Re: [PATCH] net: phy: hisi-festa: Add support for HiSilicon Festa PHYs

[Heiner Kallweit <hkallweit1@gmail.com>]

On 17.03.2023 15:30, David Yang wrote:
> HiSilicon Festa PHYs were used on some HiSilicon SoCs. This patch injects
> firmwares found on vendor kernels.
> 
What's the status of adding the firmware files to linux-firmware?
I don't see any related patch in the linux-firmware mailing list archive.

Any info on purpose of firmware? Does the PHY work normally also
w/o firmware? Or is the firmware required?

> Signed-off-by: David Yang <mmyangfl@gmail.com>
> ---
>  drivers/net/phy/Kconfig      |   5 ++
>  drivers/net/phy/Makefile     |   1 +
>  drivers/net/phy/hisi-festa.c | 169 +++++++++++++++++++++++++++++++++++
>  3 files changed, 175 insertions(+)
>  create mode 100644 drivers/net/phy/hisi-festa.c
> 
> diff --git a/drivers/net/phy/Kconfig b/drivers/net/phy/Kconfig
> index 54874555c..e7551e9b3 100644
> --- a/drivers/net/phy/Kconfig
> +++ b/drivers/net/phy/Kconfig
> @@ -177,6 +177,11 @@ config DAVICOM_PHY
>  	help
>  	  Currently supports dm9161e and dm9131
>  
> +config HISI_FESTA_PHY
> +	tristate "HiSilicon Festa PHYs"
> +	help
> +	  Supports the HiSilicon Festa PHYs.
> +
>  config ICPLUS_PHY
>  	tristate "ICPlus PHYs"
>  	help
> diff --git a/drivers/net/phy/Makefile b/drivers/net/phy/Makefile
> index b5138066b..2c5aded6b 100644
> --- a/drivers/net/phy/Makefile
> +++ b/drivers/net/phy/Makefile
> @@ -60,6 +60,7 @@ obj-$(CONFIG_DP83869_PHY)	+= dp83869.o
>  obj-$(CONFIG_DP83TC811_PHY)	+= dp83tc811.o
>  obj-$(CONFIG_DP83TD510_PHY)	+= dp83td510.o
>  obj-$(CONFIG_FIXED_PHY)		+= fixed_phy.o
> +obj-$(CONFIG_HISI_FESTA_PHY)	+= hisi-festa.o
>  obj-$(CONFIG_ICPLUS_PHY)	+= icplus.o
>  obj-$(CONFIG_INTEL_XWAY_PHY)	+= intel-xway.o
>  obj-$(CONFIG_LSI_ET1011C_PHY)	+= et1011c.o
> diff --git a/drivers/net/phy/hisi-festa.c b/drivers/net/phy/hisi-festa.c
> new file mode 100644
> index 000000000..ab54ed3ca
> --- /dev/null
> +++ b/drivers/net/phy/hisi-festa.c
> @@ -0,0 +1,169 @@
> +// SPDX-License-Identifier: GPL-2.0-or-later OR MIT
> +/*
> + * Driver for HiSilicon Festa PHYs
> + *
> + * This module does nothing than firmware injection. If you don't use firmware,
> + * simply blacklist this module.

That's not an appropriate hint. Relevant is the answer to the question
whether firmware is required.

> + *
> + * Copyright (c) 2023 David Yang
> + */
> +#include <linux/errno.h>
> +#include <linux/firmware.h>
> +#include <linux/init.h>
> +#include <linux/mii.h>
> +#include <linux/module.h>
> +#include <linux/phy.h>
> +
> +#define PHY_ID_HISILICON_FESTAV200	0x20669813
> +#define PHY_ID_HISILICON_FESTAV220	0x20669823
> +#define PHY_ID_HISILICON_FESTAV300	0x20669833
> +#define PHY_ID_HISILICON_FESTAV320	0x20669843
> +#define PHY_ID_HISILICON_FESTAV330	0x20669853
> +#define PHY_ID_HISILICON_FESTAV331	0x20669863
> +
> +#define MII_EXPMD	0x1d	/* Expanded memory data */
> +#define MII_EXPMA	0x1e	/* Expanded memory address */
> +
> +/* bus->mdio_lock should be locked when using this function */
> +static inline int hisi_festa_read_expanded(struct phy_device *phydev, u16 addr)

No inline keyword please, let the compiler decide.

> +{
> +	__phy_write(phydev, MII_EXPMA, addr);
> +	return __phy_read(phydev, MII_EXPMD);
> +}
> +
> +/* bus->mdio_lock should be locked when using this function */
> +static inline int hisi_festa_write_expanded(struct phy_device *phydev, u16 addr, u8 val)

Why return type int and not void?

> +{
> +	__phy_write(phydev, MII_EXPMA, addr);
> +	__phy_write(phydev, MII_EXPMD, val);
> +	return 0;
> +}
> +
> +/* bus->mdio_lock should be locked when using this function */
> +static inline int hisi_festa_write_expanded_mem(struct phy_device *phydev, u16 addr,
> +						const u8 *data, int len)
> +{
> +	int i;
> +
> +	for (i = 0; i < len; i++)
> +		hisi_festa_write_expanded(phydev, addr + i, data[i]);
> +	return 0;
> +}
> +
> +static int hisi_festa_write_fw(struct phy_device *phydev, const struct firmware *fw)
> +{
> +	static const u8 prologue[] = {0xbd, 0x34, 0x00, 0x39};
> +	int ret;
> +
> +	phy_lock_mdio_bus(phydev);
> +
> +	ret = __phy_set_bits(phydev, MII_BMCR, BMCR_PDOWN);

Why power down the PHY?

> +	if (ret) {
> +		phydev_err(phydev, "cannot suspend device\n");
> +		goto out;
> +	}
> +
> +	hisi_festa_write_expanded_mem(phydev, 0x33f9, prologue, sizeof(prologue));
> +	/* mask jump instruction */
> +	hisi_festa_write_expanded(phydev, 0x3400, 0x39);
> +	hisi_festa_write_expanded_mem(phydev, 0x3401, fw->data + 1, fw->size - 1);
> +	/* now release firmware */
> +	hisi_festa_write_expanded(phydev, 0x3400, fw->data[0]);
> +	hisi_festa_write_expanded(phydev, 0x33f8, 0x01);
> +
> +	ret = __phy_clear_bits(phydev, MII_BMCR, BMCR_PDOWN);

Or do a soft reset here?

> +
> +out:
> +	phy_unlock_mdio_bus(phydev);
> +	return ret;
> +}
> +
> +static int hisi_festa_patch_fw(struct phy_device *phydev)
> +{
> +	int ret;
> +	char fw_name[64];
> +	const struct firmware *fw;
> +

reverse x-mas tree

> +	snprintf(fw_name, sizeof(fw_name), "hisilicon/festa.%08x.ucode", phydev->phy_id);
> +
> +	ret = request_firmware(&fw, fw_name, &phydev->mdio.dev);

Any option to detect whether firmware is loaded already?

> +	if (ret) {
> +		/* err message already printed by request_firmware */
> +		return -EAGAIN;
> +	}
> +
> +	if (fw->data[0] != 0x01 || fw->data[1] != 0xcc) {
> +		phydev_err(phydev, "%s does not look like valid firmware; refused to load\n",
> +			   fw_name);
> +		ret = -EINVAL;
> +		goto out;
> +	}
> +
> +	ret = hisi_festa_write_fw(phydev, fw);
> +	if (ret) {
> +		phydev_err(phydev, "download firmware %s failed\n", fw_name);
> +		goto out;
> +	}
> +
> +	phydev_info(phydev, "using firmware %s\n", fw_name);
> +
> +out:
> +	release_firmware(fw);
> +	return ret;
> +}
> +
> +static int hisi_festa_config_init(struct phy_device *phydev)
> +{
> +	hisi_festa_patch_fw(phydev);
> +	/* ok, use programmed firmware */
> +	return 0;
> +}
> +
> +static struct phy_driver hisi_festa_driver[] = {
> +	{
> +		PHY_ID_MATCH_MODEL(PHY_ID_HISILICON_FESTAV200),
> +		.name        = "HiSilicon Festa v200/v210",
> +		.config_init = hisi_festa_config_init,

How about callbacks like suspend/resume?

> +	},
> +	{
> +		PHY_ID_MATCH_MODEL(PHY_ID_HISILICON_FESTAV220),
> +		.name        = "HiSilicon Festa v220",
> +		.config_init = hisi_festa_config_init,
> +	},
> +	{
> +		PHY_ID_MATCH_MODEL(PHY_ID_HISILICON_FESTAV300),
> +		.name        = "HiSilicon Festa v300",
> +		.config_init = hisi_festa_config_init,
> +	},
> +	{
> +		PHY_ID_MATCH_MODEL(PHY_ID_HISILICON_FESTAV320),
> +		.name        = "HiSilicon Festa v320",
> +		.config_init = hisi_festa_config_init,
> +	},
> +	{
> +		PHY_ID_MATCH_MODEL(PHY_ID_HISILICON_FESTAV330),
> +		.name        = "HiSilicon Festa v330",
> +		.config_init = hisi_festa_config_init,
> +	},
> +	{
> +		PHY_ID_MATCH_MODEL(PHY_ID_HISILICON_FESTAV331),
> +		.name        = "HiSilicon Festa v331",
> +		.config_init = hisi_festa_config_init,
> +	},
> +};
> +
> +module_phy_driver(hisi_festa_driver);
> +
> +static struct mdio_device_id __maybe_unused hisi_festa_tbl[] = {
> +	{ PHY_ID_MATCH_MODEL(PHY_ID_HISILICON_FESTAV200) },
> +	{ PHY_ID_MATCH_MODEL(PHY_ID_HISILICON_FESTAV220) },
> +	{ PHY_ID_MATCH_MODEL(PHY_ID_HISILICON_FESTAV300) },
> +	{ PHY_ID_MATCH_MODEL(PHY_ID_HISILICON_FESTAV320) },
> +	{ PHY_ID_MATCH_MODEL(PHY_ID_HISILICON_FESTAV330) },
> +	{ PHY_ID_MATCH_MODEL(PHY_ID_HISILICON_FESTAV331) },
> +	{ }
> +};
> +
> +MODULE_DEVICE_TABLE(mdio, hisi_festa_tbl);
> +MODULE_DESCRIPTION("HiSilicon Festa PHY driver");
> +MODULE_LICENSE("Dual MIT/GPL");