Re: [PATCH] i2c: mux: Remove unneeded platform_set_drvdata to NULL in remove

[Doug Anderson <dianders-F7+t8E8rja9g9hUCZPvPmw@public.gmane.org>]

Wolfram,

On Fri, Feb 15, 2013 at 11:53 AM, Wolfram Sang <wsa-z923LK4zBo2bacvFa/9K2g@public.gmane.org> wrote:
>
>> I'll re-send with I2C subsystem wide.  I probably won't attempt the
>> whole kernel wide at this point, but would be very happy if someone
>> else wanted to!  :)
>
> Thanks. Please double check that setting NULL is really unneeded for the
> non-platform-bus variants, too, or skip those if you are unsure. Please
> also update the commit message to reference a stronger indication than
> Stephen's "should not be necessary" ;) I do agree but for the commit
> history, a better reference would be nice to educate readers.

Sounds good.

I've it looks like non-platform-bus instances in drivers/i2c break down to:
* dev_set_drvdata() directly (even though we're a platform device)
* pci_set_drvdata()
* amba_set_drvdata()
* serio_set_drvdata()

I'll go ahead and include those in my patch.  I've looked through the
code a bit.  While I can't 100% guarantee that there's not some
strange code path that I'm missing, it all looks pretty
straightforward.  A few points to be made:

* It would be a bit hard to believe that some higher-level code could
make any particular assumptions about the value in the pointer (it
might point to static data, kmalloced data, etc).  ...so if the
higher-level code is directly looking at this value it'd really could
only be looking to validate that the driver put NULL here.  I don't
see that anywhere.

* I've checked all of the calls to xxx_get_drvdata() in the drivers
touched.  The majority are in remove, suspend, or resume, so we're
good there.  In other cases I just validated that they aren't checking
for and relying on a NULL result of xxx_get_drvdata().

* It would be possible that some subsystem tries to be helpful and
call xxx_get_drvdata() for you and then pass drvdata into a callback
(so I wouldn't see xxx_get_drvdata()).  I don't see that (it would be
a strange design anyway IMHO).  Most of the time I2C devices seem to
get their private data from adap->algo_data anyway (or from the
pointer passed to IRQ registration) and xxx_set_drvdata() doesn't
touch those.

* The __device_release_driver() function actually calls the
"dev_set_drvdata(dev, NULL)" for you anyway.  ...and I that's what's
running the remove code anyway.  ...so we'd only need to worry about
code that could show up before that...


-Doug