Now that all entries have check_feat=~0 in
kvm_check_features_against_host(), we can eliminate check_feat entirely
and make the code check all bits.
This patch shouldn't introduce any behavior change, as check_feat is set
to ~0 on all entries.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
When nested SVM is supported, the kernel returns the SVM flag on
GET_SUPPORTED_CPUID[1], so we can check the SVM flag safely in
kvm_check_features_against_host().
I don't know why the original code ignored the SVM flag. Maybe it was
because kvm_cpu_fill_host() used the CPUID instruction directly instead
of GET_SUPPORTED_CPUID
[1] Older kernels (before v2.6.37) returned the SVM flag even if nested
SVM was _not_ supported. So the only cases where this patch should
change behavior is when SVM is being requested by the user or the
CPU model, but not supported by the host. And on these cases we
really want QEMU to abort if the "enforce" option is set.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
I have no idea why PPRO_FEATURES was being ignored on the check of the
CPUID.80000001H.EDX bits. I believe it was a mistake, and it was
supposed to be ~(PPRO_FEATURES & CPUID_EXT2_AMD_ALIASES) or just
~CPUID_EXT2_AMD_ALIASES, because some time ago kvm_cpu_fill_host() used
the CPUID instruction directly (instead of
kvm_arch_get_supported_cpuid()).
But now kvm_cpu_fill_host() uses kvm_arch_get_supported_cpuid(), and
kvm_arch_get_supported_cpuid() returns all supported bits for
CPUID.80000001H.EDX, even the AMD aliases (that are explicitly copied
from CPUID.01H.EDX), so we can make the code check/enforce all the
CPUID.80000001H.EDX bits.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
We don't need any hack to ignore CPUID_EXT_HYPERVISOR anymore, because
kvm_arch_get_supported_cpuid() now sets CPUID_EXT_HYPERVISOR properly.
So, this shouldn't introduce any behavior change, but it makes the code
simpler.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
The -cpu check/enforce warnings are printing incorrect information about the
missing flags. There are no feature flags on CPUID leaves 0 and 0x80000000, but
there were references to 0 and 0x80000000 in the table at
kvm_check_features_against_host().
This changes the model_features_t struct to contain the register number as
well, so the error messages print the correct CPUID leaf+register information,
instead of wrong CPUID leaf numbers.
This also changes the format of the error messages, so they follow the
"CPUID.<leaf>.<register>.<name> [bit <offset>]" convention used in Intel
documentation. Example output:
$ qemu-system-x86_64 -machine pc-1.0,accel=kvm -cpu Opteron_G4,+ia64,enforce
warning: host doesn't support requested feature: CPUID.01H:EDX.ia64 [bit 30]
warning: host doesn't support requested feature: CPUID.01H:ECX.xsave [bit 26]
warning: host doesn't support requested feature: CPUID.01H:ECX.avx [bit 28]
warning: host doesn't support requested feature: CPUID.80000001H:ECX.abm [bit 5]
warning: host doesn't support requested feature: CPUID.80000001H:ECX.sse4a [bit 6]
warning: host doesn't support requested feature: CPUID.80000001H:ECX.misalignsse [bit 7]
warning: host doesn't support requested feature: CPUID.80000001H:ECX.3dnowprefetch [bit 8]
warning: host doesn't support requested feature: CPUID.80000001H:ECX.xop [bit 11]
warning: host doesn't support requested feature: CPUID.80000001H:ECX.fma4 [bit 16]
Unable to find x86 CPU definition
$
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
When using -cpu host, we don't need to use the kvm_default_features
variable, as the user is explicitly asking QEMU to enable all feature
supported by the host.
This changes the kvm_cpu_fill_host() code to use GET_SUPPORTED_CPUID to
initialize the kvm_features field, so we get all host KVM features
enabled.
This will also allow us to properly check/enforce KVM features inside
kvm_check_features_against_host() later. For example, we will be able to
make this:
$ qemu-system-x86_64 -cpu ...,+kvm_pv_eoi,enforce
refuse to start if kvm_pv_eoi is not supported by the host (after we fix
kvm_check_features_against_host() to check KVM flags as well).
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
The existing -cpu host code simply sets every bit inside svm_features
(initializing it to -1), and that makes it impossible to make the
enforce/check options work properly when the user asks for SVM features
explicitly in the command-line.
So, instead of initializing svm_features to -1, use GET_SUPPORTED_CPUID
to fill only the bits that are supported by the host (just like we do
for all other CPUID feature words inside kvm_cpu_fill_host()).
This will keep the existing behavior (as filter_features_for_kvm()
already uses GET_SUPPORTED_CPUID to filter svm_features), but will allow
us to properly check for KVM features inside
kvm_check_features_against_host() later.
For example, we will be able to make this:
$ qemu-system-x86_64 -cpu ...,+pfthreshold,enforce
refuse to start if the SVM "pfthreshold" feature is not supported by the
host (after we fix kvm_check_features_against_host() to check SVM flags
as well).
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
This finally makes the CPU class a subclass of the Device class,
allowing us to start using DeviceState properties on CPU subclasses.
It has no_user=1, as creating CPUs using -device doesn't work yet.
Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Commit 667d22d1ae (qdev: move bus removal
to object_unparent) made the assumption that at unparenting time
parent_bus is not NULL. This assumption is unjustified since
object_unparent() may well be called directly after object_initialize(),
without any qdev_set_parent_bus().
This did not cause any issues yet because qdev_[try_]create() does call
qdev_set_parent_bus(), falling back to SysBus if unsupplied.
While at it, ensure that this new function uses the device_ prefix and
make the name more neutral in light of this semantic change.
Reported-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Tested-by: Igor Mammedov <imammedo@redhat.com>
The code depends on some functions from qemu-option.o, so add
qemu-option.o to universal-obj-y to make sure it's included.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
The stub will be used on cases where sysbus.c is not compiled in (e.g.
*-user).
Note that code that uses NULL as the bus with qdev{_try,}_create()
implicitly uses sysbus_get_default() as the bus, and will still require
sysbus.c to be compiled in.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Add vmstate stub functions, so that qdev.o can be used without savevm.o
when vmstate support is not necessary (i.e. by *-user).
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
This will be useful for code that don't call qemu_devices_reset() (e.g.
*-user). If qemu_devices_reset() is never called, it means we don't need
to keep track of the reset handler list.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Due to disagreement on a name that is generic enough for hw/pci/pci.h,
the symbolic constants are placed in the .c files.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Some devices were missing, and we're using two PCI vendor ids.
This patch only adds devices that are already documented in hw/pci/pci.h.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
* kraxel/usb.75: (32 commits)
uhci: stop using portio lists
usbredir: Add support for buffered bulk input (v2)
exynos4210: Add EHCI support
usb/ehci: Add SysBus EHCI device for Exynos4210
usb/ehci: Move capsbase and opregbase into SysBus EHCI class
usb/ehci: Clean up SysBus and PCI EHCI split
xhci: call set-address with dummy usbpacket
usb-redir: Add debugging to bufpq save / restore
usbredir: Add usbredir_init_endpoints() helper
usbredir: Verify we have 32 bits bulk length cap when redirecting to xhci
usbredir: Add ep_stopped USBDevice method
usbredir: Add USBEP2I and I2USBEP helper macros
usbredir: Add an usbredir_stop_ep helper function
usb: Add an usb_device_ep_stopped USBDevice method
usb: Fix usb_ep_find_packet_by_id
hid: Change idle handling to use a timer
uhci: Maximize how many frames we catch up when behind
uhci: Limit amount of frames processed in one go
uhci: Add a QH_VALID define
uhci: Fix pending interrupts getting lost on migration
...
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
* stefanha/net:
rtl8139: preserve link state across device reset
e1000: no need auto-negotiation if link was down
net: clean up network at qemu process termination
e1000: Discard oversized packets based on SBP|LPE
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Helper function for dpa_w_ph, dpax_w_ph, dps_w_ph and dpsx_w_ph incorrectly
defines halfword vector elements as unsigned values. This results in wrong
output which is not triggered in the tests as they also follow this logic.
Signed-off-by: Petar Jovanovic <petarj@mips.com>
Reviewed-by: Eric Johnson <ericj@mips.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
The macros RESTORE_ROUNDING_MODE and RESTORE_FLUSH_MODE silently used
variable env from their callers. Using inline functions with env passed
as a function argument is more transparent.
This modification was proposed by Peter Maydell.
Cc: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Reviewed-by: Eric Johnson <ericj@mips.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Clear the DSP hflags at the start of compute_hflags. Otherwise access
is not properly disabled once enabled.
Signed-off-by: Eric Johnson <ericj@mips.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
MIPS only supports 31 bits of virtual address space for user space, so let's
make sure we stay within that limit with our preallocated memory block.
This fixes the MIPS user space targets when executed without command line
option.
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Buffered bulk mode is intended for bulk *input* endpoints, where the data is
of a streaming nature (not part of a command-response protocol). These
endpoints' input buffer may overflow if data is not read quickly enough.
So in buffered bulk mode the usb-host takes care of the submitting and
re-submitting of bulk transfers.
Buffered bulk mode is necessary for reliable operation with the bulk in
endpoints of usb to serial convertors. Unfortunatelty buffered bulk input
mode will only work with certain devices, therefor this patch also adds a
usb-id table to enable it for devices which need it, while leaving the
bulk ep handling for other devices unmodified.
Note that the bumping of the required usbredir from 0.5.3 to 0.6 does
not mean that we will now need a newer usbredir release then qemu-1.3,
.pc files reporting 0.5.3 have only ever existed in usbredir builds directly
from git, so qemu-1.3 needs the 0.6 release too.
Changes in v2:
-Split of quirk handling into quirks.c
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
memcpy() for overlapping regions is undefined behavior; use memmove()
instead in readline_hist_add().
[Keep tab characters since surrounding code still uses them -- Stefan]
Signed-off-by: Nickolai Zeldovich <nickolai@csail.mit.edu>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Support backend guest notifier masking in vhost-net:
create eventfd at device init, when masked,
make vhost use that as eventfd instead of
sending an interrupt.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This makes it possible to use started flag for sanity checking
of callbacks that happen during start/stop.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
As vhost started is cleared last thing on stop,
set it first things on start. This makes it
possible to use vhost_started while start is in
progress which is used by follow-up patches.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
some backends (notably vhost) can mask events
at their source in a way that is more efficient
than masking through kvm.
Specifically
- masking in kvm uses rcu write side so it has high latency
- in kvm on unmask we always send an interrupt
masking at source does not have these issues.
Add such support in virtio.h and use in virtio-pci.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Some guests mask a vector then unmask without changing it.
Store vectors to avoid kvm system calls in this case.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Pass nvqs to set_guest_notifiers. This makes it possible to
save on irqfds by not allocating one for the control vq
for virtio-net.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Peter Maydell recommended the change to be more proper. The result was tested
and shows coming up with the same proper value.
Signed-off-by: Samuel Seay <LightningTH@GMail.com>
[agraf: change subject]
Signed-off-by: Alexander Graf <agraf@suse.de>
Previously we silently exited, with subclasses we got an opcode warning.
Instead, explicitly tell the user what's wrong.
An indication for this is -cpu ? showing "host" with an all-zero PVR.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Alexander Graf <agraf@suse.de>
Since the model list is highly macrofied, keep ppc_def_t for now and
save a pointer to it in PowerPCCPUClass. This results in a flat list of
subclasses including aliases, to be refined later.
Move cpu_ppc_init() to translate_init.c and drop helper.c.
Long-term the idea is to turn translate_init.c into a standalone cpu.c.
Inline cpu_ppc_usable() into type registration.
Split cpu_ppc_register() in two by code movement into the initfn and
by turning the remaining part into a realizefn.
Move qemu_init_vcpu() call into the new realizefn and adapt
create_ppc_opcodes() to return an Error.
Change ppc_find_by_pvr() -> ppc_cpu_class_by_pvr().
Change ppc_find_by_name() -> ppc_cpu_class_by_name().
Turn -cpu host into its own subclass. This requires to move the
kvm_enabled() check in ppc_cpu_class_by_name() to avoid the class being
found via the normal name lookup in the !kvm_enabled() case.
Turn kvmppc_host_cpu_def() into the class_init and add an initfn that
asserts KVM is in fact enabled.
Implement -cpu ? and the QMP equivalent in terms of subclasses.
This newly exposes -cpu host to the user, ordered last for -cpu ?.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Alexander Graf <agraf@suse.de>
We already used to support the external proxy facility of FSL MPICs,
but only implemented it halfway correctly.
This patch adds support for
* dynamic enablement of the EPR facility
* interrupt acknowledgement only when the interrupt is delivered
This way the implementation now is closer to real hardware.
Signed-off-by: Alexander Graf <agraf@suse.de>
On e500mc, the platform doesn't provide a way for the CPU to go idle.
To still not uselessly burn CPU time, expose an idle hypercall to the guest
if kvm supports it.
Signed-off-by: Stuart Yoder <stuart.yoder@freescale.com>
[agraf: adjust for current code base, add patch description, fix non-kvm case]
Signed-off-by: Alexander Graf <agraf@suse.de>
Properly implement level-triggered interrupts by withdrawing an
interrupt from the raised queue if the interrupt source de-asserts.
Also withdraw from the raised queue if the interrupt becomes masked.
When CTPR is written, check whether we need to raise or lower the
interrupt output.
Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Besides making the code cleaner, we will need a separate way to access
IACK in order to implement EPR (external proxy) interrupt delivery.
Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Search the queue more efficiently by first looking for a non-zero word,
and then using the common bit-searching function to find the bit within
the word. It would be even nicer if bitops_ffsl() could be hooked up
to the compiler intrinsic so that bit-searching instructions could be
used, but that's another matter.
Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Previously, the sense and priority bits were masked off when writing
to IVPR, and all interrupts were treated as edge-triggered (despite
the existence of code for handling level-triggered interrupts).
Polarity is implemented only as storage. We don't simulate the
bad effects that you'd get on real hardware if you set this incorrectly,
but at least the guest sees the right thing when it reads back the register.
Sense now controls level/edge on FSL external interrupts (and all
interrupts on non-FSL MPIC). FSL internal interrupts do not have a sense
bit (reads as zero), but are level. FSL timers and IPIs do not have
sense or polarity bits (read as zero), and are edge-triggered. To
accommodate FSL internal interrupts, QEMU's internal notion of whether an
interrupt is level-triggered is separated from the IVPR bit.
Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
The two checks with abort() guard against potential QEMU-internal
problems, but the EOI check stops the guest from causing updates to queue
position -1 and other havoc if it writes EOI with no interrupt in
service.
Signed-off-by: Scott Wood <scottwood@freescale.com>
[agraf: remove hunk in code that didn't get applied yet]
Signed-off-by: Alexander Graf <agraf@suse.de>
Besides the private implementation being redundant, namespace collisions
prevented the use of other things in bitops.h.
Serialization does get a bit more awkward, unfortunately, since the
standard bitmap operations are "unsigned long" rather than "uint32_t",
though in exchange we will get faster queue lookups on 64-bit hosts once
we search a word at a time.
Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
This reverts commit a9bd83f4c65de0058659ede009fa1a241f379edd.
This counting approach is not robust against setting a bit that
was already set, or clearing a bit that was already clear. Perhaps
that is considered a bug, but besides the lack of any documentation
for that restriction, it's a pretty unpleasant way for the problem
to manifest itself.
It could be made more robust by testing the current value of the
bit before changing the count, but a later patch speeds up IRQ_check
in all cases, not just when there's nothing pending. Hopefully that
should be adequate to address performance concerns.
Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Previously the code relied on the queue's "next" field getting
set to -1 sometime between an update to the bitmap, and the next
call to IRQ_get_next. Sometimes this happened after the update.
Sometimes it happened before the check. Sometimes it didn't happen
at all.
Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Other priorities are signed, so avoid comparisons between
signed and unsigned.
Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
