When the "migratable" property was implemented, the behavior was tested
by changing the default on the code, but actually using the option on
the command-line (e.g. "-cpu host,migratable=false") doesn't work as
expected. This is a regression for a common use case of "-cpu host",
which is to enable features that are supported by the host CPU + kernel
before feature-specific code is added to QEMU.
Fix this by initializing the feature words for "-cpu host" on
x86_cpu_parse_featurestr(), right after parsing the CPU options.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Cc: qemu-stable@nongnu.org
Signed-off-by: Andreas Färber <afaerber@suse.de>
tsc_adjust migration support is already implemented (commit
f28558d3d3), so we can add it to the list
of known feature names.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Migration support for MPX is already implemented (commit
79e9ebebbf), so we can add it to the list
of known feature names.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The SDM specifies (June 2014 Vol3 11.11.5):
On a hardware reset, the P6 and more recent processors clear the
valid flags in variable-range MTRRs and clear the E flag in the
IA32_MTRR_DEF_TYPE MSR to disable all MTRRs. All other bits in the
MTRRs are undefined.
We currently do none of that, so whatever MTRR settings you had prior
to reset is what you have after reset. Usually this doesn't matter
because KVM often ignores the guest mappings and uses write-back
anyway. However, if you have an assigned device and an IOMMU that
allows NoSnoop for that device, KVM defers to the guest memory
mappings which are now stale after reset. The result is that OVMF
rebooting on such a configuration takes a full minute to LZMA
decompress the firmware volume, a process that is nearly instant on
the initial boot.
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Cc: qemu-stable@nongnu.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Chen Fan <chen.fan.fnst@cn.fujitsu.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
KVM_FEATURE_CLOCKSOURCE_STABLE_BIT is enabled by default and supported
by KVM. But not having a name defined makes QEMU treat it as an unknown
and unmigratable feature flag (as any unknown feature may possibly
require state to be migrated), and disable it by default on "-cpu host".
As a side-effect, the new name also makes the flag configurable,
allowing the user to disable it (which may be useful for testing or for
compatibility with old kernels).
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This adds a new CPU model named "Broadwell". It has all the features
from Haswell, plus PREFETCHW, RDSEED, ADX, SMAP.
PREFETCHW was already supported as "3dnowprefetch".
RDSEED, ADX was added on Linux v3.15-rc1.
SMAP was added on Linux v3.15-rc2.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Cc: Wang, Yong Y <yong.y.wang@intel.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Dugger, Donald D <donald.d.dugger@intel.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Expose "Invariant TSC" flag, if KVM is enabled. From Intel documentation:
17.13.1 Invariant TSC The time stamp counter in newer processors may
support an enhancement, referred to as invariant TSC. Processor’s
support for invariant TSC is indicated by CPUID.80000007H:EDX[8].
The invariant TSC will run at a constant rate in all ACPI P-, C-.
and T-states. This is the architectural behavior moving forward. On
processors with invariant TSC support, the OS may use the TSC for wall
clock timer services (instead of ACPI or HPET timers). TSC reads are
much more efficient and do not incur the overhead associated with a ring
transition or access to a platform resource.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
[ehabkost: redo feature filtering to use .tcg_features]
[ehabkost: add CPUID_APM_INVTSC macro, add it to .unmigratable_flags]
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Having only migratable flags reported by default on the "host" CPU model
is safer for the following reasons:
* Existing users may expect "-cpu host" to be migration-safe, if they
take care of always using compatible host CPUs, host kernels, and
QEMU versions.
* Users who don't care aboug migration and want to enable all features
supported by the host kernel can simply change their setup to use
migratable=no.
Without this change, people using "-cpu host" will stop being able to
migrate, because now "invtsc" is getting enabled by default.
We are not setting migratable=yes by default on all X86CPU subclasses,
because users should be able to get non-migratable features enabled if
they ask for them explicitly.
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
This flag will allow the user to choose between two modes:
* All flags that can be enabled on the host, even if unmigratable
(migratable=no);
* All flags that can be enabled on the host, are known to QEMU
and migratable (migratable=yes).
The default is still migratable=false, to keep current behavior, but
this will be changed to migratable=true by another patch.
My plan was to support the "migratable" flag on all CPU classes, but
have the default to "false" on all CPU models except "host". However,
DeviceClass has no mechanism to allow a child class to have a different
property default from the parent class yet, so by now only the "host"
CPU model will support the "migratable" flag.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
If enforce/check is specified in TCG mode, QEMU will ensure all CPU
features are supported by TCG, so no CPU feature is silently disabled.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
[AF: Be explicit about TCG vs. !KVM]
Signed-off-by: Andreas Färber <afaerber@suse.de>
Instead of manually filtering each feature word, add a tcg_features
field to FeatureWordInfo, and use that field to filter all feature words
in TCG mode.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Now that we have the feature word arrays, we don't need to manually copy
each array item, we can simply iterate through each feature word.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Those macros will be used in the feature_word_info array data, so need
to be defined earlier.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
TCG doesn't support any of the feature flags on FEAT_KVM and
FEAT_C000_0001_EDX feature words, so clear all bits on those feature
words.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
The TCG_7_0_EBX_FEATURES macro was defined but never used (it even had a
typo that was never noticed). Make the existing TCG feature filtering
code use it.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Cc: qemu-stable@nongnu.org
Signed-off-by: Andreas Färber <afaerber@suse.de>
Instead of an #ifdef in the middle of the code, just set
TCG_EXT2_FEATURES to a different value depending on TARGET_X86_64.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
This will allow us to re-use the feature filtering logic (and the
check/enforce flag logic) for TCG.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
This will help us simplify the code that calls
report_unavailable_features() later.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Merge filter_features_for_kvm() and kvm_check_features_against_host().
Both functions made exactly the same calculations, the only difference
was that filter_features_for_kvm() changed the bits on cpu->features[],
and kvm_check_features_against_host() did error reporting.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Instead of checking and calling unavailable_host_feature() once for each
bit, simply call the function (now renamed to
report_unavailable_features()) once for each feature word.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
[AF: Drop unused return value]
Signed-off-by: Andreas Färber <afaerber@suse.de>
KVM never supported the MONITOR flag so it doesn't make sense to have it
enabled by default when KVM is enabled.
The rationale here is similar to the cases where it makes sense to have
a feature enabled by default on all CPU models when on KVM mode (e.g.
x2apic). In this case we are having a feature disabled by default for
the same reasons.
In this case we don't need machine-type compat code because it is
currently impossible to run a KVM VM with the MONITOR flag set.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Because of the "goto out", the contents of local_err are leaked
and lost.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
* remotes/bonzini/softmmu-smap: (33 commits)
target-i386: cleanup x86_cpu_get_phys_page_debug
target-i386: fix protection bits in the TLB for SMEP
target-i386: support long addresses for 4MB pages (PSE-36)
target-i386: raise page fault for reserved bits in large pages
target-i386: unify reserved bits and NX bit check
target-i386: simplify pte/vaddr calculation
target-i386: raise page fault for reserved physical address bits
target-i386: test reserved PS bit on PML4Es
target-i386: set correct error code for reserved bit access
target-i386: introduce support for 1 GB pages
target-i386: introduce do_check_protect label
target-i386: tweak handling of PG_NX_MASK
target-i386: commonize checks for PAE and non-PAE
target-i386: commonize checks for 4MB and 4KB pages
target-i386: commonize checks for 2MB and 4KB pages
target-i386: fix coding standards in x86_cpu_handle_mmu_fault
target-i386: simplify SMAP handling in MMU_KSMAP_IDX
target-i386: fix kernel accesses with SMAP and CPL = 3
target-i386: move check_io helpers to seg_helper.c
target-i386: rename KSMAP to KNOSMAP
...
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
4MB pages can use 40-bit addresses by putting the higher 8 bits in bits
20-13 of the PDE. Bit 21 is reserved.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The latest Nvidia driver (337.88) specifically checks for KVM as the
hypervisor and reports Code 43 for the driver in a Windows guest when
found. Removing or changing the KVM signature is sufficient for the
driver to load and work. This patch adds an option to easily allow
the KVM hypervisor signature to be hidden using '-cpu kvm=off'. We
continue to expose KVM via the cpuid value by default. The state of
this option does not supercede or replace -enable-kvm or the accel=kvm
machine option. This only changes the visibility of KVM to the guest
and paravirtual features specifically tied to the KVM cpuid.
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Most MSRs, plus the FPU, MMX, MXCSR, XMM and YMM registers should not
be zeroed on INIT (Table 9-1 in the Intel SDM). Copy them out of
CPUX86State and back in, instead of special casing env->pat.
The relevant fields are already consecutive except PAT and SMBASE.
However:
- KVM and Hyper-V MSRs should be reset because they include memory
locations written by the hypervisor. These MSRs are moved together
at the end of the preserved area.
- SVM state can be moved out of the way since it is written by VMRUN.
Cc: Andreas Faerber <afaerber@suse.de>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
BND0-3, BNDCFGU, BNDCFGS, BNDSTATUS were not zeroed on reset, but they
should be (Intel Instruction Set Extensions Programming Reference
319433-015, pages 9-4 and 9-6). Same for YMM.
XCR0 should be reset to 1.
TSC and TSC_RESET were zeroed already by the memset, remove the explicit
assignments.
Cc: Andreas Faerber <afaerber@suse.de>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Now that we have a CPU object with a reset method, it is better to
keep the KVM reset close to the CPU reset. Using qemu_register_reset
as we do now keeps them far apart.
With this patch, PPC no longer calls the kvm_arch_ function, so
it can get removed there. Other arches call it from their CPU
reset handler, and the function gets an ARMCPU/X86CPU/S390CPU.
Note that ARM- and s390-specific functions are called kvm_arm_*
and kvm_s390_*, while x86-specific functions are called kvm_arch_*.
That follows the convention used by the different architectures.
Changing that is the topic of a separate patch.
Reviewed-by: Gleb Natapov <gnatapov@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Using error_is_set(ERRP) to find out whether a function failed is
either wrong, fragile, or unnecessarily opaque. It's wrong when ERRP
may be null, because errors go undetected when it is. It's fragile
when proving ERRP non-null involves a non-local argument. Else, it's
unnecessarily opaque (see commit 84d18f0).
I guess the error_is_set(errp) in the ObjectProperty set() methods are
merely fragile right now, because I can't find a call chain that
passes a null errp argument.
Make the code more robust and more obviously correct: receive the
error in a local variable, then propagate it through the parameter.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
This fixes warnings from the static code analysis (smatch).
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Most targets were using offsetof(CPUFooState, breakpoints) to determine
how much of CPUFooState to clear on reset. Use the next field after
CPU_COMMON instead, if any, or sizeof(CPUFooState) otherwise.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Note that while such functions may exist both for *-user and softmmu,
only *-user uses the CPUState hook, while softmmu reuses the prototype
for calling it directly.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Adapt the X86CPU implementation to suit the generic hook.
This involves a cleanup of error handling to cope with NULL errp.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Register separate QOM types for each x86 CPU model.
This will allow management code to more easily probe what each CPU model
provides, by simply creating objects using the appropriate class name,
without having to restart QEMU.
This also allows us to eliminate the qdev_prop_set_globals_for_type()
hack to set CPU-model-specific global properties.
Instead of creating separate class_init functions for each class, I just
used class_data to store a pointer to the X86CPUDefinition struct for
each CPU model. This should make the patch shorter and easier to review.
Later we can gradually convert each X86CPUDefinition field to lists of
per-class property defaults.
The "host" CPU model is special, as the feature flags depend on KVM
being initialized. So it has its own class_init and instance_init
function, and feature flags are set on instance_init instead of
class_init.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Tested-by: Eduardo Habkost <ehabkost@redhat.com>
[AF: Limit the host CPU type to CONFIG_KVM as build fix]
Signed-off-by: Andreas Färber <afaerber@suse.de>
When on KVM mode, enable x2apic by default on all CPU models.
Normally we try to keep the CPU model definitions as close as the real
CPUs as possible, but x2apic can be emulated by KVM without host CPU
support for x2apic, and it improves performance by reducing APIC access
overhead. x2apic emulation is available on KVM since 2009 (Linux
2.6.32-rc1), there's no reason for not enabling x2apic by default when
running KVM.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Instead of the feature-specific disable_kvm_pv_eoi() function, create a
more general function that can be used to disable other feature bits in
machine-type compat code.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
We will later make the KVM-specific code affect other feature words,
too.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Some of my recent changes introduced variable declarations in the middle
of code blocks.
Fix the code so that it compiles without warnings when using
-Wdeclaration-after-statement.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
As the new X86CPU subclass code is going to change lots of the code
invoving x86_def_t, let's rename the struct to match coding style first.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
As we will initialize the X86CPU fields on instance_init eventually,
move the code that initializes the X86CPU data based on the CPU model
name closer to the object_new() call.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
There isn't any kind of "registration" involved in cpu_x86_register()
anymore: it is simply looking up a CPU model name and loading the model
definition data into the X86CPU object. Rename it to x86_cpu_load_def()
to reflect what it does.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Default to false.
Tidy variable naming and inline cast uses while at it.
Tested-by: Jia Liu <proljc@gmail.com> (or32)
Signed-off-by: Andreas Färber <afaerber@suse.de>
* remotes/qmp-unstable/queue/qmp:
tests: test-qmp-commands: Fix double free
qapi script: do not add "_" for every capitalized char in enum
qapi script: do not allow string discriminator
qapi: convert BlockdevOptions to use enum discriminator
qapi script: support enum type as discriminator in union
qapi script: use same function to generate enum string
qapi script: code move for generate_enum_name()
qapi script: check correctness of union
qapi script: remember line number in schema parsing
qapi script: add check for duplicated key
qapi script: remember explicitly defined enum values
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
