Re: [PATCH v2 3/5] platform: surface: Add surface xbl

[Bjorn Andersson]

On Mon 08 Nov 08:44 PST 2021, Jarrett Schultz wrote:

> Introduce support for the Extensible Boot Loader driver found on the
> Surface Duo. Makes device information available to users via sysfs.
> 
> Signed-off-by: Jarrett Schultz <jaschultz@xxxxxxxxxxxxx>
> 
> ---
> 
> Changes in v2:
>  - Added types.h inclusion and removed unused inclusions
>  - Minor updates to code and acronym style
>  - Remove __packed attribute on driver struct
>  - Use .dev_groups for sysfs
>  - Added more in-depth description of driver in Kconfig
>  - Changed target KernelVersion in sysfs documentation
> 
> ---
> 
>  .../ABI/testing/sysfs-platform-surface-xbl    |  78 +++++++
>  MAINTAINERS                                   |   2 +
>  drivers/platform/surface/Kconfig              |  12 +
>  drivers/platform/surface/Makefile             |   1 +
>  drivers/platform/surface/surface-xbl.c        | 215 ++++++++++++++++++
>  5 files changed, 308 insertions(+)
>  create mode 100644 Documentation/ABI/testing/sysfs-platform-surface-xbl
>  create mode 100644 drivers/platform/surface/surface-xbl.c
> 
> diff --git a/Documentation/ABI/testing/sysfs-platform-surface-xbl b/Documentation/ABI/testing/sysfs-platform-surface-xbl
> new file mode 100644
> index 000000000000..2ae047b884d3
> --- /dev/null
> +++ b/Documentation/ABI/testing/sysfs-platform-surface-xbl
> @@ -0,0 +1,78 @@
> +What:		/sys/devices/platform/146bfa94.xbl/battery_present
> +Date:		October 2021
> +KernelVersion:	5.16
> +Contact:	jaschultz@xxxxxxxxxxxxx
> +Description:
> +		Read only. It returns whether the battery is present. Valid
> +		values are:
> +			0 - battery absent
> +			1 - battery present

Would this information not be available from some battery driver, under
/sys/class/power_supply?

> +
> +What:		/sys/devices/platform/146bfa94.xbl/board_id
> +Date:		October 2021
> +KernelVersion:	5.16
> +Contact:	jaschultz@xxxxxxxxxxxxx
> +Description:
> +		Read only. It returns the board id.

Is this a Microsoft-specific board id, or does it relate to the Qualcomm
socinfo property with the same name?

> +
> +What:		/sys/devices/platform/146bfa94.xbl/hw_init_retries
> +Date:		October 2021
> +KernelVersion:	5.16
> +Contact:	jaschultz@xxxxxxxxxxxxx
> +Description:
> +		Read only. It returns retries attempted to initialize the
> +		discrete hardware circuit.

Which description hardware circuit?

> +
> +What:		/sys/devices/platform/146bfa94.xbl/is_act_mode
> +Date:		October 2021
> +KernelVersion:	5.16
> +Contact:	jaschultz@xxxxxxxxxxxxx
> +Description:
> +		Read only. It returns whether ACT mode is enabled. Valid values
> +		are:
> +			0 - ACT disabled
> +			1 - ACT enabled
> +
> +		ACT mode is used to run checks and put the device to shipmode
> +		at factory.
> +
> +What:		/sys/devices/platform/146bfa94.xbl/is_customer_mode
> +Date:		October 2021
> +KernelVersion:	5.16
> +Contact:	jaschultz@xxxxxxxxxxxxx
> +Description:
> +		Read only. It returns whether the device is in manufacturing
> +		mode. Valid values are:
> +			0 - Not in manufacturing mode
> +			1 - In manufacturing mode
> +
> +What:		/sys/devices/platform/146bfa94.xbl/ocp_error_location
> +Date:		October 2021
> +KernelVersion:	5.16
> +Contact:	jaschultz@xxxxxxxxxxxxx
> +Description:
> +		Read only. It returns 0 or which power rail has the OCP error.

Sounds like the reason is singular, so why is this a bitmask?

> +		Valid values are:
> +			Bit(s)		Meaning
> +			15		More than one OCP error occurred
> +			14-12		PMIC
> +			11-7		SMPS
> +			6-2		LDO
> +			1-0		BOB
> +
> +What:		/sys/devices/platform/146bfa94.xbl/pmic_reset_reason
> +Date:		October 2021
> +KernelVersion:	5.16
> +Contact:	jaschultz@xxxxxxxxxxxxx
> +Description:
> +		Read only. It returns the reason for the reset. Valid values
> +		are:

Is this different from the PMIC reset reason that we read from the PMIC?
Could we make sure to expose this generically for all Qualcomm PMICs?

> +			0 - no reason lol
> +			9 - Battery driver triggered
> +
> +What:		/sys/devices/platform/146bfa94.xbl/touch_fw_version
> +Date:		October 2021
> +KernelVersion:	5.16
> +Contact:	jaschultz@xxxxxxxxxxxxx
> +Description:
> +		Read only. It returns the version of the firmware.

Why isn't this exposed by the touchscreen driver instead?


Generally I wonder how you're consuming this information in userspace.
Is it only for debugging purposes, i.e. would debugfs be a better place?

> diff --git a/MAINTAINERS b/MAINTAINERS
> index 8643546f8fab..d08b68d626f6 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -12428,7 +12428,9 @@ M:	Jarrett Schultz <jaschultz@xxxxxxxxxxxxx>
>  L:	linux-arm-msm@xxxxxxxxxxxxxxx
>  L:	platform-driver-x86@xxxxxxxxxxxxxxx
>  S:	Supported
> +F:	Documentation/ABI/testing/sysfs-platform-surface-xbl
>  F:	Documentation/devicetree/bindings/platform/microsoft/surface-xbl.yaml
> +F:	drivers/platform/surface/surface-xbl.c
>  
>  MICROSOFT SURFACE GPE LID SUPPORT DRIVER
>  M:	Maximilian Luz <luzmaximilian@xxxxxxxxx>
> diff --git a/drivers/platform/surface/Kconfig b/drivers/platform/surface/Kconfig
> index 0d3970e1d144..3a1ced269d96 100644
> --- a/drivers/platform/surface/Kconfig
> +++ b/drivers/platform/surface/Kconfig
> @@ -190,6 +190,18 @@ config SURFACE_PRO3_BUTTON
>  	help
>  	  This driver handles the power/home/volume buttons on the Microsoft Surface Pro 3/4 tablet.
>  
> +config SURFACE_XBL
> +        tristate "Surface XBL Driver"
> +        depends on ARM64 || COMPILE_TEST
> +        depends on OF
> +        help
> +          If you say 'Y' to this option, support will be included for the
> +          Surface Extensible Boot Loader (XBL) Driver. This driver exposes
> +          information about the device through sysfs.
> +
> +          This driver can also be built as a module.  If so, the module
> +          will be called surface-xbl.
> +
>  source "drivers/platform/surface/aggregator/Kconfig"
>  
>  endif # SURFACE_PLATFORMS
> diff --git a/drivers/platform/surface/Makefile b/drivers/platform/surface/Makefile
> index 32889482de55..0946266a8a73 100644
> --- a/drivers/platform/surface/Makefile
> +++ b/drivers/platform/surface/Makefile
> @@ -16,3 +16,4 @@ obj-$(CONFIG_SURFACE_GPE)		+= surface_gpe.o
>  obj-$(CONFIG_SURFACE_HOTPLUG)		+= surface_hotplug.o
>  obj-$(CONFIG_SURFACE_PLATFORM_PROFILE)	+= surface_platform_profile.o
>  obj-$(CONFIG_SURFACE_PRO3_BUTTON)	+= surfacepro3_button.o
> +obj-$(CONFIG_SURFACE_XBL)               += surface-xbl.o

All other files in this directory are named with an underscore, would be
nice to carry on with such convention.

> diff --git a/drivers/platform/surface/surface-xbl.c b/drivers/platform/surface/surface-xbl.c
> new file mode 100644
> index 000000000000..9ecec4e55a4d
> --- /dev/null
> +++ b/drivers/platform/surface/surface-xbl.c
> @@ -0,0 +1,215 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * Surface eXtensible Boot Loader (XBL)
> + *
> + * Copyright (C) 2021 Microsoft Corporation
> + * Author: Jarrett Schultz <jaschultz@xxxxxxxxxxxxx>
> + */
> +
> +#include <linux/io.h>
> +#include <linux/module.h>
> +#include <linux/of.h>
> +#include <linux/platform_device.h>
> +#include <linux/types.h>
> +
> +#define SURFACE_XBL_MAX_VERSION_LEN	16
> +#define SURFACE_XBL_BOARD_ID		0
> +#define SURFACE_XBL_BATTERY_PRESENT	1
> +#define SURFACE_XBL_HW_INIT_RETRIES	2
> +#define SURFACE_XBL_IS_CUSTOMER_MODE	3
> +#define SURFACE_XBL_IS_ACT_MODE		4
> +#define SURFACE_XBL_PMIC_RESET_REASON	5
> +#define SURFACE_XBL_TOUCH_FW_VERSION	6
> +#define SURFACE_XBL_OCP_ERROR_LOCATION		\
> +		(SURFACE_XBL_TOUCH_FW_VERSION +	\
> +		SURFACE_XBL_MAX_VERSION_LEN)
> +
> +struct surface_xbl {
> +	struct device	*dev;
> +	void __iomem	*regs;
> +
> +	u8		board_id;
> +	u8		battery_present;
> +	u8		hw_init_retries;
> +	u8		is_customer_mode;
> +	u8		is_act_mode;
> +	u8		pmic_reset_reason;
> +	char		touch_fw_version[SURFACE_XBL_MAX_VERSION_LEN];
> +	u16		ocp_error_location;
> +};
> +
> +static ssize_t
> +board_id_show(struct device *dev, struct device_attribute *attr, char *buf)

I think it would be nice to avoid some duplication by putting all these
integer ones in a single show(), see e.g. soc_info_show()

> +{
> +	struct surface_xbl	*sxbl = dev_get_drvdata(dev);
> +
> +	return sysfs_emit(buf, "%d\n", sxbl->board_id);
> +}
> +static DEVICE_ATTR_RO(board_id);
> +
> +static ssize_t
> +battery_present_show(struct device *dev, struct device_attribute *attr, char *buf)
> +{
> +	struct surface_xbl	*sxbl = dev_get_drvdata(dev);
> +
> +	return sysfs_emit(buf, "%d\n", sxbl->battery_present);
> +}
> +static DEVICE_ATTR_RO(battery_present);
> +
> +static ssize_t
> +hw_init_retries_show(struct device *dev, struct device_attribute *attr, char *buf)
> +{
> +	struct surface_xbl	*sxbl = dev_get_drvdata(dev);
> +
> +	return sysfs_emit(buf, "%d\n", sxbl->hw_init_retries);
> +}
> +static DEVICE_ATTR_RO(hw_init_retries);
> +
> +static ssize_t
> +is_customer_mode_show(struct device *dev, struct device_attribute *attr, char *buf)
> +{
> +	struct surface_xbl	*sxbl = dev_get_drvdata(dev);
> +
> +	return sysfs_emit(buf, "%d\n", sxbl->is_customer_mode);
> +}
> +static DEVICE_ATTR_RO(is_customer_mode);
> +
> +static ssize_t
> +is_act_mode_show(struct device *dev, struct device_attribute *attr, char *buf)
> +{
> +	struct surface_xbl	*sxbl = dev_get_drvdata(dev);
> +
> +	return sysfs_emit(buf, "%d\n", sxbl->is_act_mode);
> +}
> +static DEVICE_ATTR_RO(is_act_mode);
> +
> +static ssize_t
> +pmic_reset_reason_show(struct device *dev, struct device_attribute *attr, char *buf)
> +{
> +	struct surface_xbl	*sxbl = dev_get_drvdata(dev);
> +
> +	return sysfs_emit(buf, "%d\n", sxbl->pmic_reset_reason);
> +}
> +static DEVICE_ATTR_RO(pmic_reset_reason);
> +
> +static ssize_t
> +touch_fw_version_show(struct device *dev, struct device_attribute *attr, char *buf)
> +{
> +	struct surface_xbl	*sxbl = dev_get_drvdata(dev);
> +
> +	return sysfs_emit(buf, "0x%s\n", sxbl->touch_fw_version);
> +}
> +static DEVICE_ATTR_RO(touch_fw_version);
> +
> +static ssize_t
> +ocp_error_location_show(struct device *dev, struct device_attribute *attr, char *buf)
> +{
> +	struct surface_xbl	*sxbl = dev_get_drvdata(dev);
> +
> +	return sysfs_emit(buf, "%d\n", sxbl->ocp_error_location);
> +}
> +static DEVICE_ATTR_RO(ocp_error_location);
> +
> +static struct attribute *xbl_attrs[] = {
> +	&dev_attr_board_id.attr,
> +	&dev_attr_battery_present.attr,
> +	&dev_attr_hw_init_retries.attr,
> +	&dev_attr_is_customer_mode.attr,
> +	&dev_attr_is_act_mode.attr,
> +	&dev_attr_pmic_reset_reason.attr,
> +	&dev_attr_touch_fw_version.attr,
> +	&dev_attr_ocp_error_location.attr,
> +	NULL
> +};
> +
> +static const struct attribute_group xbl_attr_group = {
> +	.attrs = xbl_attrs,
> +};
> +
> +const struct attribute_group *xbl_sysfs_groups[] = {
> +	&xbl_attr_group,
> +	NULL
> +};
> +
> +static u8 surface_xbl_readb(void __iomem *base, u32 offset)
> +{
> +	return readb(base + offset);

Instead of having these helpers I think you should just call readb(base
+ offset) directly below.

The shorter function name (readb vs surface_xbl_readb) means that you
don't even need to line wrap those lines.

> +}
> +
> +static u16 surface_xbl_readw(void __iomem *base, u32 offset)
> +{
> +	return readw(base + offset);
> +}
> +
> +static int surface_xbl_probe(struct platform_device *pdev)
> +{
> +	struct surface_xbl	*sxbl;
> +	struct device		*dev;
> +	void __iomem		*regs;
> +	int			index;
> +
> +	dev = &pdev->dev;
> +	sxbl = devm_kzalloc(dev, sizeof(*sxbl), GFP_KERNEL);

This is the only use of &pdev->dev, so put that here and drop "dev" from
sxbl (and the stack).

> +	if (!sxbl)
> +		return -ENOMEM;
> +
> +	sxbl->dev = dev;
> +
> +	regs = devm_platform_ioremap_resource(pdev, 0);
> +	if (IS_ERR(regs))
> +		return PTR_ERR(regs);
> +
> +	sxbl->regs = regs;

Seems only reason you stash "regs" in sxbl is so that you can pass
sxbl->regs in below function calls. I.e. it's a local variable and you
can omit it from struct surface_xbl...

> +
> +	platform_set_drvdata(pdev, sxbl);
> +
> +	sxbl->board_id = surface_xbl_readb(sxbl->regs,
> +					   SURFACE_XBL_BOARD_ID);
> +	sxbl->battery_present = surface_xbl_readb(sxbl->regs,
> +						  SURFACE_XBL_BATTERY_PRESENT);
> +	sxbl->hw_init_retries = surface_xbl_readb(sxbl->regs,
> +						  SURFACE_XBL_HW_INIT_RETRIES);
> +	sxbl->is_customer_mode = surface_xbl_readb(sxbl->regs,
> +						   SURFACE_XBL_IS_CUSTOMER_MODE);
> +	sxbl->is_act_mode = surface_xbl_readb(sxbl->regs,
> +					      SURFACE_XBL_IS_ACT_MODE);
> +	sxbl->pmic_reset_reason = surface_xbl_readb(sxbl->regs,
> +						    SURFACE_XBL_PMIC_RESET_REASON);
> +
> +	for (index = 0; index < SURFACE_XBL_MAX_VERSION_LEN; index++)

"i" is a good succinct variable name for an index.

> +		sxbl->touch_fw_version[index] = surface_xbl_readb(sxbl->regs,
> +							SURFACE_XBL_TOUCH_FW_VERSION + index);
> +
> +	sxbl->ocp_error_location = surface_xbl_readw(sxbl->regs,
> +						     SURFACE_XBL_OCP_ERROR_LOCATION);
> +
> +	return 0;
> +}
> +
> +static int surface_xbl_remove(struct platform_device *pdev)

Your remove function is empty, simply omit it...

Regards,
Bjorn

> +{
> +	return 0;
> +}
> +
> +static const struct of_device_id surface_xbl_of_match[] = {
> +	{
> +		.compatible = "microsoft,sm8150-surface-duo-xbl"
> +	},
> +	{  }
> +};
> +MODULE_DEVICE_TABLE(of, surface_xbl_of_match);
> +
> +static struct platform_driver surface_xbl_driver = {
> +	.driver	= {
> +		.name		= "surface-xbl",
> +		.of_match_table = surface_xbl_of_match,
> +		.dev_groups	= xbl_sysfs_groups
> +	},
> +	.probe		= surface_xbl_probe,
> +	.remove		= surface_xbl_remove
> +};
> +module_platform_driver(surface_xbl_driver);
> +
> +MODULE_AUTHOR("Jarrett Schultz <jaschultz@xxxxxxxxxxxxx>");
> +MODULE_DESCRIPTION("Surface Extensible Bootloader");
> +MODULE_LICENSE("GPL");
> -- 
> 2.25.1
>