******** OS/A 1.00 ******** 23sep90, Andre Fachat
Multitasking Operating system for the 6502 CPU.
The system has a memory manager, stream handler and device handler.
Also there is an environment handler with a scheduler.
The whole stuff works like a kind of virtual machine, with the only
difference that hardware accesses are done via the kernel.
But each task and each device has its dedicated memory area.
Each task can think of having its own CPU concerning memory environment
and esp. stack and zeropage.
This only works with the CS/A builtin MMU, where the upper 4 bit are
expanded to 8 bit, thus allowing 1 MByte of virtual memory.
The memory block size is 4kByte. The possible 1 MByte is divided into
256 blocks, which can be mapped at any of the 16 possible positions
in the 64k size 6502 address space.
Each task can have up to 58 kByte of linear address space, which is
64k minus 4 kByte kernel ($f***) and 2 kByte I/O ($e800-$efff).
The communication between the tasks work via streams, i.e. asynchronous
first-in-first-out buffer. For example, a device can write into the
stream that is being used as STDIN stream, while tasks puts its data
to the STDOUT stream that might be written to file on a filesystem.
Devices are a set of subroutines that can have up to 4 kBytes, i.e.
one memory block. They are mostly called during an interrupt, which they
have to serve.
Tasks can also have their own interrupt routine. Semaphores to
synchronize different tasks are also available. As long as a
semaphore has a Pass (P) operation done on it and no Free (V),
another task trying to Pass it will be set asleep.
With SEND and RECEIVE messages up to 256 byte can be passed from one
to another task. When receiving, the sending task can also be specified.
READ and WRITE allow a task to manipulate another tasks memory.
Currently available are the NULL, SPOOLER, CENTRONICS, SER(1-2) and VIDEO(1-4)
devices. The NULL device empties all streams, until there is no further
data available (EOF). The spooler copies all input streams one after
the other to its output stream (serialization).
File handling is done with the filesystems. Currently available are
the FSDEV and the FSIEC filesystems. Filesystems are called with
SEND on task $fe. FSDEV allows the handling of devices as files.
FSIEC handles Commodore disk drives with device address 8 and 9.
******** OS/A 1.01 ******** 1oct90, Andre Fachat
Video device and shell/monitor overhauled, mainly because of better
cooperation on full screen editor.
******** OS/A 1.30 ******** 14apr91, Andre Fachat
Complete reworking:
Several negative task numbers can be assigned an
specific 'real' tasks. For each negative number only one task can
be assigned.
The operating system now even works in a computer without MMU. All tasks
work in the same environment, so that identical programs cannot be
invoked twice.
For speed reasons the task interrupts can be disabled at assemble time.
Also the availability of a system port can be flaged at assmble time.
The IEC bus filesystem can also work as a drive for the IEEE bus.
That means that e.g. a C64 can use the internal filesystems of the
CS/A computer as a disk drive.
With a new hardware board, the computer can now handle IBM PC style
disks with FSIBM. Disks from 360k up to 1.44 MByte can be handled,
with block sizes up to 1 kByte (what you get from a WD1770).
A Basic interpreter has been ported from the C64, thus making it possible
to run BASIC programs in parallel...
******** OS/A 1.31 ******** 25jun91, Andre Fachat
A special MCOPY program to copy multiple files one after each other
is available in the ROM. (Starting the copies in the shell in the
background would eat up resources). Also filename jokers (*, ?) are
available with this program.
******** OS/A 1.31 ******** (19jan92) 21feb92, Andre Fachat
Introduced some kind of STDIO library below $e800, i.e. below the I/O
area. These library routines run in the tasks own memory and handle some
things happening in most tasks.
The centronics device finally works, and FSIEC now accepts an additional
drive number for dir and and the disk commands, for the handling of
dual disk drives for example.
******** OS/A 1.31 ******** 10may92 Andre Fachat
Two new options when compiling: BATMEM allows some battery buffered
memory ($0000-$8000) to not be erased on boot. NOMIRRORMEM saves a few
byte in the kernel if it's sure that we have no mirrored memory
(no open address lines).
******** OS/A 1.31 ******** 12jul92 Andre Fachat
When using PSEM, the Carry flag now says, if the task should block
(C=0) or if the task should return immediately with an error if it
could not get the semaphore (C=1).
Negative semaphores can be used as system semaphores.
********* OS/A 1.31 ******** 15oct92 A.Fachat
Another bug in the STDIO library is fixed: The drive is set correctly
with Usedir. In the monitor, when setting registers or parameters, the
"." to ignore an entry (leave byte untouched) has been replaced by
a dash ("-"). Now it is possible to enter decimal numbers (with leading
".").
A bug in the shell DEVCMD introduced with the STDIO lib has been fixed.
The programs got too big, so now MCOPY is not in the ROM anymore.
Exchanged FSIEC and FSIBM in the ROM, so that drive C: is IEC drive 8,
which normally is the harddisk of my Atari.
******** OS/A 1.32 ******** 16oct92 A.Fachat
In the shell and in the monitor now RAM extensions are accepted not only
till $8000 but till program start. Also, if no more memory available,
the block is not set to $20 anymore.
Without an MMU, the NMI interrupt can now also be used.
If NOSINPORT is set when assembling, the system port is only used
for output, so that plugging a debug card into the bus enables debug
output (if any).
******** OS/A 1.33 ******** 17oct92 A.Fachat
In this version the previous kernel task tables (task data and the
task name) are put into two tables with the same offset for a task.
(previously the task number had to be multiplied by 24 and 8, now
by 16 for both tables). Thus now with MMU, 16 instead of 10 tasks can
be running.
The 'SYSBUF' has been eliminated. Instead of copying data from PCBUF to
SYSBUF and vice versa for (system) SENDs, REGDEV,... the MMU is now used.
Now the PCBUF is used by the kernel ony during startup and can later
be used by the programs.
******** OS/A 1.3.3 ******** apr1996 A.Fachat
Put the source code into a Unix tar archive, html-ized the docs and
put it on the World Wide Web.
******** OS/A 1.3.5 ******** apr1996 A.Fachat
Introduced a version for the C64 (on public demand... ;-)
The FSIEC has been changed to handle the original Commodore C64 IEEE488
bus interface. The terminal device handles the C64 screen and keyboard.
The kernel uses the CIA1 timer B for interrupts.
Many things have now only been ported to a system without MMU, because
that's a lot of work. Everything has to be check if using global
variables, esp. when called from different tasks, like the STDIO lib...
Development took place on the vice C64 emulator.
******** OS/A 1.3.6e ******** 09jun1996 A.Fachat
The C64 is now real 'breadbox C64' tested and debugged. The parallel
port (IEEE488) stuff had bugs that are now fixed. The serial bus is now
working as well. Some more bugs in the NOMMU port have been fixed. A
semaphore for the use of the global PCBUF send buffer has been
introduced. All filesystems and tasks now try to get the SENDBUF
semaphore first before trying to write to the send buffer.
Also system semaphores for some hardware (IEEE488 interface, WD1770 PC
disk controller) have been introduced.
******** OS/A 1.3.6f ******** 09jun1996 A.Fachat
Changed the maximum number of tasks for systems without MMU from 6 to 5.
This gives 40 instead of 32 bytes of stack space for each task.
Hopefully now occasional BRKs are gone.
FSIEC does now recognize the "/" DIRSIGN at the beginning of a directory
search mask and removes it - all filenames are global.
******** OS/A 1.3.7 ******** 30jun1996 A.Fachat
Introduced return code for device interrupt routines. It is returned
if an interrupt source has been cleared (E_OK) or not (E_NOIRQ).
This is checked, and if a source has been removed, the interrupt
routine is left - if we cannot check the irq line directly with SYSINPORT.
The C64 got a ACIA 6551 device, if it is built into the C64.
******** OS/A 1.3.8 ******** 13oct1996 A.Fachat
Added two routines (PUTB, GETB) to handle block transfers to Stream
buffers. SBLK transfers a block of data (length in a) to (c=1) or from
(c=0) a stream (xr) buffer to memory (address in zeropage address
pointed to by yr). The routine returns the number of actually
transfered data in a (?). The data must not cross page (4k) boundaries
on MMU systems.
This procedure (hopefully) avoids a lot of context switches.
******** OS/A 1.3.8d ******** 16oct1996 A.Fachat
a log says more than thousand words :-)))
----------------------------------------------------------------
root:/root>ping -c 1 192.168.0.13
PING 192.168.0.13 (192.168.0.13): 56 data bytes
64 bytes from 192.168.0.13: icmp_seq=0 ttl=64 time=4673.4 ms
--- 192.168.0.13 ping statistics ---
1 packets transmitted, 1 packets received, 0% packet loss
round-trip min/avg/max = 4673.4/4673.4/4673.4 ms
root:/root>date
Wed Oct 16 18:07:58 MET 1996
root:/root>
----------------------------------------------------------------
Well, there still is a lot to do:
- problems with lost characters
- memory management buggy
- removing DEBUG output (gets down to 1.6s for a ping reply)
******** OS/A 1.3.8e ******** 16oct1996 A.Fachat
Ironed out some bugs in SLIP memory handling, which seems stable now.
SLIP behaves strange when packet buffer is _not_ empty (i.e. filled
with zero)
******** OS/A 1.3.8f ******** 18oct1996 A.Fachat
SLIP IP layer is stable now. Added serial device driver for UART 16550A.
with FIFO. But still lost chars above 2400 baud :-(
Fixed another bug in GETB, but GETB version of SLIP is way _slower_
than GETC version!? GETC version gives PING replies in 860ms (@ 2400 baud),
GETB version in around 1750ms.
Serial line interrupts still increase cursor speed.
******** OS/A 1.3.9 ******** 19oct1996 A.Fachat
A new compile option STACKCOPY for systems without MMU allows the stack
to be copied into a save area on context switch (i.e. $80-$ff is system,
$00-$7f for task, the stack is copied only when a real task change takes
place). This now allows more than 5 tasks in systems without MMU.
******** OS/A 1.3.9a ******** 19oct1996 A.Fachat
Optimized UART 16550A driver. Now works for 9600 baud (if not much other
work to do). Then we still loose characters. Running SLIP with 19200
crashes the system :-(
******** OS/A 1.3.9b ******** 20oct1996 A.Fachat
I managed to get my selfbuilt hardware running again! So the current
stdrom is checked with MMU. OLDROM will from the next version on
refer to 1.3.9b
******** OS/A 1.3.9d ******** 21oct1996 A.Fachat
Serial device driver for the fast (i.e. 9600 baud) RS232 interface by
Daniel Dallmann started. Transmitting works.
******** OS/A 1.3.9e ******** 22oct1996 A.Fachat
Now receiving at slow byterate (though 9600 baud) works. Too many bytes
(i.e. NMIs) in a short time kill the machine...
******** OS/A 1.3.9f ******** 22oct1996 A.Fachat
Made NMI driver a bit faster, receiving at fast RS232 works better now
for low byterates, although still crashes at high byterates.
SETNMI now returns the old NMI vector, and the NMI routine has not to
RTS, but to JMP(old_vector).
******** OS/A 1.3.9g ******** 23oct1996 A.Fachat
Fixed the bug in the serial driver IRQ (not NMI!) routine, which messed
up the stack when getting above the stream low water mark...
SETNMI now sets the NMI routine directly, i.e. ($fffa) jumps to
jmp (vector), where vector can be set with SETNMI; so registers have to
be saved in the NMI routine.
TODO now: a means to disable NMI during disk or whatever access.
Maybe redefine $f003 jmp, because NMI vector is known anyway: jmp ($fffa).
PING now gives around 10 seconds reply time, dropping every second packet
due to full tx queue: tx byterate is slow due to 16 used bits per byte
(i.e. 6 additional stop bits) and interrupt latency. But usually receiving
should have higher byterate than receiving. With PING, the SLIP task
tries to send as fast as it receives, which doesn't work...
The same bug as in the fast serial driver was in the UART driver, which
now works at 19200 baud, only very occasionally loosing characters.
Also some more handshake bugs in these drivers have been fixed.
******** OS/A 1.3.10a ******** 25oct1996 A.Fachat
Added CTRLNMI system call, together with the NMI chain. There is still
a way missing to remove a driver from the chain.
The C64 fast serial driver uses this method now. Although flooding the
computer with bytes slows it down to near unusability.
******** OS/A 1.3.11 ******** jan1997 A.Fachat
Been working on TCP implementation. discard & echo ports work.
Also connection closing ceremonies. Packet Retransmission works.
Retransmission timeouts, timewait timeouts work.
chargen port works. program registers itself as filesystem.
TODO: 32-bit-modulo, some timeouts...
******** OS/A 1.3.12 ******** 2mar1997 A.Fachat
Implemented the Gecko OS ROM. It features a shell (with monitor),
the irtx program to send commands to IR devices (like CD players or
TV sets), a serial IEC filesystem. The shell runs over the serial
line (with 16550).
******** OS/A 1.3.13 ******** 09mar1997 A.Fachat
The fstcp filesystem works for reading files and directories.
Directories without directory mask, though.
The fstcp server accepts connections only from a given address now,
and (hopefully) doesn't allow accesses outside the given directory.
TODO: 32-bit modulo, fstcp writes, bugfixes
******** OS/A 1.3.13a ******** 10mar1997 A.Fachat
Writing to the fstcp filesystem seems to work now.
FSTCP can now be disabled in main.a65.
******** OS/A 1.3.13b ******** 11mar1997 A.Fachat
hacked a small (and dirty) WWW server...
******** OS/A 1.3.13d ******** 14mar1997 A.Fachat
Bug fixes...:
- It is now possible to run the WWW server together with the
FSTCP filesystem at the same time - it is _not_ allowed to let the
WWW server access a file on the FSTCP filesystem - it hangs the SLIP
process!
- The problem with the aborts is gone: I forgot to release the STREAMS
in the WWW server
- the transition from LASTACK to CLOSED wasn't detected, because ackxmit
overwrote the "tmp" variable that holds the checkack return value...
- LASTACK now also has a timeout to close the TCB after a while
- FINW2 was not resolved to TIMEW because there was no check...
so the available number of TCBs decreased...
- FINW2 now also has a timeout to closed, as when the FIN is ack'd, we
cannot resend it (removed from queue...). This is probably not as it should
be, but it works anyway.
