Why the nand chip driver "should never return -EBADMSG if it failed to read all the request data"

Hi,

During UBIFS power cut tests I get this error message:

...
UBI error: ubi_io_read: error -77 while reading 126976 bytes from PEB 21:4096, read 2048 bytes
UBIFS error (pid 484): ubifs_start_scan: cannot read 126976 bytes from LEB 18:0, error -5
mount: mounting ubi13_0 on /usr/local/appdata failed: Input/output error
...

I found that's because the nand driver return immediately in an ECC unrecoverable error. So my question is:
1. Why the nand driver must continue to read all the left data?
2. Why it still can be mount successfully with YAFFS?

And, after I patched the nand driver, in most of the situations the UBIFS can be successfully recovered, but I still get this error:

...
UBI error: ubi_io_read: error -77 while reading 126976 bytes from PEB 2:4096, read 126976 bytes
UBIFS error (pid 481): insert_node: duplicate sqnum in replay
mount: mounting ubi13_0 on /usr/local/hmt/appdata failed: Invalid argument
...

Does this mean this should never happen to UBIFS? Then what could be the reason? Thank you very much, and my test environment:

# uname -a
Linux (none) 2.6.31-3.3 #10 SMP Wed Apr 11 09:18:13 CST 2012 mips GNU/Linux

# Run this script, and cut power

# while :; do cp -a /usr/local/platform/f2 /usr/local/appdata/ && sync; done

# ls -l /usr/local/platform/f2
-rw-rw-r--    1 506      506        1048576 Apr 11  2012
/usr/local/platform/f2

# xxd f2
0000000: 0001 0203 0405 0607 0809 0a0b 0c0d 0e0f  ................
0000010: 0001 0203 0405 0607 0809 0a0b 0c0d 0e0f  ................
0000020: 0001 0203 0405 0607 0809 0a0b 0c0d 0e0f  ................
0000030: 0001 0203 0405 0607 0809 0a0b 0c0d 0e0f  ................

# mount
rootfs on / type rootfs (rw)
...
ubi12:platform on /usr/local/platform type ubifs (rw,relatime)
ubi13_0 on /usr/local/appdata type ubifs (rw,sync,relatime)
ubi14_0 on /usr/local/.appmid type ubifs (rw,relatime)

Have fun,
Mark Zhang

I see ubi_dbg_dump_stack() call in ubi_io_read() when it returns an error. If you have debugging enabled - you should have the dump in your kernel buffer.

Yeah, we should turn it into 'dump_stack()' to always have the dump. I'd gratefully accept a patch.

Best Regards,
Artem Bityutskiy