[IPNC][DM368] BBT problem after patch to new ECC layout (Bad block table not found for chip 0))

This is follow-up section of another NAND ECC topic  NAND ECC issue - ECC layout incompatible between RBL/UBL and U-Boot/Linux

After patching new ECC table, we found that "Bad block table not found for chip 0" keeps showing even if "Bad block table written to 0x07ffc000" follows every time boot up. It means that the BBT is not written correctly, in both u-boot and linux kernel.

After (painfully) tracing ECC and BBT relevant source code in u-boot, I found the root cause is the incorrect location of writing BBT/mirror patterns.

1. look at following illustration which capture from TI's wiki (http://processors.wiki.ti.com/index.php/DM365_Nand_ECC_layout)

in the file drivers/mtd/nand/nand_bbt.c, there are two nand_bbt_descr structure, bbt_main_descr and bbt_mirror_descr which contain the information how bbt is placed and searched.

originally, the structures are like below (2010-12 u-boot release, I compared with the latest one,Dec 2011, and there isn't meaningful difference regarding ECC and BBT).
===

static uint8_t bbt_pattern[] = {'B', 'b', 't', '0' };
static uint8_t mirror_pattern[] = {'1', 't', 'b', 'B' };
static struct nand_bbt_descr bbt_main_descr = {
.options = NAND_BBT_LASTBLOCK | NAND_BBT_CREATE | NAND_BBT_WRITE
| NAND_BBT_2BIT | NAND_BBT_VERSION | NAND_BBT_PERCHIP,
.offs = 8,
.len = 4,
.veroffs = 12,
.maxblocks = 4,
.pattern = bbt_pattern
};
static struct nand_bbt_descr bbt_mirror_descr = {
.options = NAND_BBT_LASTBLOCK | NAND_BBT_CREATE | NAND_BBT_WRITE
| NAND_BBT_2BIT | NAND_BBT_VERSION | NAND_BBT_PERCHIP,
.offs = 8,
.len = 4,
.veroffs = 12, .maxblocks = 4,
.pattern = mirror_pattern
};

===

2. in check_pattern()@nand_bbt.c, we know that the structure member
offs - the offset from start of oob of block, to place the bbt_pattern
len - the length of bbt pattern
veroffs - the offset from the start of oob of block, to place the bbt version number

Apparently the setting was wrong after deploying the new ECC layout (the upper table in the picture) because the bbt pattern, starting from the 8th byte of OOB, is conflict with ECC.

I change it to 16th and veroffs at 32th, and BBT is back again!  The corresponding portion in Linux source should be changed too. In appro's ipnc release, the change can be made at ti-davinci/arch/arm/mach-davinci/board-dm368-ipnc.c

[DM368][IPNC] Upgrade udev on IPNC-DM368

[Prolog]

the reason I would like to upgrade udev on IPNC-DM368 is that the latest Appro package (Ver_2.00.26, kernel 2.6.18) doesn't contain 'udevadmin', which is used to control the runtime behavior of udev, request kernel events, manage queue, and debug udev.

Appro's package includes some of udev helper functions such as udevtrigger and udevinfo. However, udevadmin is what I want to follow Embedded Linux Primer, 2nd edition, ch19.

[Main]

1. Get the udev package from kernel tree. The latest version which supports kernel 2.6.18 is udev-127.

2. Unpack the tarball

3. Configure the make envionment

CC=/opt/mv_pro_5.0.0/montavista/pro/devkit/arm/v5t_le/bin/arm_v5t_le-gcc ./configure --host=armv5tl-montavista-linux-gnueabi --prefix=/home/usr/work/udev-127/filesys

4. make & make install

5. udevd and udevadm were generated under directory ./udev

[Epilog]

Easy work! as long as the CC is set correctly, there will be no problem to finish the job.

[Extend readings]

CH19, Embedded Linux Primer, 2nd Edition

[DM368][IPNC] Using GDB on IPNC

[Prolog]

To dig into Appro's application, I decided to enable gdb on IPNC platform. There are two useful page provided by TI's wiki and they let me enable gdb on IPNC in one hour.

==============================

Debugging a Linux Application with MontaVista 5.0

Debugging remotely on DaVinci using gdb

==============================

[Main]

1. get the latest gdb from gdb download page

2. configure and make gdb and gdbserver

for gdb
==============================

host# ./configure --host=i686-pc-linux-gnu --target=armv5tl-montavista-linux-gnueabi --prefix=/opt/mv_pro_5.0.0/montavista/pro/devkit/arm/v5t_le

host# make

host# make install

==============================

for gdbserver

==============================

host# cd gdb/gdbserver
host# CC=/opt/mv_pro_5.0.0/montavista/pro/devkit/arm/v5t_le/bin/arm_v5t_le-gcc ./configure  --host=armv5tl-montavista-linux-gnueabi --prefix=/home/user/workdir/filesys
host# make && make install



==============================

there was an error during "make install". the screen showed
==============================
cannot stat 'libinproctrace.so': No such file or directory
==============================

after quick google search, it can be fixed by modifying "gdb/gdbserver/Makefile.in"
==============================
original:
if [ x$IPA_DEPFILES != x ]; then \
fixed:
if [ x${IPA_DEPFILES} != x ]; then \
==============================


3. Running gdbserver at target
==============================
target# gdbserver 192.168.100.1:1000 ./sample
==============================



4. Running gdb at host
==============================
host# /opt/mv_pro_5.0.0/montavista/pro/devkit/arm/v5t_le/bin/armv5tl-montavista-linux-gnueabi-gdb ./sample
==============================

at gdb prompt
==============================
(gdb) set solib-search-path /home/user/workdir/filesys/lib  (gdb) target remote 192.168.100.2:1000 
==============================


5. (optional) using ddd
==============================
host# ddd --debugger arm_v5t_le-gdb
==============================

6. (optional) error "Malformed packet(b) (missing colon)" 
    in Appro reference design, there is an existing gdbserver under /usr/bin. However, the installation of gdbserver is toward /bin, and the PATH setting will select the old gdbserver (GNU gdbserver 6.6.50.20070301 in my case). Easy to fix though.




[Epilog]
Easy works as far! multitread debugging could be challenge though
[Extend readings]
Debugging on DaVinci using kgdb

[DM368][IPNC][U-Boot] Porting U-boot to DM368-IPNC (1) Ethernet

[Prologue]
Appro's DM368 IPNC reference code uses u-boot version 1.3.4, which is kind of old version without USB and MMC boot support. Appro doesn't seems to upgrade it in the near future (actually it doesn't care except you are big customer). 


Couple days ago I found TI released a new dvsdk for DM365/DM368 (DVSDK_4.02 4_02_00_06), which contains U-Boot 2010.12-rc2. Even though it's not the latest version, it includes all required feature I want. I decided to port this version to DM368-IPNC.




[Main]
After scanning related config files, I decided to use include/configs/davinci_dm365evm.h as template to start the work. This article focuses on enabling Ethernet of U-Boot on IPNC. Others issues will be discussed later.

Steps:
1. create a new config file include/configs/davinci_dm368_ipnc.h with the same content of davinci_dm365evm.h

2. add board information to boards.cfg
==============================

(davinci_dm368_ipnc           arm         arm926ejs   dm365evm            davinci        davinci)

==============================

3. make u-boot!
I didn't use CodeSourcery so had to assign cross compiler.
the commands are
==============================

make CROSS_COMPILE=arm_v5t_le- distclean
make CROSS_COMPILE=arm_v5t_le- clean
make CROSS_COMPILE=arm_v5t_le- davinci_dm368_ipnc_config
make CROSS_COMPILE=arm_v5t_le-

==============================

4. found Ethernet doesn't work. after issuing "ping 192.168.1.1", which is my tftp/nfs server, IPNC hung forever without any message shown. I cannot break it by Ctrl-C either.

5.turn on Ethernet debug to get more messages
at drivers/net/davinci_emac.c, enable emac_dbg
==============================

unsigned int emac_dbg = 1;

==============================

6. reboot IPNC, and issue ping again, I got

==============================

+ emac_close
+ emac_ch_teardown
- emac_ch_teardown
+ emac_ch_teardown
- emac_ch_teardown
- emac_close
+ emac_open

==============================
and IPNC hung. "+ emac_open" was printed by davinci_eth_open()@drivers/net/davinci_emac.c. after get_link_speed statement, it broke by "return 0".
==============================

(!phy.get_link_speed(active_phy_addr))
return(0);

==============================

6. the suspicious point was the variable "active_phy_addr" above. so I added two debug print to ensure it's the root cause.
==============================

debug_emac("+ emac get_link_speed, active_phy_addr = %x\n", active_phy_addr);
if (!phy.get_link_speed(active_phy_addr))
return(0);
debug_emac("- emac get_link_speed\n");

==============================
it confirmed the problem point and active_phy_addr = 0xFF, which is strange.

7. trace code to locate where we got the active_phy_addr and found davinci_eth_phy_detect() @ drivers/net/davinci_emac.c
==============================

phy_act_state = readl(&adap_mdio->ALIVE) & EMAC_MDIO_PHY_MASK;
if (phy_act_state == 0)
return(0); /* No active PHYs */

==============================

there is no error handling code if return 0 (No active PHYs). EMAC_MDIO_PHY_MASK is a macro from EMAC_MDIO_PHY_NUM(@emac_defs.h), which is defined in config file with the value 2.

==============================

/* PHY mask - set only those phy number bits where phy is/can be connected */
#define EMAC_MDIO_PHY_NUM           CONFIG_EMAC_MDIO_PHY_NUM
#define EMAC_MDIO_PHY_MASK          (1 << EMAC_MDIO_PHY_NUM)

==============================

8. Check with schematic file of IPNC and the datasheet of PHY (DM9161), the strapping config is 1 (PHYAD0 pull high), which means the PHY address is 1. Tried to dump the register of MDIO-ALIVE(0x01d0b00c) and the value is 0x00000002. it showed the PHY address should be 1.

9. Change the config value of PHY number at davinci_dm368_ipnc

==============================

 #define CONFIG_EMAC_MDIO_PHY_NUM 1

==============================

and make u-boot again

10. Bingo!

==============================

Ethernet PHY: GENERIC @ 0x01

DaVinci-EMAC

Hit any key to stop autoboot:  0

==============================

[Epilogue]

Some of the issues are not discussed, some are not fixed yet.
To Be Post -
UART
NAND - ECC

To Be Fixed -
USB
MMC
MTD

[DM368][RTL8191] Porting RTL8191S to DM368/IPNC

The target is to install RTL8191SU, USB dongle, to davinci DM368.

Steps:
1. get official device driver from Realtek

2. in the software package, there are some documentation we would like to read first.
I suggest

HowTo support more VidPids.doc - check if VID/PID is included in code (os_intf/linux/usb_intf.c)
HowTo support new platform(including Android).doc - change Makefile 

In Makefile, I added a new platform CONFIG_PLATFORM_DM368 so the following lines were added

###
CONFIG_PLATFORM_DM368 = y
ifeq ($(CONFIG_PLATFORM_DM368), y)
EXTRA_CFLAGS += -DCONFIG_LITTLE_ENDIAN
ARCH := arm
CROSS_COMPILE = arm_v5t_le-
KSRC := [YOUR KERNEL SRC]/ipnc/ti-davinci
endif
###

3. No problem after above change, we got 8712u.ko. However, when inserting module "insmod 8712u.ko", some error messages were prompted.

# insmod 8712u.ko
8712u: Unknown symbol strcasecmp
insmod: cannot insert `8712u.ko': Unknown symbol in module (-1): No such file or directory

It's apparently that the function "strcasecmp" is not included in the C library in DM368/IPNC. I confirmed it by checking include/linux/string.h (as well as asm/string.h). I decided not to modify lib but change rtl871x_ioctl_linux.c instead.

in rtl871x_ioctl_linux.c, there are couple lines of code using "strcasecmp". I use "strcmp" with small characters instead. for example

// if(0 == strcasecmp(ext,"RSSI")){
if(0 == strcmp(ext,"rssi")){

4. After all the change, the module can successfully be installed and "wlan0" can be find by "ifconfig -a"

[DM368][RT73] Enabling Wireless Dongle

Reference information

Davinci USB WLan

Montavista Wireless Collection (although for mv pro 6, it's similar to pro 5)

 rt2x00.serialmonkey.com (driver I tried first)

Ralink Software (Linux) 

in case USB is not configured to OTG or host:

How to configure DM365 for USB host, device and OTG support

I was using D-link DWA-110, Ralink RT-73 NIC

Steps:

1. download serialmonkey's driver source since it is recommended by TI wiki Davinci USB WLan.

there was not too much trouble to make it by following the wiki. I got rt73.ko and copied it to (target)/opt/ipnc. I also copied rt73.bin to (target)/lib/firmware

However, when issuing "insmod rt73.ko", the result was always -

rt73: init
rt73: idVendor = 0x7d1, idProduct = 0x3c03
rt73: Failed to request_firmware. Check your firmware file location
rt73: Failed to load Firmware.

apparently the firmware "rt73.ko" was not found.  I am not too familiar with the procedure after the driver is installed (such as what script will be called , etc). I searched web for quick solution, some useful sites are listed below:

How to Port rt73 driver to ARM

in short, the search failed, even I place rt73.bin to various location, the firmware simply didn't get loaded.

2. Change to official Ralink driver

After reading TI's forum post "Trouble with USB WLAN driver on IPNC DM368", I decided to give up serialmonkey and go official way.

To make the official driver needs a little more effort. There was some kind of kernel inconstant with IPNC release (probably 2.0.26 or something like that). 

I got rt73_drv1.1.0.5 from ralink's download section. To make module, you have to

- modify Makefile: change LINUX_SRC

the first make resulted in missing sturct member get_wireless_stats error in rtmp_main.c, function usb_rtusb_probe()

#if (WIRELESS_EXT >= 12) 

#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20)
netdev->get_wireless_stats = rt73_get_wireless_stats;
#endif

netdev->wireless_handlers = (struct iw_handler_def *) &rt73_iw_handler_def;

#endif

I believe that my IPNC_DM036 release is of kernel 2.6.18, but the include/linux/netdevice.h, the member  get_wireless_stats doesn't exist. not sure if appro or montavista patched it and I didn't check official kernel tree.

Anyway, I changed the #if statement to KERNEL_VERSION(2,6,18) and the module can be compiled.

3. the new rt73.ko works fine, and I also found that rt73.bin is not required.

[Linux][kernel][DM368] Cannot Change Kernel Config Variable

After changing variables in the (default)config file, davinci_dm368_ipnc_defconfig, "make sys" generates .config but the .config file doesn't match the changes in davinci_dm368_ipnc_defconfig.

turning on USB_STORAGE at defconfig for example

# may also be needed; see USB_STORAGE Help for more information
#
# CONFIG_USB_STORAGE is not set
# CONFIG_USB_LIBUSUAL is not set
CONFIG_USB_STORAGE=m
CONFIG_USB_LIBUSUAL=y 

the .config will be still

# CONFIG_USB_STORAGE is not set
# CONFIG_USB_LIBUSUAL is not set

check steps:
1. track what's performed after executing "make sysall"  (make -n)
make sysall -> make lspall -> make lspclean, make lsp
make lsp -> make lspcfg
make lspcfg->
make lspbuild MAKE_TARGET=davinci_dm368_ipnc_defconfig => make ARCH=arm davinci_dm368_ipnc_defconfig   <==== I after this, .config is generated with incorrect config variables
make lspbuild MAKE_TARGET=checksetconfig

2.  locate potential trouble maker:
"make ARCH=arm davinci_dm368_ipnc_defconfig" executes "scripts/kconfig/conf -D arch/arm/configs/davinci_dm368_ipnc_defconfig arch/arm/Kconfig" to generate .config

suspicious screen output

arch/arm/configs/davinci_dm368_ipnc_defconfig:1037:warning: trying to reassign symbol USB_STORAGE
arch/arm/configs/davinci_dm368_ipnc_defconfig:1038:warning: trying to reassign symbol USB_LIBUSUAL

3. first, check if the variable USB_STORAGE has dependency problem.
the variable is defined at driver/usb/storage/Kconfig

config USB_STORAGE
    tristate "USB Mass Storage support"
    depends on USB
    select SCSI

but I am sure that CONFIG_USB is enabled in the generated .config file.

4. find more information about making kernel.

kernel 移植笔记（从omap linux-02.01.03.11 到 fred 版本kernel 学习）

史上最经典的Linux内核学习方法论

 "make ${PLATFORM}_defconfig"
     Create a ./.config file by using the default
     symbol values from
     arch/$ARCH/configs/${PLATFORM}_defconfig.
     Use "make help" to get a list of all available
     platforms of your architecture.

however, it seems not to have any problem. even I wasted couple hour to compare with the latest kernel.... no luck though

  5. back to fundamental, why "scripts/kconfig/conf -D arch/arm/configs/davinci_dm368_ipnc_defconfig arch/arm/Kconfig" outputs incorrect value.

[PATCH] Allow kconfig to accept overrides - Kernel

inspired me. so trace confdata.c
at conf_read_simple( )

    while (fgets(line, sizeof(line), in)) {
        conf_lineno++;
        sym = NULL;
        switch (line[0]) {
        case '#':
            if (memcmp(line + 2, "CONFIG_", 7))
                continue;
            p = strchr(line + 9, ' ');
            if (!p)
                continue;
            *p++ = 0;
            if (strncmp(p, "is not set", 10))
                continue;

it is surprised me that the line starting with "#" is not a comment !!

Solution
use double hash (##) as comment to "CONFIG_xxx is not set"


## CONFIG_USB_STORAGE is not set
## CONFIG_USB_LIBUSUAL is not set
CONFIG_USB_STORAGE=m
CONFIG_USB_LIBUSUAL=y  


Followups
in montavista releases, there is "Base Configuration for MontaVista Linux" file (scripts/kconfig/baseconfig), which will be compared with .config generated by make xxx_defconfig. The rules are

# If an option is marked as "m", the .config will only
# be modified if the option is turned off.
#
# If an option is marked as "y", the .config will be modified
# if the options in the "m" or the off states.
#
# If an option is marked as "n", the .config will be modified
# if the option is in the "m" or "y" states.

It could influence the final .config for making kernel.