Ticket #2059 (new bug)

Opened 1 month ago

Last modified 4 days ago

KDL during svn checkout in block notifier/writer

Reported by: anevilyak Assigned to: axeld
Priority: critical Milestone: R1/alpha1
Component: System/Kernel Version: R1 development
Cc: Platform: All

Description

After ~1GB of files having been downloaded via svn, Haiku KDLed with:

PANIC: vm_page_fault: unhandled page fault in kernel space at 0x200246 ip 0x86d9c800

stack trace for thread 8 "block notifier/writer"
<snip>usual debugger parser functions
<kernel>:panic + 0x0029
<kernel>:vm_page_fault + 0x00ab
<kernel>:page_fault_exception + 0x00b1
<kernel>:int_bottom + 0x001d (nearest)
iframe at 0x8013ad9c (end = 0x8013adf4)

eax 0x86d9c800 ebx 0x80131d68 ecx 0x1 edx 0x200246
esi 0x80131d88 edi 0x8013ae40 ebp 0x8013ae48 esp 0x8013add0
eip 0x86d9c800 eflags 0x210287
vector: 0xe, error code: 0x0

<kernel>:flush_pending_notificationsFv + 0x0069
<kernel>:block_notifier_and_writerFPv + 0x0055
<kernel>:_create_kernel_thread_kentryFv + 0x001b
<kernel>:thread_kthread_exitFv + 0x0000

This was on r24880.

Attachments

NotificationKDL2.PNG (332.1 kB) - added by bga on 04/24/08 17:39:05.
General Protection Exception

Change History

04/09/08 17:58:08 changed by anevilyak

I just realized this was with the old ide stack ; I'm usually using ata but reverted it to build an image for someone, then forgot to set it back. Not sure if that can influence it or not, but rebuilding with ata in place now to see if I can replicate it in that scenario or not.

04/09/08 20:46:02 changed by anevilyak

On further note, I've tried a complete svn checkout + rm -rf twice now with ata, and I cannot replicate this crash now ; perhaps some interplay/race condition in how the old IDE stack works?

04/10/08 03:10:41 changed by axeld

Judging from the stack crawl, it doesn't look like there is any obvious connection to ata vs. ide, at least.
This looks like the cache list was corrupted - I dunno what have caused this, though, could be theoretically everything...

04/10/08 14:20:58 changed by anevilyak

Yeah, I realize the crawl doesn't really point fingers at the ide stack directly, but I thought I'd point it out anyways since I cannot seem to replicate that crash again with the other stack. I thought perhaps it might be possible the ide stack was destroying a data block after having passed it to the cache or something along those lines ; wasn't certain how they interacted. I can try switching back to ide and see if it crashes consistently if you'd like.

04/10/08 16:23:24 changed by axeld

No need to, if it doesn't pop up anymore, we can just close this ticket. If it does, then we had reason to keep it open ;-)
I guess it'll come back. They all come back :-))

04/10/08 18:37:58 changed by anevilyak

Understood, will let you know if I run into it again since I'm probably going to be doing this kind of thing a lot the next few days :) If I do hit it again I'll leave it in the kernel debugger so you can let me know any possibly useful info I can try to trace out of it.

04/15/08 22:09:01 changed by ddew

I've just run in to what looks like the same issue on r24968 using the new ata stack. I've left it in KDL if you want me to run some tests or need more info.

04/23/08 19:46:42 changed by bonefish

#2150 is a dup of this one.

I also ran into the problem, but all I could find out is that the crash happened when the notification was removed from the list. The list had been corrupted.

The crash happened in a low-memory situation, BTW. I was running the OpenSSH test suite and due to the net buffer data header leak it had consumed virtually all memory. Furthermore another bug caused the test to write an ever-growing file at the same time.

04/24/08 06:40:21 changed by bga

I got the same crash and it was not in a low memory condition at all (unless we also consider low memory condition when the cache is taking up almost all available memory). Basically I booted Haiku and tried to create a 3 Gb file using DD in a partition that had 3.6 Gb available. After a while, I got the KDL. BTW, I managed to do it 2 in a row (didn't try more then 2 times tough) using the same steps. See bug #2151.

04/24/08 17:38:10 changed by bga

Got the crash again when checking out the source tree with svn from inside Haiku. The KDL was almost the same but the specific error it reported was different, so I am attaching a new screenshot just for reference.

04/24/08 17:39:05 changed by bga

  • attachment NotificationKDL2.PNG added.

General Protection Exception

04/28/08 04:50:11 changed by stippi

  • priority changed from normal to critical.
  • milestone changed from R1 to R1/alpha1.

This one is critical for R1/alpha.

05/12/08 06:56:49 changed by stippi

Still with us in r25463.