Hi Vincent,

On 01/07/2016 05:01 PM, Vincent Raes wrote:

Hello,

I am currently working on a TrustZone demo using Genode in the Secure world and an Android OS in the Normal world on the SABRElite board.

I've succesfully got the Genode 15.08 base-hw kernel working on the board but I am running in some issues with the Normal world. For this I am using an Android distribution customised for the SABRElite, provided by the SoC distributor.

One major problem is that I'm unable to get any output from the Android OS when it is booted by Genode. When booting Android using u-boot, there is output as expected. This supports the idea that the used Device Tree Blob is correct. The command line options for the kernel are provided during its compilation. This seems to work since booting with u-boot gets output on the specified ports.

When booting with Genode however, there is no output whatsoever. The location of the DTB in the memory is passed to the Normal world using the r2 register, which as far as I know is all that needs to be done.

Any input on this would be much appreciated since you can probably imagine that debugging the Normal world kernel without output is a grueling task

Yes, I know what you mean. I'm afraid this is the fortune of a system developer ;-). You can however check, whether you've enabled "early printk" in your Linux kernel configuration and add "earlyprintk" to your kernel command line. Nevertheless, if your guest kernel stucks before the first outputs during CPU initialization you might follow the way you already used when incorporating 'smc' calls into the kernel code. I used the same mechanism when doing first steps with TrustZone and "guest kernels" running in the normal world.

For now there exists a workaround for debugging using kernel output, thanks to the Secure world. The current debug procedure consists of entering the *smc *command in the code to check whether the Normal world reaches this portion of code. This has allowed me to get the kernel itself running but here I ran into another issue, being the L2 cache used in the SABRElite board.

The Linux kernel tries to configure which can't really be done from the Normal world. It can however make calls to the Secure world to configure the restricted registers. Once the cache is actually enabled, Genode immediately stops though. I imagine this is caused by the fact that Genode does not expect the L2 cache to actually work.

Yeah, I also realized recently that the i.MX6 specific code, which was contributed to Genode, does not make use of the L2 cache at all. In contrast to the Freescale boards, on all other Cortex A9 centered ARM platforms the external L2 cache was already enabled by firmware or u-boot, therefore Genode's base-hw kernel did not take care. When realizing that, I also tried to enable the L2 cache on i.MX6 using the Wandboard. Unfortunately it is much more complex because they used a rather old revision of Cortex A9 for their SoC with a complex initialization routine regarding SMP and cache coherency hardware. I finished the work yesterday:

https://github.com/skalk/genode/tree/issue%231312

Nevertheless, I first have to cleanup the topic branch before it will hit the staging branch of Genode's mainline repository within the next days.

Are there any pointers to correctly configure the L2 cache in Genode to keep it executing after the cache's activation? Alternatively, is it possible to run Android without the L2 cache and how would one best go about achieving this?

You can simply deactivate the L2 cache initialization within the Linux code, and continue your development. Of course, memory access is much much slower without the L2 cache (~ factor 10 here) and Android won't be much useable. Nevertheless, as soon as the L2 cache enabling for i.MX6 enters the staging branch you might switch to that state..

Regards Stefan

Best regards,

Vincent

---

---

genode-main mailing list genode-main@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/genode-main