Re: [patch] kernel events layer

[Robert Love <rml@ximian.com>]

On Sat, 2004-07-24 at 11:02 +0800, Michael Clark wrote:

> Should there be some sharing with the device naming of sysfs or are
> will we introduce a new one? ie sysfs uses:
>
> devices/system/cpu/cpu0/<blah>
>
> Would it be a better way to have a version that takes struct kobject
> to enforce consistency in the device naming scheme. This also means
> userspace would automatically know where to look in /sys if futher
> info was needed.

No, we want to give an interface that matches the sort of provider URI
used by object systems such as CORBA, D-BUS, and DCOP.  We also do _not_
want to put policy in the kernel.

The easiest way to avoid that is simply to use a name similar to the
path name.

Passing the sysfs name would probably be a good potential argument to
the signal, though.  The temperature signal in the patch is just an
example.

> Question is does it make sense to use this infrastructure without sysfs
> as hald, etc require it. ie depends CONFIG_SYSFS

That sounds like policy to me.

Especially if drivers start using this for error logging, there are no
ties to sysfs.  Configuration dependencies tend to be hard build-time
deps anyhow.

	Robert Love