Solaris guest hang/crash using SMP -> retry with 3.1.4

[rasta]

HOST:

WinXP x64 with VT enabled (E5120 quad-core Xeon processor)

6 GB RAM

NVIDIA Quadro FX 3500 video card

VBox 3.1.0

GUEST:

Solaris 10 u8 x86 64-bit

2560 MB RAM assigned

32 MB vram assigned

PAE/VT-x/IO-APIC enabled

Solaris guest additions installed.

Guest will hang or crash to GRUB if >1 processor is assigned to guest in VM settings. During hang with 2 cpus, both cpus are pegged at 100%.

Other issues:

1) Upon guest boot, I get this error once per cpu asigned to guest:

WARNING: cpu0 failed to update microcode form version 0x0 to 0xcb.

cpu1 is referenced for second processor assigned, if any.

2) Unlike previous versions of Vbox with Solaris GA installed, the guest display will not automatically resize unless screen resolution is entered into xorg.conf.

[rasta]

Log of hung Solaris guest using 2 processors after forced power-off

[rasta]

See also:

​http://forums.virtualbox.org/viewtopic.php?f=20&t=25701&sid=397f2f01315f87d9e6ac30c39f289b3d

[rasta]

Myself and SSCBrian have noted that a guest crash (or spontaneous reboot) occurs after using Sun Update Manager when using 2 processors.

[rasta]

After a crash/reboot encountered while running a large application, I typically get WARNINGS with "No SOF interrupts have been received, this USB EHCI host controller is unusable." The guest as a whole is unusable at this point, and the VM must be killed and restarted.

[rasta]

Same problem with Vbox 3.1.2. With 2 processors, 3.1.2 is completely unstable. Guests will hang and/or crash every time. Will attach log.

[rasta]

Log of hung Solaris guest using 2 processors after forced power-off - Vbox 3.1.2

[Sander van Leeuwen]

1) ignore it
2) works fine here

I can't reproduce any hangs or crashes during booting. Will try the update manager.

[rasta]

The hangs and crashes ARE NOT during booting. Booting is fine with 2 processors. They occur either during Update Manager use or during the running of a large application.

Here is a thread about the guest screen resizing problem:

​http://forums.virtualbox.org/viewtopic.php?f=20&t=25559

Even after adding my screen resolution to xorg.conf, the main Solaris login screen remains shrunken (1024x?) from my 1280x1024 screen. Others have had only intermittent problems with resizing. Do you have any ideas as to how I can diagnose/fix this? This problem only occurs on the x64 host system outlined above. This problem is minor compared to the guest hang/crash issue.

[Sander van Leeuwen]

Retry with 3.1.4. That version will include an important stability fix for SMP guests.

[rasta]

Excellent! Thank you for the update!

[Sander van Leeuwen]

You can try the new test build in #6070 if you like.

[Sander van Leeuwen]

Please check if 3.1.4 beta 1 solves the problem: ​http://forums.virtualbox.org/viewtopic.php?f=15&t=27300

[rasta]

Will do.

[rasta]

So far, the 3.1.4 beta (with Solaris additions installed) has not only not crashed or hung using 2 processors, but it is blindingly fast. I have never seen my Solaris guest move so fast. It is just AWESOME. I notice that one SMP processor is assigned to each of my two (quad) cores, unlike before, where both were on one core.

I have found no problems with my huge application, Update Manager, or anything else.

Also, I mentioned above the error on booting ("WARNING: cpu0 failed to update microcode form version 0x0 to 0xcb.") I moved the VM to my main machine (XP x64, dual Xeon E5345 processors, 6GB RAM) and that error is gone, whether I use one or more processors.

Also, with the 3.1.4 beta Solaris GA, so far I have NOT seen the screen resizing issue also mentioned above.

So far, it appears you guys (and gals?) have hit the ball out of the park on this one. Way out. Great job! I'll report back if I find any problems.

By the way, is Nested Paging supported on my processors?

[Klaus Espenlaub]

Nested paging support on Intel processor started with the Nehalem architecture, i.e. Xeon 55xx and Core i7. Some of the "stripped down" versions don't have nested paging.

So no luck.

[rasta]

That's what I thought. My guest is now so fast it doesn't matter anyway. Maybe Nested Paging should be disabled in the VM GUI for processors that don't support it?

[Sander van Leeuwen]

You can ignore the microcode warning. We do not emulate this mechanism, which results in some confusion in the Solaris AMD microcode update code. It's completely harmless.

Great to hear things have improved so much for you. Unfortunately you can only check for nested paging support with supervisor code. We'll fix that in a future release.

[rasta]

Yes, you already said I can ignore the warning, but the warning does not occur on my main system, only the older system with an E5120 processor.

[rasta]

The guest screen resize issue mentioned above appears to be exactly the same as that reported in bug #5372, fixed in 3.1.4 beta 2.