Re: [PATCH net-next v2 3/3] net: stmmac: add glue layer for T-HEAD TH1520 SoC

From: Serge Semin
Date: Tue Aug 29 2023 - 06:08:32 EST


On Tue, Aug 29, 2023 at 11:17:45AM +0800, Jisheng Zhang wrote:
> On Mon, Aug 28, 2023 at 08:30:50PM +0300, Serge Semin wrote:
> > On Mon, Aug 28, 2023 at 11:41:47PM +0800, Jisheng Zhang wrote:
> > > On Mon, Aug 28, 2023 at 04:40:19PM +0300, Serge Semin wrote:
> > > > On Sun, Aug 27, 2023 at 05:17:10PM +0800, Jisheng Zhang wrote:
> > > > > Add dwmac glue driver to support the dwmac on the T-HEAD TH1520 SoC.
> > > > >
> > > > > Signed-off-by: Jisheng Zhang <jszhang@xxxxxxxxxx>
> > > > > ---
> > > > > drivers/net/ethernet/stmicro/stmmac/Kconfig | 11 +
> > > > > drivers/net/ethernet/stmicro/stmmac/Makefile | 1 +
> > > > > .../net/ethernet/stmicro/stmmac/dwmac-thead.c | 302 ++++++++++++++++++
> > > > > 3 files changed, 314 insertions(+)
> > > > > create mode 100644 drivers/net/ethernet/stmicro/stmmac/dwmac-thead.c
> > > > >
> > > > > diff --git a/drivers/net/ethernet/stmicro/stmmac/Kconfig b/drivers/net/ethernet/stmicro/stmmac/Kconfig
> > > > > index 06c6871f8788..1bf71804c270 100644
> > > > > --- a/drivers/net/ethernet/stmicro/stmmac/Kconfig
> > > > > +++ b/drivers/net/ethernet/stmicro/stmmac/Kconfig
> > > > > @@ -216,6 +216,17 @@ config DWMAC_SUN8I
> > > > > stmmac device driver. This driver is used for H3/A83T/A64
> > > > > EMAC ethernet controller.
> > > > >
> > > > > +config DWMAC_THEAD
> > > > > + tristate "T-HEAD dwmac support"
> > > > > + depends on OF && (ARCH_THEAD || COMPILE_TEST)
> > > > > + select MFD_SYSCON
> > > > > + help
> > > > > + Support for ethernet controllers on T-HEAD RISC-V SoCs
> > > > > +
> > > > > + This selects the T-HEAD platform specific glue layer support for
> > > > > + the stmmac device driver. This driver is used for T-HEAD TH1520
> > > > > + ethernet controller.
> > > > > +
> > > > > config DWMAC_IMX8
> > > > > tristate "NXP IMX8 DWMAC support"
> > > > > default ARCH_MXC
> > > > > diff --git a/drivers/net/ethernet/stmicro/stmmac/Makefile b/drivers/net/ethernet/stmicro/stmmac/Makefile
> > > > > index 5b57aee19267..d73171ed6ad7 100644
> > > > > --- a/drivers/net/ethernet/stmicro/stmmac/Makefile
> > > > > +++ b/drivers/net/ethernet/stmicro/stmmac/Makefile
> > > > > @@ -27,6 +27,7 @@ obj-$(CONFIG_DWMAC_STI) += dwmac-sti.o
> > > > > obj-$(CONFIG_DWMAC_STM32) += dwmac-stm32.o
> > > > > obj-$(CONFIG_DWMAC_SUNXI) += dwmac-sunxi.o
> > > > > obj-$(CONFIG_DWMAC_SUN8I) += dwmac-sun8i.o
> > > > > +obj-$(CONFIG_DWMAC_THEAD) += dwmac-thead.o
> > > > > obj-$(CONFIG_DWMAC_DWC_QOS_ETH) += dwmac-dwc-qos-eth.o
> > > > > obj-$(CONFIG_DWMAC_INTEL_PLAT) += dwmac-intel-plat.o
> > > > > obj-$(CONFIG_DWMAC_GENERIC) += dwmac-generic.o
> > > > > diff --git a/drivers/net/ethernet/stmicro/stmmac/dwmac-thead.c b/drivers/net/ethernet/stmicro/stmmac/dwmac-thead.c
> > > > > new file mode 100644
> > > > > index 000000000000..85135ef05906
> > > > > --- /dev/null
> > > > > +++ b/drivers/net/ethernet/stmicro/stmmac/dwmac-thead.c
> > > > > @@ -0,0 +1,302 @@
> > > > > +// SPDX-License-Identifier: GPL-2.0
> > > > > +/*
> > > > > + * T-HEAD DWMAC platform driver
> > > > > + *
> > > > > + * Copyright (C) 2021 Alibaba Group Holding Limited.
> > > > > + * Copyright (C) 2023 Jisheng Zhang <jszhang@xxxxxxxxxx>
> > > > > + *
> > > > > + */
> > > > > +
> > > > > +#include <linux/bitfield.h>
> > > > > +#include <linux/mfd/syscon.h>
> > > > > +#include <linux/module.h>
> > > > > +#include <linux/of.h>
> > > > > +#include <linux/of_device.h>
> > > > > +#include <linux/of_net.h>
> > > > > +#include <linux/platform_device.h>
> > > > > +#include <linux/regmap.h>
> > > > > +
> > > > > +#include "stmmac_platform.h"
> > > > > +
> > > > > +#define GMAC_CLK_EN 0x00
> > > > > +#define GMAC_TX_CLK_EN BIT(1)
> > > > > +#define GMAC_TX_CLK_N_EN BIT(2)
> > > > > +#define GMAC_TX_CLK_OUT_EN BIT(3)
> > > > > +#define GMAC_RX_CLK_EN BIT(4)
> > > > > +#define GMAC_RX_CLK_N_EN BIT(5)
> > > > > +#define GMAC_EPHY_REF_CLK_EN BIT(6)
> > > > > +#define GMAC_RXCLK_DELAY_CTRL 0x04
> > > > > +#define GMAC_RXCLK_BYPASS BIT(15)
> > > > > +#define GMAC_RXCLK_INVERT BIT(14)
> > > > > +#define GMAC_RXCLK_DELAY_MASK GENMASK(4, 0)
> > > > > +#define GMAC_RXCLK_DELAY_VAL(x) FIELD_PREP(GMAC_RXCLK_DELAY_MASK, (x))
> > > > > +#define GMAC_TXCLK_DELAY_CTRL 0x08
> > > > > +#define GMAC_TXCLK_BYPASS BIT(15)
> > > > > +#define GMAC_TXCLK_INVERT BIT(14)
> > > > > +#define GMAC_TXCLK_DELAY_MASK GENMASK(4, 0)
> > > > > +#define GMAC_TXCLK_DELAY_VAL(x) FIELD_PREP(GMAC_RXCLK_DELAY_MASK, (x))
> > > > > +#define GMAC_PLLCLK_DIV 0x0c
> > > > > +#define GMAC_PLLCLK_DIV_EN BIT(31)
> > > > > +#define GMAC_PLLCLK_DIV_MASK GENMASK(7, 0)
> > > > > +#define GMAC_PLLCLK_DIV_NUM(x) FIELD_PREP(GMAC_PLLCLK_DIV_MASK, (x))
> > > > > +#define GMAC_GTXCLK_SEL 0x18
> > > > > +#define GMAC_GTXCLK_SEL_PLL BIT(0)
> > > > > +#define GMAC_INTF_CTRL 0x1c
> > > > > +#define PHY_INTF_MASK BIT(0)
> > > > > +#define PHY_INTF_RGMII FIELD_PREP(PHY_INTF_MASK, 1)
> > > > > +#define PHY_INTF_MII_GMII FIELD_PREP(PHY_INTF_MASK, 0)
> > > > > +#define GMAC_TXCLK_OEN 0x20
> > > > > +#define TXCLK_DIR_MASK BIT(0)
> > > > > +#define TXCLK_DIR_OUTPUT FIELD_PREP(TXCLK_DIR_MASK, 0)
> > > > > +#define TXCLK_DIR_INPUT FIELD_PREP(TXCLK_DIR_MASK, 1)
> > > > > +
> > > > > +#define GMAC_GMII_RGMII_RATE 125000000
> > > > > +#define GMAC_MII_RATE 25000000
> > > > > +
> > > > > +struct thead_dwmac {
> > > > > + struct plat_stmmacenet_data *plat;
> > > > > + struct regmap *apb_regmap;
> > > > > + struct device *dev;
> > > > > + u32 rx_delay;
> > > > > + u32 tx_delay;
> > > > > +};
> > > > > +
> > > > > +static int thead_dwmac_set_phy_if(struct plat_stmmacenet_data *plat)
> > > > > +{
> > > > > + struct thead_dwmac *dwmac = plat->bsp_priv;
> > > > > + u32 phyif;
> > > > > +
> > > > > + switch (plat->interface) {
> > > > > + case PHY_INTERFACE_MODE_MII:
> > > > > + phyif = PHY_INTF_MII_GMII;
> > > > > + break;
> > > > > + case PHY_INTERFACE_MODE_RGMII:
> > > > > + case PHY_INTERFACE_MODE_RGMII_ID:
> > > > > + case PHY_INTERFACE_MODE_RGMII_TXID:
> > > > > + case PHY_INTERFACE_MODE_RGMII_RXID:
> > > > > + phyif = PHY_INTF_RGMII;
> > > > > + break;
> > > > > + default:
> > > > > + dev_err(dwmac->dev, "unsupported phy interface %d\n",
> > > > > + plat->interface);
> > > > > + return -EINVAL;
> > > > > + };
> > > > > +
> > > > > + regmap_write(dwmac->apb_regmap, GMAC_INTF_CTRL, phyif);
> > > > > +
> > > > > + return 0;
> > > > > +}
> > > > > +
> > > > > +static int thead_dwmac_set_txclk_dir(struct plat_stmmacenet_data *plat)
> > > > > +{
> > > > > + struct thead_dwmac *dwmac = plat->bsp_priv;
> > > > > + u32 txclk_dir;
> > > > > +
> > > > > + switch (plat->interface) {
> > > > > + case PHY_INTERFACE_MODE_MII:
> > > > > + txclk_dir = TXCLK_DIR_INPUT;
> > > > > + break;
> > > > > + case PHY_INTERFACE_MODE_RGMII:
> > > > > + case PHY_INTERFACE_MODE_RGMII_ID:
> > > > > + case PHY_INTERFACE_MODE_RGMII_TXID:
> > > > > + case PHY_INTERFACE_MODE_RGMII_RXID:
> > > > > + txclk_dir = TXCLK_DIR_OUTPUT;
> > > > > + break;
> > > > > + default:
> > > > > + dev_err(dwmac->dev, "unsupported phy interface %d\n",
> > > > > + plat->interface);
> > > > > + return -EINVAL;
> > > > > + };
> > > > > +
> > > > > + regmap_write(dwmac->apb_regmap, GMAC_TXCLK_OEN, txclk_dir);
> > > > > +
> > > > > + return 0;
> > > > > +}
> > > > > +
> > > > > +static void thead_dwmac_fix_speed(void *priv, unsigned int speed, unsigned int mode)
> > > > > +{
> > > > > + struct thead_dwmac *dwmac = priv;
> > > > > + struct plat_stmmacenet_data *plat = dwmac->plat;
> > > > > + unsigned long rate;
> > > > > + u32 div;
> > > > > +
> > > > > + switch (plat->interface) {
> > > > > + /* For MII, rxc/txc is provided by phy */
> > > > > + case PHY_INTERFACE_MODE_MII:
> > > > > + return;
> > > > > +
> > > > > + case PHY_INTERFACE_MODE_RGMII:
> > > > > + case PHY_INTERFACE_MODE_RGMII_ID:
> > > > > + case PHY_INTERFACE_MODE_RGMII_RXID:
> > > > > + case PHY_INTERFACE_MODE_RGMII_TXID:
> > > >
> > > > > + rate = clk_get_rate(plat->stmmac_clk);
> > > > > + if (!rate || rate % GMAC_GMII_RGMII_RATE != 0 ||
> > > > > + rate % GMAC_MII_RATE != 0) {
> > > > > + dev_err(dwmac->dev, "invalid gmac rate %ld\n", rate);
> > > > > + return;
> > > > > + }
> > > > > +
> > > > > + regmap_update_bits(dwmac->apb_regmap, GMAC_PLLCLK_DIV, GMAC_PLLCLK_DIV_EN, 0);
> > > > > +
> > > > > + switch (speed) {
> > > > > + case SPEED_1000:
> > > > > + div = rate / GMAC_GMII_RGMII_RATE;
> > > > > + break;
> > > > > + case SPEED_100:
> > > > > + div = rate / GMAC_MII_RATE;
> > > > > + break;
> > > > > + case SPEED_10:
> > > > > + div = rate * 10 / GMAC_MII_RATE;
> > > > > + break;
> > > > > + default:
> > > > > + dev_err(dwmac->dev, "invalid speed %u\n", speed);
> > > > > + return;
> > > > > + }
> > > > > + regmap_update_bits(dwmac->apb_regmap, GMAC_PLLCLK_DIV,
> > > > > + GMAC_PLLCLK_DIV_MASK, GMAC_PLLCLK_DIV_NUM(div));
> > > > > +
> > > > > + regmap_update_bits(dwmac->apb_regmap, GMAC_PLLCLK_DIV,
> > > > > + GMAC_PLLCLK_DIV_EN, GMAC_PLLCLK_DIV_EN);
> > > >
> > > > This chunk looks like a hard-coded implementation of the
> > > > CLK_SET_RATE_GATE Tx-clocks rate setup which parental clock is the
> > > > "stmmaceth" clock. I suggest to move it to the respective driver, add
> > > > a "tx" clock to the bindings and use the common clock kernel API
> > > > methods here only.
> > >
> > > I did consider your solution before writing the code, here are the
> > > reasons why I dropped it:
> > >
> >
> > > There's no any clk IP here, the HW just puts several
> > > gmac related control bits here, such as rx/tx delay, bypass, invert
> > > interface choice, clk direction.
> >
> > You omitted the essential part of your code which I pointed out.
> >
> > > From this point of view, it looks more
> > > like a syscon rather than clk.
> >

> > Toggling control bits is surely the syscon work. But gating a parental
> > clock, settings up the parental clock _divider_ and ungating the clock
> > back is the clock controller function. So it means your syscon is just
> > a normal multi-function device, which one of the function is the clock
> > controller.
> >
> > It's not like your situation is unique. For instance in case of a SoC
> > I was working with recently Clock Control Unit (CCU) was actually a
> > multi-function device which had:
> > 1. PLLs and Dividers supplying the clocks to the SoC components.
>
> Hi Serge,
>
> This is the big difference between your case and TH1520 gmac.

AFAICS my PCIe-part of the syscon is a similar story as your GMAC
CSRs: enabled/disable several clock gates, setup a ref-clock divider,
multiple flags with reset function and multiple flags/fields for the
PCIe-controller tunings. All of that intermixed in a few registers. A
similar thing was for the SATA-controller.

> (PS: @Emil, I read your comments in another reply. IIUC, jh7110 puts a
> real clk IP for gmac tx clock purpose)
>

> However, There's no real clk IP in the TH1520 gmac related syscon, yep, div
> and enable are some what clock related bits, but that's all, no more, no less.

These bits/fields are the clock-controller function of the syscon.
Based on that your syscon is a multi-functional device which support
the GMAC tunings and controls a reference clock gating/dividing. The
outputs from the clock gate and divider gets to be supplied to the
TH1520 GMAC.

BTW Seeing you have only 0x20 bytes utilized in the glue driver I
guess those registers chunk is a part of a bigger CSR space. Is it?

> So even in this case, another abstraction layer via. clk subsystem is still
> preferred? IOW, a seperate clk driver for the gmac?

That's what I was told in a situation when I was trying to use a
syscon to switch between the two parental clocks:
https://lore.kernel.org/linux-ide/20220517201332.GB1462130-robh@xxxxxxxxxx/

In your case IMO it is more preferred at the very least because it
gives a more correct device description. Your bindings currently do
not define the Tx/Rx reference clocks meanwhile the TH1520 GMAC do
have them.

-Serge(y)

>
> Thanks
> >
> > 2. SoC components reset controller.
> > 3. I2C-interface controller.
> > 4. AXI-bus errors report registers.
> > 5. PCIe-controller tunings (LTSSM, link up/down, etc)
> > 6. SATA-controller tunings.
> > 7. Full SoC reset controller (syscon reboot),
> > 8. L2-cache tunings controller.
> > with the sub-functions CSRs joint in a single space. In that case the
> > PCIe-controller tunings and a lot of its reference clocks settings
> > were intermixed in a single chunk of the registers. So I had to create
> > a driver for the clocks anyway including all the PCIe reference
> > clock and refer to the syscon in the PCIe-controller device node for
> > the respective PCIe platform-specific tunings.
> >
> > >
> > > Secondly, I see other SoCs did similar for this case, such as
> > > dwmac-visconti, dwmac-meson8b, dwmac-ipq806x, dwmac-socfpga and so on.
> > > They met similar issue as the above.
> >
> > First I failed to find any clock-related things in the dwmac-socfpga
>
> I believe the ptp ref clk related is just to enable the clk by toggling
> corresponding bit. Anyway that's not important part here.
>
> > driver looking in anyway as yours. Second the dwmac-meson8b driver
> > creates a generic clock handler right in the driver. I don't think
> > it's a great solution but at the very least it registers the clock
> > handler in the kernel. But seeing the PROG_ETHERNET CSR is of 8 bytes
> > long there (0xc8834540 0x8) and defined at looking random base address
> > it's definitely a part of a Meson system controller which just
> > directly passed to the device driver. It's not correct. That part
> > should have been at least specified as a syscon too. Third the
> > dwmac-visconti driver is not a good example seeing it defines some
> > specific registers way away from the NIC CSR space. It's most likely a
> > separate device like syscon. Fourth dwmac-ipq806x driver
> > implementation looks indeed like yours.
>
> > In anyway I don't say your solution is fully wrong. At the very least
> > you have a syscon node defined. But it just makes you adding
> > incomplete device/platform bindings. Your network device do have the
> > Tx reference clock as a part of the separate system controller, but
> > you have to omit it because of the syscon property. You do have a
> > syscon node, but don't have its clock function exported. So AFAICS in
> > your case things can be implemented in a more canonical way than they
> > are now.
> >