UBIFS errors are randomly seen after reboots

[Richard Weinberger]

Chaitanya,

Am 23.01.2017 um 11:48 schrieb chaitanya vinnakota:
> Hi Richard,
> 
> We are seeing UBIFS errors even when the root-filesystem  is mounted
> read-only. But , the error is reported only once.
> Our test scenario is we are rebooting the device by calling "reboot"
> from a script ,  during data-write operations to the mtd partitions
> other than root-filesystem executed by another script.
> 
> What's more baffling is why root-filesystem UBIFS errors are seen when
> the data-write operations are  performed on the other partitions and
> most importantly when rootfs is mounted read-only,
> 
> [  155.121005] UBIFS error (pid 5040): ubifs_decompress: cannot
> decompress 2434 bytes, compressor zlib, error -22

The compressor fails to uncompress because the payload is corrupted.
This can be due to a driver bug, not strong enough ECC, etc..

> [  155.121017] UBIFS error (pid 5040): read_block: bad data node
> (block 60, inode 3484)
> [  155.121026] UBIFS error (pid 5040): do_readpage: cannot read page
> 60 of inode 3484, error -22
> 
> ECC errors are also observed sometimes when the rootfs is mounted read-only
> 
> [  154.824361] ECC: uncorrectable error 2 !!!
> [  154.824368] ECC correction failed for page 0x00014b58
> [  154.825474] ECC: uncorrectable error 2 !!!
> [  154.825479] ECC correction failed for page 0x00014b58
> [  154.825604] UBI warning: ubi_io_read: error -74 (ECC error) while

Here we have a classic ECC error. This should not happen on fresh
NANDs.

> reading 188 bytes from PEB 451:50368, read only 188 bytes, retry
> 
> The page 0x00014b58 falls in the rootfs partition. But , nanddump
> utility is not reporting any bad blocks from the rootfs partition .
> 
>  ~# nanddump /dev/mtd4
> nanddump: warning!: you did not specify a default bad-block handling
>   method. In future versions, the default will change to
>   --bb=skipbad. Use "nanddump --help" for more information.
> nanddump: warning!: in next release, nanddump will not dump OOB
>   by default. Use `nanddump --oob' explicitly to ensure
>   it is dumped.
> ECC failed: 0
> ECC corrected: 0
> Number of bad blocks: 0
> Number of bbt blocks: 0
> 
>  We ran mtd and ubi tests , mtd all tests passed but UBI one test i.e
> io_basic failed.
> 
> Can you please help us in this regard. Any inputs or suggestions ?

This is not easy. I suggest to double check all NAND/MTD settings from
group up. Timings, ECC strength, basic driver testing...

Thanks,
//richard