[PATCH] [RFC] UBI: Implement Fastmap support

[Shmulik Ladkani]

Hi Artem,

On Thu, 24 May 2012 11:17:52 +0300 Artem Bityutskiy <dedekind1 at gmail.com> wrote:
> On Tue, 2012-05-22 at 18:55 +0200, Richard Weinberger wrote:
> > >> +    e = find_early_wl_entry(&ubi->free, max_pnum);
> > >
> > > This picks the eb with the lowest pnum within 'ubi->free'.
> > >
> > > When called with INT_MAX (for the FM_DATA), why do you need to pick
> > a
> > > free eb with the minimal pnum? The FM_DATA EBs may reside everywhere
> > (as
> > > the FM_SB holds their location).
> > > So why not pick the eb with a medium EC value (as done for standard
> > > get_peb calls)? That might be better wear-leveling wise.
> > 
> > Fair point.
> > I'll fix that.
> > Artem, any comments on that?
> 
> The 'find_early_wl_entry()' function is used (currently) only at early
> stages. At these stages the we do not have the PEBs sorted by EC. We
> have just a list. This function should not be use after the WL subsystem
> is initialized.

'find_early_wl_entry' is only called from 'ubi_wl_get_fm_peb'.

'ubi_wl_get_fm_peb' is called twice from within 'ubi_update_fastmap':
First call, to get the FM_SB, with 'max_pnum' set as UBI_FM_MAX_START.
Second, to get FM_DATA pebs, with 'max_pnum' as -1, do indicate "no
matter the location, give me pebs from the free pool".

'ubi_update_fastmap' is called from 'ubi_volume_notify' and from
'ubi_wl_get_peb' - at both points, I assume ubi->free rbtree is properly
populated. Am I mistaken?

Regards,
Shmulik