In principle the files on the CP/M disks can be extracted using cpmtools. This, however, requires knowing the disk definitions.
I was able to work out the skew table by looking at the directory sectors on track 2. I guessed the directory size for the 2.2 format.
There were actually two different CP/M formats for the two different CP/M implementations on the disks: Exidy CP/M 1.4 and Software Source CP/M 2.2.
I was able to make some guesses based on the sector arrangement, and when i looked at the images i could see the order of the directory sectors. They both use a skew of 5 but start with different sectors – which explains why each CP/M only works with their own disks.
The directory size was a guess.
I thought i had nailed the format, but later i realised that there were some issues.
The two CP/Ms that I have use 2kB blocks. Each directory extent can accommodate 16 block pointers, but they are not all used. I think there was a rule that an extent could only represent a maximum of 16kB (multiple extents can be used for bigger files) so the CP/Ms that I have only used 8 pointers and just waste off the other 8.
With the original diskdefs cpmtools would read fine but when writing it would use up to 16 of the pointers. This befuddled CP/M.
The fix was not obvious but I found a comment on one of the other diskdefs suggesting that
logicalextents 1
wasted half of the entries (which is what I want) so I gave that a try and it worked.
My Sorcerer uses a Micropolis hard sectored floppy drive system. It uses 16 sectors, each of which includes a 256 byte data section and 19 bytes of bonus “metadata” used by Micropolis DOS (MDOS). This metadata is ignored by the CP/M implementations (Exidy & Software Source) that i have encountered.
The drives have an unusual track spacing of 100TPI. This is unusual so imaging has to be done using the actual drive units.
FluxEngine is a very low cost imaging solution that support Micropolis hard sector disks as used on my Sorcerer. At the time, Greaseweazle did not support hard sectors. It has since been added. I’ll do a separate post on that.
I connected to the drive using a straight through cable and using drive:0. This selects the second drive in the dual drive unit (drive 1). For the single drive unit, DS1 (the second drive select) must be selected.
I used FluxEngine on a new PC, so there would be no USB bandwidth issues. My older machine does seem to work reliably, but if it gets busy then FluxEngine will sometimes throw a bandwidth error.
Initially i used my single drive unit (which I call Drive 2) mainly because it was compact. Once I started working my way through the disks, I found I was getting lots of errors. I thought that I might do better with the dual drive unit (I call the top drive “Drive 0” and the bottom drive, “Drive 1”). I found that Drive 0 was a little better than Drive 2 and that Drive 1 was the best drive to use.
The results were a little disappointing but some further reading of the FluxEngine doco:
hinted that I could fiddle with decoder.bit_error_threshold to perhaps get better results. 0.2 is the default. 0.4 improved things a lot. 0.5 was the best. With the wrong setting, a lot of empty (?) sectors were read. I also increased the retry limit.
Drive 1 remained the best performing drive. I tried messing with the alignment on Drive 2, but I could not noticeably improve its performance. I did some poking around with the scope but no obvious differences between the dual and single units leapt out at me.
Out of 87 disks, I was able to read 19 without any errors being reported. I was able to read another 18 with only a small number of errors. A further 38 disks didn’t yield anything usable. Some were shedding oxide, but most 24 were successfully reformatted. Many were unlabelled, so perhaps no great loss.
Disks that were formatted with MDOS were read using the 275byte sector size. CP/M disks were read with 256 byte sector size. My understanding is that CP/M doesn’t use the “metadata”, and the 256 byte size is much easier to view.
There were some examples of where I was able to copy a disk on the real machine, but FluxEngine was unable to read the disk without error, eg Disk 24. The copy read fine. This meant that there was potential to improve the yield by making fresh copies on the real machine. Ie The real disk controller is still doing a better job in real time than the flux engine does in slow time. That’s the advantage of inside knowledge, perhaps.
I attempted to copy all the errant disks to intermediate disks, which I subsequently imaged. I did this for the CP/M disks using COPYDISK. This yielded another 13 good images! I think this confirms that FluxEngine is way less tolerant of problems than the sorcerer system. Even many disks that had a lot of issues came good.
I repeated the exercise for the MDOS disks, which gave another 8 good images.
I was able to write back both CP/M and MDOS images:
The write process verifies each track by reading back so even the writes need the bit_error_threshold parameter.
Writing also work through the virtual sector generator to a soft sector disk. The disk can be read by the system if the virtual sector generator is used.
I was very impressed with FluxEngine. David Given has done an amazing job.
I received about 500 8″ floppy disks and some drives from a friend at the ARC Group (thanks Andrew). I set about archiving them and extracting the files. Alan Laughton helped sort out the disk definitions. There was some interest in how i’d done it so i wrote it up here.
The system has two SASI cards that I thought might accept a SCSI2SD card.
The drive configuration comes up in two places – firstly in configuration of the master or single user system configuration program, and then again when the drive is formatted.
In both cases, the following information is required:
SASI card number: 0 worked for one card but I tried multiple numbers with the other card without success
Drive Number: It allows 1 or 2. 1 seemed to be SCSI ID 0.
The configuration also deals with partitioning. The default partition size is 4MB which is the optimal size. With large drives, that’s a bit of a nuisance because you need a lot of partitions. Having some optimal 4MB partitions and a larger sub-optimal partition seemed like a reasonable compromise.
The drive selection gave some geometry, but the specifics probably don’t matter with a SCSI2SD. The SCSI2SD was set up with a simple 32MB disk at ID 0 with 512B sectors. Termination needs to be on.
The process went like this:
Create a fresh single user floppy disk
Run the Configuration program and select modify
Set up the hard disk as above
Format the hard disk using HFORM30 with the same disk parameters
At this point the new drives were available starting at E: but when the directory was listed it appeared the disk was read only and the directory looked corrupted. It didn’t seem to matter if the format was done first and then the configuration.
The “Creating Boot Tracks” section of the System Initialisation Procedure mentioned a program called ERASEDIR but really just in the context of making faster hashed entries. Running this program on each of the drives resolved the issue. It says to run this after BOOTDISC (which writes the boot tracks).
So:
Run BOOTDISK and write to E: – only the first partition can be a boot partition. It can also be written to A:.
Run ERASEDIR on each of the new drives from e: to the last one.
Copy all the files from the A: to E: using DO DCOPY A: E:
When the system is powered up, it looks for a bootable drive. If a boot floppy is in A: it will use it; otherwise it will boot using E:.
Programs were then copied on to the solid state disk from a gotek. TurboDOS supports multiple user areas so the these can be used as directories. User 0 files marked a global can be accessed by all users.
All users are assumed to be using Televideo 950 terminals. A lot of the software on the 8″ disks was configured to use this popular terminal.
