Well, I can help you with this much - jffs2 + Android = No Go. It will
run a few apps until the actual zygote stuff needs to run. mmap is the
thing that will screw you over - you only get mmap with read only
jffs2.
yaffs2 + NOR = No Go. It will fall over and flop.
UBIFS + NOR + Android = Works well actually. Got it running on our
iMX31 board with NOR. I'll post some instructions to get UBIFS working
if you need it. UBIFS = JFFS3. It has compression and a lot of other
cool stuff. It also scales well.

On Nov 10, 4:10 pm, Markus <[EMAIL PROTECTED]> wrote:
> Hi,
>
> yes, we might change to a different file system, but actually, Android
> uses yaffs2 as main file system or at least it seems like this if you
> hack into its configuration files. To be honest, we do not know, if
> this problem is caused by the file system or something else as we are
> able to do file operations in /data/app during the booting process.
> The kernel panic occurs after Android finished the booting process. So
> our guess is that Android starts something, that watches the file
> system (especially /data/app) and that this service is causing our
> problem. Unfortunately, we could not yet locate, which tool is
> responsible... any guess, what is started in the end of the booting
> process, that might cause our problem?
>
> Below, a log of the kernel panic.
>
> Bye
> Markus
>
> busybox cp ApiDemos.apk test
> Unable to handle kernel paging request at virtual address 00100104
> pgd = c70a4000
> [00100104] *pgd=870a2031, *pte=857180dd, *ppte=8571880e
> Internal error: Oops: 81f [#1] PREEMPT
> Modules linked in:
> CPU: 0    Not tainted  (2.6.24-140-g68eb4b4 #77)
> PC is at android_unlock_suspend+0x60/0x170
> LR is at android_unlock_suspend+0x34/0x170
> pc : [<c01ff8b4>]    lr : [<c01ff888>]    psr: 60000193
> sp : c711bea8  ip : c039bac4  fp : c711bee4
> r10: c711a000  r9 : 000001e0  r8 : 60000113
> r7 : c7de20a0  r6 : c039babc  r5 : c039babc  r4 : c7de20e0
> r3 : c7c35da8  r2 : 00100100  r1 : 00200200  r0 : c7de20e0
> Flags: nZCv  IRQs off  FIQs on  Mode SVC_32  ISA ARM  Segment user
> Control: 00e5387f  Table: 870a4000  DAC: 00000015
> Process FileObserver (pid: 1675, stack limit = 0xc711a260)
> Stack: (0xc711bea8 to 0xc711c000)
> bea0:                   c710401c c7cfac40 c711bed4 c711bec0 c00603cc
> c006035c
> bec0: c73dc3c0 00000000 c73dc3d0 c7de20a0 c73dc3c0 c711a000 c711befc
> c711bee8
> bee0: c00c4f90 c01ff860 c73dc3c0 46a2cba4 c711bf4c c711bf00 c00c57d0
> c00c4f30
> bf00: c003f92c 46a2cb84 c7cfac70 00000000 c7cfac40 c005b298 c711bf18
> c711bf18
> bf20: c02bfa58 c7043ea0 46a2cb84 c711bf78 00000200 c0025004 c711a000
> 41046fc0
> bf40: c711bf74 c711bf50 c00961a4 c00c5634 c711bf74 c711bf60 c7043ea0
> fffffff7
> bf60: 00000000 00000000 c711bfa4 c711bf78 c00965ec c00960fc 00000000
> 00000000
> bf80: 001ce0b0 00000001 00000f4c ad352cd8 001cf5b8 00000003 00000000
> c711bfa8
> bfa0: c0024e80 c00965b4 00000f4c ad352cd8 0000001e 46a2cb84 00000200
> fd1fafed
> bfc0: 00000f4c ad352cd8 001cf5b8 00000003 46a2cda0 41046fd4 41046fc0
> 00000001
> bfe0: ad353458 46a2cb48 ad3414c9 afe0b50c 00000010 0000001e 00ff00ff
> 00ff00ff
> Backtrace:
> [<c01ff854>] (android_unlock_suspend+0x0/0x170) from [<c00c4f90>]
> (remove_kevent+0x6c/0x94)
> [<c00c4f24>] (remove_kevent+0x0/0x94) from [<c00c57d0>] (inotify_read
> +0x1a8/0x1e4)
>  r4:46a2cba4
> [<c00c5628>] (inotify_read+0x0/0x1e4) from [<c00961a4>] (vfs_read
> +0xb4/0x144)
> [<c00960f0>] (vfs_read+0x0/0x144) from [<c00965ec>] (sys_read
> +0x44/0x70)
>  r7:00000000 r6:00000000 r5:fffffff7 r4:c7043ea0
> [<c00965a8>] (sys_read+0x0/0x70) from [<c0024e80>] (ret_fast_syscall
> +0x0/0x2c)
>  r7:00000003 r6:001cf5b8 r5:ad352cd8 r4:00000f4c
> Code: e5965000 e5812000 e5843000 e59c3000 (e5821004)
> Kernel panic - not syncing: Fatal exception
>
> On 9 Nov., 10:16, mvniekerk <[EMAIL PROTECTED]> wrote:
>
> > Your answer lies in UBIFS. There is a port for kernel 2.6.24 up to
> > 2.6.27. UBIFS is JFFS3 if you like - and it does support mmap. If your
> > flash chip is of the NOR-type then YAFFS2 will not work - that is what
> > makes UBIFS so sweet!
> > To set up a UBI volume for UBIFS is bit of a schlep, but once done it
> > is a cool piece of equipment.
>
> > On Nov 6, 11:39 pm, Markus <[EMAIL PROTECTED]> wrote:
>
> > > Hi,
>
> > > it is the init process, that cannot start. The kernel is always
> > > booting fine and only the Android init process is not able to do its
> > > job. For yaffs2, the booting process stops like 
> > > inhttp://groups.google.com/group/android-porting/browse_thread/thread/d...
> > > - I'm sorry, that I can't post my own message at the moment, but I do
> > > not have access to the hardware right now to flash everything...
>
> > > Like in the link above, we get the same problem about the magic
> > > number, while Android tries to load the core.jar file. After 4 tries,
> > > Android resigns and reboots.
>
> > > bye
> > > Markus
>
> > > On 6 Nov., 17:22, "Gergely Kis" <[EMAIL PROTECTED]> wrote:
>
> > > > Hi,
>
> > > > Could you give more information regarding "Android was not able to
> > > > boot onyaffs2". What were the actual error messages? Did the kernel
> > > > hang, or the init process?
>
> > > > Best Regards,
> > > > Gergely
>
> > > > On Thu, Nov 6, 2008 at 4:18 PM, Markus <[EMAIL PROTECTED]> wrote:
>
> > > > > Hi,
>
> > > > > as I wrote in Android Internals, we ported Android to an i.MX31.
> > > > > Unfortunately, we have some issues with the file system.
> > > > > If I use NFS as file system with a modified init.rc config, everything
> > > > > seems to work well, but this is no option for us as permanent file
> > > > > system, so we decided to useyaffs2as file system. As this did not
> > > > > work (Android was not able to boot), we changed to jffs2. jffs2 boots
> > > > > fine as long as we use a read-only file system. After booting, we can
> > > > > start many applications, but it seems that those requiring file write
> > > > > operations fail to start, e.g. the webbrowser. If we change init.rc
> > > > > config to give file-write permissions, Android is not able to boot
> > > > > anymore.
>
> > > > > So we have decided to use a mixture ofyaffs2and jffs2, after we saw
> > > > > this idea at the armv4 port. The basic idea is, that all mmap
> > > > > operations are done onyaffs2, as jffs2 does not support them. At the
> > > > > moment, we split the file system to two parts: /data is located on our
> > > > >yaffs2partition, everything else on our jffs2 partition. The system
> > > > > boots fine and we can run every application. But now, it is getting
> > > > > confusing: As soon as Android has finished booting, it is impossible
> > > > > to write/delete files in /data/app - if we do, we get a kernel panic,
> > > > > which reports FileObserver to fail. This does not happen, if we do
> > > > > file accesses before Android has finished its booting process.
>
> > > > > Remembering that we had some cases, in which it was necessary to start
> > > > > the system with strace running in the background (and discarding the
> > > > > log), I booted theyaffs2/jffs2 system with strace in the background.
> > > > > Now, I am able to access files in /data/app, I just get "syscall:
> > > > > unknown syscall trap 0xe1a00000" reported to my debug console. In this
> > > > > mode, it is also possible to run applications directly from Eclipse on
> > > > > the target device.
>
> > > > > So can anybody tell me what is going wrong, if I use ayaffs2only
> > > > > file system? And why does strace heal those problems with ayaffs2/
> > > > > jffs2 system? It just makes the system slower...
>
> > > > > bye
> > > > > Markus
--~--~---------~--~----~------------~-------~--~----~
unsubscribe: [EMAIL PROTECTED]
website: http://groups.google.com/group/android-porting
-~----------~----~----~----~------~----~------~--~---

Reply via email to