The check on x86ms->apic_id_limit in pc_machine_done() had two problems.
Firstly, we need KVM to support the X2APIC API in order to allow IRQ
delivery to APICs >= 255. So we need to call/check kvm_enable_x2apic(),
which was done elsewhere in *some* cases but not all.
Secondly, microvm needs the same check. So move it from pc_machine_done()
to x86_cpus_init() where it will work for both.
The check in kvm_cpu_instance_init() is now redundant and can be dropped.
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Acked-by: Claudio Fontana <cfontana@suse.de>
Message-Id: <20220314142544.150555-1-dwmw2@infradead.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Some AMD processors expose the PKRU extended save state even if they do not have
the related PKU feature in CPUID. Worse, when they do they report a size of
64, whereas the expected size of the PKRU extended save state is 8, therefore
the esa->size == eax assertion does not hold.
The state is already ignored by KVM_GET_SUPPORTED_CPUID because it
was not enabled in the host XCR0. However, QEMU kvm_cpu_xsave_init()
runs before QEMU invokes arch_prctl() to enable dynamically-enabled
save states such as XTILEDATA, and KVM_GET_SUPPORTED_CPUID hides save
states that have yet to be enabled. Therefore, kvm_cpu_xsave_init()
needs to consult the host CPUID instead of KVM_GET_SUPPORTED_CPUID,
and dies with an assertion failure.
When setting up the ExtSaveArea array to match the host, ignore features that
KVM does not report as supported. This will cause QEMU to skip the incorrect
CPUID leaf instead of tripping the assertion.
Closes: https://gitlab.com/qemu-project/qemu/-/issues/916
Reported-by: Daniel P. Berrangé <berrange@redhat.com>
Analyzed-by: Yang Zhong <yang.zhong@intel.com>
Reported-by: Peter Krempa <pkrempa@redhat.com>
Tested-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Kernel allocates 4K xstate buffer by default. For XSAVE features
which require large state component (e.g. AMX), Linux kernel
dynamically expands the xstate buffer only after the process has
acquired the necessary permissions. Those are called dynamically-
enabled XSAVE features (or dynamic xfeatures).
There are separate permissions for native tasks and guests.
Qemu should request the guest permissions for dynamic xfeatures
which will be exposed to the guest. This only needs to be done
once before the first vcpu is created.
KVM implemented one new ARCH_GET_XCOMP_SUPP system attribute API to
get host side supported_xcr0 and Qemu can decide if it can request
dynamically enabled XSAVE features permission.
https://lore.kernel.org/all/20220126152210.3044876-1-pbonzini@redhat.com/
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Signed-off-by: Jing Liu <jing2.liu@intel.com>
Message-Id: <20220217060434.52460-4-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The extended state subleaves (EAX=0Dh, ECX=n, n>1).ECX[1]
indicate whether the extended state component locates
on the next 64-byte boundary following the preceding state
component when the compacted format of an XSAVE area is
used.
Right now, they are all zero because no supported component
needed the bit to be set, but the upcoming AMX feature will
use it. Fix the subleaves value according to KVM's supported
cpuid.
Signed-off-by: Jing Liu <jing2.liu@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20220217060434.52460-2-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Some cpu properties have to be set only for cpu models in builtin_x86_defs,
registered with x86_register_cpu_model_type, and not for
cpu models "base", "max", and the subclass "host".
These properties are the ones set by function x86_cpu_apply_props,
(also including kvm_default_props, tcg_default_props),
and the "vendor" property for the KVM and HVF accelerators.
After recent refactoring of cpu, which also affected these properties,
they were instead set unconditionally for all x86 cpus.
This has been detected as a bug with Nested on AMD with cpu "host",
as svm was not turned on by default, due to the wrongful setting of
kvm_default_props via x86_cpu_apply_props, which set svm to "off".
Rectify the bug introduced in commit "i386: split cpu accelerators"
and document the functions that are builtin_x86_defs-only.
Signed-off-by: Claudio Fontana <cfontana@suse.de>
Tested-by: Alexander Bulekov <alxndr@bu.edu>
Fixes: f5cc5a5c ("i386: split cpu accelerators from cpu.c,"...)
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/477
Message-Id: <20210723112921.12637-1-cfontana@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Rather than relying on the X86XSaveArea structure definition,
determine the offset of XSAVE state areas using CPUID leaf 0xd where
possible (KVM and HVF).
Signed-off-by: David Edmondson <david.edmondson@oracle.com>
Message-Id: <20210705104632.2902400-8-david.edmondson@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
i386 realizefn code is sensitive to ordering, and recent commits
aimed at refactoring it, splitting accelerator-specific code,
broke assumptions which need to be fixed.
We need to:
* process hyper-v enlightements first, as they assume features
not to be expanded
* only then, expand features
* after expanding features, attempt to check them and modify them in the
accel-specific realizefn code called by cpu_exec_realizefn().
* after the framework has been called via cpu_exec_realizefn,
the code can check for what has or hasn't been set by accel-specific
code, or extend its results, ie:
- check and evenually set code_urev default
- modify cpu->mwait after potentially being set from host CPUID.
- finally check for phys_bits assuming all user and accel-specific
adjustments have already been taken into account.
Fixes: f5cc5a5c ("i386: split cpu accelerators from cpu.c"...)
Fixes: 30565f10 ("cpu: call AccelCPUClass::cpu_realizefn in"...)
Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Claudio Fontana <cfontana@suse.de>
Message-Id: <20210603123001.17843-2-cfontana@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
It's very easy to mistakenly extend kvm_default_props to include
features that require a kernel version that's too recent. Add a
comment warning about that, pointing to the documentation file
where the minimum kernel version for KVM is documented.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200925211021.4158567-1-ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
overall, all devices' realize functions take an Error **errp, but return void.
hw/core/qdev.c code, which realizes devices, therefore does:
local_err = NULL;
dc->realize(dev, &local_err);
if (local_err != NULL) {
goto fail;
}
However, we can improve at least accel_cpu to return a meaningful bool value.
Signed-off-by: Claudio Fontana <cfontana@suse.de>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20210322132800.7470-9-cfontana@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
i386 is the first user of AccelCPUClass, allowing to split
cpu.c into:
cpu.c cpuid and common x86 cpu functionality
host-cpu.c host x86 cpu functions and "host" cpu type
kvm/kvm-cpu.c KVM x86 AccelCPUClass
hvf/hvf-cpu.c HVF x86 AccelCPUClass
tcg/tcg-cpu.c TCG x86 AccelCPUClass
Signed-off-by: Claudio Fontana <cfontana@suse.de>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
[claudio]:
Rebased on commit b8184135 ("target/i386: allow modifying TCG phys-addr-bits")
Signed-off-by: Claudio Fontana <cfontana@suse.de>
Message-Id: <20210322132800.7470-5-cfontana@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
David Woodhouse
Paolo Bonzini
Yang Zhong
Jing Liu
Claudio Fontana
David Edmondson
Eduardo Habkost