segfault in VBoxHeadless 4.1.2 while doing clonehd

[kerlerm]

Hi!

I was executing a "clonehd" command. While the system was under that load a segfault occured in VBoxHeadless. Host is a FC15 X64 system with several guest on it. Linux guests showed ATA bus resets when the error occurred, some windows guests were aborted.

I wonder if this line in the logs might be connected with that? EXT4-fs (dm-2): Unaligned AIO/DIO on inode 12845094 by VBoxHeadless; performance will be poor.

The same configuration worked flawlessly with FC12 - VBox 3.2.12

Any hints / workarounds? Thanks a lot!

Cheers Martin

[kerlerm]

VBox.log & more info

[kerlerm]

Today I updated to vbox 4.1.4. Same problem persists. Guests get randomly aborted - not only while the server is under high load. Seems to be connected to bug ticket no. 9661: https://www.virtualbox.org/ticket/9661

[kerlerm]

problem persists with version 4.1.6 installed. Guests get randomly aborted with ATA errors:

00:01:21.746 PCNet#0: Init: ss32=1 GCRDRA=0x01a46420[64]
GCTDRA=0x01a46020[64]
00:01:28.332 PIT: mode=2 count=0x4ad (1197) - 996.81 Hz (ch=0)
00:01:58.016 AHCI#0: Canceled write at offset 3803667968 (512 bytes
left) returned rc=VINF_SUCCESS
00:02:11.307 AHCI#0: Canceled read at offset 9197748224 (1024 bytes
left) returned rc=VINF_SUCCESS

[kerlerm]

problem still persists with 4.1.8 guests get ramdomly aborted:

00:49:12.629 AHCI#0: Canceled read at offset 9866681856 (2048 bytes
left) returned rc=VINF_SUCCESS
00:49:12.638 AHCI#0: Canceled read at offset 9866677760 (2048 bytes
left) returned rc=VINF_SUCCESS

Can I do anything to fix this bug?

[Frank Mehnert]

Are you still able to reproduce this bug? How easy is it for you? We fixed a bug which could be related to your problem. Would you be willing to try a test build?

[kerlerm]

Hi Frank!

Sorry for the late answer, but I'm just back from holiday.

I'm still able to reproduce this bug with version 4.1.16 installed. Usually happening under heavy load. Happens less often, when my server is running for serveral weeks. Reproducing should be easy for me.

Is the test build available in rpm format for fedora 15 64-bit?

Best Regards

[kerlerm]

Yesterday I tried to limit IO bandwidth of the guests to 6 MB/s. Things got even worse then. Guests died every few minutes. Is there a problem with IO scheduling?

[aeichner]

Can you attach a VBox.log of the VM please?

[kerlerm]

VBox log of crash

[kerlerm]

Is this bug fixed in 4.1.18?

Update notes vbox 4.1.18: "AHCI: fixed a rare bug which can cause a guest memory corruption after the guest storage controler has been reset"

I'll have a try...

[kerlerm]

Problem persists with 4.1.18 installed...

[kerlerm]

console output after crash

[aeichner]

Please attach the VBox.log of the crashed VM.

[kerlerm]

I'll add another 120713_VBox.log.

Console shows the following when the guests are crashing:

Jul 13 16:47:30 kernel: [1815837.189720] VBoxHeadless[14075]: segfault at 2b0 ip 00000000000002b0 sp 00007fcc3479aa68 error 14 in VBoxHeadless[400000+6000]

Jul 13 16:57:50 kernel: [1816457.629478] VBoxHeadless[16115] trap int3 ip:7f016c9020ef sp:7f0136db6a70 error:0[[BR]]

Jul 13 17:01:51 kernel: [1816698.664984] VBoxHeadless[17039]: segfault at 1b78000140 ip 00007fbe9ebca7b4 sp 00007fbe917d29c0 error 4 in libc-2.14.1.so[7fbe9eb52000+190000]

[kerlerm]

machine log

[kerlerm]

Any news on this?

Is this connected to https://www.virtualbox.org/ticket/9975 ?

There's definitely something fishy in Vbox AHCI code and I'm looking forward to a final solution soon. Can I assist in any way?

[kerlerm]

Hi Frank!

Will this issue be resolved in the upcoming 4.2.0 release?

[kerlerm]

Hi Frank!

FYI: Release 4.1.20 didn't solve the issue.

Regards Martin

[aeichner]

Please reopen if still relevant with a recent VirtualBox release.