Commit graph

11 commits

Author SHA1 Message Date
Stefan Agner
8fca2d8cb8 mtd: vf610_nfc: enable ONFI detection
This changes enable ONFI detection. The Read ID command now allows
one address byte which is needed for ONFI detection. To read the
ONFI parameter page, the NAND_CMD_PARAM need to be supported. The
CMD code enables one command and one address byte along with reading
data from flash using R/B#, as specified by ONFI.

Signed-off-by: Stefan Agner <stefan@agner.ch>
2015-05-24 14:27:46 -05:00
Stefan Agner
080a71e8e9 mtd: vf610_nfc: add 32-error correction option for HW ECC
Add option to choose between current 24-error correction and 32-error
correction through Kconfig. 32-error correction allow to use NAND
chips which require up to 8-bit error correction per 512 byte (when
using 2K pages).

Signed-off-by: Stefan Agner <stefan@agner.ch>
2015-05-24 14:27:46 -05:00
Stefan Agner
84d656a283 mtd: vf610_nfc: use in-band bad block table
Use in-band bad block table (NAND_BBT_NO_OOB) which allows to
use the full OOB for hardare ECC purposes. Since there is no
ECC correction on the OOB it is also safer to use in-band area
to store the bad block table marker.

Signed-off-by: Stefan Agner <stefan@agner.ch>
2015-05-24 14:26:55 -05:00
Stefan Agner
5dec286b82 mtd: vf610_nfc: implement OOB only read
Implement read of OOB area only. When using column and sector size
properties, only parts of the page can be read. However, this works
only when hardware ECC is disabled, otherwise the ECC engine would
ruin the data in the buffer. To allow OOB only reads, three points
had to be addressed:
- Set ECC mode per command.
- Handle NAND_CMD_READOOB seperate. Make sure column and sector
  size is correctly set up, while disabling ECC.
- Now, the OOB data end up at the beginning of the buffer. Remove
  the special handling of OOB (spareonly).

Especially bad block scans benefit from this change. On a 512MiB
SLC NAND device, the bad block scan took 1.5s less than before.

Signed-off-by: Stefan Agner <stefan@agner.ch>
2015-05-24 14:26:55 -05:00
Stefan Agner
d111bf99a8 mtd: vf610_nfc: allow bitflips in an empty page
Allow bit flips in a empty page up to half of the recoverable
bits (strength / 2).

Signed-off-by: Stefan Agner <stefan@agner.ch>
2015-05-24 14:26:54 -05:00
Stefan Agner
6fcfd1e890 mtd: vf610_nfc: remove read on SEQIN
Since we do not support sub-page writes anyway, reading the page
back to the controller on SEQIN command is not required. Remove
the page read on SEQIN.

However, the column/page values relevant to the SEQIN command, hence
set the column/row address on SEQIN command.

Signed-off-by: Stefan Agner <stefan@agner.ch>
2015-05-24 14:26:54 -05:00
Stefan Agner
f55bc296a8 mtd: vf610_nfc: remove caching of page in buffer
To improve performance we remember the current page in the buffer
and avoid reading it twice. This implicit page cache increases
complexity while does not increase performance in real world cases.
This patch removes that feature.

Acked-by: Bill Pringlemeir <bpringlemeir@nbsps.com>
Signed-off-by: Stefan Agner <stefan@agner.ch>
2015-05-24 14:26:54 -05:00
Stefan Agner
55765b1842 mtd: vf610_nfc: specify transfer size before each transfer
Testing showed, that commands like STATUS made the buffer dirty
when executed with NFC_SECSZ set to the page size. It looks
like the controller transfers bogus data when this register
is configured. When setting it to 0, the buffer does not get
altered while the status command still seems to work flawless.

Signed-off-by: Stefan Agner <stefan@agner.ch>
2015-03-30 23:35:27 -05:00
Stefan Agner
7653fc288a mtd: vf610_nfc: mark page as dirty on block erase
The driver tries to re-use the page buffer by storing the page
number of the current page in the buffer. The page is only read
if the requested page number is not currently in the buffer. When
a block is erased, the page number is marked as invalid if the
erased page equals the one currently in the cache. However, since
a erase block consists of multiple pages, also other page numbers
could be affected.

The commands to reproduce this issue (on a written page):
> nand dump 0x800
> nand erase 0x0 0x20000
> nand dump 0x800

The second nand dump command returns the data from the buffer,
while in fact the page is erased (0xff).

Avoid the hassle to calculate whether the page is affected or not,
but set the page buffer unconditionally to invalid instead.

Signed-off-by: Stefan Agner <stefan@agner.ch>
2015-03-30 23:33:42 -05:00
Sanchayan Maity
2260457341 mtd/nand/vf610_nfc: Disable subpage writes
This patch disables subpage writes for vf610_nfc nand
driver. This is required, as without this fix, writing
unaligned u-boot images with DFU results in a hang.
Trying to write unalgined binary images also results
in a hang, without disabling subpage writes.

Patch has been tested on a Colibri VF61 module.

Signed-off-by: Sanchayan Maity <maitysanchayan@gmail.com>
2014-11-26 19:53:49 -06:00
Stefan Agner
72d7beabf7 mtd: nand: add Freescale vf610_nfc driver
This adds initial support for Freescale NFC (NAND Flash Controller)
found in ARM Vybrid SoC's, Power Architecture MPC5125 and others.
The driver is called vf610_nfc since this is the first supported
and tested hardware platform supported by the driver.

Signed-off-by: Stefan Agner <stefan@agner.ch>
Acked-by: Bill Pringlemeir <bpringlemeir@nbsps.com>
2014-09-16 13:25:18 +02:00