KVM: s390: Fix memory slot versus run - v3

This patch fixes an incorrectness in the kvm backend for s390.
In case virtual cpus are being created before the corresponding
memory slot is being registered, we need to update the sie
control blocks for the virtual cpus.

*updates in v3*
In consideration of the s390 memslot constraints locking was changed
to trylock. These locks should never be held, as vcpu's can't run without
the single memslot we just assign when running this code. To ensure this
never deadlocks in case other code changes the code uses trylocks and bail
out if it can't get all locks.

Additionally most of the discussed special conditions for s390 like
only one memslot and no user_alloc are now checked for validity in
kvm_arch_set_memory_region.

Reported-by: Mijo Safradin <mijo@linux.vnet.ibm.com>
Signed-off-by: Carsten Otte <cotte@de.ibm.com>
Signed-off-by: Christian Ehrhardt <ehrhardt@de.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: Expand on "help" info to specify kvm intel and amd module names

Signed-off-by: Robert P. J. Day <rpjday@crashcourse.ca>
Cc: Avi Kivity <avi@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: x86: check for cr3 validity in mmu_alloc_roots

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: take mmu_lock when updating a deleted slot

kvm_handle_hva relies on mmu_lock protection to safely access
the memslot structures.

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

KVM: MMU: protect kvm_mmu_change_mmu_pages with mmu_lock

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>

KVM: Deal with interrupt shadow state for emulated instructions

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>

KVM: Replace ->drop_interrupt_shadow() by ->set_interrupt_shadow()

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: protect assigned dev workqueue, int handler and irq acker

kvm_assigned_dev_ack_irq is vulnerable to a race condition with the
interrupt handler function. It does:

        if (dev->host_irq_disabled) {
                enable_irq(dev->host_irq);
                dev->host_irq_disabled = false;
        }

If an interrupt triggers before the host->dev_irq_disabled assignment,
it will disable the interrupt and set dev->host_irq_disabled to true.

On return to kvm_assigned_dev_ack_irq, dev->host_irq_disabled is set to
false, and the next kvm_assigned_dev_ack_irq call will fail to reenable
it.

Other than that, having the interrupt handler and work handlers run in
parallel sounds like asking for trouble (could not spot any obvious
problem, but better not have to, its fragile).

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

KVM: use smp_send_reschedule in kvm_vcpu_kick

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>

KVM: Update cpuid 1.ecx reporting

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

x86: Add cpu features MOVBE and POPCNT

Add cpu feature bit support for the MOVBE and POPCNT instructions.

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

KVM: Add AMD cpuid bit: cr8_legacy, abm, misaligned sse, sse4, 3dnow prefetch

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

KVM: Fix cpuid feature misreporting

MTRR, PAT, MCE, and MCA are all supported (to some extent) but not reported.
Vista requires these features, so if userspace relies on kernel cpuid
reporting, it loses support for Vista.

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

KVM: Drop request_nmi from stats

The stats entry request_nmi is no longer used as the related user space
interface was dropped. So clean it up.

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

KVM: SVM: Don't reinject event that caused a task switch

If a task switch caused by an event remove it from the event queue.
VMX already does that.

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

KVM: SVM: Fix cross vendor migration issue in segment segment descriptor

On AMD CPUs sometimes the DB bit in the stack segment
descriptor is left as 1, although the whole segment has
been made unusable. Clear it here to pass an Intel VMX
entry check when cross vendor migrating.

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

KVM: fix apic_debug instances

Apparently nobody turned this on in a while...
setting apic_debug to something compilable, generates
some errors. This patch fixes it.

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

KVM: Trivial format fix in setup_routing_entry()

Remove extra tab.

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

KVM: VMX: Disable VMX when system shutdown

Intel TXT(Trusted Execution Technology) required VMX off for all cpu to work
when system shutdown.

CC: Joseph Cihula <joseph.cihula@intel.com>
Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: Enable snooping control for supported hardware

Memory aliases with different memory type is a problem for guest. For the guest
without assigned device, the memory type of guest memory would always been the
same as host(WB); but for the assigned device, some part of memory may be used
as DMA and then set to uncacheable memory type(UC/WC), which would be a conflict of
host memory type then be a potential issue.

Snooping control can guarantee the cache correctness of memory go through the
DMA engine of VT-d.

[avi: fix build on ia64]

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

KVM: Replace get_mt_mask_shift with get_mt_mask

Shadow_mt_mask is out of date, now it have only been used as a flag to indicate
if TDP enabled. Get rid of it and use tdp_enabled instead.

Also put memory type logical in kvm_x86_ops->get_mt_mask().

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

KVM: Wake up waitqueue before calling get_cpu()

This moves the get_cpu() call down to be called after we wake up the
waiters. Therefore the waitqueue locks can safely be rt mutex.

Signed-off-by: Jan Blunck <jblunck@suse.de>
Signed-off-by: Sven-Thorsten Dietrich <sven@thebigcorporation.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: Get rid of get_irq() callback

It just returns pending IRQ vector from the queue for VMX/SVM.
Get IRQ directly from the queue before migration and put it back
after.

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

KVM: Fix userspace IRQ chip migration

Re-put pending IRQ vector into interrupt_bitmap before migration.
Otherwise it will be lost if migration happens in the wrong time.

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

KVM: SVM: Add NMI injection support

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

KVM: Get rid of arch.interrupt_window_open & arch.nmi_window_open

They are recalculated before each use anyway.

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

KVM: Do not report TPR write to userspace if new value bigger or equal to a previous one.

Saves many exits to userspace in a case of IRQ chip in userspace.

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

KVM: sync_lapic_to_cr8() should always sync cr8 to V_TPR

Even if IRQ chip is in userspace.

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

KVM: Remove kvm_push_irq()

No longer used.

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

KVM: Remove inject_pending_vectors() callback

It is the same as inject_pending_irq() for VMX/SVM now.

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

KVM: Remove exception_injected() callback.

It always return false for VMX/SVM now.

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

KVM: SVM: Coalesce userspace/kernel irqchip interrupt injection logic

Start to use interrupt/exception queues like VMX does.
This also fix the bug that if exit was caused by a guest
internal exception access to IDT the exception was not
reinjected.

Use EVENTINJ to inject interrupts.  Use VINT only for detecting when IRQ
windows is open again.  EVENTINJ ensures
the interrupt is injected immediately and not delayed.

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

KVM: Use kvm_arch_interrupt_allowed() instead of checking interrupt_window_open directly

kvm_arch_interrupt_allowed() also checks IF so drop the check.

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

KVM: VMX: Cleanup vmx_intr_assist()

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

KVM: VMX: Consolidate userspace and kernel interrupt injection for VMX

Use the same callback to inject irq/nmi events no matter what irqchip is
in use. Only from VMX for now.

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

KVM: Make kvm_cpu_(has|get)_interrupt() work for userspace irqchip too

At the vector level, kernel and userspace irqchip are fairly similar.

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

KVM: MMU: Fix auditing code

Fix build breakage of hpa lookup in audit_mappings_page. Moreover, make
this function robust against shadow_notrap_nonpresent_pte entries.

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

KVM: ia64: ia64 vcpu_reset() do not call kmalloc() with irqs disabled

Restore local irq enabled state before calling kvm_arch_vcpu_init(),
which calls kmalloc(GFP_KERNEL).

Signed-off-by: Jes Sorensen <jes@sgi.com>
Acked-by: Xiantao Zhang <xiantao.zhang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: ia64: preserve int status through call to kvm_insert_vmm_mapping

Preserve interrupt status around call to kvm_insert_vmm_mappin()
in kvm_vcpu_pre_transition().

Signed-off-by: Jes Sorensen <jes@sgi.com>
Acked-by: Xiantao Zhang <xiantao.zhang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: ia64: restore irq state before calling kvm_vcpu_init

Make sure to restore the psr after calling kvm_insert_vmm_mapping()
which calls ia64_itr_entry() as it disables local interrupts and
kvm_vcpu_init() may sleep.

Avoids a warning from the lock debugging code.

Signed-off-by: Jes Sorensen <jes@sgi.com>
Acked-by : Xiantao Zhang <xiantao.zhang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: ia64: remove empty function vti_vcpu_load()

vti_vcpu_load() doesn't do anything, so lets get rid of it.

Signed-off-by: Jes Sorensen <jes@sgi.com>
Acked-by : Xiantao Zhang<xiantao.zhang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: ia64: Flush all TLBs once guest's memory mapping changes.

Flush all vcpu's TLB entries once changes guest's memory mapping.

Signed-off-by: Xiantao Zhang <xiantao.zhang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: x86: check for cr3 validity in ioctl_set_sregs

Matt T. Yourst notes that kvm_arch_vcpu_ioctl_set_sregs lacks validity
checking for the new cr3 value:

"Userspace callers of KVM_SET_SREGS can pass a bogus value of cr3 to
the kernel. This will trigger a NULL pointer access in gfn_to_rmap()
when userspace next tries to call KVM_RUN on the affected VCPU and kvm
attempts to activate the new non-existent page table root.

This happens since kvm only validates that cr3 points to a valid guest
physical memory page when code *inside* the guest sets cr3. However, kvm
currently trusts the userspace caller (e.g. QEMU) on the host machine to
always supply a valid page table root, rather than properly validating
it along with the rest of the reloaded guest state."

http://sourceforge.net/tracker/?func=detail&atid=893831&aid=2687641&group_id=180599

Check for a valid cr3 address in kvm_arch_vcpu_ioctl_set_sregs, triple
fault in case of failure.

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

KVM: ia64: Don't hold slots_lock in guest mode

Reorder locking to avoid holding the slots_lock when entering
the guest.

Signed-off-by: Jes Sorensen <jes@sgi.com>
Acked-by : Xiantao Zhang<xiantao.zhang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: Replace kvmclock open-coded get_cpu_var() with the real thing

Suggested by Ingo Molnar.

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

KVM: SVM: Skip instruction on a task switch only when appropriate

If a task switch was initiated because off a task gate in IDT and IDT
was accessed because of an external even the instruction should not
be skipped.

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

KVM: x86 emulator: Add new mode of instruction emulation: skip

In the new mode instruction is decoded, but not executed. The EIP
is moved to point after the instruction.

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

KVM: x86 emulator: Decode soft interrupt instructions

Do not emulate them yet.

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

KVM: x86 emulator: Completely decode in/out at decoding stage

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

KVM: x86 emulator: Add unsigned byte immediate decode

Extend "Source operand type" opcode description field to 4 bites
to accommodate new option.

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

KVM: x86 emulator: Complete decoding of call near in decode stage

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

KVM: x86 emulator: Complete short/near jcc decoding in decode stage

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

KVM: x86 emulator: Complete ljmp decoding at decode stage

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

KVM: x86 emulator: Add lcall decoding

No emulation yet.

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

KVM: x86 emulator: Add decoding of 16bit second immediate argument

Such as segment number in lcall/ljmp

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

KVM: MMU: remove global page optimization logic

Complexity to fix it not worthwhile the gains, as discussed
in http://article.gmane.org/gmane.comp.emulators.kvm.devel/28649.

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

KVM: Make kvm header C++ friendly

Two things needed fixing: 1) g++ does not allow a named structure type
within an anonymous union and 2) Avoid name clash between two padding
fields within the same struct by giving them different names as is
done elsewhere in the header.

Signed-off-by: Nathan Binkert <nate@binkert.org>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: PIT: fix count read and mode 0 handling

Commit 46ee278652f4cbd51013471b64c7897ba9bcd1b1 causes Solaris 10
to hang on boot.

Assuming that PIT counter reads should return 0 for an expired timer
is wrong: when it is active, the counter never stops (see comment on
__kpit_elapsed).

Also arm a one shot timer for mode 0.

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

KVM: ia64: make kvm depend on CONFIG_MODULES.

Since kvm-intel modue can't be built-in, make kvm depend on
CONFIG_MODULES.

Signed-off-by: Xiantao Zhang <xiantao.zhang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: x86 emulator: fix call near emulation

The length of pushed on to the stack return address depends on operand
size not address size.

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

KVM: MMU: Discard reserved bits checking on PDE bit 7-8

1. It's related to a Linux kernel bug which fixed by Ingo on
07a66d7c53a538e1a9759954a82bb6c07365eff9. The original code exists for quite a
long time, and it would convert a PDE for large page into a normal PDE. But it
fail to fit normal PDE well.  With the code before Ingo's fix, the kernel would
fall reserved bit checking with bit 8 - the remaining global bit of PTE. So the
kernel would receive a double-fault.

2. After discussion, we decide to discard PDE bit 7-8 reserved checking for now.
For this marked as reserved in SDM, but didn't checked by the processor in
fact...

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

KVM: Fix unneeded instruction skipping during task switching.

There is no need to skip instruction if the reason for a task switch
is a task gate in IDT and access to it is caused by an external even.
The problem  is currently solved only for VMX since there is no reliable
way to skip an instruction in SVM. We should emulate it instead.

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

KVM: Fix task switch back link handling.

Back link is written to a wrong TSS now.

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

KVM: VMX: Do not zero idt_vectoring_info in vmx_complete_interrupts().

We will need it later in task_switch().
Code in handle_exception() is dead. is_external_interrupt(vect_info)
will always be false since idt_vectoring_info is zeroed in
vmx_complete_interrupts().

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

KVM: VMX: Rewrite vmx_complete_interrupt()'s twisted maze of if() statements

...with a more straightforward switch().

Also fix a bug when NMI could be dropped on exit. Although this should
never happen in practice, since NMIs can only be injected, never triggered
internally by the guest like exceptions.

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

KVM: VMX: Fix handling of a fault during NMI unblocked due to IRET

Bit 12 is undefined in any of the following cases:
 If the VM exit sets the valid bit in the IDT-vectoring information field.
 If the VM exit is due to a double fault.

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

KVM: Use rsvd_bits_mask in load_pdptrs()

Also remove bit 5-6 from rsvd_bits_mask per latest SDM.

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

KVM: VMX: Fix feature testing

The testing of feature is too early now, before vmcs_config complete initialization.

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

KVM: VMX: Clean up Flex Priority related

And clean paranthes on returns.

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

KVM: remove pointless conditional before kfree() in lapic initialization

Remove pointless conditional before kfree().

Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: MMU: Use different shadows when EFER.NXE changes

A pte that is shadowed when the guest EFER.NXE=1 is not valid when
EFER.NXE=0; if bit 63 is set, the pte should cause a fault, and since the
shadow EFER always has NX enabled, this won't happen.

Fix by using a different shadow page table for different EFER.NXE bits.  This
allows vcpus to run correctly with different values of EFER.NXE, and for
transitions on this bit to be handled correctly without requiring a full
flush.

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

KVM: MMU: Emulate #PF error code of reserved bits violation

Detect, indicate, and propagate page faults where reserved bits are set.
Take care to handle the different paging modes, each of which has different
sets of reserved bits.

[avi: fix pte reserved bits for efer.nxe=0]

Signed-off-by: Eddie Dong <eddie.dong@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: ia64: enable external interrupt in vmm

Currently, the interrupt enable bit is cleared when in
the vmm.  This patch sets the bit and the external interrupts can
be dealt with when in the vmm.  This improves the I/O performance.

Signed-off-by: Yang Zhang <yang.zhang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: MMU: Fix comment in page_fault()

The original one is for the code before refactoring.

Signed-off-by: Yaozu (Eddie) Dong <eddie.dong@intel.com>
Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: VMX: Correct wrong vmcs field sizes

EXIT_QUALIFICATION and GUEST_LINEAR_ADDRESS are natural width, not 64-bit.

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

KVM: VMX: Make flexpriority module parameter reflect hardware capability

If the hardware does not support flexpriority, zero the module parameter.

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

KVM: Fix interrupt unhalting a vcpu when it shouldn't

kvm_vcpu_block() unhalts vpu on an interrupt/timer without checking
if interrupt window is actually opened.

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

KVM: Timer event should not unconditionally unhalt vcpu.

Currently timer events are processed before entering guest mode. Move it
to main vcpu event loop since timer events should be processed even while
vcpu is halted.  Timer may cause interrupt/nmi to be injected and only then
vcpu will be unhalted.

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

KVM: VMX: Fold vm_need_ept() into callers

Trivial.

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

KVM: VMX: Zero ept module parameter if ept is not present

Allows reading back hardware capability.

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

KVM: VMX: Zero the vpid module parameter if vpid is not supported

This allows reading back how the hardware is configured.

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

KVM: VMX: Annotate module parameters as __read_mostly

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

KVM: VMX: Simplify module parameter names

Instead of 'enable_vpid=1', use a simple 'vpid=1'.

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

KVM: VMX: Rename kvm_handle_exit() to vmx_handle_exit()

It is a static vmx-specific function.

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

KVM: VMX: Make module parameters readable

Useful to see how the module was loaded.

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

KVM: reuse (pop|push)_irq from svm.c in vmx.c

The prioritized bit vector manipulation functions are useful in both vmx and
svm.

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

KVM: SVM: Remove duplicate code in svm_do_inject_vector()

svm_do_inject_vector() reimplements pop_irq().

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

KVM: x86: Ignore reads to EVNTSEL MSRs

We ignore writes to the performance counters and performance event
selector registers already. Kaspersky antivirus reads the eventsel
MSR causing it to crash with the current behaviour.

Return 0 as data when the eventsel registers are read to stop the
crash.

Signed-off-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: MMU: do not free active mmu pages in free_mmu_pages()

free_mmu_pages() should only undo what alloc_mmu_pages() does.
Free mmu pages from the generic VM destruction function, kvm_destroy_vm().

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

KVM: Device assignment framework rework

After discussion with Marcelo, we decided to rework device assignment framework
together. The old problems are kernel logic is unnecessary complex. So Marcelo
suggest to split it into a more elegant way:

1. Split host IRQ assign and guest IRQ assign. And userspace determine the
combination. Also discard msi2intx parameter, userspace can specific
KVM_DEV_IRQ_HOST_MSI | KVM_DEV_IRQ_GUEST_INTX in assigned_irq->flags to
enable MSI to INTx convertion.

2. Split assign IRQ and deassign IRQ. Import two new ioctls:
KVM_ASSIGN_DEV_IRQ and KVM_DEASSIGN_DEV_IRQ.

This patch also fixed the reversed _IOR vs _IOW in definition(by deprecated the
old interface).

[avi: replace homemade bitcount() by hweight_long()]

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

KVM: make 'lapic_timer_ops' and 'kpit_ops' static

Fix this sparse warnings:
  arch/x86/kvm/lapic.c:916:22: warning: symbol 'lapic_timer_ops' was not declared. Should it be static?
  arch/x86/kvm/i8254.c:268:22: warning: symbol 'kpit_ops' was not declared. Should it be static?

Signed-off-by: Hannes Eder <hannes@hanneseder.net>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: ia64: Drop in SN2 replacement of fast path ITC emulation fault handler

Copy in SN2 RTC based ITC emulation for fast exit. The two versions
have the same size, so a dropin is simpler than patching the branch
instruction to hit the SN2 version.

Signed-off-by: Jes Sorensen <jes@sgi.com>
Acked-by: Xiantao Zhang <xiantao.zhang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: ia64: SN2 adjust emulated ITC frequency to match RTC frequency

On SN2 do not pass down the real ITC frequency, but rather patch the
values to match the SN2 RTC frequency.

Signed-off-by: Jes Sorensen <jes@sgi.com>
Acked-by: Xiantao Zhang <xiantao.zhang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: ia64: Create inline function kvm_get_itc() to centralize ITC reading.

Move all reading of special register 'AR_ITC' into two functions, one
in the kernel and one in the VMM module. When running on SN2, base the
result on the RTC rather the system ITC, as the ITC isn't
synchronized.

Signed-off-by: Jes Sorensen <jes@sgi.com>
Acked-by: Xiantao Zhang <xiantao.zhang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: ia64: Map in SN2 RTC registers to the VMM module

On SN2, map in the SN2 RTC registers to the VMM module, needed for ITC
emulation.

Signed-off-by: Jes Sorensen <jes@sgi.com>
Acked-by: Xiantao Zhang <xiantao.zhang@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: APIC: get rid of deliver_bitmask

Deliver interrupt during destination matching loop.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Acked-by: Xiantao Zhang <xiantao.zhang@intel.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: change the way how lowest priority vcpu is calculated

The new way does not require additional loop over vcpus to calculate
the one with lowest priority as one is chosen during delivery bitmap
construction.

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

KVM: consolidate ioapic/ipi interrupt delivery logic

Use kvm_apic_match_dest() in kvm_get_intr_delivery_bitmask() instead
of duplicating the same code. Use kvm_get_intr_delivery_bitmask() in
apic_send_ipi() to figure out ipi destination instead of reimplementing
the logic.

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

KVM: ioapic/msi interrupt delivery consolidation

ioapic_deliver() and kvm_set_msi() have code duplication. Move
the code into ioapic_deliver_entry() function and call it from
both places.

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

KVM: APIC: kvm_apic_set_irq deliver all kinds of interrupts

Get rid of ioapic_inj_irq() and ioapic_inj_nmi() functions.

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