KVM: Choose better candidate for directed yield

Currently, on a large vcpu guests, there is a high probability of
yielding to the same vcpu who had recently done a pause-loop exit or
cpu relax intercepted. Such a yield can lead to the vcpu spinning
again and hence degrade the performance.

The patchset keeps track of the pause loop exit/cpu relax interception
and gives chance to a vcpu which:
 (a) Has not done pause loop exit or cpu relax intercepted at all
     (probably he is preempted lock-holder)
 (b) Was skipped in last iteration because it did pause loop exit or
     cpu relax intercepted, and probably has become eligible now
     (next eligible lock holder)

Signed-off-by: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com> # on s390x
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: Note down when cpu relax intercepted or pause loop exited

Noting pause loop exited vcpu or cpu relax intercepted helps in
filtering right candidate to yield. Wrong selection of vcpu;
i.e., a vcpu that just did a pl-exit or cpu relax intercepted may
contribute to performance degradation.

Signed-off-by: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com> # on s390x
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: Add config to support ple or cpu relax optimzation

Suggested-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com> # on s390x
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: switch to symbolic name for irq_states size

Use PIC_NUM_PINS instead of hard-coded 16 for pic pins.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: x86: Fix typos in pmu.c

Signed-off-by: Guo Chao <yan@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: x86: Fix typos in lapic.c

Signed-off-by: Guo Chao <yan@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: x86: Fix typos in cpuid.c

Signed-off-by: Guo Chao <yan@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: x86: Fix typos in emulate.c

Signed-off-by: Guo Chao <yan@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: x86: Fix typos in x86.c

Signed-off-by: Guo Chao <yan@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: SVM: Fix typos

Signed-off-by: Guo Chao <yan@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: VMX: Fix typos

Signed-off-by: Guo Chao <yan@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: remove the unused parameter of gfn_to_pfn_memslot

The parameter, 'kvm', is not used in gfn_to_pfn_memslot, we can happily remove
it

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: remove is_error_hpa

Remove them since they are not used anymore

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: make bad_pfn static to kvm_main.c

bad_pfn is not used out of kvm_main.c, so mark it static, also move it near
hwpoison_pfn and fault_pfn

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: using get_fault_pfn to get the fault pfn

Using get_fault_pfn to cleanup the code

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: MMU: track the refcount when unmap the page

It will trigger a WARN_ON if the page has been freed but it is still
used in mmu, it can help us to detect mm bug early

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: x86: remove unnecessary mark_page_dirty

fix:
[  132.474633] 3.5.0-rc1+ #50 Not tainted
[  132.474634] -------------------------------
[  132.474635] include/linux/kvm_host.h:369 suspicious rcu_dereference_check() usage!
[  132.474636]
[  132.474636] other info that might help us debug this:
[  132.474636]
[  132.474638]
[  132.474638] rcu_scheduler_active = 1, debug_locks = 1
[  132.474640] 1 lock held by qemu-kvm/2832:
[  132.474657]  #0:  (&vcpu->mutex){+.+.+.}, at: [<ffffffffa01e1636>] vcpu_load+0x1e/0x91 [kvm]
[  132.474658]
[  132.474658] stack backtrace:
[  132.474660] Pid: 2832, comm: qemu-kvm Not tainted 3.5.0-rc1+ #50
[  132.474661] Call Trace:
[  132.474665]  [<ffffffff81092f40>] lockdep_rcu_suspicious+0xfc/0x105
[  132.474675]  [<ffffffffa01e0c85>] kvm_memslots+0x6d/0x75 [kvm]
[  132.474683]  [<ffffffffa01e0ca1>] gfn_to_memslot+0x14/0x4c [kvm]
[  132.474693]  [<ffffffffa01e3575>] mark_page_dirty+0x17/0x2a [kvm]
[  132.474706]  [<ffffffffa01f21ea>] kvm_arch_vcpu_ioctl+0xbcf/0xc07 [kvm]

Actually, we do not write vcpu->arch.time at this time, mark_page_dirty
should be removed.

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: MMU: Avoid handling same rmap_pde in kvm_handle_hva_range()

When we invalidate a THP page, we call the handler with the same
rmap_pde argument 512 times in the following loop:

  for each guest page in the range
    for each level
      unmap using rmap

This patch avoids these extra handler calls by changing the loop order
like this:

  for each level
    for each rmap in the range
      unmap using rmap

With the preceding patches in the patch series, this made THP page
invalidation more than 5 times faster on our x86 host: the host became
more responsive during swapping the guest's memory as a result.

Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: MMU: Push trace_kvm_age_page() into kvm_age_rmapp()

This restricts the tracing to page aging and makes it possible to
optimize kvm_handle_hva_range() further in the following patch.

Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: MMU: Add memslot parameter to hva handlers

This is needed to push trace_kvm_age_page() into kvm_age_rmapp() in the
following patch.

Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: Separate rmap_pde from kvm_lpage_info->write_count

This makes it possible to loop over rmap_pde arrays in the same way as
we do over rmap so that we can optimize kvm_handle_hva_range() easily in
the following patch.

Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: Introduce kvm_unmap_hva_range() for kvm_mmu_notifier_invalidate_range_start()

When we tested KVM under memory pressure, with THP enabled on the host,
we noticed that MMU notifier took a long time to invalidate huge pages.

Since the invalidation was done with mmu_lock held, it not only wasted
the CPU but also made the host harder to respond.

This patch mitigates this by using kvm_handle_hva_range().

Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Cc: Alexander Graf <agraf@suse.de>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: MMU: Make kvm_handle_hva() handle range of addresses

When guest's memory is backed by THP pages, MMU notifier needs to call
kvm_unmap_hva(), which in turn leads to kvm_handle_hva(), in a loop to
invalidate a range of pages which constitute one huge page:

  for each page
    for each memslot
      if page is in memslot
        unmap using rmap

This means although every page in that range is expected to be found in
the same memslot, we are forced to check unrelated memslots many times.
If the guest has more memslots, the situation will become worse.

Furthermore, if the range does not include any pages in the guest's
memory, the loop over the pages will just consume extra time.

This patch, together with the following patches, solves this problem by
introducing kvm_handle_hva_range() which makes the loop look like this:

  for each memslot
    for each page in memslot
      unmap using rmap

In this new processing, the actual work is converted to a loop over rmap
which is much more cache friendly than before.

Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Cc: Alexander Graf <agraf@suse.de>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: Introduce hva_to_gfn_memslot() for kvm_handle_hva()

This restricts hva handling in mmu code and makes it easier to extend
kvm_handle_hva() so that it can treat a range of addresses later in this
patch series.

Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Cc: Alexander Graf <agraf@suse.de>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: MMU: Use __gfn_to_rmap() to clean up kvm_handle_hva()

We can treat every level uniformly.

Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

Revert "apic: fix kvm build on UP without IOAPIC"

This reverts commit f9808b7fd422b965cea52e05ba470e0a473c53d3.
After commit 'kvm: switch to apic_set_eoi_write, apic_write'
the stubs are no longer needed as kvm does not look at apicdrivers anymore.

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

KVM guest: switch to apic_set_eoi_write, apic_write

Use apic_set_eoi_write, apic_write to avoid meedling in core apic
driver data structures directly.

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

apic: add apic_set_eoi_write for PV use

KVM PV EOI optimization overrides eoi_write apic op with its own
version. Add an API for this to avoid meddling with core x86 apic driver
data structures directly.

For KVM use, we don't need any guarantees about when the switch to the
new op will take place, so it could in theory use this API after SMP init,
but it currently doesn't, and restricting callers to early init makes it
clear that it's safe as it won't race with actual APIC driver use.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Avi Kivity <avi@redhat.com>

Merge branch 'for-upstream' of git://github.com/agraf/linux-2.6 into next

ppc queue from Alex Graf:

 * Prepare some of the booke code for 64 bit support
 * BookE: Fix ESR flag in DSI
 * BookE: Add rfci emulation

* 'for-upstream' of git://github.com/agraf/linux-2.6:
  KVM: PPC: Critical interrupt emulation support
  KVM: PPC: e500mc: Fix tlbilx emulation for 64-bit guests
  KVM: PPC64: booke: Set interrupt computation mode for 64-bit host
  KVM: PPC: bookehv: Add ESR flag to Data Storage Interrupt
  KVM: PPC: bookehv64: Add support for std/ld emulation.
  booke: Added crit/mc exception handler for e500v2
  booke/bookehv: Add host crit-watchdog exception support

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

KVM: VMX: Implement PCID/INVPCID for guests with EPT

This patch handles PCID/INVPCID for guests.

Process-context identifiers (PCIDs) are a facility by which a logical processor
may cache information for multiple linear-address spaces so that the processor
may retain cached information when software switches to a different linear
address space. Refer to section 4.10.1 in IA32 Intel Software Developer's Manual
Volume 3A for details.

For guests with EPT, the PCID feature is enabled and INVPCID behaves as running
natively.
For guests without EPT, the PCID feature is disabled and INVPCID triggers #UD.

Signed-off-by: Junjie Mao <junjie.mao@intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: Add x86_hyper_kvm to complete detect_hypervisor_platform check

While debugging I noticed that unlike all the other hypervisor code in the
kernel, kvm does not have an entry for x86_hyper which is used in
detect_hypervisor_platform() which results in a nice printk in the
syslog.  This is only really a stub function but it
does make kvm more consistent with the other hypervisors.

Signed-off-by: Prarit Bhargava <prarit@redhat.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: Gleb Natapov <gleb@redhat.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Marcelo Tostatti <mtosatti@redhat.com>
Cc: kvm@vger.kernel.org
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: PPC: Critical interrupt emulation support

rfci instruction and CSRR0/1 registers are emulated.

Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Stuart Yoder <stuart.yoder@freescale.com>
Signed-off-by: Bharat Bhushan <bharat.bhushan@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>

KVM: PPC: e500mc: Fix tlbilx emulation for 64-bit guests

tlbilxva emulation was using an u32 variable for guest effective address.
Replace it with gva_t type to handle 64-bit guests.

Signed-off-by: Mihai Caraman <mihai.caraman@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>

KVM: PPC64: booke: Set interrupt computation mode for 64-bit host

64-bit host needs to remain in 64-bit mode when an exception take place.
Set interrupt computaion mode in EPCR register.

Signed-off-by: Mihai Caraman <mihai.caraman@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>

KVM: PPC: bookehv: Add ESR flag to Data Storage Interrupt

ESR register is required by Data Storage Interrupt handling code.
Add the specific flag to the interrupt handler.

Signed-off-by: Mihai Caraman <mihai.caraman@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>

KVM: PPC: bookehv64: Add support for std/ld emulation.

Add support for std/ld emulation.

Signed-off-by: Varun Sethi <Varun.Sethi@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>

booke: Added crit/mc exception handler for e500v2

Watchdog is taken at critical exception level. So this patch
is tested with host watchdog exception happening when guest
is running.

Signed-off-by: Bharat Bhushan <bharat.bhushan@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>

booke/bookehv: Add host crit-watchdog exception support

Signed-off-by: Bharat Bhushan <bharat.bhushan@freescale.com>
Signed-off-by: Alexander Graf <agraf@suse.de>

KVM: MMU: document mmu-lock and fast page fault

Document fast page fault and mmu-lock in locking.txt

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: MMU: fix kvm_mmu_pagetable_walk tracepoint

The P bit of page fault error code is missed in this tracepoint, fix it by
passing the full error code

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: MMU: trace fast page fault

To see what happen on this path and help us to optimize it

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: MMU: fast path of handling guest page fault

If the the present bit of page fault error code is set, it indicates
the shadow page is populated on all levels, it means what we do is
only modify the access bit which can be done out of mmu-lock

Currently, in order to simplify the code, we only fix the page fault
caused by write-protect on the fast path

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: MMU: introduce SPTE_MMU_WRITEABLE bit

This bit indicates whether the spte can be writable on MMU, that means
the corresponding gpte is writable and the corresponding gfn is not
protected by shadow page protection

In the later path, SPTE_MMU_WRITEABLE will indicates whether the spte
can be locklessly updated

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: MMU: fold tlb flush judgement into mmu_spte_update

mmu_spte_update() is the common function, we can easily audit the path

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: VMX: export PFEC.P bit on ept

Export the present bit of page fault error code, the later patch
will use it

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: MMU: cleanup spte_write_protect

Use __drop_large_spte to cleanup this function and comment spte_write_protect

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: MMU: abstract spte write-protect

Introduce a common function to abstract spte write-protect to
cleanup the code

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: MMU: return bool in __rmap_write_protect

The reture value of __rmap_write_protect is either 1 or 0, use
true/false instead of these

Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: VMX: Emulate invalid guest state by default

Our emulation should be complete enough that we can emulate guests
while they are in big real mode, or in a mode transition that is not
virtualizable without unrestricted guest support.

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

KVM: x86 emulator: implement LTR

Opcode 0F 00 /3.  Encountered during Windows XP secondary processor bringup.

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

KVM: x86 emulator: make loading TR set the busy bit

Guest software doesn't actually depend on it, but vmx will refuse us
entry if we don't.  Set the bit in both the cached segment and memory,
just to be nice.

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

KVM: x86 emulator: make read_segment_descriptor() return the address

Some operations want to modify the descriptor later on, so save the
address for future use.

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

KVM: x86 emulator: emulate LLDT

Opcode 0F 00 /2. Used by isolinux durign the protected mode transition.

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

KVM: x86 emulator: emulate BSWAP

Opcodes 0F C8 - 0F CF.

Used by the SeaBIOS cdrom code (though not in big real mode).

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

KVM: VMX: Improve error reporting during invalid guest state emulation

If instruction emulation fails, report it properly to userspace.

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

KVM: VMX: Stop invalid guest state emulation on pending event

Process the event, possibly injecting an interrupt, before continuing.

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

KVM: x86 emulator: implement ENTER

Opcode C8.

Only ENTER with lexical nesting depth 0 is implemented, since others are
very rare.  We'll fail emulation if nonzero lexical depth is used so data
is not corrupted.

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

KVM: x86 emulator: split push logic from push opcode emulation

This allows us to reuse the code without populating ctxt->src and
overriding ctxt->op_bytes.

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

KVM: x86 emulator: fix byte-sized MOVZX/MOVSX

Commit 2adb5ad9fe1 removed ByteOp from MOVZX/MOVSX, replacing them by
SrcMem8, but neglected to fix the dependency in the emulation code
on ByteOp.  This caused the instruction not to have any effect in
some circumstances.

Fix by replacing the check for ByteOp with the equivalent src.op_bytes == 1.

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

KVM: x86 emulator: emulate LAHF

Opcode 9F.

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

KVM: VMX: Continue emulating after batch exhausted

If we return early from an invalid guest state emulation loop, make
sure we return to it later if the guest state is still invalid.

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

KVM: VMX: Fix interrupt exit condition during emulation

Checking EFLAGS.IF is incorrect as we might be in interrupt shadow.  If
that is the case, the main loop will notice that and not inject the interrupt,
causing an endless loop.

Fix by using vmx_interrupt_allowed() to check if we can inject an interrupt
instead.

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

KVM: x86 emulator: emulate SGDT/SIDT

Opcodes 0F 01 /0 and 0F 01 /1

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

KVM: Fix SS default ESP/EBP based addressing

We correctly default to SS when BP is used as a base in 16-bit address mode,
but we don't do that for 32-bit mode.

Fix by adjusting the default to SS when either ESP or EBP is used as the base
register.

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

KVM: x86 emulator: initialize memop

memop is not initialized; this can lead to a two-byte operation
following a 4-byte operation to see garbage values.  Usually
truncation fixes things fot us later on, but at least in one case
(call abs) it doesn't.

Fix by moving memop to the auto-initialized field area.

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

KVM: x86 emulator: emulate LEAVE

Opcode c9; used by some variants of Windows during boot, in big real mode.

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

KVM: VMX: Limit iterations with emulator_invalid_guest_state

Otherwise, if the guest ends up looping, we never exit the srcu critical
section, which causes synchronize_srcu() to hang.

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

KVM: VMX: Relax check on unusable segment

Some userspace (e.g. QEMU 1.1) munge the d and g bits of segment
descriptors, causing us not to recognize them as unusable segments
with emulate_invalid_guest_state=1.  Relax the check by testing for
segment not present (a non-present segment cannot be usable).

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

KVM: x86 emulator: fix LIDT/LGDT in long mode

The operand size for these instructions is 8 bytes in long mode, even without
a REX prefix.  Set it explicitly.

Triggered while booting Linux with emulate_invalid_guest_state=1.

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

KVM: x86 emulator: allow loading null SS in long mode

Null SS is valid in long mode; allow loading it.

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

KVM: x86 emulator: emulate cpuid

Opcode 0F A2.

Used by Linux during the mode change trampoline while in a state that is
not virtualizable on vmx without unrestricted_guest, so we need to emulate
it is emulate_invalid_guest_state=1.

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

KVM: x86 emulator: change ->get_cpuid() accessor to use the x86 semantics

Instead of getting an exact leaf, follow the spec and fall back to the last
main leaf instead.  This lets us easily emulate the cpuid instruction in the
emulator.

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

KVM: Split cpuid register access from computation

Introduce kvm_cpuid() to perform the leaf limit check and calculate
register values, and let kvm_emulate_cpuid() just handle reading and
writing the registers from/to the vcpu.  This allows us to reuse
kvm_cpuid() in a context where directly reading and writing registers
is not desired.

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

KVM: VMX: Return correct CPL during transition to protected mode

In protected mode, the CPL is defined as the lower two bits of CS, as set by
the last far jump.  But during the transition to protected mode, there is no
last far jump, so we need to return zero (the inherited real mode CPL).

Fix by reading CPL from the cache during the transition.  This isn't 100%
correct since we don't set the CPL cache on a far jump, but since protected
mode transition will always jump to a segment with RPL=0, it will always
work.

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

KVM: MMU: Force cr3 reload with two dimensional paging on mov cr3 emulation

Currently the MMU's ->new_cr3() callback does nothing when guest paging
is disabled or when two-dimentional paging (e.g. EPT on Intel) is active.
This means that an emulated write to cr3 can be lost; kvm_set_cr3() will
write vcpu-arch.cr3, but the GUEST_CR3 field in the VMCS will retain its
old value and this is what the guest sees.

This bug did not have any effect until now because:
- with unrestricted guest, or with svm, we never emulate a mov cr3 instruction
- without unrestricted guest, and with paging enabled, we also never emulate a
  mov cr3 instruction
- without unrestricted guest, but with paging disabled, the guest's cr3 is
  ignored until the guest enables paging; at this point the value from arch.cr3
  is loaded correctly my the mov cr0 instruction which turns on paging

However, the patchset that enables big real mode causes us to emulate mov cr3
instructions in protected mode sometimes (when guest state is not virtualizable
by vmx); this mov cr3 is effectively ignored and will crash the guest.

The fix is to make nonpaging_new_cr3() call mmu_free_roots() to force a cr3
reload.  This is awkward because now all the new_cr3 callbacks to the same
thing, and because mmu_free_roots() is somewhat of an overkill; but fixing
that is more complicated and will be done after this minimal fix.

Observed in the Window XP 32-bit installer while bringing up secondary vcpus.

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

KVM: handle last_boosted_vcpu = 0 case

If last_boosted_vcpu == 0, then we fall through all test cases and
may end up with all VCPUs pouncing on vcpu 0.  With a large enough
guest, this can result in enormous runqueue lock contention, which
can prevent vcpu0 from running, leading to a livelock.

Changing < to <= makes sure we properly handle that case.

Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: s390: Fix sigp sense handling.

If sigp sense doesn't have any status bits to report, it should set
cc 0 and leave the register as-is.

Since we know about the external call pending bit, we should report
it if it is set as well.

Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: s390: use sigp condition code defines

Just use the defines instead of using plain numbers and adding
a comment behind each line.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: s390: fix sigp set prefix status stored cases

If an invalid parameter is passed or the addressed cpu is in an
incorrect state sigp set prefix will store a status.
This status must only have bits set as defined by the architecture.
The current kvm implementation missed to clear bits and also did
not set the intended status bit ("and" instead of "or" operation).

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: s390: fix sigp sense running condition code handling

Only if the sensed cpu is not running a status is stored, which
is reflected by condition code 1. If the cpu is running, condition
code 0 should be returned.
Just the opposite of what the code is doing.

Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

s390/smp/kvm: unifiy sigp definitions

The smp and the kvm code have different defines for the sigp order codes.
Let's just have a single place where these are defined.
Also move the sigp condition code and sigp cpu status bits to the new
sigp.h header file.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

s390/smp: remove redundant check

condition code "status stored" for sigp sense running always implies
that only the "not running" status bit is set. Therefore no need to
check if it is set.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: Guard mmu_notifier specific code with CONFIG_MMU_NOTIFIER

In order to avoid compilation failure when KVM is not compiled in,
guard the mmu_notifier specific sections with both CONFIG_MMU_NOTIFIER
and KVM_ARCH_WANT_MMU_NOTIFIER, like it is being done in the rest of
the KVM code.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <c.dall@virtualopensystems.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: VMX: code clean for vmx_init()

Signed-off-by: Guo Chao <yan@linux.vnet.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

apic: fix kvm build on UP without IOAPIC

On UP i386, when APIC is disabled
# CONFIG_X86_UP_APIC is not set
# CONFIG_PCI_IOAPIC is not set

code looking at apicdrivers never has any effect but it
still gets compiled in. In particular, this causes
build failures with kvm, but it generally bloats the kernel
unnecessarily.

Fix by defining both __apicdrivers and __apicdrivers_end
to be NULL when CONFIG_X86_LOCAL_APIC is unset: I verified
that as the result any loop scanning __apicdrivers gets optimized out by
the compiler.

Warning: a .config with apic disabled doesn't seem to boot
for me (even without this patch). Still verifying why,
meanwhile this patch is compile-tested only.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reported-by: Randy Dunlap <rdunlap@xenotime.net>
Acked-by: Randy Dunlap <rdunlap@xenotime.net>
Acked-by: H. Peter Anvin <hpa@linux.intel.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: host side for eoi optimization

Implementation of PV EOI using shared memory.
This reduces the number of exits an interrupt
causes as much as by half.

The idea is simple: there's a bit, per APIC, in guest memory,
that tells the guest that it does not need EOI.
We set it before injecting an interrupt and clear
before injecting a nested one. Guest tests it using
a test and clear operation - this is necessary
so that host can detect interrupt nesting -
and if set, it can skip the EOI MSR.

There's a new MSR to set the address of said register
in guest memory. Otherwise not much changed:
- Guest EOI is not required
- Register is tested & ISR is automatically cleared on exit

For testing results see description of previous patch
'kvm_para: guest side for eoi avoidance'.

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

KVM: rearrange injection cancelling code

Each time we need to cancel injection we invoke same code
(cancel_injection callback).  Move it towards the end of function using
the familiar goto on error pattern.

Will make it easier to do more cleanups for PV EOI.

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

KVM: only sync when attention bits set

Commit eb0dc6d0368072236dcd086d7fdc17fd3c4574d4 introduced apic
attention bitmask but kvm still syncs lapic unconditionally.
As that commit suggested and in anticipation of adding more attention
bits, only sync lapic if(apic_attention).

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

KVM: eoi msi documentation

Document the new EOI MSR. Couldn't decide whether this change belongs
conceptually on guest or host side, so a separate patch.

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

x86, bitops: note on __test_and_clear_bit atomicity

__test_and_clear_bit is actually atomic with respect
to the local CPU. Add a note saying that KVM on x86
relies on this behaviour so people don't accidentaly break it.
Also warn not to rely on this in portable code.

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

KVM guest: guest side for eoi avoidance

The idea is simple: there's a bit, per APIC, in guest memory,
that tells the guest that it does not need EOI.
Guest tests it using a single est and clear operation - this is
necessary so that host can detect interrupt nesting - and if set, it can
skip the EOI MSR.

I run a simple microbenchmark to show exit reduction
(note: for testing, need to apply follow-up patch
'kvm: host side for eoi optimization' + a qemu patch
 I posted separately, on host):

Before:

Performance counter stats for 'sleep 1s':

            47,357 kvm:kvm_entry                                                [99.98%]
                 0 kvm:kvm_hypercall                                            [99.98%]
                 0 kvm:kvm_hv_hypercall                                         [99.98%]
             5,001 kvm:kvm_pio                                                  [99.98%]
                 0 kvm:kvm_cpuid                                                [99.98%]
            22,124 kvm:kvm_apic                                                 [99.98%]
            49,849 kvm:kvm_exit                                                 [99.98%]
            21,115 kvm:kvm_inj_virq                                             [99.98%]
                 0 kvm:kvm_inj_exception                                        [99.98%]
                 0 kvm:kvm_page_fault                                           [99.98%]
            22,937 kvm:kvm_msr                                                  [99.98%]
                 0 kvm:kvm_cr                                                   [99.98%]
                 0 kvm:kvm_pic_set_irq                                          [99.98%]
                 0 kvm:kvm_apic_ipi                                             [99.98%]
            22,207 kvm:kvm_apic_accept_irq                                      [99.98%]
            22,421 kvm:kvm_eoi                                                  [99.98%]
                 0 kvm:kvm_pv_eoi                                               [99.99%]
                 0 kvm:kvm_nested_vmrun                                         [99.99%]
                 0 kvm:kvm_nested_intercepts                                    [99.99%]
                 0 kvm:kvm_nested_vmexit                                        [99.99%]
                 0 kvm:kvm_nested_vmexit_inject                                    [99.99%]
                 0 kvm:kvm_nested_intr_vmexit                                    [99.99%]
                 0 kvm:kvm_invlpga                                              [99.99%]
                 0 kvm:kvm_skinit                                               [99.99%]
                57 kvm:kvm_emulate_insn                                         [99.99%]
                 0 kvm:vcpu_match_mmio                                          [99.99%]
                 0 kvm:kvm_userspace_exit                                       [99.99%]
                 2 kvm:kvm_set_irq                                              [99.99%]
                 2 kvm:kvm_ioapic_set_irq                                       [99.99%]
            23,609 kvm:kvm_msi_set_irq                                          [99.99%]
                 1 kvm:kvm_ack_irq                                              [99.99%]
               131 kvm:kvm_mmio                                                 [99.99%]
               226 kvm:kvm_fpu                                                  [100.00%]
                 0 kvm:kvm_age_page                                             [100.00%]
                 0 kvm:kvm_try_async_get_page                                    [100.00%]
                 0 kvm:kvm_async_pf_doublefault                                    [100.00%]
                 0 kvm:kvm_async_pf_not_present                                    [100.00%]
                 0 kvm:kvm_async_pf_ready                                       [100.00%]
                 0 kvm:kvm_async_pf_completed

       1.002100578 seconds time elapsed

After:

 Performance counter stats for 'sleep 1s':

            28,354 kvm:kvm_entry                                                [99.98%]
                 0 kvm:kvm_hypercall                                            [99.98%]
                 0 kvm:kvm_hv_hypercall                                         [99.98%]
             1,347 kvm:kvm_pio                                                  [99.98%]
                 0 kvm:kvm_cpuid                                                [99.98%]
             1,931 kvm:kvm_apic                                                 [99.98%]
            29,595 kvm:kvm_exit                                                 [99.98%]
            24,884 kvm:kvm_inj_virq                                             [99.98%]
                 0 kvm:kvm_inj_exception                                        [99.98%]
                 0 kvm:kvm_page_fault                                           [99.98%]
             1,986 kvm:kvm_msr                                                  [99.98%]
                 0 kvm:kvm_cr                                                   [99.98%]
                 0 kvm:kvm_pic_set_irq                                          [99.98%]
                 0 kvm:kvm_apic_ipi                                             [99.99%]
            25,953 kvm:kvm_apic_accept_irq                                      [99.99%]
            26,132 kvm:kvm_eoi                                                  [99.99%]
            26,593 kvm:kvm_pv_eoi                                               [99.99%]
                 0 kvm:kvm_nested_vmrun                                         [99.99%]
                 0 kvm:kvm_nested_intercepts                                    [99.99%]
                 0 kvm:kvm_nested_vmexit                                        [99.99%]
                 0 kvm:kvm_nested_vmexit_inject                                    [99.99%]
                 0 kvm:kvm_nested_intr_vmexit                                    [99.99%]
                 0 kvm:kvm_invlpga                                              [99.99%]
                 0 kvm:kvm_skinit                                               [99.99%]
               284 kvm:kvm_emulate_insn                                         [99.99%]
                68 kvm:vcpu_match_mmio                                          [99.99%]
                68 kvm:kvm_userspace_exit                                       [99.99%]
                 2 kvm:kvm_set_irq                                              [99.99%]
                 2 kvm:kvm_ioapic_set_irq                                       [99.99%]
            28,288 kvm:kvm_msi_set_irq                                          [99.99%]
                 1 kvm:kvm_ack_irq                                              [99.99%]
               131 kvm:kvm_mmio                                                 [100.00%]
               588 kvm:kvm_fpu                                                  [100.00%]
                 0 kvm:kvm_age_page                                             [100.00%]
                 0 kvm:kvm_try_async_get_page                                    [100.00%]
                 0 kvm:kvm_async_pf_doublefault                                    [100.00%]
                 0 kvm:kvm_async_pf_not_present                                    [100.00%]
                 0 kvm:kvm_async_pf_ready                                       [100.00%]
                 0 kvm:kvm_async_pf_completed

       1.002039622 seconds time elapsed

We see that # of exits is almost halved.

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

KVM: optimize ISR lookups

We perform ISR lookups twice: during interrupt
injection and on EOI. Typical workloads only have
a single bit set there. So we can avoid ISR scans by
1. counting bits as we set/clear them in ISR
2. on set, caching the injected vector number
3. on clear, invalidating the cache

The real purpose of this is enabling PV EOI
which needs to quickly validate the vector.
But non PV guests also benefit: with this patch,
and without interrupt nesting, apic_find_highest_isr
will always return immediately without scanning ISR.

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

KVM: document lapic regs field

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

KVM: Use kvm_kvfree() to free memory allocated by kvm_kvzalloc()

The following commit did not care about the error handling path:

  commit c1a7b32a14138f908df52d7c53b5ce3415ec6b50
  KVM: Avoid wasting pages for small lpage_info arrays

If memory allocation fails, vfree() will be called with the address
returned by kzalloc().  This patch fixes this issue.

Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: Introduce __KVM_HAVE_IRQ_LINE

This is a preparatory patch for the KVM/ARM implementation. KVM/ARM will use
the KVM_IRQ_LINE ioctl, which is currently conditional on
__KVM_HAVE_IOAPIC, but ARM obviously doesn't have any IOAPIC support and we
need a separate define.

Signed-off-by: Christoffer Dall <c.dall@virtualopensystems.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: use KVM_CAP_IRQ_ROUTING to protect the routing related code

The KVM code sometimes uses CONFIG_HAVE_KVM_IRQCHIP to protect
code that is related to IRQ routing, which not all in-kernel
irqchips may support.

Use KVM_CAP_IRQ_ROUTING instead.

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <c.dall@virtualopensystems.com>
Signed-off-by: Avi Kivity <avi@redhat.com>

KVM: trace events: update list of exit reasons

The list of exit reasons for the kvm_userspace_exit event was
missing recent additions; bring it into sync again.

Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: s390: Perform early event mask processing during boot

For processing under KVM it is required to detect
the actual SCLP console type in order to set it as
preferred console.

Signed-off-by: Heinz Graalfs <graalfs@linux.vnet.ibm.com>
Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Acked-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: s390: Set CPU in stopped state on initial cpu reset

The initial cpu reset sets the cpu in the stopped state.
Several places check for the cpu state (e.g. sigp set prefix) and
not setting the STOPPED state triggered errors with newer guest
kernels after reboot.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

KVM: x86: change PT_FIRST_AVAIL_BITS_SHIFT to avoid conflict with EPT Dirty bit

EPT Dirty bit use bit 9 as Intel SDM definition, to avoid conflict, change
PT_FIRST_AVAIL_BITS_SHIFT to 10.

Signed-off-by: Xudong Hao <xudong.hao@intel.com>
Signed-off-by: Xiantao Zhang <xiantao.zhang@intel.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>