I got the same error messages, when i try to migrate any online or offline machines from ESX 3.5 Update 1 ( except the june 3 patches )
to ESX 3.5 Update 1 including the June 3 patches
I have Intel duo-core HP ML370G5 servers.
I just tried another host in a different farm with the same results. This host and the others in the second farm are all the same as well. Dell PE6850's. The first farm was Dell R900's.
We did notice that the the VMs that were located on the host to be patched and offline before the patches, power on with no warnings after the patches. But vmotions brings the same error as before. All our vmotions were done online. I'll try an offline vmotion tomorrow.
Have you come up with a plan to remove the CPUID mask? Do you know what it is masking?
This is a known and pretty much benign message. It started with 3.5, and it appears that when you have a difference in version, you get this vmotion warning. I started seeing it when I vmotion'd a vm from my 3.0.2 servers which were not yet updated to 3.5, and 3.5U1 servers.
-KjB
I found KB Article #1003770 which doesn't give me a real good answer, IMO. After this mask is set, what feature are we missing out on down the road?
Is there a good way to tell which machines have the mask and which do not? We've been manually removing the mask when possible, but it requires a power off, which is strange because doing a live vmotion is when it gets set.
I think vmware is setting themselves up for a broader fix down the road. It is difficult to figure out which feature it is exactly that they mask out, but I wouldn't think you would need to remove it. As the kb article is stating, it will get set and unset as the vmotion occurs, so I would leave it alone. Just my oppinion.
-KjB
John,
The feature bit in question is called DCA, which stands for direct cache access. This feature bit was only recently defined and used by CPU vendors, so the older release of ESX that you are using enforces a strict vmotion policy for the feature bit. Newer versions of ESX have determined that this bit is not something that should prevent vmotion, so the new policy is to mask it from the guest and allow it to vary between your source and destination hosts when performing a vmotion.
In general, it is safe to remove these types of masks if you see them, but you should only do so when the VM is powered off. Their main purpose is to prevent unsafe vmotions between differing versions of ESX, and to allow future vmotions to follow the same rules as the VM used previously.
In this particular case, it is best to remove the mask, because it will allow slightly more flexible vmotions without it. If you intend to continue to vmotion this VM between the two ESX versions, however, I would recommend changing the "R" to a "0" (zero), so that you don't get this warning message and to allow the more flexible vmotion behavior no matter where the VM is started. (Note that there may be other vmotion CPUID policy differences between other versions of ESX that this is not the right solution for.)
We are actively looking to improve this mechanism in an upcoming release. Instead, we will track the necessary state internally, and when VMs are power cycled, they will pick up the behavior of the new host automatically.
Please let me know if you have any further questions.
Thanks,
Doug
Thread moved to the VI: ESX 3.5 forum
Tom Howarth
VMware Communities User Moderator