SABRE tablet camera

[Stefan Kalkowski]

Hi Riju,

On 09/12/2016 12:16 PM, Rijurekha Sen wrote:
> I can run the SABRE tablet Genode demo: https://github.com/skalk/genode/blob/i.MX53_tablet_demo/os/src/server/vmm/imx53/README. The demo is on SD card, and the android is on e-mmc. According to https://sourceforge.net/p/genode/mailman/message/33919618/, I changed the e-mmc android to imx-android-r10.4_bundle_BSP.tar.gz, so that it matches the android kernel in the genode demo.  
> 
> I can use the front camera of the SABRE tablet to take pictures, when android is started directly from emmc (sw26 and sw28 reconfigured). But when I boot the genode demo and then start android from within it, pressing the camera icon says “Cannot connect to camera”.
> 
> Question 1: I understand that the genode demo needed the display, touchscreen and capacitive sensors kernel code to be moved to secure world. Did that screw up the camera functionality of the android in the normal world? Maybe the camera needs the display for preview, which it cannot get? Did your secure-world implementation of the display/touchscreen/capacitive sensor drivers come with some APIs, that the normal world should call to access forbidden resources? Are those APIs part of the normal world kernel and uses the smc() functionality to request forbidden resources? Does that mean that the camera is failing, as it is not calling the proper API to request the display?
> 

AFAIK the camera connectors are also connected to the IPU (Image
Processing Unit), which is given to the secure world exclusively. There
is no API to access the camera. There is however a para-virtual
interface to get input events for the designated area controlled by the
normal world.
The camera is not incorporated into this API within the experiment.

> Question 2: My goal is to move the camera device driver to the secure world. At http://genode.org/documentation/articles/trustzone, under the section “Additional Device Driver”, you say that “The task of CREATING the drivers…”. I understand that the android in e-mmc already supported the display, touchscreen and capacitive sensors. So couldn’t you look at the android source and see where these drivers are, instead of creating these drivers using the documentation? Since I need to repeat the process of moving the camera device drivers from the normal to the secure world, some more details of how you approached the problem for touchscreen/display/capacitive sensor will be super useful. For example, looking at the source code of imx-android-r10.4_bundle_BSP.tar.gz, I can see 
> 
> $ls ~/myandroid/hardware/mx5x/libcamera/
> Android.mk     Camera_pmem.cpp  CaptureDeviceInterface.cpp  JpegEncoderInterface.h   PostProcessDeviceInterface.cpp  PP_ipulib.h            V4l2CsiDevice.cpp
> CameraHal.cpp  Camera_pmem.h    CaptureDeviceInterface.h    JpegEncoderSoftware.cpp  PostProcessDeviceInterface.h    V4l2CapDeviceBase.cpp  V4l2CsiDevice.h
> CameraHal.h    Camera_utils.h   JpegEncoderInterface.cpp    JpegEncoderSoftware.h    PP_ipulib.cpp                   V4l2CapDeviceBase.h    V4l2UVCDevice.h
> 
> Do some of those needed to be moved to the secure world kernel, or do we need to write a camera device driver from scratch? 
> 

Of course, you can recycle device drivers written inside the Linux
kernel to execute them within Genode. We did exactly the same to enable
a bunch of devices like USB, Wifi, Intel graphics etc.. But it is not as
simple as just moving some files into the Genode build system. You need
to build a device driver environment (DDE) that feels to be like the
Linux kernel for those modules. Some introduction to this topic can be
found here:

  http://genode.org/documentation/developer-resources/porting_device_drivers

Nevertheless those documentation seems a bit outdated. The approach
changed slighlty in release 16.05:

  http://genode.org/documentation/release-notes/16.05#Device_drivers

Whether it is more easy to port a driver from Linux, or write it from
scratch depends on the complexity of the driver. But if you're planning
to port the camera driver, I would assume you might get duplicate access
between the usage of the IPU within our framebuffer driver, and the
camera driver of Linux. So it might be better to either port all drivers
related to the IPU, or add the camera driver manually.
In fact you have to analyze the Linux kernel's driver, and decide this
on your own.

Regards
Stefan

> Thanks!
> Riju
> 
> 
> 
> 
> 
> ------------------------------------------------------------------------------
> 
> 
> 
> _______________________________________________
> genode-main mailing list
> genode-main at lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/genode-main
> 

-- 
Stefan Kalkowski
Genode Labs

https://github.com/skalk · http://genode.org/