Re: [PATCH v5 5/5] phy: hisilicon: hisi-inno-phy: add support for Hi3798MV200 INNO PHY

From: Vinod Koul
Date: Sat Apr 06 2024 - 02:19:37 EST

Next message: Vinod Koul: "Re: [PATCH] net: phy: core: make phy_class constant"
Previous message: Baolu Lu: "Re: [PATCH v4 2/9] iommu: Replace sva_iommu with iommu_attach_handle"
In reply to: Yang Xiwen: "Re: [PATCH v5 5/5] phy: hisilicon: hisi-inno-phy: add support for Hi3798MV200 INNO PHY"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On 06-04-24, 01:53, Yang Xiwen wrote:
> On 4/6/2024 12:52 AM, Vinod Koul wrote:
> > On 05-03-24, 21:32, Yang Xiwen via B4 Relay wrote:
> > > From: Yang Xiwen <forbidden405@xxxxxxxxxxx>
> > That is quite an email id!
> >
> > > Direct MMIO resgiter access is used by Hi3798MV200. For other models,
> > > of_iomap() returns NULL due to insufficient length. So they are
> > so how is that fixed... Pls describe the change...
>
>
> The commit log will be rewritten in next revision. I'll try to emphasize the
> PHY and its configuration interface briefly. Though i don't have access to
> the datasheets and TRM so most things can not be verified.
>
>
> For CV200 and MV100 INNO PHY, the configuration interface is attached to
> PERICTRL(Peripheral Control Block). So we just use a register called
> PERI_USB3 to configure the PHY. The bus reset, clock are all controlled in
> PERI_USB3 register. To read/write to a register of the PHY, a special
> sequence of register writes and reads are needed, which was implemented in
> this driver.
>
>
> But for MV200 INNO PHY, the configuration interface is attached directly to
> system bus(MMIO). The bus clocks and resets are controlled via Clock Reset
> Generator(CRG). Now we have to control them with the help of linux clk and
> reset framework because they are provided by other modules.

Okay better log is welcome

>
>
> > > unaffected.
> > >
> > > Also Hi3798MV200 INNO PHY has an extra reset required to be deasserted,
> > > switch to reset_control_array_*() APIs for that.
>
>
> The commit msg is misleading here. There is no extra reset actually. The
> reset also exist for existing users. The initial author just decided to
> manage it in the hisi_inno_phy_write_reg() routine(without using
> reset_control_* APIs) and omit it in the binding.
>
>
> > That probably should be a different patch
>
>
> I guess so. From my point of view, the whole patch is to introduce the
> support for Hi3798MV200 variant of the INNO PHY. So i've decided to squash
> the two changes into one single commit.

Not really you can build smaller reviewable changes leading up to adding
the Hi3798MV200 support

>
>
> >
> > > Signed-off-by: Yang Xiwen <forbidden405@xxxxxxxxxxx>
> > > ---
> > > drivers/phy/hisilicon/phy-hisi-inno-usb2.c | 66 ++++++++++++++++++------------
> > > 1 file changed, 40 insertions(+), 26 deletions(-)
> > >
> > > diff --git a/drivers/phy/hisilicon/phy-hisi-inno-usb2.c b/drivers/phy/hisilicon/phy-hisi-inno-usb2.c
> > > index b7e740eb4752..df154cd99ed8 100644
> > > --- a/drivers/phy/hisilicon/phy-hisi-inno-usb2.c
> > > +++ b/drivers/phy/hisilicon/phy-hisi-inno-usb2.c
> > > @@ -10,6 +10,7 @@
> > > #include <linux/io.h>
> > > #include <linux/module.h>
> > > #include <linux/of.h>
> > > +#include <linux/of_address.h>
> > > #include <linux/phy/phy.h>
> > > #include <linux/platform_device.h>
> > > #include <linux/reset.h>
> > > @@ -24,6 +25,7 @@
> > > #define PHY_TYPE_0 0
> > > #define PHY_TYPE_1 1
> > > +#define PHY_TYPE_MMIO 2
> > > #define PHY_TEST_DATA GENMASK(7, 0)
> > > #define PHY_TEST_ADDR_OFFSET 8
> > > @@ -43,6 +45,7 @@
> > > #define PHY_CLK_ENABLE BIT(2)
> > > struct hisi_inno_phy_port {
> > > + void __iomem *base;
> > > struct reset_control *utmi_rst;
> > > struct hisi_inno_phy_priv *priv;
> > > };
> > > @@ -50,7 +53,7 @@ struct hisi_inno_phy_port {
> > > struct hisi_inno_phy_priv {
> > > void __iomem *mmio;
> > > struct clk *ref_clk;
> > > - struct reset_control *por_rst;
> > > + struct reset_control *rsts;
> > > unsigned int type;
> > > struct hisi_inno_phy_port ports[INNO_PHY_PORT_NUM];
> > > };
> > > @@ -62,26 +65,31 @@ static void hisi_inno_phy_write_reg(struct hisi_inno_phy_priv *priv,
> > > u32 val;
> > > u32 value;
> > > - if (priv->type == PHY_TYPE_0)
> > > - val = (data & PHY_TEST_DATA) |
> > > - ((addr << PHY_TEST_ADDR_OFFSET) & PHY0_TEST_ADDR) |
> > > - ((port << PHY0_TEST_PORT_OFFSET) & PHY0_TEST_PORT) |
> > > - PHY0_TEST_WREN | PHY0_TEST_RST;
> > > - else
> > > - val = (data & PHY_TEST_DATA) |
> > > - ((addr << PHY_TEST_ADDR_OFFSET) & PHY1_TEST_ADDR) |
> > > - ((port << PHY1_TEST_PORT_OFFSET) & PHY1_TEST_PORT) |
> > > - PHY1_TEST_WREN | PHY1_TEST_RST;
> > > - writel(val, reg);
> > > -
> > > - value = val;
> > > - if (priv->type == PHY_TYPE_0)
> > > - value |= PHY0_TEST_CLK;
> > > - else
> > > - value |= PHY1_TEST_CLK;
> > > - writel(value, reg);
> > > -
> > > - writel(val, reg);
> > > + if (priv->ports[port].base)
> > > + /* FIXME: fill stride in priv */
> > when?
>
>
> I'm not sure. Maybe until some other users with stride other than 3? I don't
> have much knowledge about other SoCs.
>
>
> Maybe replace the FIXME here with some additional information.

Better

>
>
> >
> > > + writel(data, (u32 *)priv->ports[port].base + addr);
> > > + else {
> > > + if (priv->type == PHY_TYPE_0)
> > > + val = (data & PHY_TEST_DATA) |
> > > + ((addr << PHY_TEST_ADDR_OFFSET) & PHY0_TEST_ADDR) |
> > > + ((port << PHY0_TEST_PORT_OFFSET) & PHY0_TEST_PORT) |
> > > + PHY0_TEST_WREN | PHY0_TEST_RST;
> > > + else
> > > + val = (data & PHY_TEST_DATA) |
> > > + ((addr << PHY_TEST_ADDR_OFFSET) & PHY1_TEST_ADDR) |
> > > + ((port << PHY1_TEST_PORT_OFFSET) & PHY1_TEST_PORT) |
> > > + PHY1_TEST_WREN | PHY1_TEST_RST;
> > > + writel(val, reg);
> > > +
> > > + value = val;
> > > + if (priv->type == PHY_TYPE_0)
> > > + value |= PHY0_TEST_CLK;
> > > + else
> > > + value |= PHY1_TEST_CLK;
> > > + writel(value, reg);
> > > +
> > > + writel(val, reg);
> > val and value are very helpful variables, do consider naming them
> > better!
>
>
> I'll consider renaming them in the next revision. Maybe val and val2? They
> are just some temp vars to store register values.

Yeah, that might be better

>
>
> >
> > > + }
> > > }
> > > static void hisi_inno_phy_setup(struct hisi_inno_phy_priv *priv)
> > > @@ -104,7 +112,7 @@ static int hisi_inno_phy_init(struct phy *phy)
> > > return ret;
> > > udelay(REF_CLK_STABLE_TIME);
> > > - reset_control_deassert(priv->por_rst);
> > > + reset_control_deassert(priv->rsts);
> > > udelay(POR_RST_COMPLETE_TIME);
> > > /* Set up phy registers */
> > > @@ -122,7 +130,7 @@ static int hisi_inno_phy_exit(struct phy *phy)
> > > struct hisi_inno_phy_priv *priv = port->priv;
> > > reset_control_assert(port->utmi_rst);
> > > - reset_control_assert(priv->por_rst);
> > > + reset_control_assert(priv->rsts);
> > > clk_disable_unprepare(priv->ref_clk);
> > > return 0;
> > > @@ -158,15 +166,16 @@ static int hisi_inno_phy_probe(struct platform_device *pdev)
> > > if (IS_ERR(priv->ref_clk))
> > > return PTR_ERR(priv->ref_clk);
> > > - priv->por_rst = devm_reset_control_get_exclusive(dev, NULL);
> > > - if (IS_ERR(priv->por_rst))
> > > - return PTR_ERR(priv->por_rst);
> > > + priv->rsts = devm_reset_control_array_get_exclusive(dev);
> > > + if (IS_ERR(priv->rsts))
> > > + return PTR_ERR(priv->rsts);
> > > priv->type = (uintptr_t) of_device_get_match_data(dev);
> > > for_each_child_of_node(np, child) {
> > > struct reset_control *rst;
> > > struct phy *phy;
> > > + void __iomem *base;
> > > rst = of_reset_control_get_exclusive(child, NULL);
> > > if (IS_ERR(rst)) {
> > > @@ -174,7 +183,10 @@ static int hisi_inno_phy_probe(struct platform_device *pdev)
> > > return PTR_ERR(rst);
> > > }
> > > + base = of_iomap(child, 0);
> > > +
> > > priv->ports[i].utmi_rst = rst;
> > > + priv->ports[i].base = base;
> > > priv->ports[i].priv = priv;
> > > phy = devm_phy_create(dev, child, &hisi_inno_phy_ops);
> > > @@ -205,6 +217,8 @@ static const struct of_device_id hisi_inno_phy_of_match[] = {
> > > .data = (void *) PHY_TYPE_0 },
> > > { .compatible = "hisilicon,hi3798mv100-usb2-phy",
> > > .data = (void *) PHY_TYPE_1 },
> > > + { .compatible = "hisilicon,hi3798mv200-usb2-phy",
> > > + .data = (void *) PHY_TYPE_MMIO },
> > > { },
> > > };
> > > MODULE_DEVICE_TABLE(of, hisi_inno_phy_of_match);
> > >
> > > --
> > > 2.43.0
>
>
> --
> Regards,
> Yang Xiwen

--
~Vinod

Next message: Vinod Koul: "Re: [PATCH] net: phy: core: make phy_class constant"
Previous message: Baolu Lu: "Re: [PATCH v4 2/9] iommu: Replace sva_iommu with iommu_attach_handle"
In reply to: Yang Xiwen: "Re: [PATCH v5 5/5] phy: hisilicon: hisi-inno-phy: add support for Hi3798MV200 INNO PHY"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]