Error booting latest downloadable Beagle ROM

I get this trying to boot RISC OS on my rev C3 beagleboard with the downloadable Beagle ROM.

...

init mod MUSBdriver

init mod DisplayManagerError: UndefinedInstruction:Undefined instruction at &2000F9C4 (Error Number &80000000) *

Angstrom works fine so I assume my setup is OK.

Hmm, the latest downloadable ROM image is 3 weeks old. And ironically, the bug you’ve encountered was fixed later on in the same day that ROM was built!

If you just try rebooting the board then you might get lucky after a few attempts (it was a bug to do with the cache not being cleaned properly). Hopefully ROOL will put up a new ROM image some time soon.

I’m now getting to a supervisor prompt. Should the keyboard be recognised at this point or do i need to use the serial console?

The keyboard should be recognised. Probably you need to use a powered USB hub.

On my B6 Beagleboard on the OTG port, the USB Hub is initialised a couple of seconds after reaching the command line. Soon later the keyboard initialises and I can type ‘desktop’<enter>. Mouse and Keyboard are ‘Cherry’ The HUB is a 4 port hub with external power supply – on the PCB it says UH-634 Rev1.2 and has a NEC chip D720110AGC.

Plugging in the keyboard after boot up makes it work!

I’m going to set off some more autobuilds on Wed (hopefully) so you should see new versions of stuff on our web site by Thursday.

The latest OMAP build is now on our site. I believe this includes the Squash module.

Thanks for sorting out the autobuilds. Much appreciated.

Has anyone had any success with networking. I can see my network adapter with ifconfig but not in ‘Internet setup’. Firefox works, at the least on local files. Would be good to try it out for real!

Make sure EtherUSB is in System:Modules.Network, and then stick the following in an obeyfile in !InetSetup.Autosense:

set InetSetup$Driver$USB EtherUSB:ej0:EtherUSB:0.06:EtherUSB

I don’t think EtherUSB works with DHCP yet, so you’ll have to go with a static IP for now.

I’m quite surprised Firefox works, maybe I should try it out for myself and see how well it runs :)

EtherUSB works fine with DHCP, at least with the devices I have here as that is how I usually test them (on an Iyonix).

I’ve now got everything configured ok but still no joy. I have two network adapters one’s ax88772 based and the other is pagasus based. With the pegasus the link light comes on when I enter the desktop but does not show up in ‘ejinfo’. The ax88772 shows up in ‘ejinfo’ but I get no link light. :(

Firstly, EtherUSB writes out logging to syslog, if installed, which can help diagnose problems. It should at least log whenever a supported device is (dis)connected.

Try reinserting devices.

Does *EJInfo show non-zero bytes transmitted and/or received? Does it report the link as up?

It sounds like the pegasus device hasn’t been recognised at all, so check that its vendor/product ID pair is in *EJProducts, if not you can set a system variable as specified in the ReadMe then plug the device in again and see if that picks it up. See *Help USBDriver for commands to get USB device info.

Both adapters work fine with linux.

The pagasus device was already in EJProducts. (07A6/8515) I get this in the syslog. Explains why this isn’t working.

USB5: aue_reset: enter

USB5: aue_setmulti: enter

USB5: MII PHY 0 status = dd74

USB5: Removed as unable to start device: EtherUSB: device broken or misbehaving

For the AX88772 device (2001/3C05) the syslog just has as below. Looks like it’s worked but no link light. EJInfo shows ‘Link status: down’ and a few TX bytes but no RX.

USB11: bound to backend ‘AX88772’ as ej0.

It looks like the version of EtherUSB that comes in the archive isn’t ARMv7 safe. If I use the version in the archive my Pegasus device (which is the same vendor/product ID as yours) fails to work. But if I recompile the module with the right settings then it works fine.

James: Can you add the following extra option to the CC line in the makefile? Then the compiler should produce the correct code (no unaligned loads/stores):

Done version to 0.07. Note that the website is now at http://jamesp.drobe.co.uk

That worked. Thanks.

I posted this from Firefox using Risc OS on my Beagleboard. :)

What’s the performance of Firefox like when compared to an Iyonix (assuming you’re in a position to say, of course)?

I can’t comment on the Iyonix. Speedwise it’s not too bad. It does struggle when the site has a lot of graphics or javascript. I’m just impressed that it works at all!

I think you want to try the latest experimental version. That has quite a few performance improvements.

I’m likely to upgrade my Iyonix in the next 6 months, although I haven’t exactly decided to what. Pandora is a likely choice, but the video out situation there isn’t entirely to my liking.

It feels faster. Most sites I’ve tried have been ok.

Got Nesurf working. Had to recompile it with gcc 4.1.1 release 1b and ‘tweak’ it a bit.

I’m getting average network speeds of 1.4MB/s up and down. Tops out around 2MB/s.

I’m finding lots of programs work straight away. The Thunderbird port works. Some do strange things. !Zap doesn’t redraw the screen. !StrongED crashes typing into a task windows. !PDF displays garbage instead of text on alot of PDF files. !SparkFS can’t open zip files. If the source is available recompiling with gcc 4.1.1 seems to fix them.

I get 333 bogomips compared to 200 for RPCemu on a 6000 bogomip PC.

I’ve suddenly noticed my monitor makes an annoying buzzing sound since the beagleboard runs silent.

I’m finding lots of programs work straight away.

It’s great to hear this. Many kudos to Jeffrey!

!StrongED crashes typing into a task windows.

If you could get in touch with me then I’ll see if this can be resolved. You’ll find my email address in the Contact page of the StrongED manual.

The next autobuilt ROM image should make it easier to track down these “strange things”, as alignment exceptions are now enabled by default.

As you’ve discovered, recompiling code with GCC 4 is typically all that’s needed to make it work, because GCC 4 already defaults to the correct settings to avoid using unaligned loads/stores entirely. I’m not sure if there’s an option you can enable to force this behaviour in GCC 3; certainly I’ve never found it myself, so upgrading may be the only way to go. For Norcroft, the ”-memaccess -L22-S22-L41” should be all that’s needed to ensure the code runs on all current ARM architectures. I believe the latest version you can buy from ROOL defaults to this option, but there’s no harm in specifying the option manually if you don’t feel like buying the tools all over again!

Fred and any other developers may want to have a look at the more techie explanation of the differing behaviour of unaligned loads/stores over here

As you’ve discovered, recompiling code with GCC 4 is typically all that’s needed to make it work, because GCC 4 already defaults to the correct settings to avoid using unaligned loads/stores entirely. I’m not sure if there’s an option you can enable to force this behaviour in GCC 3;

Probably there’s a GCC 3 option, but I can’t immediately see it. In any case, I strongly recommend using GCC 4 for everything unless there’s a very strong reason otherwise. I appreciate that we still don’t have a “full” release of 4.1, but it’s a very solid compiler. Almost everything on riscos.info (except some historic legacy stuff) is built with it, so should be good to run.

Got Nesurf working. Had to recompile it with gcc 4.1.1 release 1b and ‘tweak’ it a bit.

The NetSurf autobuilder has used GCC 4 for months, so any build from there should be fine. As for tweaking it, I’d like to know exactly what you’ve changed.