AROS x86-64 Boot From Hard Disk

miker1264 · 1132

aGGreSSor

  • Member
  • ***
    • Posts: 173
    • Karma: +6/-0
    • russian transit
Reply #15 on: February 11, 2021, 07:58:27 AM
In general, the x86_64 isn't so simple. My observations:

x86_64 NEW
 
  • boot/pc/grub/i386-pc/bswap_test.mod
  • boot/pc/grub/i386-pc/cmp_test.mod
  • boot/pc/grub/i386-pc/ctz_test.mod
  • boot/pc/grub/i386-pc/div.mod
  • boot/pc/grub/i386-pc/f2fs.mod
  • boot/pc/grub/i386-pc/mui_test.mod
  • boot/pc/grub/i386-pc/pgp.mod
  • boot/pc/grub/i386-pc/random.mod
  • boot/pc/grub/i386-pc/rdmsr.mod
  • boot/pc/grub/i386-pc/shift_test.mod
  • boot/pc/grub/i386-pc/strtoull_test.mod
  • boot/pc/grub/i386-pc/wrmsr.mod
  • boot/pc/grub/i386-pc/
  • C/minigzip
  • Classes/webp.datatype
  • Developer/SDK/Extras/CUnit/share/cmake/CUnitConfig.cmake
  • Developer/SDK/Extras/CUnit/share/cmake/CUnitTargets-debug.cmake
  • Developer/SDK/Extras/CUnit/share/cmake/CUnitTargets.cmake
  • Developer/SDK/Extras/include/CUnit/
  • Developer/SDK/Extras/include/lib/libcunit.a
  • Developer/SDK/fd/
  • Devs/DataTypes/WebM
  • Devs/DataTypes/WebP
  • Devs/Drivers/bus.hidd
  • EFI/BOOT/grub/x86_64-efi
  • EFI/BOOT/BOOTX64.EFI
  • L/debug-handler
  • Prefs/Palette
  • Tools/SysExpModules/ahci.sysexp
  • Tools/SysExpModules/ata.sysexp
  • Tools/SysExpModules/pci.sysexp
  • Tools/SysExpModules/storage.sysexp

x86_64 ABSENT

  • Demos/FreeType
  • Devs/AHI/emu10kx.audio
  • Devs/AHI/envy24.audio
  • Devs/AHI/envy24ht.audio
  • Devs/Networks/nforce.device

Code: [Select]
$ file boot/pc/grub/i386-pc/bswap_test.mod
bswap_test.mod: ELF 32-bit LSB relocatable, Intel 80386, version 1 (SYSV), not stripped

$ file EFI/BOOT/grub/i386-efi/bswap_test.mod
bswap_test.mod: ELF 64-bit LSB relocatable, x86-64, version 1 (SYSV), not stripped
???

Another:

Deadwood reboot patch is absent (crashes)
AHI prefs play a test sound (with distortion, without deadwood patch)
Allocate 8G memory:
  • i386-aros, ABIv0: 2.98G graphics mem, 0b other mem
  • x86_64-aros, ABIv1: 2G graphics mem 5.98G other mem



Kill feature: Iconify on desktop like AmigaOS 4 (sometimes crashes)


« Last Edit: February 11, 2021, 08:34:01 AM by aGGreSSor »



aGGreSSor

  • Member
  • ***
    • Posts: 173
    • Karma: +6/-0
    • russian transit
Reply #16 on: February 11, 2021, 10:26:27 PM
I don't understand what kind of miracle happened, but InstallAROS started working after trying to launch not from desktop but from path AROS Live CD:Tools/InstallAROS  :o



I went through installation on a virtual disk without any problems and now AROS64 starts from a virtual disk!  :o


miker1264

  • Legendary Member
  • *****
    • Posts: 627
    • Karma: +26/-2
Reply #17 on: February 12, 2021, 02:05:20 PM
@aGGreSSor

Are you using VirtualBox or VMWare? 

What steps did you take to get to that point. Im curious.
« Last Edit: February 13, 2021, 01:31:39 AM by miker1264 »



aGGreSSor

  • Member
  • ***
    • Posts: 173
    • Karma: +6/-0
    • russian transit
Reply #18 on: February 13, 2021, 12:40:26 AM
Are you using VirtualBox or VMWare?
I ran AROS64 on VMware, but I usually use VirtualBox. At the moment AROS64 doesn't start on VirtualBox. At least since November last year I haven't been able to launch a fresh AROS64 in any virtualization. Previous stable versions of AROS64 were incredibly slow in VMware, so this is a big step forward.


trekiej

  • Member
  • ***
    • Posts: 189
    • Karma: +1/-0
Reply #19 on: February 14, 2021, 05:11:14 AM
Does this include SMP?



magorium

  • Senior Member
  • ****
    • Posts: 253
    • Karma: +7/-0
  • Convicted non contributor
Reply #20 on: February 14, 2021, 06:06:45 AM
AHI prefs play a test sound (with distortion, without deadwood patch)
see https://github.com/aros-development-team/AROS/commit/8f88c3476c2ed9738514340cd56af288b7a8403f

Quote
Allocate 8G memory:
  • i386-aros, ABIv0: 2.98G graphics mem, 0b other mem
  • x86_64-aros, ABIv1: 2G graphics mem 5.98G other mem
Not sure what your point is there. abiv0 i386 memory availability is/was limited by design. There was a change there when work was done on v1 abi. The details about what memory represented what exactly eludes me (not talking about theoretical chip vs fast-ram but actual host memory allocation). The old message-board had a thread that mentioned this and listed the details.

Quote
Kill feature: Iconify on desktop like AmigaOS 4 (sometimes crashes)
Ah, then it found its way back into the matrix  :)

Also the same note for you, please realize there are hardly any end-user related fixes being made atm. Still much work with regards to get things going (build-system, compiler, drivers) and updating other stuff of importance. Sometimes a small fix (that is visible for us) is able to find its way to trunk, which i would consider a bonus atm. The issue list is still growing each day (some of importance, while other issues can wait). Still lacking manpower there. Just be so kind to add your issues so that developers (and other users) won't forget.


magorium

  • Senior Member
  • ****
    • Posts: 253
    • Karma: +7/-0
  • Convicted non contributor
Reply #21 on: February 14, 2021, 06:21:21 AM
Does this include SMP?
In case you are talking about the reports done by user aGGreSSor then his information is conflicting.

He downloaded the non-smp version and at least started testing with that but later on reported that smp is working (but only showed AROS detecting multiple cores/threads, which AROS is capable of doing so for multiple years now). There are specific smp tests that can be run, but the results from those tests are currently absent in the reports.

So, perhaps i spoke too soon and should have simply waited for user aGGreSSor to provide a better and more clear answer for you. I just happened to be present here  when you posted your question  :)


aGGreSSor

  • Member
  • ***
    • Posts: 173
    • Karma: +6/-0
    • russian transit
Reply #22 on: February 14, 2021, 11:01:51 AM
see https://github.com/aros-development-team/AROS/commit/8f88c3476c2ed9738514340cd56af288b7a8403f
Thanks, I was wrong about the patch.. I didn't play music on AROS64 and I hear distortion (click, clicks) when I press the button "Play a test sound" in AHI Prefs. In deadwood builds the sound is clear. So I decided (wrong) that there was no patch here.

Not sure what your point is there. abiv0 i386 memory availability is/was limited by design. There was a change there when work was done on v1 abi. The details about what memory represented what exactly eludes me (not talking about theoretical chip vs fast-ram but actual host memory allocation). The old message-board had a thread that mentioned this and listed the details.
I show with screenshots how it looks in practice. I usually donít give a virtual machine more than 1 GB and discovered it by accident. To doubt is right. Let's take a look.

5G RAM for VirtualBox, i386, ABIv0 (3.41G Chip, 0 Fast)


5G RAM for VMware, i386, ABIv0 (2.98G Chip, 0 Fast)


5G RAM for VMware, x86_64, ABIv1 (2G Chip, 2.65G Fast)


Just be so kind to add your issues so that developers (and other users) won't forget.
Long ago, follow here:
« Last Edit: February 14, 2021, 11:25:31 AM by aGGreSSor »



magorium

  • Senior Member
  • ****
    • Posts: 253
    • Karma: +7/-0
  • Convicted non contributor
Reply #23 on: February 15, 2021, 04:57:20 AM
Thanks, I was wrong about the patch..
No problem. I thought i mentioned it because stating otherwise would not be entirely correct ;).

Whether or not that fixed or did not fixed the issues as experienced on deadw00ds distro, i have no clue as i haven't checked yet.

Quote
I show with screenshots how it looks in practice. I usually donít give a virtual machine more than 1 GB and discovered it by accident. To doubt is right. Let's take a look.
...
Thanks, but i did know that already :)

What i was trying to say is that 32 bit AROS is limited with regards to memory allocation, if only for the fact that it can only address 2^32  (4gb) at the most. It is simple physics. afaik AROS does not offer PAE support. Windows suffer from the same limitation(s), sometimes limited in software, sometimes limited by hardware (there are specific versions of Windows that are not allowed to be installed on hardware that allows to address more than 4GB of memory).

I can't remember what exactly is the reason for the 3GB limit, but it has been there for ages. Most likely some of the additional memory is reserved/allocated for system use, but i do not know the exact details. Of course the 64-bit version of AROS does not suffer from restrictions such as the 2^32 limit.

Therefor it has no use whatsoever to allocate more than 4GB memory to your VM for running the 32 bit version of AROS, or for that matter expecting a ginormous ramdisk to be present when you boot AROS 32-bit on your 8 GB memory equipped machine. It will simply be capped.

And the other thing i was trying to point out is that the 64 bit abiv1 version of AROS divides the memory between chip and fast-ram (at least that is what the OS reports) while the 32 bit abi-v0 version of AROS does not. There are technical reasons for that, but those reasons elude me atm.

Quote
Long ago, follow here:
...
Ah, i seem to have missed some of your reports. Thank you for having filed those (and showed me where they are).


trekiej

  • Member
  • ***
    • Posts: 189
    • Karma: +1/-0
Reply #24 on: February 16, 2021, 03:27:29 AM
Oops, I was wondering if Aros 64 bit SMP was Hard Drive installable.



aGGreSSor

  • Member
  • ***
    • Posts: 173
    • Karma: +6/-0
    • russian transit
Reply #25 on: February 16, 2021, 07:10:16 AM
Oops, I was wondering if Aros 64 bit SMP was Hard Drive installable.
More likely no than yes.
I was able to install a regular x86_64, abiv1 on the hard disk, but I couldn't repeat it and didn't find any patterns.  :(


cdimauro

  • Junior Member
  • **
    • Posts: 50
    • Karma: +1/-1
Reply #26 on: February 28, 2021, 07:38:40 AM
What i was trying to say is that 32 bit AROS is limited with regards to memory allocation, if only for the fact that it can only address 2^32  (4gb) at the most. It is simple physics. afaik AROS does not offer PAE support. Windows suffer from the same limitation(s), sometimes limited in software, sometimes limited by hardware (there are specific versions of Windows that are not allowed to be installed on hardware that allows to address more than 4GB of memory).

I can't remember what exactly is the reason for the 3GB limit, but it has been there for ages. Most likely some of the additional memory is reserved/allocated for system use, but i do not know the exact details.
The reason is that usually an o.s. uses the upper part of an application's virtual address space to map part of the kernel memory, leaving the lower part for the application.
This is needed to have a super fast context-switch of the CPU, once kernel code needs to be executed, because the processor's caches will not be flushed and reloaded every time (which means two times: first when calling kernel code, and then when going back to the application).
AROS doesn't need it anyway, since it has a single address space for everything, and calling user or kernel code is the same thing.

On 32-bit Windows usually you have 2GB for user/application and 2GB for kernel. But there's an option to enable 3GB/1GB mapping, so giving more memory to the applications. AFAIR Linux is normally 3G/1GB, but the kernel can be compiled to more a more fine-graded partition between the twos (kernel needs a minimum amount, anyway).

32-bit applications running on 64-bit Windows have full 4GB of memory available, because the kernel has no need to map its memory, since it's already mapped on the last part (usually) of the 64-bit virtual address space.
Quote
And the other thing i was trying to point out is that the 64 bit abiv1 version of AROS divides the memory between chip and fast-ram (at least that is what the OS reports) while the 32 bit abi-v0 version of AROS does not. There are technical reasons for that, but those reasons elude me atm.
Maybe because some peripherals only work with memory on the lower 4GB (32-bit) of the address space, so that space if "reserved" by AROS for them. While regular applications, which haven't that limit, can use all memory above those 4GB. Just a guess...



miker1264

  • Legendary Member
  • *****
    • Posts: 627
    • Karma: +26/-2
Reply #27 on: April 08, 2021, 01:58:07 AM
Update.

The SMP Variant of x86-64 can be installed and will be able to boot from HDD in VirtualBox with two conditions.

Hyper-V must be enabled and Primary Master (hdd) and Secondary Master (cdrom) must be set to use ata ( IDE). So IDE works but currently ahci (SATA) does NOT work.

The same is true for x86-64 version. It can be installed and boot from disk only using IDE not ahci (SATA).

That's the situation so far.