VirtualBox

Ticket #2402 (new defect)

Opened 6 years ago

Last modified 4 years ago

EComstation (OS/2) crashes during install on 2.0.2

Reported by: mgvb Owned by:
Priority: major Component: other
Version: VirtualBox 3.0.10 Keywords: ECS OS/2
Cc: Guest type: other
Host type: other

Description

Exception in module: RESOURCE TRAP 000d ERRCD=235c ERACC= ERLIM= etc...

Host Opensuse 11.0 guest ECS 2.0

Tried many options in VB and ECS, nothing works.

Attachments

VBox.log.2 Download (50.4 KB) - added by mgvb 6 years ago.
VBox.log.3 Download (54.3 KB) - added by mgvb 6 years ago.
VBox.log Download (42.9 KB) - added by mgvb 6 years ago.
VBox204.log Download (36.2 KB) - added by mgvb 5 years ago.
ecs20 Download (9.6 KB) - added by mgvb 5 years ago.
VBox.2.log Download (50.7 KB) - added by mdlueck 4 years ago.
VBox.log of crash on a 3.0.12 r54655 build booting eCS 1.2R ISO
VBox.3.log Download (56.2 KB) - added by mdlueck 4 years ago.
VBox.log from 3.1.2-56127_Ubuntu_jaunty

Change History

Changed 6 years ago by mgvb

Changed 6 years ago by mgvb

comment:1 Changed 6 years ago by mgvb

This problems was even documented installing ECS 2RC5 in August 08 at Warpstock:  http://www.warpstock.org/filemgmt_data/files/vm_inst.pdf just scan for RESOURCE.SYS.

comment:2 Changed 6 years ago by mgvb

Also just tried the 2.0.2 OSE edition had same problem but different log.

Changed 6 years ago by mgvb

comment:3 Changed 6 years ago by mgvb

More info, I was able to install XP as a guest with no problem. Also, this site is very difficult to add/update ticket. I am using Firefox and keep getting TICKET_APPEND privileges are required to perform this operation. Or other problems. Much info I typed was lost and there is no way to know when site will accept info.

comment:4 Changed 5 years ago by mgvb

Update: Tries 2.0.4 on Opensuse 11.0 with latest kernel. No change. Attached log and screen shot.

Changed 5 years ago by mgvb

Changed 5 years ago by mgvb

comment:5 Changed 5 years ago by Veit

This is an error in memdisk; the memory size reporting handler for int 15/e820 was depending on the high part of edi beeing zero.

comment:6 Changed 5 years ago by frank

Veit, can you give some more details? The VirtualBox BIOS does not use the upper part of EDI when filling the buffer.

comment:7 Changed 5 years ago by Veit

OS2LDR uses various functions to get the size of installed/available memory, including int 15/e820. To reduced the reported size by the size of the allocated memory disk, the code in memboot.bin first call the BIOS to answer the call, and after return check the filled in structure if it must be changed.

Exactly this handler code had used es:edi instead of just es:di for accessing the structure. I have no idea how this mistake happened.

                  ; extended memory ?
                  cmp eax,cs:MemDiskBIOS_Info._phys_addr        ; Anfang<=Memdisk ?
                  ja keine_aenderung_e820       ; nein.

                  add eax,es:[edi].smap.length_lo ; Ende>Memdisk ?
                  adc edx,es:[edi].smap.length_hi
                  test edx,edx
                  jnz patch_SMAP_ext            ; ja.
                  cmp eax,cs:MemDiskBIOS_Info._phys_addr
                  jb keine_aenderung_e820       ; nein.

  patch_SMAP_ext:
                  ; 128 MB -> 121 MB
                  null edx
                  mov eax,cs:MemDiskBIOS_Info._phys_addr
                  sub eax,es:[di].smap.base_lo
                  sbb edx,es:[di].smap.base_hi

                  mov es:[di].smap.length_lo,eax
                  mov es:[di].smap.length_hi,edx

  keine_aenderung_e820:

However there are two things special:

  • a 'normal machine' would not have allowed to use 32 bit offsets larger than 64 Ki to be used in real mode. And if anyone would try, it should fail with invalid opcode exception/interrupt.
  • the code that results in the high part of EDI beeing not 0000xxxx, but 000fxxxx seems to be the the 'PCI install check' in OS2LDR. at about offset 30a9
mov     ax, 0B101h
int     1Ah

jb      loc_0_30C5
cmp     edx, ' ICP'
jnz     loc_0_30C5
or      word_0_B86, 10h

the value in EDI after return in VBox is 000fadb0 (seems ok), for compare in Virtual PC it is 0000ffeb (seems wrong), and on my Thinpad T42p the BIOS does not change the register (this documented to be possible, ok).

comment:8 Changed 5 years ago by webman

Still happening in version 2.2.0

comment:9 Changed 4 years ago by -jha-

Problem still the case in version 3.0.10r54097 (Windows XP host, Intel DualCore). Tried again a lot of options/recombination of VT-x, NestedPaging, PAE/NX, ACPI, IO-APIC... tried other OS/2-Boot-CDs and ECS-Demo-Images, running all to the "RESOURCE/Resource.sys-Problem".

comment:10 Changed 4 years ago by frank

  • Version changed from VirtualBox 2.0.2 to VirtualBox 3.0.10

-jha-, please attach a VBox.log file of such a VM session with VBox 3.0.10. Thank you!

comment:11 Changed 4 years ago by mdlueck

@frank, I just tested it out and get the crash, so will try to figure out how to do attachments in this ticket system.

Host OS: Ubuntu 9.04/32-bit VB: 3.0.12 r54655 installed via the official repository method Guest: Trying to boot the latest eCS 1.2R ISO image.

Aaahh, I see the attach button at the top. So I will post this, then try to attach the log.

Changed 4 years ago by mdlueck

VBox.log of crash on a 3.0.12 r54655 build booting eCS 1.2R ISO

comment:12 Changed 4 years ago by mdlueck

Updated to 3.1.2-56127_Ubuntu_jaunty, same old crash with eCS 1.2R. Attaching log file.

Could we please have an update concerning this problem? Thanks and Happy New Year, 2010!

Changed 4 years ago by mdlueck

VBox.log from 3.1.2-56127_Ubuntu_jaunty

Note: See TracTickets for help on using tickets.

www.oracle.com
ContactPrivacy policyTerms of Use