Discussion Forum for all things Microbee
Fun with a 32mb CF image - Printable Version

+- Discussion Forum for all things Microbee (https://microbeetechnology.com.au/forum)
+-- Forum: Microbee Forum (https://microbeetechnology.com.au/forum/forum-1.html)
+--- Forum: Microbee Software and Documentation (https://microbeetechnology.com.au/forum/forum-7.html)
+--- Thread: Fun with a 32mb CF image (/thread-1014.html)

Fun with a 32mb CF image - Graham72 - 18-04-2026

using Grok and Claude I have managed to get a working 21mb image:

MicroBee 512k CP/M Compact Flash Image — Project Summary

Background
The MicroBee 512k is emulated here under the ubee512 emulator. The system boots from a Compact Flash card image, with the ROM BIOS (version 0.58/cf8) loading a secondary bootloader (BOOT.SYS) which in turn initialises the CP/M environment across multiple disk partitions. The starting point was a working 30MB image with four partitions — A: through D: — each limited to approximately 5MB, leaving roughly 11MB of the image completely unused.

Objective
To safely expand the D: partition to use as much of the unused space as possible, while keeping A:, B: and C: fully intact and bootable.

Investigation

The project required significant detective work before any modifications could be made safely. Analysis of the MBR partition table revealed the image geometry — 123 cylinders, 16 heads, 32 sectors per track — correctly decoded from the filename
Code:
cfboot_0_58_cf8_123_16_32

Early attempts at enlarging D: failed because the CHS fields in the MBR were being calculated against the wrong geometry (16 SPT instead of 32 SPT), producing partition table entries that confused the BIOS.

A key breakthrough came from the PART.COM utility output running inside CP/M, which revealed the BIOS was reporting 32 sectors per track — not 16 as initially assumed. Once the correct geometry was confirmed by reverse-engineering the original CHS values mathematically, partition table entries could be constructed accurately.

Further investigation uncovered a second constraint: an apparent signed 16-bit boundary in the BIOS or BOOT.SYS where any partition whose LBA start address exceeded 0x8000 (32,768) would become inaccessible. This explained why earlier 8MB attempts with incorrectly positioned partitions failed — D:'s start address was being interpreted as a negative number.

Testing Methodology

With the geometry corrected and D:'s LBA start address confirmed safe at 0x7600, a systematic binary search was conducted on D:'s size, testing incrementally in cylinder units (each cylinder = 512 sectors = 256KB):
  • 20 cylinders (5MB) ✓ original
  • 21 cylinders (5.25MB) ✓
  • 24 cylinders (6MB) ✓
  • 25 cylinders (6.25MB) ✓
  • 26 cylinders (6.5MB) ✗
  • 28 cylinders (7MB) ✗
  • 32 cylinders (8MB) ✗

The ceiling of 25 cylinders is consistent with the compiled-in DSM value in the CP/M BIOS Disk Parameter Block — the maximum block number the BIOS was built to address for that partition. Exceeding it causes silent failure without any error message.

Result
The final production image retains the original filename for compatibility, with the following partition layout:
Drive A:4.73MB (unchanged, bootable)
B: 5.00MB (unchanged)
C: 5.00MB (unchanged)
D: 6.25MB (expanded)
Total
~21MB usable

All four drives are accessible, existing data on A: was preserved throughout, and files copied to D: during testing survived correctly. The expansion represents the maximum achievable without recompiling the CP/M BIOS with a larger DSM value — a significantly more complex undertaking.


Cool

RE: Fun with a 32mb CF image - greybeard - 18-04-2026

Output from part for a 1gb cf (o/p on a linux box so I could screen capture) formatted on an actual microbee under cpm
I gave up after creating 23 partitions (small attention span Wink ) ! 6.5M/partition
But you can only mount 4 partitions at a time.

Expert command (m for help): p
Disk /dev/sdc: 1.91 GiB, 2048901120 bytes, 4001760 sectors
Disk model: CompactFlash
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disklabel type: dos
Disk identifier: 0x00000000

Device    Boot  Start    End Sectors Id Type              Start-C/H/S End-C/H/S Attrs
/dev/sdc1  *        63  13103  13041 db CP/M / CTOS / ...      0/1/1  12/15/63    80
/dev/sdc2        13104  26207  13104 db CP/M / CTOS / ...      13/0/1  25/15/63
/dev/sdc3        26208  39311  13104 db CP/M / CTOS / ...      26/0/1  38/15/63
/dev/sdc4        39312 4001759 3962448  5 Extended              39/0/1 897/15/63
/dev/sdc5        39375  52415  13041 db CP/M / CTOS / ...      39/1/1  51/15/63
/dev/sdc6        52479  65519  13041 db CP/M / CTOS / ...      52/1/1  64/15/63
/dev/sdc7        65583  78623  13041 db CP/M / CTOS / ...      65/1/1  77/15/63
/dev/sdc8        78687  91727  13041 db CP/M / CTOS / ...      78/1/1  90/15/63
/dev/sdc9        91791  104831  13041 db CP/M / CTOS / ...      91/1/1 103/15/63
/dev/sdc10      104895  117935  13041 db CP/M / CTOS / ...    104/1/1 116/15/63
/dev/sdc11      117999  131039  13041 db CP/M / CTOS / ...    117/1/1 129/15/63
/dev/sdc12      131103  144143  13041 db CP/M / CTOS / ...    130/1/1 142/15/63
/dev/sdc13      144207  157247  13041 db CP/M / CTOS / ...    143/1/1 155/15/63
/dev/sdc14      157311  170351  13041 db CP/M / CTOS / ...    156/1/1 168/15/63
/dev/sdc15      170415  183455  13041 db CP/M / CTOS / ...    169/1/1 181/15/63
/dev/sdc16      183519  196559  13041 db CP/M / CTOS / ...    182/1/1 194/15/63
/dev/sdc17      196623  209663  13041 db CP/M / CTOS / ...    195/1/1 207/15/63
/dev/sdc18      209727  222767  13041 db CP/M / CTOS / ...    208/1/1 220/15/63
/dev/sdc19      222831  235871  13041 db CP/M / CTOS / ...    221/1/1 233/15/63
/dev/sdc20      235935  248975  13041 db CP/M / CTOS / ...    234/1/1 246/15/63
/dev/sdc21      249039  262079  13041 db CP/M / CTOS / ...    247/1/1 259/15/63
/dev/sdc22      262143  275183  13041 db CP/M / CTOS / ...    260/1/1 272/15/63
/dev/sdc23      275247  288287  13041 db CP/M / CTOS / ...    273/1/1 285/15/63

RE: Fun with a 32mb CF image - Graham72 - 18-04-2026

i'll give that output to Claude. Grok is sulking.

RE: Fun with a 32mb CF image - Graham72 - 18-04-2026

Yes, excellent result.

"Before greybeard's data arrived we had settled on 25 cylinders = 12800 sectors = 6400KB (6.25MB) as our confirmed maximum.
Greybeard's data pushed us to 13104 sectors = 6552KB — a gain of 304 sectors = 152KB over what we had found on our own." ~ Claude.

The limits are real and proven, anything over fails silently.

RE: Fun with a 32mb CF image - shane.crozier - 19-04-2026

(18-04-2026, 07:54 PM)Graham72 Wrote: Yes, excellent result.

"Before greybeard's data arrived we had settled on 25 cylinders = 12800 sectors = 6400KB (6.25MB) as our confirmed maximum.
Greybeard's data pushed us to 13104 sectors = 6552KB — a gain of 304 sectors = 152KB over what we had found on our own." ~ Claude.

The limits are real and proven, anything over fails silently.

Does this mean other CF's could also be used with this understanding of the mechanics?

RE: Fun with a 32mb CF image - Graham72 - 24-04-2026

(19-04-2026, 06:20 PM)shane.crozier Wrote:
(18-04-2026, 07:54 PM)Graham72 Wrote: Yes, excellent result.

"Before greybeard's data arrived we had settled on 25 cylinders = 12800 sectors = 6400KB (6.25MB) as our confirmed maximum.
Greybeard's data pushed us to 13104 sectors = 6552KB — a gain of 304 sectors = 152KB over what we had found on our own." ~ Claude.

The limits are real and proven, anything over fails silently.

Does this mean other CF's could also be used with this understanding of the mechanics?
Copy and paste this thread into Claude and upload 
the CF image and request Claude to analyze and maximize partition space. I'm sure I've read somewhere of an image with 8mb partition D:

Cheers g

RE: Fun with a 32mb CF image - MbeeTech - 24-04-2026

What ever happened to figuring things out yourself?

I know AI can be a great tool, but isn't this stuff hobby driven?
Why would you want to pass stuff off to a bot to do it for you?
You might get stuff done quicker, but have you learned anything from it or
got the chance to say " yeah, I figured it out & got it done ".

RE: Fun with a 32mb CF image - Graham72 - 24-04-2026

(24-04-2026, 06:04 PM)MbeeTech Wrote: What ever happened to figuring things out yourself?

I know AI can be a great tool, but isn't this stuff hobby driven? 
Why would you want to pass stuff off to a bot to do it for you?
You might get stuff done quicker, but have you learned anything from it or
got the chance to say " yeah, I figured it out & got it done ".

I spent days unable to figure it out, Claude helped me over several hours. At present you still have to run the software yourself, create the batch files etc. Claude is a tool that helps analyse the boot sectors, cylinders et al. You have to know the right questions or else gigo still applies. Cheers g
edited:
PS Claude greeted me this morning with a 9 month precis>>


Work context
No clear professional role or workplace context has been established across graham's conversations.
Personal context
Graham has a strong interest in retro computing and nostalgia-driven technology projects, including running historic Windows versions and working with vintage hardware ecosystems. He is technically capable and hands-on, comfortable with low-level tooling, emulators, and system internals.
Top of mind
Graham has been working extensively on MicroBee emulation using ubee512, including deep ROM analysis, Z80 disassembly, CP/M BIOS patching, and disk image creation. A key unresolved thread involves patching the bn56cf8.rom to wire dead-code slave drive logic into the partition initialization chain to enable dual 20MB hard drive images via
Code:
--ide-a1
. Separately, he has been exploring PDF search tooling on his Windows 11 machine (16GB RAM, i5-1340P, 100GB free), with a planned Claude API + AnyTxt hybrid workflow and an unbuilt Windows context menu script to pass PDFs to Claude for Q&A.
Brief history
Recent months
Graham has been working through a broad range of Windows 11 technical issues, including troubleshooting Excel autosave/OneDrive sync problems, File Explorer double-copy behavior (diagnosed to shell extensions, cloud sync reparse points, and MFT issues), PowerShell execution policy and default directory configuration, and RAM compression management. He also set up the Clippy app (local LLM runner) and investigated Facebook monetization requirements, suggesting some interest in content creation. He wrote or requested an analytical essay on the use of the word "tyrant" in contemporary political and casual discourse.
Earlier context
Graham undertook a substantial retro computing session focused on running Windows 3.1, 95, and 98 under emulation on Windows 11, using Chocolatey as his package manager. He successfully installed Windows 3.11 via a locally packed Chocolatey package (win31dosbox), troubleshot the windows95 Electron app (felixrieseberg), and planned VirtualBox setups for Windows 98 and XP Pro, for which he has original CDs and product keys. He also worked on MicroBee BASIC 80-column mode via the ubee512 emulator, establishing that correct 6545 CRTC register values needed to be traced from ROM startup via the ubee512 debugger.<<

I'm glad I use Grok as well.