I've got a sporadic problem that I'm seeing using NAND/YAFFS on a Logic LV SOM using a 1928 block YAFFS filesystem. I've got the 2.6.32 kernel (L23 Poky from http://www.omappedia.org/wiki/OMAP_Poky) up and running, and sporadically in testing I observe an error where 0xff30 shows up in the data read back from the file - looks somewhat similar to: http://www.mail-archive.com/linux-omap@xxxxxxxxxxxxxxx/msg23103.html Testing involves using "dd if=/dev/zero of=/mnt/yaffs/<file> bs=1 seek=30M count=0" to create a 30MB file of zeros and then copies the file around on the flash, md5sum, syncing, etc to thrash the cache. The error I'm seeing is that when I read the file back, its md5sum does not match that of what a 30MB file of zeros should generate. To verify, I copy the file from the NAND to a temporary file in RAM, then md5sum that file and if the md5sum mimsmatches, then I hexdump the file to see where the data mismatches. This all runs fin in my test shell script, except after a while (somewhere around 30+GB read from NAND), I see: somefile.7: mismatch 666896a98683a364c10aeba0649f119c != 281ed1d5ae50e8419f9b978 aab16de83 0000000 0000 0000 0000 0000 0000 0000 0000 0000 * 1107800 ff30 ff30 ff30 ff30 ff30 ff30 ff30 ff30 * 1107a00 0000 0000 0000 0000 0000 0000 0000 0000 * 1e00000 instead of the zeros I'd expect. Originally I thought the problem was in the NAND where somehow the driver tried to read a sector of data before it was ready, but if this was the case, I'd expect an ECC error from the comparison (using Hardware generated ECC, prefetch and DMA). This is not the case (I added a printk that triggers if omap_compare_ecc() returns non-zero). So if no ECC error is reported then the data should be valid on NAND. To test if the data was not written correctly I unmounted the filesystem and remounted it, but then the md5sum does match. This is not the first I've seen of the problem. I've see it in a 2.6.28-rc8 kernel, and in the 2.6.32 kernel I've tried turning off DMA, prefetch, and that hastens when the error turns up (and the number of 0xff30 shorts seein). I modified my testing to use a unique pattern intead of zeros and found when the 0xff30 shows up, it repeats for a number of shorts at the start of a page, then I see the data that I expected from the page. I've also modified the NAND driver to use a dev_ready function (as well as statistics to track how long it waits polling the R/B# line on WAIT0 that indicate its 21.2uS +/- 8.29uS once the call to omap_device_ready is made), and still no joy. I've also run this code on a 2.6.33-rc3 kernel with the same driver set and there it works flawlessly. Unfortunately I need the Poky kernel... At this point I'm at a loss to explain what is happening: 1) Has anyone seen this type of error before? 2) Are there any OMAP35x errata that could possibly explain what I'm seeing? 3) Has anyone done exhaustive testing of NAND-based filesystem on an OMAP35x board? 4) Any suggestions where to look next? (YAFFS testing with nandsim on an x86 doesn't exhibit the problem). The following is the original test script (cd into the mountpoint of the NAND filesystem before running): #!/bin/bash # MD5sum of 30M and 1K of zeros md5_30M=281ed1d5ae50e8419f9b978aab16de83 md5_1K=0f343b0931126a20f133d67c2b018a3b # temp file to use as intermediary copy tmpfile=/dev/tmp/junk #tmpfile=/tmp/junk mkdir -p `dirname $tmpfile` mismatches=0 pass=0 passes=120 if [ "$1" != "" ]; then passes="$1" fi # $1 is file # $2 is good checksum chk_md5sum() { for cmf in $1 do cp $cmf $tmpfile ret=`md5sum $tmpfile | cut -d" " -f1` if [ "$ret" != "$2" ]; then echo "$cmf: mismatch $ret != $2" hexdump < $tmpfile | head -100 mismatches=`expr $mismatches + 1` else echo "$cmf: match $ret" fi done } # $1 is src # $2 is destination # $3 is expected md5sum of source chk_cp() { cp $1 $tmpfile cp $tmpfile $2 ret=`md5sum $tmpfile | cut -d" " -f1` if [ "$ret" != "$3" ]; then echo "$1: mismatch $ret != $3" hexdump < $tmpfile | head -100 mismatches=`expr $mismatches + 1` fi } while [ $pass -lt $passes ]; do pass=`expr $pass + 1` echo "Pass: $pass Errors: $mismatches" date # create a 30 M file echo "Create 30M file of zeros and get md5sum" dd if=/dev/zero of=somefile.1 bs=1 seek=30M count=0 chk_md5sum somefile.1 $md5_30M # create copies of file for f in 2 3 4 ; do cp somefile.1 somefile.$f done echo "Calculate md5sums for copied files" chk_md5sum "somefile.*" $md5_30M if [ "$mismatches" != "0" ]; then break; fi echo "execute sync and recalculate md5sums" sync chk_md5sum "somefile.*" $md5_30M if [ "$mismatches" != "0" ]; then break; fi echo "Delete one of the files" rm somefile.2 echo "recopy the deleted file" cp somefile.1 somefile.7 chk_md5sum "somefile.*" $md5_30M if [ "$mismatches" != "0" ]; then break; fi echo "Creating test folder and some junk files in that folder" mkdir -p test cd test dd if=/dev/random of=junk.1 bs=1 count=0 seek=1k chk_md5sum junk.1 $md5_1K if [ "$mismatches" != "0" ]; then break; fi for f in 2 3 4 5 6 7 8 9; do chk_cp junk.1 junk.$f $md5_1K if [ "$mismatches" != "0" ]; then break; fi done echo "md5sums of all files in test folder" chk_md5sum "junk.*" $md5_1K if [ "$mismatches" != "0" ]; then break; fi echo "execute sync and recalculate md5sums" sync chk_md5sum "junk.*" $md5_1K if [ "$mismatches" != "0" ]; then break; fi echo "Remove some files and recreate them" for f in 3 5 8; do rm junk.$f done for f in 8 3 5; do chk_cp junk.1 junk.$f $md5_1K if [ "$mismatches" != "0" ]; then break; fi done cd .. echo "Calculate md5sums for 30M files again" chk_md5sum "somefile.*" $md5_30M if [ "$mismatches" != "0" ]; then break; fi echo "execute sync and recalculate md5sums" sync chk_md5sum "somefile.*" $md5_30M if [ "$mismatches" != "0" ]; then break; fi if [ -f /proc/yaffs ]; then cat /proc/yaffs fi if [ -f /proc/nand-wait-stats ]; then cat /proc/nand-wait-stats fi done -- To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
