[PATCH V5 2/2] mfd: stmpe: Update DT support in stmpe driver

[Grant Likely]

On Thu, 6 Dec 2012 08:06:20 +0530, Viresh Kumar <viresh.kumar at linaro.org> wrote:
> On 6 December 2012 04:12, Grant Likely <grant.likely at secretlab.ca> wrote:
> > On Sat, 1 Dec 2012 00:33:46 +0530, Viresh Kumar <viresh.kumar at linaro.org> wrote:
> >> 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.
> 
> Grant, but isn't it true that the final device's name would not be the DT
> way of names? It would simply be "stmpe810" for us and so if we have
> multiple instances of stmpe on a board, we need to distinguish them
> ourselves? One way for that was passing id for these instances, which
> would finally be used, when we create platform devices for sub-modules
> of stmpe (gpio, keypad, ts, etc).

of_modalias_node() is based on a *heruistic*. It is a best-effort
attempt to convert the node's compatible lists into a string that will
match against an existing driver. In the simple case it works because
historically i2c has used the chip name for the driver name. We get
lucky and a lot of drivers will work with DT without changes.

However, it is in no way guaranteed. Sometimes the strings won't line up
or a certain silicon vendor will have an extra errata or feature. In
that case it makes sense to use a DT match table that can parse the
entries in the compatible list. Or a driver can call of_ helper
functions. For example, it might call of_alias_get_id() to figure out
which device id it needs. That's why the full DT parsing exists. It's
the fallback when the simple heuristic fails. Only use it when you need
to.

g.