Deliverable 4: Proposed boot process is available for review
(SoC2006)
Henrique de Moraes Holschuh
hmh at debian.org
Mon Sep 11 21:56:42 UTC 2006
On Mon, 11 Sep 2006, Henrique de Moraes Holschuh wrote:
> It is *probably* safe for all i586 and better hardware, but Debian kernels
> have the RTC as a module. That thing can fire up in the middle of a
> parallel boot because we are doing things half-assed in the early userspace
> setup right now (and I do not mean with the parallel boot complexity added
> in).
This did not came up right, so I better explain before people think I am
against parallel boot or whatever... Our early userspace setup is badly
done right now (etch/sid), even without any parallel booting or reordering
because:
1. There is no guarantee of correct system time (at least the timezone is
now fixed in Etch and Sid...) at the moment the stuff that needs it start
running:
(a) we check filesystems before it is properly read from the RTC, unless
the kernel did it for us
(b) we run all coldplug in one go, and we have no rules on what these
scripts can do. Some of them are likely to need the system time to be
correct (e.g. calls to "mount", etc).
2. There is no early rw filesystem that can be used for general tasks.
When you drop in parallel boot on top of it, you have also the added
aggravation that nothing is syncronized to udev coldplug (and doing it ain't
a picnic either :p), and stuff that must be a chokepoint (like messing with
the system clock) isn't.
Userspace is slowly moving to become fully assynchronous, but for THAT to
work right, we need to take some care to have a few essentials setup before
anything weird can happen :(
--
"One disk to rule them all, One disk to find them. One disk to bring
them all and in the darkness grind them. In the Land of Redmond
where the shadows lie." -- The Silicon Valley Tarot
Henrique Holschuh
More information about the initscripts-ng-devel
mailing list