EEE 701 Android 3.2-RC-1

[mhoppi]

Hey guys,

I've tried to boot up Android 3.2-RC-1 on my EEE 701. It just runs
well with Android 2.3-RC1-eee but I wanted to try 3.2 as its modified
for Tablet-PCs. The boot stucks at "Running busybox ash..."
The last messages it gave me are:

Type 'exit' to continue booting...

Running busybox ash...
Illegal instruction

USe Alt-F1/f2/f3 to switch between virtual consoles
Type 'exit' to enter Android...
Running busybox ash..


I wasnt fast enough to enter exit, so I dont know if it would change
something.

If someone has an Idea how to solve this problem, I really appreciate
it.
Thanks.

[Henri]

I have exactly the same problem and I did some investigations.
- As you mentioned: the gingerbread ISO does not have this problem.
- Problem occurs only on laptops with 'older processors' such as
'Intel(R) Pentium(R) III Mobile CPU 1066Mhz', cpu family 6, model 11,
stepping 1
- Problem not present on laptop with 'Intel(R) Pentium(R) M processor
1700MHz', cpu family 6, model 9, stepping 5
- All programs on the initial ramdisk contained in initrd.img work
fine
- However all programs in /android/system/bin fail with 'Illegal
instruction' including mksh.

So I think honeycomb bionic is using an cpu instruction which is not
supported on older processors, however I didn't define
TARGET_ARCH_VARIANT=x86-atom so sse2 instruction shouldn't be
generated. Maybe some sse2 *are* generated?

[Chih-Wei Huang]

2011-12-7 上午12:09 於 "Henri" <hfma...@gmail.com> 寫道：
>
> So I think honeycomb bionic is using an cpu instruction which is not
> supported on older processors, however I didn't define
> TARGET_ARCH_VARIANT=x86-atom so sse2 instruction shouldn't be
> generated. Maybe some sse2 *are* generated?

Interesting.

Could you compare the change from
gingerbread to honeycomb in bionic
to find the root cause?

[Henri]

Solved!

If a processor is used without SSE2 honeycomb won't work with the
default build, however if you add TARGET_CPU_SMP=false then it works
fine!

make -j8 iso_img TARGET_PRODUCT=generic_x86
TARGET_KERNEL_CONFIG=android-x86_hpomnibookconfig TARGET_CPU_SMP=false

When you don't set TARGET_CPU_SMP to false a multiprocessor version is
built and causes SSE2 cache control instructions such as mfence to be
generated.
It's also a good idea to select the correct processor in the kernel
configuration, which I did in android-x86_hpomnibookconfig

On Dec 7, 3:24 am, Chih-Wei Huang <cwhu...@android-x86.org> wrote:
> 2011-12-7 上午12:09 於 "Henri" <hfman...@gmail.com> 寫道：
>
>
>
> > So I think honeycomb bionic is using an cpuinstructionwhich is not

> > supported on older processors, however I didn't define

> > TARGET_ARCH_VARIANT=x86-atom so sse2instructionshouldn't be

[fuzzy7k]

I am having a similar issue with the switch from honeycomb to ics.
Unlike your situation, I am using the k8 arch, which is dual core and
has support for SSE2. With your findings I am guessing bionic in ics
is tuned for the atom arch.

[fuzzy7k]

I think I found the source of my problem, but since it is off topic
I've posted it in a more appropriate thread.
http://groups.google.com/group/android-x86/browse_thread/thread/27148b6840e06010.

Thanks for the bionic hint Henri.