int exynos_pcie_establish_link(struct exynos_pcie *ep)
>> exynos_pcie_assert_core_reset(ep);
>>
>> phy_reset(ep->phy);
>> -
>> - exynos_pcie_writel(ep->mem_res->elbi_base, 1,
>> - PCIE_PWR_RESET);
>> -
>> phy_power_on(ep->phy);
>> phy_init(ep->phy);
>>
>> exynos_pcie_deassert_core_reset(ep);
>> +
>> + val = exynos_pcie_readl(ep->elbi_base, PCIE_SW_WAKE);
>> + val &= ~PCIE_BUS_EN;
>> + exynos_pcie_writel(ep->elbi_base, val, PCIE_SW_WAKE);
>> +
>> + /*
>> + * Enable DBI_RO_WR_EN bit.
>> + * - When set to 1, some RO and HWinit bits are wriatble from
>> + * the local application through the DBI.
>> + */
>> + dw_pcie_writel_dbi(pci, PCIE_MISC_CONTROL_1_OFF, DBI_RO_WR_EN);
>> dw_pcie_setup_rc(pp);
> First thing this function does is set DBI_RO_WR_EN.

Indeed, this has been added to dw_pcie_setup_rc() in commit 3924bc2fd1b6
("PCI: dwc: Group DBI registers writes requiring unlocking"), after
initial version of this patchset. Thanks for pointing this out. I will
remove this.

>> ...
>> @@ -450,42 +347,49 @@ static int __init exynos_pcie_probe(struct platform_device *pdev)
>> if (!ep)
>> return -ENOMEM;
>>
>> - pci = devm_kzalloc(dev, sizeof(*pci), GFP_KERNEL);
>> - if (!pci)
>> - return -ENOMEM;
>> -
>> - pci->dev = dev;
>> - pci->ops = &dw_pcie_ops;
>> + ep->pci.dev = dev;
>> + ep->pci.ops = &dw_pcie_ops;
>>
>> - ep->pci = pci;
>> - ep->ops = (const struct exynos_pcie_ops *)
>> - of_device_get_match_data(dev);
>> + ep->phy = devm_of_phy_get(dev, np, NULL);
>> + if (IS_ERR(ep->phy))
>> + return PTR_ERR(ep->phy);
>>
>> - ep->reset_gpio = of_get_named_gpio(np, "reset-gpio", 0);
>> + /* External Local Bus interface (ELBI) registers */
>> + ep->elbi_base = devm_platform_ioremap_resource_byname(pdev, "elbi");
>> + if (IS_ERR(ep->elbi_base))
>> + return PTR_ERR(ep->elbi_base);
>>
>> - ep->phy = devm_of_phy_get(dev, np, NULL);
>> - if (IS_ERR(ep->phy)) {
>> - if (PTR_ERR(ep->phy) != -ENODEV)
>> - return PTR_ERR(ep->phy);
>> + /* Data Bus Interface (DBI) registers */
>> + ep->pci.dbi_base = devm_platform_ioremap_resource_byname(pdev, "dbi");
>> + if (IS_ERR(ep->pci.dbi_base))
>> + return PTR_ERR(ep->pci.dbi_base);
> This is going to get moved to the DWC core code.
Well, so far it is not there yet and other dw-pci drivers do it on their
own. Could you point a patch that does this, so I can rebase onto it?
>
>> - ep->phy = NULL;
>> + ep->clk = devm_clk_get(dev, "pcie");
>> + if (IS_ERR(ep->clk)) {
>> + dev_err(dev, "Failed to get pcie rc clock\n");
>> + return PTR_ERR(ep->clk);
>> }
>>
>> - if (ep->ops && ep->ops->get_mem_resources) {
>> - ret = ep->ops->get_mem_resources(pdev, ep);
>> - if (ret)
>> - return ret;
>> + ep->bus_clk = devm_clk_get(dev, "pcie_bus");
>> + if (IS_ERR(ep->bus_clk)) {
>> + dev_err(dev, "Failed to get pcie bus clock\n");
>> + return PTR_ERR(ep->bus_clk);
>> }
>>
>> - if (ep->ops && ep->ops->get_clk_resources &&
>> - ep->ops->init_clk_resources) {
>> - ret = ep->ops->get_clk_resources(ep);
>> - if (ret)
>> - return ret;
>> - ret = ep->ops->init_clk_resources(ep);
>> - if (ret)
>> - return ret;
>> - }
>> + ep->supplies[0].supply = "vdd18";
>> + ep->supplies[1].supply = "vdd10";
>> + ret = devm_regulator_bulk_get(dev, ARRAY_SIZE(ep->supplies),
>> + ep->supplies);
>> + if (ret)
>> + return ret;
>> +
>> + ret = exynos_pcie_init_clk_resources(ep);
>> + if (ret)
>> + return ret;
>> +
>> + ret = regulator_bulk_enable(ARRAY_SIZE(ep->supplies), ep->supplies);
>> + if (ret)
>> + return ret;
>>
>> platform_set_drvdata(pdev, ep);
>>
>> @@ -497,9 +401,9 @@ static int __init exynos_pcie_probe(struct platform_device *pdev)
>>
>> fail_probe:
>> phy_exit(ep->phy);
>> + exynos_pcie_deinit_clk_resources(ep);
>> + regulator_bulk_disable(ARRAY_SIZE(ep->supplies), ep->supplies);
>>
>> - if (ep->ops && ep->ops->deinit_clk_resources)
>> - ep->ops->deinit_clk_resources(ep);
>> return ret;
>> }
>>
>> @@ -507,32 +411,56 @@ static int __exit exynos_pcie_remove(struct platform_device *pdev)
>> {
>> struct exynos_pcie *ep = platform_get_drvdata(pdev);
>>
>> - if (ep->ops && ep->ops->deinit_clk_resources)
>> - ep->ops->deinit_clk_resources(ep);
>> + phy_power_off(ep->phy);
>> + phy_exit(ep->phy);
>> + exynos_pcie_deinit_clk_resources(ep);
>> + regulator_bulk_disable(ARRAY_SIZE(ep->supplies), ep->supplies);
>>
>> return 0;
>> }
>>
>> +static int __maybe_unused exynos_pcie_suspend_noirq(struct device *dev)
>> +{
> Why noirq variant needed? Lot's of PCI host drivers do this and I've
> yet to get a reason...
Frankly, I have no idea, but switching to SET_LATE_SYSTEM_SLEEP_PM_OPS
breaks system suspend/resume operation - the board doesn't resume from
suspend. If this is really important I will add some more logs and try
to find what happens between late/early and noirq phases.
> Adding suspend/resume should probably be a separate patch. What I'd
> like to do here is have common DWC suspend/resume functions that the
> platform drivers can use or wrap.

Okay, I can move adding suspend/resume to the separate patch if You
want. However I probably know too little about PCI to extract some
common dwc suspend/resume functions.

>> + struct exynos_pcie *ep = dev_get_drvdata(dev);
>> +
>> + phy_power_off(ep->phy);
>> + phy_exit(ep->phy);
>> + regulator_bulk_disable(ARRAY_SIZE(ep->supplies), ep->supplies);
>> +
>> + return 0;
>> +}
>> +
>> +static int __maybe_unused exynos_pcie_resume_noirq(struct device *dev)
>> +{
>> + struct exynos_pcie *ep = dev_get_drvdata(dev);
>> + struct dw_pcie *pci = &ep->pci;
>> + struct pcie_port *pp = &pci->pp;
>> + int ret;
>> +
>> + ret = regulator_bulk_enable(ARRAY_SIZE(ep->supplies), ep->supplies);
>> + if (ret)
>> + return ret;
>> + /* exynos_pcie_host_init controls ep->phy */
>> + return exynos_pcie_host_init(pp);
>> +}
>> +
>> +static const struct dev_pm_ops exynos_pcie_pm_ops = {
>> + SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(exynos_pcie_suspend_noirq,
>> + exynos_pcie_resume_noirq)
>> +};
>> +
>> static const struct of_device_id exynos_pcie_of_match[] = {
>> - {
>> - .compatible = "samsung,exynos5440-pcie",
>> - .data = &exynos5440_pcie_ops
>> - },
>> - {},
>> + { .compatible = "samsung,exynos5433-pcie", },
>> + { },
>> };
>>
>> static struct platform_driver exynos_pcie_driver = {
>> + .probe = exynos_pcie_probe,
>> .remove = __exit_p(exynos_pcie_remove),
>> .driver = {
>> .name = "exynos-pcie",
>> .of_match_table = exynos_pcie_of_match,
>> + .pm = &exynos_pcie_pm_ops,
>> },
>> };
>> -
>> -/* Exynos PCIe driver does not allow module unload */
>> -
>> -static int __init exynos_pcie_init(void)
>> -{
>> - return platform_driver_probe(&exynos_pcie_driver, exynos_pcie_probe);
>> -}
>> -subsys_initcall(exynos_pcie_init);
> Good that this is gone, but...
>
>> +builtin_platform_driver(exynos_pcie_driver);
> I would like to make all the host drivers modules.

I can check if this can be easily done. If not, I would like to keep it
builtin in this patch and leave modularization for the future.

Best regards

--
Marek Szyprowski, PhD
Samsung R&D Institute Poland