Page Fault on boot (25717)

[dustin howett]

I get a page fault at 0x0 when booting Haiku (compiled with GCC2 or GCC4), before the "rocketship" icon lights up. The stack crawl does not provide any useful information; it starts in int_bottom, and goes directly to the page fault interrupt handler.

Hardware: AMD64 X2, SATA Drive in SATA mode, Synaptics touchpad, Firewire (though the module is not in the image), nVIDIA 6150 integrated graphics and nforce 430 chipset.

[axeld]

Can you turn on on-screen-debug output in the boot loader's safe mode menu, and see where it stops? Also, you may want to try with a revision somewhere before hrev25660 and see if that helps.

[bga]

It may or may not be the same bug, but I an getting a crash during boot in the ahci driver (actually, the ahci driver calls up on the pci bus and pci crashes). I am not at home now so can't provide a backtrace, but this seems to be obviously related to the new driver model. I will capture the kdl screen later today.

[dustin howett]

Legacy SATA: controller found! vendor 0x10de, device 0x0266
Legacy SATA: controller_probe
PCI-IDE: Controller in native mode: cmd 0x30c0, ctrl 0x30b6, irq 5
PCI-IDE: init channel...
PCI-IDE: channel index 0
PCI-IDE: bus master base 0x3090
PCI-IDE: init channel done
vm_soft_fault: kernel thread accessing invalud user memory!
vm_page_fault: vm_soft_fault returned error 'Bad address' on fault at 0x0, ip 0x0, write 0, user 0, thread 0xb

I find it odd that legacy_sata is being used... ahci used to work properly for me.

[dustin howett]

22659 works.

[dustin howett]

That is.. hrev25659 works. Oops.

[anevilyak]

fyi, that trace just indicates that legacy sata found a controller that it could potentially handle...in the new driver architecture, each driver returns a float between 0 and 1.0 indicating how well it can handle the chip, the one with the highest rating wins. the 'dm' KDL command will print the device manager's tree, including what driver is loaded for what device. There should also have been a message in the log regarding which driver it loaded based off rating.

[dustin howett]

Figured it was related to the crash and nope, it never got as far as telling which driver it chose, given their ratings.

[dustin howett]

Removing legacy_sata fixes the issue; dm shows that generic_ide_pci took over. When it pagefaulted, eax and edx held the address of the second instance of the legacy_sata driver as reported by dm.

[mmlr]

This bug might be affected/fixed by hrev25867, please retest.

[dustin howett]

Still pagefaults with hrev25867.

[donn]

If it helps, same problem here as of June 10, also with Athlon dual core and nVIDIA "nForce4 Ultra" chipset. Including the "Legacy SATA" bit. Newest working archived raw hdd is hrev25633.

[tonestone57]

vm_page_fault screenshot

[tonestone57]

Latest revision gives me a vm_page_fault at boot too. Attached jpg of "sc". Not sure if this issue relates to my add-in SATA card ( VIA VT6421 )? Will look into it further & report back any new information.

[scottmc]

Can you also attach the output of listdev and sysinfo: listdev > mypcname-listdev.txt sysinfo > mypcname-sysinfo.txt This can sometimes help give further details on the specific hardware in question.

[tonestone57]

*This bug relates to a SATA issue*

My computer has onboard PATA connectors ( 80GB drive ) and an add-in PCI SATA card ( 250GB drive ). Once I removed the SATA card from the PCI slot, Haiku booted up and was working well. When I put the SATA card back in the PCI slot, Haiku gave the page fault.

Something is going on with with either ahci and/or legacy sata. SATA conflict? I *think* ahci & legacy sata are both trying to support the controller. Look at my new screenshot.

SATA card is VIA VT6421, vendor id=0x1106, device id=0x3249

[tonestone57]

SATA bug

[luroh]

oops, sorry.

[tonestone57]

Ok, with further testing it looks like the problem has to do with legacy_sata.

My PATA drive /dev/hda1 = BeOS /dev/hda3 = Haiku

I booted into BeOS, mounted Haiku, deleted ahci, rebooted, tried to boot Haiku and it crashed with page fault error. I restored/undeleted ahci, then deleted legacy_sata, rebooted the computer & Haiku boots up & works.

With legacy_sata driver removed, I'm able to boot into and use Haiku ( off my PATA ) but I don't have any access to my SATA drive.

[donn]

OK! Same here: without system/add-ons/kernel/boot/legacy_sata, hrev25969 works great on a PATA disk. SATA & PATA controllers both on nForce4 motherboard. SATA still semi-working - can detect partitions and read a little data, but eventually crashes or locks up. (I don't have listdev on hrev25969, and an older copy doesn't list anything.)

[tonestone57]

added listdev & sysinfo output. sysinfo was run off Haiku, listdev off of BeOS. Not sure why, but listdev seems to be missing from Haiku build.

@donn, if you deleted legacy_sata then your SATA drive shouldn't even show up ( what I get ); Unless ahci is somehow able to access it? Maybe this is the case for you and why SATA is semi-working? In any case, at least we know this bug relates to SATA ( legacy_sata ).

[_be]

I can confirm that the problem persists with hrev26056, tried to boot from a usb-stick here; the sc is exactly the same as the posted.I do have a SATA-disk in the pc; i'll try to switch the sata-controller off next

[ekdahl]

Same problem here. Also using Athlon 64 X2, Nforce chipset, legacy_sata.

[scottmc]

hrev26126 fixed the listdev, so if you still see this issue after hrev26126 use listdev to get the list of devices in your pc and attach it to this ticket.

[donn]

legacy_sata.c:sChannelInterface {} needs an implementation for set_channel().

I added one, modelled after generic_ide_pc.c, and that allows boot to complete without page fault.

I actually re-coded the struct initial value per generic_ide_pc.c, with order-dependent struct member values rather than assignments, and so added some NULL initializers that weren't there before. Don't know if this made any difference, but it's an advantage of this classic initializer code - static type analysis will notice some cases where a new member isn't initialized in an old module.

[stippi]

Cool, can you provide this as a patch and attach it to this ticket?

[donn]

Note that this patch only resolves my page fault problem. SATA is still not usable due to other problems.

[axeld]

(In #2464) Thanks, that made it clear - it's a duplicate of bug #2284.

[axeld]

Applied in hrev26163, thanks a lot!