/ ( mainboard and support)

Newsflash

@(#)NetBSD 1.1B (DRINSTALL) #5: Mon May 13 20:01:23 MET DST 1996
is@jocelyn.rhein.de:/bulk/s2/src/sys/arch/amiga/compile/DRINSTALL

• history of NetBSD/Amiga on the DraCo & M68060 CPU
• status of NetBSD/Amiga support for the DraCo mainboard
• status of NetBSD/Amiga support for the M68060 cpu
• known bugs

History

1996 May 7

NetBSD-DraCo compiled its own kernel and runs it:

NetBSD 1.1B (DRACO) #292: Wed May  8 10:54:59 MET DST 1996
is@jocelyn.rhein.de:/usr/src/sys/arch/amiga/compile/DRACO

---

1996 May 7

Kernel #290 has a workaround activated, which circumvents the documented store-load-shortcut bug in early CPU mask revisions. gcc works now even on my machine.

---

1996 May 5

Kernel #280 has the floating point bug fixed.

---

1996 May 3

This kernel:

@(#)NetBSD 1.1B (DRACO) #277: Fri May  3 14:31:58 MET DST 1996
is@beverly.rhein.de:/usr/src/sys/arch/amiga/compile/DRACO

went multiuser today. There are some small glitches in the floating point support/context switch code: FP using programs get killed sometimes. But I was able to newfs / and /usr, get the base11 distribution with ftp (via ARCnet), Extract() it and Configure() the machine, then reboot the installed system.

---

1996 Apr 26

@(#)NetBSD 1.1B (DRACO) #261: Fri Apr 26 14:01:39 MET DST 1996
is@beverly.rhein.de:/usr/src/sys/arch/amiga/compile/DRACO

As of today, copyback caching works... (will say, is switched off for the appropriate pages only).

---

1996 Apr 06

This kernel:

NetBSD 1.1B (DRACO) #204: Sat Apr  6 17:42:46 MET DST 1996
is@beverly.rhein.de:/usr/src/sys/arch/amiga/compile/DRACO

booted this morning into single user mode. Actually, I cheated: I had to generally switch off copyback caching... I didn't find all places yet where the kernel modifies page tables through copyback cached mappings. I was, however, able to execute a lot of programs from the inst-11.fs miniroot I tried, including init, sh, mount with mount_ffs and mount_cd9660, ls, dd, ifconfig, route. df didn't work: it uses floating point computations. newfs and fsck don't work either: they use 32/64bit multiplication.

---

1996 Feb 11

A modified NetBSD/Amiga - Kernel booted Sunday evening (1996 Feb 11) up to the point where it paniced because exec'ing init failed.

---

1996 Jan 22

First Light A modified NetBSD/Amiga - Kernel booted last night (1996 Jan 22 0145 MET) beyond the early console initialization, printed the Kernel Copyright Message and a log of the memory segment discovery onto the console and crashed (actually, ran to the Debugger prompt) with a nice, console-readable register dump due to a bus error (probably when accessing some Amiga hardware not available on a DraCo). No more background color code debugging! Yeah!

Known Bugs

• sometimes, SCSI interrupts seem to be lost.
• sometimes, the key release gets lost.

/ ( mainboard support )

General idea

Status and plans

• Modify the bootloader to recognize a DraCo and its custom chip revision level, and find its memory. [done & tested]
• Hack the Kernel to survive early console initialization so that textual debugging output is available. [done & tested]
• In this process, assign kernel virtual memory pages to DraCo mainboard hardware [done & tested]
• write the reboot code for the DraCo [done & tested]
• adapt/write device drivers:
  • Altais graphics board: From the programmers view, this is a Retina BLT Z3 with a changed product id and the ability to create vertical blank interrupts. We just need to recognize the new product id. [done & text mode tested]
  • On-board SCSI interface: Just another instance of siop. We just need to write a short frontend. [done; tested]
  • Modify CIA clock.c to use ciaa on DraCo's, and to not use the 64bit division assembler instruction which is not available on the M68060. [done ; seems to work.]
  • On-board keyboard hardware [first version done; tested]
  • On-board battery backed clock [read-only version done]
  • On-board timer, for the newer DraCo's which don't have any 8520 CIA. [done]
  • On-board multi-i/o chip (serial, parallel, floppy):
    I've implemented a minimal <machine/bus.h> for the Amiga port, and a "supio" bus (a logical direct bus with one locator, the "port"). The "drsupio" driver attaches to the mainbus, and provices an instance of the supio, attaching two coms, one lpt and one fdc at the PC "standard" port numbers.
    Currently, only com_supio is implemented.
    The Xamiga server has been enhanced to decode serial mouse events itself.
  • make the Amiga onboard mouse interface a real device driver, instead of a pseudo-device, so that mixed Amiga/DraCo kernels are possible. [done; tested]
• Add 68060 support to the kernel, so that paging and floating point stuff and unimplemented integer instructions can work. This can be done after (part of) the device drivers have been written, as the kernel itself does not page.
• Integrate the floating point emulation code, so that the DraCos can work.

/ ( support )

• make sure to enable:
  • superscalar dispatch [done]
  • branch target cache [done]
  • store buffer
• tell the kernel to modify page tables/segment tables through cache inhibited pages. [seems to finally be done and running.]
• bus error handler needs a special case for the M68060 [done]
• add branch target cache flushing at appropriate places [done]
• change context switch code to support new floating point context format and to not use fmovem fpiar/fpsr/fpcr any more. [mostly done]
• add unimplemented integer instruction handler from Motorola's 060SP [done; tested]
• add unimplemented floating point instruction & effective address handler , and general floating point support software from Motorola's 060SP [done]
• change drivers to not use movep on the M68060 [first version done for if_bah.c, which was the only one]
• tell gcc how to create 68060-optimized code [preliminary version done by jtc. --- mille grazie]