1 |
On Tuesday 31 May 2011 21:02:46 Alan McKinnon wrote: |
2 |
> Apparently, though unproven, at 21:27 on Tuesday 31 May 2011, Mick did |
3 |
> opine |
4 |
> |
5 |
> thusly: |
6 |
> > On Tuesday 31 May 2011 08:07:24 Pandu Poluan wrote: |
7 |
> > > On Tue, May 31, 2011 at 13:56, Alan McKinnon <alan.mckinnon@×××××.com> |
8 |
> > |
9 |
> > wrote: |
10 |
> > > > On Tue, May 31, 2011 at 3:14 AM, Pandu Poluan <pandu@××××××.info> |
11 |
wrote: |
12 |
> > > >> Meh, I clicked 'Send' too fast. |
13 |
> > > >> |
14 |
> > > >> *My* suggested solution: |
15 |
> > > >> |
16 |
> > > >> Generate an initramfs containing udev. The hands-down easiest way is |
17 |
> > > >> using genkernel's 'only create an initramfs' switch (sorry I forgot |
18 |
> > > >> what exactly). |
19 |
> > > > |
20 |
> > > > good god no, please, anything but genkernel. |
21 |
> > > > |
22 |
> > > > That thing is an attempt to emulate binary distros which require an |
23 |
> > > > initramfs to work properly (for any sane definition of "work") as the |
24 |
> > > > person building the installer has no idea what hardware the user will |
25 |
> > > > have. In Gentoo the user knows exactly what they have so there's no |
26 |
> > > > need for a gigantic hardware-detecting workaround at boot time. |
27 |
> > > > |
28 |
> > > >> This needs to be done exactly once throughout the life of your VM. |
29 |
> > > >> |
30 |
> > > >> (To the herd of Gentoo graybeards, feel free to CMIIW) |
31 |
> > > > |
32 |
> > > > Or wait a few days for vapier's (posting under his other name of |
33 |
> > > > spanky) sane advice to be implemented. His proposal is the sole voice |
34 |
> > > > of reason in that bug thread.... |
35 |
> > > |
36 |
> > > True. But I was having problem installing 2 servers on top of |
37 |
> > > XenServer. |
38 |
> > > |
39 |
> > > So I cheated and ran 'genkernel initramfs' exactly once. At least I |
40 |
> > > got myself a booting system. :-) |
41 |
> > > |
42 |
> > > When SpanKY's makedev gets stabilized and pushed to baselayout, I'll |
43 |
> > > then happily ditch the genkernel cheat for my next VMs :-) |
44 |
> > |
45 |
> > Are you sure that manually creating /dev/console and /dev/null isn't all |
46 |
> > that is required? The rest of the devices will be created by udev when |
47 |
> > it runs at boot time. |
48 |
> |
49 |
> null and console are the absolute irreducible minimum but there's one that |
50 |
> can be dispensed with if the correct kernel option is enabled. |
51 |
> |
52 |
> We don't need everything that makedev traditionally provided (like every |
53 |
> block device type known to man, floppys and ancient ptys) but the rest |
54 |
> number about ~250 and are useful in single-user mode if udev fails to |
55 |
> start. |
56 |
> |
57 |
> Considering that ~250 devices consumes a teeny-weeny bit of disk space and |
58 |
> they are hidden from view normally, I say it's worth it leaving them in. |
59 |
> Which is what vapier also says. |
60 |
|
61 |
I see. In my head it is as if we're going against the udev principle of |
62 |
populating required device nodes. If udev does not start, isn't it time to |
63 |
head for the nearest LiveCD, or must we ensure that every breakage is fixable |
64 |
in single-user mode? |
65 |
-- |
66 |
Regards, |
67 |
Mick |