Thank you for you post. I apologize if I was a bit harsh. It also appears I was incorrect. So please accept my apologies. The link in your blog is what caused my mistake. It took me to the part of the page at Miniand where you are probing the modules in the bash script. The part above that, which I do see now after scrolling up, is the important part. But that part gets hidden in my browserby gnexus - Allwinner A10
Oh! Thank you very much for your post! I did not realize I did not point out near the d/l link on the new image that the Ethernet module (wemac) is not loaded by default in this image. You must enable it and the SATA module manually by uncommenting the lines in /etc/modules. That info should be in the README file in the archive that you must enable the Ethernet and SATA modules on the Mele.by gnexus - Allwinner A10
You know the Debian OS install script makes a lot of sense. Combined with the SD image here on your "rescue system" it would likely make for a good "Debian OS install disk (SD card)." But, after thinking about it a while, an A10 "Rescue System" makes little to no sense to me for the A10. It makes perfect sense on the Dockstar or other embedded-only devices. But I canby gnexus - Allwinner A10
QuoteSpecifically, I'd like to know what partition your configuration file is located on It's always in partition 1 (nanda) AFAIK. The script.bin is called evb.bin on some Mele's, on some tablets it is script_??.bin or something like that. There are a LOT of A10 tablet models out there, and the mfr. can name script.bin whatever they want. Often there are several different scriby gnexus - Allwinner A10
3.0.38 kernel Looks like mnemoc has updated to 3.0.38. I will try to get a newer kernel built with it tested and benchmarked to ensure no regressions. The we can adopt it using Jeff's method. We will be using our own stable kernel based on linux-allwinner but with our own kernel config as above.by gnexus - Allwinner A10
I guess you got your Mele :) Way to go Jeff! Can't wait to play with your rescue system!by gnexus - Allwinner A10
I appreciate your thanks! Lots of questions: Quoteam wondering if you have support for USB DVB Cards in the kernel? Yes! The ones that are currently in the kernel source are all supplied as modules. Unfortunately, not all the DVB drivers are in the existing allwinner kernel sources. Many, but not all. I am planning later to port the remaining DVB and V4L drivers into our kernel image ASAPby gnexus - Allwinner A10
Sorry about the delay on these images. I didn't realize until just a few weeks ago that Wheezy also has an armhf port. My information was outdated since I always use Sid. People have been requesting Wheezy since certain applications need it. Other people seem to think that Sid is too "unstable." Personally, after using Sid for a couple of years now I can say that it is very stable.by gnexus - Allwinner A10
Is is possible to start a company that does nothing but sell overpriced hacked Plugs and phones, and then get it publicized for free on countless tech websites such as Ars, Slashdot, PCWorld, etc? YES!! Look at this: http://pwnieexpress.com/products/elite-plug Look familiar? Yes. That is a kirkwood inside! http://pwnieexpress.com/products/pwnphone Look familiar? Yes. That is a Nokiaby gnexus - Debian
QuoteI would prefer Jeff's solution. So would I, at least for now until we can also get the kernel loaded from SATA too. I would much rather be booting from SATA. SD cards are all very slow and should be avoided for the rootfs if possible. Even the class 10 ones are slower than SATA. But, (this is a rather big butt. . . and the reason I didn't think of Jeff's idea) in orderby gnexus - Allwinner A10
Unfortunately the guy in your link above has done absolutely nothing notable whatsoever. He is actually quite a way behind most people with his progress. The only thing he has done is get the Mali modules to load. We have been doing that for quite a while now here. Our kernel config already has the modules enabled in it. The rootfs and SD card images here already load the Mali modules. That is noby gnexus - Allwinner A10
I'm currently working on getting SATA and USB booting capability into u-boot like is supported on Kirkwood and other platforms. That is what is required to be able to boot those devices. But it is going to be quite a while yet. Unfortunately there are not any low-level A10 SATA, USB or NET drivers in u-boot right now. Allwinner did not implement that in their u-boot, and it seems it has notby gnexus - Allwinner A10
I just compiled the A10 SD card u-boot with a new default environment. The existing one was a joke. But a lot of work is still needed for the A10 u-boot. It has none of the functionality of ours. Here is what is needed: Add scripts to autoload different hardware configs (script.bin) Add scripts for ext2load alongside the existing fatload Add support for the PHY (Ethernet commands such as pingby gnexus - uBoot
What toolchain are you using? Is it the Ubuntu one or something else? I don't use Ubuntu. So I've never used their toolchain. But I have used CodeSourcery to cross-compile u-boot and it worked fine. AFAIK Ubuntu uses the Linaro optimizations in their toolchain. Those may be causing the problem. If you get no compiler errors then you may want to try changing the compiler flags to somethby gnexus - uBoot
Conclusion: It seems that compiling the kernel natively has little to no effect on performance. The variations in performance were very small. They are too small to not be attributable to statistical variances from test to test. That is a positive verification of this XDA MythBuster. There seems to be no advantage for a particular compiler, whether cross-compiling or native, when building tby gnexus - Allwinner A10
Here are the benchmarks for the natively compiled kernel. But first it would be best to show the GCC version: # gcc --version gcc (Debian 4.6.3-8) 4.6.3 # gcc -dumpmachine arm-linux-gnueabihf gcc was tuned to optimize for performance, not size, as much as possible, and to use the ARM Neon as much as possible, with the following compiler flags: export CFLAGS="-O3 -mfpu=neon -ftree-by gnexus - Allwinner A10
Here is more explanation about the above problem: Issue #57 lcd and hdmi as modules may lead to a blank screenby gnexus - Allwinner A10
The next test is to compile the kernel natively on the A10. I am using a Mele A2000 to compile the kernel. The kernel itself is done now. Surprisingly, it is now only 1.92 MB! That is almost 128 kB smaller with no other changes! Interesting. Edit: BTW the compiler flags used (-O3 in specific) were set to optimize for performance, not size.by gnexus - Allwinner A10
Conclusions: The benchmarks for the kernel using the custom cross-compiler flags are all just very slightly better. The differences are so small they could be statistical aberrations. But there is obviously no drawbacks to using them, and there seem to be possible advantages. Since it takes no effort no use the additional compiler flags their use is highly recommended.by gnexus - Allwinner A10
I tried compiling our kernel again using some minor config changes. The changes resulted in little or no changes in performance. Any of the small deviations could have been the result of statistical variations. Therefore the conclusion is that most kernel config changes make little to no difference in lmbench performance. Certain kernel config changes, however, can make a large difference in theby gnexus - Allwinner A10
More Preliminary Conclusions The defconfig and our config are both extremely close in performance. In some tests our config is significantly faster. In certain other tests, such as memory reads, ours is very slightly slower. But due to the nature of benchmarks any small variations could merely be a statistical aberration. They are all both too close to call in most tests. I will continue toby gnexus - Allwinner A10
Since the defconfig had no video it was not a fair comparison with the other tests. So I recompiled again using the defconfig with the only change being having the video drivers built into the kernel. Here are the results of that benchmark: ************************************************************************************************************************************ ]by gnexus - Allwinner A10
Yes! That is all that is needed. Just make sure the following modules to be built-in instead of compiled as modules: DISP Driver Support(sun4i) LCD Driver Support(sun4i) HDMI Driver Support(sun4i) That should solve any display issues for anyone experiencing them. If you don't want to configure the kernel yourself, or just want to save time, you can use this patch to enable the hdmi. Iby gnexus - Allwinner A10
HDMI video output was not working with the kernel sun4i_defconfig I just used to compile and test. It was also not working for me in an earlier defconfig. That is likely a major problem for people. So it would be good to have a defconfig with working video output. Therefore I have compiled another kernel using the defconfig with the only change being built-in video drivers. Hopefully that worksby gnexus - Allwinner A10
Preliminary Conclusions Obviously the A10 is way faster than the GoFlex. There is no contest there. The 3.0.8+ A10 kernel is a bit slower than the 3.0.36+ kernel. That is also an easy conclusion. When it comes to comparing the two different 3.0.36+ kernel configs it gets a bit more difficult to draw any major conclusions. There is no doubt that little to no performance has been lost using tby gnexus - Allwinner A10
. . . and now (drumroll) what you have all been waiting for. (All? There's nobody even here yet except hyena. But hey, I'm very interested and have been waiting patiently now for over an hour for this. So I'll quit typing and get on with the final results;) Here are the results of the A10 defconfig (But first does it even boot? Yes! But the HDMI still does not work. Hey, if theby gnexus - Allwinner A10
Quotehttp://www.memetic.org/raspbian-benchmarking-armel-vs-armhf/ I just visited that page. Unfortunately it is useless crap. There is no detail to the benchmark methodology. The benchmarks are not reproducible in their current form. So they have no credibility whatsoever. The only reproducible benchmark the guy has on his page is a regression. I'm sure that the Raspi armhf Debian distby gnexus - Allwinner A10
Quoteheres an interesting benchmarking posting on the new wheezy armhf for raspi vs old armel : hey. you're interrupting my thread ;) j/k Good post! Thanks for that! It will be really fun to throw in some Raspi benchmarks for comparison! This would actually make for a good html page once the benchmarks are complete and the data summarized. I think it would be best, however, to wait to gby gnexus - Allwinner A10
Testing Methodologies While we are waiting for the kernel to finish compiling lets talk a bit about the details of the platforms, etc. used for these tests. All testing platforms are using the Debian Sid Linux distribution. The GoFlex is running standard Sid. The A10 -based Mele A2000 is running the armhf float version of Sid. Lmbench is being compiled natively on the GoFlex by the followinby gnexus - Allwinner A10
You can see from above that there are no regressions from 3.0.8+ when using the 3.0.36+ kernel with our config. In fact it is noticeably faster in basically every benchmark. That is great. It makes my life a bit easier! But what if it is just that the 3.0.36+ code that is better. What if my new config actually sucks. Does the defconfig run much better, or is it the same or worse. Keep in miby gnexus - Allwinner A10