Avi Kivity summarized the kvm patches bound for the 2.6.25 kernel:
"Changes include performance and scalability improvements, completion of the portability work (though no new architectures are supported with this submission), support for new hardware features, using general userspace memory for kvm (which enables swapping guest memory as well as sharing memory among guests), as well as the usual cleanups and incremental fixes."
The Kernel-based Virtual Machine project, kvm, was started in mid-2006, and has been part of the Linux kernel since the 2.6.20 release in February of 2007. The recent changes can be browsed with gitweb.
From: Avi Kivity <avi@...> Subject: [GIT PULL] KVM updates for the 2.6.25 merge window Date: Jan 30, 1:18 pm 2008 Linus, Please pull the kvm updates for 2.6.25 from the repo and branch at: git://git.kernel.org/pub/scm/linux/kernel/git/avi/kvm.git for-linus Changes include performance and scalability improvements, completion of the portability work (though no new architectures are supported with this submission), support for new hardware features, using general userspace memory for kvm (which enables swapping guest memory as well as sharing memory among guests), as well as the usual cleanups and incremental fixes. Amit Shah (1): KVM: Make unloading of FPU state when putting vcpu arch-independent Anthony Liguori (6): KVM: x86 emulator: Add vmmcall/vmcall to x86_emulate (v3) KVM: Refactor hypercall infrastructure (v3) KVM: MMU: Clean up MMU functions to take struct kvm when appropriate KVM: MMU: More struct kvm_vcpu -> struct kvm cleanups KVM: Allocate userspace memory for older userspace KVM: Fix gfn_to_page() acquiring mmap_sem twice Avi Kivity (91): KVM: VMX: Further reduce efer reloads KVM: Allow not-present guest page faults to bypass kvm KVM: MMU: Make flooding detection work when guest page faults are bypassed KVM: VMX: Don't clear the vmcs if the vcpu is not loaded on any processor KVM: VMX: Simplify vcpu_clear() KVM: Move guest pte dirty bit management to the guest pagetable walker KVM: MMU: Fix nx access bit for huge pages KVM: MMU: Disable write access on clean large pages KVM: MMU: Instantiate real-mode shadows as user writable shadows KVM: MMU: Move dirty bit updates to a separate function KVM: MMU: When updating the dirty bit, inform the mmu about it KVM: Move apic timer interrupt backlog processing to common code KVM: Rename KVM_TLB_FLUSH to KVM_REQ_TLB_FLUSH KVM: MMU: Simplify page table walker KVM: Move vmx_vcpu_reset() out of vmx_vcpu_setup() KVM: Add a might_sleep() annotation to gfn_to_page() KVM: x86 emulator: don't depend on cr2 for mov abs emulation KVM: Move page fault processing to common code KVM: MMU: Topup the mmu memory preallocation caches before emulating an insn KVM: x86 emulator: Extract the common code of SrcReg and DstReg KVM: x86 emulator: centralize decoding of one-byte register access insns KVM: Simplify decode_register_operand() calling convention KVM: x86 emulator: Hoist modrm and abs decoding into separate functions KVM: VMX: Use vmx to inject real-mode interrupts KVM: VMX: Read & store IDT_VECTORING_INFO_FIELD KVM: Fix faults during injection of real-mode interrupts KVM: Simplify CPU_TASKS_FROZEN cpu notifier handling KVM: VMX: Consolidate register usage in vmx_vcpu_run() KVM: Replace 'light_exits' stat with 'host_state_reload' KVM: Add fpu_reload counter KVM: Add instruction emulation statistics KVM: Extend stats support for VM stats KVM: MMU: Add some mmu statistics KVM: MMU: Remove unused variable KVM: Export include/asm-x86/kvm.h KVM: x86 emulator: retire ->write_std() KVM: x86 emulator: prefetch up to 15 bytes of the instruction executed KVM: Split vcpu creation to avoid vcpu_load() before preemption setup KVM: MMU: Implement guest page fault bypass for nonpae KVM: Add statistic for remote tlb flushes KVM: MMU: Avoid unnecessary remote tlb flushes when guest updates a pte KVM: Don't bother the mmu if cr3 load doesn't change cr3 KVM: MMU: Introduce and use gpte_to_gfn() KVM: MMU: Move pse36 handling to the guest walker KVM: MMU: Remove extra gaddr parameter from set_pte_common() KVM: MMU: Remove set_pde() KVM: MMU: Merge set_pte() and set_pte_common() KVM: MMU: Adjust page_header_update_slot() to accept a gfn instead of a gpa KVM: MMU: Introduce gfn_to_gpa() KVM: MMU: Simplify nonpaging_map() KVM: MMU: Remove gva_to_hpa() KVM: Remove gpa_to_hpa() KVM: MMU: Rename variables of type 'struct kvm_mmu_page *' KVM: MMU: Rename 'release_page' KVM: Disallow fork() and similar games when using a VM KVM: x86 emulator: address size and operand size overrides are sticky KVM: Remove misleading check for mmio during event injection KVM: x86 emulator: Move rep processing before instruction execution KVM: x86 emulator: unify two switches KVM: x86 emulator: unify four switch statements into two KVM: Export include/linux/kvm.h only if $ARCH actually supports KVM KVM: Generalize exception injection mechanism KVM: Replace page fault injection by the generalized exception queue KVM: Replace #GP injection by the generalized exception queue KVM: Use generalized exception queue for injecting #UD KVM: x86 emulator: fix eflags preparation for emulation KVM: VMX: Avoid exit when setting cr8 if the local apic is in the kernel KVM: x86 emulator: Fix stack instructions on 64-bit mode KVM: SVM: Trap access to the cr8 register KVM: MMU: Simplify calculation of pte access KVM: MMU: Set nx bit correctly on shadow ptes KVM: MMU: Move pte access calculation into a helper function KVM: MMU: Fix inherited permissions for emulated guest pte updates KVM: MMU: No need to pick up nx bit from guest pte KVM: MMU: Pass pte dirty flag to set_pte() instead of calculating it on-site KVM: MMU: Remove walker argument to set_pte() KVM: MMU: Move set_pte() into guest paging mode independent code KVM: MMU: Adjust mmu_set_spte() debug code for gpte removal KVM: MMU: Use mmu_set_spte() for real-mode shadows KVM: Move arch dependent files to new directory arch/x86/kvm/ KVM: Move drivers/kvm/* to virt/kvm/ KVM: MMU: Add cache miss statistic KVM: Print data for unimplemented wrmsr KVM: local APIC TPR access reporting facility KVM: Accelerated apic support KVM: Disable vapic support on Intel machines with FlexPriority KVM: MMU: Avoid calling gfn_to_page() in mmu_set_spte() KVM: MMU: Move kvm_free_some_pages() into critical section KVM: Initialize the mmu caches only after verifying cpu support KVM: Fix unbounded preemption latency KVM: Move apic timer migration away from critical section Carlo Marcelo Arenas Belon (1): KVM: SVM: Remove KVM specific defines for MSR_EFER Carsten Otte (9): KVM: Move x86 msr handling to new files x86.[ch] KVM: Portability: split kvm_vcpu_ioctl KVM: Portability: Split kvm_vm_ioctl v3 KVM: Portability: Move memory segmentation to x86.c KVM: Portability: move get/set_apic_base to x86.c KVM: Portability: Move control register helper functions to x86.c KVM: Portability: Move kvm_get/set_msr[_common] to x86.c KVM: Portability: Move x86 emulation and mmio device hook to x86.c KVM: Portability: Move pio emulation functions to x86.c Christian Borntraeger (2): KVM: Use virtual cpu accounting if available for guest times. KVM: Per-architecture hypercall definitions Christian Ehrhardt (1): KVM: Portability: Move kvm_fpu to asm-x86/kvm.h Dan Kenigsberg (1): KVM: Enhance guest cpuid management Dong, Eddie (1): KVM: MMU: Merge shadow level check in FNAME(fetch) Dor Laor (1): KVM: Add make_page_dirty() to kvm_clear_guest_page() Eddie Dong (5): KVM: Export PIC reset for kernel device reset KVM: Split IOAPIC reset function and export for kernel RESET KVM: VMX: Comment VMX primary/secondary exec ctl definitions KVM: VMX: wbinvd exiting KVM: MMU: Coalesce remote tlb flushes Glauber de Oliveira Costa (1): KVM: Put kvm_para.h include outside __KERNEL__ Guillaume Thouvenin (2): KVM: x86 emulator: Make a distinction between repeat prefixes F3 and F2 KVM: x86 emulator: cmps instruction Hollis Blanchard (11): KVM: Portability: Make exported debugfs data architecture-specific KVM: Portability: Move x86 instruction emulation code to x86.c KVM: Portability: Move x86 FPU handling to x86.c KVM: Portability: Move x86 vcpu ioctl handlers to x86.c KVM: Remove unused "rmap_overflow" variable KVM: Correct consistent typo: "destory" -> "destroy" KVM: Move misplaced comment KVM: Portability: Move address types to their own header file KVM: Portability: Move IO device definitions to its own header file KVM: Portability: Stop including x86-specific headers in kvm_main.c KVM: Portability: Create kvm_arch_vcpu_runnable() function Izik Eidus (23): KVM: Remove the usage of page->private field by rmap KVM: Add general accessors to read and write guest memory KVM: Allow dynamic allocation of the mmu shadow cache size KVM: Support assigning userspace memory to the guest KVM: MMU: Add rmap_next(), a helper for walking kvm rmaps KVM: MMU: Keep a reverse mapping of non-writable translations KVM: MMU: Make gfn_to_page() always safe KVM: MMU: Partial swapping of guest memory KVM: Unmap kernel-allocated memory on slot destruction KVM: Export memory slot allocation mechanism KVM: Add kernel-internal memory slots KVM: Add ioctl to tss address from userspace, KVM: x86 emulator: remove 8 bytes operands emulator for call near instruction KVM: add kvm_is_error_hva() KVM: introduce gfn_to_hva() KVM: Change kvm_{read,write}_guest() to use copy_{from,to}_user() KVM: MMU: Change guest pte access to kvm_{read,write}_guest() KVM: Simplify kvm_clear_guest_page() KVM: MMU: Fix potential memory leak with smp real-mode KVM: MMU: Selectively set PageDirty when releasing guest memory KVM: MMU: Code cleanup KVM: MMU: mark pages that were inserted to the shadow pages table as accessed KVM: MMU: Fix dirty page setting for pages removed from rmap Jan Kiszka (1): KVM: VMX: Force seg.base == (seg.sel << 4) in real mode Jerone Young (8): KVM: Portability: Move kvm_memory_alias to asm/kvm.h KVM: Portability: Move x86 pic strutctures KVM: Portability: Move kvm_regs to <asm/kvm.h> KVM: Portability: Move structure lapic_state to <asm/kvm.h> KVM: Portability: Move kvm_segment & kvm_dtable structure to <asm/kvm.h> KVM: Portability: Move kvm_sregs and msr structures to <asm/kvm.h> KVM: Portability: Move cpuid structures to <asm/kvm.h> KVM: Add ifdef in irqchip struct for x86 only structures Joe Perches (1): KVM: Remove ptr comparisons to 0 Joerg Roedel (4): KVM: SVM: Emulate read/write access to cr8 KVM: SVM: Exit to userspace if write to cr8 and not using in-kernel apic KVM: LAPIC: minor debugging compile fix KVM: SVM: support writing 0 to K8 performance counter control registers Laurent Vivier (15): KVM: x86 emulator: remove unused functions KVM: x86 emulator: move all x86_emulate_memop() to a structure KVM: x86 emulator: move all decoding process to function x86_decode_insn() KVM: emulate_instruction() calls now x86_decode_insn() and x86_emulate_insn() KVM: Call x86_decode_insn() only when needed KVM: x86 emulator: split some decoding into functions for readability KVM: x86 emulator: remove _eflags and use directly ctxt->eflags. KVM: x86 emulator: Remove no_wb, use dst.type = OP_NONE instead KVM: Purify x86_decode_insn() error case management KVM: x86 emulator: Any legacy prefix after a REX prefix nullifies its effect KVM: Add some \n in ioapic_debug() KVM: VMX: Let gcc to choose which registers to save (x86_64) KVM: VMX: Let gcc to choose which registers to save (i386) KVM: SVM: Let gcc to choose which registers to save (x86_64) KVM: SVM: Let gcc to choose which registers to save (i386) Marcelo Tosatti (7): KVM: MMU: Remove unused prev_shadow_ent variable from fetch() KVM: MMU: Use cmpxchg for pte updates on walk_addr() KVM: MMU: Fix SMP shadow instantiation race KVM: MMU: emulated cmpxchg8b should be atomic on i386 KVM: MMU: Concurrent guest walkers KVM: Add kvm_read_guest_atomic() KVM: MMU: Switch to mmu spinlock Mike Day (1): KVM: CodingStyle cleanup Nitin A Kamble (2): KVM: x86 emulator: Implement emulation of instruction: inc & dec KVM: x86 emulator: cmc, clc, cli, sti Qing He (2): KVM: x86_emulator: no writeback for bt KVM: apic round robin cleanup Rusty Russell (3): KVM: Add kvm_free_lapic() to pair with kvm_create_lapic() KVM: Hoist kvm_create_lapic() into kvm_vcpu_init() KVM: Remove gratuitous casts from lapic.c Ryan Harper (2): KVM: MMU: Ignore reserved bits in cr3 in non-pae mode KVM: VMX: Add printk_ratelimit in vmx_intr_assist Sheng Yang (5): KVM: VMX: Enable memory mapped TPR shadow (FlexPriority) KVM: x86 emulator: modify 'lods', and 'stos' not to depend on CR2 KVM: VMX: Remove the secondary execute control dependency on irqchip KVM: x86 emulator: Rename 'cr2' to 'memop' KVM: x86 emulator: Only allow VMCALL/VMMCALL trapped by #UD Uri Lublin (1): KVM: Make mark_page_dirty() work for aliased pages too. Zhang Xiantao (40): KVM: Portability: Split kvm_vcpu into arch dependent and independent parts (part 1) KVM: Portability: Move some includes to x86.c KVM: Portability: Move kvm_x86_ops to x86.c KVM: Portability: Add vcpu and hardware management arch hooks KVM: Portability: Combine kvm_init and kvm_init_x86 KVM: Portability: Move x86 specific code from kvm_init() to kvm_arch() KVM: Portability: move KVM_CHECK_EXTENSION KVM: Portability: Make kvm_vcpu_ioctl_translate arch dependent KVM: Remove __init attributes for kvm_init_debug and kvm_init_msr_list KVM: Portability: Add two hooks to handle kvm_create and destroy vm KVM: Portability: Move kvm_vcpu_ioctl_get_dirty_log to arch-specific file KVM: Portability: MMU initialization and teardown split KVM: Portability: Move some macro definitions from kvm.h to x86.h KVM: Portability: Move struct kvm_x86_ops definition to x86.h KVM: Portability: Move vcpu regs enumeration definition to x86.h KVM: Move some static inline functions out from kvm.h into x86.h KVM: Portability: Move some function declarations to x86.h KVM: Recalculate mmu pages needed for every memory region change KVM: Portability: Split kvm_set_memory_region() to have an arch callout KVM: Portability: Move unalias_gfn to arch dependent file KVM: Portability: Move KVM_INTERRUPT vcpu ioctl to x86.c KVM: Correct kvm_init() error paths not freeing bad_pge. KVM: Replace kvm_lapic with kvm_vcpu in ioapic/lapic interface KVM: Replace dest_Lowest_Prio and dest_Fixed with self-defined macros KVM: Extend ioapic code to support iosapic KVM: Portability: Move kvm{pic,ioapic} accesors to x86 specific code KVM: Portability: Introduce kvm_vcpu_arch KVM: Portability: Split mmu-related static inline functions to mmu.h KVM: Portability: Move kvm_vcpu definition back to kvm.h KVM: Portability: Expand the KVM_VCPU_COMM in kvm_vcpu structure. KVM: Portability: Move kvm_vcpu_stat to x86.h KVM: Portability: Move memslot aliases to new struct kvm_arch KVM: Portability: Move mmu-related fields to kvm_arch KVM: Portability: move vpic and vioapic to kvm_arch KVM: Portability: Move round_robin_prev_vcpu and tss_addr to kvm_arch KVM: Portability: Move kvm_vm_stat to x86.h KVM: Move irqchip declarations into new ioapic.h and lapic.h KVM: Move ioapic code to common directory. KVM: Move kvm_vcpu_kick() to x86.c KVM: Expose ioapic to ia64 save/restore APIs npiggin@suse.de (1): KVM: Convert KVM from ->nopage() to ->fault() arch/x86/Kconfig | 3 + arch/x86/Makefile | 2 + {drivers => arch/x86}/kvm/Kconfig | 7 +- {drivers => arch/x86}/kvm/Makefile | 6 +- {drivers => arch/x86}/kvm/i8259.c | 8 +- {drivers => arch/x86}/kvm/irq.c | 22 +- arch/x86/kvm/irq.h | 88 + {drivers => arch/x86}/kvm/kvm_svm.h | 2 +- {drivers => arch/x86}/kvm/lapic.c | 216 +- arch/x86/kvm/lapic.h | 50 + arch/x86/kvm/mmu.c | 1885 +++++++++ arch/x86/kvm/mmu.h | 44 + arch/x86/kvm/paging_tmpl.h | 484 +++ arch/x86/kvm/segment_descriptor.h | 29 + {drivers => arch/x86}/kvm/svm.c | 353 +- {drivers => arch/x86}/kvm/svm.h | 3 +- {drivers => arch/x86}/kvm/vmx.c | 1079 +++--- {drivers => arch/x86}/kvm/vmx.h | 26 +- drivers/kvm/kvm_main.c => arch/x86/kvm/x86.c | 4243 +++++++++----------- arch/x86/kvm/x86_emulate.c | 1912 +++++++++ drivers/Kconfig | 2 - drivers/Makefile | 1 - drivers/kvm/irq.h | 165 - drivers/kvm/mmu.c | 1498 ------- drivers/kvm/paging_tmpl.h | 511 --- drivers/kvm/segment_descriptor.h | 17 - drivers/kvm/x86_emulate.c | 1662 -------- include/asm-x86/Kbuild | 1 + include/asm-x86/kvm.h | 191 + drivers/kvm/kvm.h => include/asm-x86/kvm_host.h | 537 +-- include/asm-x86/kvm_para.h | 105 + .../asm-x86/kvm_x86_emulate.h | 69 +- include/linux/Kbuild | 2 +- include/linux/kvm.h | 203 +- include/linux/kvm_host.h | 299 ++ include/linux/kvm_para.h | 82 +- include/linux/kvm_types.h | 54 + kernel/fork.c | 1 + {drivers => virt}/kvm/ioapic.c | 99 +- virt/kvm/ioapic.h | 95 + virt/kvm/iodev.h | 63 + virt/kvm/kvm_main.c | 1400 +++++++ 42 files changed, 9955 insertions(+), 7564 deletions(-) rename {drivers => arch/x86}/kvm/Kconfig (94%) rename {drivers => arch/x86}/kvm/Makefile (51%) rename {drivers => arch/x86}/kvm/i8259.c (98%) rename {drivers => arch/x86}/kvm/irq.c (81%) create mode 100644 arch/x86/kvm/irq.h rename {drivers => arch/x86}/kvm/kvm_svm.h (96%) rename {drivers => arch/x86}/kvm/lapic.c (83%) create mode 100644 arch/x86/kvm/lapic.h create mode 100644 arch/x86/kvm/mmu.c create mode 100644 arch/x86/kvm/mmu.h create mode 100644 arch/x86/kvm/paging_tmpl.h create mode 100644 arch/x86/kvm/segment_descriptor.h rename {drivers => arch/x86}/kvm/svm.c (84%) rename {drivers => arch/x86}/kvm/svm.h (98%) rename {drivers => arch/x86}/kvm/vmx.c (75%) rename {drivers => arch/x86}/kvm/vmx.h (96%) rename drivers/kvm/kvm_main.c => arch/x86/kvm/x86.c (52%) create mode 100644 arch/x86/kvm/x86_emulate.c delete mode 100644 drivers/kvm/irq.h delete mode 100644 drivers/kvm/mmu.c delete mode 100644 drivers/kvm/paging_tmpl.h delete mode 100644 drivers/kvm/segment_descriptor.h delete mode 100644 drivers/kvm/x86_emulate.c create mode 100644 include/asm-x86/kvm.h rename drivers/kvm/kvm.h => include/asm-x86/kvm_host.h (64%) create mode 100644 include/asm-x86/kvm_para.h rename drivers/kvm/x86_emulate.h => include/asm-x86/kvm_x86_emulate.h (83%) create mode 100644 include/linux/kvm_host.h create mode 100644 include/linux/kvm_types.h rename {drivers => virt}/kvm/ioapic.c (83%) create mode 100644 virt/kvm/ioapic.h create mode 100644 virt/kvm/iodev.h create mode 100644 virt/kvm/kvm_main.c --
without paravirtual drivers for Windows it won't succeed
Without paravirtual drivers for Windows, KVM will be just a toy for hackers and/or for tiny 1-hour test setups, nothing more.
KVM already has full virtualization...
... and paravirtualization is on its way.
http://en.wikipedia.org/wiki/Kernel-based_Virtual_Machine
And why would you even want paravirtualization when HVM is nearly ubiquitous on commodity hardware?
KVM can already boot Windows and half a dozen other OSs, it's only just getting started.
So... what exactly was your point again?
paravirtualization is different from being able to run some OS.
paravirtualization is different from being able to run some OS.
It is about:
* Running it faster
- By telling the guest about the host needs
- By telling the host about the guest needs
* Power management
* Allowing virtualization-specific new features, such as:
- copy/paste with the host
- hotplugging of new memory, cpu, pci, etc.
- mouse/keyboard integration between the host/guest
- folder sharing
- 3D
Ok, some of these are user-space centered and not driver paravirtualization, but you get the point.
The only 1 on that list that
The only 1 on that list that is not do able in HVM are near native speed 3D( you can do 3d in hvm it is just a bit slow), And hate to tell you but there are several usage patterns where a full hvm of windows out performs Windows running on the hardware. Not sure why, but I suspect it is do the hardware being written for the software and coding around the hangups. One case happens to be deploying autocad on windows. It like running in the HVM.
Being able to run another
Being able to run another instance of an OS on a vm is useful for various reasons, virtualization != running windows on a *nix box. So who the .... cares if windows doesn't have paravirtual drivers.
who cares? everyone cares
Being able to just run a virtual Windows is very different from running a virtual Windows with full performance possible.
Paravirtual drivers are needed to gain performance - without them, access to network and disks is emulated and is very slow (for a server usage, at least).
Define 'succeed' as you see
Define 'succeed' as you see it.
This is not VMWare. This is not Xen Enterprise (or whatever its called now). This is not commercial software.
Does a open source project only 'succeed' for you when you don't have to spend any cash?
"succeed", as in "popular"?
"succeed", as in "popular"?
> Does a open source project only 'succeed' for you when you don't have to spend any cash?
Huh? I wouldn't mind spending a couple of bucks on good PV Windows drivers, either for KVM or Xen.
Ah I see, without
Ah I see, without paravirtual drivers for Windows it won't be popular enough to succeed. I seriously doubt that.
Anyway I'm sure they'll get done at some point if a commercial product is being based on this.
Commerical on top of open source
In case you didn't know, a commercial product is being built on top of KVM. http://web1.qumranet.com/
So running multiple Linux
So running multiple Linux via KVM to support commercial applications in sand boxes is a "toy"?
For some people, apparently.
For some people, apparently. For the rest of us, it's a viable alternative and well worth considering. I guess some people have very specific requirements, or something. That's normal. Use the tool that works for you. KVM will work just fine for a lot of people.
*shrug*
Re: without paravirtual drivers for Windows it won't succeed
KVM as stays now is more than enough to run critical windows apps with the security of Virtual Machine Infraestructure.
Of course, it's not the fastest Virtualization technology ever but more than just enough to run 99% of business aplications.
Don't forget that is not difficult to buy a moderm desktop operating at 2 billons CPU cycles per second, with two cores, parallel operations per CPU cycle, 64 bits, 2-3 Gigabytes of RAM, and DVD burner for about 600$/400€. That's more than 50 times the power used by most enterprise hardware installed 10 years ago.
So whether KVM is 15% faster or slower is not a big issue most of the time.
I would claim all the opposite to you. KVM has all than needed for most people. Advanced paravirtualization, "copy and paste between VMs",... is just a toy for hackers.
VMware of Xen are really complex and not worth the effort from a bussiness point of view. Still some other light virtualization technologies like Virtuozzo or Solaris Containers can make lot of sense for telcos providing farms of virtual servers.
Hate to spoil the KVM party
Hate to spoil the KVM party but KVM's performance is terrible and it is no way suitable for running critical business applications. I work for a BIG software company and i did some testing with KVM and its performance was unbelievably terrible compared to other folks in virtualization business
I'll match your anecdotal
I'll match your anecdotal "evidence" with my own that it works just fine and with comparable performance when compared to the alternatives, and up the ante with a bet that you can't actually provide real figures for the statements you make in your comment. Feel (utterly) free to prove me wrong.
Facts are nice to have, as long as they're just that - facts.
Your comment is, unfortunately for your credibility, void of any.
(And no, I don't provide any of my own, either, which is part of the point. Duh.)
KVM vs others
I once heard for a speed comparison to be visually discernible it there had to be a twice or half difference in speed, In starting Windows XP and starting Adobe Elements under KVM and under VMWare, both using Intel hardware virtualization support, running under KVM is noticeably faster. What KVM doesn't do, as far as I can tell is sound and I haven't gotten the Windows VM (under KVM) to see external Windows shares....yet.