Joyce & Anne, compiled for the ARM platform

Last updated: 17-Apr-12 14:39

This is a port of the Joyce (PCW8256/8512 etc.) Amstrad PCW emulator - which also includes the 'Anne' PcW16 emulator - to various flavours of the ARM platform.

Although it works, this port should be considered incomplete, due to issues with libpng that I was not able to sort out (see below). I have fixed it!1 - but only for the EABI version: the binaries for the 2.6-kernel ARM/Versatile system compiled flawlessly. This perplexes me because yesterday, I couldn't get anything to compile without patching. Be sure to get the joyce-2.2.4-EABI.tar.gz file, because the other ones were from yesterday's error-prone compilation effort.
1: I did nothing more than stop messing with it, and it worked.

I have tested the 2.6-kernel version with an SL-C860 Zaurus, running Angstrom 2011.1 (unstable/current). The emulator was usable, but a little slow. The 'Speed' readout showed that it was running at 158%, though it didn't feel like it! XJoyce uses a lot of memory and on a 64MB Zaurus, you need to stop X and shut down any memory-hogs.


Anne

XAnne also worked, but was very slow. For the setup phase, I had trouble with the disk file emulation: where the rescue disk is requested, I used Cliff Lawson's 'pcwos.zip' disk image, which I found on the Internet Wayback Machine. Initially, the emulator wouldn't read the image and segfaulted, so I used the libdsk utilities to convert the CFI image to a DSK. That seemed to work a little better, although before it wrote the emulated EEPROM, the emulator repeatedly asked me to press the Power key, when it could not read the disk. After about four Power key presses, it accepted it and a screen appeared, showing the EEPROM-writing process. From that point on, setup was all via the Roseanne OS, which is very intuitive. I am running the display over the wireless network; you may have better luck with a 'usbnet' connection.

To start xanne, I used the PcW16 ROM from MESS, and I typed this command (in bold):

LD_LIBRARY_PATH=$PWD ./xanne -E pcw045.sys -A Disks/anne_boot112.dsk
ANNE PcW16 Emulator v2.2.4
I/O warning : failed to load external entity "/home/lex/Joyce/annehw.xml"
ANNE will emulate a PcW 16 with 2048 Mb memory
PcKeyboard::onCommand(ff)
sh: lpr: not found
The LD_LIBRARY_PATH environment variable is necessary because I don't have libstc++.so.6 on the Zaurus, so I copied it over from.. somewhere. Where did I get it? Perhaps I could edit this later, when I'm a little less tired.

Useful pages

The following text is from yesterday's session. As usual with my projects, it's upside-down.

The End

Using a Debian ARM system2 - running on QEMU - I compiled the Joyce PCW emulator for the ARM platform, with a 2.6 EABI kernel. This compiled the 'xjoyce' and 'xanne' binaries.
2: (I think that the site that I read to set-up the ARM QEmu environment was aurel32.net. I forget the full URL.)

The Start

The 2.6-kernel/EABI version - hereafter referred to as 'the EABI version' - was compiled at the end of almost 24 hours of non-stop source-code 'tweaking' and waiting, because I started by compiling this on an OpenZaurus Sharp SL-5500 (Collie) with GNU gcc/binutils and everything mounted over NFS (over Wi-Fi!). This had to be the most batty project so far, and it was a real test of my patience. Compilation was very slow -- as you'd expect for a 206 MHz StrongARM-based system with only 64MB of RAM -- and all file I/O went through a wireless link. I had to observe the entire procedure, because gcc regularly bailed-out with errors relating to recent changes in libpng, or the fact that I had all of the includes and libs in a non-standard location. Daylight arrived and I stacked up lots of boxes at the window, in a futile attempt to keep the light out, and fend off the tiredness. The next time I do this, I might put the development system on a CF card and login to the SL-5500 via USB or IrDA (yes, it is possible to have a shell running over an infra-red link!).

After a long time, the 'xjoyce' binary compiled. 'xanne', unfortunately, was not so straight-forward: after a bit of clanking and whirring, the SL-5500 (the name being 'Attraction', after the famous horse) aborted the compilation with an out-of-memory error.

Stupidity

root@attraction:/# dd if=/dev/zero of=swap bs=1M count=64
root@attraction:/# mkswap swap
root@attraction:/# swapon swap
Here is where I did something very unwise: I added a 64MB swap file, and the filesystem was on an SD card. On the SL-5500, the SD card is generally a lot slower than the CF card; because the WLAN adaptor was in the CF slot, I couldn't use CF - it had to be SD. After adding swap, I resumed the compilation with the simple command 'make'. "Well, it's all going fine!" I exclaimed, and then tried to switch screens to see how much swap it was using. Attraction did not respond. I pressed Ctrl-C: no response. I pressed Ctrl-Z, Ctrl-D and any other key that might abort it - no response..

Crash

The combination of slow SD, limited network bandwidth and lack of memory had essentially killed the system, as soon as I'd typed that 'make' command. Still optimistic about rebooting cleanly, I waited for 10 hours and ran tcpdump to see whether anything was going on, over the network. Attraction wasn't making any NFS requests and although it responding to a ping, it would not let me login: if I typed 'ssh attraction', it wouldn't reply and the connection would eventually time out. At this point, I started wondering whether a 'SysRq on ping' facility existed, where you could activate a SysRq function by using just the 'ping' command. I found out that it does, but it's only for the 2.6 kernel, and Attraction has the good ol' 2.4.18-rmk7-pxa3-embedix kernel.

Good news

Fortunately, the SD card's filesystems were okay - my use of NFS had saved the loop filesystem from damage (at least I did something right!). I should explain that OpenZaurus goes through several stages of boot:
  1. The kexecboot loader starts, from flash.
  2. The OpenZaurus kernel - or other - loads from the SD card.
  3. AltBoot runs and offers a choice of booting from internal flash, booting from SD or CF card, or running a 'rescue shell'.
  4. AltBoot loads its filesystem from a loop file on the SD card.
  5. GPE starts and the wireless card is activated.
  6. NFS mounts are initialised and the serial shell starts.

Conclusion

From this mess, one binary emerged, almost-unscathed: 'xjoyce'. Due to issues with recent changes in libpng, I was not able to get the dot-matrix printer emulation to work - so I disabled this part of the emulator. Everything else seems to work okay. I should mention that 'xanne' compiled okay (2.6-kernel only) but I have not tested it. Please note that when I compiled Joyce on 17th April, I didn't need to patch anything - so the printer emulation should work. I say 'should' because I haven't tested it.

Please see the 'Files' page for downloads, and I'm sorry that you had to read all of the above first. If you'd like to read some more 'waffling', there's the MOO transcript.

Links

Lex (mail: brooknet at imap dot cc), 16th April 2012


  • Downloads/files
  • The projects page
  • Site index page