Re: [PATCH v1 3/3] gpiolib: append SFI helpers for GPIO API

[Andy Shevchenko <andriy.shevchenko@linux.intel.com>]

On Tue, 2013-12-10 at 12:00 +0900, Alex Courbot wrote:
> On 12/09/2013 07:11 PM, Andy Shevchenko wrote:
> > On Fri, 2013-12-06 at 10:52 +0900, Alex Courbot wrote:
> >> On 12/06/2013 01:36 AM, Andy Shevchenko wrote:
> >>> To support some (legacy) firmwares and platforms let's make life easier for
> >>> their customers.
> >>>
> >>> This patch provides a function which converts sfi_gpio_table_entry to
> >>> gpio_desc. The use of it is integrated into GPIO library to enable generic
> >>> access to the SFI GPIO resources.
> >>>
> >>> Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> >>> ---
> >>>    drivers/gpio/Kconfig       |  4 ++++
> >>>    drivers/gpio/Makefile      |  1 +
> >>>    drivers/gpio/gpiolib-sfi.c | 28 ++++++++++++++++++++++++++++
> >>>    drivers/gpio/gpiolib.c     |  3 +++
> >>>    drivers/gpio/gpiolib.h     | 13 +++++++++++++
> >>>    5 files changed, 49 insertions(+)
> >>>    create mode 100644 drivers/gpio/gpiolib-sfi.c
> >>>
> >>> diff --git a/drivers/gpio/Kconfig b/drivers/gpio/Kconfig
> >>> index ae3682d..a12752a 100644
> >>> --- a/drivers/gpio/Kconfig
> >>> +++ b/drivers/gpio/Kconfig
> >>> @@ -51,6 +51,10 @@ config OF_GPIO
> >>>    	def_bool y
> >>>    	depends on OF
> >>>
> >>> +config GPIO_SFI
> >>> +	def_bool y
> >>> +	depends on SFI
> >>> +
> >>>    config GPIO_ACPI
> >>>    	def_bool y
> >>>    	depends on ACPI
> >>> diff --git a/drivers/gpio/Makefile b/drivers/gpio/Makefile
> >>> index ee95154..5373e3a 100644
> >>> --- a/drivers/gpio/Makefile
> >>> +++ b/drivers/gpio/Makefile
> >>> @@ -5,6 +5,7 @@ ccflags-$(CONFIG_DEBUG_GPIO)	+= -DDEBUG
> >>>    obj-$(CONFIG_GPIO_DEVRES)	+= devres.o
> >>>    obj-$(CONFIG_GPIOLIB)		+= gpiolib.o
> >>>    obj-$(CONFIG_OF_GPIO)		+= gpiolib-of.o
> >>> +obj-$(CONFIG_GPIO_SFI)		+= gpiolib-sfi.o
> >>>    obj-$(CONFIG_GPIO_ACPI)		+= gpiolib-acpi.o
> >>>
> >>>    # Device drivers. Generally keep list sorted alphabetically
> >>> diff --git a/drivers/gpio/gpiolib-sfi.c b/drivers/gpio/gpiolib-sfi.c
> >>> new file mode 100644
> >>> index 0000000..c804314
> >>> --- /dev/null
> >>> +++ b/drivers/gpio/gpiolib-sfi.c
> >>> @@ -0,0 +1,28 @@
> >>> +/*
> >>> + * Simple Firmware Interface (SFI) helpers for GPIO API
> >>> + *
> >>> + * Copyright (C) 2013, Intel Corporation
> >>> + * Author: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> >>> + *
> >>> + * This program is free software; you can redistribute it and/or modify
> >>> + * it under the terms of the GNU General Public License version 2 as
> >>> + * published by the Free Software Foundation.
> >>> + */
> >>> +
> >>> +#include <linux/gpio/consumer.h>
> >>> +#include <linux/sfi.h>
> >>> +#include <linux/errno.h>
> >>> +#include <linux/err.h>
> >>> +
> >>> +#include "gpiolib.h"
> >>> +
> >>> +struct gpio_desc *sfi_get_gpiod_by_name(const char *name)
> >>> +{
> >>> +	struct sfi_gpio_table_entry *pentry;
> >>> +
> >>> +	pentry = sfi_gpio_get_entry_by_name(name);
> >>> +	if (!pentry)
> >>> +		return ERR_PTR(-ENODEV);
> >>> +
> >>> +	return gpio_to_desc(pentry->pin_no);
> >>> +}
> >>> diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
> >>> index bad400c..789ae1c 100644
> >>> --- a/drivers/gpio/gpiolib.c
> >>> +++ b/drivers/gpio/gpiolib.c
> >>> @@ -2451,6 +2451,9 @@ struct gpio_desc *__must_check gpiod_get_index(struct device *dev,
> >>>    	} else if (IS_ENABLED(CONFIG_ACPI) && dev && ACPI_HANDLE(dev)) {
> >>>    		dev_dbg(dev, "using ACPI for GPIO lookup\n");
> >>>    		desc = acpi_find_gpio(dev, con_id, idx, &flags);
> >>> +	} else if (IS_ENABLED(CONFIG_SFI)) {
> >>> +		dev_dbg(dev, "using SFI for GPIO lookup\n");
> >>> +		desc = sfi_get_gpiod_by_name(con_id);
> >>
> >> Your lookup function is ignoring the dev argument. Are SFI GPIOs always
> >> supposed to be system-global?
> >
> > It's not clear. It could be device related, though SFI itself has
> > probably wrong design. I rather prefer to avoid a dev parameter check at
> > all.
> >
> >>   In this case, your if condition should
> >> likely be
> >>
> >> 	} else if (IS_ENABLED(CONFIG_SFI) && !dev) {
> >>
> >> So that a global SFI GPIO does not get mistakenly assigned to a device
> >> that has, say, a more suited platform mapping on the same con_id.
> >
> > So, for example in the driver that could be enumerated from SFI, DT, and
> > via platform data you suggest to have something like
> >
> >
> > desc = gpiod_get(dev, "con_id_device_tree");
> > ...
> >
> > if (IS_ERR(desc))
> >   desc = gpiod_get(NULL, "con_id_sfi");
> >
> > if (IS_ERR(desc))
> >   desc = gpiod_get(???, "con_id_from_platdata");
> >
> > Correct?
> 
> No, this is not what I'm suggesting. Device drivers should not care who 
> provides the GPIO, they should just ask for it, and obtain it (or not).
> 
> The scope of the problem actually depends on what SFI GPIOs are used 
> for. For instance, let's say you have two devices each using an enabling 
> GPIO which happens to be provided by SFI. Both drivers for these devices 
> obtain the enable GPIO as follows:
> 
> 	enable_gpio = gpiod_get(dev, "enable");

You rather can't do that in sfi case. Yes, the structure has (stringy)
reference to gpio chip that provides a line, but in practice the gpio
line name should be unique. Consider this is misdesign of SFI as I
mentioned earlier.

> 
> Here you actually have two problems:
> 
> 1) Since you only look for con_id, how to you discriminate the enable 
> GPIOs for these devices?

> 2) The device drivers are the one to decide which GPIO name they 
> request. How are SFI GPIO names decided? If you don't have any kind of 
> flexibility for their naming, and want to use them with devices drivers, 
> you will very likely need another naming layer that associates a 
> (device_name, con_id) pair to the right SFI GPIO, similarly to what is 
> done with platform GPIOs. If the only consumer of SFI GPIOs is platform 
> code, and SFI GPIOs are all uniquely named, then you may as well request 
> the device to be NULL for their lookup so that they don't interfere with 
> more precisely-mapped GPIOs.

So, you insist to have !dev there?

> 
> I don't know anything about SFI GPIOs, how they are defined, where their 
> name comes from, and how they are used so my vision may be incomplete. 
> But AFAICT it all comes down to one of these two scenarios:
> 
> 1) SFI GPIOs are only used in platform code -> using their pin name is 
> ok, device argument should be assumed to be NULL for their matching

Let's stick to this.

> 2) SFI GPIOs are also consumed by device drivers -> you need a way to 
> match a (device, con_id) pair to your SFI GPIOs so they can be matched 
> exactly and through the names drivers will request.

Mostly unlikely we go this way. It would mean we don't need SFI at all.


-- 
Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Intel Finland Oy