Changes v2 -> v3;
- Check for kvm_enabled() before setting cpuid_7_0_ebx_features
Changes v1 -> v2:
- Use kvm_arch_get_supported_cpuid() instead of host_cpuid() on
cpu_x86_fill_host().
We should use GET_SUPPORTED_CPUID for all bits on "-cpu host"
eventually, but I am not changing all the other CPUID leaves because
we may not be able to test such an intrusive change in time for 1.1.
Description of the bug:
Since QEMU 0.15, the CPUID information on CPUID[EAX=7,ECX=0] is being
returned unfiltered to the guest, directly from the GET_SUPPORTED_CPUID
return value.
The problem is that this makes the resulting CPU feature flags
unpredictable and dependent on the host CPU and kernel version. This
breaks live-migration badly if migrating from a host CPU that supports
some features on that CPUID leaf (running a recent kernel) to a kernel
or host CPU that doesn't support it.
Migration also is incorrect (the virtual CPU changes under the guest's
feet) if you migrate in the opposite direction (from an old CPU/kernel
to a new CPU/kernel), but with less serious consequences (guests
normally query CPUID information only once on boot).
Fortunately, the bug affects only users using cpudefs with level >= 7.
The right behavior should be to explicitly enable those features on
[cpudef] config sections or on the "-cpu" command-line arguments. Right
now there is no predefined CPU model on QEMU that has those features:
the latest Intel model we have is Sandy Bridge.
I would like to get this fixed on 1.1, so I am submitting this patch,
that enables those features only if "-cpu host" is being used (as we
don't have any pre-defined CPU model that actually have those features).
After 1.1 is released, we can make those features properly configurable
on [cpudef] and -cpu configuration.
One problem is: with this patch, users with the following setup:
- Running QEMU 1.0;
- Using a cpudef having level >= 7;
- Running a kernel that supports the features on CPUID leaf 7; and
- Running on a CPU that supports some features on CPUID leaf 7
won't be able to live-migrate to QEMU 1.1. But for these users
live-migration is already broken (they can't live-migrate to hosts with
older CPUs or older kernels, already), I don't see how to avoid this
problem.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Commit de024815e3 (target-i386: QOM'ify
CPU init) moved mce_init() call from helper.c:cpu_x86_init() into
X86CPU's cpu.c:x86_cpu_initfn().
mce_init() checks for a family >= 6 though, so we could end up with a
sequence such as for -cpu somecpu,family=6:
x86_cpu_initfn => X86CPU::family == 5
mce_init => no-op
cpu_x86_register => X86CPU::family = 6
=> MCE unexpectedly not init'ed
or for -cpu someothercpu,family=5:
x86_cpu_initfn => X86CPU::family == 6
mce_init => init'ed
cpu_x86_register => X86CPU::family = 5
=> MCE unexpectedly init'ed
Therefore partially revert the above commit. To avoid moving
mce_init() back into helper.c, foresightedly move it into a
new x86_cpu_realize() function and, in lack of ObjectClass::realize,
call it directly from cpu_x86_init().
While at it, move the qemu_init_vcpu() call that used to follow
mce_init() in cpu_x86_init() into the new realizefn as well.
Reported-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Use Hz as unit.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Using it now would incur converting the three x86_def_t vendor words
into a string for object_property_set_str(), then back to three words
in the "vendor" setter.
The built-in CPU definitions use numeric preprocessor defines to
initialize the three words in a charset-safe way, so do not change the
fields to char[12] just to use the setter.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Add the property early in the initfn so that it can be used in helpers
such as mce_init().
Signed-off-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
[AF: Add an error_free(), spotted by Michael Roth]
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
A family field value of 0xf and extended family field value of 0xff is
the maximum representable unsigned family number.
All other CPUID property values are bounds-checked, so add a check here
for symmetry before we adopt it in a property setter.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Avoids an x86_env_get_cpu() call there, to work with QOM properties.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Don't assume zeroed cpuid_model[] fields.
This didn't break anything yet but QOM properties should be able to set
the value to something else without setting an intermediate zero string.
Reviewed-by: Eduardo Habhost <ehabkost@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
[AF: Use memset() instead of for loop, suggested by Igor]
Signed-off-by: Andreas Färber <afaerber@suse.de>
Use uintptr_t instead of void * or unsigned long in
several op related functions, env->mem_io_pc and
GETPC() macro.
Reviewed-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Move code from cpu_x86_init() to new QOM x86_cpu_initfn().
Also move mce_init() to cpu.c since it's used nowhere else.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Embed CPUX86State as first member of X86CPU.
Distinguish between "x86_64-cpu" and "i386-cpu".
Drop cpu_x86_close() in favor of calling object_delete() directly.
For now let CPUClass::reset() call cpu_state_reset().
Signed-off-by: Andreas Färber <afaerber@suse.de>
Scripted conversion:
for file in *.[hc] hw/*.[hc] hw/kvm/*.[hc] linux-user/*.[hc] linux-user/m68k/*.[hc] bsd-user/*.[hc] darwin-user/*.[hc] tcg/*/*.[hc] target-*/cpu.h; do
sed -i "s/CPUState/CPUArchState/g" $file
done
All occurrences of CPUArchState are expected to be replaced by QOM CPUState,
once all targets are QOM'ified and common fields have been extracted.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Scripted conversion:
sed -i "s/CPUState/CPUX86State/g" target-i386/*.[hc]
sed -i "s/#define CPUX86State/#define CPUState/" target-i386/cpu.h
Signed-off-by: Andreas Färber <afaerber@suse.de>
Acked-by: Anthony Liguori <aliguori@us.ibm.com>
Frees the identifier cpu_reset for QOM CPUs (manual rename).
Don't hide the parameter type behind explicit casts, use static
functions with strongly typed argument to indirect.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
This was a long pending bug, now revealed by the assert in
phys_page_find that stumbled over the large page index returned by
cpu_get_phys_page_debug for NX-marked pages: We need to mask out NX and
all user-definable bits 52..62 from PDEs and the final PTE to avoid
corrupting physical addresses.
Reviewed-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
valgrind warns about padding fields which are passed
to vcpu ioctls uninitialized.
This is not an error in practice because kvm ignored padding.
Since the ioctls in question are off data path and
the cost is zero anyway, initialize padding to 0
to suppress these errors.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
* 'upstream' of git://qemu.weilnetz.de/qemu:
Move definition of HOST_LONG_BITS to qemu-common.h
target-xtensa: Clean includes
target-unicore32: Clean includes
target-sh4: Clean includes
target-s390x: Clean includes
target-ppc: Clean includes
target-mips: Clean includes
target-microblaze: Clean includes
target-m68k: Clean includes
target-lm32: Clean includes
target-i386: Clean includes
target-cris: Clean includes
target-arm: Clean includes
target-alpha: Clean includes
Remove macro HOST_LONG_SIZE
* qemu-kvm/uq/master:
pc-bios: update kvmvapic.bin
kvmvapic: Use optionrom helpers
optionsrom: Reserve space for checksum
kvmvapic: Simplify mp/up_set_tpr
kvmvapic: Introduce TPR access optimization for Windows guests
kvmvapic: Add option ROM
target-i386: Add infrastructure for reporting TPR MMIO accesses
Allow to use pause_all_vcpus from VCPU context
Process pending work while waiting for initial kick-off in TCG mode
Remove useless casts from cpu iterators
kvm: Set cpu_single_env only once
kvm: Synchronize cpu state in kvm_arch_stop_on_emulation_error()
Move the logic to transform the 48-char model ID into the 12-word model
value into a helper.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Move the logic for setting the stepping field into a helper function.
To make the function self-contained and to prepare for future
unordered/multiple uses, mask out any previous stepping values first.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Move the logic for setting the model and extended model fields
into a helper function.
To make the function self-contained and to prepare for future
unordered/multiple uses, mask out any previous model values first.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Move the logic for setting the family and extended family into a
helper function.
To make the helper self-contained and in preparation of future
unordered/multiple uses, mask out any previous family values first.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Use 'i64' instead of 'lm' and 'xd' instead of 'nx' on Intel models.
The flags have different names on Intel docs, so use those names for clarity.
This is based on a previous patch from John Cooper where this was introduced
with many other changes at the same time. Original John's patch submission is
at Message-ID: <4DDAD5E7.2020002@redhat.com>, <http://marc.info/?l=qemu-devel&m=130618871926030>.
Changes v1 -> v2:
- Rebase patch against latest Qemu git tree
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
pclmulqdq: /proc/cpuinfo on Linux and all documentation I have seen uses
pclmulqdq as the flag name. As the only document using pclmuldq seems to
be the Intel CPUID documentation (Application Note 485), it looks like a
typo and not the correct name for the flag.
ffxsr: AMD docs refer to fxsr_opt as ffxsr, so allow this named to be
used too.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
This will allow the APIC core to file a TPR access report. Depending on
the accelerator and kernel irqchip mode, it will either be delivered
right away or queued for later reporting.
In TCG mode, we can restart the triggering instruction and can therefore
forward the event directly. KVM does not allows us to restart, so we
postpone the delivery of events recording in the user space APIC until
the current instruction is completed.
Note that KVM without in-kernel irqchip will report the address after
the instruction that triggered the access.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Call to kvm_cpu_synchronize_state() is missing.
kvm_arch_stop_on_emulation_error may look at outdated registers here.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Reviewed-by: Jan Kiszka <jan.kiszka@siemens.com>
To both avoid that kvm_irqchip_in_kernel always has to be paired with
kvm_enabled and that the former ends up in a function call, implement it
like the latter. This means keeping the state in a global variable and
defining kvm_irqchip_in_kernel as a preprocessor macro.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Introduce the KVM-specific machine option kvm_shadow_mem. It allows to
set a custom shadow MMU size for the virtual machine. This is useful for
stress testing e.g.
Only x86 supports this for now, but it is in principle a generic
concept for all targets with shadow MMUs.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This introduces the alternative APIC device which makes use of KVM's
in-kernel device model. External NMI injection via LINT1 is emulated by
checking the current state of the in-kernel APIC, only injecting a NMI
into the VCPU if LINT1 is unmasked and configured to DM_NMI.
MSI is not yet supported, so we disable this when the in-kernel model is
in use.
CC: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Add the basic infrastructure to active in-kernel irqchip support, inject
interrupts into these models, and maintain IRQ routes.
Routing is optional and depends on the host arch supporting
KVM_CAP_IRQ_ROUTING. When it's not available on x86, we looe the HPET as
we can't route GSI0 to IOAPIC pin 2.
In-kernel irqchip support will once be controlled by the machine
property 'kernel_irqchip', but this is not yet wired up.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Commit 2355c16e74 introduced a new ldmxcsr
helper taking an i32 argument, but the helper is actually passed a long.
Fix that by truncating the long to i32.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
SSE rounding and flush to zero control has never been implemented. However
given that softfloat-native was using a single state for FPU and SSE and
given that glibc is setting both FPU and SSE state in fesetround(), this
was working correctly up to the switch to softfloat.
Fix that by adding an update_sse_status() function similar to
update_fpu_status(), and callin git on write to mxcsr.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
The helpers implemented dpps and dppd SSE instructions are not passing
the correct argument types to the softfloat functions. While they do
work anyway providing a correct behaviour, this patch fixes that.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
roundps and roundss SSE2 instructions have been broken when switching
target-i386 to softfloat. They use float64_round_to_int to convert a
float32, and while the implicit conversion from float32 to float64 was
correct for softfloat-native, it is not for pure softfloat. Fix that by
using the correct registers and correct functions.
Also fix roundpd and roundsd implementation at the same time, even if
these functions are behaving correctly.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
