Because PDOS/386 now supports (a subset of) the OS/2 API,
it can be considered a mini-OS/2-clone already.

However, it ships with zero LX executables.

At the end of the section "PDOS/386 hard drive image"
there is now a pdoso.zip which contains a VHD that
can be mounted on either Windows or Virtualbox.

However, when I add it Virtualbox for my ArcaOS
guest, it doesn't appear as my D drive. I don't
see it at all.

I tried running "lvm" and by going to "physical view"
I could see the disk, which supposedly is full of
"free space".

This is a FAT32, MBR and the partition starts at
sector LBA 128, not LBA 63.

Is that an issue?

Note that if I could mount it, I was going to do:

d:
path %path%;d:\dos
cd \devel\pdos\pdpclib
pdmake -f makefile.sos
cd \devel

attempt to unzip \misc\custom gcc-stag*
unzip gcc-stage*
cd gcc\gcc
copy config\i386\genfiles
pdmake -f os2.mak

Also note that although you can boot PDOS/386 from
the pdoso.vhd, it will boot fine, but you won't be
able to do the above because of bugs that have
already been fixed and a new release is expected
in the coming hours.

Any idea how to make the volume available, or what
I need to do to make it acceptable to ArcaOS?

Thanks. Paul.

On 11/03/24 22:38, Paul Edwards wrote:

> At the end of the section "PDOS/386 hard drive image"

Forgot to mention "at http://pdos.org "

> Also note that although you can boot PDOS/386 from
> the pdoso.vhd, it will boot fine, but you won't be
> able to do the above because of bugs that have
> already been fixed and a new release is expected
> in the coming hours.

The new image is on the website now and so under
PDOS/386 OS/2-flavor it works fine.

BFN. Paul.

Paul Edwards wrote:
> This is a FAT32, MBR and the partition starts at
> sector LBA 128, not LBA 63.
>
> Is that an issue?

Generally OS/2 needs CHS partitioning with the cylinders aligned. You
likely need to wipe the first sectors, then use LVM to partition it and
add the LVM stuff such as a drive letter.
The rule is to always use OS/2 tools to partition. The exception now is
with GPT partitioning, which is only available with the latest ArcaOS,
or added to earlier OS/2 versions.
Dave

On 12/03/24 01:59, Dave Yeo wrote:
> Paul Edwards wrote:
>> This is a FAT32, MBR and the partition starts at
>> sector LBA 128, not LBA 63.
>>
>> Is that an issue?
>
> Generally OS/2 needs CHS partitioning with the cylinders aligned. You
> likely need to wipe the first sectors, then use LVM to partition it and
> add the LVM stuff such as a drive letter.
> The rule is to always use OS/2 tools to partition. The exception now is
> with GPT partitioning, which is only available with the latest ArcaOS,
> or added to earlier OS/2 versions.

Thanks for that info.

I create a VHD using our own tools that we have
control over.

So I can change it to a different geometry. And
it is arbitrary. But I would like it to be 4k
aligned. 128 sectors does that.

Any suggestions on geometry that will be both 4k
and satisfy OS/2?

Thanks. Paul.

Am 11.03.24 um 18:59 schrieb Dave Yeo:
> Paul Edwards wrote:
>> This is a FAT32, MBR and the partition starts at
>> sector LBA 128, not LBA 63.
>>
>> Is that an issue?
>
> Generally OS/2 needs CHS partitioning with the cylinders aligned.

AFAIR this is not true. OS/2 fdisk and probably lvm refuses to do
changes if the partition table is wrong. But I remember that I have
booted from an HDD where the BIOS mapping was completely wrong.
To get aligned partitions on SSD with OS/2 tools working I used unusual
mappings with 32 sectors per track.

Marcel

Paul Edwards wrote:
> On 12/03/24 01:59, Dave Yeo wrote:
>> Paul Edwards wrote:
>>> This is a FAT32, MBR and the partition starts at
>>> sector LBA 128, not LBA 63.
>>>
>>> Is that an issue?
>>
>> Generally OS/2 needs CHS partitioning with the cylinders aligned. You
>> likely need to wipe the first sectors, then use LVM to partition it and
>> add the LVM stuff such as a drive letter.
>> The rule is to always use OS/2 tools to partition. The exception now is
>> with GPT partitioning, which is only available with the latest ArcaOS,
>> or added to earlier OS/2 versions.
>
> Thanks for that info.
>
> I create a VHD using our own tools that we have
> control over.
>
> So I can change it to a different geometry. And
> it is arbitrary. But I would like it to be 4k
> aligned. 128 sectors does that.
>
> Any suggestions on geometry that will be both 4k
> and satisfy OS/2?
>

Not easily. As Marcel says, you may find a mapping that works that is
not strictly legal for OS/2 or just do the math. You will still need the
LVM info on the drive.
Dave

On 12/03/24 07:12, Dave Yeo wrote:

> Not easily. As Marcel says, you may find a mapping that works that is
> not strictly legal for OS/2 or just do the math. You will still need the
> LVM info on the drive.

What LVM info?

Or is this some sort of internal OS/2 thing that
is not appropriate to be created by an external
tool, and the external tool should just use
32 sectors per track or something similar and
then let LVM accept the new drive?

Thanks. Paul.

On 12/03/24 11:59, Paul Edwards wrote:

>> Not easily. As Marcel says, you may find a mapping that works that is
>> not strictly legal for OS/2 or just do the math. You will still need the
>> LVM info on the drive.
>
> What LVM info?
>
> Or is this some sort of internal OS/2 thing that
> is not appropriate to be created by an external
> tool, and the external tool should just use
> 32 sectors per track or something similar and
> then let LVM accept the new drive?

I've done something further on this.

I'm setting my disks to 0B already, so that should
qualify as CHS.

But is it track or cylinder alignment that I need?

If I just multiply 63 * 8 then I should be 4k aligned,
and only waste about 256k - no big deal.

That's only track alignment though.

I'm doing some other tests currently (with respect to
gccwin as LX module rebuilding itself on PDOS/386) so
not yet ready to mess with that.

BFN. Paul.

Paul Edwards wrote:
> On 12/03/24 07:12, Dave Yeo wrote:
>
>> Not easily. As Marcel says, you may find a mapping that works that is
>> not strictly legal for OS/2 or just do the math. You will still need the
>> LVM info on the drive.
>
> What LVM info?

There's some stuff written to the MBR/EBR containing stuff such as the
drive letter and also in the case of the original JFS, info about
volumes that span partitions, something that isn't really recommended.
You can grow but not shrink a non-bootable JFS volume.

>
> Or is this some sort of internal OS/2 thing that
> is not appropriate to be created by an external
> tool, and the external tool should just use
> 32 sectors per track or something similar and
> then let LVM accept the new drive?
>

That's the recommended way. The only outside tool that handles the LVM
info is DFSee (dfsee.com) and even that recommends not doing it.
You will know if you got the alignment right if lvm (minilvm) accepts
the partition.
Dave

On 11/03/24 22:38, Paul Edwards wrote:

After a lot of effort, I can now answer my own questions.

> This is a FAT32, MBR and the partition starts at
> sector LBA 128, not LBA 63.

ArcaOS documents this vaguely as "CHS aligned".

When it creates a partition at beginning of free
space, it starts on LBA 63, which means it is
track-aligned, not cylinder-aligned.

However, the end of the partition is indeed
cylinder aligned.

But actually it accepts the start of the partition
not even track aligned, as in this 128 example below.

Thus this command (using the latest bug-fixed parted):

D:\scratch\xxx>parted mkpart 0x80,0x0b,128,954448 pdos.vhd

D:\scratch\xxx>hexdump pdos.vhd 0 512
....
0001A0 00000000 00000000 00000000 00000000 ................
0001B0 00000000 00000000 230CA653 00008002 ........#..S....
0001C0 03000B0F FFB28000 00005090 0E000000 ..........P.....
0001D0 00000000 00000000 00000000 00000000 ................
0001E0 00000000 00000000 00000000 00000000 ................
0001F0 00000000 00000000 00000000 000055AA ..............U.

has an end head of 0F (offset 1C3) and the FF is actually 3F,
meaning sector number 63. The high 2 bits should be ignored -
they are part of the cylinder number.

This VHD has a geometry of 16 heads, as seen here
at offset 3A:

D:\scratch\xxx>hexdump pdos.vhd 488743424
1D21A200 636F6E65 63746978 00000002 00010000 conectix........
1D21A210 FFFFFFFF FFFFFFFF 2D8A2B2E 77696E20 ........-.+.win
1D21A220 00060001 5769326B 00000000 1D21A200 ....Wi2k.....!..
1D21A230 00000000 1D21A200 03B3103F 00000002 .....!.....?....
1D21A240 FFFFEB3E 27146D29 2607290F 202E1426 ...>'.m)&.). ..&
1D21A250 496C2B22 00000000 00000000 00000000 Il+"............
1D21A260 00000000 00000000 00000000 00000000 ................

and 63 sectors per track as per offset 3B, and
number of cylinders 3B3 (big endian format at 38).

Although it seems that any start value can be used,
I chose to use 504 to get a clean track alignment,
as well as being 4k-aligned. That's a trivial amount
of wasted space (about 200k).

Here is the command I (now) use:

parted --boot devel\pdos\src\mbr.com mkpart 0x80,0x0b,504,954072 ..\pdos.vhd

However, I think I will take advantage of the fact
that 128 is accepted to return to what I was using
before.

Note that I am using x'0b' to indicate that the CHS
values are accurate - as required by ArcaOS, rather
than x'0c' to say that LBA should be used as the CHS
values may be inaccurate. It is ironic that ArcaOS
uses x'0c' in partitions it creates given that it
relies on the CHS.

Finally, ArcaOS doesn't appear to properly detect or
enforce that the end CHS is cylinder-aligned, so
doesn't always throw an error. The fact that it
doesn't throw an error doesn't mean that there isn't
a problem with failing to properly align. It may or
may not be OK. I don't have sufficient insight into
ArcaOS internals.

Anyway, the end result of this is a FAT32 VHD in
pdoso.zip (not pdos.zip) at http://pdos.org that
contains a bootable image of PDOS/386 (a partial
Windows, OS/2 and Linux clone) with native OS/2
LX executables sufficient to run a toolchain
(which includes gcc 3.2.3) so that you are in a
position to compile (or write new software) any
other C90-compliant software available. And there
is a masm (subset) compatible assembler too (as86),
so you can write assembler in traditional standard
format too.

Simply add the disk (the disk is standard FAT32 -
there is no LVM information or anything else that
is specific to ArcaOS - so you need to use ArcaOS
to add that) to the ArcaOS machine in Virtualbox,
open a command prompt and run lvmgui, select the
partition, then volume, create volume (I chose
non-bootable and compatible - other options may
work but I didn't try them and I don't know what
they mean).

When you exit, you should have a D drive or whatever.

Then you can do:

path %path%;d:\dos
d:
cd \devel\pdos\pdpclib
pdmake -f makefile.sos
pdptest abc def

to demonstrate the software.

BFN. Paul.