Date:	Wed, 05 Dec 2012 22:42:03 +0000
From:	Grant Likely <grant.likely@...retlab.ca>
To:	Viresh Kumar <viresh.kumar@...aro.org>,
	Lee Jones <lee.jones@...aro.org>
Cc:	Samuel Ortiz <sameo@...ux.intel.com>, rabin.vincent@...ricsson.com,
	shiraz.hashim@...com, devicetree-discuss@...ts.ozlabs.org,
	linux-kernel@...r.kernel.org, spear-devel@...t.st.com,
	linus.walleij@...aro.org,
	Vipul Kumar Samar <vipulkumar.samar@...com>
Subject: Re: [PATCH V5 2/2] mfd: stmpe: Update DT support in stmpe driver

On Sat, 1 Dec 2012 00:33:46 +0530, Viresh Kumar <viresh.kumar@...aro.org> wrote:
> On 30 November 2012 21:15, Lee Jones <lee.jones@...aro.org> wrote:
> > But ... I don't see how the changes in the -i2c and -spi files
> > are of benefit either. When I boot without the ID table I still
> > get "stmpe-i2c 0-0040: stmpe1601 detected, chip id: 0x212".
> >
> > What is it that actually uses the IDs?
> >
> > Perhaps Viresh can shine some light on the matter?
> 
> As you can see, i wasn't the author of this patch and when you asked
> this question, i didn't had an answer to it. I went through code and
> formed some theory/story :) .
> 
> @Grant: i need your help to check if my theory is correct or not. Question
> is about adding below code in any i2c client driver:
> 
> +#ifdef CONFIG_OF
> +static const struct of_device_id stmpe_dt_ids[] = {
> +       { .compatible = "st,stmpe610", .data = &stmpe_i2c_id[0], },
> +       { .compatible = "st,stmpe801", .data = &stmpe_i2c_id[1], },
> +       { .compatible = "st,stmpe811", .data = &stmpe_i2c_id[2], },
> +       { .compatible = "st,stmpe1601", .data = &stmpe_i2c_id[3], },
> +       { .compatible = "st,stmpe2401", .data = &stmpe_i2c_id[4], },
> +       { .compatible = "st,stmpe2403", .data = &stmpe_i2c_id[5], },
> +       { }
> +};
> +MODULE_DEVICE_TABLE(of, stmpe_dt_ids);
> +#endif
> +
>  static struct i2c_driver stmpe_i2c_driver = {
>         .driver = {
>                 .name = "stmpe-i2c",
> @@ -88,6 +102,7 @@ static struct i2c_driver stmpe_i2c_driver = {
>  #ifdef CONFIG_PM
>                 .pm = &stmpe_dev_pm_ops,
>  #endif
> +               .of_match_table = of_match_ptr(stmpe_dt_ids),
> 
> So, what is the use of this table when we already have i2c_driver.id_table
> populated.
> 
> This is my theory:
> ---------------------
> Adapter drivers supporting DT will call:
> of_i2c_register_devices()
> {
> 	for_each_child_of_node(adap->dev.of_node, node) {
> 		if (of_modalias_node(node, info.type, sizeof(info.type)) < 0)
>                      error condition
> 
>                 ...
> 		result = i2c_new_device(adap, &info);
> 
>           ...
> }
> 
> of_modalias_node(): expects compatible in child node, i.e. stmpe node in our
> case. If it is not there, then that node is skipped. then it copies
> string after ','
> to info.type. So, for us only "stmpe810" out of "st,stmpe810" is copied.
> 
> Now this name, i.e. "stmpe810" is copied as client.name in i2c_new_device()
> and device_register() is called, which will eventually call:
> 
> i2c_device_match()
> {
> 	/* Attempt an OF style match */
> 	if (of_driver_match_device(dev, drv))
> 		return 1;
> 
> 	driver = to_i2c_driver(drv);
> 	/* match on an id table if there is one */
> 	if (driver->id_table)
> 		return i2c_match_id(driver->id_table, client) != NULL;
> }
> 
> This first tries to match the table my patch added, _BUT_ the string will
> never match as we had "st,stmpe810" in table and "stmpe810" in dev.

of_driver_match_device() matches against the compatible list in
dev->of_node, not against the device name. So, if the compatible
property has a string that is in the table, then it really should match
against it.

> 
> So, we fall back to i2c_match_id(), which will match it against
> i2c_driver.id_table present in driver, which has entry for "stmpe810" and
> so strings matched.
> 
> @Lee: This is what happened in your case. :)
> 
> So, whether its DT or non DT, true is returned from here if something
> matched.
> 
> Later on, this will be called:
> 
> static int i2c_device_probe(struct device *dev)
> {
>         .....
> 	status = driver->probe(client, i2c_match_id(driver->id_table, client));

Here things are a bit wonky. Even if matched against the table, it is
possible that it also matches against i2c_match_id() and that data is
passed to the driver.

But regardless, it is the responsiblity of the probe function to go and
look if of_driver_match_device() matches against anything if it cares
about the of_match_table entries (for instance, if there is extra data
attached).

>         ....
> }
> 
> Which will again match the legacy table to find correct struct i2c_device_id *id
> to pass to probe().
> 
> So, the final question: WTF is of_match_table for?

A bit of history is valuable here. The whole of_modalias_node() thing is
really just a best-effort heuristic for figuring out which driver
*might* work against a device described in the device tree. It won't
work in all circumstances (and it was created at a time when there was
resistance to adding DT knowledge to drivers). An of_match_table is the
robust way of identifying specific devices and allows for matching
against any entry in the compatible list.

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

Powered by blists - more mailing lists