I noticed that srcu_read_lock/unlock both have a memory barrier,
so just by moving srcu_read_unlock earlier we can get rid of
one call to smp_mb() using smp_mb__after_srcu_read_unlock instead.

Unsurprisingly, the gain is small but measureable using the unit test
microbenchmark:
before
        vmcall in the ballpark of 1410 cycles
after
        vmcall in the ballpark of 1360 cycles
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>

srcu read lock/unlock include a full memory barrier
but that's an implementation detail.
Add an API for make memory fencing explicit for
users that need this barrier, to make sure we
can change it as needed without breaking all users.
Acked-by: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>

It was used in conjunction with KVM_SET_MEMORY_REGION ioctl which was
removed by b74a07be in 2010, QEMU stopped using it in 2008, so
it is time to remove the code finally.
Signed-off-by: Gleb Natapov <gleb@redhat.com>

When determining the page size we could use to map with the IOMMU, the
page size should also be aligned with the hva, not just the gfn.  The
gfn may not reflect the real alignment within the hugetlbfs file.

Most of the time, this works fine.  However, if the hugetlbfs file is
backed by non-contiguous huge pages, a multi-huge page memslot starts at
an unaligned offset within the hugetlbfs file, and the gfn is aligned
with respect to the huge page size, kvm_host_page_size() will return the
huge page size and we will use that to map with the IOMMU.

When we later unpin that same memslot, the IOMMU returns the unmap size
as the huge page size, and we happily unpin that many pfns in
monotonically increasing order, not realizing we are spanning
non-contiguous huge pages and partially unpin the wrong huge page.

Ensure the IOMMU mapping page size is aligned with the hva corresponding
to the gfn, which does reflect the alignment within the hugetlbfs file.
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Greg Edwards <gedwards@ddn.com>
Cc: stable@vger.kernel.org
Signed-off-by: Gleb Natapov <gleb@redhat.com>

Currently cpuid emulation is traced only when executed by intercept.
Move trace point so that emulator invocation is traced too.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>

Make code shorter.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>

All decode_register() callers check if instruction has rex prefix
to properly decode one byte operand. It make sense to move the check
inside.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>

Conflicts:
	arch/powerpc/include/asm/processor.h

When I was looking at RHEL5.9's failure to start with
unrestricted_guest=0/emulate_invalid_guest_state=1, I got it working with a
slightly older tree than kvm.git.  I now debugged the remaining failure,
which was introduced by commit 660696d1 (KVM: X86 emulator: fix
source operand decoding for 8bit mov[zs]x instructions, 2013-04-24)
introduced a similar mis-emulation to the one in commit 8acb4207 (KVM:
fix sil/dil/bpl/spl in the mod/rm fields, 2013-05-30).  The incorrect
decoding occurs in 8-bit movzx/movsx instructions whose 8-bit operand
is sil/dil/bpl/spl.

Needless to say, "movzbl %bpl, %eax" does occur in RHEL5.9's decompression
prolog, just a handful of instructions before finally giving control to
the decompressed vmlinux and getting out of the invalid guest state.

Because OpMem8 bypasses decode_modrm, the same handling of the REX prefix
must be applied to OpMem8.
Reported-by: Michele Baldessari <michele@redhat.com>
Cc: stable@vger.kernel.org
Cc: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>

fix up typo in comment.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Yet another instruction that we fail to emulate, this time found
in Windows 2008R2 32-bit.
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Cc: Gleb Natapov <gleb@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: KVM List <kvm@vger.kernel.org>
Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Cc: Gleb Natapov <gleb@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com>
[Some editing. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Cc: Gleb Natapov <gleb@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

mst can't be blamed for lack of switch entries: the
issue is with msrs actually.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

The loop was always using 0 as the index.  This means that
any rubbish after the first element of the array went undetected.
It seems reasonable to assume that no KVM userspace did that.
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

The KVM_SET_XCRS ioctl must accept anything that KVM_GET_XCRS
could return.  XCR0's bit 0 is always 1 in real processors with
XSAVE, and KVM_GET_XCRS will always leave bit 0 set even if the
emulated processor does not have XSAVE.  So, KVM_SET_XCRS must
ignore that bit when checking for attempts to enable unsupported
save states.
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

We currently use some ad-hoc arch variables tied to legacy KVM device
assignment to manage emulation of instructions that depend on whether
non-coherent DMA is present.  Create an interface for this, adapting
legacy KVM device assignment and adding VFIO via the KVM-VFIO device.
For now we assume that non-coherent DMA is possible any time we have a
VFIO group.  Eventually an interface can be developed as part of the
VFIO external user interface to query the coherency of a group.
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Default to operating in coherent mode.  This simplifies the logic when
we switch to a model of registering and unregistering noncoherent I/O
with KVM.
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

So far we've succeeded at making KVM and VFIO mostly unaware of each
other, but areas are cropping up where a connection beyond eventfds
and irqfds needs to be made.  This patch introduces a KVM-VFIO device
that is meant to be a gateway for such interaction.  The user creates
the device and can add and remove VFIO groups to it via file
descriptors.  When a group is added, KVM verifies the group is valid
and gets a reference to it via the VFIO external user interface.
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

This basically came from the need to be able to boot 32-bit Atom SMP
guests on an AMD host, i.e. a host which doesn't support MOVBE. As a
matter of fact, qemu has since recently received MOVBE support but we
cannot share that with kvm emulation and thus we have to do this in the
host. We're waay faster in kvm anyway. :-)

So, we piggyback on the #UD path and emulate the MOVBE functionality.
With it, an 8-core SMP guest boots in under 6 seconds.

Also, requesting MOVBE emulation needs to happen explicitly to work,
i.e. qemu -cpu n270,+movbe...

Just FYI, a fairly straight-forward boot of a MOVBE-enabled 3.9-rc6+
kernel in kvm executes MOVBE ~60K times.
Signed-off-by: Andre Przywara <andre@andrep.de>
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Add initial support for handling three-byte instructions in the
emulator.
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Call it EmulateOnUD which is exactly what we're trying to do with
vendor-specific instructions.

Rename ->only_vendor_specific_insn to something shorter, while at it.
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Add a field to the current emulation context which contains the
instruction opcode length. This will streamline handling of opcodes of
different length.
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Add a kvm ioctl which states which system functionality kvm emulates.
The format used is that of CPUID and we return the corresponding CPUID
bits set for which we do emulate functionality.

Make sure ->padding is being passed on clean from userspace so that we
can use it for something in the future, after the ioctl gets cast in
stone.

s/kvm_dev_ioctl_get_supported_cpuid/kvm_dev_ioctl_get_cpuid/ while at
it.
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Updates for KVM/ARM, take 2 including:
 - Transparent Huge Pages and hugetlbfs support for KVM/ARM
 - Yield CPU when guest executes WFE to speed up CPU overcommit

In kvm_iommu_map_pages(), we need to know the page size via call
kvm_host_page_size(). And it will check whether the target slot
is valid before return the right page size.
Currently, we will map the iommu pages when creating a new slot.
But we call kvm_iommu_map_pages() during preparing the new slot.
At that time, the new slot is not visible by domain(still in preparing).
So we cannot get the right page size from kvm_host_page_size() and
this will break the IOMMU super page logic.
The solution is to map the iommu pages after we insert the new slot
into domain.
Signed-off-by: Yang Zhang <yang.z.zhang@Intel.com>
Tested-by: Patrick Lu <patrick.lu@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

If the host supports it, we can and should expose it to the guest as
well, just like we already do with PIN_BASED_VIRTUAL_NMIS.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

__vmx_complete_interrupts stored uninjected NMIs in arch.nmi_injected,
not arch.nmi_pending. So we actually need to check the former field in
vmcs12_save_pending_event. This fixes the eventinj unit test when run
in nested KVM.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

As long as the hardware provides us 2MB EPT pages, we can also expose
them to the guest because our shadow EPT code already supports this
feature.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Support transparent huge pages in KVM/ARM and KVM/ARM64.  The
transparent_hugepage_adjust is not very pretty, but this is also how
it's solved on x86 and seems to be simply an artifact on how THPs
behave.  This should eventually be shared across architectures if
possible, but that can always be changed down the road.
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

Support huge pages in KVM/ARM and KVM/ARM64.  The pud_huge checking on
the unmap path may feel a bit silly as the pud_huge check is always
defined to false, but the compiler should be smart about this.

Note: This deals only with VMAs marked as huge which are allocated by
users through hugetlbfs only.  Transparent huge pages can only be
detected by looking at the underlying pages (or the page tables
themselves) and this patch so far simply maps these on a page-by-page
level in the Stage-2 page tables.

Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Russell King <rmk+kernel@arm.linux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

Update comments to reflect what is really going on and add the TWE bit
to the comments in kvm_arm.h.

Also renames the function to kvm_handle_wfx like is done on arm64 for
consistency and uber-correctness.
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

On an (even slightly) oversubscribed system, spinlocks are quickly
becoming a bottleneck, as some vcpus are spinning, waiting for a
lock to be released, while the vcpu holding the lock may not be
running at all.

This creates contention, and the observed slowdown is 40x for
hackbench. No, this isn't a typo.

The solution is to trap blocking WFEs and tell KVM that we're
now spinning. This ensures that other vpus will get a scheduling
boost, allowing the lock to be released more quickly. Also, using
CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT slightly improves the performance
when the VM is severely overcommited.

Quick test to estimate the performance: hackbench 1 process 1000

2xA15 host (baseline):	1.843s

2xA15 guest w/o patch:	2.083s
4xA15 guest w/o patch:	80.212s
8xA15 guest w/o patch:	Could not be bothered to find out

2xA15 guest w/ patch:	2.102s
4xA15 guest w/ patch:	3.205s
8xA15 guest w/ patch:	6.887s

So we go from a 40x degradation to 1.5x in the 2x overcommit case,
which is vaguely more acceptable.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>

drop is_hv_enabled, because that should not be a callback property
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>

This moves the kvmppc_ops callbacks to be a per VM entity. This
enables us to select HV and PR mode when creating a VM. We also
allow both kvm-hv and kvm-pr kernel module to be loaded. To
achieve this we move /dev/kvm ownership to kvm.ko module. Depending on
which KVM mode we select during VM creation we take a reference
count on respective module
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
[agraf: fix coding style]
Signed-off-by: Alexander Graf <agraf@suse.de>

Merging master into next to satisfy the dependencies.

Conflicts:
	arch/arm/kvm/reset.c

We will use that in the later patch to find the kvm ops handler
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
[agraf: squash in compile fix]
Signed-off-by: Alexander Graf <agraf@suse.de>

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>