Re: [v3,2/3] PCI: mediatek: Add new generation controller support

[Philipp Zabel]

Hi Jianjun,

On Sun, 2020-09-27 at 15:45 +0800, Jianjun Wang wrote:
> MediaTek's PCIe host controller has three generation HWs, the new
> generation HW is an individual bridge, it supoorts Gen3 speed and
> up to 256 MSI interrupt numbers for multi-function devices.
> 
> Add support for new Gen3 controller which can be found on MT8192.
> 
> Signed-off-by: Jianjun Wang <jianjun.wang@xxxxxxxxxxxx>
> Acked-by: Ryder Lee <ryder.lee@xxxxxxxxxxxx>
> ---
>  drivers/pci/controller/Kconfig              |   14 +
>  drivers/pci/controller/Makefile             |    1 +
>  drivers/pci/controller/pcie-mediatek-gen3.c | 1024 +++++++++++++++++++
>  3 files changed, 1039 insertions(+)
>  create mode 100644 drivers/pci/controller/pcie-mediatek-gen3.c
> 
[...]
> diff --git a/drivers/pci/controller/pcie-mediatek-gen3.c b/drivers/pci/controller/pcie-mediatek-gen3.c
> new file mode 100644
> index 000000000000..ad69c789b24d
> --- /dev/null
> +++ b/drivers/pci/controller/pcie-mediatek-gen3.c
> @@ -0,0 +1,1024 @@
[...]
> +static int mtk_pcie_power_up(struct mtk_pcie_port *port)
> +{
> +	struct device *dev = port->dev;
> +	int err;
> +
> +	port->phy_reset = devm_reset_control_get_optional_exclusive(dev,
> +								    "phy-rst");
> +	if (IS_ERR(port->phy_reset))
> +		return PTR_ERR(port->phy_reset);
> +
> +	reset_control_deassert(port->phy_reset);

In general, it is better to request all required resources before
starting to activate the hardware.

> +
> +	/* PHY power on and enable pipe clock */
> +	port->phy = devm_phy_optional_get(dev, "pcie-phy");
> +	if (IS_ERR(port->phy))
> +		return PTR_ERR(port->phy);

For example, if the PHY driver is not loaded yet and this returns
-EPROBE_DEFER, it was not useful to take the PHY out of reset above.
Also, phy-rst is kept deasserted if this fails.

> +
> +	err = phy_init(port->phy);
> +	if (err) {
> +		dev_notice(dev, "failed to initialize pcie phy\n");
> +		return err;

phy-rst is kept deasserted if this fails.

> +	}
> +
> +	err = phy_power_on(port->phy);
> +	if (err) {
> +		dev_notice(dev, "failed to power on pcie phy\n");
> +		goto err_phy_on;
> +	}
> +
> +	port->mac_reset = devm_reset_control_get_optional_exclusive(dev,
> +								    "mac-rst");
> +	if (IS_ERR(port->mac_reset))
> +		return PTR_ERR(port->mac_reset);

The PHY is not powered down if this fails.

> +
> +	reset_control_deassert(port->mac_reset);
> +
> +	/* MAC power on and enable transaction layer clocks */
> +	pm_runtime_enable(dev);
> +	pm_runtime_get_sync(dev);
> +
> +	err = mtk_pcie_clk_init(port);
> +	if (err) {
> +		dev_notice(dev, "clock init failed\n");
> +		goto err_clk_init;
> +	}
> +
> +	return 0;
> +
> +err_clk_init:
> +	pm_runtime_put_sync(dev);
> +	pm_runtime_disable(dev);
> +	reset_control_assert(port->mac_reset);
> +	phy_power_off(port->phy);
> +err_phy_on:
> +	phy_exit(port->phy);
> +	reset_control_assert(port->phy_reset);
> +
> +	return -EBUSY;
> +}
> +
> +static void mtk_pcie_power_down(struct mtk_pcie_port *port)
> +{
> +	phy_power_off(port->phy);
> +	phy_exit(port->phy);
> +
> +	clk_bulk_disable_unprepare(port->num_clks, port->clks);

In the power-up sequence clocks are enabled last, but here they are not
disabled before the PHY is powered off. Is this on purpose?

> +
> +	pm_runtime_put_sync(port->dev);
> +	pm_runtime_disable(port->dev);

In the power-up error path, PHY and controller resets are asserted
again, but here they are kept deasserted. Should they be asserted here
as well?

regards
Philipp