block2mtd and ubi are initialized too early when compiled in on 2.6.31-rc2

[Artem Bityutskiy]

Hi,

On Wed, 2009-07-15 at 13:52 +0200, Tobias Diedrich wrote:
> On 2.6.31-rc2 the block2mtd drivers module_init is called before any
> block devices have been registered.

Hmm, ok. Is this because block devices are registered asynchronously?
Could you please point to the code where it is done, just for reference.

> Also ubi is initialized pretty early and fails completely if an mtd
> specified on the command line could not be found.

Hmm...

> IMO ubi should at least complete initialization so that attaching
> the mtd later with ubiattach would still work.
> I'm working around this two hacky patches that create a kernel
> thread and retry every second for 20 seconds when the first try
> doesn't work.
> (Obviously this means rootdelay=$somenumber is needed)
> I tried using the async infrastructure, but apparently
> async_synchronize_full is called somewhere between registering the
> async probe thread and the target USB device being registered by the
> USB subsystem, which halts boot until my 20 second timeout, and the
> USB stick is only detected afterwards.
> 
> FWIW I want to use a erasesize aware FS on my USB stick (whose
> builtin FTL has abysmal write performance if writes are less than
> the erasesize) and also be able to use this as my root fs.
> So my setup is usb_storage->block2mtd->ubi->ubifs

Hmm, how other subsystems solve this problem? Any pointer to the code?

-- 
Best regards,
Artem Bityutskiy (Битюцкий Артём)