NMI Interrupt introduced between r34760 and r34915

[adamk]

This afternoon I updated my local svn repo to hrev34915 and installed an updated build to a spare partition I use for testing.

Upon booting up, I'm greeted with a KDL: 'PANIC: Fatal exception "NMI Interrupt" occurred! Error code: 0x0" . I've noticed other reports on here, and someone on #haiku mentioned the same problem, but it seems that others are able to 'cont' and the boot up proceeds as normal. Unfortunately, continuing is not an option for me as I end up with a vm_page_fault.

I can still boot up my hrev34760 installation on another partition on the same drive.

I have a logging via a serial cable, and grabbed a backtrace which I am attaching to this ticket.

While it is certainly possible for me to start reverting to previous commits, this will undoubtedly take a while, so I'm hoping someone can read something in my backtrace that indicates the source of the problem and figure out which commit broke Haiku for me :-)

[adamk]

NMI Interrupt backtrace.

[adamk]

Successful boot of hrev34760

[adamk]

I've attached a log from a successful boot of the older version. The crash in the newer version seems to happen at about this point:

acpi: ACPI disabled ahci: ahci_supports_device

I get the ACPI disabled in the new buggy version, and then the crash occurs before the ahci line.

[anevilyak]

From the backtrace, looks possibly related to some of Ingo's recent area management changes...if that backtrace is consistent, the output of "call 15 -3" after the NMI itself would be interesting (i.e. without trying to continue).

[adamk]

The backtrace is consistent. The output of "call 15 -3" is simply:

PANIC: Fatal exception "NMI Interrupt" occurred! Error code: 0x0

Welcome to Kernel Debugging Land... Thread 2 "idle thread 2" running on CPU 0 kdebug> call 15 -3 thread 2, idle thread 2 809a9f64 800e4530 <kernel_x86>:arch_cpu_idle(0x0 (0), 0x0 (0), 0x0 (0)) kdebug>

[adamk]

cpu1 backtrace, requested by DeadYak :-)

[adamk]

And here's 'call 18 -3' after the attached cpu1 backtrace .

kdebug> call 18 -3 thread 15, main2 81107854 800fa3ac <kernel_x86>:memset_generic(0x80c45000, 0x0 (0), 0x1000 (4096))

[anevilyak]

That unfortunately looks quite correct :( Thanks for checking though. Perhaps something's wrong with the area it just tried to create, though it'd take a bit more digging to get useful information about that one.

[bonefish]

I don't know in which situations an NMI is triggered (IIRC the IA32 specs were somewhat vague), but unlike e.g. Linux Haiku doesn't handle those yet.

Might be a duplicate of #2680 and/or #3113. Passing on this hot potato... :-)

[anevilyak]

Thanks for looking Ingo :) In that case I guess it'd be more helpful if adamk could binary search exactly what revision this starts at, since it's at least 100% reliably occurring.

[adamk]

hrev34837 does not NMI crash (but has separate app_server issues that I will detail in another ticket once I track down the bad commit here). I'll keep bisecting this bug.

[adamk]

hrev34876 is fine.

[adamk]

hrev34901 is fine.

[adamk]

34908 and 34911 are both good.

[adamk]

34912 was the first revision to cause the NMI interrupt crash.

[axeld]

When you write it as hrev34912 it is linked automatically to the corresponding changeset.

[adamk]

Thanks. FYI, it's working again as of hrev34990 :-)

Adam

[jackburton]

Closing, since it's fixed.