VMware Cloud Community
bartbogdan
Contributor
Contributor

Windows 7 32 bit VM won't boot after move from ESXi 5.1 to ESXi 5.5

Hi all,

Due to a failing ESXi 5.1 host we had to move some VM's another host running ESXi 5.5.

Our Linux VM's are running fine on 5.5 but our Windows 7 (32 bit) VM's will freeze after the login prompt.

The machines will boot into windows safe mode once. After that the machines will continuously reboot into BSOD's.

The machines were copied using scp from host to the other. The messages in the BSOD's will vary. This is one of them for example: STOP 0x0000F4.

I suspect that is has something to do with the storage controller. But the settings should be similar for both ESXi 5.1 and 5.5, so most tips in the community (regarding activation of the LSI driver) are not applicable.

Booting from the Windows 7 installation disk and performing a startup repair does not resolve the issue.

Anyone familiar with issues like this? We are running on embedded hypervisors on IBM x3530 M4 machines.

Kind regards,

Bart

Reply
0 Kudos
16 Replies
cykVM
Expert
Expert

Just a guess: Were VMWare Tools installed inside those VMs? Maybe uninstallig the 5.1 Tools before the move helps.

Reply
0 Kudos
bartbogdan
Contributor
Contributor

Thanks for the suggestion. I have removed the VMWare Tools installation, copied, and tried again. Windows still freezes after the login prompt.

Resetting the machine after that will result in a BSOD again.

PS. a VM made under ESXi 5.5 with a fresh install of Windows 7 does run without problems.

Reply
0 Kudos
cykVM
Expert
Expert

Another suggestion would be using VMWare converter instead of scp to copy the VMs over. This should take care of all relevant settings on the 5.5 server.

Reply
0 Kudos
a_p_
Leadership
Leadership

Please attach the VM's latest vmware.log file to a reply post to see whether it contains any hints on what's causing the issue.

André

Reply
0 Kudos
bartbogdan
Contributor
Contributor

I have attached the vmware.log file. You can see the "WinBSOD" occurrences in there. I am also trying to perform the copy with vCenter Converter, I will post the results when it is done.

Thanks,

Bart

Reply
0 Kudos
cykVM
Expert
Expert

The BSOD in the vmware.log file is a PAGE_FAULT_IN_NONPAGED_AREA which on a physical machine leads to somehow defective memory. Did you probably scp an existing *.vswp file over to the 5.5 server?

Maybe this (or another) file you copied with scp is corrupted?

Another thing to consider is snapshots, which probably are not copyable by scp.

Reply
0 Kudos
bartbogdan
Contributor
Contributor

I have now copied the machine using VMWare vCenter Converter. It still doesn't run properly, after the logon screen I receive a message saying that memory could not be written. See attached screenshot.

When I copy the machine from ESXi 5.1 to another ESXi 5.1 then it runs fine (also with scp).

Reply
0 Kudos
bartbogdan
Contributor
Contributor

Problem solved! I had to disable the memory hot add option for the VM. After that the machine runs fine.

Any ideas on why this option is not compatible? And why it was turned on by default?

Reply
0 Kudos
cykVM
Expert
Expert

In my 5.5 it's disabled by default. Maybe that's a bug in 5.1?

Or it was accidentally turned on as the Win7 VM was created?

Reply
0 Kudos
bartbogdan
Contributor
Contributor

Hi, it was already enabled in the original machine on ESXi 5.1. No idea why, it isn't enabled on other machines. Somehow this setting isn't problematic under 5.1, but it is under 5.5.

Reply
0 Kudos
cykVM
Expert
Expert

Very strange, just made a quick test with a Win7 64bit VM on my 5.5 server enabling hot add RAM feature and VM runs fine. no crahes or error messages. Maybe it does not fully work with 32bit?

Reply
0 Kudos
bartbogdan
Contributor
Contributor

I have also performed some tests:

* Fresh install of Win7 32 bit on ESXi with hot ram turned off -> No probems

* Fresh install of Win7 32 bit on ESXi with hot ram turned on -> Windows installer will fail with an error message saying that some .dll is not compatible with the system.

* Fresh install of Win7 32 bit on ESXi with hot ram turned off and then switched to on after installation -> Install ok, machine locked up once after the installation, will run after reboot (tested for 10 minutes).

So in the end it seems that ESXi 5.5 and hot ram addition is not compatible with Win7 32 bit (on my IBM platform). Even though this OS type was specifically selected upon creation of the VM.

Reply
0 Kudos
cykVM
Expert
Expert

Just out of curiosity I also made a

  • Fresh install of Win7 32 bit on ESXi with hot ram turned on

No error message during setup, nothing after logon and no BSOD. Reboot - no errors, crashes or freezes.

VM is running smooth and stable for over 30 minutes now.

Reply
0 Kudos
bartbogdan
Contributor
Contributor

Thanks a lot for testing! This will help in narrowing things down.

I have just tried again. I booted my test VM with hot ram turned on from the Windows install ISO. Then started a fresh install of Windows 7 32 bit, BSOD at the end of the installation process. (see attachments)

So it seems that it may be hardware related. Our machines running ESXi 5.5 are brand new and newer then our 5.1 machines. Another difference is that the new machines are equipped with 16GB RAM modules and our ESXi 5.1 machines only have 8GB modules.

These are the specs of our ESXi 5.5 machines:

IBM System x3530 M4 - 7160C3G

2x Intel Xeon CPU E5-2420v2 @ 2.2GHz

64GB RAM - DIMM1:8GB Samsung, DIMM3:8GB Hynix, DIMM5:16GB Samsung, DIMM7:8GB Hynix, DIMM9:8GB Hynix, DIMM11:16GB Samsung. There are 12 banks in total.

As you can see, it seems that the supplier constructed memory pairs with different brands (bank 1 and 7).

Our ESXi 5.1 machines are also x3530 M4 machines but an older type, 7160.

Kind regards,

Bart

Reply
0 Kudos
cykVM
Expert
Expert

Have no personal in depth experience with IBM servers, but the memory config looks a bit strange. Does the server give you any hints/warnings on boot/POST if the memory config is probably "not optimal"?

Another possible thing to consider: Are the 8GB and the 16GB models probably on different speeds?

Just took a quick look at: http://public.presalesadvisor.com/LiteratureUploads/Literature-1214.pdf

On page 7:

>Memory maximums:

>With LRDIMMs: Up to 384 GB with 12x 32 GB RDIMMs and two processors.

>With RDIMMs: Up to 192 GB with 12x 16 GB RDIMMs and two processors.

>With UDIMMs: Up to 96 GB with 12x 8 GB UDIMMs and two processors.

That sounds to me that on your system UDIMMs and RDIMMs are probably mixed?


>Memory:

>Up to 12 DIMM sockets (six DIMMs per processor). LRDIMMs, RDIMMs, and UDIMMs are

>supported, but the memory types cannot be intermixed. DIMM speeds up to 1600 MHz

If they are not mixed, the problem might be that Single-Rank 8GB RDIMMs and Dual-Rank 16GB RDIMMs are used - that might also cause problems.

(see page 13 onwards for further information). Also the DIMM per CPU (socket) should be considered.

Another addition:

Which memory mode is set in BIOS: Chipkill, Mirroring or Sparing?

(Page 12)

>Chipkill works only in independent channel mode (default operational mode) and supports only x4-based

>memory DIMMs.

>If memory mirroring is used, then DIMMs must be installed in pairs (a minimum of one pair per each

>processor, a maximum of two pairs per processor), and both DIMMs in a pair must be identical in type and

>size.

>If memory rank sparing is used, then two single-rank or dual-rank DIMMs must be installed per populated

>channel (the DIMMs do not need to be identical). In rank sparing mode, one rank of a DIMM in each

>populated channel is reserved as spare memory. The size of a rank varies depending on the DIMMs

>installed.

>Chipkill, memory mirroring, and memory rank sparing modes are mutually exclusive. Only one operational

>memory mode can be enabled on a server, and it is a system-wide setting.

At least the mixed Samsung/Hynix 8GB DIMMs are not a good idea.

Reply
0 Kudos
bartbogdan
Contributor
Contributor

Thanks for your answer, I am going to file this issue with our supplier and see if I can get the memory changed. I will post the results.

Reply
0 Kudos