[kfox]

Part of the GEOS 64 package (#61, #66, #67, #69, #70; #64 is separate). Adds a Berkeley Softworks geoRAM as a Special IO handler. GEOS 64 runs under this handler: the USB mouse (#66) and the TR block device (#69) put their registers in the IO page it leaves free.

• Shared RAM bank code (RAMBanks.c): the REU's 8K bank allocation, image preload and numbered button save move out of IOH_REU.c so GeoRAM can use them. The REU keeps its own bank array and count, and its saves still go to the next free reuNNN, now written in whole 8K banks instead of a byte at a time. A failed write is now reported instead of acked with the LED.
• IOH_GeoRAM:
  • A 256-byte window at $DE00 onto the page selected by the write-only $DFFE (page) and $DFFF (16K block) registers, decoded across $DF80–$DFFF as VICE does. $DF00–$DF7F stays free.
  • 128K to 512K, as much as fits beside a loaded crt.
  • Preloads SD:/georam.ram. The special button saves back to georam.ram through a temporary file, keeping the previous save as georam.bak.
  • No DMA. It is freed along with the REU wherever the menu reclaims RAM.

GeoRAM is the last manually selectable handler (LastSelectableIOH moves to it), so every handler number already saved in EEPROM keeps its meaning.

Tested on a TR+ in an NTSC C64, and in PAL through a VIC-II Kawari:

• Button save replaces georam.ram and keeps the previous one as georam.bak; a first-ever save creates georam.ram; with only georam.bak and a stale georam.tmp on the card, .bak survives and a new georam.ram is written.
• Relaunch preloads the saved image (NTSC and PAL).
• Save with the card out fails without touching the files on the card; preload with no card reports "Not found!" and GeoRAM still holds data.
• On an original TR in an NTSC C64: GEOS 64 runs under GeoRAM, and a file duplicated on the DeskTop survives a reset and relaunch.
• REU regression: preload path, numbered saves (reu001, reu002, ...), and a stash/fetch from BASIC.
• Freezers with the REU (Super Snapshot v5 NTSC, Final Cartridge III 101%): a byte stashed in the REU from BASIC reads back after a freeze and return.
• On a TR+ in a C128 in 128 mode (BASIC 7.0, GeoRAM left running through a reset): writes to pages in block 0 and to the last page of block 31 read back after switching between them.
• Earlier, on a TR+ in a C128 in C64 mode: a write and read-back of every page, and the same test program against VICE's 512K geoRAM.

Not run: GeoRAM beside a crt.

[SensoriumEmbedded]

Thank you Kelly for the GEOS64 implementation, this is exciting! :)

I reviewed the whole GEOS set merged onto current main (57a9d76) in the order you suggested, each PR on top of the one before it, and built both targets at every step. The note on each PR covers only that PR's own changes. As always, feel free to push back on anything that's inappropriate or would take significant memory/code to resolve.

Memory: this adds 512 B of RAM1 variables and 576 B of ITCM code to the main image on both the TR and the TR+. Free-for-locals goes from 80,700 to 80,188 on the TR+ and from 127,644 to 127,132 on the TR. MinimalBoot and the extension image are unchanged. One improvement:

• Drop GeoRAM_NoRAM (IOH_GeoRAM.c:38): 256 B of DTCM. It's never read: IO1Hndlr_GeoRAM returns whenever there are no banks, and the handler only becomes current after its init has selected a page.

One thing to look at before merging:

1. A saved GeoRAM never comes back (IOH_GeoRAM.c:25, :63)

The button save goes through SaveRAMBanks's numbered naming, so it writes georam000.ram, georam001.ram and so on. The preload always reads georam.ram. The REU's preload file is a setting, but GeoRAM has no way to point at the saved file. So a GEOS RAM disk saved with the button is never loaded again unless the user renames the file on a computer.

Suggested fix: save over georam.ram, keeping the previous one as a backup if you like. Or preload the highest-numbered georamNNN.ram when there is one.

Hardware testing still to do: for one of us to possibly get to before release, not gates, and anything I noted that isn't listed is covered by your follow-ups or fine as it is

Your test program covers the 512K sizing, the block probe, every page and the register mirror, on a TR+ in C64 mode and in VICE.

• GeoRAM under GEOS itself, with GEOS's own geoRAM RAM disk
• a plain TR
• a C128 in 128 mode
• GeoRAM alongside a generic crt, where it gets less than 512K
• a button save followed by a relaunch (after fix 1)
• a preload with no SD card in the slot
• the REU now that its bank code has moved: a preload, a button save, and the freezer + REU modes (SSv5, Retro Replay, FC3 1.01)
• a downgrade with GeoRAM selected as Special IO: the handler numbers after it shift
• PAL

Thank you!

[kfox]

Thank you for reviewing the whole GEOS stack this carefully, and for the build at every step!

I've merged current main (5f16876) into this PR and carried it up through #70, so all six merge cleanly again. Main had already made the same remote-launch fix this PR carried (find the TR menu target before freeing anything), so I took main's version and added FreeGeoRAM() beside FreeREU().

Memory: across the whole stack, free-for-locals in the main image now drops 768 B on both boards (TR 123,548 to 122,780, TR+ 80,700 to 79,932), so the TR no longer crosses the 4 KiB alignment step. On this PR:

• GeoRAM_NoRAM is gone.

1. A saved GeoRAM never comes back

Fixed: the button save now writes georam.tmp, moves the previous georam.ram to georam.bak, then renames the new one into place. If that rename fails, the old file is put back. The REU keeps its numbered saves.

Hardware testing

• GeoRAM under GEOS with its RAM disk: this can't run as shipped. CONFIGURE probes the IEC bus, which hangs with no drive, so GEOS 64 launches from the TR+ with the USB mouse, no disk drive #70 keeps GEOS from running it and GEOS never uses GeoRAM as a RAM disk. In this stack, GeoRAM is there to host the mouse and block device registers. GeoRAM itself is covered by BASIC page reads and writes, and by save and relaunch.
• Done:
  • a plain TR;
  • a C128 in 128 mode;
  • a button save followed by a relaunch (NTSC and PAL);
  • a preload with no SD card in the slot;
  • the REU after the move: preload, numbered saves, a stash/fetch from BASIC;
  • Super Snapshot v5 and FC3 101% with the REU, a stash surviving a freeze and return. I didn't have an NTSC Retro Replay image;
  • PAL.
• Not run: GeoRAM beside a generic crt, and a downgrade with GeoRAM selected.

One unrelated thing I hit on main itself: tests/easyflash.test.js fails on macOS on Apple silicon, because easyflash-harness.cpp:43 asserts that mmap returns 0x20200000, an address macOS there won't map.