Hyperion Entertainment Message Boards

Hello. I try to use the EXTMem feature but it seems not to work here on my X1000 with the latest OS4.1 Update.
When i try to fill the ram-disk to test it, the computer get locked complete at the range of 2GB.

So worked this now or is it abandoned?

Greetings
Tears

ram-handler 54.24 [Vectorport Filesystem]

Amiga Operating System (Release 4.1 - Final Edition Update 2)
Copyright (c) 1985-2020 Hyperion Entertainment CVBA.

MemPools=MP/T,ExtMem=EM/T,DisMountable=DM/T,BlockSize=BS/N,ShowSizeAsAllMemUsed=MU/T,ResetDefault=RESET/S,Report=SHOW/S: report

Current settings:
MemPools,MP = OFF (Options: ON or OFF)
ExtMem,EM = ON (Options: ON or OFF)
DisMountable,DM = OFF (Options: ON or OFF)
BlockSize,BS = 16380 (Options: 4092 to 4095996)
ShowSizeAsAllMemUsed,MU = ON (Options: ON or OFF)

I'd like to know too

If I can remember ExtMem feature is partially broken on X1000 so you can use max of 2GB of memory with it. On other hardware platforms it works without such issue. I guess nobody have bothered to fix it.

AFAIK, it is not that nobody bothered, but more that nobody has (so far) succeeded in fixing it.

Best regards,

Niels

One more nail in X1000's coffin

Extmem works just fine.

For the ram-handler, some hardware requires it to be manually enabled,
like the SAM440 and classics because they only have limited ram and
won't normally be able to run out of virtual memory addresses.

SAM460, A1222, X1000, X5000_20 & X5000_40 have it enabled by default,
as you CAN have more than 2 GB of ram, and also because it will give you
a chunk more of usable virtual address space even if you only have the
low 2 GB mapped by the kernel.

Now, here's the basic issue as I understand it...
Just because you might have 16GB of physical ram on your machine,
DOES NOT mean it is available for use in AmigaOS.

I am not sure of the exact problem at this stage, I do not work in that area.
I think the kernel only gets told about the first 4 GB of physical ram,
and only maps the first 2 GB to virtual addresses, as it would blow up some
apps that do signed pointer math, if they encountered unsigned memory
addresses between the 2 GB and 4 GB boundaries.
One of the Kernel devs can give you the more (accurate and detailed) info.

But, as the virtual address space size maps 1-1 to physical memory address size
in the low 2 GB address space, ExtMem (in the ram-handler) allows you to use
SOME of the 2 GB - 4 GB virtual addresses to memory that would be otherwise
inaccessible.

The ram-handler has been carefully written (by me) to not use signed pointers,
so that it can internally use storage in virtual addresses between 2 GB-4 GB.

As I understand it, the periferal hardware is also mapped in high 2 GB address
range too, you don't have much of that upper 2 GB address space available either.

When the kernel get the full physical memory available information, and when
it is rewritten to use 64 bit internal addressing in the Kernels memory system,
only then will the real benefit of ExtMem become really obvious.

Until then we must wait, at least the ram-handler is ready to go with virtually
unlimited capacity, as ExtMem will allow the ram-handler to store data in
otherwise inaccessible memory.
ExtMem enables the total amount of virtual address space used by RAM
to access any sized file, with only one "block" size per concurrent caller.

Full article and my example source code is available here;
https://wiki.amigaos.net/wiki/Exec_Extended_Memory

@colinw, all:

Thanks for your reply and explanation.
Unfortunately I don't really understand them.

AFAIK, it is not that nobody bothered, but more that nobody has (so far) succeeded in fixing it.

I'm sorry, I was making too quick thinking.

But this comment suggests that support for the dual memory controller won't happen either maybe ever most likely.

Maybe the community should start making its own chip set. That way the hardware and software developers could work together.

TSK wrote: Thu Apr 08, 2021 12:15 am

AFAIK, it is not that nobody bothered, but more that nobody has (so far) succeeded in fixing it.

I'm sorry, I was making too quick thinking.

But this comment suggests that support for the dual memory controller won't happen either maybe ever most likely.

You can definitely not conclude that from my comment. On the contrary, it meant that this fix is still being worked on (now probably mostly by the ExecSG team), and I for one have confidence that it will find its solution.

Best regards,

Niels