When not running in guest-debug mode (i.e. the guest controls the debug
registers, having to take an exit for each DR access is a waste of time.
If the guest gets into a state where each context switch causes DR to be
saved and restored, this can take away as much as 40% of the execution
time from the guest.

If the guest is running with vcpu->arch.db == vcpu->arch.eff_db, we
can let it write freely to the debug registers and reload them on the
next exit.  We still need to exit on the first access, so that the
KVM_DEBUGREG_WONT_EXIT flag is set in switch_db_regs; after that, further
accesses to the debug registers will not cause a vmexit.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

When not running in guest-debug mode, the guest controls the debug
registers and having to take an exit for each DR access is a waste
of time.  If the guest gets into a state where each context switch
causes DR to be saved and restored, this can take away as much as 40%
of the execution time from the guest.

After this patch, VMX- and SVM-specific code can set a flag in
switch_db_regs, telling vcpu_enter_guest that on the next exit the debug
registers might be dirty and need to be reloaded (syncing will be taken
care of by a new callback in kvm_x86_ops).  This flag can be set on the
first access to a debug registers, so that multiple accesses to the
debug registers only cause one vmexit.

Note that since the guest will be able to read debug registers and
enable breakpoints in DR7, we need to ensure that they are synchronized
on entry to the guest---including DR6 that was not synced before.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

The next patch will add another bit that we can test with the
same "if".
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Currently, this works even if the bit is not in "min", because the bit is always
set in MSR_IA32_VMX_ENTRY_CTLS.  Mention it for the sake of documentation, and
to avoid surprises if we later switch to MSR_IA32_VMX_TRUE_ENTRY_CTLS.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

It's no longer possible to enter enable_irq_window in guest mode when
L1 intercepts external interrupts and we are entering L2. This is now
caught in vcpu_enter_guest. So we can remove the check from the VMX
version of enable_irq_window, thus the need to return an error code from
both enable_irq_window and enable_nmi_window.
Signed-off-by: NJan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

According to SDM 27.2.3, IDT vectoring information will not be valid on
vmexits caused by external NMIs. So we have to avoid creating such
scenarios by delaying EXIT_REASON_EXCEPTION_NMI injection as long as we
have a pending interrupt because that one would be migrated to L1's IDT
vectoring info on nested exit.
Signed-off-by: NJan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

We cannot rely on the hardware-provided preemption timer support because
we are holding L2 in HLT outside non-root mode. Furthermore, emulating
the preemption will resolve tick rate errata on older Intel CPUs.

The emulation is based on hrtimer which is started on L2 entry, stopped
on L2 exit and evaluated via the new check_nested_events hook. As we no
longer rely on hardware features, we can enable both the preemption
timer support and value saving unconditionally.
Signed-off-by: NJan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Move the check for leaving L2 on pending and intercepted IRQs or NMIs
from the *_allowed handler into a dedicated callback. Invoke this
callback at the relevant points before KVM checks if IRQs/NMIs can be
injected. The callback has the task to switch from L2 to L1 if needed
and inject the proper vmexit events.

The rework fixes L2 wakeups from HLT and provides the foundation for
preemption timer emulation.
Signed-off-by: NJan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Merge tag 'kvm-s390-20140306' of git://git.kernel.org/pub/scm/linux/kernel/git/kvms390/linux into kvm-next

One fix for virtio-ccw, fixing a problem introduced with
"virtio_ccw: fix vcdev pointer handling issues" and noticed just
after it went into git.

During set offline processing virtio_grab_drvdata() incorrectly
calls dev_set_drvdata() to remove the virtio_ccw_device from the
parent ccw_device's driver data. This is wrong and ends up in a
hang during virtio_ccw_reset(), as the interrupt handler still
has need of the virtio_ccw_device.

A new field 'going_away' is introduced in struct virtio_ccw_device
to control the usage of the ccw_device's driver data pointer in
virtio_grab_drvdata().
Signed-off-by: NHeinz Graalfs <graalfs@linux.vnet.ibm.com>
Reviewed-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: NCornelia Huck <cornelia.huck@de.ibm.com>

Merge tag 'kvm-for-3.15-1' of git://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms into kvm-next

Merge tag 'kvm-s390-20140304' of git://git.kernel.org/pub/scm/linux/kernel/git/kvms390/linux into kvm-next

commit 0061d53d introduced a mechanism to execute a global clock
update for a vm. We can apply this periodically in order to propagate
host NTP corrections. Also, if all vcpus of a vm are pinned, then
without an additional trigger, no guest NTP corrections can propagate
either, as the current trigger is only vcpu cpu migration.
Signed-off-by: NAndrew Jones <drjones@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

When we update a vcpu's local clock it may pick up an NTP correction.
We can't wait an indeterminate amount of time for other vcpus to pick
up that correction, so commit 0061d53d introduced a global clock
update. However, we can't request a global clock update on every vcpu
load either (which is what happens if the tsc is marked as unstable).
The solution is to rate-limit the global clock updates. Marcelo
calculated that we should delay the global clock updates no more
than 0.1s as follows:

Assume an NTP correction c is applied to one vcpu, but not the other,
then in n seconds the delta of the vcpu system_timestamps will be
c * n. If we assume a correction of 500ppm (worst-case), then the two
vcpus will diverge 50us in 0.1s, which is a considerable amount.
Signed-off-by: NAndrew Jones <drjones@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Implement the new CCW_CMD_SET_IND_ADAPTER command and try to enable
adapter interrupts for every device on the first startup. If the host
does not support adapter interrupts, fall back to normal I/O interrupts.

virtio-ccw adapter interrupts use the same isc as normal I/O subchannels
and share a summary indicator for all devices sharing the same indicator
area.

Indicator bits for the individual virtqueues may be contained in the same
indicator area for different devices.
Signed-off-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: NChristian Borntraeger <borntraeger@de.ibm.com>

Add airq_iv_alloc and airq_iv_free to allocate and free consecutive
ranges of irqs from the interrupt vector.
Signed-off-by: NMartin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: NChristian Borntraeger <borntraeger@de.ibm.com>

The interrupt handler virtio_ccw_int_handler() using the vcdev pointer
is protected by the ccw_device lock. Resetting the pointer within the
ccw_device structure should be done when holding this lock.

Also resetting the vcdev pointer (under the ccw_device lock) prior to
freeing the vcdev pointer memory removes a critical path.
Signed-off-by: NHeinz Graalfs <graalfs@linux.vnet.ibm.com>
Acked-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: NChristian Borntraeger <borntraeger@de.ibm.com>

We can use kvm_get_vcpu() now and don't need the
local_int array in the floating_int struct anymore.
This also means we don't have to hold the float_int.lock
in some places.
Signed-off-by: NJens Freimann <jfrei@linux.vnet.ibm.com>
Signed-off-by: NChristian Borntraeger <borntraeger@de.ibm.com>

When SIGP SET_PREFIX is called with an illegal CPU id, it must return
the condition code 3 ("not operational") instead of 1. Also fixed the
order in which the checks are done - CC3 has a higher priority than CC1.
And while we're at it, this patch also get rid of the floating interrupt
lock here by using kvm_get_vcpu() to get the local_int struct of the
destination CPU.
Signed-off-by: NThomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: NChristian Borntraeger <borntraeger@de.ibm.com>

We don't need to loop over all cpus to get the number of
vcpus. Let's use the available counter online_vcpus instead.
Signed-off-by: NJens Freimann <jfrei@linux.vnet.ibm.com>
Signed-off-by: NChristian Borntraeger <borntraeger@de.ibm.com>

For migration/reset we want to expose the guest breaking event
address register to userspace. Lets use ONE_REG for that purpose.
Signed-off-by: NChristian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: NJason J. Herne <jjherne@linux.vnet.ibm.com>

commit d208c79d (KVM: s390: Enable
the LPP facility for guests) enabled the LPP instruction for guests.
We should expose the program parameter as a pseudo register for
migration/reset etc. Lets also reset this value on initial CPU
reset.
Signed-off-by: NChristian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: NThomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: NJason J. Herne <jjherne@linux.vnet.ibm.com>

Commit e504c909 (kvm, vmx: Fix lazy FPU on nested guest, 2013-11-13)
highlighted a real problem, but the fix was subtly wrong.

nested_read_cr0 is the CR0 as read by L2, but here we want to look at
the CR0 value reflecting L1's setup.  In other words, L2 might think
that TS=0 (so nested_read_cr0 has the bit clear); but if L1 is actually
running it with TS=1, we should inject the fault into L1.

The effective value of CR0 in L2 is contained in vmcs12->guest_cr0, use
it.

Fixes: e504c909Reported-by: NKashyap Chamarty <kchamart@redhat.com>
Reported-by: NStefan Bader <stefan.bader@canonical.com>
Tested-by: NKashyap Chamarty <kchamart@redhat.com>
Tested-by: NAnthoine Bourgeois <bourgeois@bertin.fr>
Cc: stable@vger.kernel.org
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Compiling with THP enabled leads to the following warning:

arch/arm/kvm/mmu.c: In function ‘unmap_range’:
arch/arm/kvm/mmu.c:177:39: warning: ‘pte’ may be used uninitialized in this function [-Wmaybe-uninitialized]
   if (kvm_pmd_huge(*pmd) || page_empty(pte)) {
                                        ^
Code inspection reveals that these two cases are mutually exclusive,
so GCC is a bit overzealous here. Silence it anyway by initializing
pte to NULL and testing it later on.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Acked-by: NChristoffer Dall <christoffer.dall@linaro.org>

In order to be able to detect the point where the guest enables
its MMU and caches, trap all the VM related system registers.

Once we see the guest enabling both the MMU and the caches, we
can go back to a saner mode of operation, which is to leave these
registers in complete control of the guest.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Acked-by: NCatalin Marinas <catalin.marinas@arm.com>
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>

HCR.TVM traps (among other things) accesses to AMAIR0 and AMAIR1.
In order to minimise the amount of surprise a guest could generate by
trying to access these registers with caches off, add them to the
list of registers we switch/handle.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>
Acked-by: NCatalin Marinas <catalin.marinas@arm.com>

So far, KVM/ARM used a fixed HCR configuration per guest, except for
the VI/VF/VA bits to control the interrupt in absence of VGIC.

With the upcoming need to dynamically reconfigure trapping, it becomes
necessary to allow the HCR to be changed on a per-vcpu basis.

The fix here is to mimic what KVM/arm64 already does: a per vcpu HCR
field, initialized at setup time.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>
Acked-by: NCatalin Marinas <catalin.marinas@arm.com>

Commit 240e99cb (ARM: KVM: Fix 64-bit coprocessor handling)
added an ordering dependency for the 64bit registers.

The order described is: CRn, CRm, Op1, Op2, 64bit-first.

Unfortunately, the implementation is: CRn, 64bit-first, CRm...

Move the 64bit test to be last in order to match the documentation.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>
Acked-by: NCatalin Marinas <catalin.marinas@arm.com>

Commit 240e99cb (ARM: KVM: Fix 64-bit coprocessor handling)
changed the way we match the 64bit coprocessor access from
user space, but didn't update the trap handler for the same
set of registers.

The effect is that a trapped 64bit access is never matched, leading
to a fault being injected into the guest. This went unnoticed as we
didn't really trap any 64bit register so far.

Placing the CRm field of the access into the CRn field of the matching
structure fixes the problem. Also update the debug feature to emit the
expected string in case of failing match.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>
Acked-by: NCatalin Marinas <catalin.marinas@arm.com>

In order for a guest with caches disabled to observe data written
contained in a given page, we need to make sure that page is
committed to memory, and not just hanging in the cache (as guest
accesses are completely bypassing the cache until it decides to
enable it).

For this purpose, hook into the coherent_cache_guest_page
function and flush the region if the guest SCTLR
register doesn't show the MMU and caches as being enabled.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>
Reviewed-by: NCatalin Marinas <catalin.marinas@arm.com>

When the guest runs with caches disabled (like in an early boot
sequence, for example), all the writes are diectly going to RAM,
bypassing the caches altogether.

Once the MMU and caches are enabled, whatever sits in the cache
becomes suddenly visible, which isn't what the guest expects.

A way to avoid this potential disaster is to invalidate the cache
when the MMU is being turned on. For this, we hook into the SCTLR_EL1
trapping code, and scan the stage-2 page tables, invalidating the
pages/sections that have already been mapped in.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Reviewed-by: NCatalin Marinas <catalin.marinas@arm.com>
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>

The use of p*d_addr_end with stage-2 translation is slightly dodgy,
as the IPA is 40bits, while all the p*d_addr_end helpers are
taking an unsigned long (arm64 is fine with that as unligned long
is 64bit).

The fix is to introduce 64bit clean versions of the same helpers,
and use them in the stage-2 page table code.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Acked-by: NCatalin Marinas <catalin.marinas@arm.com>
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>

In order to be able to detect the point where the guest enables
its MMU and caches, trap all the VM related system registers.

Once we see the guest enabling both the MMU and the caches, we
can go back to a saner mode of operation, which is to leave these
registers in complete control of the guest.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Reviewed-by: NCatalin Marinas <catalin.marinas@arm.com>
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>

The current handling of AArch32 trapping is slightly less than
perfect, as it is not possible (from a handler point of view)
to distinguish it from an AArch64 access, nor to tell a 32bit
from a 64bit access either.

Fix this by introducing two additional flags:
- is_aarch32: true if the access was made in AArch32 mode
- is_32bit: true if is_aarch32 == true and a MCR/MRC instruction
  was used to perform the access (as opposed to MCRR/MRRC).

This allows a handler to cover all the possible conditions in which
a system register gets trapped.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Acked-by: NChristoffer Dall <christoffer.dall@linaro.org>

In order for the guest with caches off to observe data written
contained in a given page, we need to make sure that page is
committed to memory, and not just hanging in the cache (as
guest accesses are completely bypassing the cache until it
decides to enable it).

For this purpose, hook into the coherent_icache_guest_page
function and flush the region if the guest SCTLR_EL1
register doesn't show the MMU  and caches as being enabled.
The function also get renamed to coherent_cache_guest_page.
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Reviewed-by: NCatalin Marinas <catalin.marinas@arm.com>
Reviewed-by: NChristoffer Dall <christoffer.dall@linaro.org>

Commit e504c909 (kvm, vmx: Fix lazy FPU on nested guest, 2013-11-13)
highlighted a real problem, but the fix was subtly wrong.

nested_read_cr0 is the CR0 as read by L2, but here we want to look at
the CR0 value reflecting L1's setup.  In other words, L2 might think
that TS=0 (so nested_read_cr0 has the bit clear); but if L1 is actually
running it with TS=1, we should inject the fault into L1.

The effective value of CR0 in L2 is contained in vmcs12->guest_cr0, use
it.

Fixes: e504c909Reported-by: NKashyap Chamarty <kchamart@redhat.com>
Reported-by: NStefan Bader <stefan.bader@canonical.com>
Tested-by: NKashyap Chamarty <kchamart@redhat.com>
Tested-by: NAnthoine Bourgeois <bourgeois@bertin.fr>
Cc: stable@vger.kernel.org
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

The problem occurs when the guest performs a pusha with the stack
address pointing to an mmio address (or an invalid guest physical
address) to start with, but then extending into an ordinary guest
physical address.  When doing repeated emulated pushes
emulator_read_write sets mmio_needed to 1 on the first one.  On a
later push when the stack points to regular memory,
mmio_nr_fragments is set to 0, but mmio_is_needed is not set to 0.

As a result, KVM exits to userspace, and then returns to
complete_emulated_mmio.  In complete_emulated_mmio
vcpu->mmio_cur_fragment is incremented.  The termination condition of
vcpu->mmio_cur_fragment == vcpu->mmio_nr_fragments is never achieved.
The code bounces back and fourth to userspace incrementing
mmio_cur_fragment past it's buffer.  If the guest does nothing else it
eventually leads to a a crash on a memcpy from invalid memory address.

However if a guest code can cause the vm to be destroyed in another
vcpu with excellent timing, then kvm_clear_async_pf_completion_queue
can be used by the guest to control the data that's pointed to by the
call to cancel_work_item, which can be used to gain execution.

Fixes: f78146b0Signed-off-by: NAndrew Honig <ahonig@google.com>
Cc: stable@vger.kernel.org (3.5+)
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Commit 1fcf7ce0 (arm: kvm: implement CPU PM notifier) added
support for CPU power-management, using a cpu_notifier to re-init
KVM on a CPU that entered CPU idle.

The code assumed that a CPU entering idle would actually be powered
off, loosing its state entierely, and would then need to be
reinitialized. It turns out that this is not always the case, and
some HW performs CPU PM without actually killing the core. In this
case, we try to reinitialize KVM while it is still live. It ends up
badly, as reported by Andre Przywara (using a Calxeda Midway):

[    3.663897] Kernel panic - not syncing: unexpected prefetch abort in Hyp mode at: 0x685760
[    3.663897] unexpected data abort in Hyp mode at: 0xc067d150
[    3.663897] unexpected HVC/SVC trap in Hyp mode at: 0xc0901dd0

The trick here is to detect if we've been through a full re-init or
not by looking at HVBAR (VBAR_EL2 on arm64). This involves
implementing the backend for __hyp_get_vectors in the main KVM HYP
code (rather small), and checking the return value against the
default one when the CPU notifier is called on CPU_PM_EXIT.
Reported-by: NAndre Przywara <osp@andrep.de>
Tested-by: NAndre Przywara <osp@andrep.de>
Cc: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Cc: Rob Herring <rob.herring@linaro.org>
Acked-by: NChristoffer Dall <christoffer.dall@linaro.org>
Signed-off-by: NMarc Zyngier <marc.zyngier@arm.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

No need to scan the entire VCPU array.
Signed-off-by: NTakuya Yoshikawa <yoshikawa_takuya_b1@lab.ntt.co.jp>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>

Commit "kvm: Record the preemption status of vcpus using preempt notifiers"
caused a performance regression on s390. It turned out that in the case that
if a former sleeping cpu, that was woken up, this cpu is not a yield candidate
since it gave up the cpu voluntarily. To retain this candiate its preempted
flag is set during wakeup and interrupt delivery time.

Significant performance measurement work and code analysis to solve this
issue was provided by Mao Chuan Li and his team in Beijing.
Signed-off-by: NMichael Mueller <mimu@linux.vnet.ibm.com>
Reviewed-by: NChristian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>