ELKS 0.8.0-pre2 on ESP32 FabGL/PCEmulator

[tyama501]

The latest ELKS 0.8.0-pre2 (commit 511e594 fd1440-fat.img for IBM) is booted on the ESP32 FabGL/PCEmulator.

Many thanks to the developer of the FabGL for developing the great libraries and the emulator.
https://github.com/fdivitto/FabGL

The cpu core and the bios are based on the 8086tiny.
These are also great.

[ghaerr]

Wow @tyama501, that's pretty amazing seeing ELKS run on an ESP32 board. How is disk implemented, as USB flash? Is USB implemented in an embedded OS on the ESP32 board?

I see FabGL is primarily a graphics library and that VGA is emulated. Have you tried running nano-X programs on FabGL? I am wondering whether the VGA device is emulated in software or hardware.

The cpu core and the bios are based on the 8086tiny.

The 8086tiny bios is pretty cool in that it has callouts implemented for host routines (like console I/O), and is well done. I remember that portions of the 8086tiny CPU emulator (e.g. instruction decoding) are also handled in an unusual fashion through BIOS code outside the C 8086 emulator. Is FabGL using the same mechanism? It would be interesting to see how well a recent version of ELKS runs on 8086tiny, that was the first emulator I used when I joined ELKS years ago.

[tyama501]

Yes, the export worked. Thank you.

[tyama501]

It seems that the graphic library supports VGA but the pc emulator only supports cga video modes.
That is why the current EGA/VGA nano-X driver doesn't work.

int10
https://github.com/fdivitto/FabGL/blob/master/examples/VGA/PCEmulator/bios.asm

[tyama501]

Hello @ghaerr

Currently there is no cga driver for elks nano-X right?
I may try to make one for
int10_set_vm_11:
; 640x480, 2 colour

[ghaerr]

No, but that would be great for ELKS if you wrote a CGA driver. It should be quite a bit simpler than the EGA driver.

640x480, 2 colour

Is that specific for PC-98? I thought two-color mode was 640x200. It might be nice to support 320x200 4-color also.

[tyama501]

https://github.com/fdivitto/FabGL/blob/master/examples/VGA/PCEmulator/machine.cpp

Oops
it seems that machine.c
only supports up to 640x200 vm6.

I am not sure it can scale to vga so that my LCD monitor can display it but maybe I will try it.

[tyama501]

fdisk, mkfat, and mounting 250MByte hard disk image.

[tyama501]

ttyclock
https://github.com/user-attachments/assets/50a68c7e-e64e-4c07-9f27-e993fcbd3051

It seems that redrawing once in a several seconds.

[ghaerr]

Hmm, I don't know whether that's because of ttyclock itself or just the total emulation of both 8086 kernel and app and lots of terminal output. Perhaps some other timing can be done that is not using extensive console output. I didn't think ttyclock is that slow on real 8086 though.

[tyama501]

It seems that ttyclock on i486 25MHz pc-98 emulator sometimes skips a couple of seconds,
so the several seconds in 8086 is not so bad.

Thank you.

[tyama501]

Hello @ghaerr ,

I made monochrome CGA 640x200 nano-X driver.
scr_cga.c and vgplane4_cga.c
tyama501@0c92b22

I need to use ramfont instead of romfont.
DRIVERS += drivers/scr_cga.o drivers/vgaplan4_cga.o drivers/mempl4.o
DRIVERS += drivers/ramfont.o drivers/X6x13.o drivers/elksutilasm.o

I also made scr_cga.c to pretend as colored display so that current application can draw for now.
//psd->bpp = 1;
//psd->ncolors = 2;
// To run colored applications
psd->bpp = 4;
psd->ncolors = 16;

FabGL/PCEmulator could draw it!
(Although the size of nxtetris is exceeding CGA so the bottom cannot be seen.)

nxtetris

nxworld

nxterm

nxlandmine

For nxlandmine,
it seems that psd->ncolors = 2 has better result according to DOSBox-X,
but I don't care it since I don't have a serial mouse for the device.

Thank you!

[ghaerr]

Can we use ifndef CONFIG_HW_VGA as CGA
for the makefile and the application?

Yes, that sounds like a good idea!

[tyama501]

OK, I will try and make PR if it works.
Thanks.

[tyama501]

Hello @ghaerr

I forgot about EGA.

For EGA PC
VGA driver need to be included without the CONFIG_HW_VGA right?
Then export=EGAMODE set the VGAMODE as FALSE.

https://github.com/ghaerr/elks/blob/master/elkscmd/nano-X/drivers/scr_bios.c
if(getenv("EGAMODE"))
VGAMODE = FALSE;
else VGAMODE = TRUE;

We may need new definition CONFIG_HW_CGA (or CONFIG_HW_EGA) to distinguish these.

[ghaerr]

We may need new definition CONFIG_HW_CGA (or CONFIG_HW_EGA) to distinguish these.

I would rather not go that route since the meaning of CONFIG_HW_VGA is actually about whether the BIOS supports a call to return the screen resolution, and the information isn't used by the kernel in the original way anymore. Ultimately that config variable should be going away.

VGA driver need to be included without the CONFIG_HW_VGA right?

Yes, for both VGA and EGA (=not CGA) , and it seems that setting EGAMODE in that case allows EGA resolution for compiled nano-X applications.

If CONFIG_HW_VGA is not set, the CGA driver will be included. This would mean that neither VGA or EGA is supported. Will that work, or is there another problem I am not understanding?

[tyama501]

Oh, OK. So, it means currently CONFIG_HW_VGA can be set for EGA PC too. (since it isn't used in the original way)

[lmpp1972]

Hi @tyama501
Thank you for pointing me to your work, i have download anyway and tried to run, but my IBM original Hercules card show no color or graphics only a blinking "_" , only thing i can run is matrix!!

[tyama501]

Hi @lmpp1972

Thank you for testing it!
Oh, I see.

Hi @ghaerr and @toncho11
Do you know something more to work it with the real hardware?
Or do you think the card just don't support it?

Thank you.

[ghaerr]

I think the problem @lmpp1972 is having here is that the Hercules card might not be CGA compatible, and possibly uses a different address for video memory, so nothing is displayed. Thus @tyama501's new CGA driver won't work for it. IIRC Hercules was a mono-only card that supported some better text resolutions only?

[tyama501]

Thank you @ghaerr

It seems MDA with bitmap graphic, but different from CGA.
https://en.wikipedia.org/wiki/Hercules_Graphics_Card

[tyama501]

Hello @gharr

I have a question about BIOS call of IBM PCs.
I haven't found much BIOS calls that can be used to implement graphic fuctions for ELKS basic but does IBM PC have BIOS calls to draw images? Or do we need to draw directly to vram?
Well actually there are some in
int10 but not so many as PC-98 has.

[ghaerr]

That looks great, thanks for the images!

With regards to speed, it's usually the area-fill that gets noticed when operating too slowly, compared to the line draws which are relatively quite fast.

[tyama501]

Hi @ghaerr ,

I have a couple of questions of the atexit for the basic.

Where should I put the atexit?
(Sorry I couldn' t fully understand #1939 (reply in thread))
Currently host_mode is setting mode.
https://github.com/tyama501/elks/blob/basic_cga/elkscmd/basic/host-cga.c#L74

Do I need to add some kind of handler like host_exit that calls host_mode
and assign the host_exit with atexit(host_exit)?

Or host_exit does not call host_mode but it calls int10 directly.
Then atexit(host_exit) can be in the host_mode.
Is this your idea?

One more question,
does the exit work
not only for CTRL-D (exit basic),
but also, for CTRL-C (quit program)?

For CLS command
I am thinking to clear all CGA VRAM with host_pokeb.

Thank you.

[ghaerr]

Hello @tyama501,

Sorry, I have been in a deep dive trying to understand a major bug in QEMU running large ELKS or TLVC kernels. Looks like we're getting somewhere and then I will have more time :)

Where should I put the atexit?

The atexit should go in the host_mode routine, and be called just once, the first time a graphics mode is requested.

The way atexit works, is you pass it a routine that the C library ensures will be called whenever exit() (but not _exit()) is called for any reason.

does the exit work
not only for CTRL-D (exit basic),
but also, for CTRL-C (quit program)?

Yes, when basic exits with ^D, it calls exit(), so the atexit handler will run.
If basic exits on a ^C through an interrupt handler, then if that interrupt handler calls exit, the atexit handler will run the registered routines. If _exit() is called instead, they won't. Basic should only be calling exit(), as _exit is for special purposes.

It would be possible to also have the ^C SIGINT handler call host_mode() to set back text mode. This would then make ^C exit out of graphics mode, which might (or might not) be good for users.

Or host_exit does not call host_mode but it calls int10 directly.
Then atexit(host_exit) can be in the host_mode.
Is this your idea?

Yes - host_exit will be called by atexit because host_mode called atexit(host_exit) (only if graphics mode is turned on for the first time though). Then, when host_exit is called as a result of graphics mode having been on once, it checks to see whether graphics mode is still on, and makes the INT 10 call to go back to text mode.

I hope that clears things up.

For CLS command
I am thinking to clear all CGA VRAM with host_pokeb.

That would work, but a fast method would be to use an fmemset which takes a far pointer destination and clears it to zeros. We have fmemset in the kernel under elks/arch/i86/lib/fmemory.S if you want to take a look at how that one works.

Thank you!

[tyama501]

Hello @ghaerr

I have referred to the fmemset and wrote clearing codes of vram by in-line assembler.
CLS worked well.
tyama501@c97b349

Also, I have added the atexit.
CTRL-D exit worked well. CTRL-C does not work, but that can be next step.
tyama501@da12e76

I modifed snake.bas a little to snakcga.bas to eliminate CIRCLE command and switching text mode and video mode.
(This still need debug)
tyama501@0c1001e

elksbasic.mp4

Well, I think I would like to make first PR for host-cga.c.
How should I do for the Makefile?
Do I need to add some ifdef?
or is it ok to replace host-stubs.c.

Thank you!

[ghaerr]

Hello @tyama501,

Looks nice!

hink I would like to make first PR for host-cga.c.
How should I do for the Makefile?
Do I need to add some ifdef?
or is it ok to replace host-stubs.c.

The basic executable is now built for IBM PC, PC-98, and also for testing for the Watcom library build as well as for the host using make hostbasic. Now that you have built support for CGA on IBM PC, we need another host*.c file, since the others are already being used separately and host-stubs is important to keep for when having no special machine features.

I would suggest that you rename host-cga.c to host-ibmpc.c since it is for the IBM PC platform, even though only CGA support for now. All IBM PC-specific code can go in there. Then change the Makefile so that the ELKS IBM PC build uses host-ibmpc.o instead of host-stubs.o. There are still issues as to which compiler is being used for IBM PC, see below.

Also, I have added the atexit.

Looks good, exactly what I was thinking. ^C handling should be pretty straightforward, calling the mode reset in the signal handler if needed. Since catchint is in host.c, you may need to either rename that and call it host_catchint, or leave it the same and perform a host_catchint() callout from the catchint routine. This would involved adding host_catchint to all host*.c files to keep the build running.

I have referred to the fmemset and wrote clearing codes of vram by in-line assembler.

It looks like the automatic OpenWatcom CI routine in .github/worflows/ow-libc.yml uses host-stubs.c for building, so we don't need to worry yet about GCC-specific coding, but you should plan on any non portable C to use either #if __ia16__ or #ifndef __WATCOMC__ so that both compile and work. It is too bad that there is not a common C ASM or ASM file compatibility between the two compilers. So for now we have to use ifdef. IMO it is sometimes it is better to the #ifndef __WATCOMC__ since it shows that Watcom C is not implemented, rather than the reverse for ia16, where we always assume that we have ia16-elf-gcc to work with.

Thank you!