Re: My understanding of the IDE mess, and why it does not make sense to apply the proposed patch
On Thu, Mar 09, 2006 at 11:52:01AM +0100, Jonas Smedegaard wrote:
> > I suggest going a bit further and _not_ loading ide-generic if a
> > specific driver has been loaded successfully and found some devices.
>
> I welcome the proposal of improving yaird to only load ide-generic if
> the device did not work with the possibly-problematic driver alone.
>
> yaird (unlike initramfs) wants to be certain at image build time,
> however, where IDE drivers cannot all be (unloaded and re)loaded one
> after the other until succesful. So the ide-generic module still needs
> to be included in the ramdisk for yaird to keep its principle of only
> reporting success if confident the ramdisk actually works.
To the best of my knowledge, the problematic non-probable ide drivers where
ever only a problem on x86.
Furthermore, if yaird detected that a given real driver is the right thing for
a given controller, then yaird should include this driver, and not
ide-generic. This the way yaird was designed, and we should know and fix it if
this fails.
If there is no suitable driver found for the controller handling the root
filesystem, i guess we have a problem, since this means that ide-generic may
be suitable, but there is no guarantee.
I suppose that it is more in touch with the yaird way to fail there. I don't
know how the ide-generic module would show up in sysfs though, but it is
probable yaird will be able to detect ide-generic is to be used.
This shows a fundamental problem of yaird, namely, i don't think it can adapt
and do the right thing if the currently running kernel is mis-configured.
> So for this to solve the issue at hand of yaird not working with
> powerpc kernels not building ide-generic at all, yaird would have to
> drop support for kernels older than those with known-working IDE
Well, there are no such kernels in debian anymore, so it should be no problem
to do so, but it has not been proven that there ever where such problematic
cases.
> drivers. This seems to be 2.6.15 (based on Svens code review rather
> than my using rumors from yaird upstream notes).
Notice that i participated in the discussion which lead upstream to write
those notes, and already back then i think i commented that i thought this was
not the right way of doing it.
Always look at the code, and not third-hand rumors :)
> > Reason? At least on my laptop ide-generic takes too long, about 30
> > seconds, apparently doing probing for devices that do not exist. I can
> > disable it with kernel commandline parameters but it still sucks.
> >
> > And I really doubt that there are that many users out there that have
> > their root filesystem on a disk attached with a legacy ISA controller
>
> Sounds sane that the main purpose of ide-generic is for use with ISA
> cards and other strange beasts.
>
> I believe the abuse of ide-generic to make other drivers work is a
> different matter, however.
Ah, ? Can you give any serious proof that this is not again plain FUD ? Please
take the time to go over those old bug reports, and isolate the problem so you
know it is a real problem, and you don't make a huge mountain of what is
probably nothing more than a confusion or mistake back then.
> So please file a separate wishlist bugreport against yaird to only use
> the ide-generic workaround after failed test of the driver working
> without it.
And will you be able to fix it then ? Will you back out the ide-generic patch
that was introduced in 0.0.12 ? What are your real intentions about this ?
Friendly,
Sven Luther
Reply to: