QEMU uses a fixed page size for the CPU TLB. If the guest uses large
pages then we effectively split these into multiple smaller pages, and
populate the corresponding TLB entries on demand.
When the guest invalidates the TLB by virtual address we must invalidate
all entries covered by the large page. However the address used to
invalidate the entry may not be present in the QEMU TLB, so we do not
know which regions to clear.
Implementing a full vaiable size TLB is hard and slow, so just keep a
simple address/mask pair to record which addresses may have been mapped by
large pages. If the guest invalidates this region then flush the
whole TLB.
Signed-off-by: Paul Brook <paul@codesourcery.com>
Enable all features of real CPU, unsupported features will be
trimmed depending on TCG or KVM capabilities.
Move the list of unsupported TCG features near the TCG capabilities
masks.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Bump up the xlevel number for qemu32 to allow parsing of the processor
name string for this model.
Similiarly the 486 processor should have at least the feature bit
leaf enabled.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Since 64-bit capability is just another CPUID bit we now properly
mask, there is no reason anymore to hide the 64-bit capable CPU
models from a 32-bit only QEMU. All 64-bit CPUs can be used
perfectly in 32-bit legacy mode anyway, so these models also make
sense for 32-bit.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
In KVM we trim the user provided CPUID bits to match the host CPU's
one. Introduce a similar feature to QEMU/TCG. Create a mask of TCG's
capabilities and apply it to the user bits.
This allows to let the CPU models reflect their native archetypes.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Correct me if I am wrong, but kvm_trim looks like a really bloated
implementation of a bitwise AND. So remove this function and replace
it with the real stuff(TM).
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Some CPUID feature flags had no string value, so they could not be
switched on or off from the command line.
Add names for the missing ones mentioned in the current public CPUID
specification from both Intel and AMD. Those only mentioned in the
Linux kernel source I put as comments.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
the host_cpuid function was located at the end of the file and had
a prototype before it's first use. Move it up and remove the
prototype.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
This one was accidently removed with commit
bb0300dc57
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
CPUID leaf Fn8000_0001.EDX contains a copy of many Fn0000_0001.EDX bits.
Define a name for this mask to improve readability and avoid typos.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
about half of target-i386/helper.c consist of CPUID related functions.
Only one of them is a real TCG helper function. So move the whole
CPUID stuff out of this into a separate file to get better
maintainable parts.
This is only code reordering and should not affect QEMU's
functionality.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
The commit c22549204a led movntps &
movntdq to be translated incorrectly.
Signed-off-by: TeLeMan <geleman@gmail.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Removes a set of ifdefs from exec.c.
Introduce TARGET_VIRT_ADDR_SPACE_BITS for all targets other
than Alpha. This will be used for page_find_alloc, which is
supposed to be using virtual addresses in the first place.
Signed-off-by: Richard Henderson <rth@twiddle.net>
A SIB byte with an index of 4 means "no scaled index", even if the scale
value is not 0. In 64-bit mode, if REX.X is used, an index of 4 selects
%r12. This is correctly handled by the computation of the index variable,
which includes the index bits, and also the REX.X prefix:
index = ((code >> 3) & 7) | REX_X(s);
Thanks to Avi Kivity, Jamie Lokier and Malc for the analysis of the
problem and the initial patch.
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Do not write nmi_pending, sipi_vector, and mpstate unless we at least go
through a reset. And TSC as well as KVM wallclocks should only be
written on full sync, otherwise we risk to drop some time on state
read-modify-write.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This grand cleanup drops all reset and vmsave/load related
synchronization points in favor of four(!) generic hooks:
- cpu_synchronize_all_states in qemu_savevm_state_complete
(initial sync from kernel before vmsave)
- cpu_synchronize_all_post_init in qemu_loadvm_state
(writeback after vmload)
- cpu_synchronize_all_post_init in main after machine init
- cpu_synchronize_all_post_reset in qemu_system_reset
(writeback after system reset)
These writeback points + the existing one of VCPU exec after
cpu_synchronize_state map on three levels of writeback:
- KVM_PUT_RUNTIME_STATE (during runtime, other VCPUs continue to run)
- KVM_PUT_RESET_STATE (on synchronous system reset, all VCPUs stopped)
- KVM_PUT_FULL_STATE (on init or vmload, all VCPUs stopped as well)
This level is passed to the arch-specific VCPU state writing function
that will decide which concrete substates need to be written. That way,
no writer of load, save or reset functions that interact with in-kernel
KVM states will ever have to worry about synchronization again. That
also means that a lot of reasons for races, segfaults and deadlocks are
eliminated.
cpu_synchronize_state remains untouched, just as Anthony suggested. We
continue to need it before reading or writing of VCPU states that are
also tracked by in-kernel KVM subsystems.
Consequently, this patch removes many cpu_synchronize_state calls that
are now redundant, just like remaining explicit register syncs.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
So far we synchronized any dirty VCPU state back into the kernel before
updating the guest debug state. This was a tribute to a deficite in x86
kernels before 2.6.33. But as this is an arch-dependent issue, it is
better handle in the x86 part of KVM and remove the writeback point for
generic code. This also avoids overwriting the flushed state later on if
user space decides to change some more registers before resuming the
guest.
We furthermore need to reinject guest exceptions via the appropriate
mechanism. That is KVM_SET_GUEST_DEBUG for older kernels and
KVM_SET_VCPU_EVENTS for recent ones. Using both mechanisms at the same
time will cause state corruptions.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
If you make use of hw breakpoints on a 32bit x86 linux host, qemu
will segmentation fault when processing the exception.
The problem is that the value of env is stored in $ebp in the op_helper
raise_exception() function, and it can have the wrong value when
calling it from non generated code.
It is possible to work around the problem by restoring the value of
env before calling raise_exception() using a new helper function that
takes (CPUState *) as one of the arguments.
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
helper.o(.text+0x11e0): In function `listflags':
/src/qemu/target-i386/helper.c:661: warning: sprintf() is often misused, please use snprintf()
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
This is a reimplementation of prior versions which adds
the ability to define cpu models for contemporary processors.
The added models are likewise selected via -cpu <name>,
and are intended to displace the existing convention
of "-cpu qemu64" augmented with a series of feature flags.
A primary motivation was determination of a least common
denominator within a given processor class to simplify guest
migration. It is still possible to modify an arbitrary model
via additional feature flags however the goal here was to
make doing so unnecessary in typical usage. The other
consideration was providing models names reflective of
current processors. Both AMD and Intel have reviewed the
models in terms of balancing generality of migration vs.
excessive feature downgrade relative to released silicon.
This version of the patch replaces the prior hard wired
definitions with a configuration file approach for new
models. Existing models are thus far left as-is but may
easily be transitioned to (or may be overridden by) the
configuration file representation.
Proposed new model definitions are provided here for current
AMD and Intel processors. Each model consists of a name
used to select it on the command line (-cpu <name>), and a
model_id which corresponds to a least common denominator
commercial instance of the processor class.
A table of names/model_ids may be queried via "-cpu ?model":
:
x86 Opteron_G3 AMD Opteron 23xx (Gen 3 Class Opteron)
x86 Opteron_G2 AMD Opteron 22xx (Gen 2 Class Opteron)
x86 Opteron_G1 AMD Opteron 240 (Gen 1 Class Opteron)
x86 Nehalem Intel Core i7 9xx (Nehalem Class Core i7)
x86 Penryn Intel Core 2 Duo P9xxx (Penryn Class Core 2)
x86 Conroe Intel Celeron_4x0 (Conroe/Merom Class Core 2)
:
Also added is "-cpu ?dump" which exhaustively outputs all config
data for all defined models, and "-cpu ?cpuid" which enumerates
all qemu recognized CPUID feature flags.
The pseudo cpuid flag 'check' when added to the feature flag list
will warn when feature flags (either implicit in a cpu model or
explicit on the command line) would have otherwise been quietly
unavailable to a guest:
# qemu-system-x86_64 ... -cpu Nehalem,check
warning: host cpuid 0000_0001 lacks requested flag 'sse4.2|sse4_2' [0x00100000]
warning: host cpuid 0000_0001 lacks requested flag 'popcnt' [0x00800000]
A similar 'enforce' pseudo flag exists which in addition
to the above causes qemu to error exit if requested flags are
unavailable.
Configuration data for a cpu model resides in the target config
file which by default will be installed as:
/usr/local/etc/qemu/target-<arch>.conf
The format of this file should be self explanatory given the
definitions for the above six models and essentially mimics
the structure of the static x86_def_t x86_defs.
Encoding of cpuid flags names now allows aliases for both the
configuration file and the command line which reconciles some
Intel/AMD/Linux/Qemu naming differences.
This patch was tested relative to qemu.git.
Signed-off-by: john cooper <john.cooper@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Remove all references to KVM_CR3_CACHE as it was never implemented.
Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Hi,
Kevin and I have agreed on the approach for this one now. So here is
the latest version of the patch for QEMU, submitting e820 reservation
entries via fw_cfg.
Cheers,
Jes
Use qemu-cfg to provide the BIOS with an optional table of e820 entries.
Notify the BIOS of the location of the TSS+EPT range to by reserving
it via the e820 table.
This matches a corresponding patch for Seabios, however older versions
of Seabios will default to the hardcoded address range and stay
compatible with current QEMU.
Signed-off-by: Jes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
From qemu-kvm: Kernels before 2.6.30 misreported some essential CPU
features via KVM_GET_SUPPORTED_CPUID. Fix them up.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
The final version of VCPU events in 2.6.33 will allow to skip
nmi_pending and sipi_vector on KVM_SET_VCPU_EVENTS. For now let's write
them unconditionally, which is unproblematic for upstream due to missing
SMP support. Future version which enable SMP will write them only on
reset.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
These are unused since edea5f0 (no need to define global registers in
cpu-exec.c, 2008-05-10).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Initialize KVM paravirt cpuid leaf and allow user to control guest
visible PV features through -cpu flag.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Without this qemu can even start on kvm modules with events support
since default value of exception_injected in zero and this is #DE
exception.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Now, if we inject a fatal MCE into guest OS, for example Linux, Linux
will go panic and then reboot. But if we inject another MCE now,
system will reset directly instead of go panic firstly, because
MCG_STATUS.MCIP is set to 1 and not cleared after reboot. This is does
not follow the behavior in real hardware.
This patch fixes this via set env->mcg_status to 0 during system reset.
Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Similarly to what is done in 32938e127f
for "jmp im", trunc the immediate to 32-bit when not running in 64-bit
mode.
Reported-by: Kevin O'Connor <kevin@koconnor.net>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
This reverts commit ebbc8a3d8e76d0402f8a08c10c0f32e24715d41d.
As suggested by Jan Kiszka,
"It was obsoleted by d1793b836f8f123b961c613de1bb1c0c185c84cc and now
saves/restores a useless field."
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
hw_breakpoint_type and hw_breakpoint_len used the wrong index multiplier
to extract type and len.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Marcelo correctly remarked that there are usage conflicts between QEMU
core code and KVM /wrt exception_index. So spend a separate field and
also save/restore it properly.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
The CPUID features QEMU presented to the guest were not up-to-date
with QEMU's emulated feature set.
Add the missing bits of recent (and not so recent) additions to
QEMU's emulation engine.
For stability reasons only the user mode usable bits are exposed for
now, features like Monitor or CR8LEG are left out.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Currently, the msrs involved in setting up pvclock are not saved over
migration and/or save/restore. This patch puts their value in special
fields in our CPUState, and deal with them using vmstate.
kvm also has to account for it, by including them in the msr list
for the ioctls.
This is a backport from qemu-kvm.git
[v2: sucessfully build without kerneldir ]
Signed-off-by: Glauber Costa <glommer@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
As KVM now makes use of exception_index to keep pending exceptions, we
have to save&restore this field as well.
NOTE: We have to nail the arch-independent exception_index down to a
certain bit width for proper vmstate processing, namely to 32 bit.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
The multicore CPUID code detects whether the guest is an Intel or an
AMD CPU, because the Linux kernel is picky about the CmpLegacy bit.
KVM by default passes through the host's vendor, which was not
catched by the code. So fork out the vendor determining bits into a
separate function to be used from both places and always get the real
vendor.
This fixes KVM's multicore setup on Intel CPUs.
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Reported-by: Dietmar Maurer <dietmar@proxmox.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
KVM_GET_MSR_INDEX_LIST returns -E2BIG when the provided space is too
small for all MSRs. But this is precisely the error we trigger with the
initial request in order to obtain that size. Do not fail in that case.
This caused a subtle corruption of the guest state as MSR_STAR was not
properly saved/restored. The corruption became visible with latest kvm
optimizing the MSR updates.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Paul Brook