Tmem, when called from a tmem-capable (paravirtualized) guest, makes
use of otherwise unutilized ("fallow") memory to create and manage
pools of pages that can be accessed from the guest either as
"ephemeral" pages or as "persistent" pages. In either case, the pages
are not directly addressible by the guest, only copied to and fro via
the tmem interface. Ephemeral pages are a nice place for a guest to
put recently evicted clean pages that it might need again; these pages
can be reclaimed synchronously by Xen for other guests or other uses.
Persistent pages are a nice place for a guest to put "swap" pages to
avoid sending them to disk. These pages retain data as long as the
guest lives, but count against the guest memory allocation.
Tmem pages may optionally be compressed and, in certain cases, can be
shared between guests. Tmem also handles concurrency nicely and
provides limited QoS settings to combat malicious DoS attempts.
Save/restore and live migration support is not yet provided.
Tmem is primarily targeted for an x86 64-bit hypervisor. On a 32-bit
x86 hypervisor, it has limited functionality and testing due to
limitations of the xen heap. Nearly all of tmem is
architecture-independent; three routines remain to be ported to ia64
and it should work on that architecture too. It is also structured to
be portable to non-Xen environments.
Tmem defaults off (for now) and must be enabled with a "tmem" xen boot
option (and does nothing unless a tmem-capable guest is running). The
"tmem_compress" boot option enables compression which takes about 10x
more CPU but approximately doubles the number of pages that can be
stored.
Tmem can be controlled via several "xm" commands and many interesting
tmem statistics can be obtained. A README and internal specification
will follow, but lots of useful prose about tmem, as well as Linux
patches, can be found at http://oss.oracle.com/projects/tmem .
Signed-off-by: Dan Magenheimer <dan.magenheimer@oracle.com>
Current scheduler only care performance, thus always picks pCPU from
the most idle package. This knob provides another option to pick pCPU from
least idle package, for user who want performance power balance.
Signed-off-by: Yu Ke <ke.yu@intel.com>
Signed-off-by: Tian Kevin <kevin.tian@intel.com>
Calculate cpu actual average freq according to aperf and mperf MSR and
then deliver to libxc interface
Signed-off-by: Liu, Jinsong <jinsong.liu@intel.com>
CPU topology info is necessary for power management analysis. For
example, to analysis the effect of Px state coordination, Cx
package/core coordination, the thread/core/socket topology
information is needed.
This patch add new command "get-cpu-topology" in xenpm to print the
CPU topology info:
Signed-off-by: Yu Ke <ke.yu@intel.com>
The code was attempting to use the domain's current number of pages
(info.nr_pages) as a maximum index. We then walk the memory map and
can easily over-write past the end of the nr_pages-sized array, if the
domain has more pages mapped in than earlier (live dump). Restrict
ourselves to the current number of pages.
Also fix the dump core method in xend to actually implement the crash
and live options. In particular this means that xend clients other
than xm now get non-live dumps by default.
Signed-off-by: John Levon <john.levon@sun.com>
Add a new parameter to DOMCTL_bind_pt_irq to allow Xen to know the
guest physical address of MSI-X table. Also add a new MMIO intercept
handler to intercept that gpa in order to handle MSI-X vector mask
bit operation in the hypervisor. This reduces the load of device model
considerably if the guest does mask and unmask frequently
Signed-off-by: Qing He <qing.he@intel.com>
This includes making the pagetable walker in xc_pagetab.c behave
correctly for 32-bit and 64-bit HVM guests.
Signed-off-by: Tim Deegan <Tim.Deegan@citrix.com>
Never used by a guest OS (except in IA64 hcall translation layer) and
obsoleted in the tools for ages. Recent usage by qemu-dm is now
removed.
Signed-off-by: Keir Fraser <keir.fraser@citrix.com>
(32-bit) HVM guests. 64-bit HVM guests are still not supported for
now, pending a sensible way of getting at the guest's EFER.LMA.
Signed-off-by: Tim Deegan <Tim.Deegan@citrix.com>
Implement libxc interface to PoD functionality:
* Add xc_hvm_build_target_mem(), which takes both memsize and target.
Memsize is the total memory, allocated in PoD pages and reported in
the e820; target is the size of the cache. If these are the same, the
normal funcitonality is called. (So you can use the same function to
build always, and it will decide whether to use PoD or not.)
* Add xc_domain_memory_[gs]et_pod_target(), which sets and/or returns
information about the PoD cache and p2m entries.
Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>
Some distro kernels do not handle spurious page faults so allow these
to be supressed on a per VM basis.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
When creating an HVM domain, if e.g. another domain is created before
qemu allocates video memory, the extra 8MB memory ballooning is not
available any more, because it got consumed by the other domain.
This fixes it by taking video memory from the main memory:
- make hvmloader use e820_malloc to reserve some of the main memory
and notify ioemu of its address through the Xen platform PCI card.
- add XENMAPSPACE_mfn to the xen_add_to_physmap memory op, to allow
ioemu to move the MFNs between the original position and the PCI
mapping, when LFB acceleration is disabled/enabled
- add a remove_from_physmap memory op, to allow ioemu to unmap it
completely for the case of old guests with acceleration disabled.
- add xc_domain_memory_translate_gpfn_list to libxc to allow ioemu to
get the MFNs of the video memory.
- have xend save the PCI memory space instead of ioemu: if a memory
page is there, the guest can access it like usual memory, so xend
can safely be responsible to save it. The extra benefit is that
live migration will apply the logdirty optimization there too.
- handle old saved images, populating the video memory from ioemu if
really needed.
Signed-off-by: Samuel Thibault <samuel.thibault@eu.citrix.com>
xc_gnttab_map_domain_grant_refs permits to simply map several pages
from the same domain.
Signed-off-by: Samuel Thibault <samuel.thibault@eu.citrix.com>
This allows domains which make buggy assumptions about the maximum
possible MFN to be worked around.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
The address_bits field will be limited to 8 bits and is now embedded
in the mem_flags member, which additionally contains the node number
(limited to 8 bit).
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Keir Fraser <keir.fraser@citrix.com>
This event channel will be notified when the domain transitions to the
suspended state, which can be much faster than raising VIRQ_DOM_EXC
and waiting for the notification to be propagated via xenstore.
No attempt is made here to prevent multiple subscribers (last one
wins), or to detect that the subscriber has gone away. Userspace tools
should take care.
Signed-off-by: Brendan Cully <brendan@cs.ubc.ca>
hypervisor about memory changes.
The impact on disk performance is typically making it from 71.5MBps
down to 70.5Mbps during the live migration. The impact on network
performance is actually even hard to measure.
Signed-off-by: Samuel Thibault <samuel.thibault@eu.citrix.com>
Domctl interface is extended to allow libxc retrieve device group
information from hypervisor. Vendor-specific iommu_ops is also
extended by adding a new operation "get_device_group_id()", which is
currently a null pointer but could be implemented later for vt-d.
Error will be raised from tools side when user trying to assign PCI
device with a sibling device being driven by dom0. User will keep
being prompted until he has hidden the entire device group (at least,
the sibling devices driven by dom0) in dom0 kernel
parameter. Hopefully this framework could be flexible enough to
support both amd iommu and vt-d.
The following 2 cases are not covered by this patch, but should be
easy to handle.
* Checking for hot-plug devices (maybe we can delay calling
ImageHandler.signalDeviceModel() until all checks are done?)
* Checking for splitted device group between different passthru
domains
Signed-off-by: Wei Wang <wei.wang2@amd.com>
Implement statistic interface for cx via sysctl & libxc. Provide a
easy way to collect processor cx info within dom0.
Signed-off-by: Wei Gang <gang.wei@intel.com>
Sampling and collecting dynamic Px statistic data, including
Px state value, Px count, Px residency time, Px transition
matrix, etc. Provide Px statistic data to user through libxc.
Signed-off-by: Liu Jinsong <jinsong.liu@intel.com>
This adds a new HVM op that enables tracking dirty bits of a range of
video RAM. The idea is to optimize just for the most common case
(only one guest mapping, with sometimes some temporary other
mappings), which permits to keep the overhead on shadow as low as
possible.
Signed-off-by: Samuel Thibault <samuel.thibault@eu.citrix.com>
Currently it only inercept access to MSI config space, no MSI-x support.
Signed-off-by: Jiang Yunhong <yunhong.jiang@intel.com>
Signed-off-by: Shan Haitao <haitao.shan@intel.com>
CPUID's replies are now pre-calculated in libxc and given to the
hypervisor via a new domctl.
There are two parts to this feature:
- We can overwrite cpuid's replies.
- We can check that a vm is compatible with a host's processor.
Signed-off-by: Jean Guyader <jean.guyader@eu.citrix.com>
Signed-off-by: Keir Fraser <keir.fraser@citrix.com>
This patch adds the ability to set the number of slots that may be
used for mapping grant references, using the gntdev user-space grant
reference mapping driver.
Signed-off-by: Derek Murray <Derek.Murray@cl.cam.ac.uk>
** Currently only 2 virtual hotplug pci slots(6~7) are created so more
than 2 vtd dev can't be hotplugged, but we can easily extend it in
future.
Three new commands are added:
"xm pci-list domid" show the current assigned vtd device, like:
VSlt domain bus slot func
0x6 0x0 0x02 0x00 0x0
"xm pci-detach" hot remove the specified vtd device by the virtual
slot, like:
xm pci-detach EdwinHVMDomainVtd 6
"xm pci-attach DomainID dom bus dev func [vslot]" hot add a new vtd
device in the vslot. If no vslot specified, a free slot will be picked
up. e.g. to insert '0000:03:00.0':
xm pci-attach EdwinHVMDomainVtd 0 3 0 0
** guest pci hotplug
linux: pls. use 2.6.X and enable ACPI PCI hotplug ( Bus options=> PCI
hotplug => ACPI PCI hotplug driver )
windows: 2000/xp/2003/vista are all okay
Signed-off-by: Zhai Edwin <edwin.zhai@intel.com>
mb/rmb/wmb. Instead add a xen_ prefix. Modify Xen's public headers to
expect the prefixed names instead of bare mb/rmb/wmb, but gate this
expectation on a bump of __XEN_INTERFACE_VERSION__.
Signed-off-by: Keir Fraser <keir.fraser@citrix.com>
Stubdomains (and probably other domain disagregation elements too)
need to be able to tinker with another domain. This adds IS_PRIV_FOR
that extends IS_PRIV by allowing domains to have privileges over a
given "target" domain. XEN_DOMCTL_set_target permits to set this
"target". A new 'target' configuration option makes the domain builder
use it.
Signed-off-by: Samuel Thibault <samuel.thibault@eu.citrix.com>
This makes use of the domain config option guest_os_type for
ia64 and removes the backing for the previous ACPI based
mechanism used previously. A user wanting optimal performance
for a specific type of OS guest running in an HVM domain should
make use of this new option. See updated xmexmaple.vti for
available options. All supported OSes should always work using
the default option or leaving the option unspecified. Originally
based on patch from Zhang Xin.
Signed-off-by: Alex Williamson <alex.williamson@hp.com>
guests. Use this to pin virtual framebuffer VRAM as attribute WB, even
if guest tries to map with other attributes.
Signed-off-by: Disheng Su <disheng.su@intel.com>
Keir Fraser
Alex Williamson
Keir Fraser