Re: [PATCH 2/2] i2c: i801: add support of Host Notify

[Benjamin Tissoires]

On Jul 07 2015 or thereabouts, Jean Delvare wrote:
> Hi Benjamin,
> 
> On Tue, 23 Jun 2015 14:58:19 -0400, Benjamin Tissoires wrote:
> > The i801 chip can handle the Host Notify feature since ICH 3 as mentioned
> > in http://www.intel.com/content/dam/doc/datasheet/82801ca-io-controller-hub-3-datasheet.pdf
> > 
> > Implement .toggle_smbus_host_notify() and rely on .alert() to notify
> > the actual client that it has a notification.
> > 
> > With a T440s and a Synaptics touchpad that implements Host Notify, the
> > payload data is always 0x0000, so I am not sure if the device actually
> > sends the payload or if there is a problem regarding the implementation.
> > 
> > Part of this code is inspired by i2c-smbus.c.
> > 
> > Signed-off-by: Benjamin Tissoires <benjamin.tissoires@xxxxxxxxxx>
> 
> An explanation on why a fifo is needed would be good to add, as I can't
> see any obvious reason.

I inserted the kfifo to be able to read the payload and address in the
same thread that the one handling the IRQ. But a few tests shows that I
can simply trigger the work in the interrupt, and read the data and
clear the Host Notification bit in the work thread directly. So consider
I will remove this in the v2.

> 
> No time for a full review but just a few random comments.
> 
> > ---
> >  drivers/i2c/busses/i2c-i801.c | 223 +++++++++++++++++++++++++++++++++---------
> >  drivers/i2c/i2c-core.c        |  34 +++++++
> >  drivers/i2c/i2c-smbus.c       |  17 +---
> >  include/linux/i2c.h           |   3 +
> >  4 files changed, 217 insertions(+), 60 deletions(-)
> 
> You really shouldn't mix core changes with driver specific changes.
> Anyone should be able to backport the core changes alone, and then add
> support to a different bus driver, without drawing in the i2c-i801
> specific changes (that may not be needed and may not even apply.)

OK, I will split the series for each driver/feature.

> 
> > diff --git a/drivers/i2c/busses/i2c-i801.c b/drivers/i2c/busses/i2c-i801.c
> > index 5ecbb3f..22a3218 100644
> > --- a/drivers/i2c/busses/i2c-i801.c
> > +++ b/drivers/i2c/busses/i2c-i801.c
> > (...)
> > @@ -221,6 +233,17 @@ struct i801_priv {
> >  	const struct i801_mux_config *mux_drvdata;
> >  	struct platform_device *mux_pdev;
> >  #endif
> > +
> > +	bool host_notify_requested;
> > +	struct work_struct host_notify;
> > +	struct kfifo host_notify_fifo;
> > +};
> > +
> > +#define SMBHSTNTFY_SIZE		8
> > +
> > +struct smbus_host_notify_data {
> > +	u8 addr;
> > +	u16 payload;
> >  };
> 
> This structure seems generic to the protocol and not specific to the
> Intel 82801 implementation, so it should go to <linux/i2c-smbus.h>.

fair enough

> 
> >  #define FEATURE_SMBUS_PEC	(1 << 0)
> > (...)
> > @@ -801,12 +894,35 @@ static u32 i801_func(struct i2c_adapter *adapter)
> >  	       I2C_FUNC_SMBUS_BLOCK_DATA | I2C_FUNC_SMBUS_WRITE_I2C_BLOCK |
> >  	       ((priv->features & FEATURE_SMBUS_PEC) ? I2C_FUNC_SMBUS_PEC : 0) |
> >  	       ((priv->features & FEATURE_I2C_BLOCK_READ) ?
> > -		I2C_FUNC_SMBUS_READ_I2C_BLOCK : 0);
> > +		I2C_FUNC_SMBUS_READ_I2C_BLOCK : 0) |
> > +	       ((priv->features & FEATURE_HOST_NOTIFY) ?
> > +		I2C_FUNC_SMBUS_HOST_NOTIFY : 0);
> > +}
> > +
> > +static int i801_toggle_host_ntfy(struct i2c_adapter *adapter, bool state)
> 
> Probably no longer relevant, but please spell out "notify" completely
> in function and variable names (NTFY in register defines is OK.)

Still relevant actually :)
I still need to enable the feature in the driver and on resume. So I
will amend in the v2.

> 
> > +{
> > +	struct i801_priv *priv = i2c_get_adapdata(adapter);
> > +
> > +	if (!(priv->features & FEATURE_HOST_NOTIFY))
> > +		return -EOPNOTSUPP;
> > +
> > +	if (state) {
> > +		outb_p(SMBSLVCMD_HST_NTFY_INTREN, SMBSLVCMD(priv));
> > +		/* clear Host Notify bit to allow a new notification */
> > +		outb_p(SMBSLVSTS_HST_NTFY_STS, SMBSLVSTS(priv));
> > +	} else {
> > +		outb_p(0, SMBSLVCMD(priv));
> > +	}
> > +
> > +	priv->host_notify_requested = state;
> > +
> > +	return 0;
> >  }
> >  
> >  static const struct i2c_algorithm smbus_algorithm = {
> > -	.smbus_xfer	= i801_access,
> > -	.functionality	= i801_func,
> > +	.smbus_xfer			= i801_access,
> > +	.functionality			= i801_func,
> > +	.toggle_smbus_host_notify	= i801_toggle_host_ntfy,
> >  };
> >  
> >  static const struct pci_device_id i801_ids[] = {
> > @@ -1166,6 +1282,7 @@ static int i801_probe(struct pci_dev *dev, const struct pci_device_id *id)
> >  		priv->features |= FEATURE_BLOCK_BUFFER;
> >  		/* fall through */
> >  	case PCI_DEVICE_ID_INTEL_82801CA_3:
> > +		priv->features |= FEATURE_HOST_NOTIFY;
> 
> Please add a /* fall through */ comment for consistency and clarity.

Oops, my mistake.

> 
> >  	case PCI_DEVICE_ID_INTEL_82801BA_2:
> >  	case PCI_DEVICE_ID_INTEL_82801AB_3:
> >  	case PCI_DEVICE_ID_INTEL_82801AA_3:
> > @@ -1250,6 +1367,15 @@ static int i801_probe(struct pci_dev *dev, const struct pci_device_id *id)
> >  		}
> >  	}
> >  
> > +	INIT_WORK(&priv->host_notify, i801_host_notify);
> > +	if (kfifo_alloc(&priv->host_notify_fifo,
> > +			SMBHSTNTFY_SIZE * sizeof(struct smbus_host_notify_data),
> > +			GFP_KERNEL)) {
> > +		dev_err(&dev->dev,
> > +			"%s:failed allocating host_notify_fifo\n", __func__);
> 
> Adding the function name does not add value. Other messages in this
> driver do not include it, so for consistency do not either. A leading
> capital would be good too. Also kfifo_alloc() returns an actual error
> code, which you should save, include in the error message and return to
> the caller instead of hard-coding -ENOMEM.
> 
> I think priv->host_notify_fifo is leaked if i801_probe() fails later on.

OK. But given that the fifo disappeared (hopefully), this is no longer
relevant. I also made sure the work queue is not triggered before any
return path in order not having a out_err label.

> 
> > +		return -ENOMEM;
> > +	}
> > +
> >  	if (priv->features & FEATURE_IRQ) {
> >  		init_waitqueue_head(&priv->waitq);
> >  
> > @@ -1292,6 +1418,9 @@ static void i801_remove(struct pci_dev *dev)
> >  {
> >  	struct i801_priv *priv = pci_get_drvdata(dev);
> >  
> > +	cancel_work_sync(&priv->host_notify);
> > +	kfifo_free(&priv->host_notify_fifo);
> > +
> >  	i801_del_mux(priv);
> >  	i2c_del_adapter(&priv->adapter);
> >  	pci_write_config_byte(dev, SMBHSTCFG, priv->original_hstcfg);
> > @@ -1315,8 +1444,12 @@ static int i801_suspend(struct pci_dev *dev, pm_message_t mesg)
> >  
> >  static int i801_resume(struct pci_dev *dev)
> >  {
> > +	struct i801_priv *priv = pci_get_drvdata(dev);
> > +
> >  	pci_set_power_state(dev, PCI_D0);
> >  	pci_restore_state(dev);
> > +	if (priv->host_notify_requested)
> > +		i801_toggle_host_ntfy(&priv->adapter, true);
> >  	return 0;
> >  }
> >  #else
> > diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
> > index eaaf5b0..87904ec 100644
> > --- a/drivers/i2c/i2c-core.c
> > +++ b/drivers/i2c/i2c-core.c
> > @@ -2145,6 +2145,40 @@ int i2c_master_send(const struct i2c_client *client, const char *buf, int count)
> >  EXPORT_SYMBOL(i2c_master_send);
> >  
> >  /**
> > + * i2c_alert - call an alert for the given i2c_client.
> > + * @client: Handle to slave device
> > + * @data: Payload data that will be sent to the slave
> 
> Actually this is data that was received from the "slave" device
> (temporarily acting as a master), if I understand how Host Notify works.

You are right.

> 
> > + *
> > + * Returns negative errno, or 0.
> > + */
> > +int i2c_alert(struct i2c_client *client, unsigned int data)
> > +{
> > +	struct i2c_driver *driver;
> > +
> > +	if (!client)
> > +		return -EINVAL;
> 
> Do you actually need to check for that? It seems redundant with the
> check in smbus_do_alert().

Looks like i801_do_host_notify() checks for that too. I will drop it.

> 
> > +
> > +	/*
> > +	 * Drivers should either disable alerts, or provide at least
> > +	 * a minimal handler.  Lock so the driver won't change.
> > +	 */
> > +	device_lock(&client->adapter->dev);
> > +	if (client->dev.driver) {
> > +		driver = to_i2c_driver(client->dev.driver);
> > +		if (driver->alert)
> > +			driver->alert(client, data);
> 
> So you use the same driver callback for SMBus Alert and SMBus Host
> Notify. This makes some sense, but if a given driver supports both, how
> does it know which event happened? The data is completely different and
> most probably the action required from the driver as well.

Yeah, this gets messy. I re-used the .alert() callback because of the
documentation:  "Alert callback, for example for the SMBus alert protocol".
It would seem that the alert is generic and could be re-used. But OTOH,
it is not prepared to receive anything else than a SMBus Alert.

Given that I had a toggle_host_notify() call, I figured that this was
not a problem unless you write a driver which implements both (I can not
find a sane use case for this though).

But now that this call has disappeared, we would need a way to
differentiate the too of them.

I can see two solutions out of my head right now:
- add a "protocol" parameter (with an enum) to .alert()
- add a new callback .host_notify() in struct i2c_driver.

I think I like the second option more given that it will allow to not
touch the current code in i2c_smbus.

> 
> > +		else
> > +			dev_warn(&client->dev, "no driver alert()!\n");
> > +	} else
> > +		dev_dbg(&client->dev, "alert with no driver\n");
> > +	device_unlock(&client->adapter->dev);
> > +
> > +	return 0;
> > +}
> > +EXPORT_SYMBOL_GPL(i2c_alert);
> > +
> > +
> 
> Please avoid duplicate blank lines.
> 

Sorry :(

> > +/**
> >   * i2c_master_recv - issue a single I2C message in master receive mode
> >   * @client: Handle to slave device
> >   * @buf: Where to store data read from slave
> > diff --git a/drivers/i2c/i2c-smbus.c b/drivers/i2c/i2c-smbus.c
> > index 9ebf9cb..26439a8 100644
> > --- a/drivers/i2c/i2c-smbus.c
> > +++ b/drivers/i2c/i2c-smbus.c
> > @@ -41,27 +41,14 @@ static int smbus_do_alert(struct device *dev, void *addrp)
> >  {
> >  	struct i2c_client *client = i2c_verify_client(dev);
> >  	struct alert_data *data = addrp;
> > -	struct i2c_driver *driver;
> >  
> >  	if (!client || client->addr != data->addr)
> >  		return 0;
> >  	if (client->flags & I2C_CLIENT_TEN)
> >  		return 0;
> >  
> > -	/*
> > -	 * Drivers should either disable alerts, or provide at least
> > -	 * a minimal handler.  Lock so the driver won't change.
> > -	 */
> > -	device_lock(dev);
> > -	if (client->dev.driver) {
> > -		driver = to_i2c_driver(client->dev.driver);
> > -		if (driver->alert)
> > -			driver->alert(client, data->flag);
> > -		else
> > -			dev_warn(&client->dev, "no driver alert()!\n");
> > -	} else
> > -		dev_dbg(&client->dev, "alert with no driver\n");
> > -	device_unlock(dev);
> > +	if (i2c_alert(client, data->flag))
> > +		return 0;
> >  
> >  	/* Stop iterating after we find the device */
> >  	return -EBUSY;
> > diff --git a/include/linux/i2c.h b/include/linux/i2c.h
> > index 7ffc970..fbca48a 100644
> > --- a/include/linux/i2c.h
> > +++ b/include/linux/i2c.h
> > @@ -72,6 +72,9 @@ extern int i2c_transfer(struct i2c_adapter *adap, struct i2c_msg *msgs,
> >  extern int __i2c_transfer(struct i2c_adapter *adap, struct i2c_msg *msgs,
> >  			  int num);
> >  
> > +/* call an alert. */
> 
> What a vague comment :(

sorry again :(

> 
> > +extern int i2c_alert(struct i2c_client *client, unsigned int data);
> > +
> >  /* This is the very generalized SMBus access routine. You probably do not
> >     want to use this, though; one of the functions below may be much easier,
> >     and probably just as fast.
> 

Cheers,
Benjamin
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/