u-boot 2016.01 + linux 3.0 -> NAND not detected

[kubiznak_petr]

I'm trying to use a new u-boot (v2016.01) with an old Linux kernel (3.0.15). That means there is no device tree. Everything seems to work well except of the NAND driver - the NAND device is not detected by the kernel, resulting in the following log in dmesg:

FSL NFC MTD nand Driver 1.0

No NAND device found.

mxc_nand: NAND Flash not found !

Anyway, in u-boot, the NAND device is detected and I can read/write it as required:

NAND:  256 MiB

MMC:   FSL_SDHC: 0

*** Warning - bad CRC, using default environment

---

With an old u-boot (v2011.12), the device is detected by the kernel normally:

FSL NFC MTD nand Driver 1.0

NAND device: Manufacturer ID: 0x2c, Chip ID: 0xda (Micron NAND 256MiB 3,3V 8-bit)

Bad block table not found for chip 0

Bad block table not found for chip 0

Scanning device for bad blocks

Bad eraseblock 4 at 0x000000080000

Bad eraseblock 5 at 0x0000000a0000

3 cmdlinepart partitions found on MTD device NAND

Creating 3 MTD partitions on "NAND":

0x000000000000-0x000000100000 : "mtd_uboot"

0x000000100000-0x000000600000 : "mtd_kernel"

0x000000600000-0x000010000000 : "mtd_rootfs"

In u-boot, the log contains also some bad block info:

NAND:  256 MiB

MMC:   FSL_SDHC: 0

Bad block table not found for chip 0

Bad block table not found for chip 0

Bad block table written to 0x00000ffe0000, version 0x01

Bad block table written to 0x00000ffc0000, version 0x01

*** Warning - bad CRC, using default environment

---

Now the question is what is the difference in initialization of NAND in the new u-boot that causes Linux (and only Linux) to not detect the device. I can see the driver has changed significantly but perhaps it's not difficult to make the things work, I might be missing something... Perhaps timesyssupport​ might help...?

[karina_valencia]

Stefan Agner

can you help  to comment here?

[kubiznak_petr]

Alright, I found the problem. The chip select in drivers/mtd/nand/vf610_nfc.c breaks it:

/*

* This function supports Vybrid only (MPC5125 would have full RB and four CS)

*/

static void vf610_nfc_select_chip(struct mtd_info *mtd, int chip)

{

#ifdef CONFIG_VF610

    u32 tmp = vf610_nfc_read(mtd, NFC_ROW_ADDR);

    tmp &= ~(ROW_ADDR_CHIP_SEL_RB_MASK | ROW_ADDR_CHIP_SEL_MASK);

    if (chip >= 0) {

        tmp |= 1 << ROW_ADDR_CHIP_SEL_RB_SHIFT;

        tmp |= (1 << chip) << ROW_ADDR_CHIP_SEL_SHIFT;

    }

    vf610_nfc_write(mtd, NFC_ROW_ADDR, tmp);

#endif

}

When I disable the code inside this function, everything works fine - NAND in both u-boot and Linux.

A side question remains why the code does not work. According to git blame, falstaff​ is the author of this file. May I ask for a comment?

[falstaff]

The function sets the chip select in register the NFC_RAR register. I guess when you uncomment the code, it is just the reset value (which is RB0/CS0 being set) which works fine.

Now I think what happens is that the MTD stack calls that function with -1 after flash operations (which leads to a clear of ROW_ADDR_CHIP_SEL_RB_MASK | ROW_ADDR_CHIP_SEL_MASK, see code above).  So U-Boot might leave the NFC IP with those flags cleared. So if the NFC driver in Linux 3.0 does not explicitly sets them again, I guess the driver ends up talking to no chip... IMHO, the Linux 3.0 NFC driver should implement select_chip accordingly too...

[kubiznak_petr · Accepted solution]

Thanks for your explanation, Stefan. The nfc_select_chip has empty body in Linux 3.0 for Vybrid. That driver is the same as in u-boot v2011.12. Hence the standard configuration u-boot v2011.12 + Linux 3.0 works because the chip select is never touched. My point was whether the select_chip implementation in u-boot v2016.01 might be faulty. Now I see the problem is rather in Linux 3.0. Anyway, if the kernel cannot be fixed, removing the chip select functionality from u-boot v2016.01 does a good job, too.

[karina_valencia]

timesyssupport​ can you help to review this case?