Re: [PATCH v6 01/24] i2c-mux: add common data for every i2c-mux instance
Wolfram Sang wrote: > > > wonder even more if we couldn't supply num_adapters to i2c_mux_alloc() > > > and reserve the memory statically. i2c busses are not > > > dynamic/hot-pluggable so that should be good enough? > > > > Yes, that would work, but it would take some restructuring in some of > > the drivers that currently don't know how many child adapters they > > are going to need when they call i2c_mux_alloc. > > Which ones? If you look at i2c-mux-reg.c, it currently allocates its private struct regmux, then fills it with various platform things and then when it knows how many children it needs it allocates them. After v6 it first allocates a mux core and private struct regmux in one go using i2c_mux_alloc, then continues in much the same way as before. If the number of children is needed for the i2c_mux_alloc call, then this is certainly doable, and it would probably not be all that bad, but the simplest approach would probably be to allocate the private struct regmux first, then dig through the platform data, then allocate the mux core when the number of children is known. Which would still be two allocations separated by the platform data dig. So, your suggestion would basically move the mux core allocation from generally being done early together with other private data to later when the driver has figured out how many children it's going to create. The restructuring I thought about is needed if the intention of this was to reduce number of allocations, but maybe you just wanted what I described above? Because what I did in v6 and what you are suggesting is quite similar in complexity, but your version has the advantage of not having the need for realloc. So, I have made this change locally (and the adapters->num_adapters change) and I like it. I haven't even compile-tested it yet though, but I'll get back when I have done some testing. > > Because you thought about removing i2c_mux_reserve_adapters completely, > > and not provide any means of adding more adapters than specified in > > the i2c_mux_alloc call, right? > > Yes. I assumed I2C to be static enough that such information is known in > advance. > > > > Ignoring the 80 char limit here makes the code more readable. > > > > That is only true if you actually have more than 80 characters, so I don't > > agree. Are you adamant about it? (I'm not) > > No. Keep it if you prefer it. > > > >> +EXPORT_SYMBOL_GPL(i2c_mux_one_adapter); > > > > > > Are you sure the above function pays off? Its argument list is very > > > complex and it doesn't save a lot of code. Having seperate calls is > > > probably more understandable in drivers? Then again, I assume it makes > > > the conversion of existing drivers easier. > > > > I added it in v4, you can check earlier versions if you like. Without > > it most gate-muxes (i.e. typically the muxes in drivers/media) grew > > since the i2c_add_mux_adapter call got replaced by two calls, i.e. > > i2c_mux_alloc followed by i2c_max_add_adapter, and coupled with > > error checks made it look more complex than before. So, this wasn't > > much of a cleanup from the point of those drivers. > > Hmm, v3 didn't have the driver patches posted with it. Can you push it > to your branch? I am also not too strong with this one, but having a > look how it looks without would be nice. Although I'm not sure what you meant by "driver patches", I have pushed mux-core-and-locking-2 and mux-core-and-locking-3 to https://github.com/peda-r/i2c-mux/ (note that these are the branches as they where when I posted v2 and v3 to the list, i.e. w/o fixups) Those early versions updated all drivers with each change, making each patch big, so if that was what you meant by missing "driver patches" then there simply were no driver patches. If you meant the follow-up patches to relax locking in the media drivers etc, I only compile-tested them using throwaway branches back then (if I even had branches). So, I don't have anything ready to push, sorry. Cheers, Peter -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH v6 01/24] i2c-mux: add common data for every i2c-mux instance
> > I'd suggest to rename 'adapters' into 'num_adapters' throughout this > > patch. I think it makes the code a lot easier to understand. > > Hmm, you mean just the variable names, right? And not function names > such as i2c_mux_reserve_(num_)adapters? Yes, only variable names. > > Despite that I wonder why not using some of the realloc functions, I > > When I wrote it, I found no devm_ version of realloc. I'm not finding > anything now either... Right, there is no devm_ version of it :( > > wonder even more if we couldn't supply num_adapters to i2c_mux_alloc() > > and reserve the memory statically. i2c busses are not > > dynamic/hot-pluggable so that should be good enough? > > Yes, that would work, but it would take some restructuring in some of > the drivers that currently don't know how many child adapters they > are going to need when they call i2c_mux_alloc. Which ones? > Because you thought about removing i2c_mux_reserve_adapters completely, > and not provide any means of adding more adapters than specified in > the i2c_mux_alloc call, right? Yes. I assumed I2C to be static enough that such information is known in advance. > > Ignoring the 80 char limit here makes the code more readable. > > That is only true if you actually have more than 80 characters, so I don't > agree. Are you adamant about it? (I'm not) No. Keep it if you prefer it. > >> +EXPORT_SYMBOL_GPL(i2c_mux_one_adapter); > > > > Are you sure the above function pays off? Its argument list is very > > complex and it doesn't save a lot of code. Having seperate calls is > > probably more understandable in drivers? Then again, I assume it makes > > the conversion of existing drivers easier. > > I added it in v4, you can check earlier versions if you like. Without > it most gate-muxes (i.e. typically the muxes in drivers/media) grew > since the i2c_add_mux_adapter call got replaced by two calls, i.e. > i2c_mux_alloc followed by i2c_max_add_adapter, and coupled with > error checks made it look more complex than before. So, this wasn't > much of a cleanup from the point of those drivers. Hmm, v3 didn't have the driver patches posted with it. Can you push it to your branch? I am also not too strong with this one, but having a look how it looks without would be nice. signature.asc Description: PGP signature
Re: [PATCH v6 01/24] i2c-mux: add common data for every i2c-mux instance
Hi! On 2016-04-11 22:46, Wolfram Sang wrote: > Hi Peter, > > first high-level review: > >> +int i2c_mux_reserve_adapters(struct i2c_mux_core *muxc, int adapters) > > I'd suggest to rename 'adapters' into 'num_adapters' throughout this > patch. I think it makes the code a lot easier to understand. Hmm, you mean just the variable names, right? And not function names such as i2c_mux_reserve_(num_)adapters? >> +{ >> +struct i2c_adapter **adapter; >> + >> +if (adapters <= muxc->max_adapters) >> +return 0; >> + >> +adapter = devm_kmalloc_array(muxc->dev, >> + adapters, sizeof(*adapter), >> + GFP_KERNEL); >> +if (!adapter) >> +return -ENOMEM; >> + >> +if (muxc->adapter) { >> +memcpy(adapter, muxc->adapter, >> + muxc->max_adapters * sizeof(*adapter)); >> +devm_kfree(muxc->dev, muxc->adapter); >> +} >> + >> +muxc->adapter = adapter; >> +muxc->max_adapters = adapters; >> +return 0; >> +} >> +EXPORT_SYMBOL_GPL(i2c_mux_reserve_adapters); > > Despite that I wonder why not using some of the realloc functions, I When I wrote it, I found no devm_ version of realloc. I'm not finding anything now either... > wonder even more if we couldn't supply num_adapters to i2c_mux_alloc() > and reserve the memory statically. i2c busses are not > dynamic/hot-pluggable so that should be good enough? Yes, that would work, but it would take some restructuring in some of the drivers that currently don't know how many child adapters they are going to need when they call i2c_mux_alloc. Because you thought about removing i2c_mux_reserve_adapters completely, and not provide any means of adding more adapters than specified in the i2c_mux_alloc call, right? >> -WARN(sysfs_create_link(&priv->adap.dev.kobj, &mux_dev->kobj, >> "mux_device"), >> - "can't create symlink to mux device\n"); >> +WARN(sysfs_create_link(&priv->adap.dev.kobj, &muxc->dev->kobj, >> + "mux_device"), > > Ignoring the 80 char limit here makes the code more readable. That is only true if you actually have more than 80 characters, so I don't agree. Are you adamant about it? (I'm not) >> + "can't create symlink to mux device\n"); >> >> snprintf(symlink_name, sizeof(symlink_name), "channel-%u", chan_id); >> -WARN(sysfs_create_link(&mux_dev->kobj, &priv->adap.dev.kobj, >> symlink_name), >> - "can't create symlink for channel %u\n", >> chan_id); >> +WARN(sysfs_create_link(&muxc->dev->kobj, &priv->adap.dev.kobj, >> + symlink_name), > > ditto. > >> + "can't create symlink for channel %u\n", chan_id); >> dev_info(&parent->dev, "Added multiplexed i2c bus %d\n", >> i2c_adapter_id(&priv->adap)); >> >> -return &priv->adap; >> +muxc->adapter[muxc->adapters++] = &priv->adap; >> +return 0; >> +} >> +EXPORT_SYMBOL_GPL(i2c_mux_add_adapter); >> + >> +struct i2c_mux_core *i2c_mux_one_adapter(struct i2c_adapter *parent, >> + struct device *dev, int sizeof_priv, >> + u32 flags, u32 force_nr, >> + u32 chan_id, unsigned int class, >> + int (*select)(struct i2c_mux_core *, >> + u32), > > ditto > >> + int (*deselect)(struct i2c_mux_core *, >> + u32)) > > ditto > >> +{ >> +struct i2c_mux_core *muxc; >> +int ret; >> + >> +muxc = i2c_mux_alloc(parent, dev, sizeof_priv, flags, select, deselect); >> +if (!muxc) >> +return ERR_PTR(-ENOMEM); >> + >> +ret = i2c_mux_add_adapter(muxc, force_nr, chan_id, class); >> +if (ret) { >> +devm_kfree(dev, muxc); >> +return ERR_PTR(ret); >> +} >> + >> +return muxc; >> +} >> +EXPORT_SYMBOL_GPL(i2c_mux_one_adapter); > > Are you sure the above function pays off? Its argument list is very > complex and it doesn't save a lot of code. Having seperate calls is > probably more understandable in drivers? Then again, I assume it makes > the conversion of existing drivers easier. I added it in v4, you can check earlier versions if you like. Without it most gate-muxes (i.e. typically the muxes in drivers/media) grew since the i2c_add_mux_adapter call got replaced by two calls, i.e. i2c_mux_alloc followed by i2c_max_add_adapter, and coupled with error checks made it look more complex than before. So, this wasn't much of a cleanup from the point of those drivers. > So much for now. Thanks! Yeah, thanks! Cheers, Peter -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.ke
Re: [PATCH v6 01/24] i2c-mux: add common data for every i2c-mux instance
Hi Peter, first high-level review: > +int i2c_mux_reserve_adapters(struct i2c_mux_core *muxc, int adapters) I'd suggest to rename 'adapters' into 'num_adapters' throughout this patch. I think it makes the code a lot easier to understand. > +{ > + struct i2c_adapter **adapter; > + > + if (adapters <= muxc->max_adapters) > + return 0; > + > + adapter = devm_kmalloc_array(muxc->dev, > + adapters, sizeof(*adapter), > + GFP_KERNEL); > + if (!adapter) > + return -ENOMEM; > + > + if (muxc->adapter) { > + memcpy(adapter, muxc->adapter, > +muxc->max_adapters * sizeof(*adapter)); > + devm_kfree(muxc->dev, muxc->adapter); > + } > + > + muxc->adapter = adapter; > + muxc->max_adapters = adapters; > + return 0; > +} > +EXPORT_SYMBOL_GPL(i2c_mux_reserve_adapters); Despite that I wonder why not using some of the realloc functions, I wonder even more if we couldn't supply num_adapters to i2c_mux_alloc() and reserve the memory statically. i2c busses are not dynamic/hot-pluggable so that should be good enough? > - WARN(sysfs_create_link(&priv->adap.dev.kobj, &mux_dev->kobj, > "mux_device"), > -"can't create symlink to mux device\n"); > + WARN(sysfs_create_link(&priv->adap.dev.kobj, &muxc->dev->kobj, > +"mux_device"), Ignoring the 80 char limit here makes the code more readable. > + "can't create symlink to mux device\n"); > > snprintf(symlink_name, sizeof(symlink_name), "channel-%u", chan_id); > - WARN(sysfs_create_link(&mux_dev->kobj, &priv->adap.dev.kobj, > symlink_name), > -"can't create symlink for channel %u\n", > chan_id); > + WARN(sysfs_create_link(&muxc->dev->kobj, &priv->adap.dev.kobj, > +symlink_name), ditto. > + "can't create symlink for channel %u\n", chan_id); > dev_info(&parent->dev, "Added multiplexed i2c bus %d\n", >i2c_adapter_id(&priv->adap)); > > - return &priv->adap; > + muxc->adapter[muxc->adapters++] = &priv->adap; > + return 0; > +} > +EXPORT_SYMBOL_GPL(i2c_mux_add_adapter); > + > +struct i2c_mux_core *i2c_mux_one_adapter(struct i2c_adapter *parent, > + struct device *dev, int sizeof_priv, > + u32 flags, u32 force_nr, > + u32 chan_id, unsigned int class, > + int (*select)(struct i2c_mux_core *, > +u32), ditto > + int (*deselect)(struct i2c_mux_core *, > + u32)) ditto > +{ > + struct i2c_mux_core *muxc; > + int ret; > + > + muxc = i2c_mux_alloc(parent, dev, sizeof_priv, flags, select, deselect); > + if (!muxc) > + return ERR_PTR(-ENOMEM); > + > + ret = i2c_mux_add_adapter(muxc, force_nr, chan_id, class); > + if (ret) { > + devm_kfree(dev, muxc); > + return ERR_PTR(ret); > + } > + > + return muxc; > +} > +EXPORT_SYMBOL_GPL(i2c_mux_one_adapter); Are you sure the above function pays off? Its argument list is very complex and it doesn't save a lot of code. Having seperate calls is probably more understandable in drivers? Then again, I assume it makes the conversion of existing drivers easier. So much for now. Thanks! Wolfram signature.asc Description: PGP signature
[PATCH v6 01/24] i2c-mux: add common data for every i2c-mux instance
From: Peter Rosin All i2c-muxes have a parent adapter and one or many child adapters. A mux also has some means of selection. Previously, this was stored per child adapter, but it is only needed to keep track of this per mux. Add an i2c mux core, that keeps track of this consistently. Also add some glue for users of the old interface, which will create one implicit mux core per child adapter. Signed-off-by: Peter Rosin --- drivers/i2c/i2c-mux.c | 238 ++-- include/linux/i2c-mux.h | 47 ++ 2 files changed, 237 insertions(+), 48 deletions(-) diff --git a/drivers/i2c/i2c-mux.c b/drivers/i2c/i2c-mux.c index d4022878b2f0..d95eb66e11bf 100644 --- a/drivers/i2c/i2c-mux.c +++ b/drivers/i2c/i2c-mux.c @@ -28,33 +28,34 @@ #include /* multiplexer per channel data */ +struct i2c_mux_priv_old { + void *mux_priv; + int (*select)(struct i2c_adapter *, void *mux_priv, u32 chan_id); + int (*deselect)(struct i2c_adapter *, void *mux_priv, u32 chan_id); +}; + struct i2c_mux_priv { struct i2c_adapter adap; struct i2c_algorithm algo; - - struct i2c_adapter *parent; - struct device *mux_dev; - void *mux_priv; + struct i2c_mux_core *muxc; u32 chan_id; - - int (*select)(struct i2c_adapter *, void *mux_priv, u32 chan_id); - int (*deselect)(struct i2c_adapter *, void *mux_priv, u32 chan_id); }; 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; + struct i2c_mux_core *muxc = priv->muxc; + struct i2c_adapter *parent = muxc->parent; int ret; /* Switch to the right mux port and perform the transfer. */ - ret = priv->select(parent, priv->mux_priv, priv->chan_id); + ret = muxc->select(muxc, priv->chan_id); if (ret >= 0) ret = __i2c_transfer(parent, msgs, num); - if (priv->deselect) - priv->deselect(parent, priv->mux_priv, priv->chan_id); + if (muxc->deselect) + muxc->deselect(muxc, priv->chan_id); return ret; } @@ -65,17 +66,18 @@ static int i2c_mux_smbus_xfer(struct i2c_adapter *adap, int size, union i2c_smbus_data *data) { struct i2c_mux_priv *priv = adap->algo_data; - struct i2c_adapter *parent = priv->parent; + struct i2c_mux_core *muxc = priv->muxc; + struct i2c_adapter *parent = muxc->parent; int ret; /* Select the right mux port and perform the transfer. */ - ret = priv->select(parent, priv->mux_priv, priv->chan_id); + ret = muxc->select(muxc, priv->chan_id); if (ret >= 0) ret = parent->algo->smbus_xfer(parent, addr, flags, read_write, command, size, data); - if (priv->deselect) - priv->deselect(parent, priv->mux_priv, priv->chan_id); + if (muxc->deselect) + muxc->deselect(muxc, priv->chan_id); return ret; } @@ -84,7 +86,7 @@ static int i2c_mux_smbus_xfer(struct i2c_adapter *adap, static u32 i2c_mux_functionality(struct i2c_adapter *adap) { struct i2c_mux_priv *priv = adap->algo_data; - struct i2c_adapter *parent = priv->parent; + struct i2c_adapter *parent = priv->muxc->parent; return parent->algo->functionality(parent); } @@ -102,30 +104,80 @@ static unsigned int i2c_mux_parent_classes(struct i2c_adapter *parent) return class; } -struct i2c_adapter *i2c_add_mux_adapter(struct i2c_adapter *parent, - struct device *mux_dev, - void *mux_priv, u32 force_nr, u32 chan_id, - unsigned int class, - int (*select) (struct i2c_adapter *, - void *, u32), - int (*deselect) (struct i2c_adapter *, -void *, u32)) +int i2c_mux_reserve_adapters(struct i2c_mux_core *muxc, int adapters) +{ + struct i2c_adapter **adapter; + + if (adapters <= muxc->max_adapters) + return 0; + + adapter = devm_kmalloc_array(muxc->dev, +adapters, sizeof(*adapter), +GFP_KERNEL); + if (!adapter) + return -ENOMEM; + + if (muxc->adapter) { + memcpy(adapter, muxc->adapter, + muxc->max_adapters * sizeof(*adapter)); + devm_kfree(muxc->dev, muxc->adapter); + } + + muxc->adapter = adapter; + muxc->max_adapters = adapters; + return 0; +} +EXPORT_SYMBOL_GPL(i2c_mux_reserve_adapters); + +struct i2c_mux_core *i2c_mux_alloc(struct i2c_adapter *parent, +