[android-porting] Problem porting Android to iPhone

[David Wang]

So titled because my problem is very similar in symptoms to the thread
"Problem porting Android to OMAP". My port to the iPhone runs Debian
and X.org with the multitouch screen, sound, WLAN and USB gadget ether
net working fine. In order to maintain dual-boot capability, I put the
Android images on one of the iPhone's partitions (which is HFS+ on top
of Samsung's Whimory FTL on top of a few NAND chips) and I'm planning
on booting Android from that. I built the images from a checkout with
repo using the generic instructions from http://pdk.android.com. I set
TARGET_USERIMAGES_USE_EXT2 in the environment so I get ext2 images. I
created a 1 MB ext2 image to store the ramdisk files instead of using
it as an actual initramfs. I modified the /init.rc to comment out the
mtd mounts, since my code will be mounting /system, /data and /cache
before Android's init is called.

Android's /init doesn't seem to care if it's PID 1 (I've observed from
reading this mailing list), so I've been calling it from my
busybox/buildroot ramdisk. However, it doesn't seem to work. After
awhile, it will complain "init: critical process 'servicemanager'
exited 4 times in 4 minutes" and will reboot. I commented out the
"critical" specification for the servicemanager service in init.rc and
ran /init under strace. You can see all the logs that resulted as well
as the exact commands I performed here:
http://iphwn.org/planetbeing/postmortem/

kernel-config.txt is my kernel configuration, in case there is some
option I forgot to enable or some option I must not enable. My patches
to kernel/common android-2.6.32 (done on commit
e27f17b5318851395a66cbaf1524ea89ff8f0cb9) is also present in that
directory if those are helpful. Finally, here's what df shows:

# df
Filesystem                Size      Used Available Use% Mounted on
/dev/root                 9.2M      8.1M      1.0M  89% /
tmpfs                    60.8M     96.0K     60.7M   0% /tmp
/dev/nand0p2             14.4G      1.4G     13.0G  10% /mnt
/dev/loop0             1003.0K    278.0K    674.0K  29% /android
/dev/loop1               72.1M     52.2M     20.0M  72% /android/system
/dev/loop2              193.7M     14.8M    168.9M   8% /android/data
/dev/loop3               48.4M    810.0K     45.1M   2% /android/cache

I think that's enough space for Android. /dev/loop0 is ramdisk.img.

From my analysis of the strace files, it seems as if servicemanager is
being sent SIGKILL by an unknown murderer soon after it starts
(http://iphwn.org/planetbeing/postmortem/strace.486.txt). None of
Android's init or its children seem to be to blame from the strace
outputs. Android's init DOES send a SIGKILL to servicemanager but only
after it receives a notification that it has already died. There's no
kernel messages (http://iphwn.org/planetbeing/postmortem/messages.txt)
that indicate that some OOM killer is at work (unless Android's
LOW_MEMORY_KILLER is utterly silent). I certainly did not kill the
process myself.

Ignore the stuff in messages.txt about the multitouch firmware
requests. I didn't bother to send the firmware for that test but it
doesn't make a difference if I load the firmware before running
Android's init.

I don't own an Android-capable phone so I can't even test whether the
images produced by my build are sane at all. Anyone have any hints
about what could be happening? Is there an alternate track I should be
following?


Thanks,

David

--
unsubscribe: android-porti...@googlegroups.com
website: http://groups.google.com/group/android-porting

[mizmit1222]

Hi,

I guess you got very close to make a thing happen.
ashmem - OK
binder - OK
/data fs type - ext2 - OK

Just double checking, but you do chroot to ext2 fs?

One more question.

It seems to me "adbd" is up. Is there any way to connect to your
iPhone
while booting up through USB or WiFi to get adb logcat?

Cheers,

On Apr 16, 1:00 pm, David Wang <mirrorsp...@gmail.com> wrote:
> So titled because my problem is very similar in symptoms to the thread
> "Problem porting Android to OMAP". My port to the iPhone runs Debian
> and X.org with the multitouch screen, sound, WLAN and USB gadget ether
> net working fine. In order to maintain dual-boot capability, I put the
> Android images on one of the iPhone's partitions (which is HFS+ on top
> of Samsung's Whimory FTL on top of a few NAND chips) and I'm planning
> on booting Android from that. I built the images from a checkout with

> repo using the generic instructions fromhttp://pdk.android.com. I set

[David Wang]

On Apr 16, 4:59 am, mizmit1222 <mizmit1...@gmail.com> wrote:
> Hi,
>
> I guess you got very close to make a thing happen.
>  ashmem - OK
>  binder - OK
>  /data fs type - ext2 - OK
>
> Just double checking, but you do chroot to ext2 fs?

Yes, as you can see at the bottom of http://iphwn.org/planetbeing/postmortem/,
both init and strace is run under a chroot environment.

>
> One more question.
>
> It seems to me "adbd" is up. Is there any way to connect to your
> iPhone
> while booting up through USB or WiFi to get adb logcat?

Yes, I've been connecting over wifi and running the commands. I did
actually obtain a logcat log (using cd /android; chroot . /system/bin/
logcat > /data/blah) but I forgot to upload it. It is here:
http://iphwn.org/planetbeing/postmortem/logcat.log.txt (this is from a
different run)

I don't think the logcat log says anything useful. This is after
servicemanager has started and exited a good few times. As you can
see, it seems like AndroidRuntime is being killed before it can do
anything. I'm not sure how adb works or how I can use it over the
network. This is only "day 2" of my Android experience. :)

Another funny thing I've noticed is that if I kill init, the other
processes it spawns keep working and seem to do a little better. Then
I get stuck on Waiting for SurfaceFlinger, even though there's some
messages from SurfaceFlinger inside logcat.

Also it may be important to note that the GPIO inputs have not yet
been implemented by me, and I've compiled in CONFIG_POWER_SUPPLY, but
I haven't yet put in an interface for it to communicate with the PMU
(so Android wouldn't be able to get power supply status or battery
status). The last thing is that the Android USB gadget drivers are not
working, but that is probably because the host driver I am using (Ben
Dooks's s3c-hsotg driver) seems to be almost unusably buggy on my
hardware.

[David Wang]

I've discovered a couple of more clues about the problem, but not
enough to solve it yet. Apparently any process that attempts to reset
its UID before executing execve gets killed. Since servicemanager
wants to run as 1000 (for system), it gets killed. I've tried
replacing the servicemanager executable with a shell script and the
exact strace happens, so the problem has nothing to do with the
servicemanager executable itself.

If I alter init.rc so that all services run as root, I get as far as
the animated Android logo (before an unrelated Java exception stops
the party). I don't suppose anyone has seen this setuid problem
before?

Thanks,

David

On Apr 16, 12:56 pm, David Wang <mirrorsp...@gmail.com> wrote:
> On Apr 16, 4:59 am, mizmit1222 <mizmit1...@gmail.com> wrote:
>
> > Hi,
>
> > I guess you got very close to make a thing happen.
> >  ashmem - OK
> >  binder - OK
> >  /data fs type - ext2 - OK
>
> > Just double checking, but you do chroot to ext2 fs?
>

> Yes, as you can see at the bottom ofhttp://iphwn.org/planetbeing/postmortem/,

> both init and strace is run under a chroot environment.
>
>
>
> > One more question.
>
> > It seems to me "adbd" is up. Is there any way to connect to your
> > iPhone
> > while booting up through USB or WiFi to get adb logcat?
>
> Yes, I've been connecting over wifi and running the commands. I did
> actually obtain a logcat log (using cd /android; chroot . /system/bin/

> logcat > /data/blah) but I forgot to upload it. It is here:http://iphwn.org/planetbeing/postmortem/logcat.log.txt(this is from a

[David Wang]

I found the problem. For some reason my CONFIG_DEFAULT_MMAP_MIN_ADDR
was set to 65536 (a sane default for desktops), but pretty much every
embedded program out there maps text below < 64 KB. The security
capabilities system of Linux was preventing everyone but root from
mmaping addresses below 0x10000, which pretty much prevents all
applications from running under non-root users. After I adjusted that
and allowed init.rc to use different users, stuff started working.

Now I'm getting the lock screen and a whole slew of errors like:

E/AndroidRuntime( 1963): java.lang.RuntimeException: Unable to get
provider com.android.providers.applications.ApplicationsProvider:
java.lang.ClassNotFoundException:
com.android.providers.applications.ApplicationsProvider in loader
dalvik.system.PathClassLoader@43bf49c8
E/AndroidRuntime( 1963): at
android.app.ActivityThread.installProvider(ActivityThread.java:4270)
E/AndroidRuntime( 1963): at
android.app.ActivityThread.installContentProviders(ActivityThread.java:
4042)
E/AndroidRuntime( 1963): at
android.app.ActivityThread.handleBindApplication(ActivityThread.java:
4001)
E/AndroidRuntime( 1963): at android.app.ActivityThread.access
$2900(ActivityThread.java:119)
E/AndroidRuntime( 1963): at android.app.ActivityThread
$H.handleMessage(ActivityThread.java:1901)
E/AndroidRuntime( 1963): at
android.os.Handler.dispatchMessage(Handler.java:99)
E/AndroidRuntime( 1963): at android.os.Looper.loop(Looper.java:
123)
E/AndroidRuntime( 1963): at
android.app.ActivityThread.main(ActivityThread.java:4363)
E/AndroidRuntime( 1963): at
java.lang.reflect.Method.invokeNative(Native Method)
E/AndroidRuntime( 1963): at
java.lang.reflect.Method.invoke(Method.java:521)
E/AndroidRuntime( 1963): at com.android.internal.os.ZygoteInit
$MethodAndArgsCaller.run(ZygoteInit.java:860)
E/AndroidRuntime( 1963): at
com.android.internal.os.ZygoteInit.main(ZygoteInit.java:618)
E/AndroidRuntime( 1963): at
dalvik.system.NativeStart.main(Native Method)
E/AndroidRuntime( 1963): Caused by: java.lang.ClassNotFoundException:
com.android.providers.applications.ApplicationsProvider in loader
dalvik.system.PathClassLoader@43bf49c8
E/AndroidRuntime( 1963): at
dalvik.system.PathClassLoader.findClass(PathClassLoader.java:243)
E/AndroidRuntime( 1963): at
java.lang.ClassLoader.loadClass(ClassLoader.java:573)
E/AndroidRuntime( 1963): at
java.lang.ClassLoader.loadClass(ClassLoader.java:532)
E/AndroidRuntime( 1963): at
android.app.ActivityThread.installProvider(ActivityThread.java:4255)

But that's another story.

> > logcat > /data/blah) but I forgot to upload it. It is here:http://iphwn.org/planetbeing/postmortem/logcat.log.txt(thisis from a

[nate]

Awesome job so far. By chance are you not getting to the point where
dexopt is populating /data/Dalvik-cache? That will throw class not
found exceptions. If it is, wipe the folder's contents and reboot.
Just a guess but easy to test

[David Wang]

It seemed to be a permissions problem, for some reason it managed to
create the rest of the subdirectories in /data/Dalvik-cache correctly
with the exception of two. I had to create the directories myself and
set their permissions manually. After that, everything worked.

[nate]

Good to hear it's working but I wonder what you mean by "everything."
Any chance of getting a copy of your setup?

On Apr 19, 11:59 pm, David Wang <mirrorsp...@gmail.com> wrote:
> It seemed to be a permissions problem, for some reason it managed to
> create the rest of the subdirectories in /data/Dalvik-cache correctly
> with the exception of two. I had to create the directories myself and
> set their permissions manually. After that, everything worked.
>