Full Width [alt+shift+f] Shortcuts [alt+shift+k]
Sign Up [alt+shift+s] Log In [alt+shift+l]
1
BlueSCSI Wi-Fi Desk Accessory 1.1 has been released: wifi_da-1.1.sit (StuffIt 3 archive) db06d941d7322089d8a98ebabf1f9a50b8f75ade80edbb355c127acc9a7fa629 SHA1: c9d6d3d165626ce592a65fd9e2011f361a0f78e4 Changes in this version: Fix masked password entry on Mac OS 8+ Truncate display of long Wi-Fi network names
29th Jan 2024

Stay updated

Get a weekly newsletter with the top 5 articles worth reading every week.

More from joshua stein

On AI

In 2019, I started watching Andreas Kling's programming videos, many of which were live-coding sessions writing his new operating system Serenity OS. He was always pretty fast at writing code, especially since his IDE, Qt Creator, had knowledge of all his C++ code and could auto-complete classes, variable names, and function signatures. Back then it seemed like he had much of the code written in his head and was limited just by his typing speed. Four years later, he switched to CLion which had early GitHub Copilot support for AI-assisted code completion. While initially skeptical of Copilot, he experimented with it and within months, Copilot was writing code blocks, "reading [his] mind", and writing whole functions. His IDE's auto-completion through AI seemed like it had finally caught up with his brain and he could get ideas out into code faster. While I found it fascinating watching Andreas speed up his development, none of it appealed to me for my work for years after. I didn't even use an IDE or an LSP in my editor. I was skeptical of AI and probably wrote it off too soon due to its early failures. I eventually tinkered with Claude in a web browser, asking it questions that I would have otherwise asked a search engine and been directed to a Stack Overflow answer. Once Claude Code came out, I was amazed at the power of giving it access to a directory of code on my laptop and letting it directly read and modify files. Its ability to completely (and quickly) reverse engineer a binary blob with zero context is still pretty awesome. Once local models got good and small enough, I started using Ollama on my M4 Mac Mini to do local offline work. Like many things, it's possible to like and use a technology, language, or tool while rejecting all of the hype, fraud, and negative communities that build up around them (like Rails, or Rust). Of course, you can also choose not to like or use any of those things. You can prefer to use a hand saw instead of a circular saw, or to drive a manual transmission car instead of a self-driving EV. But I think at this point it's naive to think of all AI technology as useless or bad. The AI industry on the other hand… In 2020, taking inspiration from Andreas, I started recording my own videos programming on an 8 Mhz Macintosh 512Ke (later a Macintosh Plus) in the THINK C 5 IDE. I had never actually used a classic Macintosh before and I wanted to learn what it was like to program for them as one would have 40 years ago. I made a rule for myself that I would only write code on the Macintosh itself, typing everything on its Apple M0110 keyboard. No modern computers would be involved in writing, editing, or compiling any of the code for my projects. Since starting that series, I've written (most of) an IMAP e-mail client, a revision control system, a Wikipedia reader, an IRC client, then a multi-user, multi-threaded BBS server. It all taught me a lot about the Macintosh and about C, and ignited my vintage computing hobby which has introduced me to many new friends. One thing about writing my e-mail client that bugged me was that it required a proxy server on a modern server to strip TLS and present plaintext to the poky Macintosh. So after adding Wi-Fi functionality to my Mac Plus, I added a TLS offloading device so my Mac could do TLS decryption independently. That provided the final giant piece of the puzzle needed to write a "modern" web browser (at least something newer than MacWeb that could parse HTML 5) for classic Mac OS, since everything is encrypted these days. I started working on that browser in 2024, adding other protocols such as Gopher (and begrudgingly, Gemini). A couple months ago, I was feeling particularly salty about computers and posted these notes: I have lost all enthusiasm for reading about new software projects, especially dealing with vintage computing. What would have sounded amazing a year or two ago and made me want to engage with the author, I now just assume was something AI spit out and feels hollow and boring so I skim past it. And writing software feels like being Marge in that go-kart race. Work slow and steady for a year writing something neat that had never been brought to a platform before, but someone can hear about it and use Claude to whiz by you in a week and make something with 10 times the functionality. Case in point, I saw these today: A modern web browser for Mac OS 9 with CSS, ES5 JavaScript, and TLS A 3D printer monitoring app for System 7 with video A year ago either would have sounded awesome but now I scroll past because I assume AI wrote it all Today someone posted this, a completely vibe-coded Gopher client for classic Mac OS. Claude even made a fancy website for it. And wouldn't you know it, but some of the code for it even comes from my own projects. The other week, Andreas tweeted this: Feels like “I actually enjoy writing code by hand” is about to become the big 2026 virtue signal for programmers. While I don't totally agree (and I think he's got a bit of audience capture staying on that site), it did cause me to question why I bother writing software for 40-year-old computers that only gets used by a handful of people and why I'm bothered by vibecoded software in this context. I write a lot of my software because I want it, but I'm also happy to make it for others in the community to use. Maybe I'm just being vain though, because I want my labor-of-love to be appreciated by people in the community I'm a part of and if a vibecoded version enters the chat with a ton of features and a slick interface, I'm afraid my work will get drowned out and it'll feel like a lot of wasted effort. My Comet will get Yahoo'd. So if a vibecoded tool is genuinely useful and available now while mine's been in development for years, maybe I should just be happy that other people in the community have something new to use. I can continue working on my version for myself and release it when it's ready. After all, Claude has to take that code from somewhere.

23rd Jul 2026 1 votes
Installing OpenBSD on the Pomera DM250{,XY?}

These are some notes for getting OpenBSD-current running on the Japanese-model Pomera DM250, DM250X, and DM250XY. Table of Contents Top Throat-clearing Making a backup Some DM250 hardware notes Making an OpenBSD installation SD card Installing OpenBSD Post-Installation Building a custom installation ramdisk Throat-clearing Much of my work has not yet been committed upstream so installation currently requires a custom kernel and U-Boot image which are provided here. OpenBSD support is still improving and may not be stable at any given time. Install at your own risk. These risks include: If the battery completely drains due to a software bug, the device may not power on correctly and will not be able to recharge its battery (see below). If the device cannot boot properly into U-Boot, you may need to recover it through a USB cable which may also require having to open the device (see recovery). Note: Do not use these instructions for the US-model DM250US yet. That device has a different power charging chip and keyboard layout. Making a backup This is optional but recommended. The tools and instructions at EKESETE.net should be used to make a full eMMC backup before doing anything. Some DM250 hardware notes The DM250's factory U-Boot detects Right Shift + Left Alt being pressed immediately after power-on and boots the recovery Linux kernel from its recovery_kernel partition instead of its normal kernel partition. The recovery kernel boots the initramfs image at the recovery_boot partition which mounts the SD card and executes any _sdboot.sh script found at its root directory. If you hold Shift + Alt too long at power-on, it will instead boot its graphical factory hardware test program. The recovery kernel and script do not show any output on the console while they boot and run. Hold down Right Shift + Left Alt + Power and count 3 seconds, then let go and it should boot the recovery kernel and _sdboot.sh from the SD card. Once the new U-Boot image is written to the eMMC (even before going through the OpenBSD installer), the DM250 will no longer be able to boot the original Linux kernel or recovery. You'll need to boot something via EFI to write the backup U-Boot image to eMMC to get back to the factory software. This can be done with the OpenBSD installation SD card created here by dropping to a shell. The DM250 will not fully power off with a USB-C cable connected to power (even with the factory Linux software). This means even if you halt -p and it powers itself off, it will immediately power back on. My U-Boot detects if the lid is closed when it first boots and if it's closed, it will halt the boot process so it doesn't boot into OpenBSD while it's closed on your desk trying to charge (though it will still be powered up to U-Boot). A working battery is required to be able to provide enough power even with a USB-C cable connected. If the battery is completely drained, it may not be able to boot up far enough to get to U-Boot for it to be able to tell the charging chip to go into a higher charge mode. By default it does a trickle charge but because it will always try to power on as soon as it has power connected, it may quickly drain the power it just added to the battery trying to boot. In short, don't let the battery drain completely. Making an OpenBSD installation SD card It is possible to install OpenBSD to the installation SD card and boot from it so the factory Linux software remains on the eMMC. It will run a bit slower (it runs slow even on the eMMC) and you will still need to write the new U-Boot image to eMMC, but you can easily revert back to the original U-Boot to resume booting Linux at a later time. These instructions assume you're running OpenBSD (any architecture). You'll need an SD card of a few GB, which is assumed to be sd1 here. Format an SD card with a GPT partition layout, making an EFI partition of at least 100MB and giving the rest to an OpenBSD partition. # fdisk -ygb 204800 sd1 # echo -e "a\n\n\n\n\nw\nx" | disklabel -E sd1 # newfs /dev/rsd1a # newfs_msdos /dev/rsd1i # mount /dev/sd1i /mnt U-Boot and OpenBSD will use that EFI partition as its firmware partition, but the original Pomera recovery software will see it as a normal MSDOS partition and run the /_sdboot.sh script contained on it. Fetch the standard OpenBSD armv7 EFI bootloader to the SD card's EFI partition as /efi/boot/BOOTARM.EFI. # mkdir -p /mnt/efi/boot # cd /mnt/efi/boot # ftp https://cdn.openbsd.org/pub/OpenBSD/snapshots/armv7/{BOOTARM.EFI,SHA256.sig} # signify -C -x SHA256.sig BOOTARM.EFI Fetch my uboot.img to the EFI partition as /uboot.img. This U-Boot image has an embedded copy of the binary device-tree built from my Linux tree for the DM250 (JP). # cd /mnt # ftp https://jcs.org/dm250/uboot.img Fetch this installation script to the EFI partition as /_sdboot.sh. This will get executed by the factory recovery partition's boot system once the DM250 is booted into recovery mode. The script mounts the EFI partition of the SD card, backs up all of the firmware files from the eMMC to the SD card, makes a backup of the current eMMC U-Boot partition to the SD card, and then writes the new U-Boot image from the SD card to the eMMC. # ftp https://jcs.org/dm250/_sdboot.sh Switch to the OpenBSD partition of the SD card and fetch the latest OpenBSD armv7 snapshot disk images to it. # cd / # umount /mnt # mount /dev/sd1a /mnt # cd /mnt # ftp https://cdn.openbsd.org/pub/OpenBSD/snapshots/armv7/{SHA256.sig,base79.tgz,comp79.tgz,game79.tgz,man79.tgz,xbase79.tgz,xfont79.tgz,xserv79.tgz,xshare79.tgz} # signify -C -x SHA256.sig *.tgz Fetch the current bwfm firmware so you can have working Wi-Fi after installation. # ftp http://firmware.openbsd.org/firmware/snapshots/{SHA256.sig,bwfm-firmware-20200316.1.3p5.tgz} # signify -C -x SHA256.sig bwfm-firmware-* Fetch my bsd.rd ramdisk image and bsd kernel. # ftp https://jcs.org/dm250/bsd{,.rd} Your SD card is now ready to go. Unmount it. # cd / # umount /mnt Installing OpenBSD Insert the prepped SD card into the DM250. If it's plugged into power, unplug it. If it's powered up, power it down. Hold down the Right Shift + Left Alt + Power buttons. As soon as the Pomera logo displays, wait about two seconds and let go. The Pomera logo will clear and the recovery kernel should now boot. It will execute the _sdboot.sh script on the SD card which will make a backup of the existing U-Boot partition to the SD card and write the new one. Nothing will be shown on the screen while it's working but it should only take about 30 seconds. Once it's done, it will reboot the device. Note that in OpenBSD, the default keyboard layout matches the Japanese keyboard on the DM250 so things like @ and * are not where they are on a US keyboard. At this point it will boot the new U-Boot with EFI support and video and keyboard drivers for the DM250. If it detected the eMMC and SD card properly, it will boot the OpenBSD EFI bootloader. No EFI variables loaded Loading Boot0000 'mmc 0' failed Booting: Label: mmc 1 Device path: /VenHw(...)/SD(1)/SD(0) disks: sd0* sd1 >> OpenBSD/armv7 BOOTARM 1.23 boot> Enter b bsd.rd to boot the installer. boot> b bsd.rd cannot open sd0a:/etc/random.seed: No such file or directory booting sd0a:bsd.rd: ... When prompted for the root disk, enter ? to see which is which. The eMMC should be sd1 showing 7.3G. Note: When installing to the eMMC, you'll need to manually setup the GPT partition table putting the EFI partition after ~16MB or so to leave room for U-Boot. This will not be needed in the future once the installer is able to recognize the DM250. Press Control + Z to get to a shell so you can manually fdisk. Use (W)hole disk or (E)dit the MBR? [whole] ^Z [1] + Suspended /install # cd /dev # sh MAKEDEV sd1 # fdisk -e -g sd1 Do you wish to write the new GPT? [n] y Writing GPT. Enter 'help' for information sd1: 1> Edit partition 0 which was auto-created, changing it to an ef partition at offset 32768 with size 8192. sd1: 1> e 0 #: type [ start: size ] ------------------------------------------------------------------------ 0: OpenBSD [ 64: 15269791 ] Partition id ('0' to disable) [01 - FF, <uuid>]: [A6] ef Partition offset [34 - 15269854]: [64] 32768 Partition size [1 - 15237087]: [15237087] 8192 Partition name: [OpenBSD Area] EFI sd1*: 1> Now create partition 1 as an OpenBSD (a6) type with the rest of the space. sd1*: 1> e 1 #: type [ start: size ] ------------------------------------------------------------------------ 1: Unused [ 0: 1 ] Partition id ('0' to disable) [01 - FF, <uuid>]: [00] (? for help) a6 Partition offset [34 - 15269854]: [40960] Partition size [1 - 15228895]: [15228895] Partition name: [] OpenBSD sd1*: 1> Print the table to make sure it matches, then w and q. sd1*: 1> p Disk: sd1 Usable LBA: 34 to 15269854 [15269888 Sectors] #: type [ start: size ] ------------------------------------------------------------------------ : Free [ 34: 32734 ] 0: EFI Sys [ 32768: 8192 ] 1: OpenBSD [ 40960: 15228895 ] sd1*: 1> w Writing GPT. sd1: 1> q # Now you can return to the installer, which is asking for disk info. Enter ? to get it to refresh disk information and it should now print out sd1. Enter e to edit the GPT (again). # fg install ? Disk: sd1 Usable LBA: 34 to 15269854 [15269888 Sectors] #: type [ start: size ] ------------------------------------------------------------------------ : Free [ 34: 32734 ] 0: EFI Sys [ 32768: 8192 ] 1: OpenBSD [ 40960: 15228895 ] Use (W)hole disk or (E)dit the GPT? [whole] e You'll now be back at fdisk, just q to quit. You can accept the auto-allocated partition layout or just make one big root and a swap partition. When prompted for the location of the sets, specify disk and then respond no that it's not mounted, and enter sd0 (the SD card), then partition a. The pathname to the sets will be just /. Continue installing, ignoring any errors from signify. The installer should automatically pick up the bwfm firmware and install it. At the end it will try to re-link the kernel which is using the upstream kernel object files, so it's important you copy the custom kernel back over /bsd. Exit to (S)hell, (H)alt or (R)eboot? [reboot] s To boot the new system, enter 'reboot' at the command prompt. # mount /dev/sd0a /mnt2 # cp /mnt2/bsd /mnt/bsd You'll also need to disable reorder_kernel which will run at each boot so it doesn't revert the kernel back to the upstream version. # mv /mnt/usr/libexec/reorder_kernel{,.disabled} # echo "#!/bin/sh" > /mnt/usr/libexec/reorder_kernel # chmod +x /mnt/usr/libexec/reorder_kernel (Note: the _ key is to the left of the up arrow, + is to the right of L.) You can now reboot. Post-Installation The first boot will report an error about bwfm and loadfirmware. You'll need to copy the firmware/nvram_ap6212a.txt file that was backed up to the SD card's EFI partition when U-Boot was flashed. # mount /dev/sd0i /mnt # cp /mnt/firmware/nvram_ap6212a.txt /etc/firmware/brcmfmac43430-sdio.rockchip,pomera-dm250.txt # umount /mnt U-Boot is compiled with boot logo support, so you can put a bitmap in the EFI partition's root directory as logo.bmp to get it to show at boot. To make it show OpenBSD puffy, you can use this image. # mount /dev/sd1i /mnt # cd /mnt # ftp https://jcs.org/dm250/logo.bmp # cd / # umount /mnt The two LEDs near the USB-C port can be addressed once booted. Create an /etc/rc.securelevel that sets them up so they are accessible from a normal kern.securelevel: # cat > /etc/rc.securelevel #!/bin/sh gpioctl -q gpio1 8 set out red_led gpioctl -q gpio1 12 set out green_led ^D # After rebooting, they can be addressed with the gpioctl utility: # gpioctl gpio1 8 1 pin 8: state 0 -> 1 Building a custom installation ramdisk If you want to tinker beyond installing, you'll need a Git clone of my OpenBSD rk3128 tree, my U-Boot pomera-dm250 tree with video and keyboard drivers, and my Linux tree which contains the DTB source files. This is optional and requires being able to compile on an armv7 machine (or cross-compile if you know how), or you can just download the latest image I've already built. Fetch an existing bsd.rd, extract its installation filesystem, build the RAMDISK kernel from my rk3128 tree, and then insert the filesystem into the new kernel. $ cd /tmp $ ftp https://cdn.openbsd.org/pub/OpenBSD/snapshots/armv7/bsd.rd $ rdsetroot -x bsd.rd root.fs $ cd /usr/src/sys/arch/armv7/compile/RAMDISK $ make config $ make $ cp -f obj/bsd /tmp/bsd.rd $ rdsetroot /tmp/bsd.rd /tmp/root.fs The /tmp/bsd.rd file is now a bootable installation image with a custom kernel.

9th Apr 2026 1 votes
Wallops 2.2 Released

A new release of my Wallops IRC client is available: wallops-2.2.sit (StuffIt 3 archive) adc2b473664ec0960b5b028335566831dd69f4a1de40f79e22bf1337292cca88 SHA1: 603c5080c41a0d5752cae49a08376db4e18628b1 This update includes a number of new features and bugfixes: Auto-connect with the last saved settings on startup, if they are available, unless the Command key is held down Add support for sending Pushover notifications when the user's nick is mentioned or directly messaged, when a screen saver is detected to be running; sends to Pushover's API over plain-text HTTP Improve nick highlighting when the user's nick is not at the beginning of the line Speed up parsing of nicknames when joining large channels Add support for the /monitor command which is sent directly to the server Fix bug that may have caused a crash when initially connecting, and when reconnecting

2nd Feb 2026 1 votes
Pomera DM250 Tinkering

The KING JIM Pomera DM250 "digital typewriter" is a small Linux-powered ARM computer that boots up into a custom word processor application. I've been tinkering with it to try to get OpenBSD booted on it. I'd normally wait until the end and write up a proper article explaining everything, but this process is taking a lot longer than I expected so I figured I'd document it all as I go. details { margin-bottom: 1em; } Background KING JIM has made a number of portable word processors starting with the DM5, the DM10 and DM20 with fold-out keyboards, then the DM100 and DM200 which share the form factor with its latest DM250. I only know of KING JIM because stsp@ has their Portabook x86 machine that has required a handful of tweaks to get OpenBSD working on it. The DM250 was only sold in Japan, but the manufacturer recently launched an Indiegogo campaign to launch a US version ("DM250US") with an ANSI keyboard layout and defaulting to English in the software (the Japanese model has English support in its software and can use the keyboard in English, though with its slightly different layout). I learned about this on the writerDeck subreddit which I subscribe to for some reason. The unit measures 10.35" wide by 4.72" tall, with a height of 0.7" when closed. It has a 7" 1024x600 full-color TFT LCD screen, though the DM250's custom word processing software only uses black and white. It weighs 1.4 lbs and has a soft-touch rubber coating on its case. Its DM250US now has a US layout, though the arrow keys were unfortunately moved from an inverted "T" layout on the Japanese DM250 to a horizontal layout. The DM250 is powered by a Rockchip RK3128 quad-core ARM Cortex-A7 processor with 1GB of RAM and about 8GB of eMMC storage. It has a full-size SD card slot and USB-C for charging. It has an AMPAK AP6236 Wi-Fi and Bluetooth SDIO chip which is based on the Broadcom BCM43436. 2025-03-14 I backed the Indiegogo campaign on February 19th and used Buyee to buy a Japanese model DM250 while I waited for the US campaign to end and for mine to ship out. The Japanese DM250 arrived on the 13th and with the aid of this website, I was able to boot into a Debian build and start inspecting how the device worked. I also took backups of the eMMC flash to be able to recover to it if I screw things up. I haven't really been interested in the random armv7 boards that run OpenBSD because they all seemed to be similar while also each having quirks that make them unusable for daily use due to lacking driver support or cheap hardware. The DM250 appealed to me because it was a complete computer with keyboard and screen, not just a lone board with an ethernet port. (Although I'm sure I will eventually come up short on complete driver support on this machine too.) It can turn on "instantly" due to some proprietary software called "LINEOWarp" which integrates into u-boot and the Linux kernel and basically hibernates the machine after booting and writes out its RAM to the eMMC. Upon opening the lid, u-boot directly reads the WARP image and loads it into RAM, bypassing the Linux kernel boot process. I first heard about this type of software from dosdude1's Honda infotainment video which has a similar need for "instant on". 2025-03-19 Trying to get OpenBSD loaded will require updating u-boot on the DM250 to a newer release with EFI support. EFI support was added in 2015 but the DM250 has a build from 2014. But I can't really mess with u-boot until I get access to the UART on the device and I haven't been able to find the UART pins. I tried booting to Linux and printing random garbage to the serial port while I probed every pin on the board with my Saleae looking for serial data. For some reason nothing came out anywhere. Eventually I found this page which shows where the UART pins are, which I definitely probed and found nothing while Debian was running. But once I kept leads on those pins while powering on, I could see u-boot output. Now I can actually see what's going on. U-Boot 2014.10-RK3128-06 (Mar 17 2022 - 14:28:55) CPU: rk3128 CPU's clock information: arm pll = 816000000HZ periph pll = 594000000HZ ddr pll = 600000000HZ codec pll = 400000000HZ Board: Rockchip platform Board Uboot as second level loader DRAM: Found dram banks:1 Adding bank:0000000060000000(0000000040000000) 512 MiB [...] I'm not sure why u-boot shows 512 MB of RAM there when the DM250 has 1 GB, especially when that bank output shows a size of 0x40000000 (1,073,741,824 bytes). 2025-03-31 While trying to solder wires to the UART pins, I damaged one of the pads :/ The device still works otherwise so I'll just sell this one and wait for my US model to arrive. I learned that Rockchip SoCs have a neat feature where if the firmware fails to load a bootloader from eMMC or SDMMC, it will automatically launch into a "MaskROM" mode where it becomes a ugen device over its USB-C cable and allows the attached computer to directly read and write data to the eMMC. This way the device can never really be bricked which makes me more confident testing u-boot changes. This MaskROM mode works even before SDRAM is initialized, so the first thing that has to be done is sending it a RAM training blob, then a more complete usbplug blob which allows more complicated commands over USB. This can be done with rkflashtool or xrock which both work on OpenBSD. $ doas xrock maskrom rk3128_ddr_300MHz_v2.12.bin rk3128_usbplug_v2.63.bin After uploading the blobs, the device detaches and reattaches into its USB loader mode: ugen0 at uhub3 port 3 "vendor 0x2207 product 0x310c" rev 2.00/1.00 addr 9 ugen0 detached ugen0 at uhub3 port 3 "RockChip USB-MSC" rev 2.00/1.00 addr 9 If the flashed u-boot does boot but it's broken, one can short the eMMC to ground while the board is being powered on and force it into MaskROM mode. On the DM250, this can be done by shorting TP501 to ground. 2025-04-02 My DM250US arrived. A quick teardown shows it's basically the same hardware but with a different version silkscreened. Pomera DM250 US board UART pins TP802-TP804 The keyboard keys feel slightly smaller in size and rougher in texture. u-boot appears to be the same version but the build date is newer: U-Boot 2014.10-RK3128-06 (Oct 07 2024 - 17:22:56) The kernel is still Linux 3.10.0 with WARP patches. The DTB stored on the eMMC is mostly the same but with these additions: bq27z558-battery@55 { compatible = "ti,bq27z561"; reg = <0x55>; gpios = <0x76 0x12 0x01 0x75 0x0d 0x01>; status = "okay"; }; bq256xx-charger@6b { compatible = "ti,bq25620"; reg = <0x6b>; gpios = <0x76 0x11 0x01>; ti,watchdog-timeout-ms = <0x00>; charge-current-limit-microamp = <0x2bf200>; charge-voltage-limit-microvolt = <0x408b70>; input-current-limit-microamp = <0x2dc6c0>; minimal-system-voltage-microvolt = "\0.c"; pre-charge-control-microamp = <0x8d9a0>; termination-control-microamp = <0x249f0>; ti,no-thermistor = <0x01>; status = "okay"; }; 2025-04-03 I got this pogo-pin clip from Adafruit to access the UART pins without having to solder to them and potentially damage them again. It's definitely made it much easier to reliably access the UART across multiple reboots. 2025-04-14 I've been trying to get different u-boot trees compiling and booting but none were working except the one from KING JIM. I tried rockchip-linux/u-boot and linux-rockchip/u-boot-rockchip but neither boot (or at least don't output anything to uart1. Geniatech make the XPI-3128 which is basically the rk3128-evb evaluation board that exists in u-boot. While digging around their documentation, I found this huge tarball that includes a snapshot of their u-boot tree which is based on newer 2017.09 and has most of the necessary Rockchip drivers. I'm not sure why Rockchip is so special that they can't do everything in the official upstream u-boot tree… With a few changes to build with a newer gcc, setting CONFIG_DEBUG_UART_BASE to 0x20064000 (uart1 instead of uart2), adding some custom uart initialization code to arch/arm/mach-rockchip/rk3128/rk3128.c, and adding an rk3128-specific timer driver, I now have a working build of u-boot that has EFI support! U-Boot 2017.09-g5d36672-dirty (Apr 12 2025 - 12:31:04 -0500) Model: KING JIM Pomera DM250US DRAM: 512 MiB Sysmem: init Relocation Offset: 00000000, fdt: 00000000 Using default environment dwmmc@10214000: 1, dwmmc@1021c000: 0 mmc_init: err -110, timer:41969 switch to partitions #0, OK mmc0(part 0) is current device Bootdev: mmc 0 MMC0: High Speed, 52Mhz PartType: RKPARM rockchip_get_boot_mode: Could not found misc partition boot mode: normal Found DTB in resource part DTB: rk-kernel.dtb CLK: (uboot. arm: enter 600000 KHz, init 600000 KHz, kernel 0N/A) apll 600000 KHz dpll 600000 KHz cpll 400000 KHz gpll 594000 KHz armclk 600000 KHz aclk_cpu 148500 KHz hclk_cpu 74250 KHz pclk_cpu 74250 KHz aclk_peri 148500 KHz hclk_peri 74250 KHz pclk_peri 74250 KHz => mmcinfo Device: dwmmc@1021c000 Manufacturer ID: 11 OEM: 100 Name: 008GB Timing Interface: High Speed Tran Speed: 52000000 Rd Block Len: 512 MMC version 5.1 High Capacity: Yes Capacity: 7.3 GiB Bus Width: 8-bit Erase Group Size: 512 KiB HC WP Group Size: 4 MiB User Capacity: 7.3 GiB WRREL Boot Capacity: 4 MiB ENH RPMB Capacity: 4 MiB ENH => bootefi bootefi - Boots an EFI payload from memory Usage: bootefi <image address> [fdt address] - boot EFI payload stored at address <image address>. If specified, the device tree located at <fdt address> gets exposed as EFI configuration table. Unfortunately only the eMMC (dwmmc@1021c000) is working but the probe of the SDMMC device at dwmmc@10214000 times out. This means I can't see an inserted SD card and begin to boot OpenBSD's EFI loader. I think this has to do with the device not being powered up at boot. I'm still trying to figure out what is required for this to work since it works in other older Rockchip-specific u-boot trees. 2025-04-14 (Evening) Success! U-Boot 2017.09-gcc6b241-dirty (Apr 14 2025 - 17:20:37 -0500) Model: KING JIM Pomera DM250US DRAM: 512 MiB Sysmem: init Relocation Offset: 00000000, fdt: 00000000 Using default environment dwmmc@10214000: 1, dwmmc@1021c000: 0 RKPARM: Invalid parameter part table switch to partitions #0, OK mmc1 is current device switch to partitions #0, OK mmc0(part 0) is current device Bootdev: mmc 0 MMC0: High Speed, 52Mhz PartType: RKPARM rockchip_get_boot_mode: Could not found misc partition boot mode: normal Found DTB in resource part DTB: rk-kernel.dtb In: serial Out: serial Err: serial CLK: (uboot. arm: enter 600000 KHz, init 600000 KHz, kernel 0N/A) apll 600000 KHz dpll 600000 KHz cpll 400000 KHz gpll 594000 KHz armclk 600000 KHz aclk_cpu 148500 KHz hclk_cpu 74250 KHz pclk_cpu 74250 KHz aclk_peri 148500 KHz hclk_peri 74250 KHz pclk_peri 74250 KHz => setenv fdtfile rk3128-pomera-dm250us.dtb => load mmc 1 ${kernel_addr_r} efi/boot/bootarm.efi reading efi/boot/bootarm.efi 119296 bytes read in 51 ms (2.2 MiB/s) => bootefi ${kernel_addr_r} ${fdt_addr_r} ## Starting EFI application at 62008000 ... FtlInit fffffffe Scanning disk [email protected]... rkparm_init_param_from_storage param read fail RKPARM: Invalid parameter part table Scanning disk [email protected]... Scanning disk [email protected]... Scanning disk [email protected]... rkparm_init_param_from_storage param read fail RKPARM: Invalid parameter part table Scanning disk [email protected]... rkparm_init_param_from_storage param read fail RKPARM: Invalid parameter part table Scanning disk [email protected]... rkparm_init_param_from_storage param read fail RKPARM: Invalid parameter part table Found 6 disks Adding bank: 0x60000000 - 0x80000000 (size: 0x20000000) disks: sd0* sd1 sd2 sd3 sd4 sd5 sd6 sd7 sd8 sd9 sd10 sd11 sd12 sd13 sd14 sd15 sd16 sd17 sd18 sd19 sd20 sd21 sd22 sd23 sd24 sd25 sd26 sd27 sd28 sd29 sd30 >> OpenBSD/armv7 BOOTARM 1.23 boot> cannot open sd0a:/etc/random.seed: No such file or directory booting sd0a:/bsd: 2411324+767888+11506208+484492 [188357+107+388448+214048]=0x0 I added some debug printfs to the working u-boot tree and saw that it was calling rk_iomux_config(RK_UART2_IOMUX) when initializing the storage. That ends up calling rk_uart_iomux_config() which does some magic writes to the IOMUX. Reading the GRF documentation and other pieces of code, I learned that GPIO1B needs pins 12 and 14 enabled to activate mmc0_pwren and mmc0_cmd, and GPIO1C needs pins 0, 2, 4, 6, 8, and 10 enabled to change them from JTAG and UART2 pins to those needed for eMMC. With that, the SD card is recognized and u-boot can read files from it with its built-in FAT filesystem support. The existing config on the eMMC splits up the single drive into many different partitions like kernel, warp, ro_data, etc., which each show up as separate disks to the EFI loader. The EFI loader is read from the SD card and loaded into memory with load mmc 1 ${kernel_addr_r} efi/boot/bootarm.efi, and then executed with bootefi ${kernel_addr_r} ${fdt_addr_r}. OpenBSD's BOOTARM.EFI loads successfully and can list files on the SD card and start reading and booting bsd.rd. Unfortunately it goes off into lala land there so I'm not sure what it's doing, but at least now I can move on to the OpenBSD part of this bringup. 2025-04-17 I've pushed my U-Boot tree to GitHub as it seems to be in a good state now. I split up my changes specific to the rk3128 and then added a specific board config for the DM250. Eventually this will need some work to enable the LVDS LCD at boot time like it was with the factory U-Boot. I added a uart_putc helper to OpenBSD's armv7 locore0.S to print numbers to the serial port, and then added them along the boot path to see how far it got. .globl uart_putc /* send r1 to uart */ uart_putc: ldr r0, =0x20064000 str r1, [r0] ldr r2, =0x20064000 + 0x7c /* UART_USR */ check_usr: ldr r3, [r2] tst r3, #(1<<1) /* UART_TRANSMIT_FIFO_NOT_FULL */ beq check_usr bx lr [...] start_mmu: mov r1, #'1' bl uart_putc [...] mov r1, #'2' bl uart_putc /* Enable MMU */ mrc CP15_SCTLR(r0) orr r0, r0, #CPU_CONTROL_MMU_ENABLE mcr CP15_SCTLR(r0) isb mov r1, #'3' bl uart_putc This showed it was getting to start_mmu but as soon as it wrote the SCTLR register to enable the MMU, it stopped outputting. Mark pointed out that this was because there was no mapping in the MMU page table to continue accessing the UART at 0x20064000. I added an entry for it: MMU_INIT(0x20000000, 0x20000000, 1, L1_TYPE_S|L1_S_V7_AP(AP_KRW)|L1_S_V7_AF) But it still wasn't printing '3'. After a few hours of debugging and reading more docs, I finally realized that my dumb uart_putc function was clobbering r0 and r1 which were being used inside of start_mmu so the page table wasn't getting set up right. By changing it to just a few inline instructions with no FIFO status checking and using registers that weren't in use, it could enable the MMU properly and get to '3' and beyond: ldr r4, =0x20064000 mov r5, #'3' str r4, [r5] Eventually with some more tweaks to the DTB passed from U-Boot to the EFI loader and to the kernel, the kernel could properly print to the chosen stdout-path and get to copyright. Since it is able to do this through the normal com_fdt_init_cons routine in dev/fdt/com_fdt.c which does a bus_space_map, I could remove all of my debugging from locore0 and then remove my custom UART page table entry. It can now get to copyright with no kernel changes: disks: sd0* sd1 sd2 >> OpenBSD/armv7 BOOTARM 1.23 boot> b bsd.arm cannot open sd0a:/etc/random.seed: No such file or directory booting sd0a:bsd.arm: 4910236+1012484+138796+608784 [2789902+360416+184+330342]=0x0 OpenBSD/armv7 booting ... arg0 0xc0caf850 arg1 0x0 arg2 0x9ac83000 Allocating page tables IRQ stack: p0x60cde000 v0xc0cde000 ABT stack: p0x60cdf000 v0xc0cdf000 UND stack: p0x60ce0000 v0xc0ce0000 SVC stack: p0x60ce1000 v0xc0ce1000 Creating L1 page table at 0x60cb0000 Mapping kernel Constructing L2 page tables undefined page type 0x2 pa 0x60000000 va 0x60000000 pages 0x2000 attr 0x8 type 0x7 pa 0x62000000 va 0x60000000 pages 0x6000 attr 0x8 type 0x4 pa 0x68000000 va 0x68000000 pages 0x7 attr 0x8 type 0x7 pa 0x68008000 va 0x60000000 pages 0x32c7b attr 0x8 type 0x2 pa 0x9ac83000 va 0x9ac83000 pages 0x7 attr 0x8 type 0x7 pa 0x9ac8a000 va 0x9ac8a000 pages 0x4 attr 0x8 type 0x7 pa 0x9ac8e000 va 0x9ac8e000 pages 0x2 attr 0x8 type 0x7 pa 0x9ac90000 va 0x9ac90000 pages 0x1 attr 0x8 type 0x2 pa 0x9ac91000 va 0x9ac91000 pages 0x100 attr 0x8 type 0x2 pa 0x9ad91000 va 0x9ad91000 pages 0x1e attr 0x8 type 0x6 pa 0x9adaf000 va 0x9adaf000 pages 0x1 attr 0x8000000000000008 type 0x0 pa 0x9adb0000 va 0x9adb0000 pages 0x1 attr 0x8 type 0x0 pa 0x9adb1000 va 0x9adb1000 pages 0x1 attr 0x8 type 0x0 pa 0x9adb2000 va 0x9adb2000 pages 0x1 attr 0x8 type 0x0 pa 0x9adb3000 va 0x9adb3000 pages 0x1 attr 0x8 type 0x0 pa 0x9adb4000 va 0x9adb4000 pages 0x1 attr 0x8 type 0x0 pa 0x9adb5000 va 0x9adb5000 pages 0x1 attr 0x8 type 0x0 pa 0x9adb6000 va 0x9adb6000 pages 0x1 attr 0x8 type 0x2 pa 0x9adb7000 va 0x9adb7000 pages 0x308c attr 0x8 type 0x5 pa 0x9de43000 va 0x9de43000 pages 0x1 attr 0x8000000000000008 type 0x2 pa 0x9de44000 va 0x9adb7000 pages 0x21bc attr 0x8 pmap [ using 3481620 bytes of bsd ELF symbol table ] Copyright (c) 1982, 1986, 1989, 1991, 1993 The Regents of the University of California. All rights reserved. Copyright (c) 1995-2025 OpenBSD. All rights reserved. https://www.OpenBSD.org Now I just need to figure out how far into init_main.c it's getting and why it hangs after printing the copyright line. 2025-04-21 With some instrumenting I figured out the kernel was getting as far as setting up the page tables for the MMU and would then lock up when doing a memset on the newly setup memory. By reducing the amount of memory used, I could get it to fully boot the kernel to !cold, but it crashes in userland: U-Boot 2017.09-g9333465-dirty (Apr 21 2025 - 13:38:25 -0500) Model: KING JIM Pomera DM250 DRAM: 1 GiB Sysmem: init Relocation Offset: 3ddc2000, fdt: 00000000 Using default environment Failed to load DTB Failed to get kernel dtb, ret=-1 In: serial Out: serial Err: serial Model: KING JIM Pomera DM250 dwmmc@10214000: 1, dwmmc@1021c000: 0 switch to partitions #0, OK mmc1 is current device switch to partitions #0, OK mmc0(part 0) is current device Bootdev: mmc 0 MMC0: High Speed, 52Mhz ## Unknown partition table type 0 PartType: <NULL> rockchip_get_boot_mode: Could not found misc partition boot mode: normal CLK: (uboot. arm: enter 600000 KHz, init 600000 KHz, kernel 0N/A) apll 600000 KHz dpll 600000 KHz cpll 400000 KHz gpll 594000 KHz armclk 600000 KHz aclk_cpu 148500 KHz hclk_cpu 74250 KHz pclk_cpu 74250 KHz aclk_peri 148500 KHz hclk_peri 74250 KHz pclk_peri 74250 KHz Hit key to stop autoboot('CTRL+C'): 0 switch to partitions #0, OK mmc1 is current device Scanning mmc 1:1... reading /kingjim-dm250.dtb 23239 bytes read in 6 ms (3.7 MiB/s) Found EFI removable media binary efi/boot/bootarm.efi reading efi/boot/bootarm.efi 119564 bytes read in 16 ms (7.1 MiB/s) ## Starting EFI application at 62008000 ... FtlInit fffffffe Scanning disk [email protected]... Scanning disk [email protected]... Scanning disk [email protected]... Found 3 disks Adding bank: 0x60000000 - 0xa0000000 (size: 0x40000000) disks: sd0* sd1 sd2 sd3 >> OpenBSD/armv7 BOOTARM 1.23 boot> b sd0a:/bsd.rd cannot open sd0a:/etc/random.seed: No such file or directory booting sd0a:/bsd.rd: 4916868+1014156+16731272+608976 [2791939+360736+184+330515]=0x0 OpenBSD/armv7 booting ... arg0 0xc1c8514c arg1 0x0 arg2 0x9ac82000 Allocating page tables IRQ stack: p0x61cb4000 v0xc1cb4000 ABT stack: p0x61cb5000 v0xc1cb5000 UND stack: p0x61cb6000 v0xc1cb6000 SVC stack: p0x61cb7000 v0xc1cb7000 Creating L1 page table at 0x61c88000 Mapping kernel Constructing L2 page tables undefined page type 0x2 pa 0x60000000 va 0x60000000 pages 0x2000 attr 0x8 type 0x7 pa 0x62000000 va 0x60000000 pages 0x6000 attr 0x8 initarm: added 24576 pages at 0x62000000, physmem now 32768 type 0x4 pa 0x68000000 va 0x68000000 pages 0x7 attr 0x8 type 0x7 pa 0x68008000 va 0x60000000 pages 0x32c7a attr 0x8 initarm: added 103997 pages at 0x68008000, physmem now 136765 type 0x2 pa 0x9ac82000 va 0x9ac82000 pages 0x7 attr 0x8 type 0x7 pa 0x9ac89000 va 0x9ac89000 pages 0x4 attr 0x8 type 0x7 pa 0x9ac8d000 va 0x9ac8d000 pages 0x2 attr 0x8 type 0x7 pa 0x9ac8f000 va 0x9ac8f000 pages 0x1 attr 0x8 type 0x2 pa 0x9ac90000 va 0x9ac90000 pages 0x100 attr 0x8 type 0x2 pa 0x9ad90000 va 0x9ad90000 pages 0x1e attr 0x8 type 0x6 pa 0x9adae000 va 0x9adae000 pages 0x1 attr 0x8000000000000008 type 0x0 pa 0x9adaf000 va 0x9adaf000 pages 0x1 attr 0x8 type 0x0 pa 0x9adb0000 va 0x9adb0000 pages 0x1 attr 0x8 type 0x0 pa 0x9adb1000 va 0x9adb1000 pages 0x1 attr 0x8 type 0x0 pa 0x9adb2000 va 0x9adb2000 pages 0x1 attr 0x8 type 0x0 pa 0x9adb3000 va 0x9adb3000 pages 0x1 attr 0x8 type 0x0 pa 0x9adb4000 va 0x9adb4000 pages 0x1 attr 0x8 type 0x0 pa 0x9adb5000 va 0x9adb5000 pages 0x1 attr 0x8 type 0x0 pa 0x9adb6000 va 0x9adb6000 pages 0x1 attr 0x8 type 0x2 pa 0x9adb7000 va 0x9adb7000 pages 0x308c attr 0x8 type 0x5 pa 0x9de43000 va 0x9de43000 pages 0x1 attr 0x8000000000000008 type 0x2 pa 0x9de44000 va 0x9adb7000 pages 0x21bc attr 0x8 pmap [ using 3484148 bytes of bsd ELF symbol table ] Copyright (c) 1982, 1986, 1989, 1991, 1993 The Regents of the University of California. All rights reserved. Copyright (c) 1995-2025 OpenBSD. All rights reserved. https://www.OpenBSD.org OpenBSD 7.7 (obj.amd64.armv7) #113: Fri Apr 18 11:15:57 CDT 2025 [email protected]:/usr/src/sys/arch/armv7/compile/GENERIC/obj.amd64.armv7 real mem = 560189440 (534MB) avail mem = 520486912 (496MB) random: boothowto does not indicate good seed mainbus0 at root: KING JIM Pomera DM250 cortex0 at mainbus0 psci0 at mainbus0: PSCI 0.0 syscon0 at mainbus0: can't map registers syscon1 at mainbus0: "syscon" ampintc0 at mainbus0 nirq 160, ncpu 4: "interrupt-controller" syscon2 at mainbus0: "syscon" agtimer0 at mainbus0: 24000 kHz agtimer1 at mainbus0: 24000 kHz com0 at mainbus0: dw16550, 64 byte fifo com0: probed fifo depth: 0 bytes com1 at mainbus0: dw16550 com1: console com2 at mainbus0: dw16550 ehci0 at mainbus0 usb0 at ehci0: USB revision 2.0 uhub0 at usb0 configuration 1 interface 0 "Generic EHCI root hub" rev 2.00/1.00 addr 1 ohci0 at mainbus0: version 1.0 dwmmc0 at mainbus0: 18 MHz base clock sdmmc0 at dwmmc0: 4-bit, dma dwmmc1 at mainbus0: 25 MHz base clock sdmmc1 at dwmmc1: 8-bit, dma rkiic0 at mainbus0 iic0 at rkiic0 "rockchip,rk818" at iic0 addr 0x1c not configured rkiic1 at mainbus0 iic1 at rkiic1 pcxrtc0 at iic1 addr 0x51pcxrtc0: pcxrtc_reg_read: failed to read reg0 pcxrtc0: pcxrtc_reg_write: failed to write reg0 pcxrtc0: pcxrtc_reg_read: failed to read reg2 : battery ok rkiic2 at mainbus0 iic2 at rkiic2 rkiic3 at mainbus0 iic3 at rkiic3 usb1 at ohci0: USB revision 1.0 uhub1 at usb1 configuration 1 interface 0 "Generic OHCI root hub" rev 1.00/1.00 addr 1 scsibus0 at sdmmc0: 2 targets, initiator 0 sd0 at scsibus0 targ 1 lun 0: <Sandisk, SL32G, 0080> removable sd0: 30436MB, 512 bytes/sector, 62333952 sectors scsibus1 at sdmmc1: 2 targets, initiator 0 sd1 at scsibus1 targ 1 lun 0: <Toshiba, 008GB1, 0000> removable sd1: 7456MB, 512 bytes/sector, 15269888 sectors vscsi0 at root scsibus2 at vscsi0: 256 targets softraid0 at root scsibus3 at softraid0: 256 targets bootfile: sd0a:/bsd boot device: sd0 root on rd0a swap on rd0b dump on rd0b pcxrtc0: pcxrtc_clock_read: failed to read rtc WARNING: bad clock chip time WARNING: CHECK AND RESET THE DATE! Fatal kernel mode prefetch abort at 0x00000000 trapframe: 0xcd06ba70 IFSR=00000005, IFAR=00000000, spsr=80000113 r0 =00000000, r1 =00000007, r2 =c18a0868, r3 =60000113 r4 =00000007, r5 =c93ad000, r6 =c93ad000, r7 =cd06bb10 r8 =cd06a000, r9 =00000013, r10=c08a8988, r11=cd06bb08 r12=c18e5378, ssp=cd06bac0, slr=c0780344, pc =00000000 Stopped at 0 ddb> trace 0 rlv=0xc032fd30 rfp=0xcd06bb90 exception_exit rlv=0xc0343800 rfp=0xcd06bee0 sys_execve+0x2c8 [/usr/src/sys/kern/kern_exec.c:361] rlv=0xc04c4450 rfp=0xcd06bfa8 start_init+0x254 [/usr/src/sys/kern/init_main.c:716] rlv=0xc07976ac rfp=0xc1cb8f90 Bad frame pointer: 0xc1cb8f90 I'm still not sure why the memory limiting is needed, but apparently U-boot is not passing the proper memory segment information to the EFI bootloader for the kernel to know to avoid that address space. Since I was able to reduce the custom things needed in U-boot, I tried adapting my UART, GPIO, and timer changes to mainline U-boot to see if maybe the EFI code was better there. It boots now with UART output but the SDMMC and eMMC driver fails to setup either one of them: U-Boot 2025.01-00001-g4e6a9d7df66d-dirty (Apr 19 2025 - 22:17:13 -0500) Model: KING JIM Pomera DM250 DRAM: 1 GiB Core: 30 devices, 14 uclasses, devicetree: embed MMC: mmc@10214000: 1, mmc@1021c000: 0 Loading Environment from nowhere... OK In: serial@20064000 Out: serial@20064000 Err: serial@20064000 Hit any key to stop autoboot: 0 Card did not respond to voltage select! : -110 Cannot persist EFI variables without system partition Card did not respond to voltage select! : -110 No USB controllers found I can see the udelay calls work properly (which weren't in the Rockchip-specific U-boot tree until I made the RK3128-specific timer changes), and where it's failing to respond to voltage is past the initial setup which requires responses from the controllers so it seems like they are being powered up. 2025-04-21 (Evening) I guess I should have read the kernel panic better. Fatal kernel mode prefetch abort at 0x00000000 and pc =00000000 indicate that the kernel set the program counter to 0, which meant it was probably executing a function callback that was pointing to NULL. After dozens of printfs added, kernels recompiled, SD cards swapped, and reset pins grounded, I figured out that the kernel was panicking in data_abort_handler because curcpu()->ci_flush_bp was NULL and there was no check for that (because it shouldn't really happen). Why it was NULL was much more complicated. ci_flush_bp was never initialized because arm/arm/cpu.c was not attaching to cpu0, because the reg values for cpu0-cpu3 in the FDT were 0x000-0x003, but mainbus.c expects them to be 0xf00-0xf03. They are 0x000-0x003 even in the latest U-boot tree but 0xf00-0xf03 in Linux which I guess is now the authoritative source for device trees? This is why I dislike the ARM ecosystem… cpu0 at mainbus0 mpidr f00: ARM Cortex-A7 r0p5 cpu0: 32KB 32b/line 2-way L1 VIPT I-cache, 32KB 64b/line 4-way L1 D-cache cpu0: 256KB 64b/line 8-way L2 cache Anyway, now that cpu0 actually attaches and runs cpu_identify, it sets the CPU device's ci_flush_bp callback to cpu_flush_bp_noop, which does… nothing. So the kernel isn't panicking now, but instead it just locks up (actually powers off) when it should be starting userland. I'm getting there… 2025-04-22 Oh, right, we have no clock again, so the dwmmc driver's attempt to set the frequency does nothing but this isn't handled as an error. I'll have to add rockchip,rk3128-cru support to the rkclock driver, which does not look fun doing from scratch. This menial task of translating register definitions from PDFs and cross-referencing Linux driver code is usually where my willpower fades in these types of projects. 2025-04-29 A few weeks ago I bought a Geniatech XPI-3128 which is another board based on the Rockchip 3128, but with 4 USB ports, ethernet, and HDMI. I tried flashing a new U-Boot build to it and it promptly stopped booting. When I'd try powering it on with its recovery button pressed to boot into Maskrom mode, my laptop would just log messages like this: uhub3: device problem, disabling port 3 So it was as if it was trying to attach but kept failing. The device was basically bricked, so I e-mailed Geniatech's support address for help. A couple weeks later they finally gave me the information I needed, which was that I had to desolder the Wi-Fi board, remove the CPU heatsink, and then the eMMC clock line was reachable to be shorted to ground to avoid loading U-Boot and force Maskrom mode. That allowed me to flash and test different U-Boot builds again and finally get it booting on the XPI-3128. However, the more I worked on it, the more I realized trying to do anything with my older U-Boot tree was futile. The device tree (DTB) that shipped on the DM250 (and the XPI-3128) is very old, and is configured for old U-Boot and Linux drivers. Things like the names of compatible strings and the way peripherals are described are targeting Rockchip-specific drivers in their old Linux tree, rather than what's in the current Linux kernel. Trying to write OpenBSD drivers for the way this old DTB is setup would be a bad idea, so I really needed to get RK3128 support working on the latest U-Boot and targeting the official XPI-3128 device tree with all of its compatible strings. While reading various RK3128 code, I came across Linux and U-Boot patches from Alex Bee, which led me to find their U-Boot tree with RK3128 support, but done right to eventually be upstreamed. With this tree I was finally able to boot a modern U-boot (2025.04) on the XPI-3128 (though still needing my timer init code), which allowed me to boot OpenBSD all the way to userland on a USB stick: U-Boot 2025.04-rc1-00167-g04767ba5b99f-dirty (Apr 29 2025 - 21:55:41 -0500) Model: Geniatech XPI-3128 DRAM: 1 GiB Cannot find regulator pwm init_voltage Cannot find regulator pwm init_voltage Core: 164 devices, 21 uclasses, devicetree: embed MMC: mmc@10214000: 1, mmc@1021c000: 0 Loading Environment from MMC... Reading from MMC(0)... *** Warning - bad CRC, using default environment In: serial@20064000 Out: serial@20064000 Err: serial@20064000 Model: Geniatech XPI-3128 Net: No ethernet found. Hit any key to stop autoboot: 0 Scanning for bootflows in all bootdevs Seq Method State Uclass Part Name Filename --- ----------- ------ -------- ---- ------------------------ ---------------- Scanning global bootmeth 'efi_mgr': Card did not respond to voltage select! : -110 Cannot persist EFI variables without system partition 0 efi_mgr ready (none) 0 <NULL> ** Booting bootflow '<NULL>' with efi_mgr Loading Boot0000 'mmc 0' failed EFI boot manager: Cannot load any image Boot failed (err=-14) Scanning bootdev '[email protected]': Card did not respond to voltage select! : -110 Scanning bootdev '[email protected]': Unknown uclass 'nvme' in label Unknown uclass 'scsi' in label Bus usb@10180000: USB DWC2 Bus usb@101c0000: USB EHCI 1.00 scanning bus usb@10180000 for devices... 1 USB Device(s) found scanning bus usb@101c0000 for devices... 3 USB Device(s) found Scanning bootdev 'usb_mass_storage.lun0.bootdev': 1 efi ready usb_mass_ 1 usb_mass_storage.lun0.boo /EFI/BOOT/BOOTARM.EFI ** Booting bootflow 'usb_mass_storage.lun0.bootdev.part_1' with efi Booting /\EFI\BOOT\BOOTARM.EFI disks: sd0* sd1 >> OpenBSD/armv7 BOOTARM 1.23 boot> booting sd0a:/bsd: 4915064+1013912+140528+607852 [289299+107+346480+308631]=0x0 OpenBSD/armv7 booting ... arg0 0xc0a456f8 arg1 0x0 arg2 0x9cdff000 Allocating page tables IRQ stack: p0x60a74000 v0xc0a74000 ABT stack: p0x60a75000 v0xc0a75000 UND stack: p0x60a76000 v0xc0a76000 SVC stack: p0x60a77000 v0xc0a77000 Creating L1 page table at 0x60a48000 Mapping kernel Constructing L2 page tables undefined page type 0x2 pa 0x60000000 va 0x60000000 pages 0x2000 attr 0x8 type 0x7 pa 0x62000000 va 0x62000000 pages 0x3adff attr 0x8 type 0x2 pa 0x9cdff000 va 0x9cdff000 pages 0x9 attr 0x8 type 0x7 pa 0x9ce08000 va 0x9ce08000 pages 0x1 attr 0x8 type 0x2 pa 0x9ce09000 va 0x9ce09000 pages 0x100 attr 0x8 type 0x1 pa 0x9cf09000 va 0x9cf09000 pages 0x1e attr 0x8 type 0x4 pa 0x9cf27000 va 0x9cf27000 pages 0x3 attr 0x8 type 0x9 pa 0x9cf2a000 va 0x9cf2a000 pages 0xb attr 0x8 type 0x4 pa 0x9cf35000 va 0x9cf35000 pages 0xb attr 0x8 type 0x6 pa 0x9cf40000 va 0x9cf40000 pages 0x1 attr 0x8000000000000008 type 0x4 pa 0x9cf41000 va 0x9cf41000 pages 0x1 attr 0x8 type 0x6 pa 0x9cf42000 va 0x9cf42000 pages 0x22 attr 0x8000000000000008 type 0x4 pa 0x9cf64000 va 0x9cf64000 pages 0x5 attr 0x8 type 0x3 pa 0x9cf69000 va 0x9cf69000 pages 0x1009 attr 0x8 type 0x6 pa 0x9df72000 va 0x9df72000 pages 0x1 attr 0x8000000000000008 type 0x3 pa 0x9df73000 va 0x9df73000 pages 0x1fff attr 0x8 type 0x5 pa 0x9ff72000 va 0x9ff72000 pages 0x2 attr 0x8000000000000008 type 0x3 pa 0x9ff74000 va 0x9ff74000 pages 0x8c attr 0x8 pmap [ using 945052 bytes of bsd ELF symbol table ] Copyright (c) 1982, 1986, 1989, 1991, 1993 The Regents of the University of California. All rights reserved. Copyright (c) 1995-2025 OpenBSD. All rights reserved. https://www.OpenBSD.org OpenBSD 7.7-current (GENERIC) #1: Tue Apr 29 20:43:21 MDT 2025 jcs@rk3128:/usr/src/sys/arch/armv7/compile/GENERIC real mem = 1021308928 (973MB) avail mem = 992374784 (946MB) random: good seed from bootblocks mainbus0 at root: Geniatech XPI-3128 cpu0 at mainbus0 mpidr f00: ARM Cortex-A7 r0p5 cpu0: 32KB 32b/line 2-way L1 VIPT I-cache, 32KB 64b/line 4-way L1 D-cache cpu0: 256KB 64b/line 8-way L2 cache cortex0 at mainbus0 syscon0 at mainbus0: "syscon" "power-controller" at syscon0 not configured syscon1 at mainbus0: "qos" syscon2 at mainbus0: "qos" syscon3 at mainbus0: "qos" syscon4 at mainbus0: "qos" syscon5 at mainbus0: "qos" syscon6 at mainbus0: "qos" syscon7 at mainbus0: "qos" ampintc0 at mainbus0 nirq 160, ncpu 4: "interrupt-controller" rkclock0 at mainbus0 syscon8 at mainbus0: "syscon" "usb2phy" at syscon8 not configured syscon9 at mainbus0: can't map registers agtimer0 at mainbus0: 24000 kHz ehci0 at mainbus0rk3128_enable: 0x000001d9 usb0 at ehci0: USB revision 2.0 uhub0 at usb0 configuration 1 interface 0 "Generic EHCI root hub" rev 2.00/1.00 addr 1 dwmmc0 at mainbus0rk3128_set_frequency: 68 100000000 rkclock_set_frequency(rkclock0, 68, 100000000) parent : 12 MHz base clock sdmmc0 at dwmmc0: 4-bit, sd high-speed, dma dwmmc1 at mainbus0rk3128_set_frequency: 71 100000000 rkclock_set_frequency: clk div mask 16128 rk3128_get_frequency: RK3128_XIN24M rk3128_get_frequency: RK3128_XIN24M rk3128_get_frequency: RK3128_PLL_CPLL rk3128_get_pll: 0x20 = 523462184 rk3128_get_frequency: RK3128_PLL_CPLL rk3128_get_pll: 0x20 = 523462184 rk3128_get_frequency: RK3128_XIN24M rk3128_get_frequency: RK3128_XIN24M rk3128_get_frequency: RK3128_PLL_CPLL rk3128_get_pll: 0x20 = 523462184 rk3128_get_frequency: unhandled 71 rk3128_get_frequency: RK3128_PLL_CPLL rk3128_get_pll: 0x20 = 523462184 : 43 MHz base clock sdmmc1 at dwmmc1: 8-bit, mmc high-speed, dma com0 at mainbus0: dw16550 com0: console rkiic0 at mainbus0 rk3128_get_frequency: RK3128_CLK_I2C rk3128_get_frequency: RK3128_PLL_CPLL rk3128_get_pll: 0x20 = 523462184 iic0 at rkiic0 dwge0 at mainbus0rk3128_set_frequency: 124 50000000 rkclock_set_frequency(rkclock0, 124, 50000000) rk3128_enable: 0x0000016f : rev 0x35rk3128_get_frequency: unhandled 126 rkclock_get_frequency(rkclock0, 126) , address 76:e3:5a:fa:14:d9 rk3128_set_frequency: 126 50000000 rkclock_set_frequency(rkclock0, 126, 50000000) dwge0: no PHY found! scsibus0 at sdmmc1: 2 targets, initiator 0 sd0 at scsibus0 targ 1 lun 0: <Samsung, 8GTF4R, 0000> removable sd0: 7456MB, 512 bytes/sector, 15269888 sectors uhub1 at uhub0 port 1 configuration 1 interface 0 "Genesys Logic USB2.0 Hub" rev 2.00/60.90 addr 2 umass0 at uhub1 port 1 configuration 1 interface 0 "USB SanDisk 3.2Gen1" rev 2.10/1.00 addr 3 umass0: using SCSI over Bulk-Only scsibus1 at umass0: 2 targets, initiator 0 sd1 at scsibus1 targ 1 lun 0: <USB, SanDisk 3.2Gen1, 1.00> removable serial.078155ab8107712cf658 sd1: 942480MB, 512 bytes/sector, 1930199040 sectors vscsi0 at root scsibus2 at vscsi0: 256 targets softraid0 at root scsibus3 at softraid0: 256 targets bootfile: sd0a:/bsd boot device: sd0 root on sd1a (f2059a1fe6a57770.a) swap on sd1b dump on sd1b WARNING: CHECK AND RESET THE DATE! rk3128_get_frequency: RK3128_ARMCLK rk3128_get_frequency: RK3128_PLL_APLL rk3128_get_pll: 0x0 = 211673469 rk3128_set_frequency: RK3128_ARMCLK 52918367 rk3128_set_frequency: RK3128_PLL_APLL 52918367 rk3128_set_pll: freq 52918367 rk3128_set_pll: 52918367 Hz cpu0: clock not implemented Automatic boot in progress: starting file system checks. /dev/sd1a (f2059a1fe6a57770.a): file system is clean; not checking pf enabled starting network starting early daemons: syslogd pflogd ntpd. starting RPC daemons:. savecore: no core dump checking quotas: done. clearing /tmp kern.securelevel: 0 -> 1 creating runtime link editor directory cache. preserving editor files. starting network daemons: sshd. starting local daemons: cron. Tue Apr 29 20:52:11 MDT 2025 OpenBSD/armv7 (rk3128) (console) login: That kernel was actually compiled on the XPI-3128 and then booting on it (ignore all the clock debugging output). I need to fix the clock setting for MMC and ethernet and then sdmmc and dwge devices will work. USB is working fine out of the box since it's actually booting and running off of a USB stick, but once MMC works, I can install and boot from the onboard eMMC. I'd like to write a driver for the Artasie AM1805 I2C RTC that is present on the XPI-3128 which will give it a working realtime clock. Once all of those things are working I'll hopefully commit all of this RK3128 support to OpenBSD, which will then allow me to go back to working on the DM250 and write drivers for the keyboard and LCD. 2025-07-09 You may have assumed that I gave up on this project but the sad part is that I've been working on it almost every day and getting pretty much nowhere. I have the basics working like GPIO pin control (rkpinctrl), clocks (rkclock), and regulators (rkpmic) but anything more advanced like the screen or SDIO Wi-Fi or keyboard interrupts aren't working. The main problem so far is that the DTB embedded on the eMMC is ancient and uses a lot of proprietary Rockchip properties that are specifically for Rockchip's Linux 3.10 tree, on top of which has hard-coded hacks and RK312x-specific tweaks everywhere. Also, the U-Boot on the device also has hard-coded hacks and things specific to the DM250. To make this work on OpenBSD, the DTB has to be modernized which is largely helped by this RK3128 file but there are a lot of DM250-specific components that need to be added describing the keyboard, SDIO, battery, LCD screen information, LVDS controller information, etc. I currently have two DM250s taken apart on my desk with cables hooked up to their UART pins, one running OpenBSD with current U-Boot, and one running Debian Linux 11 with the DM250 Linux 3.10.0 tree booting from the DM250-specific U-Boot. This allows me to add in some debugging printks on the Linux kernel, compile it, dd it to /dev/mmcblk0p14, reboot, and see the output. Then I can add things on the OpenBSD DM250 and reboot. But often this requires changing a pin configuration or adding something new to the DTB which then has to be written to the eMMC on the OpenBSD DM250 over a USB cable. This whole process has been going very slowly and just when I think I figured something out, I broke something else. I can turn the LCD backlight on with pwmbl and adjust its brightness, but I still can't get anything to show up on the screen. I wrote rklvds and rklcdc drivers for OpenBSD based on the Rockchip-specific code in the DM250 U-Boot tree, only to discover that the LCDC does the same as what is now called the VOP and should have used a different compatible string in the DTB. The Rockchip VOP already has an OpenBSD driver that hooks it up to wscons and rkdrm, but it needed RK3128 (RK3126 actually) support which I added. But still nothing will show on the screen. Current U-Boot even has Rockchip VOP and LVDS drivers so it should work out of the box, right? But it does the same thing as OpenBSD just enabling the backlight and unable to draw anything on the screen. The keyboard kind of works with the I2C TC3589x driver I wrote, but I can't get interrupts working. The SD card slot works, but I don't get interrupts for card-detect events even though I'm specifying the same cd-gpios information as the DTB file that shipped with the DM250. Anyway, this is all rambling and probably not very interesting but I'm getting tired of this project after a few months. If I could just get the screen and the keyboard interrupts working, I could work directly on the DM250 in OpenBSD instead of it being cracked open in pieces on my desk with wires hanging out of it working over a serial connection. 2026-03-23 tl;dr: OpenBSD with my kernel tree and U-boot with updated device-tree bindings is now working reliably on the DM250 including graphical boot early in U-boot with keyboard support, X11, interrupt-driven keyboard, battery charging and sensors, Wi-Fi, SD card eject/insertion, CPU speed adjustments, red and green power/charging LEDs, and probably other things I'm forgetting. OpenBSD-current on the Pomera DM250X I just noticed this article is now more than a year old. After many months of working on other projects, I had enough desk space to get back to the DM250. I booted my US model that had OpenBSD installed on it and through its serial console I could see it booting to the kernel copyright line and then locking up or totally powering off. I had no other usable kernels on the device so it took a while to get it back to a working state which involved cross-compiling an armv7 kernel on my ThinkPad. Once I had a new kernel booting, I was encountering the same problems that I remembered encountering half a year ago, such as it locking up when all 1GB of RAM was being initialized in OpenBSD, or the SD card not being able to be properly read in U-Boot. It took me a while to figure out (or remember) many of the issues were related to power caused by the battery not charged enough (or being completely disconnected as it sometimes was while moving everything around). I think when the system is running with all of its power regulators enabled, just having its USB port connected to a 5V power source doesn't supply enough amperage to fully power everything and it relies on a working battery for help or it crashes. That led me to figure out why the battery wasn't getting charged while idling in U-Boot or OpenBSD. After more digging through the vendor U-boot tree and using a USB-C power meter, I found that the RK818 PMIC needs to be told to enable USB charging at a higher rate or else it will just trickle charge the battery at a rate that is too low to keep up with the idle power consumption of the device. This would cause the battery to eventually drain too low to be able to boot. Luckily I had a few other DM250s so I swapped the dead battery into a device with the original U-boot firmware where it would immediately charge it at a high rate before continuing with boot. Once I figured out how to enable higher-rate charging on the RK818, I wrote an rkcharger driver that hangs off the rkpmic device, and also a driver for simple-battery devices which asks the parent device (rkcharger) to read charging and battery info and exposes it as hw.sensors values. Once there was reliable power, the random crashes and power-offs stopped and I could use the full 1GB of RAM. I also updated the simple-battery node in the DM250 device tree. I also discovered that the DM250 and DM250US aren't as identical as I thought, at least in terms of charging. The DM250 uses the RK818 to do it directly while the DM250US introduces a TI BQ25620 charging chip. This caused me a lot of frustration trying to figure out why my driver wasn't working on the DM250 (because the device wasn't even there). DM250 (JP) DM250 (US), adding a BQ25620 I fixed a bunch of other little issues, many of which stemmed from incorrect things in the unofficial DM250 device-tree that is still being worked on. Once the reset and power settings were corrected for the Wi-Fi bwfm0 at sdmmc1 just magically worked without any kernel changes. It uses brcmfmac43430-sdio.bin for firmware and it can use the brcmfmac43430-sdio.rockchip,pomera-dm250.txt NVRAM settings file from the original Linux installation on the device. I brought over the U-boot LVDS and VOP drivers from Rockchip's U-boot tree, which enabled a graphical framebuffer very early in the power-on process. I also ported my Toshiba TC3589X keyboard driver from OpenBSD so I could type on the keyboard and over the serial device at the same time. I enabled U-boot's boot logo support to get a neat OpenBSD logo (read from a .bmp file in eMMC's EFI partition) during boot before clearing the screen to show OpenBSD's EFI bootloader. Since the keyboard works in U-boot now, this also enabled the keyboard to work in OpenBSD's bootloader (at least as far as telling it to boot a different device or kernel). Once that worked, OpenBSD technically didn't need any video driver since it could use simplefb that was setup by U-boot. This shows continuous boot output from the EFI bootloader all the way to the console login, which is nice. If I enable the video drivers (rklvds, rkvop, rkdrm) to (re-)initialize the video in OpenBSD, it boots about halfway through the kernel sequence and then blacks out for a second or two as it has to wait for hardware to settle before drawing through the new output path. With video and the keyboard working, I finally reached that point where I could do development directly on the device which feels a lot different than remotely poking at something through a serial console. I've done a lot of little quality-of-life changes like implementing a US keyboard layout for the non-US model (available with wsconsctl keyboard.encoding=us), adding gpio(4) support to rkgpio so I can poke individual GPIO pins from userland with gpioctl. This allows me to turn on and off the red (gpioctl gpio1 8) and green (gpioctl gpio1 12) LEDs on the side of the device depending on whether the battery is about to die (red) or is charging (green). One thing that is odd about the DM250 is that the left Alt key and the right Shift key are directly wired up to their own GPIO pins, not going through the TC35894 like every other key. Presumably this is why the recovery sequence that the vendor's U-boot tree looks for is those two keys plus power, so their U-boot didn't have to implement a TC3589X driver. Anyway, since those two keys are not on the keyboard, I thought about how to make them work in OpenBSD without a specific hack for the DM250 or something in my tcmfd driver that had to reach into GPIO land. Since the existing OpenBSD gpiokeys driver works on armv7 and sees the entries in the device tree: gpiokeys0 at mainbus0: "Power Button", "Lid Switch", "Right Shift", "Left Alt" I added a (only-slightly-hackish) hack to it to inject unknown GPIO keys into the console wskbd device's input stream, so anything listening for keyboard input will see left Alt and right Shift as though it came from the same tc35894 device. This means Control+Alt+F# keys work as expected to change virtual terminals, for example. I still have a laundry list of things I'd like to keep working on like improving the keyboard driver, implementing some degree of suspend/resume, and supporting the external DMA engine for the MMC controller to speed up eMMC access. Our arm port also doesn't enable multiple processors but some degree of support seems there from when it was imported from NetBSD. My list of commits is getting quite long so I need to try to upstream as much of this as possible. My last attempts at committing just basic RK3128 support in various drivers were thwarted, so I'm still just hammering out stuff in my own trees for now. If you have a DM250 (non-US for now) and want to try OpenBSD on it, let me know and I can send you installation images and instructions. Note: I am frequently rebasing and squashing commits in my trees as I improve things, so the commit IDs in the trees linked here may vanish or become obsolete. OpenBSD 7.9-beta (GENERIC) #134: Mon Mar 23 16:10:06 CDT 2026 jcs@dm250x:/usr/src/sys/arch/armv7/compile/GENERIC real mem = 1018015744 (970MB) avail mem = 988418048 (942MB) random: good seed from bootblocks mainbus0 at root: Rockchip RK3128 Pomera DM250 cpu0 at mainbus0 mpidr f00: ARM Cortex-A7 r0p5 cpu0: 32KB 32b/line 2-way L1 VIPT I-cache, 32KB 64b/line 4-way L1 D-cache cpu0: 256KB 64b/line 8-way L2 cache cortex0 at mainbus0 syscon0 at mainbus0: "syscon" "power-controller" at syscon0 not configured syscon1 at mainbus0: "qos" syscon2 at mainbus0: "qos" syscon3 at mainbus0: "qos" syscon4 at mainbus0: "qos" syscon5 at mainbus0: "qos" syscon6 at mainbus0: "qos" syscon7 at mainbus0: "qos" ampintc0 at mainbus0 nirq 160, ncpu 4: "interrupt-controller" rkclock0 at mainbus0 syscon8 at mainbus0: "syscon" rkusbphy0 at syscon8: phy 0 rklvds0 at syscon8: LVDS 24-bit JEIDA rkpinctrl0 at mainbus0: "pinctrl" rkgpio0 at rkpinctrl0 gpio0 at rkgpio0: 32 pins rkgpio1 at rkpinctrl0 gpio1 at rkgpio1: 32 pins rkgpio2 at rkpinctrl0 gpio2 at rkgpio2: 32 pins rkgpio3 at rkpinctrl0 gpio3 at rkgpio3: 32 pins rkdrm0 at mainbus0 drm0 at rkdrm0 agtimer0 at mainbus0: 24000 kHz rkvop0 at mainbus0: RK3126 VOP dwctwo0 at mainbus0 dwmmc0 at mainbus0: 49 MHz base clock sdmmc0 at dwmmc0: 4-bit, sd high-speed, mmc high-speed dwmmc1 at mainbus0: 49 MHz base clock sdmmc1 at dwmmc1: 4-bit, sd high-speed dwmmc2 at mainbus0: 49 MHz base clock sdmmc2 at dwmmc2: 8-bit, mmc high-speed rklvdsphy0 at mainbus0 dwdog0 at mainbus0 rkpwm0 at mainbus0 com0 at mainbus0: dw16550, 64 byte fifo bcmbt0 at com0 com1 at mainbus0: dw16550 rkiic0 at mainbus0 iic0 at rkiic0 tcmfd0 at iic0 addr 0x45 wskbd0 at tcmfd0: console keyboard rkpmic0 at iic0 addr 0x1c: RK818 rkcharger0 at rkpmic0: 4.2V 5800mAh battery simplebat0 at rkcharger0 gpioleds0 at mainbus0: "pomera:green:power" gpiokeys0 at mainbus0: "Power Button", "Lid Switch", "Right Shift", "Left Alt" pwmbl0 at mainbus0 simplepanel0 at mainbus0: 1024x600 rkdrm0: 1024x600, 32bpp wsdisplay0 at rkdrm0 mux 1: console (std, vt100 emulation), using wskbd0 wsdisplay0: screen 1-5 added (std, vt100 emulation) usb0 at dwctwo0: USB revision 2.0 uhub0 at usb0 configuration 1 interface 0 "DWC2 DWC2 root hub" rev 2.00/1.00 addr 1 scsibus0 at sdmmc0: 2 targets, initiator 0 sd0 at scsibus0 targ 1 lun 0: <Sandisk, SD32G, 0085> removable sd0: 30436MB, 512 bytes/sector, 62333952 sectors scsibus1 at sdmmc2: 2 targets, initiator 0 sd1 at scsibus1 targ 1 lun 0: <Toshiba, 008GB0, 0000> removable sd1: 7456MB, 512 bytes/sector, 15269888 sectors bwfm0 at sdmmc1 function 1 manufacturer 0x02d0, product 0xa9a6 at sdmmc1 function 2 not configured vscsi0 at root scsibus2 at vscsi0: 256 targets softraid0 at root scsibus3 at softraid0: 256 targets bootfile: sd0a:/bsd boot device: sd0 root on sd1a (717a8af462695010.a) swap on sd1b dump on sd1b bcmbt0: address 70:4a:0e:df:xx:xx bwfm0: address 70:4a:0e:df:xx:xx

14th Mar 2025 1 votes
Adding Custom Sleep Screen Images to the Kindle Scribe

Last year I upgraded my Kindle Paperwhite to a Kindle Scribe to be able to write notes and draw diagrams while programming to help visualize things. One thing that bothered me about the Scribe was that its sleep screen images were pretty boring and because I'm now often reading PDFs or writing in a notebook, I couldn't benefit from the Kindle OS's new functionality that uses the cover of the book being read as the sleep screen image (which previously required a jailbreak and custom software). Since the process for installing custom sleep screen images is rather cumbersome and the information is scattered across old forum posts, I thought I'd document how I did it to possibly make it easier for someone else. Table of Contents Top Jailbreaking Disabling OTA Updates Installing KOReader Installing Custom Images Jailbreaking On my Kindle Paperwhite, I long-ago used a jailbreak to install a bunch of software packages that enabled custom sleep screen images. For the newer Scribe, this wasn't available until recently with the "WinterBreak" jailbreak, and the process to use custom images is now easier. To install WinterBreak on the Scribe, one needs a computer with an MTP application since the Scribe now uses MTP like an Android phone rather than attaching as a regular USB disk drive. On my Mac, I used OpenMTP. Follow the instructions at the WinterBreak page, which at the time of writing (and worked on my 1st generation Scribe running firmware 5.17.3) were: Download the current WinterBreak release Enable Airplane mode on the Kindle Connect the Scribe to the computer with a USB-C cable Extract the WinterBreak archive and copy all of the files (except .git*, README.md, and instructions.txt) to the Kindle's home directory, including the .active_content_sandbox directory which may not initially be visible in the MTP program Tap the "Disconnect" button on the Kindle and unplug it Swipe down from the top of the Kindle screen to open the settings drawer and tap "All Settings", then "Device options", then "Restart" Once restarted, tap the shopping cart icon on the Kindle to open the Kindle Store; when prompted to turn Airplane mode off, tap "Yes" Instead of the usual Kindle Store page, there will be a small icon labeled "Mesquito"; tap on it A bunch of debugging messages will print on the screen and it will say "Please install hotfix now." Plug the Kindle into the computer again with the USB-C cable Download the Hotfix .bin file and copy it to the Kindle's home directory with the MTP program Tap the "Disconnect" button on the Kindle and unplug it Swipe down from the top of the Kindle screen again and tap "All Settings", then tap the 3-dot menu button at the top and tap "Update your Kindle"; tap "Update" when prompted Once the Kindle restarts, there will be a new item in the "Your recent items" list named "KUAL" Disabling OTA Updates To prevent a future automatic update from disabling your jailbreak and possibly removing your custom software, you can install the "renameotabin" extension using the instructions shown there. Installing KOReader Once jailbroken, the KOReader application can be installed, which offers a different interface for reading books. I don't particularly like it (especially since it's a bit cumbersome to launch each time) but it's needed for the utilities included in it. Find the latest KOReader nightly release directory and download the kindlehf zip file in that directory Connect the Scribe to the computer with a USB-C cable and open the MTP program Extract the KOReader zip file and copy its koreader and extensions directories to the home directory on the Scribe (there will probably already be an extensions directory, so the extensions/koreader directory from the zip file needs to go in that extensions directory) Tap the "Disconnect" button on the Kindle and unplug it From the Kindle's home screen, tap "KUAL"; there should now be a "KOReader" entry in the list Installing Custom Images Sleep screen images for the Kindle Scribe (at least my 1st generation device) must be 8-bit grayscale PNG images, with a resolution of 1860x2480. I used GIMP to prepare some photos by scaling/cropping them to 1860x2480, converting to grayscale, and then exporting as a PNG. In GIMP's "Export Image as PNG" dialog, the "automatic.pixelformat" option should be changed to "8bpc GRAY". From what I've read, having images in a different format can cause crashes, so it's best to just force this format. When exporting your images, name them bg_ss00.png, bg_ss01.png, bg_ss02.png, etc. To install them: Connect the Scribe to the computer with a USB-C cable and open the MTP program Copy your bg_ss*.png images to the Kindle's home directory Tap the "Disconnect" button on the Kindle and unplug it From the Kindle's home screen, tap "KUAL" In KUAL, tap "KOReader", then the first "Start KOReader" option; the screen will go back to the Library and after a few seconds, KOReader will start Tap the "KOReader" header at the top to open its menu Tap the tools icon, then tap the ">" button to go to Page 2 Tap "More tools" Tap "Terminal emulator" and then "Open terminal session" You are now at a root shell, standing in an open field west of a white house, with a boarded front door; there is a small mailbox here You will need to delete the factory sleep screen images and then copy the bg_ss*.png images you transferred over to that directory. If you want to keep the factory images, don't delete anything and start your bg_ss*.png numbering at 07 since there are 7 factory sleep screen images (which start at 00). To modify the system sleep images, you'll need to remount / as read-write since it's read-only by default. In the Terminal emulator screen at the root # prompt, enter: mount -o remount,rw / If you want to delete the factory images, enter: rm /usr/share/blanket/screensaver/*.png To copy your custom images to the system directory, enter: cp /mnt/us/bg_ss* /usr/share/blanket/screensaver/ Once finished, tap the "X" button above the keyboard to close the terminal emulator. Once back at KOReader, tap the 3-line button at the top right and tap "Exit", then "Exit" again. You'll be returned to the Kindle home screen. Swipe down from the top of the Kindle screen to open the settings drawer and tap "All Settings", then "Screen and brightness". Make sure the "Show covers on lock screen" option is disabled. Tap "Device options", then "Restart" Once restarted, press the Kindle's sleep button and, if all went well, you should see a randomly-selected image from the custom ones you uploaded. A phoropter

24th Feb 2025 1 votes

More in technology

A Dick Smith VZ200 without the Dick Smith (but with a serial port)

Australians! They walk among us! Do not be deceived by those charming faces! precious bodily fluids, we must remove the scurrilous larrikin influence of Australia upon our American home computers, and I know exactly where to start! Why allow Dick Smith's name (even though he'd already sold Dick Smith Electronics' controlling interest to Woolies by then but stop ruining my intro) to corrupt this, um, rubber-keyed diminutive beige home computer when we can return it to its prior, pristine, plasticky state as it once rolled out from a glorious Asian factory? Then, to avoid wrecking it with a botched logic board repair, we'll bolt on a USB serial port using the very latest peripherals from down under, write cycle-counted Z80 assembly language to blast data to it at 57.6kbps, and hack a few games. Because only this will make the VZ200 great again! TI vs. Everybody"), was a seismic event in computing. Hong Kong entrepreneurs Allan Wong and Stephen Leung were two of many to realize that the microchip would trigger a revolution in consumer electronics, and over several years accumulated sufficient funding to establish Video Technology Ltd in 1976. Their first factory operated from the Freder Centre in Ma Tau Kok, a semi-industrial area on the west side of Kowloon Bay. Initially VTech, as it became known, concentrated on video games, primarily as an OEM for the more profitable North American and European markets. Their first products were Pong clones, released in the United Kingdom as the Grandstand Adman T.V. Game 2000 (black and white) and Adman T.V. Game 3000 (colour) in 1977, both based on the Texas Instruments TMS1965N, TI's clone of the well-known General Instrument AY-3-8500 Pong-on-a-chip (compare with the rather more complex MOS 7601). Grandstand was a brand name of Adam Imports, then a substantial toy and games importer to the UK and for a period of time New Zealand, and had other Asian contacts, notably Tomy. If the picture on VTech's history page is to be believed, and I point out there are some verifiable inaccuracies on that page, VTech continued to produce other consoles for Grandstand/Adam like the (deep breath) Grandstand Adman Colour TV Game 3600 Mk III, which used a regular AY-3-8500 and was produced until 1979. For the portable market, VTech produced a number of handheld LED, VFD, LCD games, sold under various brands and through retailers such as Radio Shack. VTech was hardly the only such Hong Kong tech company, of course; across the Bay in Kwun Tong was EACA, established in 1975 by Guangzhou escapee Eric Chung. EACA also produced consumer products such as radios and its own video games, notably the 1978 Colour TV Game, also based on the TMS1965N and variously sold under other brands such as the Sonesta Hide-Away TV Game. Creative Computing in August), but in fact ripped off from and largely compatible with the TRS-80, and advertised as such (well, the compatible part, anyway). While there were some internal differences and significant changes to the keyboard layout, EACA mostly copied the TRS-80 system ROMs for production with only minor changes, shipping it with a licensed version of Microsoft Level II BASIC. At Summer CES in Chicago it directly competed with the APF Imagination Machine and the Texas Instruments 99/4 (the original), and indirectly with the Atari 8-bits which earlier debuted at the Winter show. While it lacked the $599 model's monitor (a TV set or Tandy monitor was required), it had a better keyboard mechanism and a built-in cassette recorder, and was variously announced between $500 and $600 with 16K of RAM [$2200-$2760]. Electronics Australia, John Kennewell's National Semiconductor SC/MP-based MINI-SCAMP, running the CPU at roughly 500kHz (based on a typical 2μs cycle time) and 256 bytes of RAM expandable to 1K (64K addressable). DSE advertised the machine as "33% of the cost of the EDUC-8," an earlier Electronics Australia bit-serial TTL hobbyist system inspired by the DEC PDP-8 and designed by Jamieson Rowe — remember that name — who became an enthusiastic proponent of the new machine. Over in Silicon Valley, Palo Alto arcade game builder Exidy had developed their own Z80-based computer in 1978, the Exidy Sorcerer, a premium system featuring programmable character graphics, a faster 2.1MHz CPU and a built-in internal S-100 bus. Exidy saw the export market as a growth opportunity and aggressively inked deals with multiple foreign distributors, including DSE, who were taking ready advantage of the Whitlam government's 1973 import tariff reduction to bring more finished goods to DSE stores. The Sorcerer arguably found greater success in Europe than it ever did in the United States, particularly in the Netherlands where the licensed Compudata Sorcerer became the default government-supported educational system, and certainly in Australia due to Dick Smith's dogged promotion. Still, even in the US it was considered relatively expensive at $895 [$4580], and with tariff and import costs tacked on it didn't price itself well to the Aussie working class nerd. Conversely, the EACA Video Genie had meanwhile achieved some popularity of its own within Europe, notably West Germany, in no small part due to its lower cost. That alone made it a logical system to transition to, and better still, EACA had absolutely no objection to DSE outright rebadging it as a Dick Smith unit. Micro-80 disliked the altered keyboard (no CLEAR and TAB, no left and right arrows, up and down replaced by ESCAPE and CONTROL), complained about its changes to the character set and video circuitry, found the built-in cassette deck hideous, and noted the lack of board sockets and the missing-at-launch S-100 expansion box, but approved of the "brilliant" and "attractive" appearance, its overall functionality, and most of all its purchase price. "Even if you buy a decent tape deck from your local Big W and put it in the System 80," Hartley concluded, "you end up at least $150.00 [US$130 spot, US$520 in 2026 dollars] ahead — and that pays for your next 16K of RAM chips ... the machine has an identical computing capacity to the TRS-80, for a lot less dollars." At this point it was inevitable someone would poke the Tandy bear, and that someone was Recortec (they're still around), established in Sunnyvale, California in 1969 to specialize in magnetic tape recording technology. In 1980 Recortec, also attempting to expand their product base, opened Personal Micro Computers, Inc. (PMC) as a new venture in Mountain View, initially entering negotiations with Exidy to buy out the Sorcerer — until, through EACA's American subsidiary, they became aware of the Video Genie. To PMC/Recortec, the Genie was a remarkable opportunity, a less expensive TRS-80 compatible system already selling in Europe and entering the Australian market, and PMC now had the chance to corner it for American distribution. The company immediately backed out of the deal with Exidy, bought up Genie distribution rights in August for the entire Western Hemisphere, rebranded it as the PMC-80 and launched it in 1981 with 16K of RAM for $675 [$2330]. InfoWorld was more complimentary than Micro-80 had been, noting PMC's planned Fastload high-speed cassette scheme, a 50-40 adapter to connect Model I peripherals directly and a true lowercase conversion kit. "Tandy," said columnist Tracy Deliman, "is apparently curious now" — and in particular their attorneys, who promptly filed suit in federal court against both PMC and EACA of America, claiming, among other allegations, that the name PMC-80 infringed their trademark and that EACA and by extension PMC had committed copyright infringement as well by substantially copying the system ROMs. (Tandy didn't dispute the BASIC ROM, as that was licensed from Microsoft, but EACA and PMC dragged Microsoft into court with them anyway as a third-party defendant. Microsoft was a lot smaller then.) In their motion to dismiss, PMC and EACA did not deny that the code had been copied and that then-current copyright law covered computer programs, but argued that section 117 of the in-force 1976 Copyright Act required the infringement claim to be interpreted according to the law prior to January 1, 1978 ("this title does not afford to the owner of copyright in a work any greater or lesser rights with respect to the use of the work in conjunction with automatic systems capable of storing, processing, retrieving, or transferring information, ... than those afforded to works under the law, whether Title 17 or the common law or statutes of a State, in effect on December 31, 1977"). Robert Peckham, chief judge for the Northern District of California, disagreed in August 1981 and denied the motion, writing "that section 117, as it existed in the 1976 act, was aimed at the problem of copyrighted material inputted [sic] into a computer, such as books, magazines, and even computer programs. It was not intended to provide a loophole by which someone could duplicate a computer program fixed on a silicon chip." Moreover, even if it did, "[t]he plaintiff has suggested that the evidence may well show that the chip was duplicated by first taking a visual display or printout of the program in question ... If this method of unauthorized duplication in fact is proved, there can be no doubt that the unauthorized duplication of a visually displayed copy of the program would fall within the reach of the federal copyright laws." The case was quietly settled out of court, and although it obviously didn't enjoin EACA outside of the United States, domestically PMC replaced the line with the CP/M-based MicroMate in 1983. By then, and unknown to his backers, Eric Chung's failed investments in the Hong Kong real estate market had put him millions of dollars in debt. In October 1983 he abruptly fled to Taiwan reportedly with $10 million stuffed in a suitcase, leaving EACA to quickly fold. Simultaneously, Dick Smith sold a 60% stake in Dick Smith Electronics to Woolworths (the Australian version) later in 1980 and then the rest in 1982, leaving only his name and his bespectacled grin to remain at the company he founded. Although DSE sold the Video Genie's modestly upgraded direct successors also as System 80 variations, the System 80 family never included the Colour Genie, EACA's last system before its ignominious demise. PMC's former headquarters in Mountain View are now collectively Google Building E475. first step to computer literacy. The success of the ZX80 suggested other dirt-cheap home computers might also flourish. VTech accordingly began developing a low cost computer of their own in 1981 that could work like the TRS-80, planning to adopt its BASIC (and thus its Z80 CPU) much as crosstown rival EACA did and speed time to market, but by explicitly abandoning compatibility they were free to slash its production cost as much as practical. A substantial reduction in part count became possible immediately by using the inexpensive Motorola 6847 Video Display Generator, introduced in 1978, which replaced nearly the entire video system: with minimal support circuitry, the VDG chip could generate a 32x16 text display, comparable to the TRS-80 and Video Genie's 32-column mode for colour TV sets, 64x32 colour semigraphics reminscent of the same, and a variable high-resolution bitmap display depending on available memory. On top of that, the entire system could run from the VDG's standard 315/88 (3.58MHz) crystal, including the Z80, and part count could be reduced even more for a black-and-white low binned model by omitting the colour encoder completely. Everything else (cassette output, keyboard lines) could be supported with discrete components and a handful of TTL logic, price could be adjusted further on the basis of included RAM, edge connectors wired to the processor bus would suffice for peripherals and expansion, and any sort of keyboard would be a step up from a flat membrane. While the low-end computer was in development, VTech also hedged its bets with a higher-spec video game system, which typical for the era (see, for example, the Intellivision Keyboard Component) was convertible into a home computer of its own. This console was likewise built from off-the-shelf components, using a 2MHz Rockwell 6502 CPU and the Texas Instruments TMS9918A for graphics, 17K of RAM (1K for the 6502's zero page, stack and low memory, and the other 16K for the VDP), and controllers that doubled as a membrane keyboard when used with the optional BASIC cartridge. It shared no parts or significant engineering with the computer prototype and proceeded along a largely separate development track, released to select European test markets first as the VTech CreatiVision in 1982. Meanwhile, Sinclair Research and manufacturing partner Timex Corporation subsequently joined forces to launch the Timex Sinclair 1000 in the United States, a slightly reconfigured ZX81 with NTSC-compatible video output and 2K of RAM, but otherwise identical. It hit stores in the summer of 1982 at the same psychologically desirable price point, now US$100 [$345], though Timex Sinclair didn't get the bargain American home computer market to itself like the ZX80 mostly did in the UK: it now had to contend with Commodore, selling the VIC-20 as their well-supported low-end system, plus the ailing Atari and their Atari 400, Tandy's own TRS-80 Color Computer, and even Texas Instruments, then only months away from igniting a price war using the TI-99/4A. Nevertheless, industry observers generally believed the T/S 1000 would be a strong competitor, and its debut led to an accelerated scramble from VTech and others hoping to duplicate the ZX80/1's success. VTech identified two overall product positions, a super-low-cost black-and-white variation with Microsoft Level I BASIC for selected markets, and a colour model with full Microsoft Level II BASIC, each of which VTech intended to sell both under its own name and as an OEM. To prepare for the American launch of the nearly complete low-end computer and the CreatiVision, VTech opened a U.S. subsidiary that year in Elk Grove Village, Illinois outside Chicago (the picture above is from VTech's history page). Computer Gaming World issue 3.2), it took the Las Vegas Convention Center, the Hilton Convention Center, the Riviera Convention Center, the rest of the Riviera and the old Rotunda to contain it all. Mattel introduced the Aquarius for $199 [$670] licensed from Radofin, also in Kwun Tong (with a 3.58MHz Z80A and 4K of RAM, cassette storage, no bitmap graphics option and a rubber chiclet keyboard), Texas Instruments hawked the TI-99/2 for $99 [$330] (with a 2.7MHz TMS9995, 4K of RAM, cassette storage, no bitmap graphics option and no colour, and a plastic chiclet keyboard), Sanyo proffered the PHC-20 also for $99 as the midrange of the pocket computer PHC-10 and higher-end PHC-25 (with a 3.58MHz Z80 clone and 4K of RAM, cassette storage, no bitmap graphics option and no colour, and a rubber chiclet keyboard), and at the higher end came the Panasonic JR-200U for $349 [$1170] (with a 0.89MHz 6800 clone and 36K of RAM, cassette storage, no bitmap graphics option, and a rubber chiclet keyboard), the NEC PC-6001 also for $349 (with a 4MHz Z80 clone and 16K of RAM, cassette storage, but bitmap graphics and a rubber chiclet keyboard that was quickly replaced with a typewriter-style one), and the Spectravideo SV-318 for $299 [$1000] (with a 3.58MHz Z80 and 16K of RAM, also bitmap graphics, and a rubber chiclet keyboard). There were multiple conversion kits to turn the Atari 2600 VCS into a low-end computer of its own, several with rubber chiclet keyboards, and even a completely unlicensed ripoff of the T/S 1000, the Unisonic Futura 8300 (with a rubber chiclet keyboard) for $99. Not to be outdone, Timex Sinclair themselves announced the T/S 2000, a modified US version of the ZX Spectrum, with 16K or 48K of RAM, a 3.5MHz Z80, colour bitmap graphics and a rubber chiclet keyboard; the 16K version started at just $150 [$500]. The latecomers arrived at the Summer CES in Chicago, though by that point the rot was already setting in. Against the background of Commodore slashing prices even lower on the VIC-20 and C64 to Texas Instruments' profound discomfort, Mattel suddenly decided the Aquarius needed a sequel (i.e., the other system Radofin was developing; price point to be determined but without a rubber chiclet keyboard); Timex Sinclair replaced the T/S 2000 with the enhanced T/S 2024 and T/S 2048 with higher resolution graphics, more RAM and a plastic chiclet keyboard, plus an upgraded T/S 1500 which was a T/S 1000 with 16K of RAM and a rubber chiclet keyboard; Rabbit Computer (who? also from Hong Kong) introduced its own Z80-based Rabbit RX83 with 2K of RAM, BASIC, three-channel sound, cassette storage, bitmap graphics and a plastic chiclet keyboard for $99; and Tomy unveiled the Tomy Tutor for "under $150" [$500] with a 2.7MHz TMS9995 (from a 10.7MHz crystal), 16K of RAM, cassette storage, bitmap graphics and a rubber chiclet keyboard. In the fall Tandy, never one to be left out of a race to the bottom, delivered the MC-10 Micro Color Computer for $120 [$400] with an 0.89MHz 6803 and 4K of RAM, cassette storage and bitmap graphics, and a rubber chiclet keyboard. Even Commodore, failing to learn its lesson from the critically maligned Max Machine, was working on their own ultra-low-end family of computers to follow on to the C64 — one of which (the 116) would have a rubber chiclet keyboard. And, oh yeah, one other system made its debut at the Winter show. COMPUTE! in their March 1983 reporting called it "the first under-$100 [$330] color computer." Anticipated to hit American store shelves in April, the new VTech VZ200 featured 4K of RAM (expandable to 16K for $45 [$150] and 64K eventually), 12K of ROM (remember these numbers) with BASIC, and a simple push-pull piezo for sound. Two kilobytes of the 4K was allocated to the 6847, which used it to generate its default 32x16 text display, 64x32 semigraphics or 128x64 bitmap graphics. It had built-in jacks for cassette, TV and composite video, and shared its booth with the CreatiVision which VTech planned to sell States-side for $189 [$630], with the BASIC cartridge for $10 [$33] and an inevitable rubber chiclet keyboard for $30 [$100]. It isn't clear where the name VZ200 came from, possibly a riff on "ZX," but the computer's low cost even amongst a sea of low-cost computers still attracted positive attention. Creative Computing got a 4K VZ200 in for review in their May issue (accounting for publishing delays this would have to have arrived in February or early March), though they noted that they had no chance to try the peripherals or software. The article has some glaring technical errors — among others, they said the CPU was a 6502 — but reviewer David Ahl called the machine "a compact microcomputer with a great deal of capability and many unexpected features at a very attractive price." Although the review found the 4K of RAM "sparse" and was openly critical of the keyboard, particularly the absent space bar, single SHIFT key, nonstandard layout and the keys' inconvenient tendency to stutter, Ahl was nevertheless impressed by the full-screen editor ("a pleasure") and the 12K ROM implementation of BASIC (unbeknownst to him, secretly derived from the TRS-80 with added support for the on-board hardware), concluding the VZ200 to be "a great value for the suggested retail price of under $100." There are certain attributes of the machine shown in both the COMPUTE! and Creative Computing photos that don't match released units, but we'll address this later on. Creative Computing Winter CES cover showing the VZ200, the Timex Sinclair 2000 (in its original form), the Texas Instruments 99/2, the Mattel Aquarius and the Spectravideo SV-318. Personally, however, I lump these computers together as the "crap home computers." I use this term with only love, and this uniquely terrible subtype of the home computer was indeed greatly loved, because as the ZX80 had demonstrated, ordinary people could now finally afford them. Heck, my first computer — the Tomy Tutor, introduced at Summer CES — was one of these 1983 crap home computers because it's what we could afford. We couldn't afford a Commodore 64 right then, but we could afford that. Not for nothing did Jack Tramiel thunder, "computers for the masses, not the classes!" What these systems all had in common, other than crummy keyboards, a striking preference for the Z80 and an unabashedly low starting price, was aspirational and arguably fraudulent marketing, plus inadequate specifications requiring upgrades at additional cost to be practical — if there were any to begin with — alongside substandard quality control, a poor selection of software, and weak to non-existent customer support. Made cheap to sell cheap, most of these computers failed outright (e.g., the Mattel Aquarius) or were never even released (e.g., the TI 99/2). The glut that hit the U.S. market that year not only soured many American consumers on home computers generally, but their game-heavy libraries were also likely a contributing factor to the 1983 video game crash. Although managing to move over half a million units, Timex Sinclair was not immune to this effect, and the company became unprofitable as the sales crossfire between Commodore and Texas Instruments forced the T/S 1000's street price below $50 in mid-1983. The situation was compounded by Timex's ill-considered decision to make the more expensive T/S 2068 (the eventual sole member of the 2000 family) largely incompatible with the ZX Spectrum, robbing it of the extensive British Spectrum software library, and the joint enterprise that was once expected to dominate the American home computer market collapsed in early 1984. Even large players like Texas Instruments and Warner Communications-era Atari took hundreds of millions of dollars in losses, with only Tandy (due to their strong retail presence) and Commodore (due to the C64's prodigious installed base and their vertically integrated manufacturing) able to weather the maelstrom effectively. Family Computing's inaugural September 1983 issue the columnists mention the SV-318 and even the stillborne T/S 1500, but nothing on the Lasers, and Creative Computing around that time was only running ads for the 3000. A few VZ200s were rebadged by Texas door-to-door nuisance business Dynasty Computer Corporation as the Smart Alec Jr., though like their multi-level marketing attempt at rebadging the spent Exidy Sorcerer, they sold barely at all (the company folded in November). On the other hand, the (now) Laser 200 got off the ground in Europe under its own name and others, most notably through Sanyo, but also through Salora, Seltron and Texet. It launched there alongside the black-and-white model as an ultra-low-end system, originally dubbed the Laser 100, but after the ROM change becoming the Laser 110 with the same Level II BASIC of the colour version. However, there was one market where the VZ200 had particularly strong success, and that was of course Australia, though not exactly in its original form. History does not preserve the thought process of Dick Smith Electronics management, but DSE's probable aim was to nose past the Commodore VIC-20 on price and capability (DSE themselves even sold them for a time), which was colour and shipped with 5K RAM. That immediately excluded the black-and-white variation, since the ZX81 had since landed in Australia and the potential profit margin wasn't enough to bother, and it is instead more likely that during negotiations DSE prevailed upon VTech to strengthen the colour system and keep the price low. VTech's solution was a small 6K daughterboard retrofit that could replace the 2K system RAM chip, internally expanding the unit to a more appealing 8K (the other 2K video RAM chip was left unmolested) while still being able to use previously manufactured components. This modified VZ200, badged as a Dick Smith computer and subsequently sold elsewhere by VTech as the Laser 210, appeared in the 1983-84 catalogue as "new for 1983" at just A$199 [approximately US$220 spot and US$720 in 2026 dollars]. Australian Personal Computer April 1983 preview speciously characterised it as "to Dick Smith's specifications" (likely only the RAM complement was), he got quickly used to the keyboard, approved of the BASIC implementation (faster than the ZX Spectrum's) and full screen editor, noted the characters to be "rather like those produced by the TRS-80 Color Computer" (true!), and compared the memory loadout favourably to the VIC-20's. He was similarly pleased with the cassette tape performance and the included documentation, with about his only complaint being the weak sound. His editor Sean Howard was equally impressed, famously remarking that "I'm certainly going to buy one," which DSE promptly and repeatedly used in their advertising. many VTech rebadges DSE would eventually sell. Although nothing was going to catch the Commodore 64 by then, which had the same stratospheric sales there as it did most other places, the DSE VZ-200 had the added good fortune of ZX Spectrum manufacturing issues that eroded its availability, giving the DSE VZ-200 almost unrestrained run of the Aussie low-end market from which the VIC-20 was already fading and the ZX81 all but gone. In 1984 it remained a strong seller at its new lower price of A$169, dropping to A$99 by the end of the year. I think that suffices for a more detailed backstory; we'll talk a little more about its later history in Australia and VTech's overall as a postscript at the end. For now, we'll turn our attention to this orphaned American unit. A convention I'll establish from now on in this and future articles: although Dick Smith was not consistent on the hyphenation, variously rendering it VZ-200 and VZ200, in the few places it appears the American version was invariably written without one, so I'll write "VZ200" for the North American computer and "VZ-200" for the Australian computer. biblioteca. You can also see more clearly that the red blurb on top advertising "WITH 4K BYTES RAM" and "NTSC 4K" is in fact a sticker, so this box was almost certainly not exclusive to North America and may not have been exclusive even to 4K systems. increased to $149.95. Neither price matches its well-documented MSRP of $100. Unfortunately I can't make out the actual retailer, even with an extreme enlargement. markedly predominant French labeling might have prevented its sale in the Montréal outlet, that wouldn't have enjoined it elsewhere. As far as its provenance, however, the fact it wasn't labeled that way suggests Canadian sale was not initially contemplated, and it also has U.S. Federal Communications Commission clearance (I'll show you in a bit). COMPUTE! and Creative Computing pictures. Those earliest systems are labeled as a "VZ200 Personal Computer," not a "VZ200 Color Computer," and the colour labels over the number keys were absent. In fact, this same keyboard is used for the B&W Laser 110, though those early VZ200 systems must have been colour given that their colour capabilities were widely reported at the time. On the other hand, the VZ-200 that appeared in very early Dick Smith marketing like the flyer above was labeled a "VZ200 Color Computer" exactly like this one, including the American spelling. The keyboard consists of 45 keys, with a bottom right SPACE key in the corner, only one SHIFT on the bottom left, and no ESCape key. Necessarily, many have multiple functions accessible with the CTRL key or CTRL-ENTER key combo. Most of these alternative functions are one-touch BASIC keywords like the Sinclair machines, though unlike those computers, you are not obligated to use them and can spell keywords out if you want. Semigraphics characters are also selected by key combination, as well as moving the cursor, inserting and deleting characters, and interrupting a BASIC program. was removed, because the 16K RAM expander connects there. The cover is not in the box and I'll presume it was lost or destroyed by the previous owner. That's annoying from a preservation perspective, but no great loss functionally, because we'll have something plugged in there pretty much all the time later on. on the Wayback Machine, and is the only other NTSC VZ200 unit I have seen myself. Both his and mine have a "VZ200 Color Computer" bottom plate with a copyright date of 1982, and both have US FCC clearance tags and an RF switch to pick the channel (2 or 3). There are passive cooling vents on both sides, though given the fairly small amount of clearance its rubber feetsies afford from one's desk I hesitate to say they'd be effective. (At least they let you hear the piezo speaker.) There is also a small red sticker on the bottom which on mine is partially missing, but Bill's has a complete one, which reads "U NTSC 4K." I put a small piece of transparent tape over the sticker on mine to prevent further damage. Bill's unit also has both port covers. Although the serial number indicates his is a rather older unit, which we'll in fact confirm later, the label and keyboard are the same as mine and not those earliest CES models. Both units have FCC Part 15 clearance, specifically as a Class B computing device for home use; at the time Class B regulations were very strict and we'll see a consequence of that inside. The FCC ID is BNX84H80-0323, with an equipment authorization (EA) applied for February 10, 1983 and granted May 9, 1983. Accounting for publishing delays, this means Creative Computing must have had a pre-authorization prototype for their review. VTech later applied for a revised equipment authorization on July 11, 1983 and was granted BNX84H80-0323-1 on October 11, 1983; this change may have been for redesignation as the Laser 200. The listed address for VTech in Hong Kong appears on all entries for FCC grantee BNX and doesn't seem to have been their corporate address at the time, but the tester in both EAs is one Thomas Cokenias from Electro Service Corp., at 1116 Ninth Avenue in San Mateo, California, and Cokenias and Electro Service are seen in other EAs around that time for other Hong Kong manufacturers. However, the given address appears presently unoccupied, and the California Secretary of State indicates Electro Service is no longer in business. except its 16K RAM expansion (due to differing address mapping) would work on the older machine because many of them were effectively unchanged except for the labeling. (The address mapping difference is also why the VZ-200 16K RAM expander will work on the VZ-300, but only giving you an additional 8K.) Although there have been various other members spotted in the Laser 300 series, apparently differing only in RAM size or case type, none were reportedly sold widely if at all. The VZ-300/Laser 310 is otherwise nearly totally compatible with the VZ-200/Laser 210 except for its clock speed. As NTSC compatibility was no longer required, VTech switched to a single 17.734475MHz master crystal divided by four for the 4.43361875MHz PAL colourburst and five for the 3.546895MHz CPU clock. (The Motorola 6847 VDG in the VZ-300 and PAL systems generally is clocked with an altered signal which I'll talk about when we open the machines up.) Although that makes the VZ-300 slightly slower at 99.09% the speed of the VZ-200's 3.5795454MHz (315/88) oscillator, in practical terms the difference was imperceptible with most existing software — but of course it's going to be a problem for us later. VTech did no other upgrades, not even offering an alternate 6847 font ROM with lowercase, though we'll talk more about that when we get to the innards as well. a Christmas gift from my wife (I married well). The DSE pricetag on the back gives its price as A$9.50 but I don't know where or when it was originally purchased. While the TRM does not approach the sheer detail of, say, the Commodore 64 Programmer's Reference Guide, which even has an exhaustive memory map covering its 64K of RAM and 20K of ROM, it does include documentation on most of its important RAM locations and a full set of schematics, some of which we'll be using in this very article. A similar manual exists for the VZ-300 with its own set of schematics, which also discusses the VZ-200 and has some details not in the earlier text. (for Los Angeles residents, read "Vernon"). The North Ryde building was subsequently occupied by German truck and bus manufacturer MAN, whose old stripped logo can still be seen on the facade, and is now split between multiple tenants. Level I BASIC is quite, uh, basic, evolved by TRS-80 designer Steve Leininger from Li-Chen Wang's "copyleft" Palo Alto Tiny BASIC which Leininger substantially altered, reworking its structure and adding floating point math (because it couldn't accept Charles Tandy's salary when he typed it in), two string variables and a single array, and support for the TRS-80 hardware. VTech's version is similarly modified to support its own architecture but is otherwise the same. Level II BASIC is Microsoft BASIC, derived from Microsoft's own Extended BASIC on the Altair, and again VTech initially imported it nearly unchanged except for adding support for their hardware and graphics, which replaced some of the keywords (like TROFF with COLOR). Proof can be seen by comparing the order of keywords in their token tables, which would have no reason to match as precisely as they do unless they came from the same origin. Another persistent relic in the VZ BASIC ROM is the old Microsoft two-character error message table (e.g., ?SN ERROR instead of ?SYNTAX ERROR) even though the existing code never references it. Tandy's apparently successful legal action against EACA and PMC spooked VTech that they could be sued in the same way — but by Tandy and Microsoft. The company's solution was to dump Level I BASIC entirely, eliminating any objection from Tandy, and for Level II BASIC to secretly null out a large number of entries in the ROM BASIC keyword table potentially unusual enough to be used as evidence of copying. The tokens for those keywords still remained valid, however, and the ROM code for them also largely persisted. Disk-specific keywords were likewise omitted in the same fashion, at least until the VZ-300 disk drive debuted, but unlike the other gutted keywords were instead vectored through RAM for later expansion. These keywords (in token order) are CMD, RANDOM, DEFINT, DEFSNG, DEFDBL, RESUME, ON, OPEN, FIELD, GET, PUT, CLOSE, LOAD, NAME, KILL, LSET, RSET, SAVE, SYSTEM, DEF, DELETE, AUTO, FN, VARPTR, ERL, ERR, STRING$, INSTR, TIME$, MEM, FRE, POS, CVI, CVS, CVD, EOF, LOC, LOF, MKI$, MKS$, MKD$, CINT, CSNG, CDBL and FIX. Because their implementations often remained present, it was possible to resurrect many of them by hooking into the RAM vector for tokenizing "new" BASIC keywords, and some BASIC extensions did just that. The earliest versions of the Laser/VZ ROM use light green text on a dark green background. Without a colour encoder this would produce light text on a dark screen, which is indeed what you see on a black-and-white Laser 110. Bill's earlier NTSC VZ200 is the same way, using the same ROM 1.2. By ROM 2.0, as in this VZ-200, the display became dark green text on a light green background, like the Tandy Color Computer which uses the same MC6847 video chip. do have a 4K system, just a very late one. You may have noticed the oddly convoluted BASIC statement I entered to get that figure. That's because I had to avoid typing the number 5: the key didn't work. Normally the VZ200 will beep as you press keys, but there was no beep and no response when I did. In fact, an entire run of keys (5, T, G, B and N) was not working, and I was only able to enter the PRINT keyword by pressing SHIFT-P. We'll need to fix the keyboard now to do anything substantial with this machine. did appear to work. Still, it seemed like the most reasonable place to start, so let's crack the computer open. another VTech unit, the unrelated Laser 50) and a plastic cover sheet with slits for the top ports' card edges to reduce dirt getting inside. Also visible on the top right/northeast side is a sprawling heat sink screwed to the fin of a 7805 voltage regulator peeping out from the lower right/southeast corner. This heatsink sits under the ventilation slits in the top case. Much of the logic board is covered by a large sheet metal Faraday cage serving as an RF shield, festooned with soldered metal braids connecting everything to the ground plane, and then the whole assembly placed on an irregularly shaped metal plate on the bottom with the piezo. As mentioned, in those days the FCC was very strict about radio interference from computers and video games, particularly home units where a Class B device (as this is) had a 10dB lower maximum than a commercial Class A one. Some systems like the Atari 400 solved this problem by effectively encasing the entire system in a molded metal endoskeleton and placing as few holes in the case as possible from which radio signals could emanate. That made for a very sturdy computer but one more expensive to manufacture, so VTech went for this cheaper, hackier approach which was no doubt iterated upon until it just cleared the bar. One major component is not under the cage, however, and that is the Motorola 6847 VDG video chip, in a plastic carrier (MC6847P) with a date code of 24th week 1983. It's not precisely clear why that is, but it can be seen that the left board is connected to some of its lines. PAL synchronous demodulator, a surprising choice, since the MC6847 is usually paired with the MC1372 NTSC colour modulator. The 6847 emits YPbPr (as Y, B-Y and R-Y) video, which in the black and white Lasers only the Y (luma) signal is used. In other systems like the Tandy CoCo the MC1372 takes the YPbPr lines, encodes the colour, and emits a signal suitable for a television set and/or composite video depending on the specific components. The TBA520 can be made to do the same task, even generating NTSC colour with the right crystal, albeit with more supporting electronics. (I presume it was less expensive than an MC1372, plus there would be the advantage of not having a different chip for Euro and Aussie systems, since the TBA520 can obviously do PAL video too.) Reference B-Y and R-Y signals are generated to the TBA520 using the 3.58MHz (315/88) NTSC colourburst oscillator on the left board, while the B-Y and R-Y signals from the MC6847 are passed on different lines, with the MC6847 running on the same 3.58MHz clock. We then pulse the TBA520's line inputs at the necessary horizontal rate, approximately 15.7343kHz, causing the TBA520 to emit a single NTSC chroma signal (on its G-Y pin) from the 6847's PbPr signals. (This theoretically makes it possible to get proper S-video out of the VZ200, though we're not going to try that this time.) Discrete components then combine the luma and chroma into composite video for both the monitor connector and for feeding into the separate RF modulator. this problem), so I decided to just clip the tabs with angle cutters. One of those tabs is here over by the 7805 ... Scriptovision Super Micro Script, which has a 6802 CPU (a microcontroller version of the 6800) and a 6847 VDG, RAM, ROM and a simple keypad for entry. In that article I mentioned that those were enough to make it almost a home computer (replacing the ROM on the Super Micro Script to make it one) because home computers like the VZ-200 have a similar architecture. Now we'll prove the comparison is valid. In this view, the MC6847 is the large DIP on the far left/west side. The chips in the back, going from left to right, are a Hitachi HM6116P-2 2K static RAM used for video memory, the CPU, a clone SGS Z80 (the former state-owned SGS Microelettronica "Società Generale Semiconduttori" of Italy prior to merger into STMicroelectronics in 1987) with a date code of 19th week 1983, a Hitachi 74LS139 and 74LS32 used as part of the address decoding for the memory mapped I/O range, and lurking in the back right (east) corner a 74LS174 6-bit flip-flop serving as a latch register. In the front, also left to right, are a Hitachi 74LS245 octal bus transceiver which bridges the shared 2K video RAM between the CPU and VDG, a Hitachi 74LS04 hex inverter used for various tasks such as the CPU reset circuit, a Hitachi 74LS244 octal driver, another Hitachi 6116 2K SRAM (the system RAM this time), and the two 2364 8K system and BASIC ROMs with date codes of 27th week and 32nd week 1983 respectively. On the Dick Smith schematic using the same order, these chips are numbered U15 (the 6847), then U7, U4, U3, U2 and U1 (74LS174), then U14 (74LS245), U13, U12 (assumed), no designation for the single 2K SRAM, and U10 and U9 for the ROMs. The ROMs are unsurprisingly the newest chips in the system and mean the computer could not have been assembled earlier than then, likely making it part of the last production runs before VTech abandoned the line in North America. Obnoxiously, everything is soldered down; there are no sockets (cheap!). However, there are a number of unpopulated pads. Some extra pads near the ROMs were clearly intended to accommodate larger-capacity chips, and later machines indeed use a single 16K ROM with a change in board jumpers nearby. (Braver folks than I have extracted VZ-200 boards with 2364s and found bodge wires underneath. Apparently underpaid labour and smaller ROMs were less expensive at the time. Cheap!) There is also another set of pads near the VRAM chip between it and a big block of through-hole resistors. It's not clear what these pads were meant for, though the VZ-200 schematics show another resistor bank there serving as pull-ups. This bank is drawn on the schematic with dotted lines unlike the other set, so perhaps they were eliminated for cost reasons (cheap!). Now let's talk about what we don't see. One thing we don't see here is a 3.58MHz master crystal. That's because ... we already saw it. The entire system runs from the 3.58MHz crystal on the colour encoder, so everything is precisely synchronized to the same clock source, including the TBA520, the MC6847 and the Z80. It is therefore impossible to merely switch the colour encoder boards and turn an NTSC VZ200 into a PAL VZ-200 or vice versa because of the added line padding circuit in the PAL unit and the absence of a second system crystal in the NTSC unit (cheap!). What about the VZ-300, where there's no 3.58MHz crystal either? In that system, the VDG is entirely clocked by one of the gate array chips, providing the same line padding logic, but also using the same 3.54MHz clock as the CPU which is apparently "close enough." Another thing we don't see is something like a Motorola 6883 synchronous address multiplexer. Recall that the 6847 VDG has no externally exposed registers of its own (compare to, say, the VIC-II in a Commodore 64). Things like video modes and character attributes are twiddled by chip lines; for character attributes these lines are often wired to certain data bits from the video RAM, but to dynamically set a video mode under software control requires external hardware. Also, the VDG makes little attempt to cooperate with the CPU as to when it accesses video memory, other than indicating when it finishes a frame which is likewise asserted on one of its pins. In the Tandy Color Computers (prior to the CoCo 3, which uses a GIME), the MC6883 SAM sits between the 6809 CPU and the 6847 VDG and arbitrates all of this, managing the display mode and system timing, and servicing the VDG while the CPU is on the bus. On the other hand, this approach was judged too expensive for the cut-down Tandy MC-10, which instead uses a series of flip-flops to interleave bus access between its 6803 CPU and the 6847. A single 74LS245 bus transceiver allows the CPU unrestricted access to the video RAM when the processor is accessing memory. Neither approach is suitable in this case, however. An MC6883 would be too expensive for the VZ200 also (cheap!), and the Z80 sits on the bus longer during a CPU cycle than a 6502 or 6800-family chip would, so the MC-10's interleaved approach won't work either. As it happens, the VZ200 does nearly exactly what the Scriptovision Super Micro Script does: during CPU video RAM access, the VDG's memory fetch is immediately suppressed using its MS pin and the 74LS245 bus transceiver temporarily kicks it off the bus. The contention problem is solved in both cases with software (cheap!) by simply not doing anything with VRAM until the VDG indicates it's between frames and not reading screen memory. The only difference is how they find that out; the SMS busy-waits on the VDG's FS signal before doing a screen update, while the VZ200 just wires FS to its IRQ line — screen updates using ROM routines are batched and when the interrupt is triggered, the ROM then blits the deferred changes to the screen all at once. Of course, if you write directly to the video RAM when the VDG is accessing it you'll get intermittent artifacts, and I'll show you what that looks like, but you can just watch for the IRQ yourself if you really care about it (many programs didn't). Otherwise, the VDG's mode pins for bitmapped graphics and alternate colour selection are handled through bits in the 74LS174 latch within the memory-mapped range, which also handles the piezo and cassette output. Overall this is a good demonstration of how the SMS was almost a home computer, because here's a home computer whose video architecture was almost the same. A missed opportunity with the more upmarket VZ-300 was the potential for lowercase or at least an alternative character set, especially because it even got a word processing cartridge released for it later (we'll play with it, it works with the VZ-200 also). An external font ROM can be lashed to the 6847 with a bit of additional circuitry and the Super Micro Script has one to generate higher-quality character glyphs. It seems like VTech could have done something like that controllable by another latch bit, and with a six-bit latch there are a couple more data bits there that could be used, but I guess that was either judged too risky or not even thought about. On both systems the 6847 INT/EXT pin that would have controlled this is merely hardwired to ground. One note about the metal RF shield: if the cage is not pulled back down into position, it may distort and contact some of the pins on the 74LS174. This will cause weird graphical artifacts and knock out the piezo (no keybeep). It doesn't appear to harm the computer, but I was very careful to ensure it was bent back to as similar a position as before after this happened a couple times. Obviously this is no problem if you just completely take it off. that), which is why they could never complete a junction and be sensed. On the other hand, the 6, Y and H keys on that column are wired separately into the ribbon cable, so they were unaffected. I'm not sure how such a fault would have happened. The keys move, but they don't sweep or scour, and ordinarily they shouldn't be contacting the painted portions anyway. It also doesn't seem likely to have been a factory defect because that would have made the computer very difficult to use, and this computer was clearly used. I pondered the best way to fix it, since any repair would have to be flat or it would distort the key sheet on top (e.g., no solder blobs, no top bodge wires). I have a circuit pen I could use to draw a new trace, but it's temperamental, and I didn't want to do something I couldn't undo later in case it wasn't actually the problem. do work! Do not overtighten them or you will interfere with the conductive nubs being able to make contact. I had a couple dud keys initially after this which were returned to life by slightly loosening the screw nearest to them (to my great relief). COLOR ,1): #00ff00 and red is #ff0000 and so forth. That is definitely not the case; in fact, the default VDG palette is rather a bit muddy, with relatively poor saturation. For example, black (what the border is supposed to be) is often more like a very dark brown, buff is a dirty off-white, magenta becomes a flaccid purple where the red is a little too low, and what the documentation calls cyan comes out closer to seafoam green. Unfortunately, many simpler or older emulators provide an excessively rosy (no pun intended) simulation of what these typical home computer implementations usually generated. MAME uses the correct palette and the VDG's Wikipedia entry has a credible synthetic screenshot based on the YPbPr values in the datasheet, which you can compare with the real composite grabs above. absolutely capable of good output, but to do so it needs a quality encoder, and the VZ's ain't it. The best colour I have ever seen from a VDG is actually the Super Micro Script's, using a very high quality output stage as shown in the actual grab above, and comes out vibrant, beautifully saturated, and fabulous on a CRT. You would expect that, however — it's a $500 prosumer video titler from 1985, not a $99 crap home computer from 1983. The next order of business is software. I'd rather not use tape or audio files, and I don't have the disk drive. Fortunately, because the VZ series is so beloved in Australia, those wacky Aussies occasionally create their own modern peripherals in between prawns on the barbie. If you have a VZ-series computer, then you need the BennVenn VZ300 SD Loader. It is fairly inexpensive and provides you a way to load software into your VZ-series computer via SD card, along with topping off the RAM, even more memory with bank switching, and optional solder-yourself connectors for gamepads and I/O expansion. I figured it might be fun to build some hardware for it (and we're going to create a very simple expansion ourselves for the VZ200 in this article), so I ordered the full kit. It works well for my purposes and my wife has ordered another for the VZ-300 now at my in-laws' house in regional NSW. However, I am neither affiliated nor associated with Ben, merely an overall satisfied customer, so this is the part where I will also make three gentle constructive complaints about it. First, things like new firmware are largely delivered through a private Facebook group. This group appears to be very welcoming to new members, but it requires you to be on Facebook, and I don't want to be on Facebook. I managed to get the current firmware another way, and I will be putting it in the Github repo for this project so you don't need to join Facebook either. (If you do want to join, however, I'm sure the "VZ200 VZ300 Laser210 Laser310 fans" group would love to have you.) On the other hand, Ben was reasonably accommodating of my questions over E-mail which I did appreciate. looked like the right one until I got out the continuity tester and realized the actual fault was elsewhere. I'm not sure why some of them were routed around instead of in a straight line. Salora Fellow, a Finnish rebadge of the 4K PAL Laser 200 (by contrast, the Salora Manager is a Finnish rebadge of the CreatiVision-derived Laser 2001). This is determined by a simple memory map check in a snippet of its VHDL that Ben shared with me: x"B7FF" ) else '1'; --B800 or higher RAMarea200 <= '0' when (Address > x"8FFF" ) else '1'; --9000 or higher RAMareaSelora <= '0' when (Address > x"7FFF" ) else '1'; --8000 or higher put the reset button, nor was I particularly enthusiastic about drilling a hole in the case to make one, first due to the questionable quality of the plastic itself and second for purposes of historical preservation of an unusual machine. a Gremlin Blasto arcade board (where we had no reset circuit of any kind) that its 8080A CPU could be crudely reset by simply putting a pushbutton switch between its reset pin and ground. As the Z80 can serve as a drop-in upgrade for an 8080, it can be reset in the same way. .VZ format, which has a trivial 24-byte header indicating filename, starting address and type (BASIC or binary). The LOAD command will conveniently auto-execute binary files when the load completes. You can find many programs on Dave "Bushy" Maunder's exceptionally comprehensive site containing software, photographs, articles and documentation, and most software he offers can be copied directly to the card and used immediately. 2018AD. There's not much of a VZ200 demoscene, but there are a few out there, and this exceptional demo is unquestionably one of the best. It will not run correctly on this particular computer because the video timing is different — not because the clock speed is faster, like you'd see between an NTSC Commodore 64 which is faster than a PAL Commodore 64, but because the NTSC VDG draws the screen faster and thus fires the end-of-frame interrupt more often, messing up synchronization. For that, you'll just have to watch this YouTube recording on actual PAL hardware. It's set up like a trackmo, streaming cassette program data from one channel of a specially recorded CD audio track and using the other channel for music (admittedly a bit of a cheat but the music is excellent). If you didn't think rotozooms, raster splits and even FLD-type effects were possible on the 6847 VDG, then you're in for a treat. Another neat trick is the doubled vertical resolution while drawing the Kefrens bars. The source code is even available for your education. entirely done with semigraphics. It restores the old light-on-dark screen colours used in earlier ROMs with POKE 30744,1 (this works with twiddling the CSS pin with COLOR,1 also). Dubois and McNamara (i.e., Greg Dubois and Tricia McNamara, though Greg did all the programming), who created various titles for a number of DSE systems that were sold in stores. I should note that for many games, including this one, the more-or-less standard control keys are Q and A for up and down, M and comma for left and right, and where a fire button is used, typically SHIFT (SPACE for secondary). Note the "snow" in this shot — that's because the program was animating screen memory at the same time the VDG was trying to read it, so the VDG's memory access got briefly suppressed during that period. Because the display scan can't wait for the VDG to be enabled again, the result is a brief splat of garbage on that line until the VDG is allowed to proceed. Dubois could have simply waited for the VDG's next interframe interrupt, but there's also only so much time between frames before the VDG will start drawing again. As a result, for many programs where significant CPU time was required to do screen updates, outright ignoring the video artifacts turned out to be the least bad approach. that was) by Stephen Clarke. It shows loading from the SD card, the title and options screen (even the VZ200 had software pirates), and then playing the game, which worked fine on the keyboard. This and the other recordings I did for this entry were generated from the composite capture rig for video, but for audio using a microphone near the VZ200's piezo speaker to capture sound (since there's no audio out). You'll notice I'm pounding on the keys a bit, which the microphone faithfully picks up, though you do have to hit the keys with a bit of, shall we say, deliberateness to get them to register. Lemonade Stand: Meatpies, where you sell pies. Yes, the meaty kind, America. glasses pies you want to make, how many advertising signs you want to buy, and the price per item you want to request. Other than the pie business, Larry Taylor's port of the game seems to be heavily influenced by the well-known Apple II version and includes the same sort of simple graphics for weather reports and the like. It oozes professional quality with a slick title screen and menu, and is an extremely fun puzzler to boot. The aim is to create a factory from various primitive machines (paint, rotate, hole punch) that will generate a specific product. It is so well animated and so thoroughly polished that it deserves this short video to fully appreciate it. copious documentation for all its features, as it was a surprisingly credible word processor on par with at least, say, Color Scripsit on the Tandy Color Computer, though Color Scripsit is some years older. Although Wordpro came as a cartridge intended for the VZ-300, it would work on a VZ-200 with correspondingly less document memory available, since it occupied the slot where the RAM expander would go. On the other hand, the program seems to be calibrated for a VZ-300 keyboard since on this VZ200 the keys seem to frequently stutter. (I'll talk about how I got it to work on this 4K system later, though it would not have been possible without the BennVenn RAM expansion.) see this emulation — is also notable because the CoCo 1/2 and the VZ-200/300 all lack lowercase, so both programs solve it in the same way by displaying "capital letters" in reverse video. Wordpro's user interface is more sophisticated than Color Scripsit's, but it's also newer. The lack of a lowercase option on the VZ-300 was again a real missed opportunity, and with the number of machines DSE was buying you'd think they could have talked VTech into engineering a solution. Although Wordpro supports both disk and tape, the BennVenn cartridge currently doesn't emulate them sufficiently for Wordpro to use it. Perhaps this is a hack we can do some other time. a la Night Driver where you are apparently an alien with wings and feet ... It's only rendered at 90 degree angles, but it's fast and well-written, and deserved another video. Some games, though many were fine, would show a weird line pattern over certain sections of the screen like this port of Exidy Circus/Midway Clowns. The pattern was annoying, but in the first few games I played where it manifested, it appeared to be cosmetic (it does not restrain the jumping figures here) and I initially chalked it up to some other undiscovered difference in this NTSC unit. LDI that copies the contents of the address referenced by HL to the contents of the address referenced by DE; the instructions LDIR and LDDR expand upon it, running LDI repeatedly and decrementing the count in BC each time until it reaches zero, respectively incrementing or decrementing HL/DE on every step. The most obvious application for these instructions is copying a block of memory elsewhere, but a less obvious application is using them to fill memory. Consider this segment of actual code from Invaders (dumped with z80dismblr): Ignoring the instruction at $8ca1, you can see that this is setting the source to $7000 — i.e., the start of VRAM — and the destination to $7001 (?!), for a total of $0820 bytes (the zero test is post-decrement). Now, what would that accomplish? Just before the LDIR, we set the contents of $7000 (in HL) to zero. Let's step through the process LDIR takes manually. $7000 is first copied to $7001, which is now zero as well. HL is incremented to $7001, DE to $7002, BC decremented to $081e. Next, $7001 is copied to $7002, but $7001 had zero in it because it was copied from $7000, so all three locations are now zero. HL is incremented to $7002, DE to $7003, BC decremented to $081d. Then, $7002 is copied to $7003, so now all four locations are zero, and so on, filling all intervening locations with the immediate value before. At the end, when BC finally gets to zero, all locations from $7000 to $781f inclusive (i.e., the entirety of video memory and a little past it in system RAM) will have been zeroed out. This is how Invaders clears the hi-res screen and it is indeed faster than a naïve loop, especially for large tracts of memory. But our obnoxious little plastic beast here throws in a wrench by having a location where the RAM isn't working properly. When the "copy" gets to that point, because the copy is only between adjacent memory locations, for every subsequent location the stuck bit will be propagated forward and faithfully copied to each and every byte afterwards. That's also why the pattern doesn't cover the whole screen, because the problem doesn't actually manifest until the "copy" operation arrives there. In fact, in the process Invaders was unwittingly corrupting some of its own game variables with the same stuck bit when they should have been zero, possibly another reason why it wouldn't run correctly. The direct and most definitive solution would be to "simply" replace the VRAM chip, and I even have 6116 SRAMs in stock, but I warned you this is a very cheaply made PCB. Far better repairpersons than I have tried and failed to replace chips on these computers without requiring a lot of rework and bodges, and the prior portions of this article should have already convinced you it's only by the grace of God I haven't soldered my own fool face to the workbench yet. I did not want to try replacing that SRAM chip solely because of one stinking bad bit; I was only likely to make a bigger mess or render the computer completely inoperable. But again: we have an alternative. This fill trick was not universally used or even known by all programmers at the time. The games that do work clear the screen with a simple loop that doesn't propagate the bad bit forward, which works because the store doesn't depend on what memory contents are already "there." Likewise, VTech doesn't seem to use it in the ROMs, which is why the problem didn't manifest during the demonstration tape or with BASIC programs drawing to the screen with BASIC keywords. Most VZ programs are small enough and this code idiom distinctive enough (and usually only present once) that such code can be found and patched to use a slightly slower but functional loop. As such a loop would generally require more bytes, the patch could either direct execution to a tacked-on routine to do the clear, or we could patch it to point to a standard routine in memory. And how are we going to get a standard, always-present, stock routine into memory to do that? Easy: we're going to soft-alter the BennVenn SD loader's firmware. No, stop laughing, because we have a simple means to accomplish it. At the same time we'll combine that with a serial port loader so that we can test these programs live without having to constantly swap the SD card to and from the Talos II, so we'll also build it a bitbanged serial port (I said stop laughing). There were homebrew serial devices back in the day for these computers, so consider this one merely another entry from a venerable tradition. .VZ format. Here's a simple, complete example of "Hello World" showing how to construct that header and which can run directly from the card, demonstrated in the screenshot. This is one of several files you will find in this article's Github repo. As with all our assembler projects except for the 6502 and PowerPC, we crossbuild using the Macroassembler AS. The 24-byte header marks this as a machine language program that starts at $8000, the beginning of the extra memory furnished by the BennVenn device. Although the magic number VZF0 (for BASIC programs, which always start at $7ae9) or VZF1 would appear to be critical, the firmware doesn't seem to check it on loading or even generate it on saving, only that the byte just before the starting address word (everything is Z80 little-endian) is either $f0 or $f1. Upon execution our program then calls a "display null-terminated string" routine in the VZ ROM, the update for which is pushed to the screen during the next VDG interframe period, and returns to BASIC. The binary is assembled with AS like so, using a simple Makefile: >hello.vz (44 Bytes) % xd hello.vz 00000000 56 5a 46 31 48 45 4c 4c 4f 00 00 00 00 00 00 00 |VZF1HELLO.......| 00000010 00 00 00 00 00 f1 00 80 21 07 80 cd a7 28 c9 48 |........!....(.H| 00000020 45 4c 4c 4f 20 57 4f 52 4c 44 0d 00 |ELLO WORLD..| 0000002c LOAD"HELLO" will load and immediately execute it from the given entry address (which is both the load and execute address), as shown in the screenshot above. the Github repo I have a small program to cycle the LEDs on his proto board, which you can see in this brief video. It sets all GPIO pins to output and lights all 24 green LEDs connected to them (the red ones are check LEDs for 3.3V, 5V and 9V), then cycles a dark one through them from left to right until a key is pressed. This is done by hooking into the interrupt routine called when the VDG completes a frame, the only regular timesource on an unaltered VZ200, and used back in the day as a simple clock by various programs. Every third tick of the interrupt routine, this code runs: It rotates an in-memory image of the GPIO pin values, then emits that to the I/O locations. Because the rotation is to the left, we can see that the GPIO lines must also be oriented little-endian, i.e., the least significant bit of each GPIO output register is on the left. This then informs how we'll set up our bitbanger. A half-duplex system will suffice for downloading, since the sender will wait for us to indicate receipt between packets, and that will let us concentrate entirely on receiving until a full packet is obtained. The absolutely fastest speed we can receive at is generally limited by how quickly we can clock data bits into an accumulator from the receive line. If we connect the receive line to the least-significant input of one of the GPIO registers (we'll use the leftmost for convenience), we can do it in 23 cycles: The Z80's clock speed (in any of these systems) does not neatly divide into any standard bitrate, but theoretically 23 cycles per bit gives us a maximum possible transfer speed of ((315 000 000/88)/23) =~ 155632.4 bits per second. That suggests you might be able to get 115200bps with an unrolled loop, but at speeds this fast time required for other tasks starts to be a concern, such as storing to memory, checking how many bytes have been received, and branching back to get another, all of which together will certainly be more than 23 cycles. The other problem is the time required to sense the start bit, because this can occur at any moment, and hardware UARTs generally end up repeatedly snooping the line at some multiple of the bitrate to ensure they won't miss one. We, on the other hand, can't even check for a start bit at just twice 115200bps. In fact, the fastest we can check for a start bit (a zero) is which because of the unavoidable branch is actually longer than the time to clock in a data bit! However, these numbers do suggest that half that speed, i.e., 57600bps, is plausible. Flipping the equation around, that gives us a relatively generous ((315 000 000/88)/57600) =~ 62.1 cycles per bit, long enough to do our housekeeping tasks on each byte, and our tight startbit loop can poll the line at ((315 000 000/88)/25) =~ 143181.8 bits per second (a familiar number to some of you), which is at least twice the data rate and should be sufficient for the sort of continuous data transfer we'd experience receiving a data packet. We will target this speed. (Note from the future: an early draft used in a,(c) in the startbit loop, which is a 12-cycle instruction. This single extra cycle reduced the startbit poll rate to 137674.8bps, and at that speed multiple bytes got missed and/or corrupted. We are probably only just fast enough to make this work.) Parenthetically, VZ-300 owners in the audience will now have asked if this will work for them. If we substitute its lower clock speed at 57600bps, we get ((17734475/5)/57600) =~ 61.5 cycles per data bit and a maximum startbit poll rate of ((17734475/5)/25) == 141875.8 bits per second exactly. Because you can sample a little bit faster but never slower, we would need a separate version for the VZ-300; the same code will not work reliably on both. Sorry! That will be the subject of a future article. Note that by making receive fast, we made transmit slower: unless we occupy the least-significant bit of another GPIO register, which seems rather wasteful, the next fastest position is the second-to-least significant bit. This snippet needs no less than 31 cycles to send the next bit of a character stored in a register other than the accumulator (here we'll use B): If we had to completely guard that GPIO register from interfering with any other GPIO pins on the same register, it would be even longer because we would need to read the current state and then do the bitmasks. Mercifully we'll just refuse to support that, and as 31 cycles is still well within our 62 cycle maximum per bit, she'll be right. Texta Sharpie. If you don't have his board, you can still figure it out a little less conveniently with a voltmeter. Ensure all pins are set to output and turned off (something like FORI=68TO73:OUTI,0:NEXT will do from BASIC). The only live lines at that point should be ground, 3.3V, 5V and 9V. Find ground first, which you might do by checking for continuity with the ground test point Ben provides on the board, then use that as your common to find the voltage pins. Mark those; the rest are GPIO. Turn on pins one and two individually (OUT 71,1 or OUT 71,2) and look for voltage. don't connect the 3.3V line. Instead, for the programs below, ensure the HW-597 is already plugged into your host (such as with a USB extension cable) and showing bright status LEDs before powering on the VZ-200, or it may try to unsuccessfully power itself from the other lines and get a little daft. BITS) toggles the screen colour as it sees activity on the receive line. This is easiest to watch at a slow bit speed of around 150 baud or so. Here, I hooked it up to the Talos II, ran picocom -b150 /dev/ttyUSB0 (adjust for the path to your device), and just banged on the T2's keyboard. If you get alternating flashes of green and orange on the VZ while you do so, then your receive line at least has basic connectivity. ASCII), effectively one half of a very slow terminal program. We'll use the internal ROM routine to display a character, which will get us scrolling for free, and then force the update instead of waiting for the next IRQ — which is disabled anyway to make sure our timing remains precise. (Typing only upper case characters works; lower case shows as symbols.) This program runs at a sedate 300bps, and the reason is because the VZ ROMs are written for space efficiency, not time efficiency, at least to any extent they're efficient at all. 300 baud gives us an apparent surfeit of cycles using our formula — 11931 cycles per bit — but we may well need all of them since we've really got no idea how long it can take the ROM routines to do any arbitrary screen update. We won't be using the ROM much for our data blaster program, but a general purpose terminal emulator would have to consider a proper solution to achieve faster speeds. This is something else we might revisit in a future article. The other purpose of this ASCII test program is to mock up how we'll write the fast serial loader. Despite the fact we have over 10,000 cycles between bits at 300bps and could easily have written each bit we read as a subroutine call to save memory, I still inlined each clocked-in bit using a macro because we necessarily need to at 57.6kbps — among other things, each CALL is 17 cycles and the RET to return from it is 10, which would consume almost half our CPU budget by themselves. I'd also like to observe, again with my usual biases showing, that cycle counting isn't nearly as much fun on the Z80 as it is on the 6502. Most opcode tables will fortunately collapse the whole Z80 T-state and M-state business into a single unified cycle count, but unlike the 6502 where there are instructions with execution times of 2, 3, 4, 5, 6 or 7 cycles (so you can easily make a busywait from any combination), the Z80's cycle time options start at 4 and go as high as 23, skipping many numbers, and many of the smaller cycle times require specific conditions like not taking a branch. Having considered our little half-terminal program, here's what I settled on for 57.6kbps, written as AS macros: From our previous maximal case I turned the in b,NN instruction into a slightly slower in b,(c), which burns an additional cycle, but means we have 38 cycles left over of our 62 which we can split exactly between two ld (ix+N),b instructions of 19 cycles each. (This also lets us possibly alternate between multiple connected serial devices by changing C, but one catastrophe at a time, I always say.) The separate top and bottom waits are for situations where we have an odd number of cycles left over and need to have different wait times; consider this future expansion for the VZ-300. When the stop bit arrives, we need to dump the byte into a buffer and get ready for the next one in the same 62 cycles, since we expect the other end will be ready to fire the next start bit at us immediately. To make an interesting and vaguely useful display (as well as not requiring additional memory), the screen itself would seem like a good place, but this also imposes some constraints: we only have 512 bytes there (i.e., 32x16), some of which we also need for indicating status, meaning our received packets should really be no larger than 256 or 384 bytes to allow for a transmission log and other useful info. The protocol we select should have packets no larger than that, be easy to implement (because I'm lazy), and be something that pretty much everything can speak. While we've seen Xmodem-1K or Xmodem-CRC implemented other places (like The Newsroom's Wire Service variant), I just decided to go with good old O.G. Xmodem. That contains 132-byte packets and is easy to write and checksum, and any errors over USB between your host computer and the VZ200 would undoubtedly be from bad bit framing rather than line noise which the default checksum algorithm should detect. While it overruns memory a bit at the end, this is largely irrelevant for just loading something we intend to immediately execute. All that preamble yields us a stop bit stanza like this: Here we use the IX index register as a pointer into screen memory and the L register as the packet length countdown. A double-store of the same location onscreen once again soaks up 38 cycles, then the increment and decrement, then the branch. A nice thing about the JP instruction, which is absolute instead of relative, is that the conditional branch form requires 10 cycles regardless of whether it's taken or not, so this entire stanza always consumes precisely 62 cycles as well. We use that instruction a lot in the cycle-exact portions so that we always have predictable CPU time. Once we get a full packet, we know the sender won't do anything until we reply, so we can relax our timing and validate the packet at leisure, copy it into the correct place in memory and send the ACK for the next one. We send bytes using the same send-bit route I showed you before or a trivial variation, padded to 62 cycles per bit and also inlined. We accept .VZ-format files in this loader, so we have special handling for the first packet to make sure it has a generally correct format and note the type and memory address, which is where the rest of this packet and subsequent packets will be copied to. If it doesn't, then we send CAN and force the sender to abort. By contrast, the start bit is handled with the same 25-cycle code I showed you before, because this is the fastest way we can be sure we won't miss one. But this also means we have no way of checking the keyboard nor implementing a timeout: there is no spare time to count cycles or scan for keys, and the only free-running timer is the VDG end-of-frame IRQ which will totally mess up our timing if that runs, so during the entire transaction IRQs are disabled as well. The program therefore assumes your sender is up and ready to go the moment it starts executing. On startup it fires off the initial NAK and waits, possibly forever if the other end never gets the signal until you reset the VZ200. (We display a message to alert you that no transmission has yet been received, which is immediately overwritten on-screen by the .VZ metadata.) Let's see our loader in action. On the host side, to send the program to the VZ200 you can use any terminal program that speaks Xmodem (and just about everything does), just as long as it automatically starts the transfer as soon as the initial NAK arrives. With both my MacBook Air laptop and my Raptor Talos II workstation, I use lsx with the usb2ppp tool from BURLAP, which we earlier used to tunnel PPP over a serial line for the Brother GeoBook, but can be used to run pretty much any program over a serial port. Here's an example, substituting the path to the HW-597 that your machine uses (e.g., Fedora Linux on my Raptor Talos II uses /dev/ttyUSB0): We start the loader, which you can run as a separate program on its own and it will load anything that does not encroach on its default location at $e000. (We'll find an even better spot for it in just a minute.) Since our ASCII half-terminal program loads and executes from $8000, this is no problem; it takes up three blocks (there's apparently an off-by-one bug in lsx), so it's a good quick test of the machinery. The loader immediately sends NAK and we're off to the races. RUN you can just hit RETURN on to run a BASIC program. Ta-daaaaa! The Australian, paywalled link). Now we want to make it part of the system. To convert it to "firmware" takes advantage of a specific feature Ben built into the SD card reader for easier updates: the ability to load and run a new system "ROM" directly from the card. The default memory map for the VZ200 puts the system ROMs between $0000 and $3fff, reserving the space from $4000 to $67ff for ROM cartridges, though a cartridge could technically take over any address range above the TOM (that's how the RAM expanders worked, after all), and we'll come back to that point later. For the system to recognize code at $4000 or $6000 as part of a cartridge, a sequence $aa $55 $e7 $18 is required in that order, and execution then starts at $4004 or $6004. As shipped to you Ben's device not only fills in RAM above the TOM, it also fills RAM in from $4000 to $67ff and puts its own code there with that sequence, effectively "slushware" (a la the DECmate II, and we'll use the same term here since it's not really ROM). This code is run by the system ROM on startup like a "cartridge," because that's what it looks like to the system ROM, and this code is what does the RAM test and initializes SD card access. Once the system is up, it then does this: VZDOS.VZ it's loading from the card is "magic." If present, it will be used to temporarily replace the slushware at runtime; for a couple extra seconds spent loading it you don't have to mess around with burning it to the cartridge. But this binary is not signed or checksummed, nor does the load check if it's even a new copy of the slushware — any program will serve as long as the .VZ header loads it to $8000 but the code is written to execute from $4004 (as the onboard image would). The slushware then places a little trampoline copy routine at $a000 and runs that to copy the 8K from $8000 to $9fff to $4000, overwriting the old slushware, and jump into the new one. To make our replacement code useful, we should provide some quality-of-life features. We'll make it autostart into a transfer so that all you have to do is load up the program into your Xmodem sender and reset the VZ200, and after a polite delay it will pull down and run the program automatically. We'll also let you load multiple times if you want instead of immediately trying to execute the current file being transferred. We'll also finally put that routine in memory for the slower but more forgiving memory fill operation, and enable the VZ sticks by default in case we find something that really needs them. But more important than those, we should also let you drop back into BASIC and use the SD card loader normally without having to pop the card out. That requires us to include a copy of the actual VZDOS.VZ which we will embed in our replacement code. This adds an additional complication, because the 2.32 slushware (the most current as of this writing) is already 8068 bytes long minus the .VZ header, leaving us only a little over 2K for our own code. Moreover, if we're over 8192 bytes (and it's inevitable we will be), the loading process will overwrite at least 100 bytes of our code with the trampoline and fail to copy the rest. We'll solve this by immediately copying the remainder as the first step in our binary, and then post-processing the object to yield a new VZDOS.VZ with a 128 byte hole between the first 8K and the last 2K (remember it gets loaded to $8000, so we have plenty of space there). Part of this code will be used to make a jump table entry for our slower fill so the call will stay constant with future updates, if any. That looks like this: Since we are embedding VZDOS but we need to keep all its relative offsets intact, we skip the first 7 bytes and the .VZ header, and run those instructions later just before we jump back into it (if we do). We also do a couple patches so that VZDOS will reload (us) on a reset, but not when we execute the embedded copy, and still use an unmodified 2.32 so that you can see there's nothing up my sleeve. Time for our fill routine. 7001, etc. dec bc ld a,c or b ld a,(slobyte) ; flags kept jr nz,sloclrl ret not in VRAM, then do everything LDIR would and leave the routine with A, HL, DE and BC set as they would be at the end. (We do set the Z flag on exit, but most routines won't care about this.) Then our Invaders example, which was can be patched by overwriting the two instructions at $8caa with ld a,0:call 04008h. xor a and could just nop it.) (I couldn't cursorily find out more about this person), who also did Invaders and a number of other software releases for DSE under contract. except scrolling the attract-mode screen, because unavoidably we'll scroll up the stuck bit. I don't think there's a good general way around that. Fortunately it's purely cosmetic, and most games I converted in this fashion seemed to work fine without any glitches. WORDPRO.VZ can also be placed on the card and run from there as we did previously, though doing so doesn't enable file operations either.) You'll need copies of the actual ROMs, which do circulate. The entirety of the code — really a disguised linker script — looks like this: Remember that Wordpro was first and foremost written for the VZ-300, which has 16K of RAM, so its TOM is much higher ($b7ff). The cartridge, because it has full control of the bus, thus maps its much larger ROM in at $d000-$ffff, with 2K of $d000 also mapped to $6000 with the cartridge header sequence. This echo of the main cartridge ROM is what actually autostarts everything since the system ROM doesn't know to check anywhere else but $4000 and $6000. The same scheme works for the VZ-200, except there is no RAM between $9000 and $bfff. But with the BennVenn cartridge, we have RAM everywhere, so we load to $8000 and copy the Wordpro ROM dumps upon execution to their proper location(s), duplicating $d000-$d7ff to $6000-$67ff like a real one, and jump into the "cartridge" at $6004. This copy operation will destroy BFL-the-slushware, but we can just reset to reload it. Wordpro will get all the RAM it would expect to get on a VZ-300, even on this 4K VZ200. too efficient and constructed a general fill subroutine which lots of things called, the screen clear portion being only one of many. (Darn those efficient little assembly language programmers.) we are closed now!"), one of my favourite Intellivision titles, given a solid conversion as Hamburger Sam also by D&M. Z88 Development Kit. Arkaball by Jason Oakley, an obvious Arkanoid clone, but competently crafted and worth a video. VZ-DOOM, a Claude-written raycast Wolfenstein 3D-style game. It fortunately didn't need patching and "just worked." Your mileage may vary as to whether you think the AI usage is cheating, and this blog has a strict no-AI-article-text policy, but it performs as advertised on this NTSC machine and likewise merits a video. VZ-DOOM uses WASD for motion, comma and period to strafe, E to open and SPACE to shoot. VTech items continued to show up in Dick Smith stores, and VTech did make other Laser computers, these "later Lasers" were not closely related nor compatible with the VZ line, or each other, and most of them were much less successful — with the exception of their Apple II clones. The Laser 3000, fresh from its Summer CES 1983 debut, also made it down under to Dick Smith stores as the Dick Smith Cat. It required real Apple II ROMs in an external cartridge for compatibility, which also made it a target for Apple, fresh off their successful victory against Franklin Computer for using substantial portions of the Apple II ROM in the Franklin Ace 1000. Although VTech was still able to sell it elsewhere and the Apple II ROMs were never integrated into the base machine, Apple instead argued that the mere use of the ROMs was an infringement upon its intellectual property regardless, and successfully blocked further imports to the United States. VTech learned from this just like they learned from EACA, and developed new ROMs that were carefully clean-room reverse-engineered, additionally incorporating a licensed copy of Microsoft BASIC retrofitted to act like Applesoft BASIC. This process provided the legal assurance that not a nybble of Apple's code was even consulted, and VTech used these unencumbered ROMs to create what was introduced at Summer CES 1985 as a redesigned "90% compatible" (per Creative Computing) Laser 3000. In turn, that reworked Laser 3000 was transformed into the 1986 Laser 128, a semi-portable riff on the Apple IIe with an expansion slot and built-in 5.25" floppy (later 3.5"). VTech's work paid off handsomely: reviewers were impressed by its value for money, most software never noticed the difference, and Apple's repeated attempts to prevent its importation and sale all ended in failure. The Laser 128 family's low price and exceptional built-in functionality made them the most widely sold Apple II clones in the United States, finishing on the market as late as 1989 in an upgraded 3.6MHz version with a 3.5" disk drive and over 1MB of RAM. returned to journalism in 1987. Meanwhile, DSE was still selling enough VZ-300s to keep them in this 1987-88 catalogue, and according to Greg Dubois' contacts at Dick Smith even at that late date they continued moving over 100,000 units a year, but in the end it was actually VTech that wanted out: VTech wanted to redirect factory capacity to the Laser PC and wasn't willing to keep producing the older system, and even DSE's offer to double the order couldn't convince them otherwise. Although some software and accessories still appeared in the 1989-90 catalogue, the computer itself did not, and the line disappeared completely by 1991. For a time afterwards DSE sold IBM consumer PCs and even Commodore PC clones in the process of adopting its own DSX PC brand. In the 2000s the company failed to make a successful transition from its mail order origins to the new world of online sales, and despite several attempts to rework its retail presence Woolworths unloaded Dick Smith to Anchorage Capital Partners in 2012 in a controversial deal where much of the sale price was allegedly financed by Anchorage liquidating DSE's own assets. Anchorage took the company public in 2013, netting tens of millions, but the company did not recover and all 363 remaining stores were closed by May 2016. Today the Dick Smith brand lives on solely as a mark of online retailer Kogan, primarily selling consumer electronics. Australia's crap home computer, by golly. Much as Sinclair did in the UK and Commodore in the United States, the Aussie VZ-200 and its successor VZ-300 remain as beloved as they are because they introduced a entire generation of Australians to computers who could never buy one before. While a few importers tried to bring the also-rans to the South Pacific (DSE even had Radofin's undead zombie Aquarius in their 1985 catalogue!), the VZ was there first and in large numbers, over 20,000 VZ-200s alone, becoming the down-under standard against which all subsequent cheapo systems were measured. Indeed, when the desperately dire Tandy MC-10 got in front of Australian Personal Computer in December 1983, reviewer Surya commented that when considering it versus the VZ-200 "the MC-10 does not stand up well to this comparison." Legions of user groups and newsletters sprung up to support it, tinkerers designed all manner of expansions for it, and users wrote and sold their own software for it, ironically spawning exactly the sort of hobbyist-driven computer ecosystem post-Dick Smith that Dick Smith-era Dick Smith had previously tried to foster. Ultimately the little Hong Kong desk wedge became more of an Australian computer icon than even some truly homegrown ones. As for this NTSC VZ200, it should be very possible to clone it because the ROMs are the same as the better-known DSE flavour and it's otherwise all off-the-shelf hardware; moreover, it would be infinitely easier to maintain and repair than the ghastly PCB it's got now. The schematics for the PAL Dick Smiths are widely available and there are even fewer components needed to build an NTSC one. At least one person has already made an NTSC-compatible RC2014 workalike, though that project uses a GAL, and it seems like we could make a more straightforward knockoff just using what the original did — with the exception of the colour encoder, which would be improved and somewhat simplified by using a proper MC1372 instead of the TBA520. I know "Leaded Solder" Mike has his clone CreatiVision, so I look forward to him picking this up as a new challenge. ;) We'll be doing more with this system and particularly the VZ-300, now safely awaiting my next Southern Hemisphere trip, in future articles — along with a recently-acquired PAL CreatiVision of our own, the basis for the Dick Smith Wizzard, which we need to see if we can get up and running (I do like me a 6502 and a 9918). Meanwhile, David "Bushy" Maunder's VZ website can give you all the articles, technical information and software that you can stick on an SD card. are all on Github, including pre-built binaries and ready-to-go Bush Food Loader "slushware" you can use with your own SD card loader, all of which are under the BSD 2-clause license.

3 days ago 1 votes
Anecdotally, programmers dislike "reduce"

In short: from my experience, people like map and filter, but not reduce. I use functions like map and filter all the time. When I put that code up for review, my peers rarely complain. I get plenty of feedback about other decisions, but not about my use of map and filter. I cannot say the same for reduce. Often, when I’ve submitted a patch with reduce inside, I get a comment like, “this part is hard to read.” And I see reduce way less than map, filter, some, and so on. Anecdotally, I have come to believe that programmers don’t like reduce as much. I don’t know why, but I have a few theories: reduce is harder to read. reduce is less familiar. reduce can have worse performance compared to other options. reduce is less elegant in languages I use, like JavaScript, Python, and Swift. In my blissful stint as a Clojure developer, I did not get this feedback. I’m wrong, and I’m seeing a trend that’s not real. I usually just change reduce to something else and move on. Even though I prefer it, I don’t usually care much. But it’s a little social phenomenon I’ve observed, and I thought I’d document it. I’ve also noticed this less recently, possibly because code review is less thorough nowadays. Do you notice this? Do you like reduce? Please tell me.

3 days ago 1 votes
Microcode in Intel's 8087 floating-point chip: the scale instruction

In the 1970s, floating-point arithmetic was a mess. Computer manufacturers had a dozen incompatible arithmetic standards. Moreover, floating-point systems were designed around hardware simplicity rather than mathematical rigor, leading to problems with numerical stability. This changed when Intel introduced the 8087 floating-point coprocessor chip in 1980, designed to be as accurate as possible, even in the corner cases. The 8087 became popular because it could be installed in the IBM PC, making floating-point operations up to 100 times faster in applications ranging from spreadsheets to CAD. But more importantly, the 8087 became the floating-point standard used by most computers today. The 8087 implemented its instructions in complex low-level code called microcode. I'm part of a group, the Opcode Collective, that is reverse-engineering this microcode, and I've recently made some progress. In this post, I examine the microcode for one of the 8087's instructions—FSCALE—and describe how this microcode works. The FSCALE (Floating-point Scale) instruction provides a quick way to scale a number by a power of two, much faster than a multiplication. I figured that FSCALE was a simple, almost trivial instruction that would be straightforward to understand and explain. Spoiler: it is not simple. FSCALE uses over 140 micro-instructions and three levels of subroutine calls to handle many special cases. But the FSCALE microcode illustrates many interesting parts of the 8087, such as the shifter, the adder, and the exponent converter, and also reveals a hidden feature of the 8087, so hopefully you will find it interesting. To explore the microcode, I opened up an 8087 chip and created a high-resolution image with a microscope. The large microcode ROM is in the center, holding the 1648 micro-instructions that control the chip. The microcode engine on the left steps through the microcode, handling jumps and subroutine calls. The bottom half of the chip is the "datapath", the circuitry that performs floating-point calculations; it is split into a 16-bit datapath for the number's exponent and a 64-bit datapath for the number's significand (also known as the fractional part). Die of the Intel 8087 floating-point unit chip, with main functional blocks labeled. The die is 5mm×6mm. Click for a larger image. Zooming in on the bottom part of the chip shows the datapath circuitry; I've highlighted the relevant parts below.1 The exponent ROM holds various constants. The exponent converter is a specialized circuit that examines exponents, detects special values, and converts between exponent formats.2 The shifter is a large component; it allows a 64-bit3 value to be shifted left or right by arbitrary amounts. (I wrote about the 8087's shifter circuitry here.) The adder is the heart of the 8087's calculations; it is used in a loop for multiplication, division, and square roots. The B register holds one input to the adder, while multiple sources can provide the other input. The sum register holds the adder's output. The eight stack registers and the temporary registers hold floating-point numbers. A close-up of the 8087's datapath, showing functional blocks that are used by FSCALE. Details of the 8087 In this section, I'll explain some features of the 8087 that are important for the FSCALE microcode. To use the 8087, a programmer stores values in its eight internal registers, organized as a stack. Each register holds an 80-bit floating-point number. To optimize performance, each value in the register stack has an associated "tag" value, which is mostly invisible to the programmer.4 A tag labels a value as valid, special, zero, or empty. A "normal" floating-point value is tagged as valid. If the floating-point value is infinity, Not a Number (NaN), or a denormalized value, then it is tagged as special. A zero value is tagged as zero. Finally, if a register is empty (e.g., its value has been popped off the stack), the register is tagged as empty. The 8087 also has temporary registers that it uses internally: tmpA, tmpB, and tmpC. Like the stack registers, tmpA and tmpB are 80-bit registers, along with two tag bits. However, tmpC only holds a 64-bit significand. The 8087 supports a variety of data types: floating-point numbers of various sizes, integers, and binary-coded decimal. But internally, everything is stored as an 80-bit floating-point number called a "temporary real"; for the rest of this article, I'll only be considering temporary real values. A number has three parts: the sign bit, the 15-bit exponent, and the 64-bit significand (the fractional part), In most cases, a floating-point number is represented by sign × significand × 2exponent. The significand is a 64-bit binary number of the form 1.bbb...: a leading 1, followed by the binary point (the binary equivalent of the decimal point) and the rest of the bits.5 What makes floating-point numbers useful is that their scope covers the incredibly small to the astronomically large, thanks to the exponent, which ranges from -16382 to 16383. One important detail is that the exponent is stored with a "bias" of 16383 added to it. Thus, the stored exponent is always positive, even if the real exponent is negative.6 The 80-bit temporary real format. The triangle indicates the binary point, analogous to the decimal point. From the Intel Numerics Supplement. The 8087 supports several types of numbers that are represented as special cases with special exponents, as shown below. Zero and infinity have both positive and negative values. "Not a Number" (NaN) represents values that don't make sense, such as 0/0 or sqrt(-1); NaN has a large number of representations, not a single value. The 8087 also supports denormalized and unnormalized values, which are extremely small values where the significand doesn't have a leading 1. The encoding of special values. Based on Table S-31 in the Intel Numerics Supplement, but highly simplified. The "x" bits are arbitrary, as long as they don't conflict with another type. The 8087 has a complicated exception system with six types of exceptions to indicate if something went wrong with an arithmetic operation. The most serious is the "invalid operation", indicating that the operation does not make sense, such as 0/0 or ∞-∞. It also includes accesses to an empty register (stack overflow or underflow) or operations on a NaN value. The 8087 also has an overflow exception if a value is too large to store, an underflow exception if a value is too small, and a divide-by-zero exception (excluding 0/0). A denormalized operand exception indicates that the result is too small to store as a normal value, but can be stored as a denormalized value. Finally, a precision exception indicates that a value cannot be represented exactly and must be rounded. (Precision exceptions are very common; even 1/10 will yield one.) The 8087 provides fine-grain control over each exception type, specified by bits in the control register. If an exception is unmasked, the 8087 sends an interrupt to the 8086 processor, which handles the problem in software, for instance by terminating the program or logging an error. Alternatively, the exception can be masked and the 8087 will continue execution as best it can. For instance, an invalid result will be replaced by NaN, while an overflow or divide-by-zero will be replaced by infinity. A precision exception will result in rounding. The point of masked exceptions is that calculations continue, yielding an answer that is as accurate as possible; in most cases, this is what the programmer wants. These features make the 8087 flexible and provide accuracy, but they also make the microcode much more complicated, since the combinations of special cases need to be handled appropriately. The 8087's microcode Executing an 8087 instruction can require hundreds of internal steps to compute the result. These steps are implemented in microcode with micro-instructions that specify each step of the algorithm. (Keep in mind the two levels of instructions: the assembly language instructions used by a programmer and the undocumented low-level micro-instructions inside the chip.) The microcode ROM holds the 1648 micro-instructions that implement the 8087's instruction set. I'm working with the Opcode Collective to reverse-engineer the micro-instructions and fully understand the microcode (link). The 8087's micro-instructions are complicated, with many corner cases and ad hoc functions, but I'll provide a simplified overview. Each micro-instruction consists of 16 bits, as shown below. The first three bits specify the micro-instruction's type, which controls the meaning of the remaining bits. The first type is a transfer operation, which transfers data from one internal register to another. The two fields specify the source and destination. The three remaining bits are used for various special cases. Next is a shift operation, which uses the barrel shifter to shift a value left or right. The third type of micro-instruction controls the adder (which can also subtract). The miscellaneous instructions include stack pointer operations, tag modification, exceptions, and subroutine return. The far jump and far call micro-instructions perform a jump or subroutine call to a target micro-address in a fixed list. The condition field allows conditional jumps/calls/returns based on numerous conditions, while the last bit inverts the condition. A local jump is a relative jump to a nearby micro-instruction. Structure of an 8087 micro-instruction. The FSCALE microcode When the 8087 starts executing an instruction, the instruction decoder circuitry determines the starting address of the microcode corresponding to the instruction. This 11-bit address is loaded into the microcode engine, which starts executing the microcode.7 The microcode for FSCALE (shown below) starts at decimal address 748.8 The idea behind FSCALE is straightforward: if you want to scale a floating-point number by 2N (for an integer N), you add N to the number's exponent. This allows you to multiply or divide by a power of two much faster than using the full floating-point multiplication operation. However, the microcode for FSCALE is unexpectedly complicated and uses several microcode subroutines. In brief, the microcode first checks for arguments that are zero and then handles other special arguments. It converts the scale argument to an integer and adds it to the exponent. Finally, it handles any overflow or underflow. In more detail, the microcode routine starts by moving the first argument from the top of the stack (st(0)) to the tmpA temporary register. If the argument is zero, the routine immediately returns. (Thus, scaling 0 by anything—even NaN—will give a result of 0.) Next, the second value on the stack (the second argument) is moved to the tmpB temporary register. Likewise, the code returns if this value is 0, so scaling anything by 0 leaves the value unchanged.9 Next, a constant value is selected; selecting a constant and using it are two separate micro-instructions. (The 8087 has separate ROMs for 16-bit exponent constants and 67-bit significand constants; this one is an exponent constant.) In the normal case, execution jumps to address #0763, skipping the call to subroutine SPECIAL_TMPS. FSCALE: #0748 st(0) -> tmpA Input argument from top of stack #0749 jmp #0776 if tmpA:tag ZERO Bail if 0 #0750 stackPtr++ #0751 st(0) -> tmpB Scale argument from stack(1) #0752 stackPtr-- #0753 jmp #0776 if tmpB:tag ZERO Bail if 0 #0754 expconst 0x403e Const 403e: exp shift to convert to int #0755 jmp #0763 if not tmp empty/special/div #0756 call SPECIAL_TMPS Special handling #0757 jmp #0762 if flag #0758 jmp #0761 if not tmpB:tag SPECIAL #0759 except:invalid Invalid exception, use NaN #0760 NaN -> tmpA #0761 jmp #0776 if intr #0762 jmp #0775 if expConv[0] Return tmpA if expConv set, otherwise continue #0763 tmpB:exp -> Breg Normal path #0764 tmpB:sign,exp -> expConv ExpConv will test tmpB's sign #0765 expConst -> tmpC Const 403e #0766 adder: tmpC - Breg cin=1 403e-exp is amount to shift to convert tmpB to int #0767 sumreg:frac -> shiftcount Store in shifter control #0768 shift tmpB:frac R count byte bit Perform the shift #0769 shift R -> Breg Breg holds scale argument as an int #0770 jmp #0777 if neg Negative Breg needs separate handling #0771 adder: tmpA:exp + Breg cin=0 Add the scale to the exponent #0772 sumreg:frac -> expConv Put result in expConv to check #0773 sumreg:frac -> tmpA:exp Update exponent with sum #0774 call NONNORMAL_RESULT if not exp normal Handle overflow/underflow #0775 tmpA -> st(0) Save result back to stack #0776 RNI Done: Run Next Instruction #0777 adder: tmpA:exp - Breg cin=1 Subtract Breg #0778 jmp #0772 Continue processing Continuing at #0763, the second argument is converted from a float to an integer, which takes a few steps. For example, suppose the argument is 9, which in floating point is 1.001×23. The significand bits 1000 are "left justified", but for an integer, these bits need to be "right justified" by shifting them to the right. In general, if the exponent is n, the significand is shifted right by 63-n bits. But recall that the exponent is biased by 16383. Thus, the significand must be shifted right by 63-(exp-16383) bits, that is 0x403e-exp bits. (This explains the constant 0x403e earlier in the microcode.) Converting a float to an int by shifting. In the microcode, the subtraction takes several steps. At #0763, the exponent of the second argument is moved to the B register, one of the inputs to the adder (completely different from tmpB).10 Next, the sign and exponent are moved to the exponent converter, a circuit that, among other things, tests for overflow. Next, the constant 0x403e (selected back at #0754) is moved to the tmpC register. At #0766, the adder is activated, subtracting the exponent from the constant.11 The adder puts the result into the sum register, and this value is copied to the shift count register, which controls the shifter. This value indicates how many bits the second argument must be shifted to convert it to an integer. At #0768, the shifter is activated to shift by the desired amount, using both the bit shift part and the byte shift part. As with the adder, activating the shifter and reading the result are separate micro-instructions; the result is put into the B register. The core part of the FSCALE instruction is finally performed at #0771, adding the second argument to the first argument's exponent. The adder is activated to add the B register value (the scale) to the exponent, and the updated value is stored in tmpA's exponent. (Except if the scale factor is negative, it is subtracted via the #0777 path.)12 The value is also sent to the exponent converter circuit, which checks the exponent for overflow or underflow; if so, subroutine NONNORMAL_RESULT is called. But in the normal case, the updated value is copied from tmpA to the top-of-stack register st(0). Finally, RNI (Run Next Instruction) indicates that the microcode routine is done and the instruction is completed. Thus, even in the straightforward case, FSCALE takes about 22 micro-instructions. Handling empty or special arguments What happens if an argument accesses an empty stack location (i.e. stack underflow) or is a special value (infinity, denorm, NaN)? These cases are handled by a micro-subroutine that I'll call SPECIAL_TMPS15 because it processes special values in tmpA and/or tmpB. This subroutine is a general-purpose routine, used by basic arithmetic operations, FSCALE, FTST (test), and FPREM (partial remainder). The control flow through SPECIAL_TMPS is rather convoluted since the code must prioritize issues if, say, one argument is empty and the other is a denorm. I'll just give a brief summary; see the footnote13 for details. First, the subroutine converts any denorms to unnorms. Then it checks for access to empty stack locations, raising an exception or interrupt if so. Then it checks the two arguments again. If either is NaN, an exception or interrupt is triggered. Otherwise, it returns a status indicating the type of arguments. Unexpectedly, if both arguments are NaN, the code compares the two NaN values and returns the larger. This behavior may seem very weird, but it's a documented feature.14 You might think that NaN is a single value, but it's actually an enormous family of values. The idea was that the programmer could use different NaN values to signal where a problem occurs. For instance, you could put a different NaN in each location of an uninitialized array, so you could tell which position was accessed. For some reason, the designers of the 8087 decided that if you perform an operation with two different NaNs, the result is the larger one. Thus, the microcode needs code that detects if both operands are NaN and computes the larger, using a subtraction for the comparison (#1518). SPECIAL_TMPS (J5): #1484 call SPECIAL_VAL if tmpA:tag SPECIAL Handle special values in tmpA/tmpB #1485 xchg tmp #1486 call SPECIAL_VAL if tmpA:tag SPECIAL Handle tmpB special #1487 xchg tmp #1488 1 -> flag Flag=1 by default #1489 jmp #1500 if not tmp empty/special/div 0 -> expConv if tmps okay #1490 1 -> expConv #1491 jmp #1497 if not tmpA/B empty #1492 except:invalid Invalid if either empty #1493 jmp #1525 if compare instruction No NaN for comparison #1494 jmp #1511 if intr Return if interrupt not masked #1495 NaN -> tmpA NaN if interrupt masked #1496 return #1497 jmp #1502 if tmpA:tag SPECIAL Special cases #1498 jmp #1505 if tmpB:tag SPECIAL #1499 0 -> flag Div normal path: #1500 zero -> expConv Return flag 0, expConv 0 #1501 return #1502 call SPECIAL_VAL TmpA special #1503 jmp #1512 if not flag Jump if NaN, fallthrough if infinity #1504 jmp #1509 if not tmpB:tag SPECIAL #1505 xchg tmp TmpB special #1506 call SPECIAL_VAL #1507 xchg tmp #1508 jmp #1521 if not flag Jump if NaN, return if infinity #1509 0 -> flag Clear flag, return #1510 return #1511 RNI End instruction with interrupt #1512 jmp #1522 if not tmpB:tag SPECIAL TmpA NaN, now check tmpB #1513 xchg tmp #1514 call SPECIAL_VAL Check tmpB #1515 xchg tmp #1516 jmp #1522 if flag Jump if tmpB is not NaN #1517 except:invalid Invalid exception #1518 tmpB:frac -> Breg Both args are NaN, find larger #1519 adder: tmpA:frac - Breg cin=1 #1520 jmp #1522 if adder sign See if tmpA #1521 tmpB -> tmpA Take larger #1522 except:invalid Invalid exception #1523 jmp #1525 if compare instruction No interrupt for comparison instruction #1524 jmp #1511 if intr End instruction with interrupt #1525 1 -> flag Return with flag set #1526 return End of J5 This subroutine makes heavy use of a helper subroutine, SPECIAL_VAL,16 that processes one argument. The helper converts a denormalized argument to an unnormalized argument, raising an exception or interrupt as appropriate. It also flags an input of infinity. The hardware for the micro-instruction that exchanges tmpA and tmpB at #1485 is interesting. Instead of physically moving the values between the two registers, the micro-instruction toggles a flip-flop that exchanges the meaning of tmpA and tmpB. That is, if the flip-flop is set, a reference to tmpA goes to tmpB and vice versa. (This is a standard trick in microprocessors; the Intel 8080's XCHG instruction exchanges the DE and HL registers in a similar way. The Z80 uses the same trick for the EX and EXX instructions to exchange the regular register set with the secondary register set.) The Intel 8087 chip is packaged in a 40-pin DIP (dual in-line package), as are the 8080 and Z80. This photo is here as a break from all the microcode. Handling a non-normal result If you take a very large number and scale it larger, you can end up with overflow. If you take a very small number and scale it smaller, you can end up with a denormalized number or underflow. This will trigger an overflow, denorm, or underflow excaption, and an interrupt if unmasked. Moreover, the 8087 supports four rounding modes: round to nearest valid value, round down (toward -∞), round up (toward +∞), or round (chop) toward zero. Depending on the rounding mode, an overflow can result in either ∞ or the largest possible floating-point number. Similarly, an underflow can result in either zero or the smallest possible floating-point number. And depending on the infinity mode (affine or projective), infinity can be either signed or unsigned. Thus, the FSCALE microcode needs to handle many special cases for the result. The subroutine to handle a non-normal result in tmpA is below. One interesting micro-instruction is update overflow/underflow exceptions, which triggers an exception if appropriate. For most exceptions, a micro-instruction triggers the exception (for example, except:precision at #0346). But for the overflow and underflow exceptions, the microcode delegates the task to hardware. Specifically, the 8087's "exponent converter" circuit examines the exponent to see if an overflow or underflow exists, based on the selected floating-point precision. The micro-instruction sets the overflow and underflow flags based on these values. Thus, a complex task is performed by a single microcode instruction, thanks to the hardware support of the exponent converter. NONNORMAL_RESULT (J16): #0318 return if tmpA:tag ZERO Handle non-normal result #0319 update overflow/underflow exceptions Trigger exceptions if exp conv says to #0320 expconst 0x6000 The interrupt bias constant 0x6000 #0321 jmp #0329 if not intr #0322 expConst -> Breg Interrupt path #0323 jmp #0326 if neg #0324 adder: tmpA:exp + Breg cin=0 Add bias for underflow #0325 jmp #0327 #0326 adder: tmpA:exp - Breg cin=1 Subtract for bias overflow #0327 sumreg:frac -> tmpA:exp New exponent to tmpA #0328 return Interrupt, so done #0329 jmp #0344 if neg Masked exception #0330 tmpA:exp -> Breg Underflow #0331 adder: 1 - Breg cin=1 Amount to shift denormal #0332 call CREATE_DENORM Create a denormal #0333 adder: zero + Breg cin=0, roundmode Add zero to round #0334 call ADJUST_PRECISION Adjust to specified precision #0335 jmp #0340 if Sum register is zero If zero, return +/- zero as appropriate #0336 zero -> tmpA:exp Denorm: exponent is 0 #0337 sumreg:frac -> tmpA:frac Save denorm fraction #0338 special -> tmpA tag Tag denom as special #0339 return #0340 tmpA sign -> sign latch Return +/- zero #0341 zero -> tmpA #0342 sign latch -> tmpA sign #0343 return #0344 NaN/Inf -> tmpA:exp Overflow: maybe return infinity #0345 tmpA:frac -> tmpB:frac Save tmpA frac in tmpB #0346 except:precision Set precision exception #0347 Inf -> tmpA:frac Put infinity in frac #0348 special -> tmpA tag Mark infinity as special #0349 return if not round chop If rounding up, return infinity #0350 1 -> Breg Return max float: adjust down #0351 adder: tmpA:exp - Breg cin=1 #0352 sumreg:frac -> tmpA:exp Exp=7fff-1=7ffe #0353 adder: zero - Breg cin=1 #0354 sumreg:frac -> tmpA:frac Frac 0-1 = ff...ff #0355 norm -> tmpA tag Normal value #0356 return if tmpB:frac[63] Return max float unless unnorm #0357 tmpB:frac -> tmpA:frac Return original tmpA frac #0358 return The 8087 has interesting behavior if an overflow or underflow is unmasked and an interrupt occurs. The idea is to let the interrupt handler know what the exponent should have been. However, the proper value can't be used since it is too big or too small to fit in the exponent field (which is why the exception occurred). The solution is to add or subtract the constant 0x6000, resulting in an exponent that fits. The interrupt handler can subtract or add this constant to get the correct exponent. Lines #0322 to 0328 perform this addition or subtraction. For a masked underflow, a denorm value is created by the subroutine CREATE_DENORM. The value is rounded to the specified precision by ADJUST_PRECISION. Finally, if the value is too small for a denorm, the value +0 or -0 is returned as appropriate. For a masked overflow, the 8087 either returns Infinity or the largest-possible float, depending on the specified rounding mode. Infinity is represented by an exponent of all 1s, and a significand of 1000...; these values are loaded directly onto the bus by transistors. The maximum float, however, is computed: 1 is subtracted from the infinity exponent, and 1 is subtracted from a zero significand. Helper subroutine: creating a denormal One controversial feature of the 8087 is denormals, numbers that are smaller than "regular" floats. Recall that floating-point numbers have a significand with the first bit set to 1. But what happens if you hit the smallest possible exponent and want an even smaller number? The 8087 lets you break the rule that the significand starts with 1, producing smaller numbers known as denormalized numbers or denorms. Denorms significantly extend the range, providing numbers up to a factor of 263 smaller. However, denorms don't have as much precision since the upper bits are "wasted". Moreover, calculations with denorms can be substantially slower because special handling is required. Example of a normal number, reduced by a factor of 8, resulting in a denormal. The diagram above shows a normal number with the minimum possible exponent (-16382, which is 1 after biasing). Dividing the number by 8 (or scaling by -3) creates a denorm since the exponent can't be reduced any further. Instead, the significand is shifted 3 bits to the right. The exponent is replaced with the special value 0, indicating that the number is a denorm. In the 8087, denorms are created by a microcode subroutine that I'll call CREATE_DENORM; it is used by many arithmetic operations, not just FSCALE. This subroutine takes a normal number and a shift amount. By shifting the normal number (as in the example above), it creates a denormalized number. The microcode (below) uses the exponent converter to check if the shift is 64 or more. If so, there will be nothing left after the shift, so zero is returned. Otherwise, the value is shifted to the right and the denorm is stored in the B register. CREATE_DENORM (J20): #0522 sumreg:frac -> expConv Create denorm #0523 sumreg:frac -> shiftcount Number of bits to shift #0524 jmp #0528 if exponent[6:14] == 0 Jump if #0525 zero -> Breg No bits left, use zero #0526 shift tmpA:frac L 0 bytes, 0 bits Run through shifter? #0527 jmp #0532 #0528 shift tmpA:frac R count byte bit Shift right by the specified amount #0529 shift R -> Breg Result to Breg #0530 shift tmpA:frac L ~count byte bit Now shift back for sticky test #0531 NOP Wait for shifter #0532 rounding(h) -> Breg[grs] Store the three rounding bits in the Breg #0533 return But why is the value then shifted to the left (#0530)? The purpose of this is to get the rounding bits. One of the principles of the 8087 is to get rounding correct, which is a lot harder than it seems. In order to decide how to round up a number, you need to keep track of an impossibly large number of bits. For instance, if you calculate 1 + 0 and round up, you get 1. But if you calculate, say, 1 + 2-10000 and round up, you get a float a bit higher than 1. The problem is how do you distinguish the two sums before rounding, without storing thousands of bits? The trick is that the 8087 keeps three bits for use in rounding: the "guard" bit, the "round" bit, and the "sticky" bit. If you consider a "tail" of bits to the right of the significand, the guard bit is the most significant bit of the tail, followed by the round bit. The sticky bit is special: it is the OR of all the remaining bits in the tail, indicating if any of them are 1. Thus, 1 + 2-10000 has the sticky bit set, while 1 + 0 does not, so the two values can be rounded up differently. To generate the sticky bit, the 8087 uses a very large 64-bit NOR gate that tests the tail bits in parallel. A diagram showing how the guard, round, and sticky bits are computed from a right shift. The numbers in this example are different from the previous example. When a number is shifted to the right (e.g., when creating a denormal), bits are lost off the right. To generate the rounding bits, the value is shifted to the left, keeping all the tail bits that will eventually be discarded, and discarding the bits that will be in the final significand. The top two bits go into the guard and round bits, while the remaining bits are ORed together to generate the sticky bit from the rest.17 The diagram above is an example of this process. Suppose the value is being shifted to the right by 4 bits. The tail bits abcd (or at least d) will get lost in the shift. The rounding bits are computed by shifting the original significand to the right by 59 bits (the complement of 4). Bit 62 (a) becomes the new guard bit, bit 61 (b) becomes the new round bit, and the OR of the remaining 64 bits becomes the new sticky bit. (Note that the old guard, round, and sticky bits get ORed in too, so they aren't lost.) Merging the significand from the first shift with the rounding bits from the second shift produces the desired result. Helper subroutine: adjusting precision Although the 8087 supports three lengths of floats, it performs all calculations with 80-bit "temporary reals". At the end of an instruction, it converts the result to the desired length. (As a consequence, most instructions aren't any faster if you use a shorter float.) A microcode subroutine, which I call ADJUST_PRECISION, converts the result to the precision that is specified in the 8087's control word, using the specified rounding mode. This subroutine is used by most of the arithmetic instructions. The 8087 supports three types of real numbers. From the Intel Numerics Supplement. The first code path handles temporary reals (which have 64 bits of precision). The control word specifies one of four rounding modes. However, there are only two actions that can be taken for a particular significand: either round down (chop) or round up (chop and increment by 1). This decision is made by complicated logic circuits that examine the rounding bits, the rounding mode, and the sign to determine whether to round up or down. This simplifies the microcode but makes the hardware more complicated. The microcode performs a conditional return, returning if the significand doesn't need to be rounded up. Otherwise, the microcode increments the significand by adding 0 with a carry-in. It then checks for overflow, in which case it replaces the value with Infinity and sets a special flag.18 ADJUST_PRECISION (J11): #0299 jmp #0306 if not precision64 #0300 return if not round up, update CC1 Update condition code, maybe return #0301 adder: sumreg:frac + 0 cin=1 Add 1 to round up #0302 return if not sumreg[64] #0303 Inf -> sumreg:frac,sign Return infinity if overflow #0304 2count++ Set special flag #0305 return #0306 23/52 -> shiftcount Short or long real: get appropriate shift #0307 shift sumreg:frac,rnd L count byte bit sticky Shift to generate rounding bits #0308 NOP Wait for shifter to complete #0309 rounding(H) -> sumreg[grs] Store rounding bits #0310 shift sumreg:frac R ~count byte bit Shift right to drop excess bits #0311 shift R -> sumreg:frac #0312 jmp #0314 if not round up, update CC1 Update condition code #0313 adder: sumreg:frac + 0 cin=1 Round up if appropriate #0314 shift sumreg:frac L ~count byte bit Shift left to realign #0315 shift L -> sumreg:frac,sign #0316 return if not sumreg[64] Return if not overflow #0317 jmp #0303 Return infinity The code is more complicated when returning a smaller precision (short real or long real), since the significand must be shortened. First, the code at #0306 loads the shifter with either 23 or 52, depending on the precision specified in the control word, and then shifts the value left. This produces the rounding bits as in the previous section. Next, the value is shifted to the right, shortening it to the desired length. As before, the significand is incremented or not, depending on whether it should be rounded up or not. Finally, the value is shifted back to the left, so the most significant bit of the significand is on the left. As before, if rounding up caused an overflow, infinity is returned. One bizarre feature is that a jump with the "round up" conditional also has a side effect of updating the 8087's programmer-visible condition code register (CC1), indicating if the result was rounded up or down. That is, the 8087 has extra circuitry to detect this specific condition and load the value into the condition code latch. Strangely, the 8087 documentation doesn't describe this condition code action; Intel didn't document it until the 387SX floating-point chip in 1987.19 Conclusions Floating-point has a long history before the 8087. For instance, the IBM System/360 mainframes (1964) supported 32-bit and 64-bit floating-point numbers. In 1977, AMD introduced the Am9511 floating-point chip, supporting 16- and 32-bit floating-point numbers, along with transcendental functions. What made the 8087 revolutionary is that it was carefully designed to be as mathematically accurate as possible, largely thanks to numerical expert William Kahan. (The 8087 led to the IEEE 754 Standard, now used by almost every computer and ending the anarchy of incompatible floating-point standards.) The 8087 ended up extraordinarily complicated with three different sizes of floating-point numbers, four sizes of integers, four rounding modes, infinity modes, a collection of exceptions that could be masked or unmasked, denormalized and unnormalized numbers, signed and unsigned infinities, signed zeros, and a whole family of Not-a-Numbers. These features combine, yielding many corner cases. The 8087 deals with this complexity both through specialized circuits and through tangled microcode. How complicated is the 8087? For users who didn't have an 8087 chip, Intel sold an 8087 Support Library that exactly emulated the 8087's instructions (but much slower). The emulator took 16K bytes of 8086 code, which was a lot when a full BASIC interpreter could fit in 8K. Another way of looking at this is that the hardware of the 8087 drastically reduced the amount of software required: the 8087 itself used 3.3K of microcode, compared to the 16K for the emulator in 8086 code. I plan to continue reverse-engineering the 8087 microcode; for updates, follow me on Bluesky (@righto.com), Mastodon (@[email protected]), or RSS. I've been working on this with the members of the "Opcode Collective", especially Smartest Blob and Gloriouscow, who converted the ROM images to microcode data and extensively analyzed the contents. See the 8087 repository on GitHub for more. Notes and references The 8087 patents provide some details on the hardware, but unfortunately not the microcode. The patent diagram below shows the architecture of the 8087; I've highlighted the relevant parts. The fraction bus and exponent bus are shown in red. The adder and associated registers are in yellow. (For subtraction, the B register selector selects the complement.) The shifter is in green. The exponent constant ROM and the exponent converter are in orange. The temporary registers and stack registers are in blue. The architecture of the 8087. Based on the patent. Click this image (or any other) to magnify.  ↩ The exponent converter is surprisingly complicated because the 8087 has three different formats for floating-point numbers with three different sizes of exponent fields (8 bits, 11 bits, and 15 bits). Moreover, the different sizes of exponents are stored with different biases. Thus, converting between different sizes of exponents is not trivial. The exponent converter also recognizes overflow and underflow for the different exponent sizes, as well as special values such as infinity and NaN. I plan to describe the exponent converter in more detail later. ↩ The significand in the 8087 is nominally 64 bits wide. However, the 8087 uses three extra low-order bits for rounding, called Guard, Round, and Sticky. These bits ensure that a value is always rounded in the right direction. Some parts of the datapath have additional bits for sign or overflow: the shifter is 68 bits wide, and the adder is 69 bits wide. For the most part, I'll ignore these extra bits and refer to the datapath as 64 bits wide. ↩ Tags are normally invisible to the programmer, but can be accessed through special operations. Specifically, a programmer can dump the 8087's state to memory; the tags are stored in a 16-bit "tag word". ↩ The external representations of floating-point numbers have an implied leading one, with only the bits after the binary point explicitly stored. This provides one additional bit of resolution "for free". The internal 80-bit representation, however, has an explicit leading one to simplify calculations. ↩ One reason that the exponents are biased is that to find the larger of two floating-point numbers, you can compare them lexicographically as signed integers, rather than needing to examine the exponents separately. ↩ Most of the 8087's instructions are implemented in microcode, but a few are hard-wired. For more details on instruction decoding, see Instruction decoding in the Intel 8087 floating-point chip. ↩ I use decimal addresses for the microcode because the Opcode Collective started using decimal addresses, and it would be confusing to change now. ↩ The microcode shows that scaling 0 by anything, or scaling anything by 0, leaves the value unchanged. My view is that the designers took a shortcut here, rather than returning the "right" value. Since the 8087 defines 0×∞ as NaN, it seems to me that 0×2∞ should also be NaN, so FSCALE(0, ∞) should be NaN, not 0. The designers probably made the valid decision that nobody really cared about corner cases on the obscure FSCALE instruction. For other instructions, the behavior with denormals, unnormals, and zeros is documented (tables S-24 to S-26 in the Numerics Supplement documentation), but FSCALE is omitted. ↩ The 8087 has separate buses for the exponent and the significand, and the adder is only connected to the significand bus, so how does the exponent get to the adder? The trick is that there is a 16-bit gateway between the exponent bus and the significand bus, so the exponent can be copied over. ↩ I described the 8087's adder here. In brief, subtraction is performed by inverting the B register's value when it is fed into the adder. The carry-in to the adder is set to 1, so this in effect performs a two's-complement subtraction. ↩ Why does the microcode have separate paths to add a positive scale and subtract a negative scale? The reason is that values are stored as a sign bit and an unsigned value, not two's complement like standard integers. As a result, the adder can't perform signed addition directly. Instead, the adder circuitry must be explicitly directed to complement the B register value and perform a subtraction. ↩ This flowchart shows the SPECIAL_TMPS subroutine. The structure of this routine is complicated because paths split off and rejoin. One tricky path is the code to determine if there are 0, 1, or 2 NaN values, and take the maximum NaN if there are two. Another complication is the exception exits, which raise an interrupt if the interrupt is not masked, but not for a comparison instruction. The two return values are returned through flag and expConv. A flowchart for the SPECIAL_TMPS subroutine. Click for a larger version. The actions of SPECIAL_TMPS are summarized below. It returns status through the flag flip-flop and the exponent converter register (expConv). Its actions are: table.status {border-collapse: collapse;} table.status tr:first-child {border-bottom: 1px solid #ccc;} table.status th,td {padding: 0 10px; text-align: center;} table.status th:first-child,td:first-child {border-right: 1px solid #ccc;} InputResultflagexpConv emptyNaN, exception11 NaN(larger) NaN, exception11 infinityinfinity01 denormunnorm10 div abnormalno change00 (The last row signals an abnormal value during division computation; I'm still investigating this.) ↩ Prof. William Kahan, who guided the development of the 8087, was disappointed that some floating-point features were unused because of a vicious circle: the features didn't receive good compiler support, so programmers didn't use the features, so compiler developers claimed a lack of demand for the features and didn't implement support. Using multiple values of NaN to record how and/or where an NaN came into existence was an example of a feature that lacked software support. See Lecture Notes on the Status of IEEE Standard 754 for Binary Floating-Point Arithmetic for a detailed discussion of NaN and other issues. ↩ The 8087 makes heavy use of micro-subroutines, with a 6-level stack for microcode subroutine calls. Microcode jumps and subroutine calls get the address from a jump table. The index from the jump table comes from 6 bits of the micro-instruction. We unimaginatively named the entries in the microcode jump table as J0, J1, and so forth based on the index, but I'm adding more meaningful names as I figure them out. As for the names for micro-instructions, we don't have any information on what names were used by Intel (unlike the 8086). I invented names, influenced by the names in Gloriouscow's disassembly. ↩ A subroutine that I call SPECIAL_VAL handles denormalized values, infinity, and NaN. (This subroutine is primarily used by SPECIAL_TMPS, but is also used by FRNDINT (round to integer) and FSQRT.) First, the subroutine looks at the exponent of tmpA; if the exponent is zero, the value is denormalized. (The value could also be zero, but that was handled earlier.) If so, the denorm exception is set. Comparison instructions such as FCOM handle denorms differently, but I'll ignore that for now. The code at #1576 tests if the denorm triggered an interrupt; if so, the instruction ends with the interrupt. If the interrupt was masked, the code converts the denorm to an unnorm by changing the tag to norm and changing the exponent to 1 (which corresponds to the very negative, smallest valid value because of the exponent bias). The result of the subroutine is returned through a special flag flip-flop. SPECIAL_VAL (J12): #1572 tmpA:exp -> sumreg:frac Handle special value #1573 jmp #1581 if not Sum register is zero Test exp for denorm #1574 except:denorm #1575 jmp #1577 if compare instruction No exception for comparison #1576 jmp #1571 if DE (denormalized) interrupt RNI if exception #1577 norm -> tmpA tag Handle denorm: tag empty? or valid? #1578 1 -> tmpA:exp Change to unnorm #1579 0 -> flag Clear flag #1580 return #1581 shift tmpA:frac L 0 bytes, 1 bits Shift to check if infinity vs NaN #1582 shift L -> sumreg:frac #1583 jmp #1579 if not Sum register is zero Clear flag for NaN #1584 1 -> flag Set flag for infinity #1585 return At #1581, the code checks if the value is infinity or NaN. Interestingly, this test isn't done directly, but by manipulating the value with the shifter. Recall that infinity has a significand of 10...00, while NaN has at least one additional 1 bit. The code shifts the significand one bit to the left; a zero result indicates infinity, while a nonzero result indicates NaN. As before, the result is returned in the flag flip-flop. ↩ The logic to compute the rounding bits is more complicated than described. There are two micro-instructions with slightly different behavior depending on the expConv value, but I won't get into that here. ↩ The ADJUST_PRECISION subroutine appears to return infinity if the significand overflows after rounding up, but I'm not entirely happy with this. For instance, 1.111... should round up to 2, not infinity; the significand overflows, but that's not an overflow of the float. Presumably, this gets fixed somewhere else. ↩ I don't know why Intel failed to document the feature that a condition code indicates whether a value was rounded up or down. The 8087 documentation is very thorough with corner cases; usually, when I find a strange circuit, I can find a line in the documentation that explains why it is there. Maybe the condition code feature was buggy, so it was easier to not document it? Maybe this feature was a hidden trap to catch competitors that copied the chip? (Intel had a secret instruction in the 8086 for this purpose, but NEC's version of the 8086 didn't have it, much to the disappointment of Intel's lawyers.) Maybe Intel wasn't sure if they wanted to support the feature in later versions? (This is why some of the 8085 processor's instructions weren't documented.) For now, it's a mystery. ↩

4 days ago 1 votes
I Am a Flat-Rate Monthly Responsibility Service

Most people, when asked why they do what they do, lie. This isn’t because they’re malicious but it’s because the honest justification for a career is rarely noble. It’s usually a combination of a decent paycheck, tolerable hours, and whatever neurosis you

5 days ago 3 votes
Super ultrawide Niri

A Niri workspace with 7 visible columns. After having used practically the same xmonad configuration for a decade and a half I’ve now modernized my setup with the scrollable-tiling Wayland compositor Niri. It’s been a bit of a struggle to unlearn my old workflow but I’m really growing to love Niri’s scrollable workflow, especially on my new super ultrawide display. Samsung Odyssey Neo G9 G95NC 57” My new 57” single monitor setup. What kicked off my Niri journey was the purchase of a new super ultrawide monitor. I bought the 57” Odyssey Neo G9 as it was the largest monitor I could find. (It’s marketed as a “gaming” display but it’s really an amazing productivity display.) It replaced my old 3-monitor setup: My old 3-monitor setup. The new display is wider so I had to move the speakers around 10–15cm further apart. I was debating whether to replace the center 31.5” monitor or replace all monitors with a single one but I think I made the right choice with the ultrawide. The curvature wasn’t an issue (I’ve come to prefer it) and the extra vertical space the portrait side monitors provided wasn’t as crucial as I thought. I think an ultrawide is worth it just to get rid of the annoying bezels. Small things can be a big thing sometimes. A more dynamic workflow xmonad and Niri are similar yet different. Both automatically lay out windows as you spawn them but xmonad (at least the way I used it) follows a layout algorithm that re-flows using a “master” window and combines the rest of the windows into one space, while Niri lays out windows in columns. The change is subtle but it implies that a new window won’t change the size of other windows. This is very nice if you spawn a lot of short-lived terminals or web browsers like I do and it reduces the amount of manual reshuffling I spend time on. My xmonad workflow was more static than my Niri one. In xmonad I made heavy use of workspaces, mapping ten workspaces mentally to different programs, such as 0 Firefox and 1 terminal logs on the left monitor; 3, 4 and 5 for different Neovim instances on the center monitor; 8 as chat and 9 for music or video on the right monitor. I had no rules to enforce this; it’s an emergent behaviour that served me well for years. With Niri it’s more dynamic. I still use workspaces but they no longer have direct shortcuts, I simply go up/down in the workspace list. Maybe I’ll add them in the future but with 3–4 workspaces that’s not as necessary. I spawn workspaces/windows when I need them and remove them when I’m done. Usually it’s one workspace per project (yes, I’m now one of those who have multiple up at once) with all the related things such as editor, terminals, and browser with docs. I don’t typically utilize the full screen width and I try to keep the things I’m working on in the center, often leaving 10–30% gaps on the sides. Even though I don’t normally use the “endless scrolling” feature of Niri I re-center selected windows all the time so I can look straight ahead as much as possible. Keyboard shortcuts As a fan of keyboard layouts of course I have to spend some time tinkering with good keyboard shortcuts (especially as Niri’s recommended keybinds don’t map well with my custom keyboard or custom layout). Navigation layer What I did was add a new navigation layer that’s enabled by holding Tab (ring + middle + index on the left-hand side) with all Niri related movement and layout keybinds. In the graphics above, all green-colored keys emit Gui (which gates all window manager commands) and you can see: Long press on Close Window to close a window. The long press requirement prevents accidentally closing windows. Arrows move through columns/windows. Long press resizes them. Workspace Up/Down focuses a different workspace. Center a column. Consume/Expel to combine windows into one column. (consume-or-expel-window-left/consume-or-expel-window-right) Expand Column makes a column take up all remaining space. (expand-column-to-available-width) Audio controls. To press them I release the index finger (keeping the ring and middle finger pressed to keep the layer active) and use the index to press the audio buttons. Mouse buttons. In Niri you can move floating windows with Gui + Left Mouse and Gui + Right Mouse to resize them. As my main mouse is a trackball integrated into the keyboard I had to add them to the left-hand side. I ended up using QMK’s customizable key repress feature that allows me to: Tab combo with my three fingers (layer is active) Release only the index (layer is still active, same as with the audio controls) Press the index again (now detects the press Gui + Left Mouse key down) Use the trackball to move the window And similarly for the right mouse button to resize with the middle finger. Works great! Because there are so many commands I want to send I placed Ctrl on the thumb that provides movement-related commands like so: For example: Arrows move columns/windows in the four directions. Move columns to the neighboring workspaces. Center visible columns. (center-visible-columns) Slightly different consume/expel semantics. (consume-window-into-column/expel-window-from-column) Regular keymaps These are triggered in the “normal” way by first pressing the Super combo and then another key on the base layer (I use autoshift so I shift with a long press). Window management Super + F toggle windowed fullscreen (keep column width) Super + Shift + F fullscreen window (over the entire display) Super + M maximize column (moves other columns) Run stuff Super + Enter terminal Super + E Noctalia’s launcher (also exists on the navigation layer as Launch) Super + S show Noctalia control center Super + Shift + S show Noctalia settings Super + Q power off monitors (they wake on input) Super + Shift + Q show Noctalia session menu (reboot etc) Super + Shift + L lock screen Misc Super + H show hotkey overlay Super + P interactive screenshot Super + Shift + P screenshot selected window Tweaks to the standard CachyOS setup In the process of moving from xmonad to Niri I also moved from Void Linux to CachyOS and I let the installer install Niri and give me a basic configuration together with Noctalia (that provides a statusbar, notifications, and a bunch of things you apparently need). Center the status bar and other Noctalia windows My centered Noctalia status bar. Feels absolutely required on this screen otherwise things end up in the corners. Firefox on XWayland Force Firefox onto XWayland as the Wayland popup manager is broken: environment { MOZ_ENABLE_WAYLAND "0" } Dead keys for Ghostty For some reason dead keys were broken in Ghostty. This is bad for me as the OS keyboard is set to Swedish and it uses them to type ~ (quite a crucial character for a programmer). The fix: environment { GTK_IM_MODULE "ibus" QT_IM_MODULE "ibus" XMODIFIERS "@im=ibus" } This needs ibus installed and running. Melange colorscheme Noctalia discovers custom color schemes under ~/.config/noctalia/colorschemes/<Name>/<Name>.json, so I dropped in my trusty Melange colorscheme there: { "dark": { "mPrimary": "#EBC06D", "mOnPrimary": "#292522", "mSecondary": "#A3A9CE", "mOnSecondary": "#292522", "mTertiary": "#85B695", "mOnTertiary": "#292522", "mError": "#D47766", "mOnError": "#292522", "mSurface": "#292522", "mOnSurface": "#ECE1D7", "mSurfaceVariant": "#34302C", "mOnSurfaceVariant": "#C1A78E", "mOutline": "#867462", "mShadow": "#1a1816", "mHover": "#E49B5D", "mOnHover": "#292522", "terminal": { "normal": { "black": "#867462", "red": "#D47766", "green": "#85B695", "yellow": "#EBC06D", "blue": "#A3A9CE", "magenta": "#CF9BC2", "cyan": "#89B3B6", "white": "#ECE1D7" }, "bright": { "black": "#34302C", "red": "#BD8183", "green": "#78997A", "yellow": "#E49B5D", "blue": "#7F91B2", "magenta": "#B380B0", "cyan": "#7B9695", "white": "#C1A78E" }, "foreground": "#ECE1D7", "background": "#292522", "selectionFg": "#C1A78E", "selectionBg": "#403A36", "cursorText": "#292522", "cursor": "#EBC06D" } } } Then pick the colorscheme: "colorSchemes": { "darkMode": true, "predefinedScheme": "Melange", "useWallpaperColors": false } Layout appearance The default appearance was pretty I admit but way too much blank space and weirdness. Some tweaks: layout { // Required for noctalia-shell to set wallpaper background-color "transparent" // Never auto-center focused columns (too much movement) center-focused-column "never" // But do center a single window always-center-single-column // No extra space around it all struts {} // No gaps between windows gaps 0 // The focus ring was annoying focus-ring { off } // Use a border with consistent width for all windows instead border { on width 2 active-color "#ebc06d" inactive-color "#403a36" } // Setting widths is important with such a large screen preset-column-widths { proportion 0.15 proportion 0.3 proportion 0.4 } default-column-width { proportion 0.15; } // Heights too, why not? preset-window-heights { proportion 0.15 proportion 0.5 proportion 1.0 } } // Prevent the mouse from opening the overview in the corners gestures { hot-corners { off } } Keep windows centered Niri has the always-center-single-column option, which is nice as I want to keep as much as possible in the center of the monitor when I’m working. But I very frequently use 2–3 smaller windows and with my frequent opening and closing I’d like them centered too. Luckily, Niri has an IPC you can use to make a small program that reacts to events and does this for you. I made a small rust project using the niri-ipc crate that does this for me: The autocenter implementation [dependencies] niri-ipc = "26.4.0" use std::collections::HashMap; use std::io; use niri_ipc::socket::Socket; use niri_ipc::{Action, Event, Request, Response, Window, Workspace}; #[derive(Clone, Copy, PartialEq, Eq, Hash)] struct WindowId(u64); #[derive(Clone, Copy, PartialEq, Eq)] struct WorkspaceId(u64); struct OutputName<'a>(&'a str); struct WindowState { workspace: Option<WorkspaceId>, width: f64, } fn main() -> io::Result<()> { let mut socket = Socket::connect()?; if !matches!(socket.send(Request::EventStream)?, Ok(Response::Handled)) { eprintln!("niri rejected event stream"); std::process::exit(1); } let mut known: HashMap<WindowId, WindowState> = HashMap::new(); let mut read_event = socket.read_events(); loop { let result = match read_event()? { // A full snapshot of the current state. Just refresh our state. Event::WindowsChanged { windows } => { known = windows .into_iter() .map(|w| { ( WindowId(w.id), WindowState { workspace: w.workspace_id.map(WorkspaceId), width: w.layout.tile_size.0, }, ) }) .collect(); Ok(()) } Event::WindowOpenedOrChanged { window } => { let workspace = window.workspace_id.map(WorkspaceId); let entry = WindowState { workspace, width: window.layout.tile_size.0, }; let prev = known.insert(WindowId(window.id), entry); match prev { // Don't center floats. _ if window.is_floating => Ok(()), // New window, try to re-center. None => maybe_center_new(&window), // Window changed workspace, try to re-center. Some(state) if state.workspace != workspace => center_focused_if_fits(), // Skip other things. Some(_) => Ok(()), } } Event::WindowClosed { id } => { if known.remove(&WindowId(id)).is_some() { center_focused_if_fits() } else { Ok(()) } } Event::WindowLayoutsChanged { changes } => { // Only re-center if the width was changed, otherwise our re-center will // loop back indefinitely. let mut resized = false; for (id, layout) in changes { if let Some(state) = known.get_mut(&WindowId(id)) { if (state.width - layout.tile_size.0).abs() > 0.5 { state.width = layout.tile_size.0; resized = true; } } } if resized { center_focused_if_fits() } else { Ok(()) } } _ => Ok(()), }; if let Err(e) = result { eprintln!("autocenter: {e}"); } } } /// Center a newly created window if the workspace is focused and if there's surrounding free space left. fn maybe_center_new(window: &Window) -> io::Result<()> { let Some(workspace_id) = window.workspace_id.map(WorkspaceId) else { return Ok(()); }; let Some(focused) = focused_workspace()? else { return Ok(()); }; if WorkspaceId(focused.id) == workspace_id { center_if_fits(&focused)?; } Ok(()) } /// Center windows in the focused workspace if there's surrounding free space left. fn center_focused_if_fits() -> io::Result<()> { if let Some(focused) = focused_workspace()? { center_if_fits(&focused)?; } Ok(()) } /// Center windows in the workspace if there's surrounding free space left. fn center_if_fits(workspace: &Workspace) -> io::Result<()> { let Some(output) = workspace.output.as_deref().map(OutputName) else { return Ok(()); }; let Some(width) = output_width(output)? else { return Ok(()); }; if workspace_width(WorkspaceId(workspace.id))? < f64::from(width) { center_visible_columns()?; } Ok(()) } /// Issue a one-shot query to Niri, wait, and return the response. fn query(request: Request) -> io::Result<Response> { match Socket::connect()?.send(request)? { Ok(response) => Ok(response), Err(msg) => Err(io::Error::other(msg)), } } /// Get the focused workspace. fn focused_workspace() -> io::Result<Option<Workspace>> { match query(Request::Workspaces)? { Response::Workspaces(ws) => Ok(ws.into_iter().find(|w| w.is_focused)), _ => Ok(None), } } /// Get the width of an output (monitor). fn output_width(name: OutputName<'_>) -> io::Result<Option<u32>> { match query(Request::Outputs)? { Response::Outputs(outputs) => Ok(outputs .get(name.0) .and_then(|o| o.logical.as_ref()) .map(|l| l.width)), _ => Ok(None), } } /// Calculates the width of all columns in the workspace. fn workspace_width(workspace_id: WorkspaceId) -> io::Result<f64> { let Response::Windows(windows) = query(Request::Windows)? else { return Ok(0.0); }; let mut columns: HashMap<usize, f64> = HashMap::new(); for w in windows { if w.workspace_id.map(WorkspaceId) != Some(workspace_id) { continue; } if let Some((col, _)) = w.layout.pos_in_scrolling_layout { let width = columns.entry(col).or_insert(0.0); *width = width.max(w.layout.tile_size.0); } } Ok(columns.values().sum()) } /// Send a command to center the visible columns. fn center_visible_columns() -> io::Result<()> { if let Err(msg) = Socket::connect()?.send(Request::Action(Action::CenterVisibleColumns {}))? { eprintln!("center-visible-columns rejected: {msg}"); } Ok(()) } One catch is that if a new window overflows the monitor width, the script won’t center the columns even if there would be free space left afterwards. This is a little weird but it’s consistent with Niri’s center-visible-columns command. I had a small itch to try to hack around it but in the end I left it alone… Is Niri worth it? Yes, absolutely. Niri has been a huge upgrade for me in combination with a single wide screen. My xmonad setup worked really well with three monitors—arguably a better fit in that context than Niri—but for the big-screen use-case Niri is superior. I’m curious how it holds up on my laptop, once I gather enough energy to install CachyOS on it… But that’s a side quest. The big-screen setup I spend most of my days in is the best I’ve ever had, and I have no desire to go back.

5 days ago 2 votes
📚 BoredReading

You seem to be enjoying this.

Join free to unlock everything.

Create free account

Already have an account? Sign in