| 1 |
Luke-Jr wrote: |
| 2 |
> On Wednesday 06 April 2005 21:55, Christian Parpart wrote: |
| 3 |
> |
| 4 |
>>On Wednesday 06 April 2005 11:42 pm, Karl Zander wrote: |
| 5 |
>> |
| 6 |
>>>I was going to upgrade to a 2.6 kernel, but then found out I really |
| 7 |
>>>should stay with a 2.4 kenel. (The 2.6.x Win4Lin sources I need are hard |
| 8 |
>>>masked in portage.) I had already emerged udev before I checked this. |
| 9 |
>>>My bad for not checking. But can I go back to devfs? How? Do I need |
| 10 |
>>>to? |
| 11 |
>> |
| 12 |
>>Um, yeah, you can. Of course. But why do you want to? |
| 13 |
>>(see /etc/conf.d/rc at RC_DEVICES="devfs" / ="udev") |
| 14 |
> |
| 15 |
> |
| 16 |
> udev is less tested and doesn't work "out of the box" |
| 17 |
> |
| 18 |
> |
| 19 |
>>What do you need, you believe devfs can, but udev can't? |
| 20 |
> |
| 21 |
> |
| 22 |
> Automatic module loading, for one. Devfs will load the modules needed when you |
| 23 |
> try to access a device (eg /dev/cdroms/*). With udev, you need to load |
| 24 |
> modules to use them. While cold/hotplug will detect and load many modules, it |
| 25 |
> also lacks detection for many common devices (such as IDE CDROM drives). |
| 26 |
> And, no, I don't consider modules.autoload (or whatever file it is) to be a |
| 27 |
> solution. More of a bad hack to workaround bad detection. Thus, I don't |
| 28 |
> intend to have any of my systems make use of it. |
| 29 |
|
| 30 |
Greg Kroah-Hartman addresses this complaint in the FAQ: |
| 31 |
|
| 32 |
http://www.kernel.org/pub/linux/utils/kernel/hotplug/udev-FAQ |
| 33 |
|
| 34 |
and frankly I think it makes sense. He states that udev is intended to |
| 35 |
manage device nodes and nothing else and that "all devices present on |
| 36 |
the system should generate hotplug events, loading the appropriate |
| 37 |
driver". That seems fair enough to me; let it do one job and do it well. |
| 38 |
|
| 39 |
His assertion about the problematic nature of the "devfs approach" is |
| 40 |
quite right. I've had my fair share of nasty "append to parent" errors |
| 41 |
and other wierdness with devfs, particularly in 2.6. I also heard from a |
| 42 |
developer who had some serious issues when trying to perform various |
| 43 |
LVM2 related options ... until he moved to udev. And udev has some |
| 44 |
tricks up its sleeve too (a simple way of having a consistent naming |
| 45 |
policy for devices, for one). |
| 46 |
|
| 47 |
Regardless of the relative merits, the maintainer has gone AWOL and |
| 48 |
devfs is on the way out. I would much rather use udev in 2.6, although |
| 49 |
devfs is more functional in 2.4. |
| 50 |
|
| 51 |
Regards, |
| 52 |
|
| 53 |
--Kerin Francis Millar |
| 54 |
-- |
| 55 |
gentoo-server@g.o mailing list |
Archives