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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
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>
The in-kernel speaker emulation is only a dummy and also unneeded from
the performance point of view. Rather, it takes user space support to
generate sound output on the host, e.g. console beeps.
To allow this, introduce KVM_CREATE_PIT2 which controls in-kernel
speaker port emulation via a flag passed along the new IOCTL. It also
leaves room for future extensions of the PIT configuration interface.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
KVM provides a complete virtual system environment for guests, including
support for injecting interrupts modeled after the real exception/interrupt
facilities present on the native platform (such as the IDT on x86).
Virtual interrupts can come from a variety of sources (emulated devices,
pass-through devices, etc) but all must be injected to the guest via
the KVM infrastructure. This patch adds a new mechanism to inject a specific
interrupt to a guest using a decoupled eventfd mechnanism: Any legal signal
on the irqfd (using eventfd semantics from either userspace or kernel) will
translate into an injected interrupt in the guest at the next available
interrupt window.
Signed-off-by: Gregory Haskins <ghaskins@novell.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The problem exists only on VMX. Also currently we skip this step if
there is pending exception. The patch fixes this too.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
If we run out of cpuid entries for extended request types
we should return -E2BIG, just like we do for the standard
request types.
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Replace 0xc0010010 with MSR_K8_SYSCFG and 0xc0010015 with MSR_K7_HWCR.
Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
The related MSRs are emulated. MCE capability is exported via
extension KVM_CAP_MCE and ioctl KVM_X86_GET_MCE_CAP_SUPPORTED. A new
vcpu ioctl command KVM_X86_SETUP_MCE is used to setup MCE emulation
such as the mcg_cap. MCE is injected via vcpu ioctl command
KVM_X86_SET_MCE. Extended machine-check state (MCG_EXT_P) and CMCI are
not implemented.
Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Use standard msr-index.h's MSR declaration.
MSR_IA32_TSC is better than MSR_IA32_TIME_STAMP_COUNTER as it also solves
80 column issue.
Signed-off-by: Jaswinder Singh Rajput <jaswinderrajput@gmail.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Do not allow invalid memory types in MTRR/PAT (generating a #GP
otherwise).
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
So far, KVM copied the emulated_msrs (only MSR_IA32_MISC_ENABLE) to a
wrong address in user space due to broken pointer arithmetic. This
caused subtle corruption up there (missing MSR_IA32_MISC_ENABLE had
probably no practical relevance). Moreover, the size check for the
user-provided kvm_msr_list forgot about emulated MSRs.
Cc: stable@kernel.org
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
In commit 7fe29e0faa we ignored the
reads to the P6 EVNTSEL MSRs. That fixed crashes on Intel machines.
Ignore the reads to K7 EVNTSEL MSRs as well to fix this on AMD
hosts.
This fixes Kaspersky antivirus crashing Windows guests on AMD hosts.
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
INTn will be re-executed after migration. If we wanted to migrate
pending software interrupt we would need to migrate interrupt type
and instruction length too, but we do not have all required info on
SVM, so SVM->VMX migration would need to re-execute INTn anyway. To
make it simple never migrate pending soft interrupt.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Currently they are not requested if there is pending exception.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Re-inject event instead. This is what Intel suggest. Also use correct
instruction length when re-injecting soft fault/interrupt.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Only one interrupt vector can be injected from userspace irqchip at
any given time so no need to store it in a bitmap. Put it into interrupt
queue directly.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Verify the cr3 address stored in vcpu->arch.cr3 points to an existant
memslot. If not, inject a triple fault.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
kvm_handle_hva, called by MMU notifiers, manipulates mmu data only with
the protection of mmu_lock.
Update kvm_mmu_change_mmu_pages callers to take mmu_lock, thus protecting
against kvm_handle_hva.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
We currently unblock shadow interrupt state when we skip an instruction,
but failing to do so when we actually emulate one. This blocks interrupts
in key instruction blocks, in particular sti; hlt; sequences
If the instruction emulated is an sti, we have to block shadow interrupts.
The same goes for mov ss. pop ss also needs it, but we don't currently
emulate it.
Without this patch, I cannot boot gpxe option roms at vmx machines.
This is described at https://bugzilla.redhat.com/show_bug.cgi?id=494469
Signed-off-by: Glauber Costa <glommer@redhat.com>
CC: H. Peter Anvin <hpa@zytor.com>
CC: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch replaces drop_interrupt_shadow with the more
general set_interrupt_shadow, that can either drop or raise
it, depending on its parameter. It also adds ->get_interrupt_shadow()
for future use.
Signed-off-by: Glauber Costa <glommer@redhat.com>
CC: H. Peter Anvin <hpa@zytor.com>
CC: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
KVM uses a function call IPI to cause the exit of a guest running on a
physical cpu. For virtual interrupt notification there is no need to
wait on IPI receival, or to execute any function.
This is exactly what the reschedule IPI does, without the overhead
of function IPI. So use it instead of smp_call_function_single in
kvm_vcpu_kick.
Also change the "guest_mode" variable to a bit in vcpu->requests, and
use that to collapse multiple IPI's that would be issued between the
first one and zeroing of guest mode.
This allows kvm_vcpu_kick to called with interrupts disabled.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Gregory Haskins
