Virtual satisfaction with VMware Server and kernel 2.6.31
Initially getting VMware Server to work with Fedora 11 and kernel 2.6.30 was challenging, and then the roughly bi-weekly kernel builds to keep up with Fedora updates got tedious. Trying the same approach with Fedora 12 and kernel 2.6.31 didn’t work at all for me. I kept getting duplicate definitions of init_mm that caused link failures. I tried various #ifdef kludges to overcome the duplicates, but nothing seemed to work. All this proved to me was that I really didn’t want to be trying to build the kernel at all.
Since the old ideas didn’t work, trying KVM seemed a likely alternative. However, my main interest in production virtual machines is running Windows 2000 Professional to run old Windows stuff that I need from time to time, stuff that isn’t likely to work with more recent versions of Windows. (I’ll probably still want Win2K for these purposes even when the support life cycle ends July 13. I still find use for NT4 Server for analogous reasons.) I don’t know whether the KVM project even intends to support legacy Windows. In any case, the Win2K install process seemed to get started OK, but then would not complete. It has been a couple of months since I tried this, so I don’t remember all the details.
While looking for solutions, I kept seeing praise for VirtualBox. The VirtualBox documentation showed promise, particularly the support for existing VMware partitions. I installed an rpm, tried to fire up an existing Win2K bootable partition, but the boot process didn’t get very far. I didn’t succeed in installing Win2K with VirtualBox, either.
Sometimes the right solution is to wait. Roughly once a week, or so, I’d search to see if there were new possibilities emerging.
About a month ago, I discovered How to install VMware Server on Ubuntu. Though not of direct help, it did link to Kang’s vmware-specific-specific 5.5.x and kernel 2.6.31. I’d seen and used some of Kang’s VMware Server patches before, but the comments on that page referred to init_mm problems that I didn’t know how to solve for Fedora 12.
Aack’s comment #32, posted on January 13, provided the simplest solution I’ve seen to the init_mm problem, simply replacing one line in pgtable.h to avoid the problem altogether. That change doesn’t require kernel builds, it gets picked up when vmware-config.pl builds modules specific to the running kernel. I didn’t notice the comment for some days after it was posted, and also wanted to do some disk reconfiguration before testing, but once I got to testing with the one line change, all seemed well.
So thanks to Kang and Aack, I’m running VMware Server 1.0.10 on Fedora 12 with no apparent difficulties. Kang’s comment #31, just before Aack’s praised VirtualBox, but Kang still seems to be preparing for kernel 2.6.32 (vmware-specific-specific 5.5.x and kernel 2.6.32), so maybe things will be easy enough by the time Fedora 13 comes out.