1
0
mirror of https://github.com/hsoft/collapseos.git synced 2024-11-17 08:18:06 +11:00
collapseos/doc/glue-code.md
Virgil Dupras 2e8af376e3 pgm: new kernel module
The pgm module implements a shell hook so that when an unknown command
is typed, we look into the mounted filesystem and look for a file with
the same name as the command. If we find one, we load it in memory and
run it.
2019-05-31 14:54:15 -04:00

4.6 KiB

Writing the glue code

Collapse OS's kernel code is loosely knit. It supplies parts that you're expected to glue together in a "glue code" asm file. Here is what a minimal glue code for a shell on a Classic RC2014 with an ACIA link would look like:

; The RAM module is selected on A15, so it has the range 0x8000-0xffff
.equ    RAMSTART	0x8000
.equ    RAMEND		0xffff
.equ    ACIA_CTL	0x80	; Control and status. RS off.
.equ    ACIA_IO		0x81	; Transmit. RS on.

    jr init

; interrupt hook
.fill	0x38-$
    jp aciaInt

init:
    di
    ; setup stack
    ld hl, RAMEND
    ld sp, hl
    im 1
    call aciaInit
    xor	a
    ld	de, BLOCKDEV_GETC
    call	blkSel
    call	stdioInit
    call    shellInit
    ei
    jp      shellLoop

#include "core.asm"
.equ    ACIA_RAMSTART	RAMSTART
#include "acia.asm"
.equ	BLOCKDEV_RAMSTART	ACIA_RAMEND
.equ	BLOCKDEV_COUNT		1
#include "blockdev.asm"
; List of devices
.dw	aciaGetC, aciaPutC, 0, 0

.equ	STDIO_RAMSTART	BLOCKDEV_RAMEND
#include "stdio.asm"

.equ    SHELL_RAMSTART	STDIO_RAMEND
.equ    SHELL_EXTRA_CMD_COUNT 0
#include "shell.asm"

Once this is written, building it is easy:

zasm < glue.asm > collapseos.bin

Building zasm

Collapse OS has its own assembler written in z80 assembly. We call it zasm. Even on a "modern" machine, it is that assembler that is used, but because it is written in z80 assembler, it needs to be emulated (with libz80).

So, the first step is to build zasm. Open tools/emul/README.md and follow instructions there.

Platform constants

The upper part of the code contains platform-related constants, information related to the platform you're targeting. You might want to put it in an include file if you're writing multiple glue code that targets the same machine.

In all cases, RAMSTART are necessary. RAMSTART is the offset at which writable memory begins. This is where the different parts store their variables.

RAMEND is the offset where writable memory stop. This is generally where we put the stack, but as you can see, setting up the stack is the responsibility of the glue code, so you can set it up however you wish.

ACIA_* are specific to the acia part. Details about them are in acia.asm. If you want to manage ACIA, you need your platform to define these ports.

Header code

Then comes the header code (code at 0x0000), a task that also is in the glue code's turf. jr init means that we run our init routine on boot.

jp aciaInt at 0x38 is needed by the acia part. Collapse OS doesn't dictate a particular interrupt scheme, but some parts might. In the case of acia, we require to be set in interrupt mode 1.

Includes

This is the most important part of the glue code and it dictates what will be included in your OS. Each part is different and has a comment header explaining how it works, but there are a couple of mechanisms that are common to all.

Defines

Parts can define internal constants, but also often document a "Defines" part. These are constant that are expected to be set before you include the file.

See comment in each part for details.

RAM management

Many parts require variables. They need to know where in RAM to store these variables. Because parts can be mixed and matched arbitrarily, we can't use fixed memory addresses.

This is why each part that needs variable define a <PARTNAME>_RAMSTART constant that must be defined before we include the part.

Symmetrically, each part define a <PARTNAME>_RAMEND to indicate where its last variable ends.

This way, we can easily and efficiently chain up the RAM of every included part.

Tables grafting

A mechanism that is common to some parts is "table grafting". If a part works on a list of things that need to be defined by the glue code, it will place a label at the very end of its source file. This way, it becomes easy for the glue code to "graft" entries to the table. This approach, although simple and effective, only works for one table per part. But it's often enough.

For example, to define extra commands in the shell:

[...]
.equ    SHELL_EXTRA_CMD_COUNT 2
#include "shell.asm"
.dw myCmd1, myCmd2
[...]

Initialization

Then, finally, comes the init code. This can be pretty much anything really and this much depends on the part you select. But if you want a shell, you will usually end it with shellLoop, which never returns.