DragonFly kernel List (threaded) for 2004-04
Re: devfs vs udev/hotplug
Dave Leimbach wrote:
Matthew Dillon <dillon@xxxxxxxxxxxxxxxxxxxx> writes:
:> What do people think of the udev/hotplug solution vs devfs? At first
:> inspection, udev appears to be a little more Dragonfly-ish because of
:> its placement in userland.
:I'd vote for udev (or something like it)
I think I would prefer udev over devfs. It seems silly to try to manage
yet another fake filesystem in the kernel, just look at all the bugs
that have cropped up in existing devfs and procfs implementations, not
to mention unionfs and nullfs! You'd think we would have learned our
lesson by now!
I would definitely prefer a userland demon which performs the work based
on what the kernels tells it to do. The kernel can still assign dynamic
minor numbers for dynamic devices, and that is really the crux of the
functionality we want to have.
Doesn't this sort of design imply that the kernel is instructing a daemon
to make kernel calls to service the requests? Isn't that a lot of
address space crossing or do I not understand how a userland file "server"
would work? I realize that this will be done with fast messages but such
designs leave me with "speed" doubts. This might be a bad example for when
that speed matters though.
I don't think speed will be much of an issue, once Async
messaging and a Shared Memory Segment between userland and
kernel is implemented. One can utilize that.