diff mbox

[v6,01/12] drivers: base: Unified device connection lookup

Message ID 20180302102057.8917-2-hdegoede@redhat.com (mailing list archive)
State Awaiting Upstream, archived
Headers show

Commit Message

Hans de Goede March 2, 2018, 10:20 a.m. UTC
From: Heikki Krogerus <heikki.krogerus@linux.intel.com>

Several frameworks - clk, gpio, phy, pmw, etc. - maintain
lookup tables for describing connections and provide custom
API for handling them. This introduces a single generic
lookup table and API for the connections.

The motivation for this commit is centralizing the
connection lookup, but the goal is to ultimately extract the
connection descriptions also from firmware by using the
fwnode_graph_* functions and other mechanisms that are
available.

Signed-off-by: Heikki Krogerus <heikki.krogerus@linux.intel.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
---
Changes in v6:
-Fix data and match arguments being swapped in __device_find_connection()
 call in device_find_connection() (as noticed by Jun Li)

Changes in v5:
-Add missing documentation for @list struct devcon member

Changes in v4:
-Add Andy's Reviewed-by

Changes in v3:
-Various spelling and gramar fixes in the docs pointed out by Randy Dunlap

Changes in v2:
-Add a (struct devcon) cast to the DEVCON() macro
---
 Documentation/driver-api/device_connection.rst |  43 ++++++++
 drivers/base/Makefile                          |   3 +-
 drivers/base/devcon.c                          | 139 +++++++++++++++++++++++++
 include/linux/connection.h                     |  34 ++++++
 4 files changed, 218 insertions(+), 1 deletion(-)
 create mode 100644 Documentation/driver-api/device_connection.rst
 create mode 100644 drivers/base/devcon.c
 create mode 100644 include/linux/connection.h

Comments

Greg Kroah-Hartman March 9, 2018, 5:53 p.m. UTC | #1
On Fri, Mar 02, 2018 at 11:20:46AM +0100, Hans de Goede wrote:
> From: Heikki Krogerus <heikki.krogerus@linux.intel.com>
> 
> Several frameworks - clk, gpio, phy, pmw, etc. - maintain
> lookup tables for describing connections and provide custom
> API for handling them. This introduces a single generic
> lookup table and API for the connections.
> 
> The motivation for this commit is centralizing the
> connection lookup, but the goal is to ultimately extract the
> connection descriptions also from firmware by using the
> fwnode_graph_* functions and other mechanisms that are
> available.
> 
> Signed-off-by: Heikki Krogerus <heikki.krogerus@linux.intel.com>
> Reviewed-by: Hans de Goede <hdegoede@redhat.com>
> Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
> Signed-off-by: Hans de Goede <hdegoede@redhat.com>

Sorry for the delay, just now reviewing this patch...

The content is fine (if not scary for the obvious reason of passing
around 'struct device' of different bus types, but ok...), but the api
naming is "rough":

> --- /dev/null
> +++ b/include/linux/connection.h

"connection.h" is vague, why not just put this in device.h?

> @@ -0,0 +1,34 @@
> +// SPDX-License-Identifier: GPL-2.0
> +
> +#ifndef _LINUX_CONNECTION_H_
> +#define _LINUX_CONNECTION_H_
> +
> +#include <linux/list.h>
> +
> +struct device;
> +
> +/**
> + * struct devcon - Device Connection Descriptor
> + * @endpoint: The names of the two devices connected together
> + * @id: Unique identifier for the connection
> + * @list: List head, private for devcon internal use only
> + */
> +struct devcon {

'struct device_connection'?  Spell it out please, people might think
this is a "developer conference" :)

> +	const char		*endpoint[2];
> +	const char		*id;
> +	struct list_head	list;
> +};
> +
> +void *__device_find_connection(struct device *dev, const char *con_id,
> +			       void *data,
> +			       void *(*match)(struct devcon *con, int ep,
> +					      void *data));

Ick, __* functions are usually "no lock needed", but here you are doing
a lot "more" than the normal device_find_connection() call.  Why not
make this:
	device_connection_find_match()?

> +
> +struct device *device_find_connection(struct device *dev, const char *con_id);

device_connection_find()?

> +
> +#define DEVCON(_ep0, _ep1, _id)    (struct devcon) { { _ep0, _ep1 }, _id, }

Can you use named identifiers here?

> +
> +void add_device_connection(struct devcon *con);
> +void remove_device_connection(struct devcon *con);

device_connection_add() and device_connection_remove()?

I can make the api name changes in an add-on patch.

thanks,

greg k-h
Hans de Goede March 11, 2018, 6:24 p.m. UTC | #2
Hi all,

On 09-03-18 18:53, Greg Kroah-Hartman wrote:
> On Fri, Mar 02, 2018 at 11:20:46AM +0100, Hans de Goede wrote:
>> From: Heikki Krogerus <heikki.krogerus@linux.intel.com>
>>
>> Several frameworks - clk, gpio, phy, pmw, etc. - maintain
>> lookup tables for describing connections and provide custom
>> API for handling them. This introduces a single generic
>> lookup table and API for the connections.
>>
>> The motivation for this commit is centralizing the
>> connection lookup, but the goal is to ultimately extract the
>> connection descriptions also from firmware by using the
>> fwnode_graph_* functions and other mechanisms that are
>> available.
>>
>> Signed-off-by: Heikki Krogerus <heikki.krogerus@linux.intel.com>
>> Reviewed-by: Hans de Goede <hdegoede@redhat.com>
>> Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
>> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
> 
> Sorry for the delay, just now reviewing this patch...
> 
> The content is fine (if not scary for the obvious reason of passing
> around 'struct device' of different bus types, but ok...), but the api
> naming is "rough":


Heikki, I think it is best if you answer Greg's remarks. FWIW I'm
fine with the changes Greg proposes.

I currently have significantly less bandwidth for this due to
personal circumstances, so if a new version of this patch-set
is necessary it would be great if you (Heikki) can do a v7.

Regards,

Hans

> 
>> --- /dev/null
>> +++ b/include/linux/connection.h
> 
> "connection.h" is vague, why not just put this in device.h? >
>> @@ -0,0 +1,34 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +
>> +#ifndef _LINUX_CONNECTION_H_
>> +#define _LINUX_CONNECTION_H_
>> +
>> +#include <linux/list.h>
>> +
>> +struct device;
>> +
>> +/**
>> + * struct devcon - Device Connection Descriptor
>> + * @endpoint: The names of the two devices connected together
>> + * @id: Unique identifier for the connection
>> + * @list: List head, private for devcon internal use only
>> + */
>> +struct devcon {
> 
> 'struct device_connection'?  Spell it out please, people might think
> this is a "developer conference" :)
> 
>> +	const char		*endpoint[2];
>> +	const char		*id;
>> +	struct list_head	list;
>> +};
>> +
>> +void *__device_find_connection(struct device *dev, const char *con_id,
>> +			       void *data,
>> +			       void *(*match)(struct devcon *con, int ep,
>> +					      void *data));
> 
> Ick, __* functions are usually "no lock needed", but here you are doing
> a lot "more" than the normal device_find_connection() call.  Why not
> make this:
> 	device_connection_find_match()?
> 
>> +
>> +struct device *device_find_connection(struct device *dev, const char *con_id);
> 
> device_connection_find()?
> 
>> +
>> +#define DEVCON(_ep0, _ep1, _id)    (struct devcon) { { _ep0, _ep1 }, _id, }
> 
> Can you use named identifiers here?
> 
>> +
>> +void add_device_connection(struct devcon *con);
>> +void remove_device_connection(struct devcon *con);
> 
> device_connection_add() and device_connection_remove()?
> 
> I can make the api name changes in an add-on patch.
> 
> thanks,
> 
> greg k-h
>
Heikki Krogerus March 12, 2018, 8:13 a.m. UTC | #3
On Sun, Mar 11, 2018 at 07:24:18PM +0100, Hans de Goede wrote:
> Hi all,
> 
> On 09-03-18 18:53, Greg Kroah-Hartman wrote:
> > On Fri, Mar 02, 2018 at 11:20:46AM +0100, Hans de Goede wrote:
> > > From: Heikki Krogerus <heikki.krogerus@linux.intel.com>
> > > 
> > > Several frameworks - clk, gpio, phy, pmw, etc. - maintain
> > > lookup tables for describing connections and provide custom
> > > API for handling them. This introduces a single generic
> > > lookup table and API for the connections.
> > > 
> > > The motivation for this commit is centralizing the
> > > connection lookup, but the goal is to ultimately extract the
> > > connection descriptions also from firmware by using the
> > > fwnode_graph_* functions and other mechanisms that are
> > > available.
> > > 
> > > Signed-off-by: Heikki Krogerus <heikki.krogerus@linux.intel.com>
> > > Reviewed-by: Hans de Goede <hdegoede@redhat.com>
> > > Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
> > > Signed-off-by: Hans de Goede <hdegoede@redhat.com>
> > 
> > Sorry for the delay, just now reviewing this patch...
> > 
> > The content is fine (if not scary for the obvious reason of passing
> > around 'struct device' of different bus types, but ok...), but the api
> > naming is "rough":
> 
> Heikki, I think it is best if you answer Greg's remarks. FWIW I'm
> fine with the changes Greg proposes.

Me too.

> I currently have significantly less bandwidth for this due to
> personal circumstances, so if a new version of this patch-set
> is necessary it would be great if you (Heikki) can do a v7.

Sure thing. I'll prepare the v7.


Thanks,
Hans de Goede March 12, 2018, 9:19 a.m. UTC | #4
Hi,

On 12-03-18 09:13, Heikki Krogerus wrote:
> On Sun, Mar 11, 2018 at 07:24:18PM +0100, Hans de Goede wrote:
>> Hi all,
>>
>> On 09-03-18 18:53, Greg Kroah-Hartman wrote:
>>> On Fri, Mar 02, 2018 at 11:20:46AM +0100, Hans de Goede wrote:
>>>> From: Heikki Krogerus <heikki.krogerus@linux.intel.com>
>>>>
>>>> Several frameworks - clk, gpio, phy, pmw, etc. - maintain
>>>> lookup tables for describing connections and provide custom
>>>> API for handling them. This introduces a single generic
>>>> lookup table and API for the connections.
>>>>
>>>> The motivation for this commit is centralizing the
>>>> connection lookup, but the goal is to ultimately extract the
>>>> connection descriptions also from firmware by using the
>>>> fwnode_graph_* functions and other mechanisms that are
>>>> available.
>>>>
>>>> Signed-off-by: Heikki Krogerus <heikki.krogerus@linux.intel.com>
>>>> Reviewed-by: Hans de Goede <hdegoede@redhat.com>
>>>> Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
>>>> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
>>>
>>> Sorry for the delay, just now reviewing this patch...
>>>
>>> The content is fine (if not scary for the obvious reason of passing
>>> around 'struct device' of different bus types, but ok...), but the api
>>> naming is "rough":
>>
>> Heikki, I think it is best if you answer Greg's remarks. FWIW I'm
>> fine with the changes Greg proposes.
> 
> Me too.
> 
>> I currently have significantly less bandwidth for this due to
>> personal circumstances, so if a new version of this patch-set
>> is necessary it would be great if you (Heikki) can do a v7.
> 
> Sure thing. I'll prepare the v7.

Great, thank you. Don't forget to squash in the 2 fixes from
fengguang.wu@intel.com .

Regards,

Hans
Heikki Krogerus March 12, 2018, 9:24 a.m. UTC | #5
On Mon, Mar 12, 2018 at 10:19:34AM +0100, Hans de Goede wrote:
> > > I currently have significantly less bandwidth for this due to
> > > personal circumstances, so if a new version of this patch-set
> > > is necessary it would be great if you (Heikki) can do a v7.
> > 
> > Sure thing. I'll prepare the v7.
> 
> Great, thank you. Don't forget to squash in the 2 fixes from
> fengguang.wu@intel.com .

Ah, from build-bot!

Good that you reminded, I would not have remembered those :-)


Thanks,
diff mbox

Patch

diff --git a/Documentation/driver-api/device_connection.rst b/Documentation/driver-api/device_connection.rst
new file mode 100644
index 000000000000..64a3e5e9bb7c
--- /dev/null
+++ b/Documentation/driver-api/device_connection.rst
@@ -0,0 +1,43 @@ 
+==================
+Device connections
+==================
+
+Introduction
+------------
+
+Devices often have connections to other devices that are outside of the direct
+child/parent relationship. A serial or network communication controller, which
+could be a PCI device, may need to be able to get a reference to its PHY
+component, which could be attached for example to the I2C bus. Some device
+drivers need to be able to control the clocks or the GPIOs for their devices,
+and so on.
+
+Device connections are generic descriptions of any type of connection between
+two separate devices.
+
+Device connections alone do not create a dependency between the two devices.
+They are only descriptions which are not tied to either of the devices directly.
+A dependency between the two devices exists only if one of the two endpoint
+devices requests a reference to the other. The descriptions themselves can be
+defined in firmware (not yet supported) or they can be built-in.
+
+Usage
+-----
+
+Device connections should exist before device ``->probe`` callback is called for
+either endpoint device in the description. If the connections are defined in
+firmware, this is not a problem. It should be considered if the connection
+descriptions are "built-in", and need to be added separately.
+
+The connection description consists of the names of the two devices with the
+connection, i.e. the endpoints, and unique identifier for the connection which
+is needed if there are multiple connections between the two devices.
+
+After a description exists, the devices in it can request reference to the other
+endpoint device, or they can request the description itself.
+
+API
+---
+
+.. kernel-doc:: drivers/base/devcon.c
+   : functions: __device_find_connection device_find_connection add_device_connection remove_device_connection
diff --git a/drivers/base/Makefile b/drivers/base/Makefile
index e32a52490051..12a7f64d35a9 100644
--- a/drivers/base/Makefile
+++ b/drivers/base/Makefile
@@ -5,7 +5,8 @@  obj-y			:= component.o core.o bus.o dd.o syscore.o \
 			   driver.o class.o platform.o \
 			   cpu.o firmware.o init.o map.o devres.o \
 			   attribute_container.o transport_class.o \
-			   topology.o container.o property.o cacheinfo.o
+			   topology.o container.o property.o cacheinfo.o \
+			   devcon.o
 obj-$(CONFIG_DEVTMPFS)	+= devtmpfs.o
 obj-$(CONFIG_DMA_CMA) += dma-contiguous.o
 obj-y			+= power/
diff --git a/drivers/base/devcon.c b/drivers/base/devcon.c
new file mode 100644
index 000000000000..9f13092fa115
--- /dev/null
+++ b/drivers/base/devcon.c
@@ -0,0 +1,139 @@ 
+// SPDX-License-Identifier: GPL-2.0
+/**
+ * Device connections
+ *
+ * Copyright (C) 2018 Intel Corporation
+ * Author: Heikki Krogerus <heikki.krogerus@linux.intel.com>
+ */
+
+#include <linux/connection.h>
+#include <linux/device.h>
+
+static DEFINE_MUTEX(devcon_lock);
+static LIST_HEAD(devcon_list);
+
+/**
+ * __device_find_connection - Find physical connection to a device
+ * @dev: Device with the connection
+ * @con_id: Identifier for the connection
+ * @data: Data for the match function
+ * @match: Function to check and convert the connection description
+ *
+ * Find a connection with unique identifier @con_id between @dev and another
+ * device. @match will be used to convert the connection description to data the
+ * caller is expecting to be returned.
+ */
+void *__device_find_connection(struct device *dev, const char *con_id,
+			       void *data,
+			       void *(*match)(struct devcon *con, int ep,
+					      void *data))
+{
+	const char *devname = dev_name(dev);
+	struct devcon *con;
+	void *ret = NULL;
+	int ep;
+
+	if (!match)
+		return NULL;
+
+	rcu_read_lock();
+
+	list_for_each_entry_rcu(con, &devcon_list, list) {
+		ep = match_string(con->endpoint, 2, devname);
+		if (ep < 0)
+			continue;
+
+		if (con_id && strcmp(con->id, con_id))
+			continue;
+
+		ret = match(con, !ep, data);
+		if (ret)
+			break;
+	}
+
+	rcu_read_unlock();
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(__device_find_connection);
+
+#include <linux/platform_device.h>
+#include <linux/spi/spi.h>
+#include <linux/i2c.h>
+#include <linux/pci.h>
+
+static struct bus_type *generic_match_buses[] = {
+	&platform_bus_type,
+#ifdef CONFIG_PCI
+	&pci_bus_type,
+#endif
+#ifdef CONFIG_I2C
+	&i2c_bus_type,
+#endif
+#ifdef CONFIG_SPI_MASTER
+	&spi_bus_type,
+#endif
+	NULL,
+};
+
+/* This tries to find the device from the most common bus types by name. */
+static void *generic_match(struct devcon *con, int ep, void *data)
+{
+	struct bus_type *bus;
+	struct device *dev;
+
+	for (bus = generic_match_buses[0]; bus; bus++) {
+		dev = bus_find_device_by_name(bus, NULL, con->endpoint[ep]);
+		if (dev)
+			return dev;
+	}
+
+	/*
+	 * We only get called if a connection was found, tell the caller to
+	 * wait for the other device to show up.
+	 */
+	return ERR_PTR(-EPROBE_DEFER);
+}
+
+/**
+ * device_find_connection - Find two devices connected together
+ * @dev: Device with the connection
+ * @con_id: Identifier for the connection
+ *
+ * Find a connection with unique identifier @con_id between @dev and
+ * another device. On success returns handle to the device that is connected
+ * to @dev, with the reference count for the found device incremented. Returns
+ * NULL if no matching connection was found, or ERR_PTR(-EPROBE_DEFER) when a
+ * connection was found but the other device has not been enumerated yet.
+ */
+struct device *device_find_connection(struct device *dev, const char *con_id)
+{
+	return __device_find_connection(dev, con_id, NULL, generic_match);
+}
+EXPORT_SYMBOL_GPL(device_find_connection);
+
+/**
+ * add_device_connection - Register a collection of connection descriptions
+ * @con: Collection of connection descriptions to be registered
+ */
+void add_device_connection(struct devcon *con)
+{
+	mutex_lock(&devcon_lock);
+	list_add_tail_rcu(&con->list, &devcon_list);
+	mutex_unlock(&devcon_lock);
+}
+EXPORT_SYMBOL_GPL(add_device_connection);
+
+/**
+ * remove_device_connections - Unregister collection of connection descriptions
+ * @con: Collection of connection descriptions to be unregistered
+ */
+void remove_device_connection(struct devcon *con)
+{
+	mutex_lock(&devcon_lock);
+	list_del_rcu(&con->list);
+	mutex_unlock(&devcon_lock);
+	/* The caller may free the devcon struct immediately afterwards. */
+	synchronize_rcu();
+}
+EXPORT_SYMBOL_GPL(remove_device_connection);
diff --git a/include/linux/connection.h b/include/linux/connection.h
new file mode 100644
index 000000000000..b2af70df5a97
--- /dev/null
+++ b/include/linux/connection.h
@@ -0,0 +1,34 @@ 
+// SPDX-License-Identifier: GPL-2.0
+
+#ifndef _LINUX_CONNECTION_H_
+#define _LINUX_CONNECTION_H_
+
+#include <linux/list.h>
+
+struct device;
+
+/**
+ * struct devcon - Device Connection Descriptor
+ * @endpoint: The names of the two devices connected together
+ * @id: Unique identifier for the connection
+ * @list: List head, private for devcon internal use only
+ */
+struct devcon {
+	const char		*endpoint[2];
+	const char		*id;
+	struct list_head	list;
+};
+
+void *__device_find_connection(struct device *dev, const char *con_id,
+			       void *data,
+			       void *(*match)(struct devcon *con, int ep,
+					      void *data));
+
+struct device *device_find_connection(struct device *dev, const char *con_id);
+
+#define DEVCON(_ep0, _ep1, _id)    (struct devcon) { { _ep0, _ep1 }, _id, }
+
+void add_device_connection(struct devcon *con);
+void remove_device_connection(struct devcon *con);
+
+#endif /* _LINUX_CONNECTION_H_ */