[PATCH 4/9] gpiolib: use gpio_chips list in sysfs ops

[Alexandre Courbot]

On Sat, Feb 9, 2013 at 6:37 PM, Grant Likely <grant.likely at secretlab.ca> wrote:
> However, this code is incorrect (it was incorrect before you touched it,
> so not your fault). Moving it to a list makes it a lot worse though
> because it introduces the possibility of dereferencing an invalid
> pointer. The hooks need to grab the spinlock before parsing the list.
>
> I've got a patch which I'll post. I'd appreciate it if you can test
> also.
>
> In actual fact, I'd like to replace the spinlock with a mutex, but that
> requires some auditing to make sure there aren't any users from an
> atomic context.

I thought about doing that as well actually, but left it for later. It
looks safe at first sight, but as you mention we cannot exclude a few
users will break as a consequence of such a switch.

In any case, I plan to give it a try once that big chunk is merged.

Alex.