[Dev Kernel] 2.6.29 Support for ext2/ext3/ext4/yaffs2

raspdeep · Aug 10, 2010

Here is the kernel with yaffs2 and MTD block support, I am missing some mtd util tools like nandwrite so i cannot continue. Those that have the tools like mkyaffs2image, nandwrite, flash_eraseall can try to create an image and mount it

here are the steps that i think should work:
# mknod /dev/mtd0 c 90 0
# flash_eraseall /dev/mtd0
# mkyaffs2image /data/data yaffs2.img
# nandwrite -p /dev/mtd0 yaffs2.img
# mkdir /data/data2
# mount -t yaffs2 /dev/mtdblock0 /data/data2

If you have the tools, please attach so we can continue.. Its getting late so I am going to bed, have fun

Download
yaffs2_support.tar

darksnarfer · Aug 10, 2010

Oh my god! I was just about to post similar findings, what a coincidence!

I just had compiled a kernel with yaffs2 support. My idea was to modify /init.rc in the kernel to mount the partitions with yaffs2 instead of rfs. Problem: we can't just change the filesystem without backing up and restoring all the files inside. My idea was to do that inside the recovery. Mount a mountpoint, backup with tar to /sdcard, unmount, format with dd (in theory a partition filled with 0's is ok for yaffs2), mount again with yaffs2 and restore from tar.

I'm stuck at the point that I can't seem able to mount any partition with yaffs2 previously formatted (with /system/xbin/busybox dd if=/dev/zero of=/dev/block/stl10). If I mount (/system/xbin/busybox mount -t yaffs2 /dev/block/stl10 /dbdata) there is always an "Invalid argument" error. It happens with /dev/block/stl*, /dev/block/bml* and /dev/block/mmcblk*

I'm no kernel expert, neither I know a lot about yaffs2, but I think I read you can't mount a yaffs2 partition anywhere, it has to be an MTD flash device.

Anyway, I'm going to try with the tools you said. Wish me luck.

bub181 · Aug 10, 2010

Thanks for your work rasdeep! Is it only YAFFS2 and MTD support that you added? Because the kernel already had support for EXT2 and EXT3.

darksnarfer said:
I think I read you can't mount a yaffs2 partition anywhere, it has to be an MTD flash device.

Does the kernel that rasdeep posted load the existng flash devices/partitions as MTD devices?

I think as a start you only have to worry about formatting the /dev/block/mmcblk0p2 device which is mapped to the /data directory. With the new lag fix we are only moving directories stored in /data and that is what gives a big performance improvement. RyanZA did tests and found that the /dbdata partition was already quite fast and I don't think we have to worry about the /system partition etc. because we don't really write to them.

I think after a new firmware flash the /dev/block/mmcblk0p2 (mounted as /data) will be safe to format into YAFFS2 without loosing anything important.

I think these are the steps we can do:
1. Flash new firmware with repartition option on to erase everything.
2. Boot into recovery and format /dev/block/mmcblk0p2 as YAFFS2 filesystem.
3. Copy on new a init.rc script which will mount /dev/block/mmcblk0p2 to /data as a YAFFS2 filesystem and not a RFS filesytem to get much faster speeds.

darksnarfer · Aug 10, 2010

bub181 said:
Does the kernel that rasdeep posted load the existng flash devices/partitions as MTD devices?

Just checked, there isn't any /dev/block/mtd* so I guess the kernel still can't see them as mtd.

ivanmmj · Aug 10, 2010

/data is on the internal SD not in NAND so will we be able to use yaffs2 at all? I was working on formatting to EXT4 and adding the init.rc to match as well as compiling with EXT4 support, but if yaffs2 works, I'm fine with that.

raspdeep · Aug 11, 2010

modifying the init.rc is secondary because that is easy.. the main problem now is to be able to manually setup yaffs2 to mount properly.. if everything works like it should, init.rc can be change in matter of seconds

its funny now u said mtd not found.. i know we can mknod /dev/mtd* but there should be /dev/mtdblock* when booting with kernel right? hmm.. something to investigate further..

keep suggestions and finding coming, we may hit something that i've missed

problem is im lacking those tools that will do write to flash directly.. for example nandwrite.. anyone has that can share so i can try more stuff?

also, are my steps for "preparing" yaffs2 partitions/image correct? if anyone familiar with yaffs, please give some insight

bub181 · Aug 11, 2010

raspdeep said:
modifying the init.rc is secondary because that is easy..

Do you need to recompile the kernel to modify the init.rc? If not could you post some instructions on how to do it please? I would like to try some things in there before /data gets mounted as a RFS filesystem.

raspdeep · Aug 11, 2010

bub181 said:
Do you need to recompile the kernel to modify the init.rc? If not could you post some instructions on how to do it please? I would like to try some things in there before /data gets mounted as a RFS filesystem.

i'm afraid so for now until someone found a way to extract the initramfs from later kernel and restore it back after modification..

ivanmmj · Aug 11, 2010

I think it would be easier to start out with EXT2 since we can easily format that. Just as a proof of concept. I'm planning to do so tonight. Then I'll add support for EXT4 formatting and try that.

bub181 · Aug 11, 2010

raspdeep said:
problem is im lacking those tools that will do write to flash directly.. for example nandwrite.. anyone has that can share so i can try more stuff?

Have you seen these pages rasdeep? They seem to describe how to create the YAFFS filesystem. Also, are there any /dev/mtdblock* devices using the kernel that you created? According to what I read elsewhere it is supposed to create both /dev/mtd* and /dev/mtdblock*. I think we have to use /dev/mtdblock* to create the new filesystem.

http://processors.wiki.ti.com/index.php/Create_a_YAFFS_Target_Image
http://processors.wiki.ti.com/index.php/MTD_Utilities
http://processors.wiki.ti.com/index.php/Put_YAFFS_Image_to_Flash

Also, I know that the Nexus One uses YAFFS2 for its internal filesystem. Maybe someone has worked out a way to write a new image file for that and we can use what they have learnt?

Daemos · Aug 11, 2010

Can we potentially use this kernal (with the right tools) to format the NAND in YAFFS2 instead of RFS? Then still have way of flashing firmware onto it?

If so, that would be awesome and it would seem to have the least draw backs.

Currently the A2SD lag fix requires an external microsd card, and the one click lag fix, mounts a EXT2 image ontop of the RFS NAND, and that could drastically affect NAND life.

Hyperion82 · Aug 11, 2010

This is great !!!
Thank you raspdeep and all the devs here, this is one step closer to a stable kernel and a permanent genuine solution for the lag problem as far as a n00b like me understands

cheers !

mekwall · Aug 11, 2010

Daemos said:
Can we potentially use this kernal (with the right tools) to format the NAND in YAFFS2 instead of RFS? Then still have way of flashing firmware onto it?

If so, that would be awesome and it would seem to have the least draw backs.

Not if we can't mount the NAND as a MTD-device instead of STL-device.

Daemos said:
Currently the A2SD lag fix requires an external microsd card, and the one click lag fix, mounts a EXT2 image ontop of the RFS NAND, and that could drastically affect NAND life.

Say what? /data/ is mounted on the internal SSD, not the NAND. /dbdata/ is however mounted on the NAND. Just do a df -h and you can see all the mountpoints.

Also, mounting and using a ext2-image that reside on an rfs-partition will not damage your drive. The rfs filesystem is still the one used to store the image itself (the ones and the zeros), even though you write it inside an image that uses another filesystem.

mekwall · Aug 11, 2010

I stumbled upon these documents while Googling for STL-devices:

http://www.samsung.com/global/busin...y/downloads/LinuStoreII_GPL Compliance_10.pdf
http://www.samsung.com/global/busin...ionmemory/downloads/RFS_130_Porting_Guide.pdf

I haven't had the time to read them through yet. But if there's a porting guide, there must be a way to reverse it

Go go reverse engineers!

woeds · Aug 11, 2010

Is it really necessary to format the NAND with another filesystem? /dbdata seems fast enough as it is. Wouldn't it be enough (and easier?) to only convert the internal SD to yaffs2?

darksnarfer · Aug 11, 2010

Quote from someone that knows a lot more than I do:

yaffs2 won't work on a sdcard

http://twitter.com/wesgarner/status/20805997718

So right now we have /data that is mounted in the second partition of the internal sdcard (mmcblk0p2), where all the critical files are stored (data, dalvik-cache, etc..) but we can't use yaffs2 there because it's not MTD.

We could, however, format and mount the other partitions (/system, /dbdata, /cache, etc...) with yaffs2, but that might not help because there isn't many heavy read/write there. And we should figure out how to make the kernel see them as /dev/block/mtdblock* instead of /dev/block/stl*.

rasdeep, can you share the .config for the kernel you used? I managed to get some mtd devices (atleast in /proc/mtd) but they were virtual devices. Maybe we need some more options enabled in the kernel.

Also, anybody knows why there are /dev/block/stl* and /dev/block/bml* they seem the same though.

igniztion · Aug 11, 2010

darksnarfer said:
Also, anybody knows why there are /dev/block/stl* and /dev/block/bml* they seem the same though.

According to http://www.samsung.com/global/busin...ducts/fusionmemory/Products_XSR_Brochure.html

STL and BML are independent layers for managing I/O to the NAND devices

STL manages the sector translations while BML manages the invalid blocks (whatever that means)

Chainfire · Aug 11, 2010

Did any of you notice that in init.rc there are many references to yaffs2 ? Perhaps in engineering Samsung did use it ?

darksnarfer · Aug 11, 2010

Chainfire said:
Did any of you notice that in init.rc there are many references to yaffs2 ? Perhaps in engineering Samsung did use it ?

Yeah, I think they are just android standard (maybe they took them from a template) and commented to use their better(?) filesystem.

woeds · Aug 11, 2010

darksnarfer said:
Quote from someone that knows a lot more than I do:

http://twitter.com/wesgarner/status/20805997718

So right now we have /data that is mounted in the second partition of the internal sdcard (mmcblk0p2), where all the critical files are stored (data, dalvik-cache, etc..) but we can't use yaffs2 there because it's not MTD.

We could, however, format and mount the other partitions (/system, /dbdata, /cache, etc...) with yaffs2, but that might not help because there isn't many heavy read/write there. And we should figure out how to make the kernel see them as /dev/block/mtdblock* instead of /dev/block/stl*.

rasdeep, can you share the .config for the kernel you used? I managed to get some mtd devices (atleast in /proc/mtd) but they were virtual devices. Maybe we need some more options enabled in the kernel.

Also, anybody knows why there are /dev/block/stl* and /dev/block/bml* they seem the same though.

Ok, so what filesystem does the internal SD use currently? Would it help if we change that to EXT4? That would emulate the current mimocan fix, but without having to use an external SD card.

[Dev Kernel] 2.6.29 Support for ext2/ext3/ext4/yaffs2

Senior Member

Senior Member

Senior Member

Senior Member

Retired Moderator

Senior Member

Senior Member

Senior Member

Retired Moderator

Senior Member

Senior Member

Senior Member

Senior Member

Senior Member

Senior Member

Senior Member

Senior Member

Moderator Emeritus / Senior Recognized Developer

Senior Member

Senior Member

Similar threads