In order to be able to use "xl info" before Xenstore has been started
or after it has crashed, hide obtaining the available Xenstore features
behind the new option "-x".
Fixes: ecce7970cf ("tools/xl: add available Xenstore features to xl info output")
Reported-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
Acked-by: Anthony PERARD <anthony.perard@vates.tech>
Avoid dereferencing NULL in introduce_domain() when called during
live update.
Coverity ID: 1665111
Fixes: f78895ad78 ("tools/xenstored: Read event channel from xenstored page")
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
Update get_implicit_path to return the correct value for a non-dom0
xenstored domain.
Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
posix.c and minios.c implement the same named functions serving slightly
different purposes.
For xenbus_map()
posix.c maps the local /dev/xen/xsd_kva
minios.c maps dom0 via grant and there is open coding for stub_domid in
map_interface.
Change xenbus_map() to map the local domain's interface. The default
grant table mapping is performed otherwise.
Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
The dom0_domid variable is misnamed and conflates purposes. If we have
xenstored running in a Linux domain that is not dom0, this variable
controls the lookup of /proc/xen/xsd_kva and the event channel.
Rename to store_domid to better show its purpose.
One implication of this change is that the xenstore domain is not
privileged by virtue of considering store_domid as privileged.
domain_is_unprivileged() removes the dom0_domid/store_domid check, so
xenstore domain is no longer considered privileged.
onearg_domain() is updated to return EINVAL for store_domid or priv_domid
to maintain the ability to call XS_RESUME.
xenbus_master_domid() is removed with store_domid being used instead.
Add a description of the -m/--master-domid options while
doing this.
Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Usually, priv_domid == dom0_domid == 0, and that is what is expected.
If we rename s/dom0_domid/store_domid/, it seems more likely we want to
actually have the priv_domid as the owner.
That leads to follow on changes to ensure that the priv_domid is created
first.
Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
With separate control and xenstore domains, the xenstore domain itself
is no longer privileged. Local socket connections do not have
conn->domain assigned, but domain_chk_quota() will go ahead and
dereference it. The !domid_is_unprivileged(conn->id) check usually
means the function would have already returned.
Bail out on !conn->domain to avoid faulting. This adds a similar check
to that in domain_max_chk()
Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Replace dom0_init() with init_domains() which uses libxenmanage to
iterate through all existing domains, storing them in a list. The xenstore
domain is introduced first, and then all the other domains are
introduced. The xenstore domain needs to be introduced first to setup
structures needed for firing watches.
dom0_domid is updated with the xenstore domain, since it really
indicates the local domain.
priv_domid is set to the control domain. This makes it limited to a
single domain.
These features let xenstore automatically connect any existing domains,
which means it doesn't need to be done manually from init-dom0less.
For a legacy dom0, the result should be unchanged.
For a late xenstore stubdom it should also be the same, but priv_domid
would be set automatically to control domain (which default to 0
normally).
Always signal the event channel for initial domains. This gets dom0 (a
local xenstored domain) to connect.
Also always write XENSTORE_CONNECTED since we know we are connected at
this point.
To support ARM dom0less domains with xen,enhanced = "no-xenstore" a
failed introduce_domain() becomes non-fatal. Normally,
HVM_PARAM_STORE_EVTCHN is used to identify .
priv_domid from the command line is used, or the first control domain is
used.
dom0_domid will set to the last xenstore found. This will handle dom0
or dom0less, where only 1 xenstore domain can exist, or stubdom, where
dom0 and dom1 exist, and we want to take the stubdom.
Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Add helpers to lookup the event channel for a domid. This hides some
of the differences between dom0 and stubdom xenstored. Each version
defines its own.
It highlights the different meanings between get_xenbus_evtchn() in a
stubdom, where it looks up dom0's event channel, and dom0, where it
looks up the local event channel.
get_domain_evtchn() replaces get_xenbus_evtchn(), and
get_xenbus_evtchn() is removed from minios.c as it is inlined in the new
function.
The default return 0 will be fine as any other auto-introduced domain
will needs the event channel populated in the grant.
Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Make introduce_domain() use an event channel from the the xenstore page.
It is only used if non-zero. Otherwise the passed in event channel port
is used.
The is useful for a xenstored stubdom to configure domains autonomously.
Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
With the appearance of Intel Sierra Forest and Granite Rapids it's now
possible to get a production x86 host with the following memory map:
SRAT: Node 0 PXM 0 [0000000000000000, 000000007fffffff]
SRAT: Node 0 PXM 0 [0000000100000000, 000000807fffffff]
SRAT: Node 1 PXM 1 [0000063e80000000, 000006be7fffffff]
SRAT: Node 2 PXM 2 [00000c7e80000000, 00000cfe7fffffff]
SRAT: Node 3 PXM 3 [000012be80000000, 0000133e7fffffff]
This is from a four socket Granite Rapids system, with each node having
512GB of memory. The total amount of RAM on the system is 2TB, but without
enabling CONFIG_BIGMEM the last range is not accessible, as it's above the
16TB boundary covered by the frame table. Sierra Forest and Granite Rapids
are socket compatible, however Sierra Forest only supports 2 socket
configurations, while Granite Rapids can go up to 8 sockets.
Note that while the memory map is very sparse, it couldn't be compressed
using the current PDX_MASK compression algorithm, which relies on all
ranges having a shared zeroed region of bits that can be removed.
The memory map presented above has the property of all regions being
similarly spaced between each other, and all having also a similar size.
Use a lookup table to store the offsets to translate from/to PFN and PDX
spaces. Such table is indexed based on the input PFN or PDX to translated.
The example PFN layout about would get compressed using the following:
PFN compression using PFN lookup table shift 29 and PDX region size 0x10000000
range 0 [0000000000000, 0x0000807ffff] PFN IDX 0 : 0000000000000
range 1 [0x00063e80000, 0x0006be7ffff] PFN IDX 3 : 0x00053e80000
range 2 [0x000c7e80000, 0x000cfe7ffff] PFN IDX 6 : 0x000a7e80000
range 3 [0x0012be80000, 0x00133e7ffff] PFN IDX 9 : 0x000fbe80000
Note how the tow ranges belonging to node 0 get merged into a single PDX
region by the compression algorithm.
The default size of lookup tables currently set in Kconfig is 64 entries,
and the example memory map consumes 10 entries. Such memory map is from a
4 socket Granite Rapids host, which in theory supports up to 8 sockets
according to Intel documentation. Assuming the layout of a 8 socket system
is similar to the 4 socket one, it would require 21 lookup table entries to
support it, way below the current default of 64 entries.
The valid range of lookup table size is currently restricted from 1 to 512
elements in Kconfig.
An extra array is used to keep track of the base PFN for each translated
range. Non used slots are set to ~0UL, so that in mfn_valid() the mfn <
base check always fails, thus reporting the mfn as invalid.
Introduce __init_or_pdx_mask and use it on some shared functions between
PDX mask and offset compression, as otherwise some code becomes unreachable
after boot if PDX offset compression is used. Mark the code as __init in
that case, so it's pruned after boot.
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Otherwise Coverity complains about possibly shifting an integer more than
31 bits.
This also reverts the previous attempt to fix this Coverity reported
issue, commit 4dd3230290.
Suggested-by: Jan Beulich <jbeulich@suse.com>
Coverity ID: 1662707
Fixes: cb50e40337 ('test/pdx: add PDX compression unit tests')
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
Introduce a set of unit tests for PDX compression. The unit tests contains
both real and crafted memory maps that are then compressed using the
selected PDX algorithm. Note the build system for the unit tests has been
done in a way to support adding new compression algorithms easily. That
requires generating a new test-pdx-<compress> executable that's build with
the selected PDX compression enabled.
Currently the only generated executable is test-pdx-mask that tests PDX
mask compression.
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Anthony PERARD <anthony.perard@vates.tech>
Add a guest config parameter "xenstore_feature_mask" allowing to limit
the Xenstore features the guest can see and use. This can be needed in
order to allow migrating a guest to a host running a Xenstore version
providing less features than the source host.
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
Acked-by: Nick Rosbrook <enr0n@ubuntu.com> # tools/golang
Reviewed-by: Anthony PERARD <anthony.perard@vates.tech>
This patch adds a systemd sleep hook script to stop the xen-watchdog
service before system suspend and start it again after resume.
Stopping the watchdog before a system suspend operation may look unsafe.
Let's imagine the following situation: 'systemctl suspend' does not
interact with the running service at all. In such a case, the Xen
watchdog daemon freezes just before suspend. If this happens, for
example, right before sending a ping, and Xen has not yet marked the
domain as suspended (is_shutting_down), the Xen watchdog timer may
trigger a false alert.
This is an almost impossible situation, because typically:
ping time = watchdog timeout / 2
and the watchdog timeout is usually set to a relatively large value
(dozens of seconds).
Still, this is more likely with very short watchdog timeouts. It may
happen in the following scenarios:
* Significant delays occur between freezing Linux tasks and
triggering the ACPI or PSCI sleep request or handler.
* Long delays happen inside Xen between the entrance to the sleep
trigger and the actual forwarding of the sleep request further.
A similar situation may occur on resume with short timeouts. During the
resume operation, Xen restores timers and the domain context. The Xen
watchdog timer also resumes. If it schedules the domain right before the
watchdog timeout expires, and the daemon responsible for pinging is not
yet running, a timeout might occur.
Both scenarios are rare and typically require very small watchdog
timeouts combined with significant delays in Xen or the Linux kernel
during suspend/resume flows.
Conceptually, however, if activating and pinging the Xen watchdog is the
responsibility of the domain and its services, then the domain should
also manage the watchdog service/daemon lifecycle. This is similar to
what is already done by the Xen watchdog driver inside the Linux kernel.
Signed-off-by: Mykola Kvach <mykola_kvach@epam.com>
Reviewed-by: Denis Mukhin <dmukhin@ford.com>
Reviewed-by: Anthony PERARD <anthony.perard@vates.tech>
When removing qemu-trad support, libxl__domain_build_info_setdefault()
was modified to only allow qemu-xen as a device model. This change
resulted in an error at domain creation when qemu-xen isn't there.
As this is a perfectly fine situation for PV and PVH guests without
any qemu based backends, handle this case gracefully.
Fixes: 19ab8356ab ("tools: remove support for running a guest with qemu-traditional")
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Anthony PERARD <anthony.perard@vates.tech>
Add the Xenstore feature value to the output of "xl info" in order to
prepare for a future capability to limit Xenstore features visible by
a guest.
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
Acked-by: Anthony PERARD <anthony.perard@vates.tech>
SYMBOL_NAME_SIZE is 124, not 128. Use the symbolic name there as well,
to avoid any disconnect.
While there correct indentation on the following line as well.
Coverity ID: 1659391
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Alejandro Vallejo <alejandro.garciavallejo@amd.com>
Acked-by: Anthony PERARD <anthony.perard@vates.tech>
When running xl in a domU, it doesn't have access to the Xen command
line. Before the non-truncating xc_xenver_cmdline(), it was always set
with strdup, possibly of an empty string. Now it's NULL. Treat it the
same as empty cmdline, as it was before. Autoballoon isn't relevant for
xl devd in a domU anyway.
Fixes: 75f9160762 ("tools: Introduce a non-truncating xc_xenver_cmdline()")
Signed-off-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
Acked-by: Anthony PERARD <anthony.perard@vates.tech>
Pass the live_update bool to stubdom_init() in order to be able to
handle it properly under Mini-OS.
As stubdom_init() will (re-)create the 9pfs device needed for
reading the saved state, call stubdom_init() before lu_read_state().
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
After performing a live update in stubdom environment some information
about inter-domain event channels is lost. In order to set this
information again, call xenevtchn_bind() from new_domain() in the
restore case.
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
In order to reactivate an event channel after kexec() of Mini-OS,
libxenevtchn needs to allocate the port data for the event channel
and set the handler again. Add a new interface xenevtchn_bind()
for that purpose, available under Mini-OS only.
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
Acked-by: Anthony PERARD <anthony.perard@vates.tech>
When running in a stubdom environment xenevtchn_fdopen() won't work,
as any file descriptor state is lost across kexec().
Use a wrapper to replace the call of xenevtchn_fdopen() with the
really needed xenevtchn_open() when running on top of Mini-OS.
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
With Mini-OS now supporting the O_CLOEXEC flag for the kexec case,
set the related cloexec flag in minios.c.
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
Acked-by: Anthony PERARD <anthony.perard@vates.tech>
With Mini-OS now supporting O_CLOEXEC for the kexec case, no longer
ignore the XENEVTCHN_NO_CLOEXEC flag in minios.c.
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
Acked-by: Anthony PERARD <anthony.perard@vates.tech>
Mini-OS doesn't support using mmap() for accessing a file. In order
to support reading the live update state from a 9pfs based file, use
fread() instead of mmap().
While adding the offset member to lu_state, change the type of "size"
to size_t in order to avoid problems with state exceeding 4GB.
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
Add an additional "caps" argument to the libxenmanage functions to
obtain a domains capabilities - control, hardware, and xenstore.
Update the xenstored callers at the same time.
Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Acked-by: Anthony PERARD <anthony.perard@vates.tech>
Having multiple values wrapped onto as few lines as practical is good for
space efficiency, but causes complex collisions for hypercall backports and
local policy changes. Reformat to use one value per line.
No functional change, only whitespace changes.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by: Daniel P. Smith <dpsmith@apertussolutions.com>
Most indentation is with tabs, but a few spaces have slipped in. Switch them
back to tabs.
No functional change.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by: Daniel P. Smith <dpsmith@apertussolutions.com>
Hypercalls are not free; cache colouring is an experimental feature, and
ignoring an -EOPNOTSUPP is bad form.
Now that Xen has been fixed to initialise colouring information correctly for
domains, xc_domain_set_llc_colors() only needs calling for domains with
explicit configuration.
Rearrange the logic to avoid the hypercall in the general case, leaving a
comment explaining why it is performed so early.
Fixes: 748bd725fb ("tools: add support for cache coloring configuration")
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Stefano Stabellini <sstabellini@kernel.org>
Add support for XS_SET_FEATURE and XS_GET_FEATURE to xenstored.
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
Add per domain server features, which are initialized by the supported
features at domain introduction, or by live update from the migration
stream. This requires to add the DOMAIN_DATA record to the migration
stream, but for now it will only contain the feature word.
Advertise the Xenstore features to guests by setting the appropriate
bits in the ring page.
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
Add all the missing migration record structures and defines to the
xenstore_state.h header file.
Update the version of the migration stream to "2". In order to allow
receiving version 1 streams, change the version check of the header
to reject only streams with a version higher than the current known
one.
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
In order to prepare xenstored supporting migration stream versions
other than 1, add a parameter to the live update command allowing to
specify the version of the migration stream.
This will allow going back from xenstored using version 2 per default
to a xenstored only accepting version 1.
For now only version 1 is supported.
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
Add functions for getting and setting Xenstore features to libxenstore:
xs_get_features_supported(): return the features supported by the
running Xenstore implementation as defined in xs_wire.h via the
XENSTORE_SERVER_FEATURE_* macros.
xs_get_features_domain(): return the features offered for a specific
domain.
xs_set_features_domain(): set the features available for a specific
domain.
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
The current strtoll() implementation in hvmloader requires hex number to be
prefixed with 0x, otherwise strtoll() won't parse them correctly even when
calling the function with base == 16.
Fix this by not unconditionally setting the base to 10 when the string is
not 0 prefixed, this also allows parsing octal numbers not 0 prefixed.
While there also handle '0X' as a valid hex number prefix, together with
'0x'.
No functional change intended to the existing call sites.
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Introduce a new altp2m_count parameter to control the maximum number of altp2m
views a domain can use. By default, if altp2m_count is unspecified and altp2m
is enabled, the value is set to 10, reflecting the legacy behavior.
This change is preparatory; it establishes the groundwork for the feature but
does not activate it.
Signed-off-by: Petr Beneš <w1benny@gmail.com>
Acked-by: Nick Rosbrook <enr0n@ubuntu.com>
Reviewed-by: Anthony PERARD <anthony.perard@vates.tech>
For the final step of live-update in stubdom, call kexec().
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
Add the required configuration settings to enable kexec in the PVH
variant of xenstore-stubdom. This is required to support live update.
Set the number of kexec module pages to 18 in order to support using
the kexec module memory for the stubdom's 9pfs granted pages.
In Xenstore code test the CONFIG_KEXEC setting in order to determine
whether stubdom live update is supported (PV xenstore-stubdom doesn't
have live update support, as it lacks kexec).
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
"ondemand" is more suitable to describe a preset in which epp value is set
with medium(CPPC_ENERGY_PERF_BALANCE), showing no preference over performance or
powersave, minimum with lowest and maximum with highest.
Signed-off-by: Penny Zheng <Penny.Zheng@amd.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
Introduce a new altp2m_count parameter to control the maximum number of altp2m
views a domain can use. By default, if altp2m_count is unspecified and altp2m
is enabled, the value is set to 10, reflecting the legacy behavior.
This change is preparatory; it establishes the groundwork for the feature but
does not activate it.
Signed-off-by: Petr Beneš <w1benny@gmail.com>
Reviewed-by: Anthony PERARD <anthony.perard@vates.tech>
Acked-by: Nick Rosbrook <enr0n@ubuntu.com>
Encapsulate the altp2m options within a struct. This change is preparatory
and sets the groundwork for introducing additional parameter in subsequent
commit.
Signed-off-by: Petr Beneš <w1benny@gmail.com>
Acked-by: Christian Lindig <christian.lindig@cloud.com>
Acked-by: Anthony PERARD <anthony.perard@vates.tech>
Acked-by: Julien Grall <jgrall@amazon.com> # arm
Reviewed-by: Jan Beulich <jbeulich@suse.com> # hypervisor
As a result of the addition of a new field in libxl domain build info
structure the golang types need to be regnerated, this was missing as part
of 22650d6054.
Regenerate the headers now.
Reported-by: Juergen Gross <jgross@suse.com>
Fixes: 22650d6054 ('x86/hvmloader: select xen platform pci MMIO BAR UC or WB MTRR cache attribute')
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Acked-by: Nick Rosbrook <enr0n@ubuntu.com>
With introduction of the new byteswap infrastructure the build of
libxenguest for stubdoms was broken. Fix that again.
Fixes: 60dcff871e ("xen/decompressors: Remove use of *_to_cpup() helpers")
Signed-off-by: Juergen Gross <jgross@suse.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jason Andryuk <jason.andryuk@amd.com>
Acked-by: Anthony PERARD <anthony.perard@vates.tech>
The Xen platform PCI device (vendor ID 0x5853) exposed to x86 HVM guests
doesn't have the functionality of a traditional PCI device. The exposed
MMIO BAR is used by some guests (including Linux) as a safe place to map
foreign memory, including the grant table itself.
Traditionally BARs from devices have the uncacheable (UC) cache attribute
from the MTRR, to ensure correct functionality of such devices. hvmloader
mimics this behavior and sets the MTRR attributes of both the low and high
PCI MMIO windows (where BARs of PCI devices reside) as UC in MTRR.
This however causes performance issues for users of the Xen platform PCI
device BAR, as for the purposes of mapping remote memory there's no need to
use the UC attribute. On Intel systems this is worked around by using
iPAT, that allows the hypervisor to force the effective cache attribute of
a p2m entry regardless of the guest PAT value. AMD however doesn't have an
equivalent of iPAT, and guest PAT values are always considered.
Linux commit:
41925b105e34 xen: replace xen_remap() with memremap()
Attempted to mitigate this by forcing mappings of the grant-table to use
the write-back (WB) cache attribute. However Linux memremap() takes MTRRs
into account to calculate which PAT type to use, and seeing the MTRR cache
attribute for the region being UC the PAT also ends up as UC, regardless of
the caller having requested WB.
As a workaround to allow current Linux to map the grant-table as WB using
memremap() introduce an xl.cfg option (xen_platform_pci_bar_uc=0) that can
be used to select whether the Xen platform PCI device BAR will have the UC
attribute in MTRR. Such workaround in hvmloader should also be paired with
a fix for Linux so it attempts to change the MTRR of the Xen platform PCI
device BAR to WB by itself.
Overall, the long term solution would be to provide the guest with a safe
range in the guest physical address space where mappings to foreign pages
can be created.
Some vif throughput performance figures provided by Anthoine from a 8
vCPUs, 4GB of RAM HVM guest(s) running on AMD hardware:
Without this patch:
vm -> dom0: 1.1Gb/s
vm -> vm: 5.0Gb/s
With the patch:
vm -> dom0: 4.5Gb/s
vm -> vm: 7.0Gb/s
Reported-by: Anthoine Bourgeois <anthoine.bourgeois@vates.tech>
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Reviewed-by: Oleksii Kurochko<oleksii.kurochko@gmail.com>
Acked-by: Jan Beulich <jbeulich@suse.com> # hvmloader
Reviewed-by: Anthony PERARD <anthony.perard@vates.tech>
When a device is passed through to a xl domU, the iomem ranges may
overlap with the extended regions. Remove iomem from extended regions.
Signed-off-by: Stewart Hildebrand <stewart.hildebrand@amd.com>
Reviewed-by: Anthony PERARD <anthony.perard@vates.tech>
Juergen Gross
Jason Andryuk
Roger Pau Monne
Mykola Kvach
Jan Beulich
Marek Marczykowski-Górecki
Andrew Cooper
Petr Beneš
Penny Zheng
Stewart Hildebrand