pr_warn from fs/jffs2/scan.c line 671 in new 4.4 kernel

[mdmillerii]

The jffs2 file system over the ast2400 fmc driver is working fine on the dev-4.3 branch 35d2cfb . But the file system is spewing pr_warn in the dev-4.4 branch 1500e26 .

The spew is from fs/jffs2/scan.c line 671. The whole scan.c file has not changed between revisions. A quick web search shows this is likely due to either file system corruption or a bad driver. After scanning file system for tens of minutes the file system typically reports an error. Returning to a previous 4.3 kernel build the flash is able to be read.

The error could be caused by flash partition alignment, by flash driver, or by something elsewhere in the kernel.

I did a very quick web search and found plenty of hits the first few I looked at were recovered with erasing and formating the whole partition. I did a git diff on the file and there have been zero changes since Febuary 2012 in kernel v3.4-rc1. I did a git diff between drivers/mtd/spi-nor/ and see changes including moving the struct mtd around, but no obvious errors on first scan.

It has been suggested a bisect be attempted, but that will require cherry picking each aspeed support patch stack for each rev (outside the add support options). I feel an attempt to recreate on an existing mtd platform in qemu or similar be attempted first, and/or logging the IO to the flash to compare the pattern.

[mdmillerii]

Doing a git bisect run with the test being the rofs could be the root file system yields:
git bisect start
git bisect good 35d2cfb
git bisect bad 1500e26
git bisect good 6a13feb
git bisect good ab1228e
git bisect good ea5c58e
git bisect bad 9b1b61c
git bisect bad bd4f203
git bisect good 2d49dcb
git bisect bad 7130098
git bisect bad 02f0d3f
git bisect skip 0df4155
git bisect bad 8e2c992
git bisect bad 271707b
git bisect bad 0791a5f
git bisect bad 0791a5f
git bisect good 6ff33f3
git bisect bad 11bff0b
git bisect good e747dbe
git bisect good 7eadd47
git bisect good f9b97fe
git bisect good f9b97fe

Showing commit 11bff0b
as the first bad commit. I reverifed I could mount on f9b97fe
and failed on 11bff0b , even after adding
chip->nor.flash_node = child; and removing the assignment to chip->mtd.priv.

prep:

git diff v4.4.6 openbmc/dev-4.4 > aspeed.patch
git diff openbmc/dev-4.4 openbmc/dev-4.3 -- drivers/mtd/spi-nor/aspeed_smc.c > aspeed-spi-nor-4.3

u-boot boot prep

setenv test 'setenv tftp cuImage_mdm ; bootm ${loadaddr}'
setenv justro 'setenv bootargs ${bootargs} root=/dev/mtdblock5 rootfstype=squashfs ro saveenv'

each step

git clean -e obj -d -f
git reset --hard
git bisect <blah>
git apply --whitespace nowarn --reject aspeed.patch
git checkout openbmc/dev-4.3 arch/arm/configs/aspeed_defconfig
git checkout openbmc/dev-4.4 drivers/clocksource/moxart_timer.c drivers/irqchip/Makefile
make O=obj/ -k -j15  ARCH=arm CROSS_COMPILE=${ARM_CROSS_COMPILE} aspeed_defconfig arch/arm/boot/dts/aspeed-bmc-opp-palmetto.dtb
# remove CONFIG_I2C_ASPEED  as it will encounter a NULL pointer from rejects
vi arch/arm/mach-aspeed/Kconfig obj/.config
make O=obj/ -k -j15  ARCH=arm CROSS_COMPILE=${ARM_CROSS_COMPILE} oldconfig 
make O=obj/ -k -j15  ARCH=arm CROSS_COMPILE=${ARM_CROSS_COMPILE} zImage aspeed-bmc-opp-palmetto.dtb

in u-boot

run justro test

[mdmillerii]

It turns out I identified the fix (filling out flash_node) but missed a step copying the image over for test.

[shenki]

This is fixed in aeb4718