As opposed to the existing virCPUCompare{,XML} this function does not
use CPU model definitions from CPU map. It relies on CPU model usability
info from a hypervisor with a list of blockers that make the selected
CPU model unusable. Explicitly requested features are checked against
the hypervisor's view of a host CPU.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
On x86 the function returns whether an old style compat check mode
should be used for a specified CPU model according to the CPU map. All
other architectures will always use compat mode.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
CPU models in the CPU map may be marked with <check partial="compat"/>
to indicate a backward compatible partial check (comparing our
definition of the model with the host CPU) should be performed. Other
models will be checked using just runnability info from QEMU.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
qemu supports this enlightenment since version 7.10.
From the qemu commit:
Hyper-V specification allows to pass parameters for certain hypercalls
using XMM registers ("XMM Fast Hypercall Input"). When the feature is
in use, it allows for faster hypercalls processing as KVM can avoid
reading guest's memory.
Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
qemu supports this enlightenment since version 7.10.
From the qemu commit:
The newly introduced enlightenment allow L0 (KVM) and L1 (Hyper-V)
hypervisors to collaborate to avoid unnecessary updates to L2
MSR-Bitmap upon vmexits.
Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
When starting a domain on a host which lacks a vmx-* CPU feature which
is expected to be enabled by the CPU model specified in the domain XML,
libvirt properly marks such feature as disabled in the active domain
XML. But migrating the domain to a similar host which lacks the same
vmx-* feature will fail with libvirt reporting the feature as missing.
This is because of a bug in the hack ensuring backward compatibility
libvirt running on the destination thinks the missing feature is
expected to be enabled.
https://issues.redhat.com/browse/RHEL-40899
Fixes: v10.1.0-85-g5fbfa5ab8a
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
virCPUUpdate check the CPU definition for features that were marked as
removed in the specified CPU model and explicitly adds those that were
not mentioned in the definition. So far such features were added with
VIR_CPU_FEATURE_DISABLE policy, but the caller may want to use a
different policy in some situations, which is now possible via the
removedPolicy parameter.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The virCPUDefAddFeatureInternal helper function only fails if it is
called with VIR_CPU_ADD_FEATURE_MODE_EXCLUSIVE, which is only used in
virCPUDefAddFeature. The other callers (virCPUDefUpdateFeature and
virCPUDefAddFeatureIfMissing) will never get anything but 0 from
virCPUDefAddFeatureInternal and their return type can be changed to
void.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Add loongarch cpu support, Define new cpu type 'loongarch64'
and implement it's driver functions.
Signed-off-by: Xianglai Li <lixianglai@loongson.cn>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
This is not a good idea in general, but we can (and have to) do it in
specific cases when a feature has always been part of a CPU model in
hypervisor's definition, but we ignored it and did not include the
feature in our definition.
Blindly adding the features to the CPU map and not adding them to
existing CPU models breaks migration between old and new libvirt in both
directions. New libvirt would complain the features got unexpectedly
enabled (as they were not mentioned in the incoming domain XML) even
though they were also enabled on the source and the old libvirt just
didn't know about them. On the other hand, old libvirt would refuse to
accept incoming migration of a domain started by new libvirt because the
domain XML would contain CPU features unknown to the old libvirt.
This is exactly what happened when several vmx-* features were added a
few releases back. Migration between libvirt releases before and after
the addition is now broken.
This patch adds support for added these features to existing CPU models
by marking them with added='yes'. The features will not be considered
part of the CPU model and will be described explicitly via additional
<feature/> elements, but the compatibility check will not complain if
they are enabled by the hypervisor even though they were not explicitly
mentioned in the CPU definition and incoming migration from old libvirt
will succeed.
To fix outgoing migration to old libvirt, we also need to drop all those
features from domain XML unless they were explicitly requested by the
user. This will be handled by a later patch.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
Recently a kernel bug caused QEMU to report a CPU feature as enabled
while listing it in the "unavailable-features" list of features that
were requested, but could not be enabled. The feature was actually
enabled, but we marked it as disabled when starting a domain. Later when
the domain is migrated, the destination requests the feature to be
disabled, which breaks the guest ABI or if we are lucky QEMU just fails
to load the migration stream.
Let's make similar bugs more visible in the future by refusing to even
start the domain.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
'virXPathNodeSet' returns -1 only when 'ctxt' or 'xpath' are NULL or
when the 'xpath' string is invalid. Both are programming errors. It
doesn't make sense for the code to overwrite the error message for
anything supposedly more relevant.
The majority of calls to 'virXPathNodeSet' already didn't do this, so
this patch fixes the rest to prevent it from spreading again.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The default number of CPU clusters is 1, and values other than
that one are currently rejected by all hypervisor drivers.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
virCPUx86DataAddItem() return value is invariant, so change it
type and remove all dependent checks.
Functions changed to void:
virCPUx86DataAddItem()
x86DataAdd()
virCPUx86DataAdd()
x86DataAddSignature()
virCPUx86DataSetSignature()
libxlCapsAddCPUID()
cpuidSetLeaf4()
cpuidSetLeaf7()
cpuidSetLeafB()
cpuidSetLeafD()
cpuidSetLeafResID()
cpuidSetLeaf12()
cpuidSetLeaf14()
cpuidSetLeaf17()
Signed-off-by: Artem Chernyshev <artem.chernyshev@red-soft.ru>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
While glibc provides qsort(), which usually is just a mergesort,
until sorting arrays so huge that temporary array used by
mergesort would not fit into physical memory (which in our case
is never), we are not guaranteed it'll use mergesort. The
advantage of mergesort is clear - it's stable. IOW, if we have an
array of values parsed from XML, qsort() it and produce some
output based on those values, we can then compare the output with
some expected output, line by line.
But with newer glibc this is all history. After [1], qsort() is
no longer mergesort but introsort instead, which is not stable.
This is suboptimal, because in some cases we want to preserve
order of equal items. For instance, in ebiptablesApplyNewRules(),
nwfilter rules are sorted by their priority. But if two rules
have the same priority, we want to keep them in the order they
appear in the XML. Since it's hard/needless work to identify
places where stable or unstable sorting is needed, let's just
play it safe and use stable sorting everywhere.
Fortunately, glib provides g_qsort_with_data() which indeed
implement mergesort and it's a drop in replacement for qsort(),
almost. It accepts fifth argument (pointer to opaque data), that
is passed to comparator function, which then accepts three
arguments.
We have to keep one occurance of qsort() though - in NSS module
which deliberately does not link with glib.
1: https://sourceware.org/git/?p=glibc.git;a=commitdiff;h=03bf8357e8291857a435afcc3048e0b697b6cc04
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
Error messages are exempt from the 80 columns rule. Move them
onto one line.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
This is a more concise approach and guarantees there is
no time window where the struct is uninitialized.
Generated using the following semantic patch:
@@
type T;
identifier X;
@@
- T X;
+ T X = { 0 };
... when exists
(
- memset(&X, 0, sizeof(X));
|
- memset(&X, 0, sizeof(T));
)
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Claudio Fontana <cfontana@suse.de>
Commit be1b7d5b18 introduced parsing /proc/cpuinfo for "address size"
which is not including on S390 and therefore reports an internal error.
Lets remove the parsing on S390.
Signed-off-by: Boris Fiuczynski <fiuczy@linux.ibm.com>
Reviewed-by: Marc Hartmayer <mhartmay@linux.ibm.com>
Reviewed-by: Collin Walling <walling@linux.ibm.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
At this moment it is not possible to launch a 'riscv64' domain if a CPU
definition is presented in the domain. For example, adding this CPU
definition:
<cpu mode='custom' match='exact' check='none'>
<model fallback='forbid'>rv64</model>
</cpu>
Will trigger the following error:
$ sudo ./run tools/virsh start riscv-virt1
error: Failed to start domain 'riscv-virt1'
error: this function is not supported by the connection driver:
cannot update guest CPU for riscv64 architecture
The error comes from virCPUUpdate(), via qemuProcessUpdateGuestCPU(),
and it's caused by the absence of the 'update' API in the existing
RISC-V driver.
Add an 'update' API impl to the RISC-V driver to allow for CPU
definitions to be declared in RISC-V domains. This API was copied from
the ARM driver (virCPUarmUpdate()) since it's a good enough
implementation to get us going.
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
Make all callers always pass a valid pointer which in turn allows us to
remove return value check from the callers.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Make all callers always pass a valid pointer which in turn allows us to
remove return value check from the callers.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
When running "virsh domcapabilities" on a s390x host, all the CPU
models show up with vendor='unknown' - which sounds kind of weird
since the vendor of these mainframe CPUs is well known: IBM.
All CPUs starting with either "z" or "gen" match a real mainframe
CPU by IBM, so let's return the string "IBM" for those now.
The only remaining ones are now the artifical "qemu" and "max"
models from QEMU itself, so it should be OK to get an "unknown"
vendor for those two.
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
Signed-off-by: Boris Fiuczynski<fiuczy@linux.ibm.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
There are tests in qemuxml2argvtest that will fail if we enable RISC-V
testing, with an error like the following:
"cpuGetSubDriver:64 : this function is not supported by the connection
driver: 'riscv64' architecture is not supp orted by CPU driver"
This happens because we don't have a RISC-V driver yet.
Add a barebone RISC-V driver to allow tests to be executed. The only 2
callbacks implemented here are 'compare' and 'validateFeatures', both
acting as a no-op. More callbacks and features will be added in the
future.
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
These format are left unchanged when convert 'unsigned long' to
'unsigned long long', which caused compile warning.
Signed-off-by: Jiang Jiacheng <jiangjiacheng@huawei.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Ján Tomko <jtomko@redhat.com>
Based on qemu commit e1f9a8e8c90ae54387922e33e5ac4fd759747d01 introduce
the hv-avic feature in leaf 0x40000004, EAX 0x00000200 (1 << 9).
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
'VIR_CPU_x86_HV_STIMER_DIRECT' is reported under leaf 0x40000003,
but the data is in the EDX register. Create a new group for such
features and move them after the 0x40000003 EAX group.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
A later patch will add alias names to the feature map. They will be used
in virQEMUCapsCPUFeatureTranslate and for synchronizing the list with QEMU.
Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
We don't need to do the extra XPath lookups and we can use the proper
type right away.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
This patch is effectively a no-op, but I wanted to initialize
.getVendorForModel explicitly as implementing this function does not
even make sense on ARM. The CPU models in our CPU map are only used for
describing host CPU in capabilities XML and cannot be used for guest CPU
definition in domain XML anyway. The CPU models listed as supported in
domain capabilities XML are just passed through from QEMU.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
So far QEMU driver does not get CPU model vendor from QEMU directly and
it has to ask the CPU driver for the info stored in CPU map.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The ppc64 CPU code still has to load and parse the CPU map everytime it
needs to look at it, which can make some operations pretty slow. Other
archs already switched to loading the CPU map once and keeping the
parsed structure in memory. Let's switch ppc64 as well.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The G_GNUC_NO_INLINE macro will eventually be marked as
deprecated [1] and we are recommended to use G_NO_INLINE instead.
Do the switch now, rather than waiting for compile time warning
to occur.
1: 15cd0f0461
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
Internally we already collect x86 host family + model + stepping
numeric values. This exposed them in capabilities CPU output.
Example:
$ sudo virsh capabilities | grep -A1 -B1 signature
<microcode version='240'/>
<signature family='6' model='94' stepping='3'/>
<counter name='tsc' frequency='3408010000' scaling='no'/>
Users need to know these values to calculate an expected.
SEV-ES/SEV-SNP launch measurement.
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
Signed-off-by: Cole Robinson <crobinso@redhat.com>
While we don't want to aim for the shortest list of disabled features in
the baseline result (it would select a very old model), we want to do so
while looking at any of the input models for which we're trying to
compute a baseline CPU model. Given a set of input models, we always
want to take the least capable one of them (i.e., the one with shortest
list of disabled features) or a better model which is not one of the
input models.
So when considering an input model, we just check whether its list of
disabled features is shorter than the currently best one. When looking
at other models we check both enabled and disabled features while
penalizing disabled features as implemented by the previous patch.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
For finding the best matching CPU model for a given set of features
while we don't know the CPU signature (i.e., when computing a baseline
CPU model) we've been using a "shortest list of features" heuristics.
This works well if new CPU models are supersets of older models, but
that's not always the case. As a result it may actually select a new CPU
model as a baseline while removing some features from it to make it
compatible with older models. This is in general worse than using an old
CPU model with a bunch of added features as a guest OS or apps may crash
when using features that were disabled.
On the other hand we don't want to end up with a very old model which
would guarantee no disabled features as it could stop a guest OS or apps
from using some features provided by the CPU because they would not
expect them on such an old CPU.
This patch changes the heuristics to something in between. Enabled and
disabled features are counted separately so that a CPU model requiring
some features to be disabled looks worse than a model with fewer
disabled features even if its complete list of features is longer. The
penalty given for each additional disabled feature gets bigger to make
longer list of disabled features look even worse.
https://bugzilla.redhat.com/show_bug.cgi?id=1851227
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
It will become more complicated and so it deserves to be separated into
a new function.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Checking the signature in two different places makes no sense since the
code in between can only mark the candidate as the best option so far,
which is what the second signature match does as well.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
These two pointers can never be NULL since they are initialised to a reference
of a struct. This became apparent when commit 210a19539447 added a VIR_DEBUG
which used both pointers because due to the concise condition the compiler saw
that if the "and" part of the condition did short-circuit (and it assumed that
can happen) the second variable would not be initialised, but it is used in the
debugging message, so the build failed with:
In file included from ../src/cpu/cpu_x86.c:27:
../src/cpu/cpu_x86.c: In function ‘virCPUx86DataIsIdentical’:
../src/util/virlog.h:79:5: error: ‘bdata’ may be used uninitialized in this
function [-Werror=maybe-uninitialized]
Fix this by just assigning the helper pointers and remove the condition
altogether.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
The function returns 'virCPUData' but doesn't do two important steps
which other code takes:
1) leaves with all-zero data is stripped from the XML output
2) the data is expected to be sorted in the array
Now the 'virHostCPUGetCPUID' helper returns both all 0 leaves and
doesn't order them as we expect.
If this is then used in conjunction with 'virCPUx86DataIsIdentical'
together with data which made a roundtrip to XML and back the result
will be always false even if the data itself is identical.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Without this it's impossible to debug scenarios when this function
returns a mismatch but the formatted data looks identical.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
There are couple of places where virTristateBoolTypeFromString()
is called. Well, the same result can be achieved by
virXMLPropTristateBool() and on fewer lines.
Note there are couple of places left untouched because those
don't care about error reporting and thus are shorter they way
they are now.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
