On Wed, 1 Oct 2008, Grant Grundler wrote:

quoted text
> >

> > And then we can create a separate set of these dependency points, so that

> > X and Y don't have to know about each other, they just have to have some

> > knowledge about some common synchronization point - one that exists

> > regardless of whether X or Y are even compiled in!

>

> We already do this today. :)

> Definitions are in include/linux/init.h.

Absolutely. 
The problem with the current  isn't that it doesn't work -

it's worked pretty well for a long time - it's that we continually tend to

hit the limits of the few fixed points.
I would just extend on that notion a bit, and also make the markers a bit

more dynamic. Instead of having just 7-8 levels of initcalls, and a very

fixed naming that is a bit misleading, I'd like to extend it to maybe 50

levels, and the levels would be named by the subsystems and then just have

one single place that orders them.
The old 7 levels (plus the "pure" one) would still exist, so it wouldn't

need to be a flag-day event.
For example: why would you use "fs_initcall()" for the PnP init? The

reason is simple: it's not a filesystem init, but it's the one that comes

after the subsys init and before the actual low-leveld drivers. We used to

initialize the core filesystem data (VFS) at that stage (we still do,

although most of the actual filesystems are actually just done using the

default module-init much later), which explains the naming, but the

problem is that
 (a) we want to order things at a finer granularity than that

 (b) we want to have a better and more descriptive naming
but the basic notion of having a fixed ordering certainly isn't wrong (and

I already argued against any _dynamic_ one).
> Point A would be "early" ("run before initialing SMP")

quoted text
> The rest could use better definitions and AFAICT aren't that much better

> than being named "Point B".

It would be *much* better to give them symbolic names (easy enough - we

just turn them into sections that are symbolic anyway), and then have a

list of ordering for those symbolic names.
So they sure as hell would be much better than being named "Point B". We

could get rid of
	subsys_initcall(pci_init)
and instead do
	initcall(pci_init, "pci");

	..

	initcall(pnp_init, "pnp");
and then just list the levels for the linker scripts in one place. So that

we wouldn't really have to worry about link ordering: if the link ordering

is wrong, we just add a new initcall level and insert it in the right

place, and then we can look at the ordering and see it explicitly in one

place instead of looking at the makefiles and checking the order we add

object files to the list in!
		Linus

--

unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in

the body of a message to majordomo@vger.kernel.org

More majordomo info at  http://vger.kernel.org/majordomo-info.html

Please read the FAQ at  http://www.tux.org/lkml/