On Thu, 2012-06-28 at 20:53 +0300, Shmulik Ladkani wrote:
> On Mon, 25 Jun 2012 17:08:38 +0200 Richard Genoud <richard.genoud@gmail.com> wrote:
> > diff --git a/drivers/mtd/ubi/misc.c b/drivers/mtd/ubi/misc.c
> > index f6a7d7a..c2c6db0 100644
> > --- a/drivers/mtd/ubi/misc.c
> > +++ b/drivers/mtd/ubi/misc.c
> > @@ -100,8 +100,9 @@ void ubi_calculate_reserved(struct ubi_device *ubi)
> >  {
> >  	ubi->beb_rsvd_level = ubi->good_peb_count/100;
> >  	ubi->beb_rsvd_level *= CONFIG_MTD_UBI_BEB_RESERVE;
> 
> BTW Artem, I've always had an unresolved issue with this original
> calculation... maybe you can shed some light here.
> 
> Why 'beb_rsvd_level' is set to RESERVE percent of the 'good_peb_count'?

I thought it was a good idea I guess :-)

> As the device gets worn, number of 'good_peb_count' will lessen over
> time - and as such, the 'beb_rsvd_level' may vary (lessen) over time.

Yeah, probably you are right.

> I'd expect a fixed number of 'beb_rsvd_level' PEBs for a given mtd
> partition, or more correctly, as Richard suggests, the *sum* of bad PEBs
> plus the beb reserved PEBs should be constant for a partition - as I
> do not expect more than a known constant of blocks to go bad during
> device's (and thus, partition's) lifetime.

Those days we did not have this "vendor-guaranteed max. bad blocks
count" thing and I thought that UBI would try to always maintain a pool
of reserved PEBs.

Would you send a patch?

Note, one thing: if I already marked max. possible amount of PEBs as
bad, I just do not reserve any more. But if I have a plenty of available
PEBs anyway (my volumes are smaller than they could be, or I shrink one
of them), and another PEB goes bad, I should just mark it as bad. I may
print a warning, but should not panic.

-- 
Best Regards,
Artem Bityutskiy