Getting started with Xilinx V4 PPC?

[David Summers]

Thanks for all of the great advice.


I think my plan is to stick with the 2.4 kernel for now until I run
into a compelling reason to switch to 2.6.  It sounds like the 2.6
support is getting very close to maturity and if I get some free time
I would love to pitch in and help.


Is anyone out there using JFFS2 on a NAND flash device in 2.4?  If
this works in 2.4 then I should be OK.  My dev board only has 4MB of
NOR flash so this is one part of the processor subsystem that I cannot
test.

-David Summers


On 1/31/06, David H. Lynch Jr. <dhlii at dlasys.net> wrote:
>
> David;
>
> I just completed a port of 2.6 to the Pico E-12 which is a Xilinx
> Virtex-4 with very little hardware.
>
> I had the advantage of a system that already had a elf loader (not
> U-Boot), but not a Linux loader.
> If I could load Linux as an elf file then I did not need another loader.
> In the end I had to make some very trivial modifications to Pico's
> loader to pass a board information structure. But otherwise the decision
> to use the existing loader saved me having to port u-boot.
>
> I also chose to boot into a ramdisk. Using the existing loader precluded
> having the ramdisk as a separate file.
> the intramfs feature of 2.6 allowed me to wrap the ramdisk into the same
> file as the kernel. Documentation on initramfs is spartan, and somewhat
> opaque. The good news was that despite the documentation initramfs
> proved trivially easy to use.
>
> After that I took the 2.6 kernel source, and litterally went through
> looking for all references to the xilinx ml300.
> Wherever there was an ml300 specific file I created an new one for the
> pico_e12. Wherever there was an ml300 configuration option I created a
> PICO_E12 one.
>
> The E-12 had substatnitially less hardware than the ml300 (or the ml403)
> so mostly I just ended up ripping code out of my private version of the
> ml300 code that was transmuting into the pico_e12 code.
>
> The next obstacle I bumped into was that the e-12 had two devices
> suitable for a console - a highspeed fifo interface to a host
> development system called the keyhole port, and the xilinx uartlite
> serial port. Neither are supported in 2.6. I looked at the Xilinx
> Uartlite 2.4 driver and it was coded substantially different from other
> serial drivers. I decided not to try to port it to 2.6. I started from
> scratch using the 8250 code as a base. I picked the 8250 because:
> It must be the most heavily used linux serial driver and therefore I
> hoped the most uptodate and well debugged.
> The 8250 had boot through console IO support. Most other serial devices
> require getting fairly far into the boot process before you get any output.
>
> The really early IO proved fairly simple. Implimenting xxxx_dbg.c and
> xxxx_tty.c for both the uartlite and the keyhole was fairly trivial.
>
> Additionally the keyhole port had a "debug" feature where if you output
> a single 32 bit word to one of the fifo ports it would be displayed in
> hex on the host. As this could be done in an assembler macro that proved
> indispensiable early on.
>
> I was stalled for about 3 weeks trying to get the sucker into virtual
> mode. In the end I had to disable Machine Check exceptions to make the
> transition to virtual mode. Pico claims there is not a hardware problem,
> but if I do nto disable machine checks on the e-12 it will never make
> the tansition to virtual mode.
>
> The full serial driver for the Uartlite (and Keyhole) proved more
> daunting than anything else. In some ways the 8250 proved a bad choice
> as a template there as there are more permutations, busses, etc
> associated with the 8250 than anything else. I ended up stripping out
> huge hunks of code. Eventually, I used the m32r_sio as a simpler
> template. The driver was much simpler, and it had the critical features
> I needed - both interupt driven and polled IO support - some
> implimentations of the E12 do not include a PIC. The serial drivers took
> much longer than the whole rest of the port combined - including the
> machine check detour.
>
> Just as I was finishing up the E12 port, Grant posted a set of patches
> for the ml403. There were some fundimental differences at a fairly low
> level between Grant's approach and mine. And I think Grant's Virtex-4
> code appears to reflect more of the direction things are going, so I am
> in the process of remodeling my port for the E12 on his for the ml403.
> But I have not completed that.
>
> Absent the machine check issue and having to write serial drivers thus
> was atmost 2 weeks work.
>
>
> BTW while you appear to be a hardware guy with alot of software
> experience, I am a software guy with alot of hardware experience.
>
>
> The E-12 has flash, but it uses a proprietary flash file system, and
> using it would have required writing a filesystem driver - which I would
> be happy to do if I got paid for it. Write performance on flash can be
> extremely slow. If you are recording a significant amount of data in
> realtime, you may want to skip the flash and just use a ramdisk. However
> the ramdisk will not survive the system loosing power - which might be
> an issue in a rocket.
>
> If you are going to start with the 2.6 Kernel, I would get git. use it
> to DL Linus's current 2.6 git repository, adn apply all of Grant's ml403
> patches from the mailing list to it.
> Then I would start an approach somewhat similar to what I described
> above - except using Grant's ml403 as the base to create your own board
> spec from instead of the ml300 that I used.
> How tightly constrained are you for gates in the FPGA ?
> If you are not tight, use the Xilinx 16550 serial IP and then you can
> use standard Kernel 8250 dirvers (making Grant's patches work with
> Uartlite is one of the things I am having trouble with) It would be my
> guess that if you use the 16550 serial IP, you may be able to use
> Grant's ml403 stuff asis.
>
>
>
>
>
> David Summers wrote:
>
> >I am starting a new project where I need to have a flash file system
> >and an ethenet interface (HTTP and FTP).  The project is a solar
> >physics experiment that will be launched on a sub-orbital rocket
> >flight this fall.  The idea is that the experiment will write data
> >files to the flash file system while in flight, and then I can
> >download the data using ethernet after the experiment returns to
> >earth.
> >
> >I think that Linux is the way to go for this project because of the
> >JFFS2 filesystem and the strong networking support.  The only problem
> >is that I am totally new to embedded linux.
> >
> >I am prototyping my system on an Avnet FX12 development board ( Specs
> >here: http://tinyurl.com/4gfdv ) which is pretty similar to the Xilinx
> >ML403 board except with less memory and no SystemACE slot.
> >
> >I will eventually build my own custom board with the same Xilinx
> >Virtex 4 FX12 FPGA and additional flash memory and some custom
> >interface hardware. My background is primarily as a hardware designer,
> >and I am very confortable with the FPGA part of this project.   I am
> >comfortable as a Linux user, and I am a pretty good C coder (for a
> >hardware guy anyway :).  I have never built a linux system (embedded
> >or desktop) before, and I need some help getting started with embedded
> >linux.
> >
> >I have already found the following websites which have been a big help so far:
> >
> >http://www.klingauf.de/v2p/index.phtml
> >http://splish.ee.byu.edu/projects/LinuxFPGA/configuring.htm
> >http://www.crhc.uiuc.edu/IMPACT/gsrc/hardwarelab/docs/kernel-HOWTO.html#toc1
> >
> >I have a few questions that I hope someone can help me with:
> >
> >1. What is the best linux distribution to start out with?  I am
> >currently working with the 2.4 kernel code from
> >ppc.bkbits.net/linuxppc_2_4_devel, but I'm not sure that this version
> >be being updated.  (MontaVista also seems to be making it rather
> >difficult to download their free Linux Preview Kit, so I wouldn't mind
> >finding another distribution)    Does the DENX distribution have good
> >PPC support?  Are there any others that I should look at, or should I
> >just download the kernel source directly from kernel.org?
> >
> >
> >2. kernel 2.4 or 2.6?
> >
> >It is my understanding that the latest version of MTD and JFFS2 have
> >dropped support for kernel 2.4.  I would like to use JFFS2 on a NAND
> >flash device, so it seems like I should use 2.6.  (Can I use JFFS2 on
> >a NAND device with kernel v 2.4?)
> >
> >Being a complete newbie, am I biting off more that I can handle by
> >trying to use 2.6?
> >
> >I have been lurking on the mailing list for a while, and I know that
> >there are several people working on 2.6.x patches for the Xilinx
> >virtex 4 parts.  Could someone point me to a list of the patches that
> >I need to get me started?  I think that the ML403 patches should work
> >with my board, but I don't know which ones I need to download.
> >
> >
> >Thank you for your help,
> >
> >David Summers
> >University of Colorado
> >_______________________________________________
> >Linuxppc-embedded mailing list
> >Linuxppc-embedded at ozlabs.org
> >https://ozlabs.org/mailman/listinfo/linuxppc-embedded
> >
> >
> >
>
> --
> Dave Lynch                                              DLA Systems
> Software Development:                                Embedded Linux
> 717.627.3770    dhlii at dlasys.net         http://www.dlasys.net:8888
>
>