| 1 |
On Wed, 2006-12-13 at 12:25 -0800, Greg KH wrote: |
| 2 |
> > Here's a question, then. Is there a way, after boot, to cause udev to |
| 3 |
> > attempt to detect and load modules? Basically, is there a way to |
| 4 |
> > emulate the old behavior of the "coldplug" init script? This would be |
| 5 |
> > very useful for our LiveCD builds, since we've lost the ability to do |
| 6 |
> > "nohotplug" on the command line to disable cold plugging. |
| 7 |
> |
| 8 |
> You can re-run /sbin/udevtrigger if you want to. |
| 9 |
|
| 10 |
Nice. |
| 11 |
|
| 12 |
> But why do it later? Why not just let udev do it at the start of boot? |
| 13 |
|
| 14 |
Well, it's pretty simple, we don't always want udev to do it. |
| 15 |
|
| 16 |
> What do you gain by delaying this? |
| 17 |
|
| 18 |
Delaying? Nothing. |
| 19 |
|
| 20 |
Allowing it to be disabled from the kernel command line is what I am |
| 21 |
looking for, here. Basically, we would disable udev cold plugging by |
| 22 |
default. Then, when we're actually to the point of being a bit more in |
| 23 |
control of the system, we check the kernel command line. If "nohotplug" |
| 24 |
is there, we do nothing. If it isn't, then we'd run /sbin/udevtrigger. |
| 25 |
This would "coldplug" the devices, but still allow us to retain |
| 26 |
functionality that we used to have with older coldplug versions. |
| 27 |
|
| 28 |
The only issue that we're now facing is that to disable coldplug also |
| 29 |
disables their services being auto-started, which means that even if we |
| 30 |
do run udevtrigger later, the services for those devices won't be |
| 31 |
started. I'm sure there's some solution to this, but this is really why |
| 32 |
I've been dreading newer udev versions going stable. While the newer |
| 33 |
versions are better, no doubt, it means we have to rethink how most of |
| 34 |
the auto-detection and configuration is done on the LiveCD. |
| 35 |
|
| 36 |
There have been many times where coldplug tried to load a driver that |
| 37 |
caused problems for the user, and being able to disable that driver |
| 38 |
being loaded has allowed them to continue to boot. With a coldplug that |
| 39 |
cannot be disabled arbitrarily, we cannot do this. |
| 40 |
|
| 41 |
-- |
| 42 |
Chris Gianelloni |
| 43 |
Release Engineering Strategic Lead |
| 44 |
Alpha/AMD64/x86 Architecture Teams |
| 45 |
Games Developer/Council Member/Foundation Trustee |
| 46 |
Gentoo Foundation |
Archives