11 years agoKVM: add ioeventfd support
Gregory Haskins [Tue, 7 Jul 2009 21:08:49 +0000]
KVM: add ioeventfd support

ioeventfd is a mechanism to register PIO/MMIO regions to trigger an eventfd
signal when written to by a guest.  Host userspace can register any
arbitrary IO address with a corresponding eventfd and then pass the eventfd
to a specific end-point of interest for handling.

Normal IO requires a blocking round-trip since the operation may cause
side-effects in the emulated model or may return data to the caller.
Therefore, an IO in KVM traps from the guest to the host, causes a VMX/SVM
"heavy-weight" exit back to userspace, and is ultimately serviced by qemu's
device model synchronously before returning control back to the vcpu.

However, there is a subclass of IO which acts purely as a trigger for
other IO (such as to kick off an out-of-band DMA request, etc).  For these
patterns, the synchronous call is particularly expensive since we really
only want to simply get our notification transmitted asychronously and
return as quickly as possible.  All the sychronous infrastructure to ensure
proper data-dependencies are met in the normal IO case are just unecessary
overhead for signalling.  This adds additional computational load on the
system, as well as latency to the signalling path.

Therefore, we provide a mechanism for registration of an in-kernel trigger
point that allows the VCPU to only require a very brief, lightweight
exit just long enough to signal an eventfd.  This also means that any
clients compatible with the eventfd interface (which includes userspace
and kernelspace equally well) can now register to be notified. The end
result should be a more flexible and higher performance notification API
for the backend KVM hypervisor and perhipheral components.

To test this theory, we built a test-harness called "doorbell".  This
module has a function called "doorbell_ring()" which simply increments a
counter for each time the doorbell is signaled.  It supports signalling
from either an eventfd, or an ioctl().

We then wired up two paths to the doorbell: One via QEMU via a registered
io region and through the doorbell ioctl().  The other is direct via
ioeventfd.

You can download this test harness here:

ftp://ftp.novell.com/dev/ghaskins/doorbell.tar.bz2

The measured results are as follows:

qemu-mmio:       110000 iops, 9.09us rtt
ioeventfd-mmio: 200100 iops, 5.00us rtt
ioeventfd-pio:  367300 iops, 2.72us rtt

I didn't measure qemu-pio, because I have to figure out how to register a
PIO region with qemu's device model, and I got lazy.  However, for now we
can extrapolate based on the data from the NULLIO runs of +2.56us for MMIO,
and -350ns for HC, we get:

qemu-pio:      153139 iops, 6.53us rtt
ioeventfd-hc: 412585 iops, 2.37us rtt

these are just for fun, for now, until I can gather more data.

Here is a graph for your convenience:

http://developer.novell.com/wiki/images/7/76/Iofd-chart.png

The conclusion to draw is that we save about 4us by skipping the userspace
hop.

--------------------

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: make io_bus interface more robust
Gregory Haskins [Tue, 7 Jul 2009 21:08:44 +0000]
KVM: make io_bus interface more robust

Today kvm_io_bus_regsiter_dev() returns void and will internally BUG_ON
if it fails.  We want to create dynamic MMIO/PIO entries driven from
userspace later in the series, so we need to enhance the code to be more
robust with the following changes:

   1) Add a return value to the registration function
   2) Fix up all the callsites to check the return code, handle any
      failures, and percolate the error up to the caller.
   3) Add an unregister function that collapses holes in the array

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: add module parameters documentation
Andre Przywara [Fri, 10 Jul 2009 12:20:35 +0000]
KVM: add module parameters documentation

Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: PIT support for HPET legacy mode
Beth Kon [Tue, 7 Jul 2009 15:50:38 +0000]
KVM: PIT support for HPET legacy mode

When kvm is in hpet_legacy_mode, the hpet is providing the timer
interrupt and the pit should not be. So in legacy mode, the pit timer
is destroyed, but the *state* of the pit is maintained. So if kvm or
the guest tries to modify the state of the pit, this modification is
accepted, *except* that the timer isn't actually started. When we exit
hpet_legacy_mode, the current state of the pit (which is up to date
since we've been accepting modifications) is used to restart the pit
timer.

The saved_mode code in kvm_pit_load_count temporarily changes mode to
0xff in order to destroy the timer, but then restores the actual
value, again maintaining "current" state of the pit for possible later
reenablement.

[avi: add some reserved storage in the ioctl; make SET_PIT2 IOW]
[marcelo: fix memory corruption due to reserved storage]

Signed-off-by: Beth Kon <eak@us.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Always report x2apic as supported feature
Gleb Natapov [Sun, 12 Jul 2009 13:10:55 +0000]
KVM: Always report x2apic as supported feature

We emulate x2apic in software, so host support is not required.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: No need to kick cpu if not in a guest mode
Gleb Natapov [Tue, 7 Jul 2009 12:27:32 +0000]
KVM: No need to kick cpu if not in a guest mode

This will save a couple of IPIs.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Acked-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Add trace points in irqchip code
Gleb Natapov [Tue, 7 Jul 2009 13:00:57 +0000]
KVM: Add trace points in irqchip code

Add tracepoint in msi/ioapic/pic set_irq() functions,
in IPI sending and in the point where IRQ is placed into
apic's IRR.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: ignore msi request if !level
Michael S. Tsirkin [Sun, 5 Jul 2009 18:27:58 +0000]
KVM: ignore msi request if !level

Irqfd sets level for interrupt to 1 and then to 0.
For MSI, check level so that a single message is sent.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: fix MMIO_CONF_BASE MSR access
Andre Przywara [Thu, 2 Jul 2009 13:04:14 +0000]
KVM: fix MMIO_CONF_BASE MSR access

Some Windows versions check whether the BIOS has setup MMI/O for
config space accesses on AMD Fam10h CPUs, we say "no" by returning 0 on
reads and only allow disabling of MMI/O CfgSpace setup by igoring "0" writes.

Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Trace shadow page lifecycle
Avi Kivity [Mon, 6 Jul 2009 12:58:14 +0000]
KVM: Trace shadow page lifecycle

Create, sync, unsync, zap.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Document basic API
Avi Kivity [Tue, 9 Jun 2009 09:37:58 +0000]
KVM: Document basic API

Document the basic API corresponding to the 2.6.22 release.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: MMU: Trace guest pagetable walker
Avi Kivity [Mon, 6 Jul 2009 09:21:32 +0000]
KVM: MMU: Trace guest pagetable walker

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoRevert "KVM: x86: check for cr3 validity in ioctl_set_sregs"
Jan Kiszka [Wed, 1 Jul 2009 18:52:03 +0000]
Revert "KVM: x86: check for cr3 validity in ioctl_set_sregs"

This reverts commit 6c20e1442bb1c62914bb85b7f4a38973d2a423ba.

To my understanding, it became obsolete with the advent of the more
robust check in mmu_alloc_roots (89da4ff17f). Moreover, it prevents
the conceptually safe pattern

 1. set sregs
 2. register mem-slots
 3. run vcpu

by setting a sticky triple fault during step 1.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: handle AMD microcode MSR
Andre Przywara [Fri, 3 Jul 2009 14:00:14 +0000]
KVM: handle AMD microcode MSR

Windows 7 tries to update the CPU's microcode on some processors,
so we ignore the MSR write here. The patchlevel register is already handled
(returning 0), because the MSR number is the same as Intel's.

Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Fix apic_mmio_write return for unaligned write
Sheng Yang [Mon, 6 Jul 2009 03:05:39 +0000]
KVM: Fix apic_mmio_write return for unaligned write

Some in-famous OS do unaligned writing for APIC MMIO, and the return value
has been missed in recent change, then the OS hangs.

Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Use temporary variable to shorten lines.
Gleb Natapov [Sun, 5 Jul 2009 15:48:12 +0000]
KVM: Use temporary variable to shorten lines.

Cosmetic only. No logic is changed by this patch.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: x2apic interface to lapic
Gleb Natapov [Sun, 5 Jul 2009 14:39:36 +0000]
KVM: x2apic interface to lapic

This patch implements MSR interface to local apic as defines by x2apic
Intel specification.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Add Directed EOI support to APIC emulation
Gleb Natapov [Sun, 5 Jul 2009 14:39:35 +0000]
KVM: Add Directed EOI support to APIC emulation

Directed EOI is specified by x2APIC, but is available even when lapic is
in xAPIC mode.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Trace apic registers using their symbolic names
Avi Kivity [Wed, 1 Jul 2009 15:25:44 +0000]
KVM: Trace apic registers using their symbolic names

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Trace mmio
Avi Kivity [Wed, 1 Jul 2009 13:01:02 +0000]
KVM: Trace mmio

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Ignore PCI ECS I/O enablement
Andre Przywara [Wed, 24 Jun 2009 13:37:05 +0000]
KVM: Ignore PCI ECS I/O enablement

Linux guests will try to enable access to the extended PCI config space
via the I/O ports 0xCF8/0xCFC on AMD Fam10h CPU. Since we (currently?)
don't use ECS, simply ignore write and read attempts.

Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Trace irq level and source id
Avi Kivity [Wed, 1 Jul 2009 09:09:41 +0000]
KVM: Trace irq level and source id

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: fix lock imbalance
Jiri Slaby [Mon, 29 Jun 2009 16:05:10 +0000]
KVM: fix lock imbalance

There is a missing unlock on one fail path in ioapic_mmio_write,
fix that.

Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: document lock nesting rule
Michael S. Tsirkin [Mon, 29 Jun 2009 19:24:45 +0000]
KVM: document lock nesting rule

Document kvm->lock nesting within kvm->slots_lock

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: remove in_range from io devices
Michael S. Tsirkin [Mon, 29 Jun 2009 19:24:32 +0000]
KVM: remove in_range from io devices

This changes bus accesses to use high-level kvm_io_bus_read/kvm_io_bus_write
functions. in_range now becomes unused so it is removed from device ops in
favor of read/write callbacks performing range checks internally.

This allows aliasing (mostly for in-kernel virtio), as well as better error
handling by making it possible to pass errors up to userspace.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: convert bus to slots_lock
Michael S. Tsirkin [Mon, 29 Jun 2009 19:24:26 +0000]
KVM: convert bus to slots_lock

Use slots_lock to protect device list on the bus.  slots_lock is already
taken for read everywhere, so we only need to take it for write when
registering devices.  This is in preparation to removing in_range and
kvm->lock around it.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: switch pit creation to slots_lock
Michael S. Tsirkin [Mon, 29 Jun 2009 19:24:21 +0000]
KVM: switch pit creation to slots_lock

switch pit creation to slots_lock. slots_lock is already taken for read
everywhere, so we only need to take it for write when creating pit.
This is in preparation to removing in_range and kvm->lock around it.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: switch coalesced mmio changes to slots_lock
Michael S. Tsirkin [Mon, 29 Jun 2009 19:24:14 +0000]
KVM: switch coalesced mmio changes to slots_lock

switch coalesced mmio slots_lock.  slots_lock is already taken for read
everywhere, so we only need to take it for write when changing zones.
This is in preparation to removing in_range and kvm->lock around it.

[avi: fix build]

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: document locking for kvm_io_device_ops
Michael S. Tsirkin [Mon, 29 Jun 2009 19:24:07 +0000]
KVM: document locking for kvm_io_device_ops

slots_lock is taken everywhere when device ops are called.
Document this as we will use this to rework locking for io.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: use vcpu_id instead of bsp_vcpu pointer in kvm_vcpu_is_bsp
Marcelo Tosatti [Wed, 17 Jun 2009 13:07:59 +0000]
KVM: use vcpu_id instead of bsp_vcpu pointer in kvm_vcpu_is_bsp

Change kvm_vcpu_is_bsp to use vcpu_id instead of bsp_vcpu pointer, which
is only initialized at the end of kvm_vm_ioctl_create_vcpu.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: remove old KVMTRACE support code
Marcelo Tosatti [Thu, 18 Jun 2009 14:47:28 +0000]
KVM: remove old KVMTRACE support code

Return EOPNOTSUPP for KVM_TRACE_ENABLE/PAUSE/DISABLE ioctls.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: powerpc: convert marker probes to event trace
Marcelo Tosatti [Thu, 18 Jun 2009 14:47:27 +0000]
KVM: powerpc: convert marker probes to event trace

[avi: make it build]
[avi: fold trace-arch.h into trace.h]

CC: Hollis Blanchard <hollisb@us.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: introduce module parameter for ignoring unknown MSRs accesses
Andre Przywara [Thu, 25 Jun 2009 10:36:49 +0000]
KVM: introduce module parameter for ignoring unknown MSRs accesses

KVM will inject a #GP into the guest if that tries to access unhandled
MSRs. This will crash many guests. Although it would be the correct
way to actually handle these MSRs, we introduce a runtime switchable
module param called "ignore_msrs" (defaults to 0). If this is Y, unknown
MSR reads will return 0, while MSR writes are simply dropped. In both cases
we print a message to dmesg to inform the user about that.

You can change the behaviour at any time by saying:

 # echo 1 > /sys/modules/kvm/parameters/ignore_msrs

Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: ignore reads from AMDs C1E enabled MSR
Andre Przywara [Wed, 24 Jun 2009 10:44:34 +0000]
KVM: ignore reads from AMDs C1E enabled MSR

If the Linux kernel detects an C1E capable AMD processor (K8 RevF and
higher), it will access a certain MSR on every attempt to go to halt.
Explicitly handle this read and return 0 to let KVM run a Linux guest
with the native AMD host CPU propagated to the guest.

Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: ignore AMDs HWCR register access to set the FFDIS bit
Andre Przywara [Wed, 24 Jun 2009 10:44:33 +0000]
KVM: ignore AMDs HWCR register access to set the FFDIS bit

Linux tries to disable the flush filter on all AMD K8 CPUs. Since KVM
does not handle the needed MSR, the injected #GP will panic the Linux
kernel. Ignore setting of the HWCR.FFDIS bit in this MSR to let Linux
boot with an AMD K8 family guest CPU.

Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: x86: missing locking in PIT/IRQCHIP/SET_BSP_CPU ioctl paths
Marcelo Tosatti [Tue, 23 Jun 2009 18:05:14 +0000]
KVM: x86: missing locking in PIT/IRQCHIP/SET_BSP_CPU ioctl paths

Correct missing locking in a few places in x86's vm_ioctl handling path.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Prepare memslot data structures for multiple hugepage sizes
Joerg Roedel [Fri, 19 Jun 2009 13:16:23 +0000]
KVM: Prepare memslot data structures for multiple hugepage sizes

[avi: fix build on non-x86]

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agohugetlbfs: export vma_kernel_pagsize to modules
Joerg Roedel [Fri, 19 Jun 2009 13:16:22 +0000]
hugetlbfs: export vma_kernel_pagsize to modules

This function is required by KVM.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: s390: Fix memslot initialization for userspace_addr != 0
Christian Borntraeger [Tue, 23 Jun 2009 15:24:06 +0000]
KVM: s390: Fix memslot initialization for userspace_addr != 0

Since
commit 854b5338196b1175706e99d63be43a4f8d8ab607
Author: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
    KVM: s390: streamline memslot handling

s390 uses the values of the memslot instead of doing everything in the arch
ioctl handler of the KVM_SET_USER_MEMORY_REGION. Unfortunately we missed to
set the userspace_addr of our memslot due to our s390 ifdef in
__kvm_set_memory_region.
Old s390 userspace launchers did not notice, since they started the guest at
userspace address 0.
Because of CONFIG_DEFAULT_MMAP_MIN_ADDR we now put the guest at 1M userspace,
which does not work. This patch makes sure that new.userspace_addr is set
on s390.
This fix should go in quickly. Nevertheless, looking at the code we should
clean up that ifdef in the long term. Any kernel janitors?

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: x86 emulator: Add sysexit emulation
Andre Przywara [Thu, 18 Jun 2009 10:56:02 +0000]
KVM: x86 emulator: Add sysexit emulation

Handle #UD intercept of the sysexit instruction in 64bit mode returning to
32bit compat mode on an AMD host.
Setup the segment descriptors for CS and SS and the EIP/ESP registers
according to the manual.

Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: x86 emulator: Add sysenter emulation
Andre Przywara [Thu, 18 Jun 2009 10:56:01 +0000]
KVM: x86 emulator: Add sysenter emulation

Handle #UD intercept of the sysenter instruction in 32bit compat mode on
an AMD host.
Setup the segment descriptors for CS and SS and the EIP/ESP registers
according to the manual.

Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: x86 emulator: add syscall emulation
Andre Przywara [Thu, 18 Jun 2009 10:56:00 +0000]
KVM: x86 emulator: add syscall emulation

Handle #UD intercept of the syscall instruction in 32bit compat mode on
an Intel host.
Setup the segment descriptors for CS and SS and the EIP/ESP registers
according to the manual. Save the RIP and EFLAGS to the correct registers.

[avi: fix build on i386 due to missing R11]

Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: x86 emulator: Prepare for emulation of syscall instructions
Andre Przywara [Wed, 17 Jun 2009 13:50:33 +0000]
KVM: x86 emulator: Prepare for emulation of syscall instructions

Add the flags needed for syscall, sysenter and sysexit to the opcode table.
Catch (but for now ignore) the opcodes in the emulation switch/case.

Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: x86 emulator: Add missing EFLAGS bit definitions
Andre Przywara [Wed, 17 Jun 2009 13:50:32 +0000]
KVM: x86 emulator: Add missing EFLAGS bit definitions

Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Allow emulation of syscalls instructions on #UD
Andre Przywara [Wed, 17 Jun 2009 13:50:31 +0000]
KVM: Allow emulation of syscalls instructions on #UD

Add the opcodes for syscall, sysenter and sysexit to the list of instructions
handled by the undefined opcode handler.

Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: convert custom marker based tracing to event traces
Marcelo Tosatti [Wed, 17 Jun 2009 12:22:14 +0000]
KVM: convert custom marker based tracing to event traces

This allows use of the powerful ftrace infrastructure.

See Documentation/trace/ for usage information.

[avi, stephen: various build fixes]
[sheng: fix control register breakage]

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: SVM: Improve nested interrupt injection
Alexander Graf [Mon, 15 Jun 2009 13:21:25 +0000]
KVM: SVM: Improve nested interrupt injection

While trying to get Hyper-V running, I realized that the interrupt injection
mechanisms that are in place right now are not 100% correct.

This patch makes nested SVM's interrupt injection behave more like on a
real machine.

Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: SVM: Implement INVLPGA
Alexander Graf [Mon, 15 Jun 2009 13:21:24 +0000]
KVM: SVM: Implement INVLPGA

SVM adds another way to do INVLPG by ASID which Hyper-V makes use of,
so let's implement it!

For now we just do the same thing invlpg does, as asid switching
means we flush the mmu anyways. That might change one day though.

Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Implement MSRs used by Hyper-V
Alexander Graf [Mon, 15 Jun 2009 13:21:23 +0000]
KVM: Implement MSRs used by Hyper-V

Hyper-V uses some MSRs, some of which are actually reserved for BIOS usage.

But let's be nice today and have it its way, because otherwise it fails
terribly.

[jaswinder: fix build for linux-next changes]

Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agox86: Add definition for IGNNE MSR
Alexander Graf [Mon, 15 Jun 2009 13:21:22 +0000]
x86: Add definition for IGNNE MSR

Hyper-V accesses MSR_IGNNE while running under KVM.

Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: SVM: Don't save/restore host cr2
Avi Kivity [Tue, 16 Jun 2009 09:36:33 +0000]
KVM: SVM: Don't save/restore host cr2

The host never reads cr2 in process context, so are free to clobber it.  The
vmx code does this, so we can safely remove the save/restore code.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: VMX: Only reload guest cr2 if different from host cr2
Avi Kivity [Tue, 16 Jun 2009 09:33:56 +0000]
KVM: VMX: Only reload guest cr2 if different from host cr2

cr2 changes only rarely, and writing it is expensive.  Avoid the costly cr2
writes by checking if it does not already hold the desired value.

Shaves 70 cycles off the vmexit latency.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Drop useless atomic test from timer function
Jan Kiszka [Tue, 9 Jun 2009 13:37:03 +0000]
KVM: Drop useless atomic test from timer function

The current code tries to optimize the setting of
KVM_REQ_PENDING_TIMER but used atomic_inc_and_test - which always
returns true unless pending had the invalid value of -1 on entry. This
patch drops the test part preserving the original semantic but
expressing it less confusingly.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Fix racy event propagation in timer
Jan Kiszka [Tue, 9 Jun 2009 13:37:01 +0000]
KVM: Fix racy event propagation in timer

Minor issue that likely had no practical relevance: the kvm timer
function so far incremented the pending counter and then may reset it
again to 1 in case reinjection was disabled. This opened a small racy
window with the corresponding VCPU loop that may have happened to run
on another (real) CPU and already consumed the value.

Fix it by skipping the incrementation in case pending is already > 0.
This opens a different race windows, but may only rarely cause lost
events in case we do not care about them anyway (!reinject).

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Optimize searching for highest IRR
Gleb Natapov [Thu, 11 Jun 2009 08:06:51 +0000]
KVM: Optimize searching for highest IRR

Most of the time IRR is empty, so instead of scanning the whole IRR on
each VM entry keep a variable that tells us if IRR is not empty. IRR
will have to be scanned twice on each IRQ delivery, but this is much
more rare than VM entry.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Replace pending exception by PF if it happens serially
Gleb Natapov [Thu, 11 Jun 2009 08:26:11 +0000]
KVM: Replace pending exception by PF if it happens serially

Replace previous exception with a new one in a hope that instruction
re-execution will regenerate lost exception.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: VMX: conditionally disable 2M pages
Marcelo Tosatti [Thu, 11 Jun 2009 15:07:44 +0000]
KVM: VMX: conditionally disable 2M pages

Disable usage of 2M pages if VMX_EPT_2MB_PAGE_BIT (bit 16) is clear
in MSR_IA32_VMX_EPT_VPID_CAP and EPT is enabled.

[avi: s/largepages_disabled/largepages_enabled/ to avoid negative logic]

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: VMX: EPT misconfiguration handler
Marcelo Tosatti [Thu, 11 Jun 2009 15:07:43 +0000]
KVM: VMX: EPT misconfiguration handler

Handler for EPT misconfiguration which checks for valid state
in the shadow pagetables, printing the spte on each level.

The separate WARN_ONs are useful for kerneloops.org.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: MMU: add kvm_mmu_get_spte_hierarchy helper
Marcelo Tosatti [Thu, 11 Jun 2009 15:07:42 +0000]
KVM: MMU: add kvm_mmu_get_spte_hierarchy helper

Required by EPT misconfiguration handler.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: MMU: make for_each_shadow_entry aware of largepages
Marcelo Tosatti [Thu, 11 Jun 2009 15:07:41 +0000]
KVM: MMU: make for_each_shadow_entry aware of largepages

This way there is no need to add explicit checks in every
for_each_shadow_entry user.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: VMX: more MSR_IA32_VMX_EPT_VPID_CAP capability bits
Marcelo Tosatti [Thu, 11 Jun 2009 15:07:40 +0000]
KVM: VMX: more MSR_IA32_VMX_EPT_VPID_CAP capability bits

Required for EPT misconfiguration handler.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Move performance counter MSR access interception to generic x86 path
Andre Przywara [Fri, 12 Jun 2009 20:01:29 +0000]
KVM: Move performance counter MSR access interception to generic x86 path

The performance counter MSRs are different for AMD and Intel CPUs and they
are chosen mainly by the CPUID vendor string. This patch catches writes to
all addresses (regardless of VMX/SVM path) and handles them in the generic
MSR handler routine. Writing a 0 into the event select register is something
we perfectly emulate ;-), so don't print out a warning to dmesg in this
case.
This fixes booting a 64bit Windows guest with an AMD CPUID on an Intel host.

Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: MMU audit: largepage handling
Marcelo Tosatti [Wed, 10 Jun 2009 15:27:08 +0000]
KVM: MMU audit: largepage handling

Make the audit code aware of largepages.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: MMU audit: audit_mappings tweaks
Marcelo Tosatti [Wed, 10 Jun 2009 15:27:07 +0000]
KVM: MMU audit: audit_mappings tweaks

- Fail early in case gfn_to_pfn returns is_error_pfn.
- For the pre pte write case, avoid spurious "gva is valid but spte is notrap"
  messages (the emulation code does the guest write first, so this particular
  case is OK).

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: MMU audit: nontrapping ptes in nonleaf level
Marcelo Tosatti [Wed, 10 Jun 2009 15:27:06 +0000]
KVM: MMU audit: nontrapping ptes in nonleaf level

It is valid to set non leaf sptes as notrap.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: MMU audit: update audit_write_protection
Marcelo Tosatti [Wed, 10 Jun 2009 15:27:05 +0000]
KVM: MMU audit: update audit_write_protection

- Unsync pages contain writable sptes in the rmap.
- rmaps do not exclusively contain writable sptes anymore.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: MMU audit: update count_writable_mappings / count_rmaps
Marcelo Tosatti [Wed, 10 Jun 2009 15:27:04 +0000]
KVM: MMU audit: update count_writable_mappings / count_rmaps

Under testing, count_writable_mappings returns a value that is 2 integers
larger than what count_rmaps returns.

Suspicion is that either of the two functions is counting a duplicate (either
positively or negatively).

Modifying check_writable_mappings_rmap to check for rmap existance on
all present MMU pages fails to trigger an error, which should keep Avi
happy.

Also introduce mmu_spte_walk to invoke a callback on all present sptes visible
to the current vcpu, might be useful in the future.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: MMU: introduce is_last_spte helper
Marcelo Tosatti [Wed, 10 Jun 2009 15:27:03 +0000]
KVM: MMU: introduce is_last_spte helper

Hiding some of the last largepage / level interaction (which is useful
for gbpages and for zero based levels).

Also merge the PT_PAGE_TABLE_LEVEL clearing loop in unlink_children.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Return to userspace on emulation failure
Avi Kivity [Thu, 11 Jun 2009 12:43:28 +0000]
KVM: Return to userspace on emulation failure

Instead of mindlessly retrying to execute the instruction, report the
failure to userspace.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Use macro to iterate over vcpus.
Gleb Natapov [Tue, 9 Jun 2009 12:56:29 +0000]
KVM: Use macro to iterate over vcpus.

[christian: remove unused variables on s390]

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Break dependency between vcpu index in vcpus array and vcpu_id.
Gleb Natapov [Tue, 9 Jun 2009 12:56:28 +0000]
KVM: Break dependency between vcpu index in vcpus array and vcpu_id.

Archs are free to use vcpu_id as they see fit. For x86 it is used as
vcpu's apic id. New ioctl is added to configure boot vcpu id that was
assumed to be 0 till now.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Use pointer to vcpu instead of vcpu_id in timer code.
Gleb Natapov [Tue, 9 Jun 2009 12:56:27 +0000]
KVM: Use pointer to vcpu instead of vcpu_id in timer code.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Introduce kvm_vcpu_is_bsp() function.
Gleb Natapov [Tue, 9 Jun 2009 12:56:26 +0000]
KVM: Introduce kvm_vcpu_is_bsp() function.

Use it instead of open code "vcpu_id zero is BSP" assumption.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: MMU: s/shadow_pte/spte/
Avi Kivity [Wed, 10 Jun 2009 11:24:23 +0000]
KVM: MMU: s/shadow_pte/spte/

We use shadow_pte and spte inconsistently, switch to the shorter spelling.

Rename set_shadow_pte() to __set_spte() to avoid a conflict with the
existing set_spte(), and to indicate its lowlevelness.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: MMU: Adjust pte accessors to explicitly indicate guest or shadow pte
Avi Kivity [Wed, 10 Jun 2009 11:12:05 +0000]
KVM: MMU: Adjust pte accessors to explicitly indicate guest or shadow pte

Since the guest and host ptes can have wildly different format, adjust
the pte accessor names to indicate on which type of pte they operate on.

No functional changes.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: MMU: Fix is_dirty_pte()
Avi Kivity [Wed, 10 Jun 2009 09:56:54 +0000]
KVM: MMU: Fix is_dirty_pte()

is_dirty_pte() is used on guest ptes, not shadow ptes, so it needs to avoid
shadow_dirty_mask and use PT_DIRTY_MASK instead.

Misdetecting dirty pages could lead to unnecessarily setting the dirty bit
under EPT.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: VMX: Move rmode structure to vmx-specific code
Avi Kivity [Tue, 9 Jun 2009 11:10:45 +0000]
KVM: VMX: Move rmode structure to vmx-specific code

rmode is only used in vmx, so move it to vmx.c

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Reorder ioctls in kvm.h
Avi Kivity [Tue, 9 Jun 2009 08:33:36 +0000]
KVM: Reorder ioctls in kvm.h

Somehow the VM ioctls got unsorted; resort.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: VMX: Support Unrestricted Guest feature
Nitin A Kamble [Mon, 8 Jun 2009 18:34:16 +0000]
KVM: VMX: Support Unrestricted Guest feature

"Unrestricted Guest" feature is added in the VMX specification.
Intel Westmere and onwards processors will support this feature.

    It allows kvm guests to run real mode and unpaged mode
code natively in the VMX mode when EPT is turned on. With the
unrestricted guest there is no need to emulate the guest real mode code
in the vm86 container or in the emulator. Also the guest big real mode
code works like native.

  The attached patch enhances KVM to use the unrestricted guest feature
if available on the processor. It also adds a new kernel/module
parameter to disable the unrestricted guest feature at the boot time.

Signed-off-by: Nitin A Kamble <nitin.a.kamble@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: switch irq injection/acking data structures to irq_lock
Marcelo Tosatti [Thu, 4 Jun 2009 18:08:24 +0000]
KVM: switch irq injection/acking data structures to irq_lock

Protect irq injection/acking data structures with a separate irq_lock
mutex. This fixes the following deadlock:

CPU A                               CPU B
kvm_vm_ioctl_deassign_dev_irq()
  mutex_lock(&kvm->lock);            worker_thread()
  -> kvm_deassign_irq()                -> kvm_assigned_dev_interrupt_work_handler()
    -> deassign_host_irq()               mutex_lock(&kvm->lock);
      -> cancel_work_sync() [blocked]

[gleb: fix ia64 path]

Reported-by: Alex Williamson <alex.williamson@hp.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: introduce irq_lock, use it to protect ioapic
Marcelo Tosatti [Thu, 4 Jun 2009 18:08:23 +0000]
KVM: introduce irq_lock, use it to protect ioapic

Introduce irq_lock, and use to protect ioapic data structures.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: move coalesced_mmio locking to its own device
Marcelo Tosatti [Thu, 4 Jun 2009 18:08:22 +0000]
KVM: move coalesced_mmio locking to its own device

Move coalesced_mmio locking to its own device, instead of relying on
kvm->lock.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Grab pic lock in kvm_pic_clear_isr_ack
Marcelo Tosatti [Thu, 4 Jun 2009 18:08:21 +0000]
KVM: Grab pic lock in kvm_pic_clear_isr_ack

isr_ack is protected by kvm_pic->lock.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Cleanup LAPIC interface
Jan Kiszka [Fri, 5 Jun 2009 14:13:34 +0000]
KVM: Cleanup LAPIC interface

None of the interface services the LAPIC emulation provides need to be
exported to modules, and kvm_lapic_get_base is even totally unused
today.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: ppc: e500: Add MMUCFG and PVR emulation
Liu Yu [Fri, 5 Jun 2009 06:54:31 +0000]
KVM: ppc: e500: Add MMUCFG and PVR emulation

Latest kernel started to use these two registers.

Signed-off-by: Liu Yu <yu.liu@freescale.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: ppc: e500: Directly pass pvr to guest
Liu Yu [Fri, 5 Jun 2009 06:54:30 +0000]
KVM: ppc: e500: Directly pass pvr to guest

Signed-off-by: Liu Yu <yu.liu@freescale.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: ppc: e500: Move to Book-3e MMU definitions
Liu Yu [Fri, 5 Jun 2009 06:54:29 +0000]
KVM: ppc: e500: Move to Book-3e MMU definitions

According to commit 70fe3af8403f85196bb74f22ce4813db7dfedc1a.

Signed-off-by: Liu Yu <yu.liu@freescale.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Calculate available entries in coalesced mmio ring
Avi Kivity [Thu, 4 Jun 2009 15:09:08 +0000]
KVM: Calculate available entries in coalesced mmio ring

Instead of checking whether we'll wrap around, calculate how many entries
are available, and check whether we have enough (just one) for the pending
mmio.

By itself, this doesn't change anything, but it paves the way for making
this function lockless.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: VMX: Fix reporting of unhandled EPT violations
Avi Kivity [Wed, 3 Jun 2009 11:12:10 +0000]
KVM: VMX: Fix reporting of unhandled EPT violations

Instead of returning -ENOTSUPP, exit normally but indicate the hardware
exit reason.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Cache pdptrs
Avi Kivity [Sun, 31 May 2009 19:58:47 +0000]
KVM: Cache pdptrs

Instead of reloading the pdptrs on every entry and exit (vmcs writes on vmx,
guest memory access on svm) extract them on demand.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: VMX: Simplify pdptr and cr3 management
Avi Kivity [Sun, 31 May 2009 15:41:29 +0000]
KVM: VMX: Simplify pdptr and cr3 management

Instead of reading the PDPTRs from memory after every exit (which is slow
and wrong, as the PDPTRs are stored on the cpu), sync the PDPTRs from
memory to the VMCS before entry, and from the VMCS to memory after exit.
Do the same for cr3.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: VMX: Avoid duplicate ept tlb flush when setting cr3
Avi Kivity [Sun, 31 May 2009 15:29:59 +0000]
KVM: VMX: Avoid duplicate ept tlb flush when setting cr3

vmx_set_cr3() will call vmx_tlb_flush(), which will flush the ept context.
So there is no need to call ept_sync_context() explicitly.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: do not register i8254 PIO regions until we are initialized
Gregory Haskins [Mon, 1 Jun 2009 16:54:56 +0000]
KVM: do not register i8254 PIO regions until we are initialized

We currently publish the i8254 resources to the pio_bus before the devices
are fully initialized.  Since we hold the pit_lock, its probably not
a real issue.  But lets clean this up anyway.

Reported-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
Acked-by: Chris Wright <chrisw@sous-sol.org>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: cleanup io_device code
Gregory Haskins [Mon, 1 Jun 2009 16:54:50 +0000]
KVM: cleanup io_device code

We modernize the io_device code so that we use container_of() instead of
dev->private, and move the vtable to a separate ops structure
(theoretically allows better caching for multiple instances of the same
ops structure)

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
Acked-by: Chris Wright <chrisw@sous-sol.org>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: Clean up coalesced_mmio destruction
Gregory Haskins [Mon, 1 Jun 2009 16:54:45 +0000]
KVM: Clean up coalesced_mmio destruction

We invoke kfree() on a data member instead of the structure.  This works today
because the kvm_io_device is the first element of the private structure, but
this could change in the future, so lets clean this up.

Signed-off-by: Gregory Haskins <ghaskins@novell.com>
Acked-by: Chris Wright <chrisw@sous-sol.org>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: powerpc: fix some init/exit annotations
Stephen Rothwell [Tue, 2 Jun 2009 01:46:14 +0000]
KVM: powerpc: fix some init/exit annotations

Fixes a couple of warnings like this one:

WARNING: arch/powerpc/kvm/kvm-440.o(.text+0x1e8c): Section mismatch in reference from the function kvmppc_44x_exit() to the function .exit.text:kvmppc_booke_exit()
The function kvmppc_44x_exit() references a function in an exit section.
Often the function kvmppc_booke_exit() has valid usage outside the exit section
and the fix is to remove the __exit annotation of kvmppc_booke_exit.

Also add some __init annotations on obvious routines.

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: SVM: Fold kvm_svm.h info svm.c
Avi Kivity [Sun, 31 May 2009 15:15:37 +0000]
KVM: SVM: Fold kvm_svm.h info svm.c

kvm_svm.h is only included from svm.c, so fold it in.

Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: remove redundant declarations
Christian Ehrhardt [Fri, 29 May 2009 10:58:50 +0000]
KVM: remove redundant declarations

Changing s390 code in kvm_arch_vcpu_load/put come across this header
declarations. They are complete duplicates, not even useful forward
declarations as nothing using it is in between (maybe it was that in
the past).

This patch removes the two dispensable lines.

Signed-off-by: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: No disable_irq for MSI/MSI-X interrupt on device assignment
Sheng Yang [Thu, 30 Apr 2009 02:58:42 +0000]
KVM: No disable_irq for MSI/MSI-X interrupt on device assignment

Disable interrupt at interrupt handler and enable it when guest ack is for
the level triggered interrupt, to prevent reinjected interrupt. MSI/MSI-X don't
need it.

One possible problem is multiply same vector interrupt injected between irq
handler and scheduled work handler would be merged as one for MSI/MSI-X.
But AFAIK, the drivers handle it well.

The patch fixed the oplin card performance issue(MSI-X performance is half of
MSI/INTx).

Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

11 years agoKVM: SVM: use explicit 64bit storage for sysenter values
Andre Przywara [Thu, 28 May 2009 09:56:31 +0000]
KVM: SVM: use explicit 64bit storage for sysenter values

Since AMD does not support sysenter in 64bit mode, the VMCB fields storing
the MSRs are truncated to 32bit upon VMRUN/#VMEXIT. So store the values
in a separate 64bit storage to avoid truncation.

[andre: fix amd->amd migration]

Signed-off-by: Christoph Egger <christoph.egger@amd.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>