Opened 9 years ago

Last modified 6 years ago

#12252 assigned bug

Boot Loader Death Land with AHCI

Reported by: IIsi50MHz Owned by: nobody
Priority: normal Milestone: Unscheduled
Component: System/Boot Loader Version: R1/Development
Keywords: boot-failure Cc:
Blocked By: Blocking:
Platform: All

Description

Motherboard: Gigabyte GA-MA785GM-US2H BIOS version: F12f CPU: AMD Phenom II X4 965 <CPUID:00100F43 Patch ID: 0068> Memory: 4094 MB DDR2 800

This motherboard ran Haiku previously with a Phenom II X2 545 and BIOS F7.

hrev's tested: 47259, 47628, 47665, 48350, 49162, 49180

BLDL Screen:

Page-Fault Exception: write fault at address: 0x81127000
Welcome to Boot Loader Death Land!

 eax 0x3c          ebx 0x81127000     ecx 0x8         edx 0x3c0800
 esi 0x0           edi 0x11           ebp 0xf4d0      esp 0xf468
 eip 0x3ecf     eflags 0x211016

       frame    return address
  0x0000f4d0        0x0003e3cf
  0x0000f540        0x0003c3a0
  0x0000fdf0        0x00030cc0
  0x0000fe60        0x000309a6
  0x0000fe90        0x00030d91
  0x0000fed0        0x00029109
  0x0000ff10        0x000291c4
  0x0000ff50        0x00024512
  0x0000ffa0        0x0001ee20
  0x0000fff0        0x000105d9

Press a key to reboot.

Ineffective Changes:

x Disabling various CPU cores in BIOS x Underclock 5%, 10%, 20% x Reset BIOS for defaults x Load BIOS optimised defaults x Reverting to older hrevs which used to work x Using CD burnt from ISO (works on another machine) x Using USB with AnyBoot (works on another machine)

Works with:

  • Ubuntu (Zorin OS, Elementary OS) Live USB (need PLOP Boot Mgr due to 2GB USB limit)
  • Lubuntu 9 CD
  • FreeDOS
  • Windows XP HD

Change History (14)

comment:1 by axeld, 9 years ago

Did you change anything since last time? What revision did work for you?

comment:2 by IIsi50MHz, 9 years ago

I think hrev 47665 was the last one to work on this motherboard, but that was with the previous CPU. I can't revert the CPU, but I can revert the BIOS. Have just been reluctant to do so.

I know BIOS F12f is a beta, but it's probably the last that will ever be released, and the problem exists with F10 as well.

What I'd prefer to do is step through the boot process, and find out exactly what the boot loader is trying to do when it dies. I expect this requires heavy wizardry.

comment:3 by IIsi50MHz, 9 years ago

Neglected to mention motherboard version is 1.0, and AMD 785G + SB710 Chipset 

comment:4 by IIsi50MHz, 9 years ago

Changing from "AHCI" to "Native IDE" mode lets Haiku boot on this hardware & BIOS. A release note for BIOS F12f reports "Update AHCI ROM 3.1.0.0".

comment:5 by pulkomandy, 9 years ago

Summary: Boot Loader Death LandBoot Loader Death Land with AHCI

comment:6 by IIsi50MHz, 9 years ago

This issue has mysteriously gone away from one device:

Today I a did full-backup from a Vaio B100B laptop to this USB disc by running Installer. The Vaio was running hrev49180. I then booted the desktop system previously specified above with SATA mode set to Native IDE.

I then ran the following:

pkgman refresh pkgman update sync

followed by reboot.

Bootloader procedes to boot menu, where it declares no boot volume found. Fearing that maybe the update had written something to the internal HD instead of only the USB disc, I returned BIOS SATA mode to "AHCI" and "As SATA Type", then booted Windows successfully. I then forgot to set SATA mode to Native IDE.

After inserting this mis-updated USB disc into the Vaio, I ran Installer to restore the USB disc to hrev49180. This disc had previously received multiple copies of all previously mentioned hrevs, both by writing Anyboot images from Windows and by running Installer from working systems. Each of these hrevs would boot other hardware, but not this one desktop. Deleting partitions and reformating had also been tried previously, with no effect.

I put the restored USB into the affected system and booted successfully. Only then did I remember this shouldn't be possible, because I had not set Native IDE mode in BIOS, which had been necessary. I verified that the BIOS is indeed set to AHCI mode, and that the disc now boots in both AHCI and IDE modes.

Somehow, the failed update, followed by running the hrev49180 Installer, made just this one USB volume bootable in the affected system without needing to set SATA to Native IDE mode.

However, the Haiku CD-ROM made from the hrev49180 ISO image still exhibits the Boot Loader Death Land unless Native IDE mode is set in the BIOS.

comment:7 by kallisti5, 8 years ago

AHCI was reworked recently. Is this still an issue?

comment:8 by IIsi50MHz, 8 years ago

Test:

  • Use ISO from haiku-nightly-hrev50398-x86_gcc2_hybrid-cd
  • Burn to rewritable CD using Windows 7 6.1.7601 default burning utility
  • Set target machine BIOS to AHCI mode
  • Boot from CD

Machine Specifications:

  • As originally posted.

Result:

Page-Fault Exception: write fault at address: 0x81127000
Welcome to Boot Loader Death Land!

 eax 0x12          ebx 0x81127000     ecx 0x81126e80  edx 0x501
 esi 0x3           edi 0xd            ebp 0xf490      esp 0xf418
 eip 0x3e1fb    eflags 0x211097

       frame    return address
  0x0000f490        0x0003e1fb
  0x0000f530        0x0003c257
  0x0000fde0        0x00030bd4
  0x0000fe50        0x000308ab
  0x0000fe80        0x00030ca1
  0x0000fec0        0x00029059
  0x0000ff00        0x0002910c
  0x0000ff40        0x00024426
  0x0000ffa0        0x0001ec46
  0x0000fff0        0x000105d1

Press a key to reboot.

Workaround:

  • Set target machine BIOS to IDE mode. (Caveat: This may prevent other installed OS from booting, if the other OS was installed with AHCI enabled.)

comment:9 by axeld, 8 years ago

Owner: changed from axeld to nobody
Status: newassigned

comment:10 by waddlesplash, 6 years ago

Keywords: boot-failure added

comment:11 by IIsi50MHz, 6 years ago

Still present.

Test:

  • Use anyboot from haiku-nightly-hrev52375 x86-gcc2_hybrid
  • Burn to rewritable CD using Solus 3.9999 64-bit (Linux 4.9.128-108 lts x86_64) & Brasero v3.12.2 burning utility
  • Set target machine BIOS to AHCI mode
  • Boot from CD

Machine Specifications:

  • As originally posted, except 6GB DDR2

Result:

Page-Fault Exception: write fault at address: 0x81127000
Welcome to Boot Loader Death Land!

 eax 0x3          ebx 0x81127000      ecx 0x9         edx 0x30900
 esi 0x1          edi 0x15            ebp 0xf4c0      esp 0xf458
 eip 0x3eda4   eflags 0x211016

       frame    return address
  0x0000f4c0        0x0003eda4
  0x0000f530        0x0003cb7e
  0x0000fde0        0x000312c0
  0x0000fe50        0x00030fa6
  0x0000fe80        0x00031395
  0x0000fec0        0x000293f1
  0x0000ff00        0x000294ac
  0x0000ffa0        0x000246a3
  0x0000fff0        0x0001ee68
  0x0000fff0        0x000105d9

Press a key to reboot.

Workaround remains:

  • Set target machine BIOS to IDE mode. (Caveat: This may prevent other installed OS from booting, if the other OS was installed with AHCI enabled.)

comment:12 by tqh, 6 years ago

Does the nightly image still fit on a CD?

comment:13 by IIsi50MHz, 6 years ago

hrev52375 does, and the disc boots on two other machines. Also boots on that machine with BIOS set to native IDE mode. (And I thought simply leaving sourcecode packages out was enough for R1b1 to fit CD.)

comment:14 by korli, 6 years ago

The fault happens in inflate()->inflate_fast() called from TarFS::Volume::_Inflate(). What's curious is that the fault address is the same since 3 years.

Note: See TracTickets for help on using tickets.