The 2.6.32 kernel is out [LWN.net]

The 2.6.32 kernel is out

Posted Dec 3, 2009 15:27 UTC (Thu) by sbergman27 (subscriber, #10767) [Link]

Arjan van de Ven knows a little about fast booting:

and he says:

"""
This devtmpfs is not need to boot fast. Really.
This is a workaround for a certain distros crappy mkinitrd basically, and nothing more; if
you do the initrd correct (or if you don't use an initrd at all), you don't need this "solution"
and you'll even boot faster....
"""

http://lwn.net/Articles/332190/

Boot speed being GKH's main pitch point for the "feature". I'm just a lowly sysadmin, and
not a kernel developer. But I'm not at all convinced that the current methods need to be
ripped out and replaced with this. And I wouldn't be even if it did save a second or two at
boot. Fix what we have, like Arjan suggests.

People sometimes criticize Linux development as running around in circles, ripping out and
replacing subsystems, willy-nilly. And in this particular case, that criticism looks valid.

The 2.6.32 kernel is out

Posted Dec 3, 2009 15:31 UTC (Thu) by Cyberax (subscriber, #52523) [Link]

"But I'm not at all convinced that the current methods need to be
ripped out and replaced with this."

WTF? Devtmpfs does not 'rip out' anything. Final /dev tree is still managed by udev, as usual.

The 2.6.32 kernel is out

Posted Dec 3, 2009 17:28 UTC (Thu) by dmk (subscriber, #50141) [Link]

GKH insists that devtmpfs is useful even if it dosn't in itself improve bootspeed. it's a convenience so that you are not dependent on udev being operable... (e.g. init=/bin/bash boot) and you can easily now skip the cold-plug step that udev needed until now. if the kernel discovers the device, creates sysfs-nodes, why shouldn't it also provide some basic device-nodes?

and as stated in another comment: udev still manages as before (rights/permissions). it just dosn't create the node if it is already there..

the only point that could be made against devtmpfs would be, that the kernel has to have no say in device-permissions and even default-device permissions are too much policy decisions... but i don't support this view.

if one feels strongly about devtmpfs... it is possible to just say no at configure time....

The 2.6.32 kernel is out

Posted Dec 3, 2009 18:43 UTC (Thu) by arjan (subscriber, #36785) [Link]

Greg has other reasons for devtmps, and on those merits it got merged.

yes boot speed is a red herring (and that is the ONLY thing I responsed about); the other reasons stand on their own.

The 2.6.32 kernel is out

Posted Dec 4, 2009 4:47 UTC (Fri) by SEJeff (subscriber, #51588) [Link]

I wasn't fully convinced by Arjan's argument in relation to a _general
purpose_ distro. This especially holds true because he is the rockstar who
helped Moblin boot in 5 seconds on very specific hardware. Moblin is no
Fedora, it is no Ubuntu, or opensuse, etc. He thinks that modular kernels are
fail because modprobe is slow. Besides more TLBs for separate modules, they
make perfect sense for general purpose distros.

devtmpfs is kind of like another tool in a swiss army knife. Just because you
don't always use the can opener doesn't mean someone else doesn't and won't.
Them using it more than you doesn't make them wrong. Each achieves the same
end goal. That being said, Arjan is certainly doing amazing things for Linux
as a whole and hopefully he keeps it up.

The 2.6.32 kernel is out

Posted Dec 4, 2009 19:45 UTC (Fri) by kjp (subscriber, #39639) [Link]

I for one am happy to see devtmpfs merged. It shrinks your initramfs if you use one... you don't need udev there. You only need udev on your root partition (general distro case here).

Given all the other crap the kernel puts by default in sysfs, it seems like a no brainer to me to also create a default device node.

The 2.6.32 kernel is out

Posted Dec 4, 2009 17:06 UTC (Fri) by crimsun (subscriber, #13750) [Link]

What do you mean by "latest"? 2.6.32 certainly doesn't ship what's in sound-2.6 (origin/master) HEAD. It's for-linus, which is essentially 1.0.21 and some fixes.

The 2.6.32 kernel is out

Posted Dec 4, 2009 23:03 UTC (Fri) by Tuxie (guest, #47191) [Link]

By "latest" I meant version 1.0.21, which is supposedly the first version to have working more-than-2 channel LPCM audio over HDMI on NVidia chipsets.