[gentoo-user] Gentoo segfaults on virtualbox-4.1.14 when running on AMD bulldozer

[gentoo-user] Gentoo segfaults on virtualbox-4.1.14 when running on AMD bulldozer

[Markos Chandras]

Hi all

This is a copy of the thread I recently opened on Gentoo forums[1].

I replaced my Phenom II cpu with a new 6-core AMD bulldozer. However, I
noticed that all of my Gentoo virtual machines throw (compiler)
segmentation faults when building or running any application. This
happens on a Gentoo testing VM, on a Gentoo stable VM and on a brand new
Gentoo installation. I have no idea where to look for that problem.

System:

Motherboard: Asus M5A78L-M/USB3
CPU: AMD FX(tm)-6100 Six-Core Processor
Host: Gentoo x86_64 testing (has not problem at all, so I presume HW
problem is eliminated)

Any ideas?

-- 
Regards,
Markos Chandras / Gentoo Linux Developer / Key ID: B4AFF2C2

[gentoo-user] Re: Gentoo segfaults on virtualbox-4.1.14 when running on AMD bulldozer

[Markos Chandras]

On 04/27/2012 08:35 PM, Markos Chandras wrote:
> Hi all
> 
> This is a copy of the thread I recently opened on Gentoo forums[1].
> 
> I replaced my Phenom II cpu with a new 6-core AMD bulldozer. However, I
> noticed that all of my Gentoo virtual machines throw (compiler)
> segmentation faults when building or running any application. This
> happens on a Gentoo testing VM, on a Gentoo stable VM and on a brand new
> Gentoo installation. I have no idea where to look for that problem.
> 
> System:
> 
> Motherboard: Asus M5A78L-M/USB3
> CPU: AMD FX(tm)-6100 Six-Core Processor
> Host: Gentoo x86_64 testing (has not problem at all, so I presume HW
> problem is eliminated)
> 
> Any ideas?
> 
On the other hand, these virtual machines work fine on a Windows 7
64-bit host

-- 
Regards,
Markos Chandras / Gentoo Linux Developer / Key ID: B4AFF2C2

[gentoo-user] Re: Gentoo segfaults on virtualbox-4.1.14 when running on AMD bulldozer

[Nikos Chantziaras]

On 27/04/12 22:35, Markos Chandras wrote:
> I replaced my Phenom II cpu with a new 6-core AMD bulldozer. However, I
> noticed that all of my Gentoo virtual machines throw (compiler)
> segmentation faults when building or running any application.

How do you get it to boot?  How do you manage to log in?  Are the 
binaries from the stage tarball not affected?

Re: [gentoo-user] Re: Gentoo segfaults on virtualbox-4.1.14 when running on AMD bulldozer

[Markos Chandras]

On 04/27/2012 11:45 PM, Nikos Chantziaras wrote:
> On 27/04/12 22:35, Markos Chandras wrote:
>> I replaced my Phenom II cpu with a new 6-core AMD bulldozer. However, I
>> noticed that all of my Gentoo virtual machines throw (compiler)
>> segmentation faults when building or running any application.
> 
> How do you get it to boot?  How do you manage to log in?  Are the
> binaries from the stage tarball not affected?
> 
> 
What do you mean? I was using these VMs before I replace my CPU and I
was kinda expecting to keep working on my new CPU as well. The same VMs
work on the same host when using Windows 7 64-bit as host OS and not
Gentoo. I also tried to install Gentoo on a fresh new VM and I got a gcc
segmentation fault right in the middle of the kernel compilation. So,
yes stage binaries are affected as well. Is that what you were asking?

-- 
Regards,
Markos Chandras / Gentoo Linux Developer / Key ID: B4AFF2C2

Re: [gentoo-user] Re: Gentoo segfaults on virtualbox-4.1.14 when running on AMD bulldozer

[Matthew Marlowe]

On Fri, Apr 27, 2012 at 3:54 PM, Markos Chandras <hwoarang@gentoo.org> wrote:
> On 04/27/2012 11:45 PM, Nikos Chantziaras wrote:
>> On 27/04/12 22:35, Markos Chandras wrote:
>>> I replaced my Phenom II cpu with a new 6-core AMD bulldozer. However, I
>>> noticed that all of my Gentoo virtual machines throw (compiler)
>>> segmentation faults when building or running any application.
>>

I'm not familiar with virtualbox, but I've seen similar issues occur
with vmware and the solution was to at least temporarily mask whatever
new cpu flags the new hypervisor was passing to the guest.  In vmware,
one could limit the cpu flags to maintain compatibility with various
cpu releases which was especially helpful in clusters.... Yes, your
gentoo vms should have been fine ..but at least until you track down
the issue, see if virtualbox has a similar feature?

-- 
Matthew Marlowe
matt@professionalsysadmin.com
https://www.twitter.com/deploylinux
1-805-857-9144

[gentoo-user] Re: Gentoo segfaults on virtualbox-4.1.14 when running on AMD bulldozer

[walt]

On 04/27/2012 12:35 PM, Markos Chandras wrote:
> Hi all
> 
> This is a copy of the thread I recently opened on Gentoo forums[1].
> 
> I replaced my Phenom II cpu with a new 6-core AMD bulldozer. However, I
> noticed that all of my Gentoo virtual machines throw (compiler)
> segmentation faults when building or running any application. This
> happens on a Gentoo testing VM, on a Gentoo stable VM and on a brand new
> Gentoo installation. I have no idea where to look for that problem.

VirtualBox 4.1.14 just landed on ~amd64 today and I haven't had time to
test it yet.  How long ago did you upgrade VirtualBox?

Re: [gentoo-user] Re: Gentoo segfaults on virtualbox-4.1.14 when running on AMD bulldozer

[Markos Chandras]

On 04/28/2012 01:24 AM, Matthew Marlowe wrote:
> On Fri, Apr 27, 2012 at 3:54 PM, Markos Chandras
> <hwoarang@gentoo.org> wrote:
>> On 04/27/2012 11:45 PM, Nikos Chantziaras wrote:
>>> On 27/04/12 22:35, Markos Chandras wrote:
>>>> I replaced my Phenom II cpu with a new 6-core AMD bulldozer.
>>>> However, I noticed that all of my Gentoo virtual machines
>>>> throw (compiler) segmentation faults when building or running
>>>> any application.
>>> 
> 
> I'm not familiar with virtualbox, but I've seen similar issues
> occur with vmware and the solution was to at least temporarily mask
> whatever new cpu flags the new hypervisor was passing to the guest.
> In vmware, one could limit the cpu flags to maintain compatibility
> with various cpu releases which was especially helpful in
> clusters.... Yes, your gentoo vms should have been fine ..but at
> least until you track down the issue, see if virtualbox has a
> similar feature?
> 
I don't see an obvious way to manipulate the cpu flags in virtualbox

-- 
Regards,
Markos Chandras / Gentoo Linux Developer / Key ID: B4AFF2C2

Re: [gentoo-user] Re: Gentoo segfaults on virtualbox-4.1.14 when running on AMD bulldozer

[Markos Chandras]

On 04/28/2012 01:36 AM, walt wrote:
> On 04/27/2012 12:35 PM, Markos Chandras wrote:
>> Hi all
>>
>> This is a copy of the thread I recently opened on Gentoo forums[1].
>>
>> I replaced my Phenom II cpu with a new 6-core AMD bulldozer. However, I
>> noticed that all of my Gentoo virtual machines throw (compiler)
>> segmentation faults when building or running any application. This
>> happens on a Gentoo testing VM, on a Gentoo stable VM and on a brand new
>> Gentoo installation. I have no idea where to look for that problem.
> 
> VirtualBox 4.1.14 just landed on ~amd64 today and I haven't had time to
> test it yet.  How long ago did you upgrade VirtualBox?
> 
> 
> 
I updated yesterday but the same thing happened in 4.1.12 as well

-- 
Regards,
Markos Chandras / Gentoo Linux Developer / Key ID: B4AFF2C2

Re: [gentoo-user] Re: Gentoo segfaults on virtualbox-4.1.14 when running on AMD bulldozer

[Markos Chandras]

On 04/28/2012 01:24 AM, Matthew Marlowe wrote:
> On Fri, Apr 27, 2012 at 3:54 PM, Markos Chandras <hwoarang@gentoo.org> wrote:
>> On 04/27/2012 11:45 PM, Nikos Chantziaras wrote:
>>> On 27/04/12 22:35, Markos Chandras wrote:
>>>> I replaced my Phenom II cpu with a new 6-core AMD bulldozer. However, I
>>>> noticed that all of my Gentoo virtual machines throw (compiler)
>>>> segmentation faults when building or running any application.
>>>
> 
> I'm not familiar with virtualbox, but I've seen similar issues occur
> with vmware and the solution was to at least temporarily mask whatever
> new cpu flags the new hypervisor was passing to the guest.  In vmware,
> one could limit the cpu flags to maintain compatibility with various
> cpu releases which was especially helpful in clusters.... Yes, your
> gentoo vms should have been fine ..but at least until you track down
> the issue, see if virtualbox has a similar feature?
> 
Interestingly this seems to be caused when using my wireless card to
bridge the virtualbox interfaces onto. I can't reproduce (yet) any
segfaults when I use the onboard ethernet card. I have the following
wireless card (supported by the rtl8180 kernel module)

04:06.0 Ethernet controller: Belkin F5D7000 v7000 Wireless G Desktop
Card [Realtek RTL8185] (rev 20)

-- 
Regards,
Markos Chandras / Gentoo Linux Developer / Key ID: B4AFF2C2

Re: [gentoo-user] Re: Gentoo segfaults on virtualbox-4.1.14 when running on AMD bulldozer

[Michael Mol]

[-- Attachment #1: Type: text/plain, Size: 1628 bytes --]

Neat. Random guess, but it could be a bug in Bulldozer's memory controller
or IOMMU. Try disabling IOMMU support in your kernel?
On Apr 29, 2012 3:29 PM, "Markos Chandras" <hwoarang@gentoo.org> wrote:

> On 04/28/2012 01:24 AM, Matthew Marlowe wrote:
> > On Fri, Apr 27, 2012 at 3:54 PM, Markos Chandras <hwoarang@gentoo.org>
> wrote:
> >> On 04/27/2012 11:45 PM, Nikos Chantziaras wrote:
> >>> On 27/04/12 22:35, Markos Chandras wrote:
> >>>> I replaced my Phenom II cpu with a new 6-core AMD bulldozer. However,
> I
> >>>> noticed that all of my Gentoo virtual machines throw (compiler)
> >>>> segmentation faults when building or running any application.
> >>>
> >
> > I'm not familiar with virtualbox, but I've seen similar issues occur
> > with vmware and the solution was to at least temporarily mask whatever
> > new cpu flags the new hypervisor was passing to the guest.  In vmware,
> > one could limit the cpu flags to maintain compatibility with various
> > cpu releases which was especially helpful in clusters.... Yes, your
> > gentoo vms should have been fine ..but at least until you track down
> > the issue, see if virtualbox has a similar feature?
> >
> Interestingly this seems to be caused when using my wireless card to
> bridge the virtualbox interfaces onto. I can't reproduce (yet) any
> segfaults when I use the onboard ethernet card. I have the following
> wireless card (supported by the rtl8180 kernel module)
>
> 04:06.0 Ethernet controller: Belkin F5D7000 v7000 Wireless G Desktop
> Card [Realtek RTL8185] (rev 20)
>
> --
> Regards,
> Markos Chandras / Gentoo Linux Developer / Key ID: B4AFF2C2
>
>

[-- Attachment #2: Type: text/html, Size: 2102 bytes --]

Re: [gentoo-user] Re: Gentoo segfaults on virtualbox-4.1.14 when running on AMD bulldozer

[Markos Chandras]

On 04/30/2012 02:33 PM, Michael Mol wrote:
> Neat. Random guess, but it could be a bug in Bulldozer's memory 
> controller or IOMMU. Try disabling IOMMU support in your kernel?
> 
> On Apr 29, 2012 3:29 PM, "Markos Chandras" <hwoarang@gentoo.org 
> <mailto:hwoarang@gentoo.org>> wrote:
> 
> On 04/28/2012 01:24 AM, Matthew Marlowe wrote:
>> On Fri, Apr 27, 2012 at 3:54 PM, Markos Chandras
> <hwoarang@gentoo.org <mailto:hwoarang@gentoo.org>> wrote:
>>> On 04/27/2012 11:45 PM, Nikos Chantziaras wrote:
>>>> On 27/04/12 22:35, Markos Chandras wrote:
>>>>> I replaced my Phenom II cpu with a new 6-core AMD
>>>>> bulldozer.
> However, I
>>>>> noticed that all of my Gentoo virtual machines throw
>>>>> (compiler) segmentation faults when building or running any
>>>>> application.
>>>> 
>> 
>> I'm not familiar with virtualbox, but I've seen similar issues
>> occur with vmware and the solution was to at least temporarily
>> mask whatever new cpu flags the new hypervisor was passing to the
>> guest.  In vmware, one could limit the cpu flags to maintain
>> compatibility with various cpu releases which was especially
>> helpful in clusters.... Yes, your gentoo vms should have been
>> fine ..but at least until you track down the issue, see if
>> virtualbox has a similar feature?
>> 
> Interestingly this seems to be caused when using my wireless card
> to bridge the virtualbox interfaces onto. I can't reproduce (yet)
> any segfaults when I use the onboard ethernet card. I have the
> following wireless card (supported by the rtl8180 kernel module)
> 
> 04:06.0 Ethernet controller: Belkin F5D7000 v7000 Wireless G
> Desktop Card [Realtek RTL8185] (rev 20)
> 
> -- Regards, Markos Chandras / Gentoo Linux Developer / Key ID:
> B4AFF2C2
> 
AMD IOMMU (Device Drivers -> Hardware IOMMU ) makes no difference. I
will have to move the discussion to virtualbox forums/ML as this seems
a driver or virtualbox problem. Bridge networking on that wireless
card work flawlessly when using Windows 7 as host.

-- 
Regards,
Markos Chandras / Gentoo Linux Developer / Key ID: B4AFF2C2

Re: [gentoo-user] Re: Gentoo segfaults on virtualbox-4.1.14 when running on AMD bulldozer

[Michael Mol]

On Mon, Apr 30, 2012 at 3:03 PM, Markos Chandras <hwoarang@gentoo.org> wrote:
> On 04/30/2012 02:33 PM, Michael Mol wrote:
>> Neat. Random guess, but it could be a bug in Bulldozer's memory
>> controller or IOMMU. Try disabling IOMMU support in your kernel?
>>
>> On Apr 29, 2012 3:29 PM, "Markos Chandras" <hwoarang@gentoo.org
>> <mailto:hwoarang@gentoo.org>> wrote:
>>
>> On 04/28/2012 01:24 AM, Matthew Marlowe wrote:
>>> On Fri, Apr 27, 2012 at 3:54 PM, Markos Chandras
>> <hwoarang@gentoo.org <mailto:hwoarang@gentoo.org>> wrote:
>>>> On 04/27/2012 11:45 PM, Nikos Chantziaras wrote:
>>>>> On 27/04/12 22:35, Markos Chandras wrote:
>>>>>> I replaced my Phenom II cpu with a new 6-core AMD
>>>>>> bulldozer.
>> However, I
>>>>>> noticed that all of my Gentoo virtual machines throw
>>>>>> (compiler) segmentation faults when building or running any
>>>>>> application.
>>>>>
>>>
>>> I'm not familiar with virtualbox, but I've seen similar issues
>>> occur with vmware and the solution was to at least temporarily
>>> mask whatever new cpu flags the new hypervisor was passing to the
>>> guest.  In vmware, one could limit the cpu flags to maintain
>>> compatibility with various cpu releases which was especially
>>> helpful in clusters.... Yes, your gentoo vms should have been
>>> fine ..but at least until you track down the issue, see if
>>> virtualbox has a similar feature?
>>>
>> Interestingly this seems to be caused when using my wireless card
>> to bridge the virtualbox interfaces onto. I can't reproduce (yet)
>> any segfaults when I use the onboard ethernet card. I have the
>> following wireless card (supported by the rtl8180 kernel module)
>>
>> 04:06.0 Ethernet controller: Belkin F5D7000 v7000 Wireless G
>> Desktop Card [Realtek RTL8185] (rev 20)
>>
>> -- Regards, Markos Chandras / Gentoo Linux Developer / Key ID:
>> B4AFF2C2
>>
> AMD IOMMU (Device Drivers -> Hardware IOMMU ) makes no difference. I
> will have to move the discussion to virtualbox forums/ML as this seems
> a driver or virtualbox problem. Bridge networking on that wireless
> card work flawlessly when using Windows 7 as host.

Report back when you find out the meat of the problem. I'm intensely curious.

-- 
:wq

Re: [gentoo-user] Gentoo segfaults on virtualbox-4.1.14 when running on AMD bulldozer

[Alan McKinnon]

On Fri, 27 Apr 2012 20:35:47 +0100
Markos Chandras <hwoarang@gentoo.org> wrote:

> Hi all
> 
> This is a copy of the thread I recently opened on Gentoo forums[1].
> 
> I replaced my Phenom II cpu with a new 6-core AMD bulldozer. However,
> I noticed that all of my Gentoo virtual machines throw (compiler)
> segmentation faults when building or running any application. This
> happens on a Gentoo testing VM, on a Gentoo stable VM and on a brand
> new Gentoo installation. I have no idea where to look for that
> problem.
> 
> System:
> 
> Motherboard: Asus M5A78L-M/USB3
> CPU: AMD FX(tm)-6100 Six-Core Processor
> Host: Gentoo x86_64 testing (has not problem at all, so I presume HW
> problem is eliminated)
> 
> Any ideas?
> 

I ran into the same system here: VMs crash soon after being launched
when they first start doing some real work. I only tested Gentoo guests
though.

But, I have Intel i7 cpu.

This is an ~amd64 (unstable)  host, running amd64 (stable) guests.

The only clue is in messages:

May  1 16:25:01 khamul kernel: [73366.634971] VBoxHeadless[2642]:
segfault at 10 ip 00007fa35e40599f sp 00007fa35c5b8b00 error 4 in
VBoxDD.so[7fa35e311000+144000]

I downgraded virtual box from 4.1.14 to 4.1.12 and initial
tests are good. A guest is doing a full emerge world and the first
package is glibc. Previously it wouldn't even manage to download
sources fully.

It's too early to draw any real conclusions though. I'm still rather
suspicious of glibc on the host (just as hunch, as that version seems
to be popping up in many more bug reports than usual)


-- 
Alan McKinnnon
alan.mckinnon@gmail.com