Hi Michael, On Tue, 04 May 2010 14:46:57 +0200, Michael Lawnick wrote: > Add multiplexed bus core support. I2C multiplexer and switches > like pca954x get instantiated as new adapters per output. > > Signed-off-by: Michael Lawnick <[email protected]> > > --- > Based on kernel 2.6.33 + > [PATCH] i2c-core: Use per-adapter userspace device lists > by Jean Delware <[email protected]> > <http://article.gmane.org/gmane.linux.drivers.i2c/5994/>
Future patches should be based on 2.6.34 at least... 2.6.35-rc2 would be even better. Review follows. Need some more work I fear, but not that much... Please let me know if you are going to submit another iteration, or if you prefer to let me fix things the way I like (in which case I'll still ask you to test my version of the patch when I'm done.) > > drivers/i2c/Kconfig | 8 ++ > drivers/i2c/Makefile | 1 + > drivers/i2c/i2c-core.c | 93 ++++++++++++++++++++++-- > drivers/i2c/i2c-dev.c | 46 ++++++++++++- > drivers/i2c/i2c-mux.c | 184 > +++++++++++++++++++++++++++++++++++++++++++++++ > include/linux/i2c-mux.h | 46 ++++++++++++ > include/linux/i2c.h | 11 +++ > 7 files changed, 381 insertions(+), 8 deletions(-) > create mode 100755 drivers/i2c/i2c-mux.c > create mode 100755 include/linux/i2c-mux.h > > diff --git a/drivers/i2c/Kconfig b/drivers/i2c/Kconfig > index 8d8a00e..2cd6d78 100755 > --- a/drivers/i2c/Kconfig > +++ b/drivers/i2c/Kconfig > @@ -36,6 +36,14 @@ config I2C_COMPAT > other user-space package which expects i2c adapters to be class > devices. If you don't know, say Y. > > +config I2C_MUX > + tristate "I2C bus multiplexing support" > + depends on I2C Dependency on I2C isn't needed: this configuration item is already in an "if I2C" block. A dependency on EXPERIMENTAL might be a good idea for the time being though. > + help > + Say Y here if you want the I2C core to support the ability to > + handle multiplexed I2C bus topologies, by presenting each > + multiplexed segment as a I2C adapter. For support that can be build as a module, we usually add a sentence saying so and giving the name of the module. > + I would move config I2C_MUX below config I2C_CHARDEV, to make it consistent with Makefile, and also because users expect new options to be added after existing options. > config I2C_CHARDEV > tristate "I2C device interface" > help > diff --git a/drivers/i2c/Makefile b/drivers/i2c/Makefile > index ba26e6c..a488e8b 100755 > --- a/drivers/i2c/Makefile > +++ b/drivers/i2c/Makefile > @@ -5,6 +5,7 @@ > obj-$(CONFIG_I2C_BOARDINFO) += i2c-boardinfo.o > obj-$(CONFIG_I2C) += i2c-core.o > obj-$(CONFIG_I2C_CHARDEV) += i2c-dev.o > +obj-$(CONFIG_I2C_MUX) += i2c-mux.o > obj-y += busses/ chips/ algos/ > > ifeq ($(CONFIG_I2C_DEBUG_CORE),y) > diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c > index fc93e28..682966a 100755 > --- a/drivers/i2c/i2c-core.c > +++ b/drivers/i2c/i2c-core.c > @@ -20,7 +20,10 @@ > /* With some changes from Kyösti Mälkki <[email protected]>. > All SMBus-related things are written by Frodo Looijaard <[email protected]> > SMBus 2.0 support by Mark Studebaker <[email protected]> and > - Jean Delvare <[email protected]> */ > + Jean Delvare <[email protected]> > + Mux support by Rodolfo Giometti <[email protected]> and > + Michael Lawnick <[email protected]> */ > + Extra blank line. > > #include <linux/module.h> > #include <linux/kernel.h> > @@ -288,6 +291,12 @@ struct i2c_client *i2c_verify_client(struct device *dev) > } > EXPORT_SYMBOL(i2c_verify_client); > > +static int i2c_parent_is_i2c_adapter(const struct i2c_adapter *adapter) > +{ > + return adapter->dev.parent != NULL > + && adapter->dev.parent->bus == &i2c_bus_type; > +} > + > > /** > * i2c_new_device - instantiate an i2c device > @@ -624,6 +633,7 @@ static int i2c_register_adapter(struct i2c_adapter *adap) > > rt_mutex_init(&adap->bus_lock); > INIT_LIST_HEAD(&adap->userspace_clients); > + INIT_LIST_HEAD(&adap->mux_list_head); > > /* Set default timeout to 1 second if not already set */ > if (adap->timeout == 0) > @@ -949,9 +959,78 @@ static int __i2c_check_addr(struct device *dev, void > *addrp) > return 0; > } > > +/* walk up mux tree */ > +static int i2c_check_mux_parents(struct i2c_adapter *adapter, int addr) > +{ > + int result; > + > + result = device_for_each_child(&adapter->dev, &addr, __i2c_check_addr); > + > + if (!result && i2c_parent_is_i2c_adapter(adapter)) > + result = i2c_check_mux_parents( > + to_i2c_adapter(adapter->dev.parent), addr); > + > + return result; > +} > + > +/* recurse down mux tree */ > +static int i2c_check_mux_childs(struct i2c_adapter *adapter, int addr) "childs" doesn't exist in proper English, the right work is "children" ;) > +{ > + int result = 0; > + struct i2c_adapter *child, *next; > + > + list_for_each_entry_safe(child, next, &adapter->mux_list_head, > + mux_list) { > + result = device_for_each_child(&child->dev, &addr, > + __i2c_check_addr); > + if (result) > + break; > + result = i2c_check_mux_childs(child, addr); This value is ignored, except for the last child. I think you should break here too. > + } > + return result; > +} > + > static int i2c_check_addr(struct i2c_adapter *adapter, int addr) > { > - return device_for_each_child(&adapter->dev, &addr, __i2c_check_addr); > + int result; > + > + result = i2c_check_mux_parents(adapter, addr); > + > + if (!result) > + result = i2c_check_mux_childs(adapter, addr); > + > + return result; > +} I am worried about (the lack of) locking when the above function is called. The problem isn't new with your patch, but your patch makes it worse: beforehand, the driver core would complain if we hit a race here, because we would try to create two clients with the same bus_id as a result. But now, we may create 2 conflicting clients with different bus_id, so the driver core will not complain. I'll think about it some more, and add locking if needed. > +static void i2c_mux_tree_lock(struct i2c_adapter *adap) > +{ > + if (i2c_parent_is_i2c_adapter(adap)) > + i2c_mux_tree_lock(to_i2c_adapter(adap->dev.parent)); > + else > + i2c_lock_adapter(adap); > +} > + > +static int i2c_mux_tree_trylock(struct i2c_adapter *adap) > +{ > + int ret; > + struct i2c_adapter *parent; > + > + if (i2c_parent_is_i2c_adapter(adap)) { > + parent = to_i2c_adapter(adap->dev.parent); > + ret = i2c_mux_tree_trylock(parent); > + } else { > + ret = i2c_trylock_adapter(adap); > + } > + > + return ret; > +} > + > +static void i2c_mux_tree_unlock(struct i2c_adapter *adap) > +{ > + if (i2c_parent_is_i2c_adapter(adap)) > + i2c_mux_tree_unlock(to_i2c_adapter(adap->dev.parent)); > + else > + i2c_unlock_adapter(adap); > } > > /** > @@ -1104,12 +1183,12 @@ int i2c_transfer(struct i2c_adapter *adap, struct > i2c_msg *msgs, int num) > #endif > > if (in_atomic() || irqs_disabled()) { > - ret = rt_mutex_trylock(&adap->bus_lock); > + ret = i2c_mux_tree_trylock(adap); > if (!ret) > /* I2C activity is ongoing. */ > return -EAGAIN; > } else { > - rt_mutex_lock(&adap->bus_lock); > + i2c_mux_tree_lock(adap); > } > > /* Retry automatically on arbitration loss */ > @@ -1121,7 +1200,7 @@ int i2c_transfer(struct i2c_adapter *adap, struct > i2c_msg *msgs, int num) > if (time_after(jiffies, orig_jiffies + adap->timeout)) > break; > } > - rt_mutex_unlock(&adap->bus_lock); > + i2c_mux_tree_unlock(adap); > > return ret; > } else { > @@ -1857,7 +1936,7 @@ s32 i2c_smbus_xfer(struct i2c_adapter *adapter, u16 > addr, unsigned short flags, > flags &= I2C_M_TEN | I2C_CLIENT_PEC; > > if (adapter->algo->smbus_xfer) { > - rt_mutex_lock(&adapter->bus_lock); > + i2c_mux_tree_lock(adapter); > > /* Retry automatically on arbitration loss */ > orig_jiffies = jiffies; > @@ -1871,7 +1950,7 @@ s32 i2c_smbus_xfer(struct i2c_adapter *adapter, u16 > addr, unsigned short flags, > orig_jiffies + adapter->timeout)) > break; > } > - rt_mutex_unlock(&adapter->bus_lock); > + i2c_mux_tree_unlock(adapter); > } else > res = i2c_smbus_xfer_emulated(adapter,addr,flags,read_write, > command, protocol, data); I'm worried that this approach is risky. We now use i2c_mux_tree_*lock internally, but still expose the i2c_*lock_adapter() API externally. This API is _not_ mux-aware, and will do the wrong thing if called on a bus segment behind a mux. I'd rather not have two APIs when we really only need one. Let's just update the current public API to be mux-aware, and use that everywhere. > diff --git a/drivers/i2c/i2c-dev.c b/drivers/i2c/i2c-dev.c > index f4110aa..da5327f 100755 > --- a/drivers/i2c/i2c-dev.c > +++ b/drivers/i2c/i2c-dev.c > @@ -193,12 +193,56 @@ static int i2cdev_check(struct device *dev, void *addrp) > return dev->driver ? -EBUSY : 0; > } > > +static int i2cdev_parent_is_i2c_adapter(const struct i2c_adapter *adapter) > +{ > + return adapter->dev.parent != NULL > + && adapter->dev.parent->bus == &i2c_bus_type; > +} This is an exact duplicate of the same function in i2c-core. I'd rather make it an inline function in <linux/i2c.h>, or export the one from i2c-core.c. We have enough duplicate code in i2c-dev.c already :( > + > +/* walk up mux tree */ > +static int i2cdev_check_mux_parents(struct i2c_adapter *adapter, int addr) > +{ > + int result; > + > + result = device_for_each_child(&adapter->dev, &addr, i2cdev_check); > + > + if (!result && i2cdev_parent_is_i2c_adapter(adapter)) > + result = i2cdev_check_mux_parents( > + to_i2c_adapter(adapter->dev.parent), addr); > + > + return result; > +} > + > +/* recurse down mux tree */ > +static int i2cdev_check_mux_childs(struct i2c_adapter *adapter, int addr) children > +{ > + int result = 0; > + struct i2c_adapter *child, *next; > + > + list_for_each_entry_safe(child, next, &adapter->mux_list_head, > + mux_list) { > + result = device_for_each_child(&child->dev, &addr, > + i2cdev_check); > + if (result) > + break; > + result = i2cdev_check_mux_childs(child, addr); Same as in the i2c-core variant, this value is ignored, I think this isn't correct. > + } > + return result; > +} > + > /* This address checking function differs from the one in i2c-core > in that it considers an address with a registered device, but no > driver bound to it, as NOT busy. */ > static int i2cdev_check_addr(struct i2c_adapter *adapter, unsigned int addr) > { > - return device_for_each_child(&adapter->dev, &addr, i2cdev_check); > + int result; > + > + result = i2cdev_check_mux_parents(adapter, addr); > + > + if (!result) > + result = i2cdev_check_mux_childs(adapter, addr); > + > + return result; > } > > static noinline int i2cdev_ioctl_rdrw(struct i2c_client *client, > diff --git a/drivers/i2c/i2c-mux.c b/drivers/i2c/i2c-mux.c > new file mode 100755 > index 0000000..f88d29a > --- /dev/null > +++ b/drivers/i2c/i2c-mux.c > @@ -0,0 +1,184 @@ > +/* > + * Multiplexed I2C bus driver. > + * > + * Copyright (c) 2008-2009 Rodolfo Giometti <[email protected]> > + * Copyright (c) 2008-2009 Eurotech S.p.A. <[email protected]> > + * > + * Simplifies access to complex multiplexed I2C bus topologies, by presenting > + * each multiplexed bus segment as an additional I2C adapter. > + * Supports multi-level mux'ing (mux behind a mux). > + * > + * Based on: > + * i2c-virt.c from Kumar Gala <[email protected]> > + * i2c-virtual.c from Ken Harrenstien, Copyright (c) 2004 Google, Inc. > + * i2c-virtual.c from Brian Kuschak <[email protected]> > + * > + * This file is licensed under the terms of the GNU General Public > + * License version 2. This program is licensed "as is" without any > + * warranty of any kind, whether express or implied. > + */ > + > +/* Adapted to kernel 2.6.33 by Michael Lawnick <[email protected]> > */ I'd rather have you add your copyright with the other ones already present above. > + > +#include <linux/kernel.h> > +#include <linux/module.h> > +#include <linux/slab.h> > +#include <linux/i2c.h> > +#include <linux/i2c-mux.h> > + > +/* multiplexer per channel data */ > +struct i2c_mux_priv { > + struct i2c_adapter adap; > + struct i2c_algorithm algo; > + > + struct i2c_adapter *parent; > + void *mux_dev; /* the mux chip/device */ > + u32 chan_id; /* the channel id */ > + > + int (*select)(struct i2c_adapter *adap, void *mux_dev, u32 chan_id); > + int (*deselect)(struct i2c_adapter *adap, void *mux_dev, u32 chan_id); The two "adap" can go away... I insisted on having parameter names for the other two; for structures, it's usually pretty clear what we're talking about. > +}; > + > +#define VIRT_TIMEOUT (HZ/2) > +#define VIRT_RETRIES 3 These two are no longer used and can thus be deleted. > + > +static int i2c_mux_master_xfer(struct i2c_adapter *adap, > + struct i2c_msg msgs[], int num) > +{ > + struct i2c_mux_priv *priv = adap->algo_data; > + struct i2c_adapter *parent = priv->parent; > + int ret; > + > + /* Switch to the right mux port and perform the transfer.*/ Missing space at end of comment. > + > + ret = priv->select(parent, priv->mux_dev, priv->chan_id); > + if (ret >= 0) > + ret = parent->algo->master_xfer(parent, msgs, num); > + if (priv->deselect) > + priv->deselect(parent, priv->mux_dev, priv->chan_id); > + > + return ret; > +} > + > +static int i2c_mux_smbus_xfer(struct i2c_adapter *adap, > + u16 addr, unsigned short flags, > + char read_write, u8 command, > + int size, union i2c_smbus_data *data) > +{ > + struct i2c_mux_priv *priv = adap->algo_data; > + struct i2c_adapter *parent = priv->parent; > + int ret; > + > + /* Select the right mux port and perform the transfer.*/ Ditto. > + > + ret = priv->select(parent, priv->mux_dev, priv->chan_id); > + if (ret >= 0) > + ret = parent->algo->smbus_xfer(parent, addr, flags, > + read_write, command, size, data); Strange alignment. > + if (priv->deselect) > + priv->deselect(parent, priv->mux_dev, priv->chan_id); > + > + return ret; > +} > + > +/* Return the parent's functionality for the virtual adapter */ Please avoid referring to "virtual adapters". The fact here is that all bus segments of a given tree share the same functionality, because it is a characteristic of the controller - nothing virtual here. > +static u32 i2c_mux_functionality(struct i2c_adapter *adap) > +{ > + struct i2c_mux_priv *priv = adap->algo_data; > + struct i2c_adapter *parent = priv->parent; > + > + return parent->algo->functionality(parent); > +} > + > +struct i2c_adapter *i2c_add_mux_adapter(struct i2c_adapter *parent, > + void *mux_dev, u32 force_nr, u32 chan_id, > + int (*select) (struct i2c_adapter *, > + void *, u32), > + int (*deselect) (struct i2c_adapter *, > + void *, u32)) > +{ > + struct i2c_mux_priv *priv; > + int ret; > + > + priv = kzalloc(sizeof(struct i2c_mux_priv), GFP_KERNEL); > + if (!priv) > + return NULL; > + > + /* Set up private adapter data */ > + priv->parent = parent; > + priv->mux_dev = mux_dev; > + priv->chan_id = chan_id; > + priv->select = select; > + priv->deselect = deselect; > + > + /* Need to do algo dynamically because we don't know ahead > + * of time what sort of physical adapter we'll be dealing with. > + */ > + if (parent->algo->master_xfer) > + priv->algo.master_xfer = i2c_mux_master_xfer; > + if (parent->algo->smbus_xfer) > + priv->algo.smbus_xfer = i2c_mux_smbus_xfer; > + priv->algo.functionality = i2c_mux_functionality; > + > + /* Now fill out new adapter structure */ > + snprintf(priv->adap.name, sizeof(priv->adap.name), > + "i2c-%d-mux (chan_id %d)", i2c_adapter_id(parent), chan_id); I'm only half happy with this name. One problem is that it isn't unique: if you have 2 mux chips on the same root segment, this will lead to duplicate segments with the same name. Another problem is that it completely hides the name of the controller (which, I admit, we may recover by walking sysfs, but still...) This isn't a blocker point, but something to keep in mind for further improvements. It might even make sense to let the caller provide the format for the segment names. > + priv->adap.owner = THIS_MODULE; > + priv->adap.id = parent->id; > + priv->adap.algo = &priv->algo; > + priv->adap.algo_data = priv; > + priv->adap.dev.parent = &parent->dev; > + > + if (force_nr) { > + priv->adap.nr = force_nr; > + ret = i2c_add_numbered_adapter(&priv->adap); > + } else { > + ret = i2c_add_adapter(&priv->adap); > + } > + if (ret < 0) { > + dev_err(&parent->dev, "failed to add mux-adapter (error=%d)\n", > + ret); > + kfree(priv); > + return NULL; > + } > + > + list_add_tail(&priv->adap.mux_list, &parent->mux_list_head); The driver core maintains a device tree already, do you really need to keep your own? We've tried hard to remove all redundancy between i2c-core and the driver core in the last couple years, so I really would like us to not add such redundancy back now. If you have a good reason to have your own list, please explain. If not, please get rid of it. > + > + dev_info(&parent->dev, "i2c-mux-%d: Multiplexed I2C bus on " > + "parent bus i2c-%d, chan_id %d\n", > + i2c_adapter_id(&priv->adap), i2c_adapter_id(parent), chan_id); Have you looked at the output? I expect some redundancy, as you use dev_info() on the parent, i2c_adapter_id(parent) is already included. I think I'd rather call dev_info() on the newly spawned child, so you can get rid of the leading "i2c-mux-%d". BTW, please use dev_name() instead of hard-coding i2c-%d for the parent. > + > + return &priv->adap; > +} > +EXPORT_SYMBOL_GPL(i2c_add_mux_adapter); > + > +int i2c_del_mux_adapter(struct i2c_adapter *adap) > +{ > + struct i2c_mux_priv *priv = adap->algo_data; > + int ret; > + struct i2c_adapter *child, *next, *parent; Would look nicer with all structs in front and ret last, IMHO. > + > + if (adap->dev.parent != NULL > + && adap->dev.parent->bus == &i2c_bus_type) { This is a copy of i2c_parent_is_i2c_adapter(), isn't it? That's one more reason to export it from i2c-core. > + parent = to_i2c_adapter(adap->dev.parent); Note that we could have i2c_parent_is_i2c_adapter() return the parent adapter, if it proves useful in practice. > + list_for_each_entry_safe(child, next, &parent->mux_list_head, > + mux_list) { > + if (child == adap) { > + list_del(&child->mux_list); > + break; > + } > + } > + } That being said... Using the driver core's list would save you the hassle altogether. > + > + ret = i2c_del_adapter(adap); > + if (ret < 0) > + return ret; > + kfree(priv); > + > + return 0; > +} > +EXPORT_SYMBOL_GPL(i2c_del_mux_adapter); > + > +MODULE_AUTHOR("Rodolfo Giometti <[email protected]>"); > +MODULE_DESCRIPTION("I2C driver for multiplexed I2C busses"); > +MODULE_LICENSE("GPL v2"); > diff --git a/include/linux/i2c-mux.h b/include/linux/i2c-mux.h > new file mode 100755 > index 0000000..0d5d6c8 > --- /dev/null > +++ b/include/linux/i2c-mux.h > @@ -0,0 +1,46 @@ > +/* > + * > + * i2c-mux.h - functions for the i2c-bus mux support > + * > + * Copyright (c) 2008-2009 Rodolfo Giometti <[email protected]> > + * Copyright (c) 2008-2009 Eurotech S.p.A. <[email protected]> > + * Michael Lawnick <[email protected]> > + * > + * This program is free software; you can redistribute it and/or modify > + * it under the terms of the GNU General Public License as published by > + * the Free Software Foundation; either version 2 of the License, or > + * (at your option) any later version. > + * > + * This program is distributed in the hope that it will be useful, > + * but WITHOUT ANY WARRANTY; without even the implied warranty of > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the > + * GNU General Public License for more details. > + * > + * You should have received a copy of the GNU General Public License > + * along with this program; if not, write to the Free Software > + * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. > + */ > + > +#ifndef _LINUX_I2C_MUX_H > +#define _LINUX_I2C_MUX_H > + > +#ifdef __KERNEL__ > + > +/* > + * Called to create a i2c bus on a multiplexed bus segment. > + * The mux_dev and chan_id parameters are passed to the select > + * and deselect callback functions to perform hardware-specific > + * mux control. > + */ > +struct i2c_adapter *i2c_add_mux_adapter(struct i2c_adapter *parent, > + void *mux_dev, u32 force_nr, u32 chan_id, > + int (*select) (struct i2c_adapter *adap, > + void *mux_dev, u32 chan_id), > + int (*deselect) (struct i2c_adapter *, > + void *mux_dev, u32 chan_id)); > + > +int i2c_del_mux_adapter(struct i2c_adapter *adap); > + > +#endif /* __KERNEL__ */ > + > +#endif /* _LINUX_I2C_H */ You actually mean /* _LINUX_I2C_MUX_H */. > diff --git a/include/linux/i2c.h b/include/linux/i2c.h > index 1fd7b12..898a8c1 100755 > --- a/include/linux/i2c.h > +++ b/include/linux/i2c.h > @@ -349,6 +349,8 @@ struct i2c_adapter { > struct completion dev_released; > > struct list_head userspace_clients; > + struct list_head mux_list_head; > + struct list_head mux_list; As said before, I would much prefer if we didn't have to introduce these. > }; > #define to_i2c_adapter(d) container_of(d, struct i2c_adapter, dev) > > @@ -380,6 +382,15 @@ static inline void i2c_unlock_adapter(struct i2c_adapter > *adapter) > rt_mutex_unlock(&adapter->bus_lock); > } > > +/** > + * i2c_trylock_adapter - Try to prevent access to an I2C bus segment > + * @adapter: Target I2C bus segment > + */ > +static inline int i2c_trylock_adapter(struct i2c_adapter *adapter) > +{ > + return rt_mutex_trylock(&adapter->bus_lock); > +} > + > /*flags for the client struct: */ > #define I2C_CLIENT_PEC 0x04 /* Use Packet Error Checking */ > #define I2C_CLIENT_TEN 0x10 /* we have a ten bit chip > address */ Thanks, -- Jean Delvare -- To unsubscribe from this list: send the line "unsubscribe linux-i2c" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
