tl;dr - How do I build a Linux kernel from source for the beagle bone (white), install it, then debug it?
I am trying to break into kernel hacking - specifically porting an experimental kernel feature that has been implemented on x86 over to ARM. I chose the Beagleboard white as a development platform because of the built in J-Tag over USB.
I want to get a development environment set up using openocd and gdb so that I can debug running kernels, and nail down a routine for building Linux from source and installing it onto the beagle board. I have truly scoured the internet and although I feel like all of the pieces may be out there somewhere, I'm having trouble putting everything together.
I hope that responses to this question might be a roadmap for those who come after me and want to start doing some kernel development on ARM.
Thanks!
Things I have tried:
I have NOT tried CCS5 because, as far as I can tell, it is not open source, does not have a command-line interface, and is not fully supported on Linux.
I HAVE run openOCD. I can halt and resume the processor on the beagle bone through a gdb session connected to the OpenOCD process. I have NOT been able to load any meaningful symbol table into that GDB session and therefore have not been able to break on a system call (a milestone I am shooting for).
I HAVE installed many premade kernel images onto the beagle board and have run through the tutorial at http://eewiki.net/display/linuxonarm/BeagleBone, but it did not start from stock linux and hid the necessary configuration options to build for the BeagleBone.
I understand my question is broad, but I asked that way by design in the hopes that people who come after me can find a solid starting point rather than specific snippets of info from people who are stuck at various points along the process. If that is not the proper way to use SO, my sincerest apologies.
Related
I have Fedora installed on my PC and I have a Friendly ARM Mini2440 board. I have successfully installed Linux kernel and everything is working. Now I have some image processing program, which I want to run on the board without OS. The only process running on board should be my program. And in that program how can I access the on board camera to take image from, and serial port to send output to the PC.
You're talking about what is often called a bare-metal environment. Google can help you, for example here. In a bare-metal environment you have to have a good understanding of your hardware because you have to take care of a lot of things that the OS normally handles.
I've been working (off and on) on bare-metal support for my ELLCC cross development tool-chain. I have the ARM implementation pretty far along but there is still quite a bit of work to do. I have written about some of my experiences on my blog.
First off, you have to get your program started. You'll need to write some start-up code, usually in assembly, to handle the initialization of the processor as it comes out of reset (or is powered on). The start-up code then typically passes control to code written in C that ultimately directly or indirectly calls your main() function. Getting to main() is a huge step in your bare-metal adventure!
Next, you need to decide how to support your hardware's I/O devices which in your case include the camera and serial port. How much of the standard C (or C++) library does your image processing require? You might need to add some support for functions like printf() or malloc() that normally need some kind of OS support. A simple "hello world" would be a good thing to try next.
ELLCC has examples of various levels of ARM bare-metal in the examples directory. They range from a simple main() up to and including MMU and TCP/IP support. The source for all of it can be browsed here.
I started writing this before I left for work this morning and didn't have time to finish. Both dwelch and Clifford had good suggestions. A bootloader might make your job a lot simpler and documentation on your hardware is crucial.
First you must realise that without an OS, you are responsible for bringing the board up from reset including configuring the PLL and SDRAM, and also for the driver code for every device on the board you wish to use. To do that required adequate documentation of the board and it devices.
It is possible that you can use the existing bootloader to configure the core and SDRAM, but that may not meet your requirement for the only process running on the board should be your image processing program.
Additionally you will need some means of loading and bootstrapping; again the existing Linux bootstrapper may suit.
It is by no means straightforward and cannot really be described in detail here.
I'm using Arm DS-5 and Xilinx SDK for developing programs on Zynq board.
I'm trying to boot Zynq 702 board from Qspi Flash.
What I've done so far is generating FSBL project from Xilinx SDK, and combining it with my application using Bootgen tool in SDK, then program it into the flash.
There are several questions in my mind.
DS-5 produces an .axf file, Bootgen requires an .elf file. Can I use
the .axf file by just changing its extension to .elf or do I require
some more steps?
Is there a tool that shows the inner structure of an .axf file?
Showing what is where?
And how can I debug if I managed to boot from QSPI. For example I want to debug my application from the beginning of FSBL, is it possible? Because in Qspi Boot, When I power on the board, my application would start running and when I connect with JTAG, it would be in somewhere in my application.
An AXF might have some extra ARM-toolchain magic in it (I'm not sure off-hand), but at heart it's an ELF file - the ARM toolchain provides fromelf for poking around inside them, but other tools like readelf and objdump also work.
I'm not familiar with the Zynq platform so I don't know any specific debugger tricks, but a general one is just to put an infinite loop at the start of your code (possibly using volatile or inline asm trickery if necessary to prevent optimisation) - once the debugger's connected and broken into it, you just move the PC past the loop and continue.
You can totally halt QSPI-booted Zynq via JTAG and do whatever you want with it. However, there are some quirks. Sometimes Zynq goes into some kind of lockup, and JTAG doesn't work at all, and you need to power-cycle before retrying. Some not-so-well-written peripherial might die after starting software over JTAG, so you might need to re-load bitstream first. And there are some Vivado-related bugs (like the one where you cannot re-flash the board unless you downgrade to 2017.2 or change MIO2-6 pulls or patch the FSBL) but i'm not sure if they apply in your case.
I'm into something about writing a "Mock GPU driver" for Linux based systems. What I mean is that, simply I want to write a driver (Behind X-server obviously) to answer X's API calls with some debugging messages.
In other words I want to fool Linux about having an actual GPU. So I can make a test-bed for GUI-accelerated packages in console based systems.
Right now, if I execute a GUI-accelerated package in Linux console based systems; it'll simply dies due to lack of a real GPU (or a GPU driver better I'd say).
So I want to know:
Is it even possible? (Writing a GPU driver to fool Linux about having an actual GPU)
What resources do you recommend before getting my hands dirty in code?
Is there any similar projects around the net?
PS: I'm an experienced ANSI-C programmer but I don't have any clue in real Kernel/Driver development under *nix (read some tutorials about USB driver development though), so any resources about these areas will be really appreciated as well. Thanks in advance.
What you are looking for is actually part of Xorg server suite, and it is called Xvfb (virtual framebuffer).
If you're not afraid of a bit complex bash, you can take a look at Gentoo's virtualx.eclass for an use example (we use it to run tests which require X11).
A good place to start is the Mesa project - it implements OpenGL in software. It has a way to trick the OS into thinking that it is the OpenGL driver.
I am looking for a modern system to do some bare bones Assembly programming (for fun/learning) that does not have the legacy burden of x86 platforms (where you still have to deal with BIOS, switching to protected mode, VESA horrors to be able to output pixels to the screen in modern resolutions/colordepths etc.). Do such systems even exist? I suspect it is not even possible today to do low-level graphics programming without dealing with proprietary hardware.
qemu is likely what you want if you dont want to have to build that stuff in. You wont get as much visibility as to what is going on in the guts of it.
For hardware, beagleboard (dont get the old one get the new one with reasonable connectors, etc), or the open-rd board. I was disappointed with the plug computer thing. The hawkboard I like better than the beagleboard, but am concerned about the big banner about a pcb design problem. The raspberry pi will be out at some point and will also provide what you are looking for. Note that for beagleboard, etc, you dont have to run linux or anything like that, you can write your own binary and xmodem it over or use the network and then just run it, not a problem at all.
The stellaris eval boards all/most have oled displays, monochrome and small but graphics, not sure how much you were after.
earth-lcd used to have an arm based board with a decent sized panel on it.
there is of course the gameboy advance and the nintendo ds. flash/developer cartridges are under $20. the gba is better to start with IMO, as the nds is like two gbas competing for shared resources and a little confusing. with a ez flash cartridge (open source software to program), was easy to put a bootloader on the gba and for like another $20 create a serial cable, I have a serial based bootloader for loading the programs. If you have an interest in this path start with the visual boy advance emulator to get your feet wet and see how you feel about the platform.
If you go to sparkfun.com there are likely a number of boards that either already have lcd connectors that you would mate up with a display or definitely displays and breakout boards that you could connect to a number of microcontroller development boards. Other than the insanely painful blue leds, and the implication that there is 64KB (there is but non-linear 32KB+16+16) the mbed board is nice, up to 100mhz, cortex-m3. I have some mbed samples at github as well that walk you through building an arm binary too boot an arm from flash for those that have not done it (and want to learn that rather than call some apis in a sandbox).
the armmite pro and the maple (sparkfun) are arm based arduino footprint platforms, so for example you can get the color lcd shield or the gameduino
There is the open pandora project. I was quite dissappointed with the experience, after over a year paid another fee to get the unit and it failed within a few minutes. Sent it back and I need to check my credit card statement, maybe we took the return and give it to someone who wants it path. I have used the gamepark gp32 and gpx2, but not the wiz, the gpx2 was fine other than some memory I/O problem in the chip that caused chaotic timing. the thing would run just fine but memory performance was all over the map and non-deterministic. the gp32 is not what you are looking for but the gpx2 might be, finding connectors for a serial cable might be more difficult now that the cell phone cable folks used to cut up is not as readily available.
gen 1 ipod nanos can still be had easily, as well as the older gen ipod classics. easy to homebrew, the lcd panels are easy to get at. grayscale only, maybe only black and white I dont remember. All the programming info is had from the ipodlinux folks.
I have not tried it yet but the barns and noble folks are homebrew friendly or as friendly as anyone on that scale has been so far. the nook color can easily be turned into a generic android device, so I assume that also means you could develop homebrew on the metal, not sure though, have not studied it.
You might look at always innovating, my experience with them was similar to the open pandora folks. These folks started with a modified beagleboard in a box with a display and batteries, then added a couple more products, any one of them should be very open, and homebrew friendly so you can write whatever level you want, boot and run on the metal, no problem. For the original product it was one of those wait for several months things.
I am hoping the raspberry pi becomes the next beagleboard but better.
BTW all hardware is proprietary, it is just a matter of whether they choose to provide programming information or not. vesa came about because no two vendors did it the same way, and that has not changed, you have to still read the dataseets and programmers reference manuals. But as you can see above I have only scratched the surface, and covered the sub or close to $100 items. If you are willing to pay in the thousands of dollars that greatly opens the door to graphics based development platforms that are well documented and relatively sandbox free. many are arm based since arm is the choice for phones, etc and these are phone-like, tablet-like, eval platforms.
The Android emulator is such a beast; it runs a linux kernel and driver stack (including /dev/fb) that one can log into via the android debugger bridge, and run (statically linked) arm-linux-eabi applications. Framebuffer access is possible; see example.
The meta-question rather is, what do you mean by "low-level" graphics programming; no emulator is going to expose all the register and chip state complexity that's behind a modern graphics chip pipeline. But simple framebuffer contents manipulation (pixel buffer access) is surely simple enough, as is experimenting with software rendering in ARM assembly.
Of course, things that you can do with the Android emulator you can also do with cheap physical ARM hardware, like the beagleboard and similar. Real complexity only begins when you want to access "advanced" things - that's anything accelerated functionality beyond just reading/writing framebuffer contents.
New Answer
I recently came across this while looking for emulators to run NetBSD on, but there's a project called GXemul that provides a full-system computer architecture emulation with support for a variety of virtual devices and CPUs. The primary and most up-to-date core looks to be MIPS-based, but it also lists support for emulating the ARM architecture. It even includes an integrated debugger and it sounds like you can just assemble your code into a raw binary with some bootstrapping code and boot it as a kernel inside the emulator from the commandline.
Previous Answer
This isn't an emulator, but if you're interested in having a complete, ARM-based computer that you can develop whatever you want on that doesn't cost much, you should keep an eye on the Raspberry Pi project. They're very close to selling a complete, tiny, low-power ARM-based computer for $25 a piece. It has USB ports, ethernet, video out, and an SD card reader, and can boot Linux, although in your case you'd probably want to boot your own code and access the hardware directly.
EDIT: Looks like Erik already mentioned it.
I bought a Hawkboard and went looking for a JTAG emulator to use for debugging. The only one I seemed certain about was the Spectrum Digital XDS100v2, because the pins matched and I had read about others using it with a Hawkboard. I had hoped to use a GCC ARM toolchain and OpenOCD, but the XDS100v2 apparently only works with TI Code Composer Studio. I was fine with that, because the Hawkboard uses a TI processor anyway and I figured a TI compiler would be able to optimize really well for it. After I received the JTAG emulator, I installed TI CCSv4...
I absolutely HATE IT.
It has scattered files throughout my hard drive, cluttered up my user directory, is a massive pain in the ass to configure, and now it won't even uninstall properly. I really, really want to just switch to a GCC toolchain and OpenOCD/GDB for debugging, but I can't find any way to do that with the XDS100v2.
There was some recent discussion about this on the OpenOCD mailing list, but it looks like licensing issues prevent the team from including direct support for the XDS100v2. I also found a Git commit made around the same time as the discussion that appears to include code for supporting the XDS100v2, but I don't know if this is official or not. I can't really test it, either, because the XDS100v2 doesn't actually install correctly. I have to install CCSv4 to get the drivers, but I refuse to do this on my other machine because I don't want it to get cluttered like the first one. The discussion mentions that the XDS100v2 is actually just a FTDI device, so I tried using a generic FTDI driver, but Windows didn't recognize it.
I guess what I'm asking is this: Is there some way that I can easily get OpenOCD to support the XDS100v2 by somehow using a generic FTDI driver or another method? I spent $80 on this JTAG emulator and I really hate to let it go to waste.
Getting OpenOCD to work with this will be tricky...
First you need to add the USB IDs of you XDS100v2 to the driver inf file. Please note that
you have to choose between the FTDI drivers and libusb drivers depending how you compiled OpenOCD. If you downloaded a binary OpenOCD version, you should use the drivers shipped with it. Once you added the correct USB Vendor and Product ID to inf file, the driver will install (you have to tell windows the correct path). This step is only needed on Windows platforms.
The device manager will tell you the ID numbers on its "Details" page as "Hardware IDs" Property. Is VID_xxxx and PID_yyyy where xxxx is the Vendor id (VID) and yyyy is the Product id (PID).
Next step is to tell OpenOCD the USB ID (the same you used in the .inf file) - look at other interface/*.cfg files that have the line "interface ft2232". The "layout" is tricky,
just use try-and-error on these.
Final step is to make a complete board definition - look for boards that contain the same or similar cpu chips. If reset does not work, try "reset_config none".