A modern RISC workstation for a modern (government) era.

While discussing various non x86 boards the topic came up about the Chinese government MIPS based processors, namely the Loongson-3A4000. I’ve tried several times on the past to buy one with no success however 2020 would make up with it as 2 separate leads yielded boards.

To me, there is a great deal of confusion around this setup as it goes by different names, and is sometimes spelt in English, other times Chinese (Simplified & Traditional) with different part numbers and vendors making it kind of confusing.

Various names include, but not limited to:

  • Lemote loongson
  • loongson-3-desktop
  • LEMOTE-LS3A4000-7A1000-1w-V01-pc
  • ICT Loongson-3A R4(loongson-3A4000)
  • Lemote LX-6901
  • Lemote A1901
  • 龙芯3A4000+7A台式机主板

I don’t want to sound like I’m going to just shit all over this thing, but it’s not free, it’s not subsided, and it’s not cheap. Weighing in at 元3,500 RMB the LX-6901 is not a machine for the masses, however it is a non x86 machine for the classes.

Thanks to my day job I was able to get my buyer to trace down several companies, and ads placed on TaoBao and another from AliExpress. Last time I tried both routes, along with the board manufacter Lemote, I was unable to get anything. This time however both leads would work out so now I have 2.

I was told that it would take upwards of 2 weeks for the order(s) to be fulfilled and I should have them in 3 weeks. However 4 business days later this box arrived in my office:

presents from the orient

I suspect they need to temper people’s expectations on shipping, but luckily for me I’m not over seas, although shipping from China to Hong Kong does require a special permit for electronics.

Clearly the tape had been opened several times for various inspections as this shipment was destined to be exported. But props on the dragon tape!

As expected, two boards! Oddly and confusing enough both suppliers insist the boards are different.

Natrually, they are of course identical.

Here is a better glimpse at the board.

The Lemote LX-6901 is not without faults however, it has a memory controller issue and cannot operate correctly with 2 sticks of ram. Luckily my DDR4 extras are 16gb so it’s not bad for messing around. The board also can post ATI boards. However it doesn’t like my Asus Radeon R9 380 Strix board, although it posted fine. I have a few of the FirePro W2100 cards, not a remarkable card, but it does work.

In addition you do need a specific OS for the board, on vendor on AliExpress was unwilling to send me anything, while another on TaoBao was willing to send me UnionTech’s UOS.

System BIOS

the M.2 slot works fine and I was able to boot from USB, and install UOS. The BIOS is very ‘PC’ like, pressing ENTER will enter the bios, and you can change boot priorities, or drop to the UEFI shell if you so please.

BIOS initalization

Installation is pretty easy and straight forward. There is only a few options during install, the desination, if you should accept the default layout (why not)? and a language

The USB stick is slow, but it didn’t matter as I only needed to install twice. The first time I had both memory slots populated, and the board crashed at 5% of the install. I was able to do some searching around and found out about the bad memory controller, so popping out one of the DIMMs and I was able to install and use the machine.

UOS for the MIPS however is a seeming commercial product that is very difficult to buy outside of China (it may very well be allocated only for certain circles as you need a Chinese cellphone number, government ID, and some kind of project id?), although I’m still trying.

Not authorized

The phone support was useless, and I’ve had a few email exchanges on asking if it’s available for purchase, and if so how much. I’ll update if I can figure it out.

What does suck is that while UOS is not authorized you can’t get any OS updates, nor can you enable to root user. So yeah you can’t effectively own the machine with UOS. There is a ‘trial’ mode to enable a 90 day ‘grace’ period so at least I can add GCC and at least build software. Clearly if I can’t sort out UOS I’ll have to dump the kernel / initrd and restore another MIPS64EL Linux distro over the filesystem. Debian has MIPS64EL, and seems to be working to mainline the LX-6901.

UOS bills itself as a Windows replacement, and I have to say that I do enjoy using it. It has some rather ‘Windows’ qualities to it, like the sound mixer, and ease of installing apps from the ‘store’ however if it’s not in the store and it’s Not authorized you are out of luck. As much as I dislike distro of the week nonsense, I do like the idea of thousands of people being hired to flesh it out to make Linux usable, but only time will tell how much of it is translation to Chinese, and how much is developing software.

When it comes to performance the 3A4000 running at 1.8Ghz is faster than a Raspberry Pi 4, however not significantly faster CPU wise. However the big plus the MIPS does have is that it has a far more capable bus with M.2 and PCI-E slots along with SATA giving the board much better IO than any SOC solution like the Pis.

I built the BYTE Benchmark 5.1.3 and did this graph with data from running on a Zero for comparison. I had to adjust scales for some of this so its more visible, however the important data is here that CPU wise they are close together, but in the area of IO the Dragon pulls far ahead. For those who like the Linux boot score, the CPU ‘clocks’ in at 3594.02 BogoMIPS per core.

It’s been a large ‘discovery’ thing, and a long time since I’ve tried to make Linux a ‘daily driver’ and of course the scarcity of MIPS binaries on Linux is going to be an issue, but I’ll have to explore the apparent ARM/x86 compatibility as I can find more information about it.

Since this has been such a learning curve for me as I learn more things I’ll add them to this list:

I don’t know how much (if any) reach my blog has in China, but if anyone knows any more or can help, drop a line!

GXemul for Win32

Luna m88k booted off RAM disk

Don’t get all to excited, it’s a terrible port, but it’s to the point where it can barely run stuff. Although I don’t know how much is me, and how much is GXemul. I probably should have tested on Linux first.

Anyways it’s enough to boot the Luna m88k OpenBSD ram disk up to the single user mode, and poke around. The hard disk doesn’t pick up, and I haven’t even tried the NIC, although the address is looking pretty bogus.

I wanted to try the PMAX version of Mach, but then it hit me, that there is no server to load. And porting the system level from Mach 3.0 to 2.5 looks way more involved than Mach 3.0 being ‘something minor’.

Back on the 88k front, the Luna shipped with something called UniOS-Mach, but good luck finding that in this day & age. I guess I’ll have to go back to Japan.

For the crazy among us, go ahead and try gxemul-0.6.2-ultra-primative.zip The name says just how stable it is.

In the meantime here is a super low resolution capture of the screensaver from a Luna via http://www.nk-home.net/~aoyama/luna88k/

As an update, I added in the timer code from PCemu, and now that the timers appear to be firing some stuff like OS/F 1.0 get’s further!

OS/F 1.0 in single user mode

I need to go through the setup stuff a lot better as this is just untar’d and not setup at all. Not that it’s useful, but here, osf1-barely.7z .

So if anyone downloaded gxemul prior to this update, re-download it again! I put the m88k ramdisk kernel in there too so you can quickly test the Luna 88k emulation.

Tyne RISC PC Deskstation

Wow, another MIPS running Windows NT appears in the wild!

This is a great tear down, and look at a MIPS desktop from back in the day, sold to run Windows NT.  It’s hardware shares a lot from PC’s of the era, from the AT keyboard, serial mouse, ISA slots, with a VL S3 video card.  One day I hope to get a MIPS NT machine, although they are incredibly rare to find.

N64 cross GCC / Binutils for Win32

Building GCC & Binutils for the Nintendo 64

I had a request to help get a GCC+Binutils running as native win32 exe’s something comperable to the ancient ‘ultra’ N64 toolchain done by Kyoto Microcomputer (resume pdf).  One interesting thing about their toolchain is that they used a common object format for MS-DOS, DOS/V and MS-DOS on the PC-98 format, along with Win32.  However the Win32 runtime doesn’t like Win64 environments.  On Win64 the exew32 driver just complains:

Can’t allocate memory (Error Code=487)

However the stubs in all the exe’s reference exegcc98 exegccv DOS extender’s along with a exegcc.  However googling around yields nothing.

Running on a x86 version of Windows, however the tools run and report gcc 2.7.2 release 1.2 and the binutils version is simply 2.6 with BFD version 2.6.  So going with this, and the request to keep it 1997 vintage I went ahead with Gcc and Binutils 2.8.1 as they are the end of the line in both trains of code.

To configure is really a snap, as both support the Windows NT platform directly

sh configure --host=i386-winnt3.5 --target=mips-elf

I guess I should add that I build with TDM GCC 5.1, and I use the incredibly ancient MSYS-1.0.11-rc-1.  But it’s enough to bootstrap and build with!  Since my GCC is much newer, I did have to finagle some things.  Here is a quick list of my notes on what I had changed, and some justification.

Binutils 2.8.1 notes:


make sure this uses MS-DOS rb wb type constraints!


There is no sbrk on my MinGW32 … so comment out all the sbrk stuff.


My sed LOVES UNIX style text files, so this one shouldn’t be in MS-DOS CRLF format.


mkdir only accepts the path on Win32.  Also there is now chown.

Gcc notes:


‘__inline’ for is_reserved_word needs to be commented out.


Set like the following for both ASM_FINAL_SPEC to prevent the t-mips from trying to be run.

#define ASM_FINAL_SPEC “\



Just because we are on Windows NT, doesn’t mean we want an .obj object suffix.


__spawnv : __spawnvp work better as _spawnv : _spawnvp


*((void **)__o->next_free)++ = ((void *)datum);

confuses newer compilers, with this error message:

obstack.h:341:32: error: lvalue required as increment operand

replace it with with:

*(__o->next_free)++ = ((void *)datum);

So at the end I have a cross compiler, and I can generate object files, and link files that the final tool MILD can then use and produce N64 ROM images.  It’s not a 100% solution, as I don’t see any mention of MILD being GNU, however the compiler and binutils is running on Windows 10 x64!

GCC cross compiling to the N64 target

GCC cross compiling to the N64 target

I built a few demos and tested with the 1964 emulator.

And there you have it.  For anyone who cares, you can download the toolchain + source here: winnt3.5_i386-mips_elf-gcc-

Torbjörn Granlund’s Excellent resource on running free OS’s on Qemu

Ever get tired of x86 on x86?  yeah me too.

How to solve that problem?

Simple, grab QEMU, and jump off into all those cool RISC processors of the 1990’s that were going to save us all from the WINTEL hegemony!

Lots of instructions, samples, images, and hints here:


It’s really more comprehensive than I’ve sat down to do, so yeah it’s awesome!

Supported platforms include:



So I stumbled onto LiteBSD while I was trying to see if I can cross compile 386BSD 0.0 from Windows (it compiles, but triple faults on boot.)

LiteBSD is a 4.4 BSD derived OS for PIC32MZ microcontrollers.

And to make things more fun, Serge Vakulenko has a Qemu fork that includes these simulators so you can run them on Linux and OS X.

So what about us poor Windows users?

Well a few tweeks, and only one annoying bug remains, but it’s easy enough to sidestep and it runs!



Even better, I got the console to kind of work, although you can still control+c it to kill Qemu.  I guess I could capture the signal being kind of UNIXy.

For some reason when opening the SD card image, it’s already open by the time pic32_sdcard_init is called.  Or so I suspect.  It gets the file handle of 3 which tells me that it shouldn’t be open.  So my fix is lame but it works.  Since something is holding the file that I can’t see, I launch Qemu like this:

qemu-system-mipsel.exe -machine pic32mz-wifire -nographic -serial vc -serial vc -serial vc -serial mon:stdio  -bios boot-wifire.hex -kernel vmunix.hex -hda litebsd.img

with the SD/HDA being litebsd.img but in pic32_sdcard_init I do this:


So you need a dummy file named litebsd.img (it’s just junk but it needs to exist), so whatever is blocking it will block it, then let pic32_sdcard_init open the file litebsd.img.SD which is the real file.

C:\litebsd>qemu-system-mipsel.exe -machine pic32mz-wifire -nographic -serial vc -serial vc -serial vc -serial mon:stdio -bios boot-wifire.hex -kernel vmunix.hex -hda litebsd.img
WARNING: Image format was not specified for ‘litebsd.img’ and probing guessed raw.
Automatically detecting the format is dangerous for raw images, write operations on block 0 will be restricted.
Specify the ‘raw’ format explicitly to remove the restrictions.
Board: chipKIT WiFire
Processor: microAptivP
RAM size: 512 kbytes
Load file: ‘boot-wifire.hex’, 6916 bytes
Load file: ‘vmunix.hex’, 522408 bytes
sdcard: opened d->fd 3
Card0 image ‘litebsd.img’, 339969 kbytes
Copyright (c) 1982, 1986, 1989, 1991, 1993
The Regents of the University of California. All rights reserved.

4.4BSD-Lite build 13 compiled 2015-01-20
cpu: PIC32MZ2048ECG100 rev A4, 200 MHz
oscillator: system PLL div 1:6 mult x50
real mem = 512 kbytes
avail mem = 344 kbytes
using 18 buffers containing 73728 bytes of memory
spi1 at pins sdi=D14/sdo=C1/sck=D1
spi2 at pins sdi=F0/sdo=D11/sck=G6
spi3 at pins sdi=B10/sdo=C4/sck=B14
spi4 at pins sdi=F5/sdo=G0/sck=D10
uart1 at pins rx=F1/tx=D15, interrupts 112/113/114
uart4 at pins rx=F2/tx=F8, interrupts 170/171/172, console
sd0 at port spi3, pin cs=C3
sd0: type I, size 339968 kbytes, speed 12 Mbit/sec
sd0a: partition type b7, sector 2, size 204800 kbytes
sd0b: partition type b8, sector 409602, size 32768 kbytes
sd0c: partition type b7, sector 475138, size 102400 kbytes
bpf: sl0 attached
bpf: lo0 attached
WARNING: preposterous clock chip time — CHECK AND RESET THE DATE!

starting file system checks.
/dev/rsd0a: file system is clean; not checking
starting network
clearing /tmp
standard daemons: update.
Wed Dec 10 21:06:39 PST 2014

4.4BSD-Lite (bsd.net) (tty4)


So there it is!  As always, you can do the whole telnet console, on port 2023 like the SPARC with:

qemu-system-mipsel.exe -machine pic32mz-wifire -nographic -serial vc -serial vc -serial vc -serial mon:telnet:,server,wait  -bios boot-wifire.hex -kernel vmunix.hex

In this case, I prefer to use the ‘wait’ portion of the server, so I can watch it boot.  Maybe I’m just weird.  But this way you can control+c to your hearts content.

As always, you can download my image here.

Also for those who like graphical serial connections (???) you can launch it like this:

qemu-system-mipsel.exe -machine pic32mz-wifire  -serial vc -serial vc -serial vc -serial vc  -bios boot-wifire.hex -kernel vmunix.hex -sd litebsd.img

And use control+alt and hunt around for s3, and you’ll have your console….. That you can’t paste into.

**EDIT I found out I forgot to link this with static libgcc so there were missing DLL’s.  sorry, I’ve re-linked and now it’ll just work out of the box (tested with clearing my path on Windows 10).  Next I need to add curses support.

Restoring the MIPS Magnum in Qemu 1.6.0

As many of you Windows NT MIPS fans may know (apparently there are more than 3 of us now!) the MIPS Magnum target is broken in the current build of Qemu.  The problem lies in the firmware as it accesses unassigned memory.  Luckily I just received an email from Hervé that details how to fix this!

The good news is that the fix is VERY easy, all you need to do is comment out a single line in target-mips/op_helper.c

In the function mips_cpu_unassigned_access:

void mips_cpu_unassigned_access(CPUState *cs, hwaddr addr,
bool is_write, bool is_exec, int unused,
unsigned size)
MIPSCPU *cpu = MIPS_CPU(cs);
CPUMIPSState *env = &cpu->env;

if (is_exec) {
helper_raise_exception(env, EXCP_IBE);
} else {
helper_raise_exception(env, EXCP_DBE);

Simply comment out the line

helper_raise_exception(env, EXCP_DBE);

And you’ll be able to boot up the NT PROM.

MIPS Firmware

MIPS Firmware

Remember you’ll want to run it something like this:

qemu-system-mips64el -L /tmp -M magnum -m 64 -cdrom WindowsNT4.0-MIPS.iso -hda MIPS.disk -net nic -net user -global ds1225y.filename=nvram  -global ds1225y.size=8200

Where I’m keeping my NT PROM is /tmp (although that is probably a bad idea…)  But you’ll need the NVRam stuff to add extra space for the ethernet MAC address.  The clock is always trashed but at least it is doing something this time!

For the uninitiated, some installation notes can be found here.

GoldenEye’s hidden ZX Spectrum 48x emulator

Unbeknownst to the world, retail versions of GoldenEye contain a ZX Spectrum 48x emulator, merely lacking a program monitor and user access. This video displays the results of a nifty little patch to re-enable the content. In total, nine games are embedded. Each is mapped to keypresses on controller 3. The original access point couldn’t be determined, so now you load the games while on the folder select menu (menu 5).
c left Sabre Wulf
c right Atic Atac
c up Jetpac
c down Lunar Jetman
+ left Alien 8
+ right Gun Fright
+ up Underwurlde
+ down Knight Lore
A button Pssst
(default) Cookie

The video displays short sections from each of the games. (Forgot to map the A button, so Pssst won’t appear in this video.) The emulator is fully-compartmentilzed so you can leave, load another game, or play GoldenEye normally.

The emulator was a side-project of one of the programmers, experimentally inserted in the current game under production to see if Spectrum emulation on the N64 was possible. Access to it was removed before release, but the emulation code itself remained embedded, much like debug and devtool support, within a large chunk of assembly code. The patch, exclusively for the North American release (NGEE), is available here:

Don’t pirate ROMs! In most countries you can legally make your own backup copies and apply the patch to those. No direct links to ROMs of any kind, patched or otherwise. Respect the Fuzz!


Steve Ellis, who originally created the emulator, sent an email to clarify how the original Spectrum ROM was set up. Since it wasn’t included and the copyright has been lifted by Amstrad I’ve included the complete one with the patch. If there is any further issue I’ll reduce it to a copyright-unrestricted version.
Here’s the letter though, and be certain to check out Crash Lab. Really!

I see various posts about this on the web now. I thought I’d clarify the point about spec_rom.seg.rz – the contents of this file fill the bottom 16kb of the Spectrum’s memory. On the original Spectrum, the bottom 16kb would have included the whole operating system (the BASIC programming language, functions for loading from and saving to cassettes, etc.). Since the ROM is copyrighted we weren’t able to use it. However, some of the games that we were emulating wanted to call one or two Spectrum ROM functions. The solution to this was to create an empty ROM (99.9% filled with NOP’s), but with newly-created (copyright-free) replacements for the few short functions that the games needed to call. The games should run with this minimal replacement ROM.

BTW, if you’d like to send any people in the direction of @CrashLab or facebook.com/CrashLab, I’d be grateful. We’re going to release something later this year that hopefully should appeal to fans of “old-school” games.


Steve Ellis
Crash Lab



Check the thread at:
For Steve Ellis’s letter and a snippet debunking DK64 Spectrum emulation.

Also, check out Steve’s latest venture at Crash Lab. Expect some great stuff out of there!

So I thought I’d have some fun with jsmips

Hello World!

Hello World!

The Javascript MIPS emulator.. So after snagging the source, and all went well building the binutils, gcc, then gcc again, the libc. All seemed well except I couldn’t work out exactly what goes where…

So after a bunch of messing around I managed to get it run..  kind of.  I think the php file that proxies the filesystem is broken as I get something like this from the now broken main js page..

function genroot(p){var dir=new Directory(p);function genbin(p){var dir=new Directory(p);dir.children["=listusers"]=remoteFile("./bin/listusers","http://codu.org/jsmips/server/dir.php?f=.%2Fbin%2Flistusers");dir.children["=apropos"]=remoteFile("./bin/apropos","http://codu.org/jsmips/server/dir.php?f=.%2Fbin%2Fapropos");dir.children["=banner"]=remoteFile("./bin/banner","http://codu.org/jsmips/server/dir.php?f=.%2Fbin%2Fbanner");

And yet I get this kind of output:

{error: false, type: ‘d’, cont: [“dtextc.dat+”,”hi+”,”var/”,”hi.c+”,”source/”,”bin/”,”dungn27s.zip+”,”dungeon+”,”./”, “../”]}

Which.. isn’t right enough.  I’m not sure what is up with dir.php ..

Oh well, I was able to build a simple hello world type program, but anything that hopes to pull data off the drive won’t work.  If anyone thinks they can do better my archive of all the bits is here (48MB), and the ‘runnable’ version is here .. hi is about as much fun as it’ll get.