Re: udevsynthesize as a udevstart+coldplug replacement

[md@Linux.IT (Marco d'Itri)]

[-- Attachment #1: Type: text/plain, Size: 1183 bytes --]

On Aug 25, Bill Nottingham <notting@redhat.com> wrote:

> USB controller scans bus, sends events from kernel.
This is something I have been thinking about recently: events for a bus
should be synthesized only if the bus driver was not loaded at the time
the coldplug program is started, or they would be duplicated.
What needs to be checked in sysfs?
I think that all existing coldplug implementations suffer of this bug.

> So, in theory, this 'udevd is done' could be signaled at
> either A, B, or C, as I understand things.
I meant to do it for A, which would be like how udevstart currently
works.

> Speaking of that, what prioritization is there in udevsynthesize?
> Realistically, I think you'd always want to load USB and firewire
> controller modules last when scanning the PCI bus, as a means of
> enforcing 'polite' device ordering (while you can't always
> ensure LANANA device names are consistent, it's polite to make
> sure your USB stick doesn't jump to /dev/sda ahead of your SATA drive.)
I agree that this is desired (eth1394 becoming eth0 is a classic
example), but I am not sure about how hard it would be to implement.

-- 
ciao,
Marco

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]