Kernel/user space (Was: Re: redhat problems)

[Gregory Nowak]

One problem I see with this idea is that if you get a kernel panic before the rest of the package loads, you're up a creek without a sightling (grin).

Greg


On Wed, Apr 16, 2003 at 09:24:30PM -0500, Luke Davis wrote:
> Okay, since we're going to have this discussion, let's have it, under a
> better subject...
> 
> Long before I started using Speakup, I was apposed to having it in the
> kernel, for all manner of reasons.
> 
> However, after using it a bit, and learning more about how it worked, I
> became less attached to that idea, and started looking at it as more of a
> driver, of the display type, and thus as something that needed to be in
> the kernel.  At least, my arguments against it, lost some major weight.
> 
> As it stands, I am happy with Speakup as it is--in the kernel.  I still
> maintain, however, that there may be a better way.
> 
> What I am looking at (unless Kirc, et al already did), is whether a hybrid
> solution is possible--part in the kernel, and part in user space.
> 
> The only part that (and this is said with an admited lack of knowledge on
> certain things, and is as such subject to change without notice) needs to
> be in the kernel, is what is, at minimum, required to access the consoles,
> and *maybe* talk to the synths.
> I am hoping that some parts can be moved out of the kernel, while still
> retaining the full functionality of Speakup as it is.  At the very least,
> buffering of data will be necessary.  What I mean is, that when output
> starts to a console (such as when booting starts), data will be buffered
> until the rest of the package, or at least what is needed to speak, is
> loaded, either from initrd, or from the root fs.  This is necessary for
> some hardware synths (DEC PC), and software synths, anyway, and as such
> should not be as strange as it initially looks.
> 
> The questions, as I see them, are:
> (1) is this a feasable idea; and
> (2) can enough of Speakup be moved into user space to make this worth
> doing?
> 
> I do not know the answers to those yet.
> 
> However, that is my active idea.  Pleese feel free to change my mind.
> 
> What I am thinking, is that if this can be made to happen, the kernel
> parts can be made highly stable, while the rest (managing the data which
> the kernel parts provide), can be as unstable as may be necessary.
> 
>  Luke
> 
>