[9fans] kernel modules

[9fans] kernel modules

[Fco.J.Ballesteros]

Maybe I'm wrong, because I'm kind of sleepy, but I think
that you said that the plan was to add loadable module
support and after that, maybe put usb (and perhaps others)
into the kernel.

I've been thinking about it and I think it's not a problem
at all to have everything compiled inside the kernel.
I know this does not include all allocated memory in use
by the kernel, but

872411t + 1555144d + 108904b = 2536459	9pcdisk

2.5 Mbytes (say, 5 Mbytes while running)
is a pretty low ammount of memory this days. Of course this
does not account for fossil et al.

Isn't it enought just to be able to power on/off the
different devices? Even when they contend for a resource,
we can shut down all but the one that we want. There's no
need to load/unload them.

Is the complexity added worth? For instance,
my kernel is a 1.9% of the machine memory?? (maybe a 4%
while running, but that's in a 128M machine, which
is pretty low these days).

I'd love to see this thing stay as simple as feasible;
just had to say it.

Re: [9fans] kernel modules

[Charles Forsyth]

dynamically-loaded modules are useful--perhaps even essential--
for some things, but most usb work still doesn't really require
kernel support for much more than managing the IO streams once the
device is configured.  the interface to those streams
needs to be improved, partly because the spec was ambiguous
about one point, and partly because of experience with it,
but having looked at the usb support in other systems, i think
it's good to aim to keep it out of the kernel.

Re: [9fans] kernel modules

[ron minnich]

On Sat, 8 Nov 2003, Fco.J.Ballesteros wrote:

> Is the complexity added worth? For instance,
> my kernel is a 1.9% of the machine memory?? (maybe a 4%
> while running, but that's in a 128M machine, which
> is pretty low these days).

my modules tree for my laptop, which is hardly complete, is 14MB. Other
machines have 24 MB.

The idea still might work, though. If modules can be avoided that is all
to the good ...

ron

Re: [9fans] kernel modules

[Fco.J.Ballesteros]

[-- Attachment #1: Type: text/plain, Size: 140 bytes --]

Ok. So you'd turn my 128M machine into a, say, 100M machine.
Not a big deal... given that we indeed need those extra Mbytes
for devices.

[-- Attachment #2: Type: message/rfc822, Size: 2730 bytes --]

From: ron minnich <rminnich@lanl.gov>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] kernel modules
Date: Mon, 10 Nov 2003 07:34:30 -0700 (MST)
Message-ID: <Pine.LNX.4.44.0311100731200.5100-100000@maxroach.lanl.gov>

On Sat, 8 Nov 2003, Fco.J.Ballesteros wrote:

> Is the complexity added worth? For instance,
> my kernel is a 1.9% of the machine memory?? (maybe a 4%
> while running, but that's in a 128M machine, which
> is pretty low these days).

my modules tree for my laptop, which is hardly complete, is 14MB. Other
machines have 24 MB.

The idea still might work, though. If modules can be avoided that is all
to the good ...

ron

Re: [9fans] kernel modules

[Sape Mullender]

> Is the complexity added worth? For instance,
> my kernel is a 1.9% of the machine memory?? (maybe a 4%
> while running, but that's in a 128M machine, which
> is pretty low these days).

It could be worth while in embedded systems using Plan 9...

Re: [9fans] kernel modules

[Fco.J.Ballesteros]

[-- Attachment #1: Type: text/plain, Size: 122 bytes --]

But then, isn't it reasonable to keep a 9embed
kernel compiled with just a subset of devices for that
sort of platforms?

[-- Attachment #2: Type: message/rfc822, Size: 1892 bytes --]

From: Sape Mullender <sape@plan9.bell-labs.com>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] kernel modules
Date: Mon, 10 Nov 2003 09:38:18 -0500
Message-ID: <ef8639e23f04a0595330978dd74d1c0e@plan9.bell-labs.com>

> Is the complexity added worth? For instance,
> my kernel is a 1.9% of the machine memory?? (maybe a 4%
> while running, but that's in a 128M machine, which
> is pretty low these days).

It could be worth while in embedded systems using Plan 9...

Re: [9fans] kernel modules

[ron minnich]

On Mon, 10 Nov 2003, Fco.J.Ballesteros wrote:

> Ok. So you'd turn my 128M machine into a, say, 100M machine.
> Not a big deal... given that we indeed need those extra Mbytes
> for devices.


yes, the only other issue is that even on my little laptop this is 430
devices, so probing will slow up just a bit, as you do the mxn search for
'what device is that anyway'. We probably want a better way to figure out
what's on the bus and what driver it is. Probably something like 24.5M of
that 26M kernel image will be unused. If you don't care, I guess I don't
either.

ron

Re: [9fans] kernel modules

[ron minnich]

On Mon, 10 Nov 2003, Fco.J.Ballesteros wrote:

> But then, isn't it reasonable to keep a 9embed
> kernel compiled with just a subset of devices for that
> sort of platforms?


I think so. Embedded stuff is very fixed in configuration. So lock that
down.

Given the 10-20 second compile time of plan 9 kernels, seems like you
figure out that maximum set of devices you care about for your laptop, and
build that. Then hot plug becomes, as you point out, just turning a driver
on and off, not unloading/loading a module.

ron

Re: [9fans] kernel modules

[Sape Mullender]

> > It could be worth while in embedded systems using Plan 9...

> But then, isn't it reasonable to keep a 9embed
> kernel compiled with just a subset of devices for that
> sort of platforms?

Probably.  But it's a hassle to maintain lots of different kernels.