Re: [PATCH v4] uio/pdrv_genirq: Add OF support

[Grant Likely <grant.likely-s3s/WqlpOiPyB63q8FvJNQ@public.gmane.org>]

On Wed, May 4, 2011 at 1:47 PM, Hans J. Koch <hjk-vqZO0P4V72/QD6PfKP4TzA@public.gmane.org> wrote:
> On Wed, May 04, 2011 at 03:21:55PM +0200, Michal Simek wrote:
>> Hans J. Koch wrote:
>> >On Tue, May 03, 2011 at 10:34:12PM +0200, Wolfram Sang wrote:
>> >>On Mon, May 02, 2011 at 08:51:55AM +0200, Michal Simek wrote:
>> >>>Adding OF binding to genirq.
>> >>>Version string is setup to the "devicetree".
>> >>>
>> >>>Compatible string is not setup for now but you can add your
>> >>>custom compatible string to uio_of_genirq_match structure.
>> >>>
>> >>>For example with "vendor,device" compatible string:
>> >>>static const struct of_device_id __devinitconst uio_of_genirq_match[] = {
>> >>>   { .compatible = "vendor,device", },
>> >>>   { /* empty for now */ },
>> >>>};
>> >>>
>> >>>Signed-off-by: Michal Simek <monstr-pSz03upnqPeHXe+LvDLADg@public.gmane.org>
>> >>[...]
>> >>
>> >>>+          /* alloc uioinfo for one device */
>> >>>+          uioinfo = kzalloc(sizeof(*uioinfo), GFP_KERNEL);
>> >>kfree in remove?
>> >
>> >Oh yes. Missed that one. It should probably look like the "bad0" case in probe().
>>
>> Yes, freeing uioinfo in uio_pdrv_genirq_remove make sense for CONFIG_OF.
>>
>> Please correct me if I am wrong dev.of_node is not NULL for OF. I
>> think yes that's why I would prefer to use this construct instead of
>> #ifdef CONFIG_OF.
>>
>>       if (pdev->dev.of_node)
>>               kfree(pdev->dev.platform_data);
>
> Huh? You didn't allocate platform_data, so you shouldn't free it.
> It's uioinfo you allocated.

*never* set or modify the pdev->dev.platform_data pointer from a
device driver.  If the data needs to be modified, or a new structure
allocated, then rework the driver to store it in it's private data
structure.  platform_data must be considered immutable by drivers.
Otherwise you'll break unbindng and rebinding device drivers.

g.