[U-Boot] Porting UBI fixes (specially fastmap's) to U-Boot

[Heiko Schocher]

Hello Ezequiel,

Am 19.10.2015 um 15:47 schrieb Ezequiel Garcia:
> On 19 October 2015 at 02:17, Heiko Schocher <hs at denx.de> wrote:
>> Hello Ezequiel,
>>
>> Am 17.10.2015 um 20:07 schrieb Ezequiel Garcia:
>>>
>>> Hi Heiko,
>>>
>>> On 9 October 2015 at 09:30, Heiko Schocher <hs at denx.de> wrote:
>>> [..]
>>>>
>>>>
>>>>
>>>> I just updated the "ubi_sync_with_linux" branch on u-boot-ubi.
>>>>
>>>> It seems UBI/UBIFS now work with the NAND on the aristainetos2
>>>> board, but my stomach says, there are some subtile issues ...
>>>>
>>>> Still needs more testing, also not tested yet UBI on the SPI NOR on
>>>> this board.
>>>>
>>>> If I "nand erase" the mtd device, and try "ubi part ..." it fails
>>>>
>>>> :-(
>>>>
>>>> But If I flash_eraseall under linux the device, and then make
>>>> "ubi part..." in U-Boot, commmand succeeds ...
>>>>
>>>
>>> Tested it on my custom AM335x board. Haven't used mainline U-Boot
>>> but cherry-picked the following commits:
>>>
>>>     linux, compat: add missing definitions for ubi
>>>     ubi/ubifs: some bugfixes
>>>     ubi,ubifs: sync with linux v4.2
>>>     ubi: reset mtd_devs when ubi part fail
>>>
>>> Commits apply cleanly except for "linux, compat: add missing
>>> definitions for ubi"
>>> which was trivial to backport anyway.
>>>
>>> U-Boot built fine, but there was a warning:
>>>
>>> drivers/mtd/ubi/fastmap.c: In function 'ubi_attach_fastmap':
>>> drivers/mtd/ubi/fastmap.c:816:2: warning: pointer targets in passing
>>> argument 3 of 'scan_pool' differ in signedness [-Wpointer-sign]
>>>     ret = scan_pool(ubi, ai, fmpl->pebs, pool_size, &max_sqnum, &free);
>>>
>>> Just sent this patch which should fix it:
>>> http://patchwork.ozlabs.org/patch/531842/
>>
>>
>> Thanks for fixing. With it, I see no compiler warning anymore.
>> Applied to u-boot-ubi.git ubi_sync_with_linux
>>
>> Tried this branch on the aristainetos2 board with ubi/ubifs on nand and
>> spi nor flash, worked fine, see log:
>>
>> http://xeidos.ddns.net/buildbot/builders/ari_ubi/builds/7
>> http://xeidos.ddns.net/buildbot/builders/ari_ubi/builds/7/steps/shell/logs/tbotlog
>>
>>> The board seems to work fine, and I can even to "nand erase" and "ubi
>>> part" without
>>> any problem.
>>
>>
>> Did you tried "ubi part" more than once in a row?
>>
>
> Yup, I can do "ubi part $mypart" several times. And can also do "nand
> erase.part foo"
> and then "ubi part foo" several times.

Thanks for checking!

> The patches seem to work fine here, so:
>
> Tested-by: Ezequiel Garcia <ezequiel at vanguardiasur.com.ar>

Updated it to u-boot-ubi.git ubi_sync_with_linux

>>> However, I'm still seeing the same warning I had before, when my partition
>>> is attached:
>>>
>>> ubi0: default fastmap pool size: 200
>>> ubi0: default fastmap WL pool size: 100
>>> ubi0: attaching mtd1
>>> WARNING in drivers/mtd/ubi/fastmap.c line 846
>>> ubi0: scanning is finished
>>> ubi0: attached mtd1 (name "mtd=7", size 509 MiB)
>>> ubi0: PEB size: 131072 bytes (128 KiB), LEB size: 129024 bytes
>>> ubi0: min./max. I/O unit sizes: 2048/2048, sub-page size 512
>>> ubi0: VID header offset: 512 (aligned 512), data offset: 2048
>>> ubi0: good PEBs: 4072, bad PEBs: 4, corrupted PEBs: 0
>>> ubi0: user volume: 4, internal volumes: 1, max. volumes count: 128
>>> ubi0: max/mean erase counter: 5/3, WL threshold: 4096, image sequence
>>> number: 2068197800
>>> ubi0: available PEBs: 3504, total reserved PEBs: 568, PEBs reserved
>>> for bad PEB handling: 76
>>>
>>> @Richard, any ideas? Do you know which of the fixes should fix that?
>>
>
> After some further investigation, printing the counters as Richard suggested
> I found it was a bug on my side :-( The Linux partition and the U-Boot partition
> had different size (i.e. PEB count) and so Fastmap complained :-(
>
> We trimmed the Linux partition size, and forgot to change U-Boot's
> (or thought it wouldn't matter).
>
> Sorry for the noise guys!

Ah, ok.

>> Does this warning raise also in linux? I do not see it on the aristainetos2
>> board.
>>
>> U-Boot code:
>>
>>         /*
>>           * If fastmap is leaking PEBs (must not happen), raise a
>>           * fat warning and fall back to scanning mode.
>>           * We do this here because in ubi_wl_init() it's too late
>>           * and we cannot fall back to scanning.
>>           */
>> #ifndef __UBOOT__
>>          if (WARN_ON(count_fastmap_pebs(ai) != ubi->peb_count -
>>                      ai->bad_peb_count - fm->used_blocks))
>>                  goto fail_bad;
>> #else
>>          if (count_fastmap_pebs(ai) != ubi->peb_count -
>>                      ai->bad_peb_count - fm->used_blocks) {
>>                  WARN_ON(1);
>>                  goto fail_bad;
>>          }
>> #endif
>>
>> Hmm.. could not remember, why I did this U-Boot specific ...
>> But that;s a good example, why I like the "#ifndef __UBOOT__"
>> in Code ... I immediately see, if its linux or u-boot specific
>> code ... maybe its worth to try the original linux code?
>>
>
> Linux WARN_xxx macros return a value, and so can be used inside
> if statements. U-Boot on the other side, just prints a warning:
>
> #define WARN_ON(x) if (x) {printf("WARNING in %s line %d\n" \
>                                    , __FILE__, __LINE__); }
>
> AFAICS, the above ifndef __UBOOT__ is needed, unless you fix
> your WARN_ON in U-Boot.

Yes, correct. Patches are welcome ;-)

bye,
Heiko
-- 
DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany