Re: [lm-sensors] [PATCH] hwmon class driver registration with a device number

[Himanshu Chauhan <hschauhan@nulltrace.org>]

On Thu, Oct 06, 2011 at 09:43:02AM +0200, Jean Delvare wrote:
>

Thanks Jean for summing up the questions :)
 
> I again agree. Which means that Himanshu is still 3 steps away from
> getting his patch accepted:
> * Explaining why the current sysfs interface is insufficient and can't
>   be fixed.

AFAIK, the current sysfs interface is to read standard attributes from
the device. My reasoning is:

1. This will give hwmon devices to have their "dev" in their sysfs folder
   as like other classes and in turn be visible in /dev by mdev or udev.

2. Implementing IOCTL if required. I know that kernel hackers believe
   they are "backdoors" and "can't be done right" but sometimes there
   isn't any get away with them. I don't know when their support will
   be completely removed. Rather, I see that instead of deprecation,
   new IOCTL vectors like compat_ioctl have been introduced.

> * Getting official device node numbers registered for hwmon use.

Jean, as long as drivers get major device numbers on the fly, I don't
think this is a requirement. Isn't it?

> * Defining an API for these device nodes.
My idea is just this:
1. Get a major number from kernel on the fly.
2. During probing, for each device, call "hwmon_device_register_numbered"
   and let mdev create a /dev node for it. I don't say that this interface
   be imposed on drivers. If they want they can still call "hwmon_device_register"
   if they don't want to implement standard ioctl, read, write calls.
 
> Before then, there's no point in resending this patch, it will be
> rejected.
> 

Thanks for reviewing. These are my thoughts. If you guys think the reasoning
isn't strong enough to get this in mainline, I am happy to accept that :)
At least I know what can get rejected for sure and I am happy to build a
stack of rejected patches for that. :)

-Himanshu