On Wednesday 20 February 2013 07:34:09 Steven Hein wrote:
> On 02/18/2013 04:34 PM, Charles Manning wrote:
> > On Saturday 16 February 2013 11:30:43 Steven Hein wrote:
> >> I've had a situation where NAND blocks are being marked bad
> >> invalidly due to a HW/driver issue, and I am trying to recover them.
> >>
> >> I've patched the MTD code to allow bad block erase, and I've
> >> the scrounged together a MEMSETGOODBLOCK ioctl patch to
> >> allow bad blocks to be unmarked in the BBT. All of this is working:
> >> I can erase bad blocks (and verify that data and OOB are all 0xff's),
> >> and I can unmark the blocks bad in the BBT.
> >>
> >> When I have a YAFFS2 filesystem that has bad blocks, I use this
> >> sequence to attempt to recover the blocks:
> >>
> >> (1) unmount the filesystem
> >> (2) erase the bad blocks
> >> (3) unmark the blocks bad
> >> (4) mount the filesystem
> >>
> >> After this sequence, I have found that I have files in the
> >> filesystem that are corrupted or missing, but they were
> >> correct before I started the sequence.
> >
> > Adding erased blocks to a yaffs partition should not cause any problems
> > so long as you mount without the scheckpoint, thus forcing a rescan to
> > find the erased blocks.
> >
> > I suspect your procedure might not be working properly. Sometimes drivers
> > will not erase a bad block.
> >
> > Try changing the order:
> > unmount,
> > unmark then erase bad blocks.
> > check that the block really is erased
> > remount.
> >
> > -- Charles
>
> Thanks Charles! Checkpointing was the issue.
> If I mount with "no-checkpoint-read" after the erase, then my
> FS is intact.
>
> Other than longer mount times, is there any other down side to
> using no-checkpoint-read? Would you expect the non-checkpointed
> mount delay to be significant for a 256MiB filesystem on a 2K block
> NAND device (size 512MiB)?
There are varied speed options:
Fastest: Use checkpointing
Next fastest: Scan using block summaries
Slowest: Scan without chunk summaries
A scan with summaries is generally only a touch slower than checkpointing but
is very dependent on the hardware etc.
From something I posted 16th May 2011:
"On the weekend I set up an Overo board (equivalent to a slower Beagleboard)
with yaffs to get a datapoint for the block summary speed up. Flash speed was
quite slow due to a slow mtd implementation.
I filled almost 250Mbytes with 1MB files and observed the following mount
times:
Mount with full scan using no checkpoint or block summary info: 8.67 seconds
Mount with block summary: 1.02 seconds
Mount with checkpoint: 0.17 seconds"
-- Charles
(text/plain)