Hi Scott,
On Mon, 3 Nov 2014 15:40:59 -0600 Scott Wood scottwood@freescale.com wrote:
On Thu, 2014-10-23 at 21:25 +0900, Masahiro Yamada wrote:
Since commit ff94bc40af34 (mtd, ubi, ubifs: resync with Linux-3.14), chip->scan_bbt() is called at the end of nand_scan_tail(). It means the first read access happens immediately after the generic NAND initialization process.
It causes a problem to some SoCs of UniPhier platform because some of their register values need to be fixed up after the general initialization procedure has been finished. Otherwise, read asccess fails. Such a fix-up is SoC-specific enough to be written in a board file rather than in driver code.
One of possible and clean enough ways to work around this issue is postpone the BBT scanning until necessary fix-up is done in board_late_init() or somewhere else.
CONFIG_MTD_NAND_SKIP_BBTSCAN, if enabled, allows to skip the BBT scanning at the end of nand_scan_tail().
Signed-off-by: Masahiro Yamada yamada.m@jp.panasonic.com Cc: Scott Wood scottwood@freescale.com
Rotislav Lisovy already posted a patch to fix the regression, but I don't think you should be relying on this for your problem.
If you are willing to apply Rostislav's patch, I am fine. There is no reason (at least) for me to insist on this patch any more.
Why don't you want to put SoC-specific knowledge in the driver?
There are two reasons:
[1] This problem occurs on some of Panasonic UniPhier SoCs. Denali driver (drivers/mtd/nand/denali.c) is used by SOCFPGA as well as UniPhier. I do not think it is a good idea to put dirty and SoC-specific work around into the driver.
[2] There is no good place to insert a callback to an SoC file. I need to write parameters such as page_size to hardware registers. (You can see my code, nand_denali_fixup() in arch/arm/cpu/armv7/uniphier/board_late_init.c)
The NAND init procedure of U-Boot is like this:
(1) board_nand_init() (drivers/mtd/nand/denali.c) (2) nand_scan_ident() (drivers/mtd/nand/nand_base.c) (3) nand_scan_tail() (drivers/mtd/nand/nand_base.c)
(2) detects the device size and set mtd->write_size, mtd->erase_size, mtd->oob_size. I need to set these values to the Denali hardware, but the Denali driver code is called at (1) which is called before the detection of the device size.
In Linux, nand_scan_ident() and nand_scan_tail() are called from each of NAND drivers, so we can use the values which have been set during nand_scan_ident().
In U-Boot, I think it is impossible.
At least make it a callback from the driver to an SoC file, rather than hoping the NAND won't be touched for a certain amount of time after the driver has told the subsystem "I'm here and ready to be used".
-Scott
Best Regards Masahiro Yamada