The progress of the device registry is ok
but could be better. Coding of the framework is finished,
I still have to add the new registration stuff to the
drivers and then hope that everything works fine without
too much debugging. Beside that I started writing the
documentation for driver developers. I think this is the
right time to do this since I'm pretty confident that most
of the kernel API will be stable from now on.
config fs: I'm planning to abandon my proc fs
changes in a future version and instead write new
filesystem based on the proc fs code. Proc FS is a mess.
While most of the files
in /proc are quite convenient to read for human readers
they are almost unusuable for software. The problem is not
that parsing would be so difficult but that it is
impossible to add new fields once you have defined a text
format without breaking all software that reads it. Binary
files suffer from the same compatibility problems. A few
months ago several people convinced me on LKML to go the
"one-value-per-file" way for devreg, and the new
filesystem will enforce this even more than my proc fs
extensions. Every file will be strictly typed (string,
enum), dynamic directories will be possible, the type of a
can be retrieved using ioctls, and unlike my current proc
changes writing will be supported, too.
The more interesting part of the plan is that I want to
support both persistent and non-persistent files. So there
will be some daemon waiting, similar to devfsd, for files
that have been flagged as persistent and will then store
their new values when they have been changed. On
of a file the daemon will fill it with a previously stored
value. Dynamic directories can have keys that can be used
to find the correct one. I'm currently calling this
concept config fs. Beside the replacement of proc fs for
everything but processes it can be used for the
configuration of the kernel and drivers. The current way
of configuration definitely sucks, module options are a
bad idea for most purposes because you cannot change them
at runtime and they are not very useful if your driver
handles several devices at the same time. They can only be
used for the driver's global configuration, but not for
the configuration of driver instances. IMHO module options
should disappear for anything but boot-time configuration.
The only frightening thing about the config fs idea is how
close it is to Win2k's registry. It's a little bit less
arcane because you can access it as a filesystem, but
beside that there are only few differences between them.
It could even make sense to use it for the general
configuration of system software. bah...