UBI: badblock management confusion!

[Adrian Hunter]

Nancy wrote:
>> Well, the number of initial bad blocks does not affect how many will become
>> bad over time.  So it is not part of the calculation.  For example, if you
>> have 2048 PEBs and 17 are bad initially and you want to reserve 1% for
>> (more)
>> bad blocks then the number reserved is (2048 - 17) / 100 = 20.  That makes
>> the
>> default maximum volume size 2048 - 17 - 20 = 2011.  Whereas, if you are
>> lucky
>> and there are no initial bad blocks, then the number reserved for bad block
>> handling is 2048 / 100 = 20, and the default maximum volume size is
>> 2048 - 20 = 2028.
>>
>> Just because 17 are initially bad in the first case does not mean only 3
>> more
>> will become bad in the time it takes for 20 to become bad in the second
>> case.
>>
>> So 20 are reserved for bad block handling in both cases.
> No, your implement says the initial bad blocks are part of the
> reserved badblocks. As you figure out
>      UBI warning: ubi_eba_init_scan: cannot reserve enough PEBs for
> bad PEB handling, reserved 19, need 20
> just Warning, it need reserved 20, but found there's one more badblock
> comes out during the long using time, so reserved only 19.
> Same code, same operation. Can you tell me how UBI figure out when is
> the real initial time?

Initial bad blocks is the number of bad blocks when the flash comes from
the factory.

> I think it has no way to know, it have to admit
> each attache MTD partition time is initial time. You said initial
> time's badblocks Nand has are not part of reserved badblocks, but the
> practice tell it is part of reserved badblock PEBs( the warning is a
> prof). If it is part of reserved badblocks, then why the max volume
> size have to - current badblocks (it is already in reserved
> badblocks). That is a waist. That is what I confused about!

There are two numbers and they both depend on the number of bad blocks.

The two numbers are:

	beb_rsvd_pebs
	beb_rsvd_level

These numbers are not written to the flash, they are calculated as follows:

beb_rsvd_level is the number of good PEBs multiplied by the bad block
handling percentage.

beb_rsvd_pebs is the same as beb_rsvd_level unless there are not enough
PEBs.  That is, if the sum of the volume sizes plus beb_rsvd_pebs is
greater than the number of good PEBs, then beb_rsvd_pebs is reduced.

So the number of PEBs reserved for bad block handling depends on the
volume size(s) and the number of bad blocks.

When a new volume is created there must be enough PEBs for it plus the
minimum bad block handling reserve (which is 2 PEBs minimum).

When a new volume is created with ubimkvol, you can use the -m option and
the size is calculated automatically to give the biggest volume size where
beb_rsvd_pebs = beb_rsvd_level.

When the volume size is specified with ubinize you can flag it as "autoresize"
and it will be increased if possible.

>> And that means that the maximum volume size varies.
>>
>> What we do here is we don't have the same partition table.  We fix the size
>> of
>> every partition except the last and let the start of each partition be
>> adjusted
>> if there are bad blocks.  For the last partition, it ends up being smaller
>> by the
>> total number of bad blocks.  We then use the "autoresize" of UBI to
>> automatically
>> fit the volume size of the last volume to the partition size.  Since our
>> last
>> volume is UBIFS, there is no problem because UBIFS also automatically
>> adjusts
>> (so long as the volume size is larger that the flashed image created by
>> mkfs.ubifs).
> Oh, too complicated! I don't quite understand.

Sorry that wasn't very clear.  Please just ignore it.

> en...
> fix size of every partition except the last!  if last partitions is
> full of data, then it can't give any free PEBs to other volumes right?
> or you just keep last volume blank, do not use it, like a space
> storehouse for other volume's need, if like that, why not simply
> reserved the last volume size.

The image for the last partition or volume must always be small enough to fit.

>> If you want to have the same volume size, then you can but you must make the
>> size smaller than the maximum.  You will need to consult your documentation
>> about what the maximum number of initial bad blocks is.  For example, for
>> OneNAND with 2048 PEBs, when a OneNAND comes from the factory it may have
>> as many as 40 PEBs that are bad.  So 2048 - 40 = 2008 good PEBs minimum.
>> 2008 / 100 = 20 PEBs reserved for (more) bad blocks (assuming you use the 1%
>> setting).  2008 - 20 = 1988 PEBs for the volume size.  The exact volume size
>> is
>> then 1988 * LEB size = 1988 * 258048 (in your case) = 512999424.
> 
> Compare with reserved badblocks(including the init and future go bad)
> by a percentage UBI user can configure by wish. What good autoresize
> has?

autoresize is for when you are creating images with ubinize and you want to
use the maximum volume size.  Because the maximum volume size depends on
the number of bad blocks, it changes from device to device.

autoresize means you can specify the worst-case volume size and tell UBI to
make the volume bigger if there are fewer bad blocks than the worst case.

If you are using ubimkvol, you can just use the -m option to get the
maximum size.

If you are using ubimkvol, you can find UBI values in /sys/class/ubi/ubix
where x is the UBI device number (0 if you only have 1 UBI device).

For example, the following is the same as using the -m option:

	modprobe ubi mtd=5

	ebcnt=`cat /sys/class/ubi/ubi0/avail_eraseblocks`
	ebsize=`cat /sys/class/ubi/ubi0/eraseblock_size`

	volsize=`expr $ebcnt \* $ebsize`

	ubimkvol /dev/ubi0 -Nvol -s$volsize