[v13,03/10] mux: minimal mux subsystem and gpio-based mux controller

Submitted by Peter Rosin on April 13, 2017, 4:43 p.m.

Details

Message ID 1492101794-13444-4-git-send-email-peda@axentia.se
State Superseded
Headers show

Commit Message

Peter Rosin April 13, 2017, 4:43 p.m.
Add a new minimalistic subsystem that handles multiplexer controllers.
When multiplexers are used in various places in the kernel, and the
same multiplexer controller can be used for several independent things,
there should be one place to implement support for said multiplexer
controller.

A single multiplexer controller can also be used to control several
parallel multiplexers, that are in turn used by different subsystems
in the kernel, leading to a need to coordinate multiplexer accesses.
The multiplexer subsystem handles this coordination.

This new mux controller subsystem initially comes with a single backend
driver that controls gpio based multiplexers. Even though not needed by
this initial driver, the mux controller subsystem is prepared to handle
chips with multiple (independent) mux controllers.

Reviewed-by: Jonathan Cameron <jic23@kernel.org>
Signed-off-by: Peter Rosin <peda@axentia.se>
---
 Documentation/driver-model/devres.txt |   8 +
 MAINTAINERS                           |   2 +
 drivers/Kconfig                       |   2 +
 drivers/Makefile                      |   1 +
 drivers/mux/Kconfig                   |  34 +++
 drivers/mux/Makefile                  |   6 +
 drivers/mux/mux-core.c                | 422 ++++++++++++++++++++++++++++++++++
 drivers/mux/mux-gpio.c                | 114 +++++++++
 include/linux/mux.h                   | 252 ++++++++++++++++++++
 9 files changed, 841 insertions(+)
 create mode 100644 drivers/mux/Kconfig
 create mode 100644 drivers/mux/Makefile
 create mode 100644 drivers/mux/mux-core.c
 create mode 100644 drivers/mux/mux-gpio.c
 create mode 100644 include/linux/mux.h

Comments

Philipp Zabel April 18, 2017, 8:34 a.m.
Hi Peter,

On Thu, 2017-04-13 at 18:43 +0200, Peter Rosin wrote:
[...]
> diff --git a/include/linux/mux.h b/include/linux/mux.h
> new file mode 100644
> index 000000000000..febdde4246df
> --- /dev/null
> +++ b/include/linux/mux.h
[...]

Consider separating mux.h into a consumer header and a driver header.
Right now there is no separation between the consumer API and the
framework internals.

regards
Philipp

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Greg KH April 18, 2017, 8:51 a.m.
On Thu, Apr 13, 2017 at 06:43:07PM +0200, Peter Rosin wrote:
> +config MUX_GPIO
> +	tristate "GPIO-controlled Multiplexer"
> +	depends on OF && GPIOLIB

Why have the gpio and mux core in the same patch?

And why does this depend on OF?


> +	help
> +	  GPIO-controlled Multiplexer controller.
> +
> +	  The driver builds a single multiplexer controller using a number
> +	  of gpio pins. For N pins, there will be 2^N possible multiplexer
> +	  states. The GPIO pins can be connected (by the hardware) to several
> +	  multiplexers, which in that case will be operated in parallel.
> +
> +	  To compile the driver as a module, choose M here: the module will
> +	  be called mux-gpio.
> +
> +endif
> diff --git a/drivers/mux/Makefile b/drivers/mux/Makefile
> new file mode 100644
> index 000000000000..bb16953f6290
> --- /dev/null
> +++ b/drivers/mux/Makefile
> @@ -0,0 +1,6 @@
> +#
> +# Makefile for multiplexer devices.
> +#
> +
> +obj-$(CONFIG_MULTIPLEXER)	+= mux-core.o
> +obj-$(CONFIG_MUX_GPIO)		+= mux-gpio.o
> diff --git a/drivers/mux/mux-core.c b/drivers/mux/mux-core.c
> new file mode 100644
> index 000000000000..66a8bccfc3d7
> --- /dev/null
> +++ b/drivers/mux/mux-core.c
> @@ -0,0 +1,422 @@
> +/*
> + * Multiplexer subsystem
> + *
> + * Copyright (C) 2017 Axentia Technologies AB
> + *
> + * Author: Peter Rosin <peda@axentia.se>
> + *
> + * 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.
> + */
> +
> +#define pr_fmt(fmt) "mux-core: " fmt
> +
> +#include <linux/device.h>
> +#include <linux/err.h>
> +#include <linux/export.h>
> +#include <linux/idr.h>
> +#include <linux/init.h>
> +#include <linux/module.h>
> +#include <linux/mux.h>
> +#include <linux/of.h>
> +#include <linux/of_platform.h>
> +#include <linux/slab.h>
> +
> +/*
> + * The idle-as-is "state" is not an actual state that may be selected, it
> + * only implies that the state should not be changed. So, use that state
> + * as indication that the cached state of the multiplexer is unknown.
> + */
> +#define MUX_CACHE_UNKNOWN MUX_IDLE_AS_IS
> +
> +static struct class mux_class = {
> +	.name = "mux",
> +	.owner = THIS_MODULE,
> +};

No Documentation/ABI/ update for your sysfs files?  Please do so.

> +
> +static int __init mux_init(void)
> +{
> +	return class_register(&mux_class);
> +}
> +
> +static DEFINE_IDA(mux_ida);

When your module is unloaded, you forgot to clean this structure up with
what was done with it.

> +
> +static void mux_chip_release(struct device *dev)
> +{
> +	struct mux_chip *mux_chip = to_mux_chip(dev);
> +
> +	ida_simple_remove(&mux_ida, mux_chip->id);
> +	kfree(mux_chip);
> +}
> +
> +static struct device_type mux_type = {
> +	.name = "mux-chip",
> +	.release = mux_chip_release,
> +};
> +
> +struct mux_chip *mux_chip_alloc(struct device *dev,
> +				unsigned int controllers, size_t sizeof_priv)
> +{
> +	struct mux_chip *mux_chip;
> +	int i;
> +
> +	if (WARN_ON(!dev || !controllers))
> +		return NULL;
> +
> +	mux_chip = kzalloc(sizeof(*mux_chip) +
> +			   controllers * sizeof(*mux_chip->mux) +
> +			   sizeof_priv, GFP_KERNEL);
> +	if (!mux_chip)
> +		return NULL;

You don't return PTR_ERR(-ENOMEM)?  Ok, why not?  (I'm not arguing for
it, just curious...)

> +
> +	mux_chip->mux = (struct mux_control *)(mux_chip + 1);
> +	mux_chip->dev.class = &mux_class;
> +	mux_chip->dev.type = &mux_type;
> +	mux_chip->dev.parent = dev;
> +	mux_chip->dev.of_node = dev->of_node;
> +	dev_set_drvdata(&mux_chip->dev, mux_chip);
> +
> +	mux_chip->id = ida_simple_get(&mux_ida, 0, 0, GFP_KERNEL);
> +	if (mux_chip->id < 0) {
> +		pr_err("muxchipX failed to get a device id\n");
> +		kfree(mux_chip);
> +		return NULL;
> +	}
> +	dev_set_name(&mux_chip->dev, "muxchip%d", mux_chip->id);
> +
> +	mux_chip->controllers = controllers;
> +	for (i = 0; i < controllers; ++i) {
> +		struct mux_control *mux = &mux_chip->mux[i];
> +
> +		mux->chip = mux_chip;
> +		init_rwsem(&mux->lock);
> +		mux->cached_state = MUX_CACHE_UNKNOWN;
> +		mux->idle_state = MUX_IDLE_AS_IS;
> +	}
> +
> +	device_initialize(&mux_chip->dev);

Why are you not registering the device here as well?  Why have this be a
two step process?

> +
> +	return mux_chip;
> +}
> +EXPORT_SYMBOL_GPL(mux_chip_alloc);
> +
> +static int mux_control_set(struct mux_control *mux, int state)
> +{
> +	int ret = mux->chip->ops->set(mux, state);
> +
> +	mux->cached_state = ret < 0 ? MUX_CACHE_UNKNOWN : state;
> +
> +	return ret;
> +}
> +
> +int mux_chip_register(struct mux_chip *mux_chip)
> +{
> +	int i;
> +	int ret;
> +
> +	for (i = 0; i < mux_chip->controllers; ++i) {
> +		struct mux_control *mux = &mux_chip->mux[i];
> +
> +		if (mux->idle_state == mux->cached_state)
> +			continue;
> +
> +		ret = mux_control_set(mux, mux->idle_state);
> +		if (ret < 0) {
> +			dev_err(&mux_chip->dev, "unable to set idle state\n");
> +			return ret;
> +		}
> +	}
> +
> +	ret = device_add(&mux_chip->dev);
> +	if (ret < 0)
> +		dev_err(&mux_chip->dev,
> +			"device_add failed in mux_chip_register: %d\n", ret);

Did you run checkpatch.pl in strict mode on this new file?  Please do so :)

> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(mux_chip_register);
> +
> +void mux_chip_unregister(struct mux_chip *mux_chip)
> +{
> +	device_del(&mux_chip->dev);
> +}
> +EXPORT_SYMBOL_GPL(mux_chip_unregister);
> +
> +void mux_chip_free(struct mux_chip *mux_chip)
> +{
> +	if (!mux_chip)
> +		return;
> +
> +	put_device(&mux_chip->dev);
> +}
> +EXPORT_SYMBOL_GPL(mux_chip_free);
> +
> +static void devm_mux_chip_release(struct device *dev, void *res)
> +{
> +	struct mux_chip *mux_chip = *(struct mux_chip **)res;
> +
> +	mux_chip_free(mux_chip);
> +}
> +
> +struct mux_chip *devm_mux_chip_alloc(struct device *dev,
> +				     unsigned int controllers,
> +				     size_t sizeof_priv)
> +{
> +	struct mux_chip **ptr, *mux_chip;
> +
> +	ptr = devres_alloc(devm_mux_chip_release, sizeof(*ptr), GFP_KERNEL);
> +	if (!ptr)
> +		return NULL;
> +
> +	mux_chip = mux_chip_alloc(dev, controllers, sizeof_priv);
> +	if (!mux_chip) {
> +		devres_free(ptr);
> +		return NULL;
> +	}
> +
> +	*ptr = mux_chip;
> +	devres_add(dev, ptr);
> +
> +	return mux_chip;
> +}
> +EXPORT_SYMBOL_GPL(devm_mux_chip_alloc);


Having devm functions that create/destroy other struct devices worries
me, do we have other examples of this happening today?  Are you sure you
got the reference counting all correct?

> +
> +static int devm_mux_chip_match(struct device *dev, void *res, void *data)
> +{
> +	struct mux_chip **r = res;
> +
> +	if (WARN_ON(!r || !*r))

How can this happen?

> +		return 0;
> +
> +	return *r == data;
> +}
> +
> +void devm_mux_chip_free(struct device *dev, struct mux_chip *mux_chip)
> +{
> +	WARN_ON(devres_release(dev, devm_mux_chip_release,
> +			       devm_mux_chip_match, mux_chip));

What can someone do with these WARN_ON() splats in the kernel log?


> +}
> +EXPORT_SYMBOL_GPL(devm_mux_chip_free);
> +
> +static void devm_mux_chip_reg_release(struct device *dev, void *res)
> +{
> +	struct mux_chip *mux_chip = *(struct mux_chip **)res;
> +
> +	mux_chip_unregister(mux_chip);
> +}
> +
> +int devm_mux_chip_register(struct device *dev,
> +			   struct mux_chip *mux_chip)
> +{
> +	struct mux_chip **ptr;
> +	int res;
> +
> +	ptr = devres_alloc(devm_mux_chip_reg_release, sizeof(*ptr), GFP_KERNEL);
> +	if (!ptr)
> +		return -ENOMEM;
> +
> +	res = mux_chip_register(mux_chip);
> +	if (res) {
> +		devres_free(ptr);
> +		return res;
> +	}
> +
> +	*ptr = mux_chip;
> +	devres_add(dev, ptr);
> +
> +	return res;
> +}
> +EXPORT_SYMBOL_GPL(devm_mux_chip_register);
> +
> +void devm_mux_chip_unregister(struct device *dev, struct mux_chip *mux_chip)
> +{
> +	WARN_ON(devres_release(dev, devm_mux_chip_reg_release,
> +			       devm_mux_chip_match, mux_chip));
> +}
> +EXPORT_SYMBOL_GPL(devm_mux_chip_unregister);
> +
> +int mux_control_select(struct mux_control *mux, int state)
> +{
> +	int ret;
> +
> +	if (down_read_trylock(&mux->lock)) {
> +		if (mux->cached_state == state)
> +			return 0;
> +
> +		/* Sigh, the mux needs updating... */
> +		up_read(&mux->lock);
> +	}
> +
> +	/* ...or it's just contended. */
> +	down_write(&mux->lock);

Why use a read/write lock at all?  Have you tested this to verify it
really is faster and needed?

> +
> +	if (mux->cached_state == state) {
> +		/*
> +		 * Hmmm, someone else changed the mux to my liking.
> +		 * That makes me wonder how long I waited for nothing?
> +		 */
> +		downgrade_write(&mux->lock);

Oh that always scares me...  Are you _sure_ this is correct?  And
needed?

> +		return 0;
> +	}
> +
> +	ret = mux_control_set(mux, state);
> +	if (ret < 0) {
> +		if (mux->idle_state != MUX_IDLE_AS_IS)
> +			mux_control_set(mux, mux->idle_state);
> +
> +		up_write(&mux->lock);
> +		return ret;
> +	}
> +
> +	downgrade_write(&mux->lock);
> +
> +	return 1;
> +}
> +EXPORT_SYMBOL_GPL(mux_control_select);
> +
> +int mux_control_deselect(struct mux_control *mux)
> +{
> +	int ret = 0;
> +
> +	if (mux->idle_state != MUX_IDLE_AS_IS &&
> +	    mux->idle_state != mux->cached_state)
> +		ret = mux_control_set(mux, mux->idle_state);
> +
> +	up_read(&mux->lock);

You require a lock to be held for a "global" function?  Without
documentation?  Or even a sparse marking?  That's asking for trouble...

> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(mux_control_deselect);
> +
> +static int of_dev_node_match(struct device *dev, const void *data)
> +{
> +	return dev->of_node == data;
> +}
> +
> +static struct mux_chip *of_find_mux_chip_by_node(struct device_node *np)
> +{
> +	struct device *dev;
> +
> +	dev = class_find_device(&mux_class, NULL, np, of_dev_node_match);
> +
> +	return dev ? to_mux_chip(dev) : NULL;
> +}
> +
> +struct mux_control *mux_control_get(struct device *dev, const char *mux_name)
> +{
> +	struct device_node *np = dev->of_node;
> +	struct of_phandle_args args;
> +	struct mux_chip *mux_chip;
> +	unsigned int controller;
> +	int index = 0;
> +	int ret;
> +
> +	if (mux_name) {
> +		index = of_property_match_string(np, "mux-control-names",
> +						 mux_name);
> +		if (index < 0) {
> +			dev_err(dev, "mux controller '%s' not found\n",
> +				mux_name);
> +			return ERR_PTR(index);
> +		}
> +	}
> +
> +	ret = of_parse_phandle_with_args(np,
> +					 "mux-controls", "#mux-control-cells",
> +					 index, &args);
> +	if (ret) {
> +		dev_err(dev, "%s: failed to get mux-control %s(%i)\n",
> +			np->full_name, mux_name ?: "", index);
> +		return ERR_PTR(ret);
> +	}
> +
> +	mux_chip = of_find_mux_chip_by_node(args.np);
> +	of_node_put(args.np);
> +	if (!mux_chip)
> +		return ERR_PTR(-EPROBE_DEFER);
> +
> +	if (args.args_count > 1 ||
> +	    (!args.args_count && (mux_chip->controllers > 1))) {
> +		dev_err(dev, "%s: wrong #mux-control-cells for %s\n",
> +			np->full_name, args.np->full_name);
> +		return ERR_PTR(-EINVAL);
> +	}
> +
> +	controller = 0;
> +	if (args.args_count)
> +		controller = args.args[0];
> +
> +	if (controller >= mux_chip->controllers) {
> +		dev_err(dev, "%s: bad mux controller %u specified in %s\n",
> +			np->full_name, controller, args.np->full_name);
> +		return ERR_PTR(-EINVAL);
> +	}
> +
> +	get_device(&mux_chip->dev);
> +	return &mux_chip->mux[controller];
> +}
> +EXPORT_SYMBOL_GPL(mux_control_get);
> +
> +void mux_control_put(struct mux_control *mux)
> +{
> +	put_device(&mux->chip->dev);
> +}
> +EXPORT_SYMBOL_GPL(mux_control_put);
> +
> +static void devm_mux_control_release(struct device *dev, void *res)
> +{
> +	struct mux_control *mux = *(struct mux_control **)res;
> +
> +	mux_control_put(mux);
> +}
> +
> +struct mux_control *devm_mux_control_get(struct device *dev,
> +					 const char *mux_name)
> +{
> +	struct mux_control **ptr, *mux;
> +
> +	ptr = devres_alloc(devm_mux_control_release, sizeof(*ptr), GFP_KERNEL);
> +	if (!ptr)
> +		return ERR_PTR(-ENOMEM);
> +
> +	mux = mux_control_get(dev, mux_name);
> +	if (IS_ERR(mux)) {
> +		devres_free(ptr);
> +		return mux;
> +	}
> +
> +	*ptr = mux;
> +	devres_add(dev, ptr);
> +
> +	return mux;
> +}
> +EXPORT_SYMBOL_GPL(devm_mux_control_get);
> +
> +static int devm_mux_control_match(struct device *dev, void *res, void *data)
> +{
> +	struct mux_control **r = res;
> +
> +	if (WARN_ON(!r || !*r))
> +		return 0;

Same here, how can this happen?

> +
> +	return *r == data;
> +}
> +
> +void devm_mux_control_put(struct device *dev, struct mux_control *mux)
> +{
> +	WARN_ON(devres_release(dev, devm_mux_control_release,
> +			       devm_mux_control_match, mux));
> +}
> +EXPORT_SYMBOL_GPL(devm_mux_control_put);
> +
> +/*
> + * Using subsys_initcall instead of module_init here to ensure - for the
> + * non-modular case - that the subsystem is initialized when mux consumers
> + * and mux controllers start to use it /without/ relying on link order.
> + * For the modular case, the ordering is ensured with module dependencies.
> + */
> +subsys_initcall(mux_init);

Even with subsys_initcall you are relying on link order, you do realize
that?  What about other subsystems that rely on this?  :)


> +
> +MODULE_DESCRIPTION("Multiplexer subsystem");
> +MODULE_AUTHOR("Peter Rosin <peda@axentia.se>");
> +MODULE_LICENSE("GPL v2");
> diff --git a/drivers/mux/mux-gpio.c b/drivers/mux/mux-gpio.c
> new file mode 100644
> index 000000000000..227d3572e6db
> --- /dev/null
> +++ b/drivers/mux/mux-gpio.c
> @@ -0,0 +1,114 @@
> +/*
> + * GPIO-controlled multiplexer driver
> + *
> + * Copyright (C) 2017 Axentia Technologies AB
> + *
> + * Author: Peter Rosin <peda@axentia.se>
> + *
> + * 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/err.h>
> +#include <linux/gpio/consumer.h>
> +#include <linux/module.h>
> +#include <linux/mux.h>
> +#include <linux/of_platform.h>
> +#include <linux/platform_device.h>
> +#include <linux/property.h>
> +
> +struct mux_gpio {
> +	struct gpio_descs *gpios;
> +	int *val;
> +};
> +
> +static int mux_gpio_set(struct mux_control *mux, int state)
> +{
> +	struct mux_gpio *mux_gpio = mux_chip_priv(mux->chip);
> +	int i;
> +
> +	for (i = 0; i < mux_gpio->gpios->ndescs; i++)
> +		mux_gpio->val[i] = (state >> i) & 1;
> +
> +	gpiod_set_array_value_cansleep(mux_gpio->gpios->ndescs,
> +				       mux_gpio->gpios->desc,
> +				       mux_gpio->val);
> +
> +	return 0;
> +}
> +
> +static const struct mux_control_ops mux_gpio_ops = {
> +	.set = mux_gpio_set,
> +};
> +
> +static const struct of_device_id mux_gpio_dt_ids[] = {
> +	{ .compatible = "gpio-mux", },
> +	{ /* sentinel */ }
> +};
> +MODULE_DEVICE_TABLE(of, mux_gpio_dt_ids);
> +
> +static int mux_gpio_probe(struct platform_device *pdev)
> +{
> +	struct device *dev = &pdev->dev;
> +	struct mux_chip *mux_chip;
> +	struct mux_gpio *mux_gpio;
> +	int pins;
> +	s32 idle_state;
> +	int ret;
> +
> +	pins = gpiod_count(dev, "mux");
> +	if (pins < 0)
> +		return pins;
> +
> +	mux_chip = devm_mux_chip_alloc(dev, 1, sizeof(*mux_gpio) +
> +				       pins * sizeof(*mux_gpio->val));
> +	if (!mux_chip)
> +		return -ENOMEM;
> +
> +	mux_gpio = mux_chip_priv(mux_chip);
> +	mux_gpio->val = (int *)(mux_gpio + 1);
> +	mux_chip->ops = &mux_gpio_ops;
> +
> +	mux_gpio->gpios = devm_gpiod_get_array(dev, "mux", GPIOD_OUT_LOW);
> +	if (IS_ERR(mux_gpio->gpios)) {
> +		ret = PTR_ERR(mux_gpio->gpios);
> +		if (ret != -EPROBE_DEFER)
> +			dev_err(dev, "failed to get gpios\n");
> +		return ret;
> +	}
> +	WARN_ON(pins != mux_gpio->gpios->ndescs);
> +	mux_chip->mux->states = 1 << pins;
> +
> +	ret = device_property_read_u32(dev, "idle-state", (u32 *)&idle_state);
> +	if (ret >= 0 && idle_state != MUX_IDLE_AS_IS) {
> +		if (idle_state < 0 || idle_state >= mux_chip->mux->states) {
> +			dev_err(dev, "invalid idle-state %u\n", idle_state);
> +			return -EINVAL;
> +		}
> +
> +		mux_chip->mux->idle_state = idle_state;
> +	}
> +
> +	ret = devm_mux_chip_register(dev, mux_chip);
> +	if (ret < 0)
> +		return ret;
> +
> +	dev_info(dev, "%u-way mux-controller registered\n",
> +		 mux_chip->mux->states);
> +
> +	return 0;
> +}
> +
> +static struct platform_driver mux_gpio_driver = {
> +	.driver = {
> +		.name = "gpio-mux",
> +		.of_match_table	= of_match_ptr(mux_gpio_dt_ids),
> +	},
> +	.probe = mux_gpio_probe,
> +};
> +module_platform_driver(mux_gpio_driver);
> +
> +MODULE_DESCRIPTION("GPIO-controlled multiplexer driver");
> +MODULE_AUTHOR("Peter Rosin <peda@axentia.se>");
> +MODULE_LICENSE("GPL v2");
> diff --git a/include/linux/mux.h b/include/linux/mux.h
> new file mode 100644
> index 000000000000..febdde4246df
> --- /dev/null
> +++ b/include/linux/mux.h
> @@ -0,0 +1,252 @@
> +/*
> + * mux.h - definitions for the multiplexer interface
> + *
> + * Copyright (C) 2017 Axentia Technologies AB
> + *
> + * Author: Peter Rosin <peda@axentia.se>
> + *
> + * 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.
> + */
> +
> +#ifndef _LINUX_MUX_H
> +#define _LINUX_MUX_H
> +
> +#include <linux/device.h>
> +#include <linux/rwsem.h>
> +
> +struct mux_chip;
> +struct mux_control;
> +struct platform_device;
> +
> +/**
> + * struct mux_control_ops -	Mux controller operations for a mux chip.
> + * @set:			Set the state of the given mux controller.
> + */
> +struct mux_control_ops {
> +	int (*set)(struct mux_control *mux, int state);
> +};
> +
> +/* These defines match the constants from the dt-bindings. On purpose. */

Why on purpose?

> +#define MUX_IDLE_AS_IS      (-1)
> +#define MUX_IDLE_DISCONNECT (-2)
> +
> +/**
> + * struct mux_control -	Represents a mux controller.
> + * @lock:		Protects the mux controller state.
> + * @chip:		The mux chip that is handling this mux controller.
> + * @states:		The number of mux controller states.
> + * @cached_state:	The current mux controller state, or -1 if none.
> + * @idle_state:		The mux controller state to use when inactive, or one
> + *			of MUX_IDLE_AS_IS and MUX_IDLE_DISCONNECT.
> + */
> +struct mux_control {
> +	struct rw_semaphore lock; /* protects the state of the mux */
> +
> +	struct mux_chip *chip;
> +
> +	unsigned int states;
> +	int cached_state;
> +	int idle_state;
> +};
> +
> +/**
> + * struct mux_chip -	Represents a chip holding mux controllers.
> + * @controllers:	Number of mux controllers handled by the chip.
> + * @mux:		Array of mux controllers that are handled.
> + * @dev:		Device structure.
> + * @id:			Used to identify the device internally.
> + * @ops:		Mux controller operations.
> + */
> +struct mux_chip {
> +	unsigned int controllers;
> +	struct mux_control *mux;
> +	struct device dev;
> +	int id;
> +
> +	const struct mux_control_ops *ops;
> +};
> +
> +#define to_mux_chip(x) container_of((x), struct mux_chip, dev)
> +
> +/**
> + * mux_chip_priv() - Get the extra memory reserved by mux_chip_alloc().
> + * @mux_chip: The mux-chip to get the private memory from.
> + *
> + * Return: Pointer to the private memory reserved by the allocator.
> + */
> +static inline void *mux_chip_priv(struct mux_chip *mux_chip)
> +{
> +	return &mux_chip->mux[mux_chip->controllers];
> +}
> +
> +/**
> + * mux_chip_alloc() - Allocate a mux-chip.
> + * @dev: The parent device implementing the mux interface.
> + * @controllers: The number of mux controllers to allocate for this chip.
> + * @sizeof_priv: Size of extra memory area for private use by the caller.
> + *
> + * Return: A pointer to the new mux-chip, NULL on failure.
> + */
> +struct mux_chip *mux_chip_alloc(struct device *dev,
> +				unsigned int controllers, size_t sizeof_priv);
> +

Don't put kernel doc comments in a .h file, they normally go into the .c
file, next to the code itself.  That makes it easier to fix up and
realise when they need to be changed when the code changes.  The .h file
rarely changes.


> +/**
> + * mux_chip_register() - Register a mux-chip, thus readying the controllers
> + *			 for use.
> + * @mux_chip: The mux-chip to register.
> + *
> + * Do not retry registration of the same mux-chip on failure. You should
> + * instead put it away with mux_chip_free() and allocate a new one, if you
> + * for some reason would like to retry registration.
> + *
> + * Return: Zero on success or a negative errno on error.
> + */
> +int mux_chip_register(struct mux_chip *mux_chip);
> +
> +/**
> + * mux_chip_unregister() - Take the mux-chip off-line.
> + * @mux_chip: The mux-chip to unregister.
> + *
> + * mux_chip_unregister() reverses the effects of mux_chip_register().
> + * But not completely, you should not try to call mux_chip_register()
> + * on a mux-chip that has been registered before.
> + */
> +void mux_chip_unregister(struct mux_chip *mux_chip);
> +
> +/**
> + * mux_chip_free() - Free the mux-chip for good.
> + * @mux_chip: The mux-chip to free.
> + *
> + * mux_chip_free() reverses the effects of mux_chip_alloc().
> + */
> +void mux_chip_free(struct mux_chip *mux_chip);
> +
> +/**
> + * devm_mux_chip_alloc() - Resource-managed version of mux_chip_alloc().
> + * @dev: The parent device implementing the mux interface.
> + * @controllers: The number of mux controllers to allocate for this chip.
> + * @sizeof_priv: Size of extra memory area for private use by the caller.
> + *
> + * See mux_chip_alloc() for more details.
> + *
> + * Return: A pointer to the new mux-chip, NULL on failure.
> + */
> +struct mux_chip *devm_mux_chip_alloc(struct device *dev,
> +				     unsigned int controllers,
> +				     size_t sizeof_priv);
> +
> +/**
> + * devm_mux_chip_register() - Resource-managed version mux_chip_register().
> + * @dev: The parent device implementing the mux interface.
> + * @mux_chip: The mux-chip to register.
> + *
> + * See mux_chip_register() for more details.
> + *
> + * Return: Zero on success or a negative errno on error.
> + */
> +int devm_mux_chip_register(struct device *dev, struct mux_chip *mux_chip);
> +
> +/**
> + * devm_mux_chip_unregister() - Resource-managed version mux_chip_unregister().
> + * @dev: The device that originally registered the mux-chip.
> + * @mux_chip: The mux-chip to unregister.
> + *
> + * See mux_chip_unregister() for more details.
> + *
> + * Note that you do not normally need to call this function.

Odd, then why is it exported???


> + */
> +void devm_mux_chip_unregister(struct device *dev, struct mux_chip *mux_chip);
> +
> +/**
> + * devm_mux_chip_free() - Resource-managed version mux_chip_free().
> + * @dev: The device that originally got the mux-chip.
> + * @mux_chip: The mux-chip to free.
> + *
> + * See mux_chip_free() for more details.
> + *
> + * Note that you do not normally need to call this function.
> + */
> +void devm_mux_chip_free(struct device *dev, struct mux_chip *mux_chip);
> +
> +/**
> + * mux_control_select() - Select the given multiplexer state.
> + * @mux: The mux-control to request a change of state from.
> + * @state: The new requested state.
> + *
> + * Make sure to call mux_control_deselect() when the operation is complete and
> + * the mux-control is free for others to use, but do not call
> + * mux_control_deselect() if mux_control_select() fails.
> + *
> + * Return: 0 if the requested state was already active, or 1 it the
> + * mux-control state was changed to the requested state. Or a negative
> + * errno on error.
> + *
> + * Note that the difference in return value of zero or one is of
> + * questionable value; especially if the mux-control has several independent
> + * consumers, which is something the consumers should perhaps not be making
> + * assumptions about.

I don't understand this note, what is a user of this api supposed to do
differently between 1 and 0?  Why make the difference at all?

And I agree with the comment to split this up into 2 different .h files,
if possible.

thanks,

greg k-h
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Peter Rosin April 18, 2017, 10:59 a.m.
On 2017-04-18 10:51, Greg Kroah-Hartman wrote:
> On Thu, Apr 13, 2017 at 06:43:07PM +0200, Peter Rosin wrote:
>> +config MUX_GPIO
>> +	tristate "GPIO-controlled Multiplexer"
>> +	depends on OF && GPIOLIB
> 
> Why have the gpio and mux core in the same patch?

One is not usable w/o the other. I can split them if you want to?

> And why does this depend on OF?

That's historical, I was originally using of_property_read_u32.
I'll remove the dep...

>> +	help
>> +	  GPIO-controlled Multiplexer controller.
>> +
>> +	  The driver builds a single multiplexer controller using a number
>> +	  of gpio pins. For N pins, there will be 2^N possible multiplexer
>> +	  states. The GPIO pins can be connected (by the hardware) to several
>> +	  multiplexers, which in that case will be operated in parallel.
>> +
>> +	  To compile the driver as a module, choose M here: the module will
>> +	  be called mux-gpio.
>> +
>> +endif
>> diff --git a/drivers/mux/Makefile b/drivers/mux/Makefile
>> new file mode 100644
>> index 000000000000..bb16953f6290
>> --- /dev/null
>> +++ b/drivers/mux/Makefile
>> @@ -0,0 +1,6 @@
>> +#
>> +# Makefile for multiplexer devices.
>> +#
>> +
>> +obj-$(CONFIG_MULTIPLEXER)	+= mux-core.o
>> +obj-$(CONFIG_MUX_GPIO)		+= mux-gpio.o
>> diff --git a/drivers/mux/mux-core.c b/drivers/mux/mux-core.c
>> new file mode 100644
>> index 000000000000..66a8bccfc3d7
>> --- /dev/null
>> +++ b/drivers/mux/mux-core.c
>> @@ -0,0 +1,422 @@
>> +/*
>> + * Multiplexer subsystem
>> + *
>> + * Copyright (C) 2017 Axentia Technologies AB
>> + *
>> + * Author: Peter Rosin <peda@axentia.se>
>> + *
>> + * 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.
>> + */
>> +
>> +#define pr_fmt(fmt) "mux-core: " fmt
>> +
>> +#include <linux/device.h>
>> +#include <linux/err.h>
>> +#include <linux/export.h>
>> +#include <linux/idr.h>
>> +#include <linux/init.h>
>> +#include <linux/module.h>
>> +#include <linux/mux.h>
>> +#include <linux/of.h>
>> +#include <linux/of_platform.h>
>> +#include <linux/slab.h>
>> +
>> +/*
>> + * The idle-as-is "state" is not an actual state that may be selected, it
>> + * only implies that the state should not be changed. So, use that state
>> + * as indication that the cached state of the multiplexer is unknown.
>> + */
>> +#define MUX_CACHE_UNKNOWN MUX_IDLE_AS_IS
>> +
>> +static struct class mux_class = {
>> +	.name = "mux",
>> +	.owner = THIS_MODULE,
>> +};
> 
> No Documentation/ABI/ update for your sysfs files?  Please do so.

Ok I'll look into it. Wasn't even aware that I added any. But there's the
new class of course...

>> +
>> +static int __init mux_init(void)
>> +{
>> +	return class_register(&mux_class);
>> +}
>> +
>> +static DEFINE_IDA(mux_ida);
> 
> When your module is unloaded, you forgot to clean this structure up with
> what was done with it.

I was under the impression that not providing an exit function for modules
made the module infrastructure prevent unloading (by bumping some reference
counter). Maybe that is a misconception?

>> +
>> +static void mux_chip_release(struct device *dev)
>> +{
>> +	struct mux_chip *mux_chip = to_mux_chip(dev);
>> +
>> +	ida_simple_remove(&mux_ida, mux_chip->id);
>> +	kfree(mux_chip);
>> +}
>> +
>> +static struct device_type mux_type = {
>> +	.name = "mux-chip",
>> +	.release = mux_chip_release,
>> +};
>> +
>> +struct mux_chip *mux_chip_alloc(struct device *dev,
>> +				unsigned int controllers, size_t sizeof_priv)
>> +{
>> +	struct mux_chip *mux_chip;
>> +	int i;
>> +
>> +	if (WARN_ON(!dev || !controllers))
>> +		return NULL;
>> +
>> +	mux_chip = kzalloc(sizeof(*mux_chip) +
>> +			   controllers * sizeof(*mux_chip->mux) +
>> +			   sizeof_priv, GFP_KERNEL);
>> +	if (!mux_chip)
>> +		return NULL;
> 
> You don't return PTR_ERR(-ENOMEM)?  Ok, why not?  (I'm not arguing for
> it, just curious...)

There's no particular reason. Do you think I should change it?

>> +
>> +	mux_chip->mux = (struct mux_control *)(mux_chip + 1);
>> +	mux_chip->dev.class = &mux_class;
>> +	mux_chip->dev.type = &mux_type;
>> +	mux_chip->dev.parent = dev;
>> +	mux_chip->dev.of_node = dev->of_node;
>> +	dev_set_drvdata(&mux_chip->dev, mux_chip);
>> +
>> +	mux_chip->id = ida_simple_get(&mux_ida, 0, 0, GFP_KERNEL);
>> +	if (mux_chip->id < 0) {
>> +		pr_err("muxchipX failed to get a device id\n");
>> +		kfree(mux_chip);
>> +		return NULL;
>> +	}
>> +	dev_set_name(&mux_chip->dev, "muxchip%d", mux_chip->id);
>> +
>> +	mux_chip->controllers = controllers;
>> +	for (i = 0; i < controllers; ++i) {
>> +		struct mux_control *mux = &mux_chip->mux[i];
>> +
>> +		mux->chip = mux_chip;
>> +		init_rwsem(&mux->lock);
>> +		mux->cached_state = MUX_CACHE_UNKNOWN;
>> +		mux->idle_state = MUX_IDLE_AS_IS;
>> +	}
>> +
>> +	device_initialize(&mux_chip->dev);
> 
> Why are you not registering the device here as well?  Why have this be a
> two step process?

Because of idle state handling. The drivers are expected to fill in
the desired idle state(s) after allocating the mux controller(s).
Then, when registering, the desired idle state is activated (if the
idle state is not idle-as-is, of course) and as a last step the mux
is "advertised".

>> +
>> +	return mux_chip;
>> +}
>> +EXPORT_SYMBOL_GPL(mux_chip_alloc);
>> +
>> +static int mux_control_set(struct mux_control *mux, int state)
>> +{
>> +	int ret = mux->chip->ops->set(mux, state);
>> +
>> +	mux->cached_state = ret < 0 ? MUX_CACHE_UNKNOWN : state;
>> +
>> +	return ret;
>> +}
>> +
>> +int mux_chip_register(struct mux_chip *mux_chip)
>> +{
>> +	int i;
>> +	int ret;
>> +
>> +	for (i = 0; i < mux_chip->controllers; ++i) {
>> +		struct mux_control *mux = &mux_chip->mux[i];
>> +
>> +		if (mux->idle_state == mux->cached_state)
>> +			continue;
>> +
>> +		ret = mux_control_set(mux, mux->idle_state);
>> +		if (ret < 0) {
>> +			dev_err(&mux_chip->dev, "unable to set idle state\n");
>> +			return ret;
>> +		}
>> +	}
>> +
>> +	ret = device_add(&mux_chip->dev);
>> +	if (ret < 0)
>> +		dev_err(&mux_chip->dev,
>> +			"device_add failed in mux_chip_register: %d\n", ret);
> 
> Did you run checkpatch.pl in strict mode on this new file?  Please do so :)

I did, and did it again just to be sure, and I do not get any complaints.
So, what's wrong?

$ scripts/checkpatch.pl --strict mux-13/0003-mux-minimal-mux-subsystem-and-gpio-based-mux-control.patch
total: 0 errors, 0 warnings, 0 checks, 860 lines checked

mux-13/0003-mux-minimal-mux-subsystem-and-gpio-based-mux-control.patch has no obvious style problems and is ready for submission.

The chackpatch warnings I am aware of are three instances of (for the
'prop', 's' and 'i' arguments):

mux-13/0006-iio-multiplexer-new-iio-category-and-iio-mux-driver.patch
---------------------------------------------------------------------
CHECK: Macro argument reuse 'prop' - possible side-effects?
#433: FILE: drivers/iio/multiplexer/iio-mux.c:326:
+#define of_property_for_each_string_index(np, propname, prop, s, i)    \
+       for (prop = of_find_property(np, propname, NULL),               \
+            s = of_prop_next_string(prop, NULL),                       \
+            i = 0;                                                     \
+            s;                                                         \
+            s = of_prop_next_string(prop, s),                          \
+            i++)

But those kinds of warnings are also present in the code I plagiarized,
so I don't feel too bad...

And then there are a couple of false positives about files added w/o
adding an entry to MAINTAINERS (the files are covers by wildcards).

>> +	return ret;
>> +}
>> +EXPORT_SYMBOL_GPL(mux_chip_register);
>> +
>> +void mux_chip_unregister(struct mux_chip *mux_chip)
>> +{
>> +	device_del(&mux_chip->dev);
>> +}
>> +EXPORT_SYMBOL_GPL(mux_chip_unregister);
>> +
>> +void mux_chip_free(struct mux_chip *mux_chip)
>> +{
>> +	if (!mux_chip)
>> +		return;
>> +
>> +	put_device(&mux_chip->dev);
>> +}
>> +EXPORT_SYMBOL_GPL(mux_chip_free);
>> +
>> +static void devm_mux_chip_release(struct device *dev, void *res)
>> +{
>> +	struct mux_chip *mux_chip = *(struct mux_chip **)res;
>> +
>> +	mux_chip_free(mux_chip);
>> +}
>> +
>> +struct mux_chip *devm_mux_chip_alloc(struct device *dev,
>> +				     unsigned int controllers,
>> +				     size_t sizeof_priv)
>> +{
>> +	struct mux_chip **ptr, *mux_chip;
>> +
>> +	ptr = devres_alloc(devm_mux_chip_release, sizeof(*ptr), GFP_KERNEL);
>> +	if (!ptr)
>> +		return NULL;
>> +
>> +	mux_chip = mux_chip_alloc(dev, controllers, sizeof_priv);
>> +	if (!mux_chip) {
>> +		devres_free(ptr);
>> +		return NULL;
>> +	}
>> +
>> +	*ptr = mux_chip;
>> +	devres_add(dev, ptr);
>> +
>> +	return mux_chip;
>> +}
>> +EXPORT_SYMBOL_GPL(devm_mux_chip_alloc);
> 
> 
> Having devm functions that create/destroy other struct devices worries
> me, do we have other examples of this happening today?  Are you sure you
> got the reference counting all correct?

drivers/iio/industrialio_core.c:devm_iio_device_alloc

Or is the iio case different in some subtle way that I'm missing?

>> +
>> +static int devm_mux_chip_match(struct device *dev, void *res, void *data)
>> +{
>> +	struct mux_chip **r = res;
>> +
>> +	if (WARN_ON(!r || !*r))
> 
> How can this happen?

It shouldn't. I copied the pattern from the iio subsystem.

>> +		return 0;
>> +
>> +	return *r == data;
>> +}
>> +
>> +void devm_mux_chip_free(struct device *dev, struct mux_chip *mux_chip)
>> +{
>> +	WARN_ON(devres_release(dev, devm_mux_chip_release,
>> +			       devm_mux_chip_match, mux_chip));
> 
> What can someone do with these WARN_ON() splats in the kernel log?

Don't know. Again, I copied the pattern from the iio subsystem.

> 
>> +}
>> +EXPORT_SYMBOL_GPL(devm_mux_chip_free);
>> +
>> +static void devm_mux_chip_reg_release(struct device *dev, void *res)
>> +{
>> +	struct mux_chip *mux_chip = *(struct mux_chip **)res;
>> +
>> +	mux_chip_unregister(mux_chip);
>> +}
>> +
>> +int devm_mux_chip_register(struct device *dev,
>> +			   struct mux_chip *mux_chip)
>> +{
>> +	struct mux_chip **ptr;
>> +	int res;
>> +
>> +	ptr = devres_alloc(devm_mux_chip_reg_release, sizeof(*ptr), GFP_KERNEL);
>> +	if (!ptr)
>> +		return -ENOMEM;
>> +
>> +	res = mux_chip_register(mux_chip);
>> +	if (res) {
>> +		devres_free(ptr);
>> +		return res;
>> +	}
>> +
>> +	*ptr = mux_chip;
>> +	devres_add(dev, ptr);
>> +
>> +	return res;
>> +}
>> +EXPORT_SYMBOL_GPL(devm_mux_chip_register);
>> +
>> +void devm_mux_chip_unregister(struct device *dev, struct mux_chip *mux_chip)
>> +{
>> +	WARN_ON(devres_release(dev, devm_mux_chip_reg_release,
>> +			       devm_mux_chip_match, mux_chip));
>> +}
>> +EXPORT_SYMBOL_GPL(devm_mux_chip_unregister);
>> +
>> +int mux_control_select(struct mux_control *mux, int state)
>> +{
>> +	int ret;
>> +
>> +	if (down_read_trylock(&mux->lock)) {
>> +		if (mux->cached_state == state)
>> +			return 0;
>> +
>> +		/* Sigh, the mux needs updating... */
>> +		up_read(&mux->lock);
>> +	}
>> +
>> +	/* ...or it's just contended. */
>> +	down_write(&mux->lock);
> 
> Why use a read/write lock at all?  Have you tested this to verify it
> really is faster and needed?

For one of the HW configuration that drove the development, the same mux
controller is used to mux both an I2C channel and a couple of ADC lines.

If there is no kind of reader/writer locking going on, there is no way to
do ADC readings concurrently with an I2C transfer even when the consumers
want the mux in the same position. With an ordinary mutex controlling the
mux position, the consumers will unconditionally get serialized, which
seems like a waste to me. Or maybe I'm missing something?

>> +
>> +	if (mux->cached_state == state) {
>> +		/*
>> +		 * Hmmm, someone else changed the mux to my liking.
>> +		 * That makes me wonder how long I waited for nothing?
>> +		 */
>> +		downgrade_write(&mux->lock);
> 
> Oh that always scares me...  Are you _sure_ this is correct?  And
> needed?

It might not be needed, and it would probably work ok to just fall
through and call mux_control_set unconditionally. What is it that
always scares you exactly? Relying on cached state to be correct?
Downgrading writer locks?

>> +		return 0;
>> +	}
>> +
>> +	ret = mux_control_set(mux, state);
>> +	if (ret < 0) {
>> +		if (mux->idle_state != MUX_IDLE_AS_IS)
>> +			mux_control_set(mux, mux->idle_state);
>> +
>> +		up_write(&mux->lock);
>> +		return ret;
>> +	}
>> +
>> +	downgrade_write(&mux->lock);
>> +
>> +	return 1;
>> +}
>> +EXPORT_SYMBOL_GPL(mux_control_select);
>> +
>> +int mux_control_deselect(struct mux_control *mux)
>> +{
>> +	int ret = 0;
>> +
>> +	if (mux->idle_state != MUX_IDLE_AS_IS &&
>> +	    mux->idle_state != mux->cached_state)
>> +		ret = mux_control_set(mux, mux->idle_state);
>> +
>> +	up_read(&mux->lock);
> 
> You require a lock to be held for a "global" function?  Without
> documentation?  Or even a sparse marking?  That's asking for trouble...

Documentation I can handle, but where should I look to understand how I
should add sparse markings?

The mux needs to be locked somehow. But as I stated in the cover letter
the rwsem isn't a perfect fit.

	I'm using an rwsem to lock a mux, but that isn't really a
	perfect fit. Is there a better locking primitive that I don't
	know about that fits better? I had a mutex at one point, but
	that didn't allow any concurrent accesses at all. At least
	the rwsem allows concurrent access as long as all users
	agree on the mux state, but I suspect that the rwsem will
	degrade to the mutex situation pretty quickly if there is
	any contention.

Also, the lock doesn't add anything if there is only one consumer of
a mux controller. Maybe there should be some mechanism for shortcutting
the locking for the (more common?) single-consumer case?

But again, I need the locking for my multi-consumer use case.

>> +
>> +	return ret;
>> +}
>> +EXPORT_SYMBOL_GPL(mux_control_deselect);
>> +
>> +static int of_dev_node_match(struct device *dev, const void *data)
>> +{
>> +	return dev->of_node == data;
>> +}
>> +
>> +static struct mux_chip *of_find_mux_chip_by_node(struct device_node *np)
>> +{
>> +	struct device *dev;
>> +
>> +	dev = class_find_device(&mux_class, NULL, np, of_dev_node_match);
>> +
>> +	return dev ? to_mux_chip(dev) : NULL;
>> +}
>> +
>> +struct mux_control *mux_control_get(struct device *dev, const char *mux_name)
>> +{
>> +	struct device_node *np = dev->of_node;
>> +	struct of_phandle_args args;
>> +	struct mux_chip *mux_chip;
>> +	unsigned int controller;
>> +	int index = 0;
>> +	int ret;
>> +
>> +	if (mux_name) {
>> +		index = of_property_match_string(np, "mux-control-names",
>> +						 mux_name);
>> +		if (index < 0) {
>> +			dev_err(dev, "mux controller '%s' not found\n",
>> +				mux_name);
>> +			return ERR_PTR(index);
>> +		}
>> +	}
>> +
>> +	ret = of_parse_phandle_with_args(np,
>> +					 "mux-controls", "#mux-control-cells",
>> +					 index, &args);
>> +	if (ret) {
>> +		dev_err(dev, "%s: failed to get mux-control %s(%i)\n",
>> +			np->full_name, mux_name ?: "", index);
>> +		return ERR_PTR(ret);
>> +	}
>> +
>> +	mux_chip = of_find_mux_chip_by_node(args.np);
>> +	of_node_put(args.np);
>> +	if (!mux_chip)
>> +		return ERR_PTR(-EPROBE_DEFER);
>> +
>> +	if (args.args_count > 1 ||
>> +	    (!args.args_count && (mux_chip->controllers > 1))) {
>> +		dev_err(dev, "%s: wrong #mux-control-cells for %s\n",
>> +			np->full_name, args.np->full_name);
>> +		return ERR_PTR(-EINVAL);
>> +	}
>> +
>> +	controller = 0;
>> +	if (args.args_count)
>> +		controller = args.args[0];
>> +
>> +	if (controller >= mux_chip->controllers) {
>> +		dev_err(dev, "%s: bad mux controller %u specified in %s\n",
>> +			np->full_name, controller, args.np->full_name);
>> +		return ERR_PTR(-EINVAL);
>> +	}
>> +
>> +	get_device(&mux_chip->dev);
>> +	return &mux_chip->mux[controller];
>> +}
>> +EXPORT_SYMBOL_GPL(mux_control_get);
>> +
>> +void mux_control_put(struct mux_control *mux)
>> +{
>> +	put_device(&mux->chip->dev);
>> +}
>> +EXPORT_SYMBOL_GPL(mux_control_put);
>> +
>> +static void devm_mux_control_release(struct device *dev, void *res)
>> +{
>> +	struct mux_control *mux = *(struct mux_control **)res;
>> +
>> +	mux_control_put(mux);
>> +}
>> +
>> +struct mux_control *devm_mux_control_get(struct device *dev,
>> +					 const char *mux_name)
>> +{
>> +	struct mux_control **ptr, *mux;
>> +
>> +	ptr = devres_alloc(devm_mux_control_release, sizeof(*ptr), GFP_KERNEL);
>> +	if (!ptr)
>> +		return ERR_PTR(-ENOMEM);
>> +
>> +	mux = mux_control_get(dev, mux_name);
>> +	if (IS_ERR(mux)) {
>> +		devres_free(ptr);
>> +		return mux;
>> +	}
>> +
>> +	*ptr = mux;
>> +	devres_add(dev, ptr);
>> +
>> +	return mux;
>> +}
>> +EXPORT_SYMBOL_GPL(devm_mux_control_get);
>> +
>> +static int devm_mux_control_match(struct device *dev, void *res, void *data)
>> +{
>> +	struct mux_control **r = res;
>> +
>> +	if (WARN_ON(!r || !*r))
>> +		return 0;
> 
> Same here, how can this happen?

Same response as above.

>> +
>> +	return *r == data;
>> +}
>> +
>> +void devm_mux_control_put(struct device *dev, struct mux_control *mux)
>> +{
>> +	WARN_ON(devres_release(dev, devm_mux_control_release,
>> +			       devm_mux_control_match, mux));
>> +}
>> +EXPORT_SYMBOL_GPL(devm_mux_control_put);
>> +
>> +/*
>> + * Using subsys_initcall instead of module_init here to ensure - for the
>> + * non-modular case - that the subsystem is initialized when mux consumers
>> + * and mux controllers start to use it /without/ relying on link order.
>> + * For the modular case, the ordering is ensured with module dependencies.
>> + */
>> +subsys_initcall(mux_init);
> 
> Even with subsys_initcall you are relying on link order, you do realize
> that?  What about other subsystems that rely on this?  :)

Yes, that is true, but if others start relying on this, that's their problem,
right? :-)

> 
>> +
>> +MODULE_DESCRIPTION("Multiplexer subsystem");
>> +MODULE_AUTHOR("Peter Rosin <peda@axentia.se>");
>> +MODULE_LICENSE("GPL v2");
>> diff --git a/drivers/mux/mux-gpio.c b/drivers/mux/mux-gpio.c
>> new file mode 100644
>> index 000000000000..227d3572e6db
>> --- /dev/null
>> +++ b/drivers/mux/mux-gpio.c
>> @@ -0,0 +1,114 @@
>> +/*
>> + * GPIO-controlled multiplexer driver
>> + *
>> + * Copyright (C) 2017 Axentia Technologies AB
>> + *
>> + * Author: Peter Rosin <peda@axentia.se>
>> + *
>> + * 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/err.h>
>> +#include <linux/gpio/consumer.h>
>> +#include <linux/module.h>
>> +#include <linux/mux.h>
>> +#include <linux/of_platform.h>
>> +#include <linux/platform_device.h>
>> +#include <linux/property.h>
>> +
>> +struct mux_gpio {
>> +	struct gpio_descs *gpios;
>> +	int *val;
>> +};
>> +
>> +static int mux_gpio_set(struct mux_control *mux, int state)
>> +{
>> +	struct mux_gpio *mux_gpio = mux_chip_priv(mux->chip);
>> +	int i;
>> +
>> +	for (i = 0; i < mux_gpio->gpios->ndescs; i++)
>> +		mux_gpio->val[i] = (state >> i) & 1;
>> +
>> +	gpiod_set_array_value_cansleep(mux_gpio->gpios->ndescs,
>> +				       mux_gpio->gpios->desc,
>> +				       mux_gpio->val);
>> +
>> +	return 0;
>> +}
>> +
>> +static const struct mux_control_ops mux_gpio_ops = {
>> +	.set = mux_gpio_set,
>> +};
>> +
>> +static const struct of_device_id mux_gpio_dt_ids[] = {
>> +	{ .compatible = "gpio-mux", },
>> +	{ /* sentinel */ }
>> +};
>> +MODULE_DEVICE_TABLE(of, mux_gpio_dt_ids);
>> +
>> +static int mux_gpio_probe(struct platform_device *pdev)
>> +{
>> +	struct device *dev = &pdev->dev;
>> +	struct mux_chip *mux_chip;
>> +	struct mux_gpio *mux_gpio;
>> +	int pins;
>> +	s32 idle_state;
>> +	int ret;
>> +
>> +	pins = gpiod_count(dev, "mux");
>> +	if (pins < 0)
>> +		return pins;
>> +
>> +	mux_chip = devm_mux_chip_alloc(dev, 1, sizeof(*mux_gpio) +
>> +				       pins * sizeof(*mux_gpio->val));
>> +	if (!mux_chip)
>> +		return -ENOMEM;
>> +
>> +	mux_gpio = mux_chip_priv(mux_chip);
>> +	mux_gpio->val = (int *)(mux_gpio + 1);
>> +	mux_chip->ops = &mux_gpio_ops;
>> +
>> +	mux_gpio->gpios = devm_gpiod_get_array(dev, "mux", GPIOD_OUT_LOW);
>> +	if (IS_ERR(mux_gpio->gpios)) {
>> +		ret = PTR_ERR(mux_gpio->gpios);
>> +		if (ret != -EPROBE_DEFER)
>> +			dev_err(dev, "failed to get gpios\n");
>> +		return ret;
>> +	}
>> +	WARN_ON(pins != mux_gpio->gpios->ndescs);
>> +	mux_chip->mux->states = 1 << pins;
>> +
>> +	ret = device_property_read_u32(dev, "idle-state", (u32 *)&idle_state);
>> +	if (ret >= 0 && idle_state != MUX_IDLE_AS_IS) {
>> +		if (idle_state < 0 || idle_state >= mux_chip->mux->states) {
>> +			dev_err(dev, "invalid idle-state %u\n", idle_state);
>> +			return -EINVAL;
>> +		}
>> +
>> +		mux_chip->mux->idle_state = idle_state;
>> +	}
>> +
>> +	ret = devm_mux_chip_register(dev, mux_chip);
>> +	if (ret < 0)
>> +		return ret;
>> +
>> +	dev_info(dev, "%u-way mux-controller registered\n",
>> +		 mux_chip->mux->states);
>> +
>> +	return 0;
>> +}
>> +
>> +static struct platform_driver mux_gpio_driver = {
>> +	.driver = {
>> +		.name = "gpio-mux",
>> +		.of_match_table	= of_match_ptr(mux_gpio_dt_ids),
>> +	},
>> +	.probe = mux_gpio_probe,
>> +};
>> +module_platform_driver(mux_gpio_driver);
>> +
>> +MODULE_DESCRIPTION("GPIO-controlled multiplexer driver");
>> +MODULE_AUTHOR("Peter Rosin <peda@axentia.se>");
>> +MODULE_LICENSE("GPL v2");
>> diff --git a/include/linux/mux.h b/include/linux/mux.h
>> new file mode 100644
>> index 000000000000..febdde4246df
>> --- /dev/null
>> +++ b/include/linux/mux.h
>> @@ -0,0 +1,252 @@
>> +/*
>> + * mux.h - definitions for the multiplexer interface
>> + *
>> + * Copyright (C) 2017 Axentia Technologies AB
>> + *
>> + * Author: Peter Rosin <peda@axentia.se>
>> + *
>> + * 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.
>> + */
>> +
>> +#ifndef _LINUX_MUX_H
>> +#define _LINUX_MUX_H
>> +
>> +#include <linux/device.h>
>> +#include <linux/rwsem.h>
>> +
>> +struct mux_chip;
>> +struct mux_control;
>> +struct platform_device;
>> +
>> +/**
>> + * struct mux_control_ops -	Mux controller operations for a mux chip.
>> + * @set:			Set the state of the given mux controller.
>> + */
>> +struct mux_control_ops {
>> +	int (*set)(struct mux_control *mux, int state);
>> +};
>> +
>> +/* These defines match the constants from the dt-bindings. On purpose. */
> 
> Why on purpose?

I sure wasn't an accident? :-)

Want me to remove it?

>> +#define MUX_IDLE_AS_IS      (-1)
>> +#define MUX_IDLE_DISCONNECT (-2)
>> +
>> +/**
>> + * struct mux_control -	Represents a mux controller.
>> + * @lock:		Protects the mux controller state.
>> + * @chip:		The mux chip that is handling this mux controller.
>> + * @states:		The number of mux controller states.
>> + * @cached_state:	The current mux controller state, or -1 if none.
>> + * @idle_state:		The mux controller state to use when inactive, or one
>> + *			of MUX_IDLE_AS_IS and MUX_IDLE_DISCONNECT.
>> + */
>> +struct mux_control {
>> +	struct rw_semaphore lock; /* protects the state of the mux */
>> +
>> +	struct mux_chip *chip;
>> +
>> +	unsigned int states;
>> +	int cached_state;
>> +	int idle_state;
>> +};
>> +
>> +/**
>> + * struct mux_chip -	Represents a chip holding mux controllers.
>> + * @controllers:	Number of mux controllers handled by the chip.
>> + * @mux:		Array of mux controllers that are handled.
>> + * @dev:		Device structure.
>> + * @id:			Used to identify the device internally.
>> + * @ops:		Mux controller operations.
>> + */
>> +struct mux_chip {
>> +	unsigned int controllers;
>> +	struct mux_control *mux;
>> +	struct device dev;
>> +	int id;
>> +
>> +	const struct mux_control_ops *ops;
>> +};
>> +
>> +#define to_mux_chip(x) container_of((x), struct mux_chip, dev)
>> +
>> +/**
>> + * mux_chip_priv() - Get the extra memory reserved by mux_chip_alloc().
>> + * @mux_chip: The mux-chip to get the private memory from.
>> + *
>> + * Return: Pointer to the private memory reserved by the allocator.
>> + */
>> +static inline void *mux_chip_priv(struct mux_chip *mux_chip)
>> +{
>> +	return &mux_chip->mux[mux_chip->controllers];
>> +}
>> +
>> +/**
>> + * mux_chip_alloc() - Allocate a mux-chip.
>> + * @dev: The parent device implementing the mux interface.
>> + * @controllers: The number of mux controllers to allocate for this chip.
>> + * @sizeof_priv: Size of extra memory area for private use by the caller.
>> + *
>> + * Return: A pointer to the new mux-chip, NULL on failure.
>> + */
>> +struct mux_chip *mux_chip_alloc(struct device *dev,
>> +				unsigned int controllers, size_t sizeof_priv);
>> +
> 
> Don't put kernel doc comments in a .h file, they normally go into the .c
> file, next to the code itself.  That makes it easier to fix up and
> realise when they need to be changed when the code changes.  The .h file
> rarely changes.

I'll move the lot over.

>> +/**
>> + * mux_chip_register() - Register a mux-chip, thus readying the controllers
>> + *			 for use.
>> + * @mux_chip: The mux-chip to register.
>> + *
>> + * Do not retry registration of the same mux-chip on failure. You should
>> + * instead put it away with mux_chip_free() and allocate a new one, if you
>> + * for some reason would like to retry registration.
>> + *
>> + * Return: Zero on success or a negative errno on error.
>> + */
>> +int mux_chip_register(struct mux_chip *mux_chip);
>> +
>> +/**
>> + * mux_chip_unregister() - Take the mux-chip off-line.
>> + * @mux_chip: The mux-chip to unregister.
>> + *
>> + * mux_chip_unregister() reverses the effects of mux_chip_register().
>> + * But not completely, you should not try to call mux_chip_register()
>> + * on a mux-chip that has been registered before.
>> + */
>> +void mux_chip_unregister(struct mux_chip *mux_chip);
>> +
>> +/**
>> + * mux_chip_free() - Free the mux-chip for good.
>> + * @mux_chip: The mux-chip to free.
>> + *
>> + * mux_chip_free() reverses the effects of mux_chip_alloc().
>> + */
>> +void mux_chip_free(struct mux_chip *mux_chip);
>> +
>> +/**
>> + * devm_mux_chip_alloc() - Resource-managed version of mux_chip_alloc().
>> + * @dev: The parent device implementing the mux interface.
>> + * @controllers: The number of mux controllers to allocate for this chip.
>> + * @sizeof_priv: Size of extra memory area for private use by the caller.
>> + *
>> + * See mux_chip_alloc() for more details.
>> + *
>> + * Return: A pointer to the new mux-chip, NULL on failure.
>> + */
>> +struct mux_chip *devm_mux_chip_alloc(struct device *dev,
>> +				     unsigned int controllers,
>> +				     size_t sizeof_priv);
>> +
>> +/**
>> + * devm_mux_chip_register() - Resource-managed version mux_chip_register().
>> + * @dev: The parent device implementing the mux interface.
>> + * @mux_chip: The mux-chip to register.
>> + *
>> + * See mux_chip_register() for more details.
>> + *
>> + * Return: Zero on success or a negative errno on error.
>> + */
>> +int devm_mux_chip_register(struct device *dev, struct mux_chip *mux_chip);
>> +
>> +/**
>> + * devm_mux_chip_unregister() - Resource-managed version mux_chip_unregister().
>> + * @dev: The device that originally registered the mux-chip.
>> + * @mux_chip: The mux-chip to unregister.
>> + *
>> + * See mux_chip_unregister() for more details.
>> + *
>> + * Note that you do not normally need to call this function.
> 
> Odd, then why is it exported???

You normally don't call the devm_foo_{free,release,unregister,etc} functions.
The intention is of course that the resourse cleans up automatically. But there
are no cases where the manual clean up is not available, at least not that I can
find?

>> + */
>> +void devm_mux_chip_unregister(struct device *dev, struct mux_chip *mux_chip);
>> +
>> +/**
>> + * devm_mux_chip_free() - Resource-managed version mux_chip_free().
>> + * @dev: The device that originally got the mux-chip.
>> + * @mux_chip: The mux-chip to free.
>> + *
>> + * See mux_chip_free() for more details.
>> + *
>> + * Note that you do not normally need to call this function.
>> + */
>> +void devm_mux_chip_free(struct device *dev, struct mux_chip *mux_chip);
>> +
>> +/**
>> + * mux_control_select() - Select the given multiplexer state.
>> + * @mux: The mux-control to request a change of state from.
>> + * @state: The new requested state.
>> + *
>> + * Make sure to call mux_control_deselect() when the operation is complete and
>> + * the mux-control is free for others to use, but do not call
>> + * mux_control_deselect() if mux_control_select() fails.
>> + *
>> + * Return: 0 if the requested state was already active, or 1 it the
>> + * mux-control state was changed to the requested state. Or a negative
>> + * errno on error.
>> + *
>> + * Note that the difference in return value of zero or one is of
>> + * questionable value; especially if the mux-control has several independent
>> + * consumers, which is something the consumers should perhaps not be making
>> + * assumptions about.
> 
> I don't understand this note, what is a user of this api supposed to do
> differently between 1 and 0?  Why make the difference at all?

If the consumer somehow *knows* that it is the only user of a mux controller,
it can use the return value to shortcut (perhaps costly) actions only needed
when the mux changes. The 1/0 difference was also a "free" extra given the
current implementation of mux_control_select. But it's cheep for the consumer
to keep track of this by itself if it needs it.

It's when there are several (independent) consumers of a mux controller that
the information in the return value is questionable and can't be used to
shortcut actions only needed on mux changes.

Now that you point the finger at it and I have been made to spell out the
problem, I think it's probably wise to remove the distiction so that users
do not start to use the return value when they shouldn't. It's generally
better to keep track of the expected mux state in the consumer and use that
local information to shortcut those (perhaps costly) actions instead.

> And I agree with the comment to split this up into 2 different .h files,
> if possible.

Will split.

> thanks,
> 
> greg k-h
> 

I'll wait for further feedback before posting a v14.

Cheers and thanks,
peda

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Greg KH April 18, 2017, 11:44 a.m.
On Tue, Apr 18, 2017 at 12:59:50PM +0200, Peter Rosin wrote:
> On 2017-04-18 10:51, Greg Kroah-Hartman wrote:
> > On Thu, Apr 13, 2017 at 06:43:07PM +0200, Peter Rosin wrote:
> >> +config MUX_GPIO
> >> +	tristate "GPIO-controlled Multiplexer"
> >> +	depends on OF && GPIOLIB
> > 
> > Why have the gpio and mux core in the same patch?
> 
> One is not usable w/o the other. I can split them if you want to?

Then why are they two different config options?  Add the core code in
one patch, and then add the gpio controled mutiplexer in a separate
patch.

> >> +static struct class mux_class = {
> >> +	.name = "mux",
> >> +	.owner = THIS_MODULE,
> >> +};
> > 
> > No Documentation/ABI/ update for your sysfs files?  Please do so.
> 
> Ok I'll look into it. Wasn't even aware that I added any. But there's the
> new class of course...

Hint, you have files, the devices that belong to the class :)

> >> +static int __init mux_init(void)
> >> +{
> >> +	return class_register(&mux_class);
> >> +}
> >> +
> >> +static DEFINE_IDA(mux_ida);
> > 
> > When your module is unloaded, you forgot to clean this structure up with
> > what was done with it.
> 
> I was under the impression that not providing an exit function for modules
> made the module infrastructure prevent unloading (by bumping some reference
> counter). Maybe that is a misconception?

Ah, messy, don't do that.  Make it so you can unload your module please,
why wouldn't you want that to happen?

> >> +	mux_chip = kzalloc(sizeof(*mux_chip) +
> >> +			   controllers * sizeof(*mux_chip->mux) +
> >> +			   sizeof_priv, GFP_KERNEL);
> >> +	if (!mux_chip)
> >> +		return NULL;
> > 
> > You don't return PTR_ERR(-ENOMEM)?  Ok, why not?  (I'm not arguing for
> > it, just curious...)
> 
> There's no particular reason. Do you think I should change it?

What does the caller do with an error?  Pass it up to where?  Who gets
it?  Don't you want the caller to know you are out of memory?

> >> +
> >> +	device_initialize(&mux_chip->dev);
> > 
> > Why are you not registering the device here as well?  Why have this be a
> > two step process?
> 
> Because of idle state handling. The drivers are expected to fill in
> the desired idle state(s) after allocating the mux controller(s).
> Then, when registering, the desired idle state is activated (if the
> idle state is not idle-as-is, of course) and as a last step the mux
> is "advertised".

Ok, is that documented in the functions somewhere?

> >> +	ret = device_add(&mux_chip->dev);
> >> +	if (ret < 0)
> >> +		dev_err(&mux_chip->dev,
> >> +			"device_add failed in mux_chip_register: %d\n", ret);
> > 
> > Did you run checkpatch.pl in strict mode on this new file?  Please do so :)
> 
> I did, and did it again just to be sure, and I do not get any complaints.
> So, what's wrong?

You list the function name in the printk string, it should complain
that __func__ should be used.  Oh well, it's just a perl script, it
doesn't always catch everything.
isn't always correct :)

> >> +EXPORT_SYMBOL_GPL(devm_mux_chip_alloc);
> > 
> > 
> > Having devm functions that create/destroy other struct devices worries
> > me, do we have other examples of this happening today?  Are you sure you
> > got the reference counting all correct?
> 
> drivers/iio/industrialio_core.c:devm_iio_device_alloc
> 
> Or is the iio case different in some subtle way that I'm missing?

I don't know, hopefully you got it all correct, I haven't audited that
code path in a long time :)

> >> +
> >> +static int devm_mux_chip_match(struct device *dev, void *res, void *data)
> >> +{
> >> +	struct mux_chip **r = res;
> >> +
> >> +	if (WARN_ON(!r || !*r))
> > 
> > How can this happen?
> 
> It shouldn't. I copied the pattern from the iio subsystem.

Then it should be removed there too...

> >> +void devm_mux_chip_free(struct device *dev, struct mux_chip *mux_chip)
> >> +{
> >> +	WARN_ON(devres_release(dev, devm_mux_chip_release,
> >> +			       devm_mux_chip_match, mux_chip));
> > 
> > What can someone do with these WARN_ON() splats in the kernel log?
> 
> Don't know. Again, I copied the pattern from the iio subsystem.

If you don't know what it should be used for, don't copy it!

Cargo-cult coding is horrible, please no.

> >> +	/* ...or it's just contended. */
> >> +	down_write(&mux->lock);
> > 
> > Why use a read/write lock at all?  Have you tested this to verify it
> > really is faster and needed?
> 
> For one of the HW configuration that drove the development, the same mux
> controller is used to mux both an I2C channel and a couple of ADC lines.
> 
> If there is no kind of reader/writer locking going on, there is no way to
> do ADC readings concurrently with an I2C transfer even when the consumers
> want the mux in the same position. With an ordinary mutex controlling the
> mux position, the consumers will unconditionally get serialized, which
> seems like a waste to me. Or maybe I'm missing something?

Why is serializing things a "waste"?  Again, a rw semaphore is slower,
takes more logic to get correct, and is very complex at times.  If you
are not SURE you need it, and that it matters, don't use it.  And if you
do use it, document the heck out of it how you need it and why.

> >> +
> >> +	if (mux->cached_state == state) {
> >> +		/*
> >> +		 * Hmmm, someone else changed the mux to my liking.
> >> +		 * That makes me wonder how long I waited for nothing?
> >> +		 */
> >> +		downgrade_write(&mux->lock);
> > 
> > Oh that always scares me...  Are you _sure_ this is correct?  And
> > needed?
> 
> It might not be needed, and it would probably work ok to just fall
> through and call mux_control_set unconditionally. What is it that
> always scares you exactly? Relying on cached state to be correct?
> Downgrading writer locks?

downgrading a writer lock scares me, especially for something as
"simple" as this type of interface.  Again, don't use it unless you
_have_ to.  Simple is good, start with that always.

> >> +	if (mux->idle_state != MUX_IDLE_AS_IS &&
> >> +	    mux->idle_state != mux->cached_state)
> >> +		ret = mux_control_set(mux, mux->idle_state);
> >> +
> >> +	up_read(&mux->lock);
> > 
> > You require a lock to be held for a "global" function?  Without
> > documentation?  Or even a sparse marking?  That's asking for trouble...
> 
> Documentation I can handle, but where should I look to understand how I
> should add sparse markings?

Run sparse on the code and see what it says :)

> The mux needs to be locked somehow. But as I stated in the cover letter
> the rwsem isn't a perfect fit.
> 
> 	I'm using an rwsem to lock a mux, but that isn't really a
> 	perfect fit. Is there a better locking primitive that I don't
> 	know about that fits better? I had a mutex at one point, but
> 	that didn't allow any concurrent accesses at all. At least
> 	the rwsem allows concurrent access as long as all users
> 	agree on the mux state, but I suspect that the rwsem will
> 	degrade to the mutex situation pretty quickly if there is
> 	any contention.
> 
> Also, the lock doesn't add anything if there is only one consumer of
> a mux controller. Maybe there should be some mechanism for shortcutting
> the locking for the (more common?) single-consumer case?
> 
> But again, I need the locking for my multi-consumer use case.

Go back to a mutex, and having a function that requires it to be held
is, icky.

> >> +/*
> >> + * Using subsys_initcall instead of module_init here to ensure - for the
> >> + * non-modular case - that the subsystem is initialized when mux consumers
> >> + * and mux controllers start to use it /without/ relying on link order.
> >> + * For the modular case, the ordering is ensured with module dependencies.
> >> + */
> >> +subsys_initcall(mux_init);
> > 
> > Even with subsys_initcall you are relying on link order, you do realize
> > that?  What about other subsystems that rely on this?  :)
> 
> Yes, that is true, but if others start relying on this, that's their problem,
> right? :-)

Yes, but no need to document something that isn't true.  You are relying
on link order here.

> >> +struct mux_control_ops {
> >> +	int (*set)(struct mux_control *mux, int state);
> >> +};
> >> +
> >> +/* These defines match the constants from the dt-bindings. On purpose. */
> > 
> > Why on purpose?
> 
> I sure wasn't an accident? :-)
> 
> Want me to remove it?

At least explain _why_ you are doing this, that would help, right?

thanks,

greg k-h
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Peter Rosin April 18, 2017, 9:53 p.m.
On 2017-04-18 13:44, Greg Kroah-Hartman wrote:
> On Tue, Apr 18, 2017 at 12:59:50PM +0200, Peter Rosin wrote:
>> On 2017-04-18 10:51, Greg Kroah-Hartman wrote:
>>> On Thu, Apr 13, 2017 at 06:43:07PM +0200, Peter Rosin wrote:
>>>> +config MUX_GPIO
>>>> +	tristate "GPIO-controlled Multiplexer"
>>>> +	depends on OF && GPIOLIB
>>>
>>> Why have the gpio and mux core in the same patch?
>>
>> One is not usable w/o the other. I can split them if you want to?
> 
> Then why are they two different config options?  Add the core code in
> one patch, and then add the gpio controled mutiplexer in a separate
> patch.

Ah, I meant when there are not yet any other mux drivers. I'll just
split the patch.

>>>> +static struct class mux_class = {
>>>> +	.name = "mux",
>>>> +	.owner = THIS_MODULE,
>>>> +};
>>>
>>> No Documentation/ABI/ update for your sysfs files?  Please do so.
>>
>> Ok I'll look into it. Wasn't even aware that I added any. But there's the
>> new class of course...
> 
> Hint, you have files, the devices that belong to the class :)

Right.

>>>> +static int __init mux_init(void)
>>>> +{
>>>> +	return class_register(&mux_class);
>>>> +}
>>>> +
>>>> +static DEFINE_IDA(mux_ida);
>>>
>>> When your module is unloaded, you forgot to clean this structure up with
>>> what was done with it.
>>
>> I was under the impression that not providing an exit function for modules
>> made the module infrastructure prevent unloading (by bumping some reference
>> counter). Maybe that is a misconception?
> 
> Ah, messy, don't do that.  Make it so you can unload your module please,
> why wouldn't you want that to happen?

What made me do it was the worry that mux drivers might be left dangling
w/o the core. But that can probably only happen if someone is very
deliberately trying to break stuff by forcing unloads??

>>>> +	mux_chip = kzalloc(sizeof(*mux_chip) +
>>>> +			   controllers * sizeof(*mux_chip->mux) +
>>>> +			   sizeof_priv, GFP_KERNEL);
>>>> +	if (!mux_chip)
>>>> +		return NULL;
>>>
>>> You don't return PTR_ERR(-ENOMEM)?  Ok, why not?  (I'm not arguing for
>>> it, just curious...)
>>
>> There's no particular reason. Do you think I should change it?
> 
> What does the caller do with an error?  Pass it up to where?  Who gets
> it?  Don't you want the caller to know you are out of memory?

The current callers return -ENOMEM when NULL is returned here. Looks
like I'm going to be doing some fairly major changes anyway so I'll
just change this too while at it...

>>>> +
>>>> +	device_initialize(&mux_chip->dev);
>>>
>>> Why are you not registering the device here as well?  Why have this be a
>>> two step process?
>>
>> Because of idle state handling. The drivers are expected to fill in
>> the desired idle state(s) after allocating the mux controller(s).
>> Then, when registering, the desired idle state is activated (if the
>> idle state is not idle-as-is, of course) and as a last step the mux
>> is "advertised".
> 
> Ok, is that documented in the functions somewhere?

I'll make sure to add it if it's missing.

>>>> +	ret = device_add(&mux_chip->dev);
>>>> +	if (ret < 0)
>>>> +		dev_err(&mux_chip->dev,
>>>> +			"device_add failed in mux_chip_register: %d\n", ret);
>>>
>>> Did you run checkpatch.pl in strict mode on this new file?  Please do so :)
>>
>> I did, and did it again just to be sure, and I do not get any complaints.
>> So, what's wrong?
> 
> You list the function name in the printk string, it should complain
> that __func__ should be used.  Oh well, it's just a perl script, it
> doesn't always catch everything.
> isn't always correct :)

Ah, ok.

>>>> +EXPORT_SYMBOL_GPL(devm_mux_chip_alloc);
>>>
>>>
>>> Having devm functions that create/destroy other struct devices worries
>>> me, do we have other examples of this happening today?  Are you sure you
>>> got the reference counting all correct?
>>
>> drivers/iio/industrialio_core.c:devm_iio_device_alloc
>>
>> Or is the iio case different in some subtle way that I'm missing?
> 
> I don't know, hopefully you got it all correct, I haven't audited that
> code path in a long time :)

Looks as ok to me as it did before. Moving on... :-)

>>>> +
>>>> +static int devm_mux_chip_match(struct device *dev, void *res, void *data)
>>>> +{
>>>> +	struct mux_chip **r = res;
>>>> +
>>>> +	if (WARN_ON(!r || !*r))
>>>
>>> How can this happen?
>>
>> It shouldn't. I copied the pattern from the iio subsystem.
> 
> Then it should be removed there too...

Ok, I'll see if I can find time to suggest some patch(es) to Jonathan.

>>>> +void devm_mux_chip_free(struct device *dev, struct mux_chip *mux_chip)
>>>> +{
>>>> +	WARN_ON(devres_release(dev, devm_mux_chip_release,
>>>> +			       devm_mux_chip_match, mux_chip));
>>>
>>> What can someone do with these WARN_ON() splats in the kernel log?
>>
>> Don't know. Again, I copied the pattern from the iio subsystem.
> 
> If you don't know what it should be used for, don't copy it!
> 
> Cargo-cult coding is horrible, please no.

Ok, I'll just drop those WARNs...

>>>> +	/* ...or it's just contended. */
>>>> +	down_write(&mux->lock);
>>>
>>> Why use a read/write lock at all?  Have you tested this to verify it
>>> really is faster and needed?
>>
>> For one of the HW configuration that drove the development, the same mux
>> controller is used to mux both an I2C channel and a couple of ADC lines.
>>
>> If there is no kind of reader/writer locking going on, there is no way to
>> do ADC readings concurrently with an I2C transfer even when the consumers
>> want the mux in the same position. With an ordinary mutex controlling the
>> mux position, the consumers will unconditionally get serialized, which
>> seems like a waste to me. Or maybe I'm missing something?
> 
> Why is serializing things a "waste"?  Again, a rw semaphore is slower,
> takes more logic to get correct, and is very complex at times.  If you
> are not SURE you need it, and that it matters, don't use it.  And if you
> do use it, document the heck out of it how you need it and why.

It's a waste of time because two independent mux consumers of the same
mux controller can't do things concurrently even if they want the same
thing from the mux. Let's say that one mux consumer is an iio-mux and
one is an i2c-mux. Also, let's say that you for some reason need to get
a lot of samples at a determined rate through the iio-mux. With a mutex,
that can't happen if there is an access through the i2c-mux taking
"forever" in the eyes of the ADC/iio-mux, even if they both want the
shared mux controller to be in the same position. The ADC/iio-mux and
the i2c-mux would be waiting for each other for no purpose at all.

>>>> +
>>>> +	if (mux->cached_state == state) {
>>>> +		/*
>>>> +		 * Hmmm, someone else changed the mux to my liking.
>>>> +		 * That makes me wonder how long I waited for nothing?
>>>> +		 */
>>>> +		downgrade_write(&mux->lock);
>>>
>>> Oh that always scares me...  Are you _sure_ this is correct?  And
>>> needed?
>>
>> It might not be needed, and it would probably work ok to just fall
>> through and call mux_control_set unconditionally. What is it that
>> always scares you exactly? Relying on cached state to be correct?
>> Downgrading writer locks?
> 
> downgrading a writer lock scares me, especially for something as
> "simple" as this type of interface.  Again, don't use it unless you
> _have_ to.  Simple is good, start with that always.

Some kind of lock needs to be grabbed in mux_control_select and
released in mux_control_deselect that fixes the mux state while the
mux consumer goes about its business. For the reasons stated above I
went with a reader/writer lock instead of the mutex I had originally.

Agreed, the code in mux_control_select is a few more lines than I
like, but I suspected the big issue to be holding *any* lock over
a pair of "global" functions. Changing from holding a rw-lock as
reader to instead holding a mutex changes very little in my eyes.
mux_control_select is simply not *that* complicated...

>>>> +	if (mux->idle_state != MUX_IDLE_AS_IS &&
>>>> +	    mux->idle_state != mux->cached_state)
>>>> +		ret = mux_control_set(mux, mux->idle_state);
>>>> +
>>>> +	up_read(&mux->lock);
>>>
>>> You require a lock to be held for a "global" function?  Without
>>> documentation?  Or even a sparse marking?  That's asking for trouble...
>>
>> Documentation I can handle, but where should I look to understand how I
>> should add sparse markings?
> 
> Run sparse on the code and see what it says :)

Will do.

>> The mux needs to be locked somehow. But as I stated in the cover letter
>> the rwsem isn't a perfect fit.
>>
>> 	I'm using an rwsem to lock a mux, but that isn't really a
>> 	perfect fit. Is there a better locking primitive that I don't
>> 	know about that fits better? I had a mutex at one point, but
>> 	that didn't allow any concurrent accesses at all. At least
>> 	the rwsem allows concurrent access as long as all users
>> 	agree on the mux state, but I suspect that the rwsem will
>> 	degrade to the mutex situation pretty quickly if there is
>> 	any contention.
>>
>> Also, the lock doesn't add anything if there is only one consumer of
>> a mux controller. Maybe there should be some mechanism for shortcutting
>> the locking for the (more common?) single-consumer case?
>>
>> But again, I need the locking for my multi-consumer use case.
> 
> Go back to a mutex, and having a function that requires it to be held
> is, icky.

But how do you propose that the ickyness is avoided? It's a requirement
that any waiters are released when the mux is available...

*thinking/coding for a bit*

I'm going to experiment with the below (untested) code which AFAICT
has the issues that starvation is possible, that it isn't first-come
first-serve, and that once there is contention the waiters may wait
for a longer time than needed...

On the positive side there are no actual locks held from select over
to deselect and there's no rwsem. But a missing (or an extra) deselect
still messes things up pretty fatally, since the below is really just
some kind of open coded locking thingy that I thought was one of the
things I should stay away from. But maybe you like it better?

(In the below, mux->wait_sem is assumed to be initialized to zero.)

int mux_control_select(struct mux_control *mux, int state)
{
	int ret = 0;

again:
	mutex_lock(&mux->lock);

	if (mux->cached_state == state) {
		/* The mux is already correct, just bump the user count. */
		++mux->users;
		goto done;
	}
	if (mux->users) {
		/* The mux needs updating but is in use, wait... */
		++mux->waiters;
		mutex_unlock(&mux->lock);
		down(&mux->wait_sem);
		goto again;
	}

	/* The mux needs updating and is unused. */
	ret = mux_control_set(mux, state);
	if (ret >= 0) {
		++mux->users;
		goto done;
	}

	/* The mux update failed, try to revert if appropriate... */
	if (mux->idle_state != MUX_IDLE_AS_IS)
		mux_control_set(mux, mux->idle_state);

	/* ...and release a waiter if there is one. */
	if (mux->waiters) {
		--mux->waiters;
		up(&mux->wait_sem);
	}

done:
	mutex_unlock(&mux->lock);

	return ret;
}

int mux_control_deselect(struct mux_control *mux)
{
	int ret = 0;

	mutex_lock(&mux->lock);
	if (--mux->users)
		goto done;

	/* This was the last user, idle the mux if appropriate... */
	if (mux->idle_state != MUX_IDLE_AS_IS &&
	    mux->idle_state != mux->cached_state)
		ret = mux_control_set(mux, mux->idle_state);

	/* ...and release a waiter if there is one. */
	if (mux->waiters) {
		--mux->waiters;
		up(&mux->wait_sem);
	}

done:
	mutex_unlock(&mux->lock);

	return ret;
}

>>>> +/*
>>>> + * Using subsys_initcall instead of module_init here to ensure - for the
>>>> + * non-modular case - that the subsystem is initialized when mux consumers
>>>> + * and mux controllers start to use it /without/ relying on link order.
>>>> + * For the modular case, the ordering is ensured with module dependencies.
>>>> + */
>>>> +subsys_initcall(mux_init);
>>>
>>> Even with subsys_initcall you are relying on link order, you do realize
>>> that?  What about other subsystems that rely on this?  :)
>>
>> Yes, that is true, but if others start relying on this, that's their problem,
>> right? :-)
> 
> Yes, but no need to document something that isn't true.  You are relying
> on link order here.

Well, am I? If I change it to module_init I do get runtime errors for a
non-modular build (when mux consumers get initialized before the mux core).
With subsys_init the mux core get initialized before all (current)
consumers. I don't really see how the link order *currently* matters?

So, just making sure that we are on the same page, the thing that relies on
link order are any mux consumers/drivers that in the future may be added as
part of some other subsys_init call (or earlier). Right?

Hmmm, or are you perhaps referring to the fact that the mux core depends on
other subsystems being initialized first?

>>>> +struct mux_control_ops {
>>>> +	int (*set)(struct mux_control *mux, int state);
>>>> +};
>>>> +
>>>> +/* These defines match the constants from the dt-bindings. On purpose. */
>>>
>>> Why on purpose?
>>
>> I sure wasn't an accident? :-)
>>
>> Want me to remove it?
> 
> At least explain _why_ you are doing this, that would help, right?

Should I perhaps just #include <linux/dt-bindings/mux/mux.h> instead?
Is that an OK thing to do? I didn't because I feared it might come back
to haunt me at some point if the bindings header needed stuff in the
future that made it incompatible...

It's also not terribly common to include bindings from an include...

Cheers,
peda

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Joe Perches April 19, 2017, 2:23 a.m.
On Tue, 2017-04-18 at 23:53 +0200, Peter Rosin wrote:
> On 2017-04-18 13:44, Greg Kroah-Hartman wrote:
> > On Tue, Apr 18, 2017 at 12:59:50PM +0200, Peter Rosin wrote:
[]
> > > > > +	ret = device_add(&mux_chip->dev);
> > > > > +	if (ret < 0)
> > > > > +		dev_err(&mux_chip->dev,
> > > > > +			"device_add failed in mux_chip_register: %d\n", ret);
> > > > 
> > > > Did you run checkpatch.pl in strict mode on this new file?  Please do so :)
> > > 
> > > I did, and did it again just to be sure, and I do not get any complaints.
> > > So, what's wrong?
> > 
> > You list the function name in the printk string, it should complain
> > that __func__ should be used.  Oh well, it's just a perl script, it
> > doesn't always catch everything.
> > isn't always correct :)
> 
> Ah, ok.

Also, please use the checkpatch in -next as it has a
slightly better mechanism to identify functions and
uses in strings.

$ ./scripts/checkpatch.pl ~/1.patch
WARNING: Prefer using '"%s...", __func__' to using 'mux_chip_register', this function's name, in a string
#302: FILE: drivers/mux/mux-core.c:134:
+			"device_add failed in mux_chip_register: %d\n", ret);


--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Philipp Zabel April 19, 2017, 9:06 a.m.
On Thu, 2017-04-13 at 18:43 +0200, Peter Rosin wrote:
> Add a new minimalistic subsystem that handles multiplexer controllers.
> When multiplexers are used in various places in the kernel, and the
> same multiplexer controller can be used for several independent things,
> there should be one place to implement support for said multiplexer
> controller.
> 
> A single multiplexer controller can also be used to control several
> parallel multiplexers, that are in turn used by different subsystems
> in the kernel, leading to a need to coordinate multiplexer accesses.
> The multiplexer subsystem handles this coordination.
> 
> This new mux controller subsystem initially comes with a single backend
> driver that controls gpio based multiplexers. Even though not needed by
> this initial driver, the mux controller subsystem is prepared to handle
> chips with multiple (independent) mux controllers.
> 
> Reviewed-by: Jonathan Cameron <jic23@kernel.org>
> Signed-off-by: Peter Rosin <peda@axentia.se>
> ---
>  Documentation/driver-model/devres.txt |   8 +
>  MAINTAINERS                           |   2 +
>  drivers/Kconfig                       |   2 +
>  drivers/Makefile                      |   1 +
>  drivers/mux/Kconfig                   |  34 +++
>  drivers/mux/Makefile                  |   6 +
>  drivers/mux/mux-core.c                | 422 ++++++++++++++++++++++++++++++++++
>  drivers/mux/mux-gpio.c                | 114 +++++++++
>  include/linux/mux.h                   | 252 ++++++++++++++++++++
>  9 files changed, 841 insertions(+)
>  create mode 100644 drivers/mux/Kconfig
>  create mode 100644 drivers/mux/Makefile
>  create mode 100644 drivers/mux/mux-core.c
>  create mode 100644 drivers/mux/mux-gpio.c
>  create mode 100644 include/linux/mux.h
> 
> diff --git a/Documentation/driver-model/devres.txt b/Documentation/driver-model/devres.txt
> index efb8200819d6..e2343d9cbec7 100644
> --- a/Documentation/driver-model/devres.txt
> +++ b/Documentation/driver-model/devres.txt
> @@ -337,6 +337,14 @@ MEM
>  MFD
>    devm_mfd_add_devices()
>  
> +MUX
> +  devm_mux_chip_alloc()
> +  devm_mux_chip_free()
> +  devm_mux_chip_register()
> +  devm_mux_chip_unregister()
> +  devm_mux_control_get()
> +  devm_mux_control_put()
> +
>  PER-CPU MEM
>    devm_alloc_percpu()
>    devm_free_percpu()
> diff --git a/MAINTAINERS b/MAINTAINERS
> index 7fc06739c8ad..591eba737678 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -8563,6 +8563,8 @@ M:	Peter Rosin <peda@axentia.se>
>  S:	Maintained
>  F:	Documentation/devicetree/bindings/mux/
>  F:	include/linux/dt-bindings/mux/
> +F:	include/linux/mux.h
> +F:	drivers/mux/
>  
>  MULTISOUND SOUND DRIVER
>  M:	Andrew Veliath <andrewtv@usa.net>
> diff --git a/drivers/Kconfig b/drivers/Kconfig
> index 117ca14ccf85..a7ea13e1b869 100644
> --- a/drivers/Kconfig
> +++ b/drivers/Kconfig
> @@ -204,4 +204,6 @@ source "drivers/fpga/Kconfig"
>  
>  source "drivers/fsi/Kconfig"
>  
> +source "drivers/mux/Kconfig"
> +
>  endmenu
> diff --git a/drivers/Makefile b/drivers/Makefile
> index 2eced9afba53..c0436f6dd5a9 100644
> --- a/drivers/Makefile
> +++ b/drivers/Makefile
> @@ -177,3 +177,4 @@ obj-$(CONFIG_ANDROID)		+= android/
>  obj-$(CONFIG_NVMEM)		+= nvmem/
>  obj-$(CONFIG_FPGA)		+= fpga/
>  obj-$(CONFIG_FSI)		+= fsi/
> +obj-$(CONFIG_MULTIPLEXER)	+= mux/
> diff --git a/drivers/mux/Kconfig b/drivers/mux/Kconfig
> new file mode 100644
> index 000000000000..41dfe08ead84
> --- /dev/null
> +++ b/drivers/mux/Kconfig
> @@ -0,0 +1,34 @@
> +#
> +# Multiplexer devices
> +#
> +
> +menuconfig MULTIPLEXER
> +	tristate "Multiplexer subsystem"
> +	help
> +	  Multiplexer controller subsystem. Multiplexers are used in a
> +	  variety of settings, and this subsystem abstracts their use
> +	  so that the rest of the kernel sees a common interface. When
> +	  multiple parallel multiplexers are controlled by one single
> +	  multiplexer controller, this subsystem also coordinates the
> +	  multiplexer accesses.
> +
> +	  To compile the subsystem as a module, choose M here: the module will
> +	  be called mux-core.
> +
> +if MULTIPLEXER
> +
> +config MUX_GPIO
> +	tristate "GPIO-controlled Multiplexer"
> +	depends on OF && GPIOLIB
> +	help
> +	  GPIO-controlled Multiplexer controller.
> +
> +	  The driver builds a single multiplexer controller using a number
> +	  of gpio pins. For N pins, there will be 2^N possible multiplexer
> +	  states. The GPIO pins can be connected (by the hardware) to several
> +	  multiplexers, which in that case will be operated in parallel.
> +
> +	  To compile the driver as a module, choose M here: the module will
> +	  be called mux-gpio.
> +
> +endif
> diff --git a/drivers/mux/Makefile b/drivers/mux/Makefile
> new file mode 100644
> index 000000000000..bb16953f6290
> --- /dev/null
> +++ b/drivers/mux/Makefile
> @@ -0,0 +1,6 @@
> +#
> +# Makefile for multiplexer devices.
> +#
> +
> +obj-$(CONFIG_MULTIPLEXER)	+= mux-core.o
> +obj-$(CONFIG_MUX_GPIO)		+= mux-gpio.o
> diff --git a/drivers/mux/mux-core.c b/drivers/mux/mux-core.c
> new file mode 100644
> index 000000000000..66a8bccfc3d7
> --- /dev/null
> +++ b/drivers/mux/mux-core.c
> @@ -0,0 +1,422 @@
> +/*
> + * Multiplexer subsystem
> + *
> + * Copyright (C) 2017 Axentia Technologies AB
> + *
> + * Author: Peter Rosin <peda@axentia.se>
> + *
> + * 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.
> + */
> +
> +#define pr_fmt(fmt) "mux-core: " fmt
> +
> +#include <linux/device.h>
> +#include <linux/err.h>
> +#include <linux/export.h>
> +#include <linux/idr.h>
> +#include <linux/init.h>
> +#include <linux/module.h>
> +#include <linux/mux.h>
> +#include <linux/of.h>
> +#include <linux/of_platform.h>
> +#include <linux/slab.h>
> +
> +/*
> + * The idle-as-is "state" is not an actual state that may be selected, it
> + * only implies that the state should not be changed. So, use that state
> + * as indication that the cached state of the multiplexer is unknown.
> + */
> +#define MUX_CACHE_UNKNOWN MUX_IDLE_AS_IS
> +
> +static struct class mux_class = {
> +	.name = "mux",
> +	.owner = THIS_MODULE,
> +};
> +
> +static int __init mux_init(void)
> +{
> +	return class_register(&mux_class);
> +}
> +
> +static DEFINE_IDA(mux_ida);
> +
> +static void mux_chip_release(struct device *dev)
> +{
> +	struct mux_chip *mux_chip = to_mux_chip(dev);
> +
> +	ida_simple_remove(&mux_ida, mux_chip->id);
> +	kfree(mux_chip);
> +}
> +
> +static struct device_type mux_type = {
> +	.name = "mux-chip",
> +	.release = mux_chip_release,
> +};
> +
> +struct mux_chip *mux_chip_alloc(struct device *dev,
> +				unsigned int controllers, size_t sizeof_priv)
> +{
> +	struct mux_chip *mux_chip;
> +	int i;
> +
> +	if (WARN_ON(!dev || !controllers))
> +		return NULL;
> +
> +	mux_chip = kzalloc(sizeof(*mux_chip) +
> +			   controllers * sizeof(*mux_chip->mux) +
> +			   sizeof_priv, GFP_KERNEL);
> +	if (!mux_chip)
> +		return NULL;
> +
> +	mux_chip->mux = (struct mux_control *)(mux_chip + 1);
> +	mux_chip->dev.class = &mux_class;
> +	mux_chip->dev.type = &mux_type;
> +	mux_chip->dev.parent = dev;
> +	mux_chip->dev.of_node = dev->of_node;
> +	dev_set_drvdata(&mux_chip->dev, mux_chip);
> +
> +	mux_chip->id = ida_simple_get(&mux_ida, 0, 0, GFP_KERNEL);
> +	if (mux_chip->id < 0) {
> +		pr_err("muxchipX failed to get a device id\n");
> +		kfree(mux_chip);
> +		return NULL;
> +	}
> +	dev_set_name(&mux_chip->dev, "muxchip%d", mux_chip->id);
> +
> +	mux_chip->controllers = controllers;
> +	for (i = 0; i < controllers; ++i) {
> +		struct mux_control *mux = &mux_chip->mux[i];
> +
> +		mux->chip = mux_chip;
> +		init_rwsem(&mux->lock);
> +		mux->cached_state = MUX_CACHE_UNKNOWN;
> +		mux->idle_state = MUX_IDLE_AS_IS;
> +	}
> +
> +	device_initialize(&mux_chip->dev);
> +
> +	return mux_chip;
> +}
> +EXPORT_SYMBOL_GPL(mux_chip_alloc);
> +
> +static int mux_control_set(struct mux_control *mux, int state)
> +{
> +	int ret = mux->chip->ops->set(mux, state);
> +
> +	mux->cached_state = ret < 0 ? MUX_CACHE_UNKNOWN : state;
> +
> +	return ret;
> +}
> +
> +int mux_chip_register(struct mux_chip *mux_chip)
> +{
> +	int i;
> +	int ret;
> +
> +	for (i = 0; i < mux_chip->controllers; ++i) {
> +		struct mux_control *mux = &mux_chip->mux[i];
> +
> +		if (mux->idle_state == mux->cached_state)
> +			continue;
> +
> +		ret = mux_control_set(mux, mux->idle_state);
> +		if (ret < 0) {
> +			dev_err(&mux_chip->dev, "unable to set idle state\n");
> +			return ret;
> +		}
> +	}
> +
> +	ret = device_add(&mux_chip->dev);
> +	if (ret < 0)
> +		dev_err(&mux_chip->dev,
> +			"device_add failed in mux_chip_register: %d\n", ret);
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(mux_chip_register);
> +
> +void mux_chip_unregister(struct mux_chip *mux_chip)
> +{
> +	device_del(&mux_chip->dev);
> +}
> +EXPORT_SYMBOL_GPL(mux_chip_unregister);
> +
> +void mux_chip_free(struct mux_chip *mux_chip)
> +{
> +	if (!mux_chip)
> +		return;
> +
> +	put_device(&mux_chip->dev);
> +}
> +EXPORT_SYMBOL_GPL(mux_chip_free);
> +
> +static void devm_mux_chip_release(struct device *dev, void *res)
> +{
> +	struct mux_chip *mux_chip = *(struct mux_chip **)res;
> +
> +	mux_chip_free(mux_chip);
> +}
> +
> +struct mux_chip *devm_mux_chip_alloc(struct device *dev,
> +				     unsigned int controllers,
> +				     size_t sizeof_priv)
> +{
> +	struct mux_chip **ptr, *mux_chip;
> +
> +	ptr = devres_alloc(devm_mux_chip_release, sizeof(*ptr), GFP_KERNEL);
> +	if (!ptr)
> +		return NULL;
> +
> +	mux_chip = mux_chip_alloc(dev, controllers, sizeof_priv);
> +	if (!mux_chip) {
> +		devres_free(ptr);
> +		return NULL;
> +	}
> +
> +	*ptr = mux_chip;
> +	devres_add(dev, ptr);
> +
> +	return mux_chip;
> +}
> +EXPORT_SYMBOL_GPL(devm_mux_chip_alloc);
> +
> +static int devm_mux_chip_match(struct device *dev, void *res, void *data)
> +{
> +	struct mux_chip **r = res;
> +
> +	if (WARN_ON(!r || !*r))
> +		return 0;
> +
> +	return *r == data;
> +}
> +
> +void devm_mux_chip_free(struct device *dev, struct mux_chip *mux_chip)
> +{
> +	WARN_ON(devres_release(dev, devm_mux_chip_release,
> +			       devm_mux_chip_match, mux_chip));
> +}
> +EXPORT_SYMBOL_GPL(devm_mux_chip_free);
> +
> +static void devm_mux_chip_reg_release(struct device *dev, void *res)
> +{
> +	struct mux_chip *mux_chip = *(struct mux_chip **)res;
> +
> +	mux_chip_unregister(mux_chip);
> +}
> +
> +int devm_mux_chip_register(struct device *dev,
> +			   struct mux_chip *mux_chip)
> +{
> +	struct mux_chip **ptr;
> +	int res;
> +
> +	ptr = devres_alloc(devm_mux_chip_reg_release, sizeof(*ptr), GFP_KERNEL);
> +	if (!ptr)
> +		return -ENOMEM;
> +
> +	res = mux_chip_register(mux_chip);
> +	if (res) {
> +		devres_free(ptr);
> +		return res;
> +	}
> +
> +	*ptr = mux_chip;
> +	devres_add(dev, ptr);
> +
> +	return res;
> +}
> +EXPORT_SYMBOL_GPL(devm_mux_chip_register);
> +
> +void devm_mux_chip_unregister(struct device *dev, struct mux_chip *mux_chip)
> +{
> +	WARN_ON(devres_release(dev, devm_mux_chip_reg_release,
> +			       devm_mux_chip_match, mux_chip));
> +}
> +EXPORT_SYMBOL_GPL(devm_mux_chip_unregister);
> +
> +int mux_control_select(struct mux_control *mux, int state)

If we let two of these race, ...

> +{
> +	int ret;
> +
> +	if (down_read_trylock(&mux->lock)) {
> +		if (mux->cached_state == state)
> +			return 0;
> +		/* Sigh, the mux needs updating... */
> +		up_read(&mux->lock);

... and both decide the mux needs updating ...

> +	}
> +
> +	/* ...or it's just contended. */
> +	down_write(&mux->lock);

... then the last to get to down_write will just wait here forever (or
until the first consumer calls mux_control_deselect, which may never
happen)?

> +
> +	if (mux->cached_state == state) {
> +		/*
> +		 * Hmmm, someone else changed the mux to my liking.
> +		 * That makes me wonder how long I waited for nothing?
> +		 */
> +		downgrade_write(&mux->lock);
> +		return 0;
> +	}
> +
> +	ret = mux_control_set(mux, state);
> +	if (ret < 0) {
> +		if (mux->idle_state != MUX_IDLE_AS_IS)
> +			mux_control_set(mux, mux->idle_state);
> +
> +		up_write(&mux->lock);
> +		return ret;
> +	}
> +
> +	downgrade_write(&mux->lock);
> +
> +	return 1;
> +}
> +EXPORT_SYMBOL_GPL(mux_control_select);

I wonder if these should be called mux_control_lock/unlock instead,
which would allow for try_lock and lock_timeout variants.

regards
Philipp

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Peter Rosin April 19, 2017, noon
On 2017-04-19 11:06, Philipp Zabel wrote:
> On Thu, 2017-04-13 at 18:43 +0200, Peter Rosin wrote:
>> Add a new minimalistic subsystem that handles multiplexer controllers.
>> When multiplexers are used in various places in the kernel, and the
>> same multiplexer controller can be used for several independent things,
>> there should be one place to implement support for said multiplexer
>> controller.
>>
>> A single multiplexer controller can also be used to control several
>> parallel multiplexers, that are in turn used by different subsystems
>> in the kernel, leading to a need to coordinate multiplexer accesses.
>> The multiplexer subsystem handles this coordination.
>>
>> This new mux controller subsystem initially comes with a single backend
>> driver that controls gpio based multiplexers. Even though not needed by
>> this initial driver, the mux controller subsystem is prepared to handle
>> chips with multiple (independent) mux controllers.
>>
>> Reviewed-by: Jonathan Cameron <jic23@kernel.org>
>> Signed-off-by: Peter Rosin <peda@axentia.se>
>> ---

*snip*

>> +int mux_control_select(struct mux_control *mux, int state)
> 
> If we let two of these race, ...

The window for this "race" is positively huge. If there are several
mux consumers of a single mux controller, it is self-evident that
if one of them grabs the mux for a long time, the others will suffer.

The design is that the rwsem is reader-locked for the full duration
of a select/deselect operation by the mux consumer.

>> +{
>> +	int ret;
>> +
>> +	if (down_read_trylock(&mux->lock)) {
>> +		if (mux->cached_state == state)
>> +			return 0;
>> +		/* Sigh, the mux needs updating... */
>> +		up_read(&mux->lock);
> 
> ... and both decide the mux needs updating ...
> 
>> +	}
>> +
>> +	/* ...or it's just contended. */
>> +	down_write(&mux->lock);
> 
> ... then the last to get to down_write will just wait here forever (or
> until the first consumer calls mux_control_deselect, which may never
> happen)?

It is vital that the mux consumer call _deselect when it is done with
the mux. Not doing so will surely starve out any other mux consumers.
The whole thing is designed around the fact that mux consumers should
deselect the mux as soon as it's no longer needed.

It's simply not possible to share something as fundamental as a mux
without some cooperation. It's not like suffering mux consumers can
go off and use some other mux, and it's also not possible for a
"competing" mux consumer to just clobber the mux state.

>> +
>> +	if (mux->cached_state == state) {
>> +		/*
>> +		 * Hmmm, someone else changed the mux to my liking.
>> +		 * That makes me wonder how long I waited for nothing?
>> +		 */
>> +		downgrade_write(&mux->lock);
>> +		return 0;
>> +	}
>> +
>> +	ret = mux_control_set(mux, state);
>> +	if (ret < 0) {
>> +		if (mux->idle_state != MUX_IDLE_AS_IS)
>> +			mux_control_set(mux, mux->idle_state);
>> +
>> +		up_write(&mux->lock);
>> +		return ret;
>> +	}
>> +
>> +	downgrade_write(&mux->lock);
>> +
>> +	return 1;
>> +}
>> +EXPORT_SYMBOL_GPL(mux_control_select);
> 
> I wonder if these should be called mux_control_lock/unlock instead,
> which would allow for try_lock and lock_timeout variants.

Maybe, I'm not totally against it. Do others care to opine?

But mux_control_try_select and mux_control_select_timeout does not
look all that bad either. But maybe foo_lock is making it clearer
that a foo_unlock is needed, if you compared it to foo_select and
foo_unselect? I'm probably not the best person to make the call,
as I know all to well what to expect from the functions...

Cheers,
peda

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Philipp Zabel April 19, 2017, 1:49 p.m.
On Wed, 2017-04-19 at 14:00 +0200, Peter Rosin wrote:
[...]
> >> +int mux_control_select(struct mux_control *mux, int state)
> > 
> > If we let two of these race, ...
>
> The window for this "race" is positively huge. If there are several
> mux consumers of a single mux controller, it is self-evident that
> if one of them grabs the mux for a long time, the others will suffer.
> 
> The design is that the rwsem is reader-locked for the full duration
> of a select/deselect operation by the mux consumer.

I was not clear. I meant: I think this can also happen if we let them
race with the same state target.

> >> +{
> >> +	int ret;
> >> +
> >> +	if (down_read_trylock(&mux->lock)) {
> >> +		if (mux->cached_state == state)
> >> +			return 0;

This check makes it clear that a second select call is not intended to
block if the intended state is already selected. But if the instance we
will lose the race against has not yet updated cached_state, ...

> >> +		/* Sigh, the mux needs updating... */
> >> +		up_read(&mux->lock);
> >> +	}
> >> +
> >> +	/* ...or it's just contended. */
> >> +	down_write(&mux->lock);

... we are blocking here until the other instance calls up_read. Even
though in this case (same state target) we would only have to block
until the other instance calls downgrade_write after the mux control is
set to the correct state.

Basically there is a small window before down_write with no lock at all,
where multiple instances can already have decided they must change the
mux (to the same state). If this happens, they go on to block each other
unnecessarily.

> > ... then the last to get to down_write will just wait here forever (or
> > until the first consumer calls mux_control_deselect, which may never
> > happen)?
> 
> It is vital that the mux consumer call _deselect when it is done with
> the mux. Not doing so will surely starve out any other mux consumers.
> The whole thing is designed around the fact that mux consumers should
> deselect the mux as soon as it's no longer needed.

I'd like to use this for video bus multiplexers. Those would not be
selected for the duration of an i2c transfer, but for the duration of a
running video capture stream, or for the duration of an enabled display
path. While I currently have no use case for multiple consumers
controlling the same mux, this scenario is what shapes my perspective.
For such long running selections the consumer should have the option to
return -EBUSY instead of blocking when the lock can't be taken.

regards
Philipp

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Peter Rosin April 19, 2017, 9:04 p.m.
On 2017-04-19 15:49, Philipp Zabel wrote:
> On Wed, 2017-04-19 at 14:00 +0200, Peter Rosin wrote:
> [...]
>>>> +int mux_control_select(struct mux_control *mux, int state)
>>>
>>> If we let two of these race, ...
>>
>> The window for this "race" is positively huge. If there are several
>> mux consumers of a single mux controller, it is self-evident that
>> if one of them grabs the mux for a long time, the others will suffer.
>>
>> The design is that the rwsem is reader-locked for the full duration
>> of a select/deselect operation by the mux consumer.
> 
> I was not clear. I meant: I think this can also happen if we let them
> race with the same state target.

Right, but there is another glaring problem with the v13 implementation
of select/deselect. If there are three consumers and the first one
holds the mux while the other two tries to select it to some other
position, then even if the two "new" consumers agree on the mux state,
then both of them will end up in the "it's just contended" case and
then be serialized. So, yes, the select/deselect implementation is not
perfect. To quote the cover letter:

	I'm using an rwsem to lock a mux, but that isn't really a
	perfect fit. Is there a better locking primitive that I don't
	know about that fits better? I had a mutex at one point, but
	that didn't allow any concurrent accesses at all. At least
	the rwsem allows concurrent access as long as all users
	agree on the mux state, but I suspect that the rwsem will
	degrade to the mutex situation pretty quickly if there is
	any contention.

But with your discovery that there's a race when two consumers go
for the same state in a free mux, in addition to the above contention
problem, I'm about ready to go with Greg's suggestion and just use a
mutex. I.e. ignore the desire to allow concurrent use. Because it's
not like the sketchy thing I threw out in the other part of the thread
solves any of these problems. I can live without the concurrency, and
I guess I can also live with passing the buck to the poor sod that
eventually needs it.

>>>> +{
>>>> +	int ret;
>>>> +
>>>> +	if (down_read_trylock(&mux->lock)) {
>>>> +		if (mux->cached_state == state)
>>>> +			return 0;
> 
> This check makes it clear that a second select call is not intended to
> block if the intended state is already selected. But if the instance we
> will lose the race against has not yet updated cached_state, ...
> 
>>>> +		/* Sigh, the mux needs updating... */
>>>> +		up_read(&mux->lock);
>>>> +	}
>>>> +
>>>> +	/* ...or it's just contended. */
>>>> +	down_write(&mux->lock);
> 
> ... we are blocking here until the other instance calls up_read. Even
> though in this case (same state target) we would only have to block
> until the other instance calls downgrade_write after the mux control is
> set to the correct state.
> 
> Basically there is a small window before down_write with no lock at all,
> where multiple instances can already have decided they must change the
> mux (to the same state). If this happens, they go on to block each other
> unnecessarily.
> 
>>> ... then the last to get to down_write will just wait here forever (or
>>> until the first consumer calls mux_control_deselect, which may never
>>> happen)?
>>
>> It is vital that the mux consumer call _deselect when it is done with
>> the mux. Not doing so will surely starve out any other mux consumers.
>> The whole thing is designed around the fact that mux consumers should
>> deselect the mux as soon as it's no longer needed.
> 
> I'd like to use this for video bus multiplexers. Those would not be
> selected for the duration of an i2c transfer, but for the duration of a
> running video capture stream, or for the duration of an enabled display
> path. While I currently have no use case for multiple consumers
> controlling the same mux, this scenario is what shapes my perspective.
> For such long running selections the consumer should have the option to
> return -EBUSY instead of blocking when the lock can't be taken.

I'll see if I can implement a try variant for the mutex based select I
will probably end up with in v14...

Cheers,
peda
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Peter Rosin April 20, 2017, 9:53 p.m.
On 2017-04-18 23:53, Peter Rosin wrote:
> On 2017-04-18 13:44, Greg Kroah-Hartman wrote:
>> On Tue, Apr 18, 2017 at 12:59:50PM +0200, Peter Rosin wrote:
>>> On 2017-04-18 10:51, Greg Kroah-Hartman wrote:
>>>> On Thu, Apr 13, 2017 at 06:43:07PM +0200, Peter Rosin wrote:

*snip*

>>>>> +	if (mux->idle_state != MUX_IDLE_AS_IS &&
>>>>> +	    mux->idle_state != mux->cached_state)
>>>>> +		ret = mux_control_set(mux, mux->idle_state);
>>>>> +
>>>>> +	up_read(&mux->lock);
>>>>
>>>> You require a lock to be held for a "global" function?  Without
>>>> documentation?  Or even a sparse marking?  That's asking for trouble...
>>>
>>> Documentation I can handle, but where should I look to understand how I
>>> should add sparse markings?
>>
>> Run sparse on the code and see what it says :)
> 
> Will do.

I just did, and even went through the trouble of getting the bleeding
edge sparse from the git repo when sparse 0.5.0 came up empty, but it's
all silence for me. So, how do I add sparse markings?

Cheers,
peda

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Philipp Zabel April 21, 2017, 2:18 p.m.
Hi Peter,

On Thu, 2017-04-13 at 18:43 +0200, Peter Rosin wrote:
[...]
> +int mux_control_select(struct mux_control *mux, int state)

state could be unsigned int for the consumer facing API.

> +{
> +	int ret;

And mux_control_select should check that (0 <= state < mux->states).

regards
Philipp


--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Philipp Zabel April 21, 2017, 2:23 p.m.
On Thu, 2017-04-13 at 18:43 +0200, Peter Rosin wrote:
[...]
> +int mux_chip_register(struct mux_chip *mux_chip)
> +{
> +	int i;
> +	int ret;
> +
> +	for (i = 0; i < mux_chip->controllers; ++i) {
> +		struct mux_control *mux = &mux_chip->mux[i];
> +
> +		if (mux->idle_state == mux->cached_state)
> +			continue;

I think this should be changed to
 
-               if (mux->idle_state == mux->cached_state)
+               if (mux->idle_state == mux->cached_state ||
+                   mux->idle_state == MUX_IDLE_AS_IS)
                        continue;

or the following mux_control_set will be called with state ==
MUX_IDLE_AS_IS. Alternatively, mux_control_set should return when passed
this value.

> +		ret = mux_control_set(mux, mux->idle_state);
> +		if (ret < 0) {
> +			dev_err(&mux_chip->dev, "unable to set idle state\n");
> +			return ret;
> +		}
> +	}
> +
> +	ret = device_add(&mux_chip->dev);
> +	if (ret < 0)
> +		dev_err(&mux_chip->dev,
> +			"device_add failed in mux_chip_register: %d\n", ret);
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(mux_chip_register);

regards
Philipp

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Peter Rosin April 21, 2017, 2:32 p.m.
On 2017-04-21 16:23, Philipp Zabel wrote:
> On Thu, 2017-04-13 at 18:43 +0200, Peter Rosin wrote:
> [...]
>> +int mux_chip_register(struct mux_chip *mux_chip)
>> +{
>> +	int i;
>> +	int ret;
>> +
>> +	for (i = 0; i < mux_chip->controllers; ++i) {
>> +		struct mux_control *mux = &mux_chip->mux[i];
>> +
>> +		if (mux->idle_state == mux->cached_state)
>> +			continue;
> 
> I think this should be changed to
>  
> -               if (mux->idle_state == mux->cached_state)
> +               if (mux->idle_state == mux->cached_state ||
> +                   mux->idle_state == MUX_IDLE_AS_IS)
>                         continue;
> 
> or the following mux_control_set will be called with state ==
> MUX_IDLE_AS_IS. Alternatively, mux_control_set should return when passed
> this value.

That cannot happen because ->cached_state is initialized to -1
in mux_chip_alloc, so should always be == MUX_IDLE_AS_IS when
registering. And drivers are not supposed to touch ->cached_state.
I.e., ->cached_state is "owned" by the core.

Cheers,
peda

>> +		ret = mux_control_set(mux, mux->idle_state);
>> +		if (ret < 0) {
>> +			dev_err(&mux_chip->dev, "unable to set idle state\n");
>> +			return ret;
>> +		}
>> +	}
>> +
>> +	ret = device_add(&mux_chip->dev);
>> +	if (ret < 0)
>> +		dev_err(&mux_chip->dev,
>> +			"device_add failed in mux_chip_register: %d\n", ret);
>> +	return ret;
>> +}
>> +EXPORT_SYMBOL_GPL(mux_chip_register);
> 
> regards
> Philipp
> 

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Philipp Zabel April 21, 2017, 2:41 p.m.
On Fri, 2017-04-21 at 16:32 +0200, Peter Rosin wrote:
> On 2017-04-21 16:23, Philipp Zabel wrote:
> > On Thu, 2017-04-13 at 18:43 +0200, Peter Rosin wrote:
> > [...]
> >> +int mux_chip_register(struct mux_chip *mux_chip)
> >> +{
> >> +	int i;
> >> +	int ret;
> >> +
> >> +	for (i = 0; i < mux_chip->controllers; ++i) {
> >> +		struct mux_control *mux = &mux_chip->mux[i];
> >> +
> >> +		if (mux->idle_state == mux->cached_state)
> >> +			continue;
> > 
> > I think this should be changed to
> >  
> > -               if (mux->idle_state == mux->cached_state)
> > +               if (mux->idle_state == mux->cached_state ||
> > +                   mux->idle_state == MUX_IDLE_AS_IS)
> >                         continue;
> > 
> > or the following mux_control_set will be called with state ==
> > MUX_IDLE_AS_IS. Alternatively, mux_control_set should return when passed
> > this value.
> 
> That cannot happen because ->cached_state is initialized to -1
> in mux_chip_alloc, so should always be == MUX_IDLE_AS_IS when
> registering. And drivers are not supposed to touch ->cached_state.
> I.e., ->cached_state is "owned" by the core.

So this was caused by me filling cached_state from register reads in the
mmio driver. Makes me wonder why I am not allowed to do this, though, if
I am able to read back the initial state?

regards
Philipp

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Peter Rosin April 21, 2017, 2:55 p.m.
On 2017-04-21 16:41, Philipp Zabel wrote:
> On Fri, 2017-04-21 at 16:32 +0200, Peter Rosin wrote:
>> On 2017-04-21 16:23, Philipp Zabel wrote:
>>> On Thu, 2017-04-13 at 18:43 +0200, Peter Rosin wrote:
>>> [...]
>>>> +int mux_chip_register(struct mux_chip *mux_chip)
>>>> +{
>>>> +	int i;
>>>> +	int ret;
>>>> +
>>>> +	for (i = 0; i < mux_chip->controllers; ++i) {
>>>> +		struct mux_control *mux = &mux_chip->mux[i];
>>>> +
>>>> +		if (mux->idle_state == mux->cached_state)
>>>> +			continue;
>>>
>>> I think this should be changed to
>>>  
>>> -               if (mux->idle_state == mux->cached_state)
>>> +               if (mux->idle_state == mux->cached_state ||
>>> +                   mux->idle_state == MUX_IDLE_AS_IS)
>>>                         continue;
>>>
>>> or the following mux_control_set will be called with state ==
>>> MUX_IDLE_AS_IS. Alternatively, mux_control_set should return when passed
>>> this value.
>>
>> That cannot happen because ->cached_state is initialized to -1
>> in mux_chip_alloc, so should always be == MUX_IDLE_AS_IS when
>> registering. And drivers are not supposed to touch ->cached_state.
>> I.e., ->cached_state is "owned" by the core.
> 
> So this was caused by me filling cached_state from register reads in the
> mmio driver. Makes me wonder why I am not allowed to do this, though, if
> I am able to read back the initial state?

You gain fairly little by reading back the original state. If the mux
should idle-as-is, you can avoid a maximum of one mux update if the first
consumer happens to starts by requesting the previously active state.
Similarly, if the mux should idle in a specific state, you can avoid a
maximum of one mux update.

In both cases it costs one unconditional read of the mux state.

Sure, in some cases reads are cheaper than writes, but I didn't think
support for seeding the cache was worth it. Is it worth it?

Cheers,
peda

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Peter Rosin April 21, 2017, 3:08 p.m.
On 2017-04-21 16:18, Philipp Zabel wrote:
> Hi Peter,
> 
> On Thu, 2017-04-13 at 18:43 +0200, Peter Rosin wrote:
> [...]
>> +int mux_control_select(struct mux_control *mux, int state)
> 
> state could be unsigned int for the consumer facing API.
> 
>> +{
>> +	int ret;
> 
> And mux_control_select should check that (0 <= state < mux->states).

Yes, that makes sense. I worried that we might end up with
signed/unsigned comparisons since the internal state still needs
to be signed, but that didn't happen when I tried...

I'll include this change in v14.

Cheers,
peda
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Philipp Zabel April 21, 2017, 3:19 p.m.
On Fri, 2017-04-21 at 16:55 +0200, Peter Rosin wrote:
> On 2017-04-21 16:41, Philipp Zabel wrote:
> > On Fri, 2017-04-21 at 16:32 +0200, Peter Rosin wrote:
> >> On 2017-04-21 16:23, Philipp Zabel wrote:
> >>> On Thu, 2017-04-13 at 18:43 +0200, Peter Rosin wrote:
> >>> [...]
> >>>> +int mux_chip_register(struct mux_chip *mux_chip)
> >>>> +{
> >>>> +	int i;
> >>>> +	int ret;
> >>>> +
> >>>> +	for (i = 0; i < mux_chip->controllers; ++i) {
> >>>> +		struct mux_control *mux = &mux_chip->mux[i];
> >>>> +
> >>>> +		if (mux->idle_state == mux->cached_state)
> >>>> +			continue;
> >>>
> >>> I think this should be changed to
> >>>  
> >>> -               if (mux->idle_state == mux->cached_state)
> >>> +               if (mux->idle_state == mux->cached_state ||
> >>> +                   mux->idle_state == MUX_IDLE_AS_IS)
> >>>                         continue;
> >>>
> >>> or the following mux_control_set will be called with state ==
> >>> MUX_IDLE_AS_IS. Alternatively, mux_control_set should return when passed
> >>> this value.
> >>
> >> That cannot happen because ->cached_state is initialized to -1
> >> in mux_chip_alloc, so should always be == MUX_IDLE_AS_IS when
> >> registering. And drivers are not supposed to touch ->cached_state.
> >> I.e., ->cached_state is "owned" by the core.
> > 
> > So this was caused by me filling cached_state from register reads in the
> > mmio driver. Makes me wonder why I am not allowed to do this, though, if
> > I am able to read back the initial state?
> 
> You gain fairly little by reading back the original state. If the mux
> should idle-as-is, you can avoid a maximum of one mux update if the first
> consumer happens to starts by requesting the previously active state.
> Similarly, if the mux should idle in a specific state, you can avoid a
> maximum of one mux update.
> 
> In both cases it costs one unconditional read of the mux state.
> 
> Sure, in some cases reads are cheaper than writes, but I didn't think
> support for seeding the cache was worth it. Is it worth it?

Probably not, I'll just drop the cached_state initialization. It should
be documented in the mux.h that this field is framework internal and not
to be touched by the drivers. At least I was surprised.

regards
Philipp

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Peter Rosin April 21, 2017, 11:28 p.m.
On 2017-04-20 23:53, Peter Rosin wrote:
> On 2017-04-18 23:53, Peter Rosin wrote:
>> On 2017-04-18 13:44, Greg Kroah-Hartman wrote:
>>> On Tue, Apr 18, 2017 at 12:59:50PM +0200, Peter Rosin wrote:
>>>> On 2017-04-18 10:51, Greg Kroah-Hartman wrote:
>>>>> On Thu, Apr 13, 2017 at 06:43:07PM +0200, Peter Rosin wrote:
> 
> *snip*
> 
>>>>>> +	if (mux->idle_state != MUX_IDLE_AS_IS &&
>>>>>> +	    mux->idle_state != mux->cached_state)
>>>>>> +		ret = mux_control_set(mux, mux->idle_state);
>>>>>> +
>>>>>> +	up_read(&mux->lock);
>>>>>
>>>>> You require a lock to be held for a "global" function?  Without
>>>>> documentation?  Or even a sparse marking?  That's asking for trouble...
>>>>
>>>> Documentation I can handle, but where should I look to understand how I
>>>> should add sparse markings?
>>>
>>> Run sparse on the code and see what it says :)
>>
>> Will do.
> 
> I just did, and even went through the trouble of getting the bleeding
> edge sparse from the git repo when sparse 0.5.0 came up empty, but it's
> all silence for me. So, how do I add sparse markings?

I looked some more into this, and the markings I find that seem related
are __acquire() and __release(). But neither mutex_lock() nor up_read()
has markings like that, so adding them when using those kinds of locks
in an imbalanced way seems like a sure way of *getting* sparse messages
about context imbalance...

So, either that, or you are talking about __must_check markings?

I feel like I'm missing something, please advise further.

Cheers,
peda

--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Peter Rosin May 5, 2017, 1:19 p.m.
On 2017-04-18 12:59, Peter Rosin wrote:
> On 2017-04-18 10:51, Greg Kroah-Hartman wrote:
>> On Thu, Apr 13, 2017 at 06:43:07PM +0200, Peter Rosin wrote:
>>> +/**
>>> + * devm_mux_chip_unregister() - Resource-managed version mux_chip_unregister().
>>> + * @dev: The device that originally registered the mux-chip.
>>> + * @mux_chip: The mux-chip to unregister.
>>> + *
>>> + * See mux_chip_unregister() for more details.
>>> + *
>>> + * Note that you do not normally need to call this function.
>>
>> Odd, then why is it exported???
> 
> You normally don't call the devm_foo_{free,release,unregister,etc} functions.
> The intention is of course that the resourse cleans up automatically. But there
> are no cases where the manual clean up is not available, at least not that I can
> find?

I had a second look and there are of course plenty of examples of missing clean up
versions for devm function. I simply hadn't looked very hard at all.

So, for v15 I intend to remove all of devm_mux_chip_unregister, devm_mux_chip_free
and devm_mux_control_put. They are all just sitting there with no callers. And the
mux-mmio/video-mux drivers by Philipp Zabel that build on top of this series don't
need them either. Besides, easy to resurrect if needed...

I will do v15 with the above, the change from mutex to semaphore for locking the
mux controller state [1] and a few small documentation improvements. That will be
rebased onto v4.12-rc1 and sent in 10 days or so, or whenever v4.12-rc1 is out.

Meanwhile, what I currently intend for v15 but based on v4.11 is available from
https://gitlab.com/peda-linux/mux.git in the "mux" branch.

Cheers,
peda

[1] https://lkml.org/lkml/2017/4/25/411
--
To unsubscribe from this list: send the line "unsubscribe linux-i2c" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch hide | download patch | download mbox

diff --git a/Documentation/driver-model/devres.txt b/Documentation/driver-model/devres.txt
index efb8200819d6..e2343d9cbec7 100644
--- a/Documentation/driver-model/devres.txt
+++ b/Documentation/driver-model/devres.txt
@@ -337,6 +337,14 @@  MEM
 MFD
   devm_mfd_add_devices()
 
+MUX
+  devm_mux_chip_alloc()
+  devm_mux_chip_free()
+  devm_mux_chip_register()
+  devm_mux_chip_unregister()
+  devm_mux_control_get()
+  devm_mux_control_put()
+
 PER-CPU MEM
   devm_alloc_percpu()
   devm_free_percpu()
diff --git a/MAINTAINERS b/MAINTAINERS
index 7fc06739c8ad..591eba737678 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -8563,6 +8563,8 @@  M:	Peter Rosin <peda@axentia.se>
 S:	Maintained
 F:	Documentation/devicetree/bindings/mux/
 F:	include/linux/dt-bindings/mux/
+F:	include/linux/mux.h
+F:	drivers/mux/
 
 MULTISOUND SOUND DRIVER
 M:	Andrew Veliath <andrewtv@usa.net>
diff --git a/drivers/Kconfig b/drivers/Kconfig
index 117ca14ccf85..a7ea13e1b869 100644
--- a/drivers/Kconfig
+++ b/drivers/Kconfig
@@ -204,4 +204,6 @@  source "drivers/fpga/Kconfig"
 
 source "drivers/fsi/Kconfig"
 
+source "drivers/mux/Kconfig"
+
 endmenu
diff --git a/drivers/Makefile b/drivers/Makefile
index 2eced9afba53..c0436f6dd5a9 100644
--- a/drivers/Makefile
+++ b/drivers/Makefile
@@ -177,3 +177,4 @@  obj-$(CONFIG_ANDROID)		+= android/
 obj-$(CONFIG_NVMEM)		+= nvmem/
 obj-$(CONFIG_FPGA)		+= fpga/
 obj-$(CONFIG_FSI)		+= fsi/
+obj-$(CONFIG_MULTIPLEXER)	+= mux/
diff --git a/drivers/mux/Kconfig b/drivers/mux/Kconfig
new file mode 100644
index 000000000000..41dfe08ead84
--- /dev/null
+++ b/drivers/mux/Kconfig
@@ -0,0 +1,34 @@ 
+#
+# Multiplexer devices
+#
+
+menuconfig MULTIPLEXER
+	tristate "Multiplexer subsystem"
+	help
+	  Multiplexer controller subsystem. Multiplexers are used in a
+	  variety of settings, and this subsystem abstracts their use
+	  so that the rest of the kernel sees a common interface. When
+	  multiple parallel multiplexers are controlled by one single
+	  multiplexer controller, this subsystem also coordinates the
+	  multiplexer accesses.
+
+	  To compile the subsystem as a module, choose M here: the module will
+	  be called mux-core.
+
+if MULTIPLEXER
+
+config MUX_GPIO
+	tristate "GPIO-controlled Multiplexer"
+	depends on OF && GPIOLIB
+	help
+	  GPIO-controlled Multiplexer controller.
+
+	  The driver builds a single multiplexer controller using a number
+	  of gpio pins. For N pins, there will be 2^N possible multiplexer
+	  states. The GPIO pins can be connected (by the hardware) to several
+	  multiplexers, which in that case will be operated in parallel.
+
+	  To compile the driver as a module, choose M here: the module will
+	  be called mux-gpio.
+
+endif
diff --git a/drivers/mux/Makefile b/drivers/mux/Makefile
new file mode 100644
index 000000000000..bb16953f6290
--- /dev/null
+++ b/drivers/mux/Makefile
@@ -0,0 +1,6 @@ 
+#
+# Makefile for multiplexer devices.
+#
+
+obj-$(CONFIG_MULTIPLEXER)	+= mux-core.o
+obj-$(CONFIG_MUX_GPIO)		+= mux-gpio.o
diff --git a/drivers/mux/mux-core.c b/drivers/mux/mux-core.c
new file mode 100644
index 000000000000..66a8bccfc3d7
--- /dev/null
+++ b/drivers/mux/mux-core.c
@@ -0,0 +1,422 @@ 
+/*
+ * Multiplexer subsystem
+ *
+ * Copyright (C) 2017 Axentia Technologies AB
+ *
+ * Author: Peter Rosin <peda@axentia.se>
+ *
+ * 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.
+ */
+
+#define pr_fmt(fmt) "mux-core: " fmt
+
+#include <linux/device.h>
+#include <linux/err.h>
+#include <linux/export.h>
+#include <linux/idr.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/mux.h>
+#include <linux/of.h>
+#include <linux/of_platform.h>
+#include <linux/slab.h>
+
+/*
+ * The idle-as-is "state" is not an actual state that may be selected, it
+ * only implies that the state should not be changed. So, use that state
+ * as indication that the cached state of the multiplexer is unknown.
+ */
+#define MUX_CACHE_UNKNOWN MUX_IDLE_AS_IS
+
+static struct class mux_class = {
+	.name = "mux",
+	.owner = THIS_MODULE,
+};
+
+static int __init mux_init(void)
+{
+	return class_register(&mux_class);
+}
+
+static DEFINE_IDA(mux_ida);
+
+static void mux_chip_release(struct device *dev)
+{
+	struct mux_chip *mux_chip = to_mux_chip(dev);
+
+	ida_simple_remove(&mux_ida, mux_chip->id);
+	kfree(mux_chip);
+}
+
+static struct device_type mux_type = {
+	.name = "mux-chip",
+	.release = mux_chip_release,
+};
+
+struct mux_chip *mux_chip_alloc(struct device *dev,
+				unsigned int controllers, size_t sizeof_priv)
+{
+	struct mux_chip *mux_chip;
+	int i;
+
+	if (WARN_ON(!dev || !controllers))
+		return NULL;
+
+	mux_chip = kzalloc(sizeof(*mux_chip) +
+			   controllers * sizeof(*mux_chip->mux) +
+			   sizeof_priv, GFP_KERNEL);
+	if (!mux_chip)
+		return NULL;
+
+	mux_chip->mux = (struct mux_control *)(mux_chip + 1);
+	mux_chip->dev.class = &mux_class;
+	mux_chip->dev.type = &mux_type;
+	mux_chip->dev.parent = dev;
+	mux_chip->dev.of_node = dev->of_node;
+	dev_set_drvdata(&mux_chip->dev, mux_chip);
+
+	mux_chip->id = ida_simple_get(&mux_ida, 0, 0, GFP_KERNEL);
+	if (mux_chip->id < 0) {
+		pr_err("muxchipX failed to get a device id\n");
+		kfree(mux_chip);
+		return NULL;
+	}
+	dev_set_name(&mux_chip->dev, "muxchip%d", mux_chip->id);
+
+	mux_chip->controllers = controllers;
+	for (i = 0; i < controllers; ++i) {
+		struct mux_control *mux = &mux_chip->mux[i];
+
+		mux->chip = mux_chip;
+		init_rwsem(&mux->lock);
+		mux->cached_state = MUX_CACHE_UNKNOWN;
+		mux->idle_state = MUX_IDLE_AS_IS;
+	}
+
+	device_initialize(&mux_chip->dev);
+
+	return mux_chip;
+}
+EXPORT_SYMBOL_GPL(mux_chip_alloc);
+
+static int mux_control_set(struct mux_control *mux, int state)
+{
+	int ret = mux->chip->ops->set(mux, state);
+
+	mux->cached_state = ret < 0 ? MUX_CACHE_UNKNOWN : state;
+
+	return ret;
+}
+
+int mux_chip_register(struct mux_chip *mux_chip)
+{
+	int i;
+	int ret;
+
+	for (i = 0; i < mux_chip->controllers; ++i) {
+		struct mux_control *mux = &mux_chip->mux[i];
+
+		if (mux->idle_state == mux->cached_state)
+			continue;
+
+		ret = mux_control_set(mux, mux->idle_state);
+		if (ret < 0) {
+			dev_err(&mux_chip->dev, "unable to set idle state\n");
+			return ret;
+		}
+	}
+
+	ret = device_add(&mux_chip->dev);
+	if (ret < 0)
+		dev_err(&mux_chip->dev,
+			"device_add failed in mux_chip_register: %d\n", ret);
+	return ret;
+}
+EXPORT_SYMBOL_GPL(mux_chip_register);
+
+void mux_chip_unregister(struct mux_chip *mux_chip)
+{
+	device_del(&mux_chip->dev);
+}
+EXPORT_SYMBOL_GPL(mux_chip_unregister);
+
+void mux_chip_free(struct mux_chip *mux_chip)
+{
+	if (!mux_chip)
+		return;
+
+	put_device(&mux_chip->dev);
+}
+EXPORT_SYMBOL_GPL(mux_chip_free);
+
+static void devm_mux_chip_release(struct device *dev, void *res)
+{
+	struct mux_chip *mux_chip = *(struct mux_chip **)res;
+
+	mux_chip_free(mux_chip);
+}
+
+struct mux_chip *devm_mux_chip_alloc(struct device *dev,
+				     unsigned int controllers,
+				     size_t sizeof_priv)
+{
+	struct mux_chip **ptr, *mux_chip;
+
+	ptr = devres_alloc(devm_mux_chip_release, sizeof(*ptr), GFP_KERNEL);
+	if (!ptr)
+		return NULL;
+
+	mux_chip = mux_chip_alloc(dev, controllers, sizeof_priv);
+	if (!mux_chip) {
+		devres_free(ptr);
+		return NULL;
+	}
+
+	*ptr = mux_chip;
+	devres_add(dev, ptr);
+
+	return mux_chip;
+}
+EXPORT_SYMBOL_GPL(devm_mux_chip_alloc);
+
+static int devm_mux_chip_match(struct device *dev, void *res, void *data)
+{
+	struct mux_chip **r = res;
+
+	if (WARN_ON(!r || !*r))
+		return 0;
+
+	return *r == data;
+}
+
+void devm_mux_chip_free(struct device *dev, struct mux_chip *mux_chip)
+{
+	WARN_ON(devres_release(dev, devm_mux_chip_release,
+			       devm_mux_chip_match, mux_chip));
+}
+EXPORT_SYMBOL_GPL(devm_mux_chip_free);
+
+static void devm_mux_chip_reg_release(struct device *dev, void *res)
+{
+	struct mux_chip *mux_chip = *(struct mux_chip **)res;
+
+	mux_chip_unregister(mux_chip);
+}
+
+int devm_mux_chip_register(struct device *dev,
+			   struct mux_chip *mux_chip)
+{
+	struct mux_chip **ptr;
+	int res;
+
+	ptr = devres_alloc(devm_mux_chip_reg_release, sizeof(*ptr), GFP_KERNEL);
+	if (!ptr)
+		return -ENOMEM;
+
+	res = mux_chip_register(mux_chip);
+	if (res) {
+		devres_free(ptr);
+		return res;
+	}
+
+	*ptr = mux_chip;
+	devres_add(dev, ptr);
+
+	return res;
+}
+EXPORT_SYMBOL_GPL(devm_mux_chip_register);
+
+void devm_mux_chip_unregister(struct device *dev, struct mux_chip *mux_chip)
+{
+	WARN_ON(devres_release(dev, devm_mux_chip_reg_release,
+			       devm_mux_chip_match, mux_chip));
+}
+EXPORT_SYMBOL_GPL(devm_mux_chip_unregister);
+
+int mux_control_select(struct mux_control *mux, int state)
+{
+	int ret;
+
+	if (down_read_trylock(&mux->lock)) {
+		if (mux->cached_state == state)
+			return 0;
+
+		/* Sigh, the mux needs updating... */
+		up_read(&mux->lock);
+	}
+
+	/* ...or it's just contended. */
+	down_write(&mux->lock);
+
+	if (mux->cached_state == state) {
+		/*
+		 * Hmmm, someone else changed the mux to my liking.
+		 * That makes me wonder how long I waited for nothing?
+		 */
+		downgrade_write(&mux->lock);
+		return 0;
+	}
+
+	ret = mux_control_set(mux, state);
+	if (ret < 0) {
+		if (mux->idle_state != MUX_IDLE_AS_IS)
+			mux_control_set(mux, mux->idle_state);
+
+		up_write(&mux->lock);
+		return ret;
+	}
+
+	downgrade_write(&mux->lock);
+
+	return 1;
+}
+EXPORT_SYMBOL_GPL(mux_control_select);
+
+int mux_control_deselect(struct mux_control *mux)
+{
+	int ret = 0;
+
+	if (mux->idle_state != MUX_IDLE_AS_IS &&
+	    mux->idle_state != mux->cached_state)
+		ret = mux_control_set(mux, mux->idle_state);
+
+	up_read(&mux->lock);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(mux_control_deselect);
+
+static int of_dev_node_match(struct device *dev, const void *data)
+{
+	return dev->of_node == data;
+}
+
+static struct mux_chip *of_find_mux_chip_by_node(struct device_node *np)
+{
+	struct device *dev;
+
+	dev = class_find_device(&mux_class, NULL, np, of_dev_node_match);
+
+	return dev ? to_mux_chip(dev) : NULL;
+}
+
+struct mux_control *mux_control_get(struct device *dev, const char *mux_name)
+{
+	struct device_node *np = dev->of_node;
+	struct of_phandle_args args;
+	struct mux_chip *mux_chip;
+	unsigned int controller;
+	int index = 0;
+	int ret;
+
+	if (mux_name) {
+		index = of_property_match_string(np, "mux-control-names",
+						 mux_name);
+		if (index < 0) {
+			dev_err(dev, "mux controller '%s' not found\n",
+				mux_name);
+			return ERR_PTR(index);
+		}
+	}
+
+	ret = of_parse_phandle_with_args(np,
+					 "mux-controls", "#mux-control-cells",
+					 index, &args);
+	if (ret) {
+		dev_err(dev, "%s: failed to get mux-control %s(%i)\n",
+			np->full_name, mux_name ?: "", index);
+		return ERR_PTR(ret);
+	}
+
+	mux_chip = of_find_mux_chip_by_node(args.np);
+	of_node_put(args.np);
+	if (!mux_chip)
+		return ERR_PTR(-EPROBE_DEFER);
+
+	if (args.args_count > 1 ||
+	    (!args.args_count && (mux_chip->controllers > 1))) {
+		dev_err(dev, "%s: wrong #mux-control-cells for %s\n",
+			np->full_name, args.np->full_name);
+		return ERR_PTR(-EINVAL);
+	}
+
+	controller = 0;
+	if (args.args_count)
+		controller = args.args[0];
+
+	if (controller >= mux_chip->controllers) {
+		dev_err(dev, "%s: bad mux controller %u specified in %s\n",
+			np->full_name, controller, args.np->full_name);
+		return ERR_PTR(-EINVAL);
+	}
+
+	get_device(&mux_chip->dev);
+	return &mux_chip->mux[controller];
+}
+EXPORT_SYMBOL_GPL(mux_control_get);
+
+void mux_control_put(struct mux_control *mux)
+{
+	put_device(&mux->chip->dev);
+}
+EXPORT_SYMBOL_GPL(mux_control_put);
+
+static void devm_mux_control_release(struct device *dev, void *res)
+{
+	struct mux_control *mux = *(struct mux_control **)res;
+
+	mux_control_put(mux);
+}
+
+struct mux_control *devm_mux_control_get(struct device *dev,
+					 const char *mux_name)
+{
+	struct mux_control **ptr, *mux;
+
+	ptr = devres_alloc(devm_mux_control_release, sizeof(*ptr), GFP_KERNEL);
+	if (!ptr)
+		return ERR_PTR(-ENOMEM);
+
+	mux = mux_control_get(dev, mux_name);
+	if (IS_ERR(mux)) {
+		devres_free(ptr);
+		return mux;
+	}
+
+	*ptr = mux;
+	devres_add(dev, ptr);
+
+	return mux;
+}
+EXPORT_SYMBOL_GPL(devm_mux_control_get);
+
+static int devm_mux_control_match(struct device *dev, void *res, void *data)
+{
+	struct mux_control **r = res;
+
+	if (WARN_ON(!r || !*r))
+		return 0;
+
+	return *r == data;
+}
+
+void devm_mux_control_put(struct device *dev, struct mux_control *mux)
+{
+	WARN_ON(devres_release(dev, devm_mux_control_release,
+			       devm_mux_control_match, mux));
+}
+EXPORT_SYMBOL_GPL(devm_mux_control_put);
+
+/*
+ * Using subsys_initcall instead of module_init here to ensure - for the
+ * non-modular case - that the subsystem is initialized when mux consumers
+ * and mux controllers start to use it /without/ relying on link order.
+ * For the modular case, the ordering is ensured with module dependencies.
+ */
+subsys_initcall(mux_init);
+
+MODULE_DESCRIPTION("Multiplexer subsystem");
+MODULE_AUTHOR("Peter Rosin <peda@axentia.se>");
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/mux/mux-gpio.c b/drivers/mux/mux-gpio.c
new file mode 100644
index 000000000000..227d3572e6db
--- /dev/null
+++ b/drivers/mux/mux-gpio.c
@@ -0,0 +1,114 @@ 
+/*
+ * GPIO-controlled multiplexer driver
+ *
+ * Copyright (C) 2017 Axentia Technologies AB
+ *
+ * Author: Peter Rosin <peda@axentia.se>
+ *
+ * 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/err.h>
+#include <linux/gpio/consumer.h>
+#include <linux/module.h>
+#include <linux/mux.h>
+#include <linux/of_platform.h>
+#include <linux/platform_device.h>
+#include <linux/property.h>
+
+struct mux_gpio {
+	struct gpio_descs *gpios;
+	int *val;
+};
+
+static int mux_gpio_set(struct mux_control *mux, int state)
+{
+	struct mux_gpio *mux_gpio = mux_chip_priv(mux->chip);
+	int i;
+
+	for (i = 0; i < mux_gpio->gpios->ndescs; i++)
+		mux_gpio->val[i] = (state >> i) & 1;
+
+	gpiod_set_array_value_cansleep(mux_gpio->gpios->ndescs,
+				       mux_gpio->gpios->desc,
+				       mux_gpio->val);
+
+	return 0;
+}
+
+static const struct mux_control_ops mux_gpio_ops = {
+	.set = mux_gpio_set,
+};
+
+static const struct of_device_id mux_gpio_dt_ids[] = {
+	{ .compatible = "gpio-mux", },
+	{ /* sentinel */ }
+};
+MODULE_DEVICE_TABLE(of, mux_gpio_dt_ids);
+
+static int mux_gpio_probe(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+	struct mux_chip *mux_chip;
+	struct mux_gpio *mux_gpio;
+	int pins;
+	s32 idle_state;
+	int ret;
+
+	pins = gpiod_count(dev, "mux");
+	if (pins < 0)
+		return pins;
+
+	mux_chip = devm_mux_chip_alloc(dev, 1, sizeof(*mux_gpio) +
+				       pins * sizeof(*mux_gpio->val));
+	if (!mux_chip)
+		return -ENOMEM;
+
+	mux_gpio = mux_chip_priv(mux_chip);
+	mux_gpio->val = (int *)(mux_gpio + 1);
+	mux_chip->ops = &mux_gpio_ops;
+
+	mux_gpio->gpios = devm_gpiod_get_array(dev, "mux", GPIOD_OUT_LOW);
+	if (IS_ERR(mux_gpio->gpios)) {
+		ret = PTR_ERR(mux_gpio->gpios);
+		if (ret != -EPROBE_DEFER)
+			dev_err(dev, "failed to get gpios\n");
+		return ret;
+	}
+	WARN_ON(pins != mux_gpio->gpios->ndescs);
+	mux_chip->mux->states = 1 << pins;
+
+	ret = device_property_read_u32(dev, "idle-state", (u32 *)&idle_state);
+	if (ret >= 0 && idle_state != MUX_IDLE_AS_IS) {
+		if (idle_state < 0 || idle_state >= mux_chip->mux->states) {
+			dev_err(dev, "invalid idle-state %u\n", idle_state);
+			return -EINVAL;
+		}
+
+		mux_chip->mux->idle_state = idle_state;
+	}
+
+	ret = devm_mux_chip_register(dev, mux_chip);
+	if (ret < 0)
+		return ret;
+
+	dev_info(dev, "%u-way mux-controller registered\n",
+		 mux_chip->mux->states);
+
+	return 0;
+}
+
+static struct platform_driver mux_gpio_driver = {
+	.driver = {
+		.name = "gpio-mux",
+		.of_match_table	= of_match_ptr(mux_gpio_dt_ids),
+	},
+	.probe = mux_gpio_probe,
+};
+module_platform_driver(mux_gpio_driver);
+
+MODULE_DESCRIPTION("GPIO-controlled multiplexer driver");
+MODULE_AUTHOR("Peter Rosin <peda@axentia.se>");
+MODULE_LICENSE("GPL v2");
diff --git a/include/linux/mux.h b/include/linux/mux.h
new file mode 100644
index 000000000000..febdde4246df
--- /dev/null
+++ b/include/linux/mux.h
@@ -0,0 +1,252 @@ 
+/*
+ * mux.h - definitions for the multiplexer interface
+ *
+ * Copyright (C) 2017 Axentia Technologies AB
+ *
+ * Author: Peter Rosin <peda@axentia.se>
+ *
+ * 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.
+ */
+
+#ifndef _LINUX_MUX_H
+#define _LINUX_MUX_H
+
+#include <linux/device.h>
+#include <linux/rwsem.h>
+
+struct mux_chip;
+struct mux_control;
+struct platform_device;
+
+/**
+ * struct mux_control_ops -	Mux controller operations for a mux chip.
+ * @set:			Set the state of the given mux controller.
+ */
+struct mux_control_ops {
+	int (*set)(struct mux_control *mux, int state);
+};
+
+/* These defines match the constants from the dt-bindings. On purpose. */
+#define MUX_IDLE_AS_IS      (-1)
+#define MUX_IDLE_DISCONNECT (-2)
+
+/**
+ * struct mux_control -	Represents a mux controller.
+ * @lock:		Protects the mux controller state.
+ * @chip:		The mux chip that is handling this mux controller.
+ * @states:		The number of mux controller states.
+ * @cached_state:	The current mux controller state, or -1 if none.
+ * @idle_state:		The mux controller state to use when inactive, or one
+ *			of MUX_IDLE_AS_IS and MUX_IDLE_DISCONNECT.
+ */
+struct mux_control {
+	struct rw_semaphore lock; /* protects the state of the mux */
+
+	struct mux_chip *chip;
+
+	unsigned int states;
+	int cached_state;
+	int idle_state;
+};
+
+/**
+ * struct mux_chip -	Represents a chip holding mux controllers.
+ * @controllers:	Number of mux controllers handled by the chip.
+ * @mux:		Array of mux controllers that are handled.
+ * @dev:		Device structure.
+ * @id:			Used to identify the device internally.
+ * @ops:		Mux controller operations.
+ */
+struct mux_chip {
+	unsigned int controllers;
+	struct mux_control *mux;
+	struct device dev;
+	int id;
+
+	const struct mux_control_ops *ops;
+};
+
+#define to_mux_chip(x) container_of((x), struct mux_chip, dev)
+
+/**
+ * mux_chip_priv() - Get the extra memory reserved by mux_chip_alloc().
+ * @mux_chip: The mux-chip to get the private memory from.
+ *
+ * Return: Pointer to the private memory reserved by the allocator.
+ */
+static inline void *mux_chip_priv(struct mux_chip *mux_chip)
+{
+	return &mux_chip->mux[mux_chip->controllers];
+}
+
+/**
+ * mux_chip_alloc() - Allocate a mux-chip.
+ * @dev: The parent device implementing the mux interface.
+ * @controllers: The number of mux controllers to allocate for this chip.
+ * @sizeof_priv: Size of extra memory area for private use by the caller.
+ *
+ * Return: A pointer to the new mux-chip, NULL on failure.
+ */
+struct mux_chip *mux_chip_alloc(struct device *dev,
+				unsigned int controllers, size_t sizeof_priv);
+
+/**
+ * mux_chip_register() - Register a mux-chip, thus readying the controllers
+ *			 for use.
+ * @mux_chip: The mux-chip to register.
+ *
+ * Do not retry registration of the same mux-chip on failure. You should
+ * instead put it away with mux_chip_free() and allocate a new one, if you
+ * for some reason would like to retry registration.
+ *
+ * Return: Zero on success or a negative errno on error.
+ */
+int mux_chip_register(struct mux_chip *mux_chip);
+
+/**
+ * mux_chip_unregister() - Take the mux-chip off-line.
+ * @mux_chip: The mux-chip to unregister.
+ *
+ * mux_chip_unregister() reverses the effects of mux_chip_register().
+ * But not completely, you should not try to call mux_chip_register()
+ * on a mux-chip that has been registered before.
+ */
+void mux_chip_unregister(struct mux_chip *mux_chip);
+
+/**
+ * mux_chip_free() - Free the mux-chip for good.
+ * @mux_chip: The mux-chip to free.
+ *
+ * mux_chip_free() reverses the effects of mux_chip_alloc().
+ */
+void mux_chip_free(struct mux_chip *mux_chip);
+
+/**
+ * devm_mux_chip_alloc() - Resource-managed version of mux_chip_alloc().
+ * @dev: The parent device implementing the mux interface.
+ * @controllers: The number of mux controllers to allocate for this chip.
+ * @sizeof_priv: Size of extra memory area for private use by the caller.
+ *
+ * See mux_chip_alloc() for more details.
+ *
+ * Return: A pointer to the new mux-chip, NULL on failure.
+ */
+struct mux_chip *devm_mux_chip_alloc(struct device *dev,
+				     unsigned int controllers,
+				     size_t sizeof_priv);
+
+/**
+ * devm_mux_chip_register() - Resource-managed version mux_chip_register().
+ * @dev: The parent device implementing the mux interface.
+ * @mux_chip: The mux-chip to register.
+ *
+ * See mux_chip_register() for more details.
+ *
+ * Return: Zero on success or a negative errno on error.
+ */
+int devm_mux_chip_register(struct device *dev, struct mux_chip *mux_chip);
+
+/**
+ * devm_mux_chip_unregister() - Resource-managed version mux_chip_unregister().
+ * @dev: The device that originally registered the mux-chip.
+ * @mux_chip: The mux-chip to unregister.
+ *
+ * See mux_chip_unregister() for more details.
+ *
+ * Note that you do not normally need to call this function.
+ */
+void devm_mux_chip_unregister(struct device *dev, struct mux_chip *mux_chip);
+
+/**
+ * devm_mux_chip_free() - Resource-managed version mux_chip_free().
+ * @dev: The device that originally got the mux-chip.
+ * @mux_chip: The mux-chip to free.
+ *
+ * See mux_chip_free() for more details.
+ *
+ * Note that you do not normally need to call this function.
+ */
+void devm_mux_chip_free(struct device *dev, struct mux_chip *mux_chip);
+
+/**
+ * mux_control_select() - Select the given multiplexer state.
+ * @mux: The mux-control to request a change of state from.
+ * @state: The new requested state.
+ *
+ * Make sure to call mux_control_deselect() when the operation is complete and
+ * the mux-control is free for others to use, but do not call
+ * mux_control_deselect() if mux_control_select() fails.
+ *
+ * Return: 0 if the requested state was already active, or 1 it the
+ * mux-control state was changed to the requested state. Or a negative
+ * errno on error.
+ *
+ * Note that the difference in return value of zero or one is of
+ * questionable value; especially if the mux-control has several independent
+ * consumers, which is something the consumers should perhaps not be making
+ * assumptions about.
+ */
+int mux_control_select(struct mux_control *mux, int state);
+
+/**
+ * mux_control_deselect() - Deselect the previously selected multiplexer state.
+ * @mux: The mux-control to deselect.
+ *
+ * Return: 0 on success and a negative errno on error. An error can only
+ * occur if the mux has an idle state. Note that even if an error occurs, the
+ * mux-control is unlocked and is thus free for the next access.
+ */
+int mux_control_deselect(struct mux_control *mux);
+
+/**
+ * mux_control_get_index() - Get the index of the given mux controller
+ * @mux: The mux-control to get the index for.
+ *
+ * Return: The index of the mux controller within the mux chip the mux
+ * controller is a part of.
+ */
+static inline unsigned int mux_control_get_index(struct mux_control *mux)
+{
+	return mux - mux->chip->mux;
+}
+
+/**
+ * mux_control_get() - Get the mux-control for a device.
+ * @dev: The device that needs a mux-control.
+ * @mux_name: The name identifying the mux-control.
+ *
+ * Return: A pointer to the mux-control, or an ERR_PTR with a negative errno.
+ */
+struct mux_control *mux_control_get(struct device *dev, const char *mux_name);
+
+/**
+ * mux_control_put() - Put away the mux-control for good.
+ * @mux: The mux-control to put away.
+ *
+ * mux_control_put() reverses the effects of mux_control_get().
+ */
+void mux_control_put(struct mux_control *mux);
+
+/**
+ * devm_mux_control_get() - Get the mux-control for a device, with resource
+ *			    management.
+ * @dev: The device that needs a mux-control.
+ * @mux_name: The name identifying the mux-control.
+ *
+ * Return: Pointer to the mux-control, or an ERR_PTR with a negative errno.
+ */
+struct mux_control *devm_mux_control_get(struct device *dev,
+					 const char *mux_name);
+
+/**
+ * devm_mux_control_put() - Resource-managed version mux_control_put().
+ * @dev: The device that originally got the mux-control.
+ * @mux: The mux-control to put away.
+ *
+ * Note that you do not normally need to call this function.
+ */
+void devm_mux_control_put(struct device *dev, struct mux_control *mux);
+
+#endif /* _LINUX_MUX_H */