KVM - User contributions [en]

TODO

2013-10-16T04:02:16Z

ConradPino: force TOC location

=ToDo=
__TOC__
The following items need some love. Please post to the list if you are interested in helping out:

* Emulate MSR_IA32_DEBUGCTL for guests which use it
* Bring up Windows 95 and Windows 98 guests
* Implement ACPI memory hotplug
* Improve ballooning to try to use 2MB pages when possible ( in progress - kern.devel@gmail.com )

==== Networking TODO: ====
* Has its [[NetworkingTodo|own page]]

==== PCI TODO: ====
* Has its [[PCITodo|own page]]

==== MMU related: ====
* Improve mmu page eviction algorithm (currently FIFO, change to approximate LRU).
* Add a read-only memory type.
** possible using mprotect()?
* Implement AM20 for dos and the like.
* O(1) write protection by protecting the PML4Es, then on demand PDPTEs, PDEs, and PTEs
* Simpler variant: don't drop large ptes when write protecting; just write protect them. When taking a write fault, either drop the large pte, or convert it to small ptes and write protect those (like O(1) write protection).
* O(1) mmu invalidation using a generation number

==== x86 emulator updates: ====
* On-demand register access, really, copying all registers all the time is gross.
** Can be done by adding 'available' and 'dirty' bitmasks
* Implement mmx and sse memory move instructions; useful for guests that use multimedia extensions for accessing vga (partially done)
* Implement an operation queue for the emulator. The emulator often calls userspace to perform a read or a write, but due to inversion of control it actually restarts instead of continuing. The queue would allow it to replay all previous operations until it reaches the point it last stopped.
** if this is done, we can retire ->read_std() in favour of ->read_emulated().
* convert more instructions to direct dispatch (function pointer in decode table)
* move init_emulate_ctxt() into x86_decode_insn() and other emulator entry points

==== Interactivity improvements: ====
* If for several frames in a row a large proportion of the framebuffer pages are changing, then for the next few frames don't bother to get the dirty page log from kvm, but instead assume that all pages are dirty. This will reduce page fault overhead on highly interactive workloads.
* When detecting keyboard/video/mouse activity, scale up the frame rate; when activity dies down, scale it back down (applicable to qemu as well).

==== Pass-through/VT-d related: ====
* Enhance KVM QEMU to return error messages if user attempts to pass-through unsupported devices:
** Devices with shared host IOAPIC interrupt
** Conventional PCI devices
** Devices without FLR capability
* QEMU PCI pass-through patch needs to be enhanced to same functionality as corresponding file in Xen
** Remove direct HW access by QEMU for probing PCI BAR size
** PCI handling of various PCI configuration registers
** Other enhancements that was done in Xen
* Host shared interrupt support
* VT-d2 support (WIP in Linux Kernel)
** Queued invalidation
** Interrupt remapping
** ATS
* USB 2.0 (EHCI) support

==== Bug fixes: ====
* Less sexy but ever important, fixing bugs is one of the most important contributions

==== Random improvements ====
* Utilize the SVM interrupt queue to avoid extra exits when guest interrupts are disabled

==== For the adventurous: ====
* Emulate the VMX instruction sets on qemu. This would be very beneficial to debugging kvm ( working on this - kern.devel@gmail.com ).
* Add [http://lagarcavilla.org/vmgl/ vmgl] support to qemu. Port to virtio. Write a Windows driver.
* Keep this TODO up to date

==== Nested VMX ====
* Implement performance features such as EPT and VPID

== KVM Safe Mode ==

An ioctl() from userspace that tells KVM to disable one or more of the following features:

* shadow paging (force direct mapping)
* instruction emulation (require virtio or mmio hypercall)
* task switches
* mode switches (long mode / legacy mode / real mode)
* IDT/GDT/LDT changes
* IDT/GDT/LDT write protect
* write protect important MSRs (*STAR etc)

The idea is both to protect the guest from attacks, and to protect the host from the guest.

__NOTOC__

FAQ

2013-10-16T03:58:33Z

ConradPino: force TOC location

=FAQ=
__TOC__
== Preparing to use KVM ==
=== What do I need to use KVM? ===
You will need an x86 machine running a recent Linux kernel on an Intel processor with VT (virtualization technology) extensions, or an AMD processor with SVM extensions (also called AMD-V). Xen has a [http://wiki.xensource.com/xenwiki/HVM_Compatible_Processors complete list] of compatible processors. For Intel processors, see also [http://ark.intel.com/VTList.aspx the Intel® Virtualization Technology List].

----------
=== Are 64bits processors supported under KVM? ===
Yes they are supported and will allow you to run 32bits and 64 bits guests.

See also '''Can KVM run a 32-bit guest on a 64-bit host? What about PAE?''' below.

----------

=== What is Intel VT / AMD-V / hvm? ===
[http://www.intel.com/technology/itj/2006/v10i3/1-hardware/6-vt-x-vt-i-solutions.htm Intel VT] and [http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_8826_14287,00.html AMD's AMD-V] are instruction set extensions that provide hardware assistance to virtual machine monitors. They enable running fully isolated virtual machines at native hardware speeds, for some workloads.

HVM (for Hardware Virtual Machine) is a vendor-neutral term often used to designate the x86 instruction set extensions.

----------

=== Where do I get my kvm kernel modules from? ===

See the [[Getting the kvm kernel modules]] page.

----------

=== How can I tell if I have Intel VT or AMD-V? ===
With a recent enough Linux kernel, run the command:

. egrep '^flags.*(vmx|svm)' /proc/cpuinfo

If something shows up, you have VT. You can also check the processor model name (in `/proc/cpuinfo`) in the vendor's web site.

Note:

* You'll never see (vmx|svm) in /proc/cpuinfo if you're currently running in in a dom0 or domU.<br /> The Xen hypervisor suppresses these flags in order to prevent hijacking.
* Some manufacturers disable VT in the machine's BIOS, in such a way that it cannot be re-enabled.
* `/proc/cpuinfo` only shows virtualization capabilities starting with Linux 2.6.15 (Intel) and Linux 2.6.16 (AMD). Use the `uname -r` command to query your kernel version.
In case of doubt, contact your hardware vendor.

----------

=== "KVM: disabled by BIOS" error ===
Check if there is an option to enable it in the BIOS. If not, look for a more recent BIOS on the vendor's web site.

Note:

* On some hardware (e-g HP nx6320), you need to power-off/power-on the machine after enabling virtualization in the BIOS.
* Enabling some BIOS features may break VT support on some hardware (e-g Enabling Intel AMT on a Thinkpad T500 will prevent kvm-intel from loading with "disabled by bios")
* On some Dell hardware, you also need to disable "Trusted Execution", otherwise VT will not be enabled.

----------

=== How can I use AMD-V extension? ===
modprobe kvm-amd

----------

=== What user space tools does KVM use? ===
KVM uses a slightly modified [http://www.nongnu.org/qemu QEMU] program to instantiate the virtual machine. Once running, a virtual machine is just a regular process. You can use `top(1), kill(1), taskset(1)` and similar tools to manage virtual machines.

----------

=== What virtual disk formats can KVM use? ===
KVM inherits a wealth of disk formats support from QEMU; it supports raw images, the native QEMU format (qcow2), VMware format, and many more.

----------

=== How do I use KVM on a headless machine (without a local GUI?) ===
Install a management tool such as virt-manager on a remote machine.

----------

=== Are there management tools available to help me manage my virtual machines? ===
Yes. Please see the [[Management Tools]] page for some links.

----------

== Using KVM ==
=== How can I use KVM with a non-privileged user? ===
The cleanest way is probably to create a group, say ''kvm'', and add the user(s) to that group. Then you will need change /dev/kvm to owned by group ''kvm''.

On a system that runs udev, you will probably need to add the following line somewhere in your udev configuration so it will automatically give the right group to the newly created device (i-e for ubuntu add a line to ''/etc/udev/rules.d/40-permissions.rules'').

KERNEL=="kvm", GROUP="kvm"

----------

=== How can I get the most performance out of KVM? ===

See the [[Tuning KVM]] page.

----------

=== Is KVM stable? ===
KVM is stable and used in production. As with most open source projects, development snapshots are less stable than the stable release series.

If your name is Andreas Mohr, you're reporting bugs in the wrong place.
----------

=== That's alright, but can I really use it for my daily use? ===
Sure. We continuously run the most often-used OSes and configurations and if anything breaks for the developers, it's fixed as soon as it was broken. See the [[Guest Support Status]] and [[Host Support Status]] pages to find out more. Please update them with success stories so that new users would benefit from the experience of the community.

----------

=== How about production use? ===
For production use, it's recommended you use the KVM modules shipped by the distribution you're using to ensure stability. As mentioned above, it's tempting to use new features, but you never know of (unwanted) surprises hidden away. It'll be best if you can run the development snapshots with non-critical production load, so that the latest releases are stable for you when you decide to deploy them.

----------

=== What happens if I kill -9 a VM process? ===
From the guest's perspective, it is as if you yanked the power cord out. From the host's perspective, the process is killed and all resources it uses are reclaimed.

----------

=== I need help to setup the network for my guest ===
You can have a look to the [[Networking]] page of this wiki for informations on the most classical networking setup for the guests. You can also refer to the QEMU documentation.

----------

=== Where can I find more documention... ===
Most usability issues are covered in the QEMU [http://www.nongnu.org/qemu/user-doc.html documentation]. There is also an extensive [http://qemu-buch.de/cgi-bin/moin.cgi/FrequentlyAskedQuestions FAQ] (old vanished link: [http://kidsquid.com/cgi-bin/moin.cgi/FrequentlyAskedQuestions FAQ]).

----------

== Troubleshooting ==
=== How can I check that I'm not falling back to QEMU with no hardware acceleration? ===

If you think that you might no be using the hardware acceleration provided by the KVM module, here are a few steps to help you check this.

First of all, check that you don't have messages such as:

qemu-system-x86_64 -hda myvm.qcow2
open /dev/kvm: No such file or directory
Could not initialize KVM, will disable KVM support

In that case, you can check that:
* the modules are correctly loaded <code>lsmod|grep kvm
* you don't have a "KVM: disabled by BIOS" line in the output of dmesg
* /dev/kvm exists and you have the correct rights to use it

Other ways to do the diagnostic:
* if you have access to the QEMU monitor (Ctrl-Alt-2, use Ctrl-Alt-1 to get back to the VM display), enter the "info kvm" command and it should respond with "KVM support: enabled"
* the right-end columns of the output from <code>lsmod|grep kvm</code> on the host system, once the VM is started should show only non zero values. The value on the line corresponding to the architecture specific module (e-g kvm_intel, kvm_amd) show the number of VM using the module. For instance, if I have 2 VM running using the KVM module on a machine with vt, it will report:

lsmod|grep kvm
kvm_intel 44896 2
kvm 159656 1 kvm_intel

=== "rect too big" Message when using VNC Display ===
When connection to a VNC Terminal, a "rect too big" message appears, and the VNC Session disconnects.

This happens because of a VNC protocol flaw on the way on-the-fly pixel format changes are handled (more info at [http://www.mail-archive.com/qemu-devel@nongnu.org/msg04879.html this thread]). If you are using TigerVNC, you can avoid this problem by disabling on-the-fly selection of pixel encoding, using the <tt>-AutoSelect=0</tt> command-line option of vncviewer. You may also want to check the encoding options on the vncviewer man page, as this will disable automatic selection of encoding based on connection speed.

----------
'''How do I set up the network such that my guest is accessible from other machines?''' or

=== My guest network is stuck what should I do? ===
KVM uses QEMU for its device emulation. Consult the [http://qemu-project.org/Documentation/Networking QEMU network wiki page] for detailed network setup instructions.

One would probably be interested in the Root Networking Mode page and the Network Bridge page.

Guest-side network lockups (fortunately restartable) may be happening due to tun/tap bridging erroneous MAC address reconfiguration on host side, see RHEL bug #571991 and others.

----------

=== I'm experiencing timer drift issues in my VM guests, what to do? ===

Especially in case of networked systems (e.g. via NFS or Samba) it is very important to ensure stable operation of timing (both system timer and RTC).
Tell-tale signs of related trouble in VMs (apparently qemu/KVM/VMWare etc. are all affected) are e.g.
"make[2]: Warning: File `XXXXX/cmakelists_rebuilder.stamp' has modification time 0.37 s in the future"
"Clock skew detected. Your build may be incomplete."

[http://maemovmware.garage.maemo.org/2nd_edition/requirements_documentation.html Maemo docs] state that it's important to disable UTC and set the correct time zone, however I don't really see how that would help in case of diverging host/guest clocks.
IMHO much more useful and important is to configure properly working NTP server (chrony recommended, or ntpd) on both host and guest.
The single most decisive trick IMHO is to specify the '''host''' NTP server as the main entry within guest VM instead of "foreign" NTP servers, to make sure to achieve the most precise coupling between these two related systems (timing drift vs. other systems does not matter nearly as much as a tight time precision for inner host/guest system interaction e.g. in the case of NFS/Samba shares etc.).
For verification, see chronyc "sources -v", "tracking" ("System time" row) commands.

After having applied this very tight NTP coupling, this seems to finally have gotten rid of make's time drift warnings.

Perhaps qemu's -tdf (timing drift fix) option magically manages to help in your case, too.

See also [https://espace.cern.ch/it-faqs/Lists/faqs/DispForm.aspx?ID=368 Faqs: I received a message about "clock skew"].

----------
=== I get "rtc interrupts lost" messages, and the guest is very slow? ===
Try setting <code>CONFIG_HPET_EMULATE_RTC=y</code> in your host <code>.config</code>.

----------

=== I get an "Exception 13" or "Exception 12" message while booting a guest OS on my Intel host ===
See the [[Intel Real Mode Emulation Problems]] page.

----------

=== I have VMware/Parallels/VirtualBox installed and when I modprobe KVM, my system deadlocks. ===
Neither Intel VT nor AMD-V provide a mechanism to determine whether software is currently using the hardware virtualization extensions. This means that if you have two kernel modules loaded attempting to use hardware virtualization extensions, very bad things will happen. If you are using another type of virtualization software and experience any sort of weirdness with KVM, make sure you can reproduce the problem without the kernel modules for that software loaded before you report a bug in KVM.

----------

=== There's nothing on QEMU/KVM screen, but it's not hanged! I'm trying to install Kubuntu. ===
Try to run kvm with -std-vga option. It helps if guest operating system uses framebuffer mode like Kubuntu/Ubuntu.

----------

=== When I click the guest operating system window, mouse is grabbed. How can I get mouse to not to do that? OR Mouse doesn't show up / doesn't work in the guest. What do I do? ===
From #qemu wiki, try to run kvm/qemu with

-usb -usbdevice tablet
If that doesn't work, try this:

$ export SDL_VIDEO_X11_DGAMOUSE=0
(from http://wiki.clug.org.za/wiki/QEMU_mouse_not_working )

----------

== General KVM information ==
=== What is the difference between KVM and Xen? ===
Xen is an external hypervisor; it assumes control of the machine and divides resources among guests. On the other hand, KVM is part of Linux and uses the regular Linux scheduler and memory management. This means that KVM is much smaller and simpler to use; it is also more featureful; for example KVM can swap guests to disk in order to free RAM.

KVM only run on processors that supports x86 hvm (vt/svm instructions set) whereas Xen also allows running modified operating systems on non-hvm x86 processors using a technique called paravirtualization. KVM does not support paravirtualization for CPU but may support paravirtualization for device drivers to improve I/O performance.

----------

=== What is the difference between KVM and VMware? ===
VMware is a proprietary product. KVM is Free Software released under the GPL.

----------

=== What is the difference between KVM and QEMU? ===
QEMU uses emulation; KVM uses processor extensions (HVM) for virtualization.

----------

=== Do you have a port of KVM for Windows? ===
Not officially.

Kazushi Takahashi has been working on an experimental version though, called WinKVM, available [http://github.com/ddk50/winkvm here].

----------

=== What kernel version does it work with? ===
It depends on what version of KVM you are using. The last release of KVM should work with any recent kernel (2.6.17 and above), older releases even older kernels.

----------

=== How much RAM do I need? ===
You will need enough memory to let the guest run comfortably while keeping enough for the host. 1GB is probably a minimum configuration for the host OS.

----------

=== Is dynamic memory management for guests supported? ===
This is a broad topic covering a few areas.

A. KVM only allocates memory as the guest tries to use it. Once it's allocated, KVM keeps it. Some guests (namely Microsoft guests) zero all memory at boot time. So they will use all memory.

B. Certain guests (only Linux at the moment) have a balloon driver, so the host can have the guest allocate a certain amount of memory which the guest won't be able to use anymore and it can then be freed on the host. Ballooning is controlled in the host via the [http://www.nongnu.org/qemu/qemu-doc.html#SEC12 balloon monitor command].

C. Some hosts (presently only RHEL5.4 / CentOS 5.4) have a feature called KSM (Kernel Sharedpage Merging), which collapses together identical pages; this requires kernel support on the host, as well as a kvm new enough to opt in to the behavior. As some guest platforms (most notably Windows) zero out free'd memory, such pages are trivially collapsed. The ksmctl command needs to be used to enable KSM; alternately, the ksmtuned service found in Fedora 12 can be run to dynamically adjust KSM's aggressiveness based on the amount of free memory available

----------

=== What OSs can I run inside KVM VM? ===
Several. See the [[Guest Support Status]] page for details. Note that several Linux flavors are known to hang on Intel processors during startup. Workaround is to disable splash screens in grub.

----------

=== Does KVM support a live migration feature to move virtual machines from one host to another without downtime? ===
Yes. See the [[Migration]] page for details.

----------

=== Does KVM support live migration from an AMD host to an Intel host and back? ===
Yes. There may be issues on 32-bit Intel hosts which don't support NX (or XD), but for 64-bit hosts back and forth migration should work well. Migration of 32-bit guests should work between 32-bit hosts and 64-bit hosts.
If one of your hosts does not support NX, you may consider disabling NX when starting the guest on a NX-capable system. You can do it by passing "-cpu qemu64,-nx" parameter to the guest.

----------

=== Can KVM run a 32-bit guest on a 64-bit host? What about PAE? ===
KVM supports 32-bit guests on 64-bit hosts, and any combination of PAE and non-PAE guests and hosts. The only unsupported combination is a 64-bit guest on a 32-bit host.

If you are running a Windows Virtual Machine and have problems enabling PAE in your guest see the [[Windows PAE Workaround]] page.

----------

=== Is it possible to use USB devices with a guest OS? ===
Yes, look up how to do it with QEMU, it's the same way.

----------

=== Can I have higher or widescreen resolutions (eg 1680 x 1050) in KVM? ===
Use the -vga std parameter while starting the VM to allow high resolution and widescreen displays.

If the resolution you want to use is not available, you can patch the corresponding source files (see http://article.gmane.org/gmane.comp.emulators.kvm.devel/13557 as a reference), or send a mail to the KVM mailing list if you are not able to patch the source yourself.

When using Windows as guest OS and not having issues with people violating the GPL you might want to use the driver from the VBEMP x86 project (http://www.bearwindows.boot-land.net/vbemp.htm) which is based on ReactOS code.

----------

=== Does KVM support SMP hosts? ===
Yes.

----------

=== Does KVM support SMP guests? ===
Yes. Up to 16 CPUs can be specified using the -smp option.

----------

=== Is the name 'KVM' trademarked? ===
No.

----------

User talk:ConradPino

2013-10-16T03:53:08Z

ConradPino:

==Talk:Conrad T. Pino==

User:ConradPino

2013-10-16T03:52:10Z

ConradPino: point to home

==Conrad T. Pino==
http://www.pino.com/

NetworkingTodo

2013-10-16T03:49:41Z

ConradPino: add missing heading close markup

This page should cover all networking related activity in KVM,
currently most info is related to virtio-net.

TODO: add bugzilla entry links.

=== projects in progress. contributions are still very wellcome! ===

* vhost-net scalability tuning: threading for many VMs
Plan: switch to workqueue shared by many VMs
http://www.mail-archive.com/kvm@vger.kernel.org/msg69868.html

http://domino.research.ibm.com/library/cyberdig.nsf/1e4115aea78b6e7c85256b360066f0d4/479e3578ed05bfac85257b4200427735!OpenDocument

Developer: Bandan Das
Testing: netperf guest to guest

* multiqueue support in macvtap
multiqueue is only supported for tun.
Add support for macvtap.
Developer: Jason Wang

* support more queues
We limit TUN to 8 queues, but we really want
1 queue per guest CPU. The limit comes from net
core, need to teach it to allocate array of
pointers and not array of queues.
Jason has an draft patch to use flex array.
Another thing is to move the flow caches out of tun_struct.
Developer: Jason Wang

* enable multiqueue by default
Multiqueue causes regression in some workloads, thus
it is off by default. Detect and enable/disable
automatically so we can make it on by default.
This is because GSO tends to batch less when mq is enabled.
https://patchwork.kernel.org/patch/2235191/
Developer: Jason Wang

* rework on flow caches
Current hlist implementation of flow caches has several limitations:
1) at worst case, linear search will be bad
2) not scale
https://patchwork.kernel.org/patch/2025121/
Developer: Jason Wang

* eliminate the extra copy in virtio-net driver
We need do an extra copy of 128 bytes for every packets.
This could be eliminated for small packets by:
1) use build_skb() and head frag
2) bigger vnet header length ( >= NET_SKB_PAD + NET_IP_ALIGN )
Or use a dedicated queue for small packet receiving ? (reordering)
Developer: Jason Wang

* make pktgen works for virtio-net ( or partially orphan )
virtio-net orphan the skb during tx,
which will makes pktgen wait for ever to the refcnt.
Jason's idea: introduce a flat to tell pktgen not for wait
Discussion here: https://patchwork.kernel.org/patch/1800711/
MST's idea: add a .ndo_tx_polling not only for pktgen
Developer: Jason Wang

* Add HW_VLAN_TX support for tap
Eliminate the extra data moving for tagged packets
Developer: Jason Wang

* Announce self by guest driver
Send gARP by guest driver. Guest part is finished.
Qemu is ongoing.
V7 patches is here:
http://lists.nongnu.org/archive/html/qemu-devel/2013-03/msg01127.html
Developer: Jason Wang

* guest programmable mac/vlan filtering with macvtap
Developer: Amos Kong
qemu: https://bugzilla.redhat.com/show_bug.cgi?id=848203
libvirt: https://bugzilla.redhat.com/show_bug.cgi?id=848199
https://git.kernel.org/cgit/virt/kvm/mst/qemu.git/patch/?id=1c0fa6b709d02fe4f98d4ce7b55a6cc3c925791c
Status: [[GuestProgrammableMacVlanFiltering]]

* bridge without promisc mode in NIC
given hardware support, teach bridge
to program mac/vlan filtering in NIC
Helps performance and security on noisy LANs
http://comments.gmane.org/gmane.linux.network/266546
Developer: Vlad Yasevich

* reduce networking latency:
allow handling short packets from softirq or VCPU context
Plan:
We are going through the scheduler 3 times
(could be up to 5 if softirqd is involved)
Consider RX: host irq -> io thread -> VCPU thread ->
guest irq -> guest thread.
This adds a lot of latency.
We can cut it by some 1.5x if we do a bit of work
either in the VCPU or softirq context.
Testing: netperf TCP RR - should be improved drastically
netperf TCP STREAM guest to host - no regression
Developer: MST

* Flexible buffers: put virtio header inline with packet data
https://patchwork.kernel.org/patch/1540471/
Developer: MST

* device failover to allow migration with assigned devices
https://fedoraproject.org/wiki/Features/Virt_Device_Failover
Developer: Gal Hammer, Cole Robinson, Laine Stump, MST

* Reuse vringh code for better maintainability
Developer: Rusty Russell

* Improve stats, make them more helpful for per analysis
Developer: Sriram Narasimhan

* Bug: e1000 & rtl8139: Change macaddr in guest, but not update to qemu (info network)
Developer: Amos Kong
https://bugzilla.redhat.com/show_bug.cgi?id=922589

* Enable GRO for packets coming to bridge from a tap interface
Developer: Dmitry Fleytman

* Better support for windows LRO
Extend virtio-header with statistics for GRO packets:
number of packets coalesced and number of duplicate ACKs coalesced
Developer: Dmitry Fleytman

* IPoIB infiniband bridging
Plan: implement macvtap for ipoib and virtio-ipoib
Developer: MST

* netdev polling for virtio.
There are two kinds of netdev polling:
- netpoll - used for debugging
- proposed low latency net polling
See http://lkml.indiana.edu/hypermail/linux/kernel/1303.0/00553.html
Developer: Jason Wang

* sharing config interrupts
Support mode devices by sharing a single msi vector
between multiple virtio devices.
(Applies to virtio-blk too).
Developer: Amos Kong

* use kvm eventfd support for injecting level interrupts,
enable vhost by default for level interrupts
Developer: Amos Kong

=== projects that are not started yet - no owner ===

* receive side zero copy
The ideal is a NIC with accelerated RFS support,
So we can feed the virtio rx buffers into the correct NIC queue.
Depends on non promisc NIC support in bridge.
Search for "Xin Xiaohui: Provide a zero-copy method on KVM virtio-net"
for a very old prototype

* RDMA bridging

* DMA emgine (IOAT) use in tun
Old patch here: [PATCH RFC] tun: dma engine support
It does not speed things up. Need to see why and
what can be done.

* virtio API extension: improve small packet/large buffer performance:
support "reposting" buffers for mergeable buffers,
support pool for indirect buffers

* more GSO type support:
Kernel not support more type of GSO: FCOE, GRE, UDP_TUNNEL

* ring aliasing:
using vhost-net as a networking backend with virtio-net in QEMU
being what's guest facing.
This gives you the best of both worlds: QEMU acts as a first
line of defense against a malicious guest while still getting the
performance advantages of vhost-net (zero-copy).
In fact a bit of complexity in vhost was put there in the vague hope to
support something like this: virtio rings are not translated through
regular memory tables, instead, vhost gets a pointer to ring address.
This allows qemu acting as a man in the middle,
verifying the descriptors but not touching the packet data.

* non-virtio device support with vhost
Use vhost interface for guests that don't use virtio-net

=== vague ideas: path to implementation not clear ===

* ring redesign:
find a way to test raw ring performance
fix cacheline bounces
reduce interrupts

* irq/numa affinity:
networking goes much faster with irq pinning:
both with and without numa.
what can be done to make the non-pinned setup go faster?

* reduce conflict with VCPU thread
if VCPU and networking run on same CPU,
they conflict resulting in bad performance.
Fix that, push vhost thread out to another CPU
more aggressively.

* rx mac filtering in tun
the need for this is still not understood as we have filtering in bridge
we have a small table of addresses, need to make it larger
if we only need filtering for unicast (multicast is handled by IMP filtering)

* vlan filtering in tun
the need for this is still not understood as we have filtering in bridge

* vlan filtering in bridge
kernel part is done (Vlad Yasevich)
teach qemu to notify libvirt to enable the filter (still to do) (existed NIC_RX_FILTER_CHANGED event contains vlan-tables)

* tx coalescing
Delay several packets before kick the device.

* interrupt coalescing
Reduce the number of interrupt

* bridging on top of macvlan
add code to forward LRO status from macvlan (not macvtap)
back to the lowerdev, so that setting up forwarding
from macvlan disables LRO on the lowerdev

* virtio: preserve packets exactly with LRO
LRO is not normally compatible with forwarding.
virtio we are getting packets from a linux host,
so we could thinkably preserve packets exactly
even with LRO. I am guessing other hardware could be
doing this as well.

* vxlan
What could we do here?

* bridging without promisc mode with OVS

=== testing projects ===
Keeping networking stable is highest priority.

* Write some unit tests for vhost-net/vhost-scsi
* Run weekly test on upstream HEAD covering test matrix with autotest
* Measure the effect of each of the above-mentioned optimizations
- Use autotest network performance regression testing (that runs netperf)
- Also test any wild idea that works. Some may be useful.
* Migrate some of the performance regression autotest functionality into Netperf
- Get the CPU-utilization of the Host and the other-party, and add them to the report. This is also true for other Host measures, such as vmexits, interrupts, ...
- Run Netperf in demo-mode, and measure only the time when all the sessions are active (could be many seconds after the beginning of the tests)
- Packaging of Netperf in Fedora / RHEL (exists in Fedora). Licensing could be an issue.
- Make the scripts more visible

=== non-virtio-net devices ===
* e1000: stabilize

=== test matrix ===

DOA test matrix (all combinations should work):
vhost: test both on and off, obviously
test: hotplug/unplug, vlan/mac filtering, netperf,
file copy both ways: scp, NFS, NTFS
guests: linux: release and debug kernels, windows
conditions: plain run, run while under migration,
vhost on/off migration
networking setup: simple, qos with cgroups
host configuration: host-guest, external-guest