Background (moar info!)

Starting with the Galaxy S2, Samsung began tracking the number of times you flashed custom kernels to your device. This includes the Samsung Galaxy Note, the Galaxy Tab 7 Plus, Galaxy Tab 7.7, and probably many others. If the flash count is greater than 0, your device is marred by some sort of mark to betray the fact that you’ve been tinkering; on the Galaxy S2 it’s a yellow triangle, on the Tab 7 Plus it’s a number displayed during download mode, etc.

After some research by teamhacksung and ChainfireXDA, it was found that the flash counter stored its count in one of the device’s boot partitions (/sys/block/mmcblk0boot0), and that it could be reset. Chainfire even wrote a handy GUI tool called TriangleAway for querying/resetting the counter. The catch is that this partition is only exposed on Ice Cream Sandwich kernels.

TriangeAway… before and after!

As we recently got the Galaxy S2 Exynos 4210 kernel booting on our tablets, I had a hunch this would work. Sure enough, it did!

Thanks to teamhacksung and ChainfireXDA for discovering this ish. Actually, I couldn’t give a flying hoot about the flash counter, but it’s nice to know it can be reset if we need it.

Dunno why it finally decided to work today, but at least I’ve seen that it CAN work; we can only go up from here!

The road ahead

In other news, we’re learning that our Exynos 4210-based tabs have a lot of wireless stuff in common with the HP Touchpad (aka “Tenderloin” in the CyanogenMod world), so hopefully we can all work together to get stable Wi-Fi on these devices.

As far as the tools we’re using, compat-wireless is proving to be a promising route for us, as they have an actively-maintained ath6kl driver (which supports our wireless chipset). Also, the Samsung Origen development board also uses an Atheros chipset and has a working compat-wireless implementation, and Linaro has some literature about this chipset.

Stay tuned!

I just got it working on my alpha CM9 port for the Galaxy Tab 7.0 Plus…

I don’t know how accurate it is, but it does appear to work; you can see it has named and shamed Google Play Store and Chrome.

Background

The Android platform apparently requires the kernel network data logging functionality, but many pre-ICS devices using Linux kernels < 3.0 don't have it. The Honeycomb-based Galaxy Tab 7.0 Plus and Tab 7.7, for example, are using Linux kernel 2.6.36.

Android is quite noisy about the missing stats when using a device without the appropriate kernel support:

F/NetworkStats( 3434): problem reading network stats
F/NetworkStats( 3434): java.lang.IllegalStateException: problem parsing idx 1
F/NetworkStats( 3434):  at com.android.internal.net.NetworkStatsFactory.readNetworkStatsDetail(NetworkStatsFactory.java:300)
F/NetworkStats( 3434):  at com.android.server.NetworkManagementService.getNetworkStatsUidDetail(NetworkManagementService.java:1284)
F/NetworkStats( 3434):  at com.android.server.net.NetworkStatsService.performPollLocked(NetworkStatsService.java:831)
F/NetworkStats( 3434):  at com.android.server.net.NetworkStatsService.updateIfacesLocked(NetworkStatsService.java:743)
F/NetworkStats( 3434):  at com.android.server.net.NetworkStatsService.updateIfaces(NetworkStatsService.java:721)
F/NetworkStats( 3434):  at com.android.server.net.NetworkStatsService.access$000(NetworkStatsService.java:128)
F/NetworkStats( 3434):  at com.android.server.net.NetworkStatsService$8.handleMessage(NetworkStatsService.java:1633)
F/NetworkStats( 3434):  at android.os.Handler.dispatchMessage(Handler.java:95)
F/NetworkStats( 3434):  at android.os.Looper.loop(Looper.java:137)
F/NetworkStats( 3434):  at android.os.HandlerThread.run(HandlerThread.java:60)
F/NetworkStats( 3434): Caused by: java.io.FileNotFoundException: /proc/net/xt_qtaguid/stats: open failed: ENOENT (No such file or directory)
F/NetworkStats( 3434):  at libcore.io.IoBridge.open(IoBridge.java:406)
F/NetworkStats( 3434):  at java.io.FileInputStream.(FileInputStream.java:78)
F/NetworkStats( 3434):  at com.android.internal.net.NetworkStatsFactory.readNetworkStatsDetail(NetworkStatsFactory.java:269)
F/NetworkStats( 3434):  ... 9 more
F/NetworkStats( 3434): Caused by: libcore.io.ErrnoException: open failed: ENOENT (No such file or directory)
F/NetworkStats( 3434):  at libcore.io.Posix.open(Native Method)
F/NetworkStats( 3434):  at libcore.io.BlockGuardOs.open(BlockGuardOs.java:110)
F/NetworkStats( 3434):  at libcore.io.IoBridge.open(IoBridge.java:390)
F/NetworkStats( 3434):  ... 11 more

The fix(es)

You need to patch your kernel to have the appropriate "Quota, tag, owner" match and stats support for netfilter. If you're lucky you'll find someone who's already backported the support for your kernel version (or at least something close). In our case, I found that Cyanogen himself had already patched another 2.6.36 kernel with this functionality and, much to my surprise, the patches applied cleanly to our kernel.

After that, you need to make sure CONFIG_NETFILTER_XT_MATCH_QTAGUID=y in your kernel config, and then you should be golden.

With a load of ~75 and 32 physical cores, the CPU usage on this server is oversubscribed 2.3 times a “normal” 100% load (where 100% of 32 CPUs would be a load of 32.0). Luckily nobody else is using it right now!

The job in question is the highly-parallelizable blastx; a sequence search/alignment tool used by Bioinformaticians to determine if a given protein sequence matches proteins for a known nucleotide coding region.

I’ve been having some problems with Ubuntu 11.10′s default GCC (4.6.2) for some time now. First, I was having problems with self-kang CyanogenMod 9 builds on my Samsung Tab 7 Plus (specifically, adb wouldn’t work). Second, I couldn’t boot Linux kernel 3.3 on my Sandy Bridge Core i7-2600 build server.

I didn’t know the problems were because of GCC until recently, but now that I’ve switched back to 4.4 both of those problems have gone away; I can only assume that something in Ubuntu’s GCC fork is broken. The solution: set GCC 4.4 as the default compiler.

It’s pretty easy to do this in Debian-based distros using the alternatives system.

Install GCC 4.4

First, install GCC 4.4 (and friends):

apt-get install gcc-4.4 g++-4.4 g++-4.4-multilib gcc-4.4-multilib

Set 4.4 to be the default

Then set 4.4 to be higher priority than 4.6:

update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-4.4 100
update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-4.6 50
update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-4.4 100
update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-4.6 50
update-alternatives --install /usr/bin/cpp cpp-bin /usr/bin/cpp-4.4 100
update-alternatives --install /usr/bin/cpp cpp-bin /usr/bin/cpp-4.6 50

Verify that it has worked:

gcc -v
Using built-in specs.
Target: x86_64-linux-gnu
Configured with: ../src/configure -v --with-pkgversion='Ubuntu/Linaro 4.4.6-11ubuntu2' --with-bugurl=file:///usr/share/doc/gcc-4.4/README.Bugs --enable-languages=c,c++,fortran,objc,obj-c++ --prefix=/usr --program-suffix=-4.4 --enable-shared --enable-linker-build-id --with-system-zlib --libexecdir=/usr/lib --without-included-gettext --enable-threads=posix --with-gxx-include-dir=/usr/include/c++/4.4 --libdir=/usr/lib --enable-nls --with-sysroot=/ --enable-clocale=gnu --enable-libstdcxx-debug --enable-objc-gc --disable-werror --with-arch-32=i686 --with-tune=generic --enable-checking=release --build=x86_64-linux-gnu --host=x86_64-linux-gnu --target=x86_64-linux-gnu
Thread model: posix
gcc version 4.4.6 (Ubuntu/Linaro 4.4.6-11ubuntu2)

Done!

Calls, SMS, and USSD requests all seem to be working reasonably well. Here’s a picture of me demonstrating how it works to my girlfriend:

Data seems to work too. Here are a few screenshots from the web browser:

If you’re curious how I got it up and running (like maybe you’re starting a new CyanogenMod port for your device), check out these commits to my p6200 repo on github:

• extract-files.sh: Add RIL stuff
• extract-files.sh: Add more RIL libraries
• Add initial system.prop with RIL settings
• Enable GSM functionality
• Add system properties required for mobile data

Going further…

Basically, you need to make sure your device is inheriting AOSP and CyanogenMod components required for telephony, excerpt from p6200.mk:

$(call inherit-product, vendor/cm/config/gsm.mk)

$(call inherit-product, $(SRC_TARGET_DIR)/product/full_base_telephony.mk)

Also, you’ll need to make sure you’ve extracted/copied proprietary RIL bits from your phone’s stock ROM, excerpt from extract-files.sh:

# RIL
adb pull /system/bin/rild ../../../vendor/$MANUFACTURER/$DEVICE/proprietary/rild
adb pull /system/lib/libril.so ../../../vendor/$MANUFACTURER/$DEVICE/proprietary/libril.so
adb pull /system/lib/libsec-ril.so ../../../vendor/$MANUFACTURER/$DEVICE/proprietary/libsec-ril.so
adb pull /system/lib/libsecril-client.so ../../../vendor/$MANUFACTURER/$DEVICE/proprietary/libsecril-client.so
...
# Prebuilt libraries that are needed to build open-source libraries
PRODUCT_COPY_FILES += \\
    vendor/__MANUFACTURER__/__DEVICE__/proprietary/libril.so:obj/lib/libril.so \\
    vendor/__MANUFACTURER__/__DEVICE__/proprietary/libsecril-client.so:obj/lib/libsecril-client.so

# RIL
PRODUCT_COPY_FILES += \\
    vendor/__MANUFACTURER__/__DEVICE__/proprietary/rild:system/bin/rild \\
    vendor/__MANUFACTURER__/__DEVICE__/proprietary/libril.so:system/lib/libril.so \\
    vendor/__MANUFACTURER__/__DEVICE__/proprietary/libsecril-client.so:system/lib/libsecril-client.so

EOF

Lastly, make sure you’re specifying the correct RIL-related properties (hint: check your stock ROM), excerpt from rild.libpath=/system/lib/libsec-ril.so:

rild.libargs=-d /dev/ttyS0
ro.telephony.ril_class=SamsungHCRIL
ro.telephony.sends_barcount=1
ro.telephony.ril.v3=signalstrength,usehcradio

Hope that helps some future space time traveler looking to port AOSP/CM to his/her device.

I know it’s ONLY a Core i7-2600 (no overclocking!), but maybe I’ll throw an SSD in next… Or! Maybe a Beowulf cluster of mulika mwizis!

As you can see, it is based on Android 4.0.4:

What works

The device is properly recognized as a tablet, so tablet-mode stuff (like the bottom bar) work properly. It’s incredibly fast; most transitions are buttery smooth, applications launch fast, etc. Hardware buttons (power and volume) work properly, and trigger appropriate Android OS functionality.

What doesn’t work

Lots of stuff is still broken. The list is very long!

• Audio
• Bluetooth
• Hardware-accelerated video decoding
• Telephony (voice, sms, data)
• Wifi
• Camera
• GPS
• Ambient light sensor

Here are a few more screenshots to tide you over…

Anyone interested in more details (developers or curious users), should head over to the XDA thread. Sources are available at my github account.

After a few hours of staring at Linux kernel code, reading documentation, and cross referencing fixes for other touch screen drivers, I finally figured out what to do. It’s all in this commit, diff here:

diff --git a/drivers/input/touchscreen/melfas_ts.c b/drivers/input/touchscreen/melfas_ts.c
index 2bb5b95..3b6e9c1 100755
--- a/drivers/input/touchscreen/melfas_ts.c
+++ b/drivers/input/touchscreen/melfas_ts.c
@@ -144,7 +144,8 @@
 	input_report_abs(ts->input_dev, ABS_MT_POSITION_X, x);             \
 	input_report_abs(ts->input_dev, ABS_MT_POSITION_Y, y);             \
 	input_report_abs(ts->input_dev, ABS_MT_TOUCH_MAJOR, amplitude);         \
-	input_report_abs(ts->input_dev, ABS_MT_WIDTH_MAJOR, width); \
+	input_report_abs(ts->input_dev, ABS_MT_PRESSURE, amplitude); \
+	input_report_key(ts->input_dev, BTN_TOUCH, 1); \
 	input_mt_sync(ts->input_dev);                                      \
 } while (0)

@@ -2100,7 +2101,7 @@ static int melfas_ts_probe(struct i2c_client *client, const struct i2c_device_id
 	input_set_abs_params(ts->input_dev, ABS_MT_POSITION_Y, 0, TS_MAX_Y_COORD, 0, 0);
 	input_set_abs_params(ts->input_dev, ABS_MT_TOUCH_MAJOR, 0, TS_MAX_Z_TOUCH, 0, 0);
 	input_set_abs_params(ts->input_dev, ABS_MT_TRACKING_ID, 0, MELFAS_MAX_TOUCH - 1, 0, 0);
-	input_set_abs_params(ts->input_dev, ABS_MT_WIDTH_MAJOR, 0, TS_MAX_W_TOUCH, 0, 0);
+	input_set_abs_params(ts->input_dev, ABS_MT_PRESSURE, 0, 255, 0, 0);

 //	__set_bit(EV_SYN, ts->input_dev->evbit);
 //	__set_bit(EV_KEY, ts->input_dev->evbit);

Basically, Ice Cream Sandwich expects ABS_MT_PRESSURE data and BTN_TOUCH events to be sent from the touch screen driver. Without a properly-patched driver, you have to do funny things to get touch events to work (like using two fingers and mashing icons repeatedly on the screen). Very annoying!

Now that we’re back in business… here’s a picture to prove that it’s up:

Great progress, but still a lot more to do! btw, technically I got ICS on this device before Samsung . w00t.

At first glance it appears I only have 310 megabytes of RAM free. Wrong! The devil is in the details:

If you want to know how much RAM your computer really has free, this is a much better indicator; it’s the amount of RAM which is free if file caches are not counted (notice the “-/+ buffers/cache” row heading on the left side).

Oh noes!! Where is all my RAM?

Computers use memory for both running programs (like Firefox, Rhythmbox, ls, cp, etc), as well as for caching recently-used files. When you unzip linux-3.2.tar.bz2, for example, the file is read into RAM and every single file inside the archive passes through RAM as it is extracted (generally speaking, CPUs can only access data which is in RAM!). The same goes for when you compile Android from source, or when your late-night cron jobs run through the file system cleaning log files and indexing things.

Caches save you boat loads time when accessing files which you’ve recently accessed. It’s by design that the kernel does this, and it’s a good thing.

Waaa! I want `top` to show me all my RAM!

Calm down, br0, it’s not the end of the world. If you so desire, you can instruct the kernel to drop caches immediately:

root@ndechu:~# echo 3 > /proc/sys/vm/drop_caches

Now you see numbers which are probably more like what you were expecting in the first place:

If you're interested, there are various knobs you can tweak to modify the kernel's memory management, caching, swapping, etc; vfs_cache_pressure and swappiness are two good places to start. Also, I just stumbled upon this hilarious site which should help quite a bit: Help! Linux Ate My RAM!.