ubi_wl_get_peb: Unable to get a free PEB from user WL pool

[Jörg Krause]

On Do, 2015-08-06 at 09:30 +0200, Richard Weinberger wrote:
> Jörg,
> 
> Am 06.08.2015 um 00:17 schrieb Jörg Krause:
> > > > > Then just disable it. :)
> > > > 
> > > > I did :) I thought fastmap is used automatically by the UBI 
> > > > driver?
> > > 
> > > It will *use* fastmap automatically but not convert non-fastmap 
> > > UBI 
> > > images
> > > to fastmap.
> > > You can enable ubi.fm_autoconvert to do so.
> > 
> > Sorry for asking, but how do I create an UBI image with fastmap
> > support? I've never read something about this.
> 
> You have to use the ubi.fm_autoconvert parameter. The UBI will 
> automatically
> write down a fastmap if needed.
> 
> Currently it is not possible to create a fastmap-enabled UBI image
> using ubinize.
> The reason is that fastmap stores the LEB->PEB mappings to allow
> a fast attach. ubinize cannot know which PEBs are bad.
> So, we'd need a tool which creates the fastmap while flashing to the 
> nand.
> 

I see!

> > > > > Meanwhile I'll find out what exactly is going on.
> > > > 
> > > > Many thanks!
> > > 
> > > Can you please apply the attached debug patch?
> > > I wonder why exactly the auto resize function is failing.
> > 
> > autoresize: cannot auto-resize volume 0: -28
> 
> I have good and bad news.
> The good news is that I know what is going on, the bad one
> is that I don't have an good solution for it.
> 
> UBI is requesting a new PEB but your system is still busy with
> erasing them. With fastmap disabled UBI can use produce_free_peb()
> and wait for a new erased PEB.
> But fastmap cannot use this helper as it releases ->wl_lock
> and would lead to a inconsistent fastmap state.
> Kernels < 4.1 did so and faced interesting fastmap corruptions.
> I hoped using the current retry logic the issue you're facing cannot
> happen. But as it seems I was wrong.
> 
> Can you please give the attached patch a try?
> 

I tried the "dirty hack", but it does not solves the problem. I also
tested with 100 retries, but the error remains.

I will turn fastmap off for now, but feel free to send more patches to
try.

Jörg