Whenever we want to wake up a VCPU (e.g. when injecting an IRQ), we
have to kick it out of vsie, so the request will be handled faster.
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
As soon as guest 2 is allowed to use the vector facility (indicated via
STFLE), it can also enable it for guest 3. We have to take care of the
sattellite block that might be used when not relying on lazy vector
copying (not the case for KVM).
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
This patch adds basic support for nested virtualization on s390x, called
VSIE (virtual SIE) and allows it to be used by the guest if the necessary
facilities are supported by the hardware and enabled for the guest.
In order to make this work, we have to shadow the sie control block
provided by guest 2. In order to gain some performance, we have to
reuse the same shadow blocks as good as possible. For now, we allow
as many shadow blocks as we have VCPUs (that way, every VCPU can run the
VSIE concurrently).
We have to watch out for the prefix getting unmapped out of our shadow
gmap and properly get the VCPU out of VSIE in that case, to fault the
prefix pages back in. We use the PROG_REQUEST bit for that purpose.
This patch is based on an initial prototype by Tobias Elpelt.
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Nested virtualization will have to enable own gmaps. Current code
would enable the wrong gmap whenever scheduled out and back in,
therefore resulting in the wrong gmap being enabled.
This patch reenables the last enabled gmap, therefore avoiding having to
touch vcpu->arch.gmap when enabling a different gmap.
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
If guest-storage-limit-suppression is not available, we would for now
have a valid guest address space with size 0. So let's simply set the
origin to 0 and the limit to hamax.
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
For now, we only have an interface to query and configure facilities
indicated via STFL(E). However, we also have features indicated via
SCLP, that have to be indicated to the guest by user space and usually
require KVM support.
This patch allows user space to query and configure available cpu features
for the guest.
Please note that disabling a feature doesn't necessarily mean that it is
completely disabled (e.g. ESOP is mostly handled by the SIE). We will try
our best to disable it.
Most features (e.g. SCLP) can't directly be forwarded, as most of them need
in addition to hardware support, support in KVM. As we later on want to
turn these features in KVM explicitly on/off (to simulate different
behavior), we have to filter all features provided by the hardware and
make them configurable.
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Store hypervisor information is a valid instruction not only in
supervisor state but also in problem state, i.e. the guest's
userspace. Its execution is not only computational and memory
intensive, but also has to get hold of the ipte lock to write to the
guest's memory.
This lock is not intended to be held often and long, especially not
from the untrusted guest userspace. Therefore we apply rate limiting
of sthyi executions per VM.
Signed-off-by: Janosch Frank <frankja@linux.vnet.ibm.com>
Acked-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Store Hypervisor Information is an emulated z/VM instruction that
provides a guest with basic information about the layers it is running
on. This includes information about the cpu configuration of both the
machine and the lpar, as well as their names, machine model and
machine type. This information enables an application to determine the
maximum capacity of CPs and IFLs available to software.
The instruction is available whenever the facility bit 74 is set,
otherwise executing it results in an operation exception.
It is important to check the validity flags in the sections before
using data from any structure member. It is not guaranteed that all
members will be valid on all machines / machine configurations.
Signed-off-by: Janosch Frank <frankja@linux.vnet.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
This commit introduces code that handles operation exception
interceptions. With this handler we can emulate instructions by using
illegal opcodes.
Signed-off-by: Janosch Frank <frankja@linux.vnet.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
on s390 we disabled the halt polling with commit 920552b213
("KVM: disable halt_poll_ns as default for s390x"), as floating
interrupts would let all CPUs have a successful poll, resulting
in much higher CPU usage (on otherwise idle systems).
With the improved selection of polls we can now retry halt polling.
Performance measurements with different choices like 25,50,80,100,200
microseconds showed that 80 microseconds seems to improve several cases
without increasing the CPU costs too much. Higher values would improve
the performance even more but increased the cpu time as well.
So let's start small and use this value of 80 microseconds on s390 until
we have a better understanding of cost/benefit of higher values.
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Some wakeups should not be considered a sucessful poll. For example on
s390 I/O interrupts are usually floating, which means that _ALL_ CPUs
would be considered runnable - letting all vCPUs poll all the time for
transactional like workload, even if one vCPU would be enough.
This can result in huge CPU usage for large guests.
This patch lets architectures provide a way to qualify wakeups if they
should be considered a good/bad wakeups in regard to polls.
For s390 the implementation will fence of halt polling for anything but
known good, single vCPU events. The s390 implementation for floating
interrupts does a wakeup for one vCPU, but the interrupt will be delivered
by whatever CPU checks first for a pending interrupt. We prefer the
woken up CPU by marking the poll of this CPU as "good" poll.
This code will also mark several other wakeup reasons like IPI or
expired timers as "good". This will of course also mark some events as
not sucessful. As KVM on z runs always as a 2nd level hypervisor,
we prefer to not poll, unless we are really sure, though.
This patch successfully limits the CPU usage for cases like uperf 1byte
transactional ping pong workload or wakeup heavy workload like OLTP
while still providing a proper speedup.
This also introduced a new vcpu stat "halt_poll_no_tuning" that marks
wakeups that are considered not good for polling.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Acked-by: Radim Krčmář <rkrcmar@redhat.com> (for an earlier version)
Cc: David Matlack <dmatlack@google.com>
Cc: Wanpeng Li <kernellwp@gmail.com>
[Rename config symbol. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We only have one cpuid for all VCPUs, so let's directly use the one in the
cpu model. Also always store it directly as u64, no need for struct cpuid.
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
but lots of architecture-specific changes.
* ARM:
- VHE support so that we can run the kernel at EL2 on ARMv8.1 systems
- PMU support for guests
- 32bit world switch rewritten in C
- various optimizations to the vgic save/restore code.
* PPC:
- enabled KVM-VFIO integration ("VFIO device")
- optimizations to speed up IPIs between vcpus
- in-kernel handling of IOMMU hypercalls
- support for dynamic DMA windows (DDW).
* s390:
- provide the floating point registers via sync regs;
- separated instruction vs. data accesses
- dirty log improvements for huge guests
- bugfixes and documentation improvements.
* x86:
- Hyper-V VMBus hypercall userspace exit
- alternative implementation of lowest-priority interrupts using vector
hashing (for better VT-d posted interrupt support)
- fixed guest debugging with nested virtualizations
- improved interrupt tracking in the in-kernel IOAPIC
- generic infrastructure for tracking writes to guest memory---currently
its only use is to speedup the legacy shadow paging (pre-EPT) case, but
in the future it will be used for virtual GPUs as well
- much cleanup (LAPIC, kvmclock, MMU, PIT), including ubsan fixes.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)
iQEcBAABAgAGBQJW5r3BAAoJEL/70l94x66D2pMH/jTSWWwdTUJMctrDjPVzKzG0
yOzHW5vSLFoFlwEOY2VpslnXzn5TUVmCAfrdmFNmQcSw6hGb3K/xA/ZX/KLwWhyb
oZpr123ycahga+3q/ht/dFUBCCyWeIVMdsLSFwpobEBzPL0pMgc9joLgdUC6UpWX
tmN0LoCAeS7spC4TTiTTpw3gZ/L+aB0B6CXhOMjldb9q/2CsgaGyoVvKA199nk9o
Ngu7ImDt7l/x1VJX4/6E/17VHuwqAdUrrnbqerB/2oJ5ixsZsHMGzxQ3sHCmvyJx
WG5L00ubB1oAJAs9fBg58Y/MdiWX99XqFhdEfxq4foZEiQuCyxygVvq3JwZTxII=
=OUZZ
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Pull KVM updates from Paolo Bonzini:
"One of the largest releases for KVM... Hardly any generic
changes, but lots of architecture-specific updates.
ARM:
- VHE support so that we can run the kernel at EL2 on ARMv8.1 systems
- PMU support for guests
- 32bit world switch rewritten in C
- various optimizations to the vgic save/restore code.
PPC:
- enabled KVM-VFIO integration ("VFIO device")
- optimizations to speed up IPIs between vcpus
- in-kernel handling of IOMMU hypercalls
- support for dynamic DMA windows (DDW).
s390:
- provide the floating point registers via sync regs;
- separated instruction vs. data accesses
- dirty log improvements for huge guests
- bugfixes and documentation improvements.
x86:
- Hyper-V VMBus hypercall userspace exit
- alternative implementation of lowest-priority interrupts using
vector hashing (for better VT-d posted interrupt support)
- fixed guest debugging with nested virtualizations
- improved interrupt tracking in the in-kernel IOAPIC
- generic infrastructure for tracking writes to guest
memory - currently its only use is to speedup the legacy shadow
paging (pre-EPT) case, but in the future it will be used for
virtual GPUs as well
- much cleanup (LAPIC, kvmclock, MMU, PIT), including ubsan fixes"
* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (217 commits)
KVM: x86: remove eager_fpu field of struct kvm_vcpu_arch
KVM: x86: disable MPX if host did not enable MPX XSAVE features
arm64: KVM: vgic-v3: Only wipe LRs on vcpu exit
arm64: KVM: vgic-v3: Reset LRs at boot time
arm64: KVM: vgic-v3: Do not save an LR known to be empty
arm64: KVM: vgic-v3: Save maintenance interrupt state only if required
arm64: KVM: vgic-v3: Avoid accessing ICH registers
KVM: arm/arm64: vgic-v2: Make GICD_SGIR quicker to hit
KVM: arm/arm64: vgic-v2: Only wipe LRs on vcpu exit
KVM: arm/arm64: vgic-v2: Reset LRs at boot time
KVM: arm/arm64: vgic-v2: Do not save an LR known to be empty
KVM: arm/arm64: vgic-v2: Move GICH_ELRSR saving to its own function
KVM: arm/arm64: vgic-v2: Save maintenance interrupt state only if required
KVM: arm/arm64: vgic-v2: Avoid accessing GICH registers
KVM: s390: allocate only one DMA page per VM
KVM: s390: enable STFLE interpretation only if enabled for the guest
KVM: s390: wake up when the VCPU cpu timer expires
KVM: s390: step the VCPU timer while in enabled wait
KVM: s390: protect VCPU cpu timer with a seqcount
KVM: s390: step VCPU cpu timer during kvm_run ioctl
...
We can fit the 2k for the STFLE interpretation and the crypto
control block into one DMA page. As we now only have to allocate
one DMA page, we can clean up the code a bit.
As a nice side effect, this also fixes a problem with crycbd alignment in
case special allocation debug options are enabled, debugged by Sascha
Silbe.
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
For now, only the owning VCPU thread (that has loaded the VCPU) can get a
consistent cpu timer value when calculating the delta. However, other
threads might also be interested in a more recent, consistent value. Of
special interest will be the timer callback of a VCPU that executes without
having the VCPU loaded and could run in parallel with the VCPU thread.
The cpu timer has a nice property: it is only updated by the owning VCPU
thread. And speaking about accounting, a consistent value can only be
calculated by looking at cputm_start and the cpu timer itself in
one shot, otherwise the result might be wrong.
As we only have one writing thread at a time (owning VCPU thread), we can
use a seqcount instead of a seqlock and retry if the VCPU refreshed its
cpu timer. This avoids any heavy locking and only introduces a counter
update/check plus a handful of smp_wmb().
The owning VCPU thread should never have to retry on reads, and also for
other threads this might be a very rare scenario.
Please note that we have to use the raw_* variants for locking the seqcount
as lockdep will produce false warnings otherwise. The rq->lock held during
vcpu_load/put is also acquired from hardirq context. Lockdep cannot know
that we avoid potential deadlocks by disabling preemption and thereby
disable concurrent write locking attempts (via vcpu_put/load).
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Architecturally we should only provide steal time if we are scheduled
away, and not if the host interprets a guest exit. We have to step
the guest CPU timer in these cases.
In the first shot, we will step the VCPU timer only during the kvm_run
ioctl. Therefore all time spent e.g. in interception handlers or on irq
delivery will be accounted for that VCPU.
We have to take care of a few special cases:
- Other VCPUs can test for pending irqs. We can only report a consistent
value for the VCPU thread itself when adding the delta.
- We have to take care of STP sync, therefore we have to extend
kvm_clock_sync() and disable preemption accordingly
- During any call to disable/enable/start/stop we could get premeempted
and therefore get start/stop calls. Therefore we have to make sure we
don't get into an inconsistent state.
Whenever a VCPU is scheduled out, sleeping, in user space or just about
to enter the SIE, the guest cpu timer isn't stepped.
Please note that all primitives are prepared to be called from both
environments (cpu timer accounting enabled or not), although not completely
used in this patch yet (e.g. kvm_s390_set_cpu_timer() will never be called
while cpu timer accounting is enabled).
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
The problem:
On -rt, an emulated LAPIC timer instances has the following path:
1) hard interrupt
2) ksoftirqd is scheduled
3) ksoftirqd wakes up vcpu thread
4) vcpu thread is scheduled
This extra context switch introduces unnecessary latency in the
LAPIC path for a KVM guest.
The solution:
Allow waking up vcpu thread from hardirq context,
thus avoiding the need for ksoftirqd to be scheduled.
Normal waitqueues make use of spinlocks, which on -RT
are sleepable locks. Therefore, waking up a waitqueue
waiter involves locking a sleeping lock, which
is not allowed from hard interrupt context.
cyclictest command line:
This patch reduces the average latency in my tests from 14us to 11us.
Daniel writes:
Paolo asked for numbers from kvm-unit-tests/tscdeadline_latency
benchmark on mainline. The test was run 1000 times on
tip/sched/core 4.4.0-rc8-01134-g0905f04:
./x86-run x86/tscdeadline_latency.flat -cpu host
with idle=poll.
The test seems not to deliver really stable numbers though most of
them are smaller. Paolo write:
"Anything above ~10000 cycles means that the host went to C1 or
lower---the number means more or less nothing in that case.
The mean shows an improvement indeed."
Before:
min max mean std
count 1000.000000 1000.000000 1000.000000 1000.000000
mean 5162.596000 2019270.084000 5824.491541 20681.645558
std 75.431231 622607.723969 89.575700 6492.272062
min 4466.000000 23928.000000 5537.926500 585.864966
25% 5163.000000 1613252.750000 5790.132275 16683.745433
50% 5175.000000 2281919.000000 5834.654000 23151.990026
75% 5190.000000 2382865.750000 5861.412950 24148.206168
max 5228.000000 4175158.000000 6254.827300 46481.048691
After
min max mean std
count 1000.000000 1000.00000 1000.000000 1000.000000
mean 5143.511000 2076886.10300 5813.312474 21207.357565
std 77.668322 610413.09583 86.541500 6331.915127
min 4427.000000 25103.00000 5529.756600 559.187707
25% 5148.000000 1691272.75000 5784.889825 17473.518244
50% 5160.000000 2308328.50000 5832.025000 23464.837068
75% 5172.000000 2393037.75000 5853.177675 24223.969976
max 5222.000000 3922458.00000 6186.720500 42520.379830
[Patch was originaly based on the swait implementation found in the -rt
tree. Daniel ported it to mainline's version and gathered the
benchmark numbers for tscdeadline_latency test.]
Signed-off-by: Daniel Wagner <daniel.wagner@bmw-carit.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: linux-rt-users@vger.kernel.org
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/1455871601-27484-4-git-send-email-wagi@monom.org
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Since commit 9977e886cb ("s390/kernel: lazy restore fpu registers"),
vregs in struct sie_page is unsed. We can safely remove the field and
the definition.
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
The kernel now always uses vector registers when available, however KVM
has special logic if support is really enabled for a guest. If support
is disabled, guest_fpregs.fregs will only contain memory for the fpu.
The kernel, however, will store vector registers into that area,
resulting in crazy memory overwrites.
Simply extending that area is not enough, because the format of the
registers also changes. We would have to do additional conversions, making
the code even more complex. Therefore let's directly use one place for
the vector/fpu registers + fpc (in kvm_run). We just have to convert the
data properly when accessing it. This makes current code much easier.
Please note that vector/fpu registers are now always stored to
vcpu->run->s.regs.vrs. Although this data is visible to QEMU and
used for migration, we only guarantee valid values to user space when
KVM_SYNC_VRS is set. As that is only the case when we have vector
register support, we are on the safe side.
Fixes: b5510d9b68 ("s390/fpu: always enable the vector facility if it is available")
Cc: stable@vger.kernel.org # v4.4 d9a3a09af5 s390/kvm: remove dependency on struct save_area definition
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
[adopt to d9a3a09af5]
Since the numbers now overlap, it makes sense to enumerate
them in asm/kvm_host.h rather than linux/kvm_host.h. Functions
that refer to architecture-specific requests are also moved
to arch/.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch adds runtime instrumentation support for KVM guest. We need to
setup a save area for the runtime instrumentation-controls control block(RICCB)
and implement the necessary interfaces to live migrate the guest settings.
We setup the sie control block in a way, that the runtime
instrumentation instructions of a guest are handled by hardware.
We also add a capability KVM_CAP_S390_RI to make this feature opt-in as
it needs migration support.
Signed-off-by: Fan Zhang <zhangfan@linux.vnet.ibm.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
While the userspace interface requests the maximum size the gmap code
expects to get a maximum address.
This error resulted in bigger page tables than necessary for some guest
sizes, e.g. a 2GB guest used 3 levels instead of 2.
At the same time we introduce KVM_S390_NO_MEM_LIMIT, which allows in a
bright future that a guest spans the complete 64 bit address space.
We also switch to TASK_MAX_SIZE for the initial memory size, this is a
cosmetic change as the previous size also resulted in a 4 level pagetable
creation.
Reported-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
This patch allows s390 to have more than 64 VCPUs for a guest (up to
248 for memory usage considerations), if supported by the underlaying
hardware (sclp.has_esca).
Signed-off-by: Eugene (jno) Dvurechenski <jno@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
This patch adds code that performs transparent switch to Extended
SCA on addition of 65th VCPU in a VM. Disposal of ESCA is added too.
The entier ESCA functionality, however, is still not enabled.
The enablement will be provided in a separate patch.
This patch also uses read/write lock protection of SCA and its subfields for
possible disposal at the BSCA-to-ESCA transition. While only Basic SCA needs such
a protection (for the swap), any SCA access is now guarded.
Signed-off-by: Eugene (jno) Dvurechenski <jno@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
This patch updates the routines (sca_*) to provide transparent access
to and manipulation on the data for both Basic and Extended SCA in use.
The kvm.arch.sca is generalized to (void *) to handle BSCA/ESCA cases.
Also the kvm.arch.use_esca flag is provided.
The actual functionality is kept the same.
Signed-off-by: Eugene (jno) Dvurechenski <jno@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
This patch adds new structures and updates some existing ones to
provide the base for Extended SCA functionality.
The old sca_* structures were renamed to bsca_* to keep things uniform.
The access to fields of SIGP controls were turned into bitfields instead
of hardcoded bitmasks.
Signed-off-by: Eugene (jno) Dvurechenski <jno@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
handling.
PPC: Mostly bug fixes.
ARM: No big features, but many small fixes and prerequisites including:
- a number of fixes for the arch-timer
- introducing proper level-triggered semantics for the arch-timers
- a series of patches to synchronously halt a guest (prerequisite for
IRQ forwarding)
- some tracepoint improvements
- a tweak for the EL2 panic handlers
- some more VGIC cleanups getting rid of redundant state
x86: quite a few changes:
- support for VT-d posted interrupts (i.e. PCI devices can inject
interrupts directly into vCPUs). This introduces a new component (in
virt/lib/) that connects VFIO and KVM together. The same infrastructure
will be used for ARM interrupt forwarding as well.
- more Hyper-V features, though the main one Hyper-V synthetic interrupt
controller will have to wait for 4.5. These will let KVM expose Hyper-V
devices.
- nested virtualization now supports VPID (same as PCID but for vCPUs)
which makes it quite a bit faster
- for future hardware that supports NVDIMM, there is support for clflushopt,
clwb, pcommit
- support for "split irqchip", i.e. LAPIC in kernel + IOAPIC/PIC/PIT in
userspace, which reduces the attack surface of the hypervisor
- obligatory smattering of SMM fixes
- on the guest side, stable scheduler clock support was rewritten to not
require help from the hypervisor.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)
iQEcBAABAgAGBQJWO2IQAAoJEL/70l94x66D/K0H/3AovAgYmJQToZlimsktMk6a
f2xhdIqfU5lIQQh5uNBCfL3o9o8H9Py1ym7aEw3fmztPHHJYc91oTatt2UEKhmEw
VtZHp/dFHt3hwaIdXmjRPEXiYctraKCyrhaUYdWmUYkoKi7lW5OL5h+S7frG2U6u
p/hFKnHRZfXHr6NSgIqvYkKqtnc+C0FWY696IZMzgCksOO8jB1xrxoSN3tANW3oJ
PDV+4og0fN/Fr1capJUFEc/fejREHneANvlKrLaa8ht0qJQutoczNADUiSFLcMPG
iHljXeDsv5eyjMtUuIL8+MPzcrIt/y4rY41ZPiKggxULrXc6H+JJL/e/zThZpXc=
=iv2z
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Pull KVM updates from Paolo Bonzini:
"First batch of KVM changes for 4.4.
s390:
A bunch of fixes and optimizations for interrupt and time handling.
PPC:
Mostly bug fixes.
ARM:
No big features, but many small fixes and prerequisites including:
- a number of fixes for the arch-timer
- introducing proper level-triggered semantics for the arch-timers
- a series of patches to synchronously halt a guest (prerequisite
for IRQ forwarding)
- some tracepoint improvements
- a tweak for the EL2 panic handlers
- some more VGIC cleanups getting rid of redundant state
x86:
Quite a few changes:
- support for VT-d posted interrupts (i.e. PCI devices can inject
interrupts directly into vCPUs). This introduces a new
component (in virt/lib/) that connects VFIO and KVM together.
The same infrastructure will be used for ARM interrupt
forwarding as well.
- more Hyper-V features, though the main one Hyper-V synthetic
interrupt controller will have to wait for 4.5. These will let
KVM expose Hyper-V devices.
- nested virtualization now supports VPID (same as PCID but for
vCPUs) which makes it quite a bit faster
- for future hardware that supports NVDIMM, there is support for
clflushopt, clwb, pcommit
- support for "split irqchip", i.e. LAPIC in kernel +
IOAPIC/PIC/PIT in userspace, which reduces the attack surface of
the hypervisor
- obligatory smattering of SMM fixes
- on the guest side, stable scheduler clock support was rewritten
to not require help from the hypervisor"
* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (123 commits)
KVM: VMX: Fix commit which broke PML
KVM: x86: obey KVM_X86_QUIRK_CD_NW_CLEARED in kvm_set_cr0()
KVM: x86: allow RSM from 64-bit mode
KVM: VMX: fix SMEP and SMAP without EPT
KVM: x86: move kvm_set_irq_inatomic to legacy device assignment
KVM: device assignment: remove pointless #ifdefs
KVM: x86: merge kvm_arch_set_irq with kvm_set_msi_inatomic
KVM: x86: zero apic_arb_prio on reset
drivers/hv: share Hyper-V SynIC constants with userspace
KVM: x86: handle SMBASE as physical address in RSM
KVM: x86: add read_phys to x86_emulate_ops
KVM: x86: removing unused variable
KVM: don't pointlessly leave KVM_COMPAT=y in non-KVM configs
KVM: arm/arm64: Merge vgic_set_lr() and vgic_sync_lr_elrsr()
KVM: arm/arm64: Clean up vgic_retire_lr() and surroundings
KVM: arm/arm64: Optimize away redundant LR tracking
KVM: s390: use simple switch statement as multiplexer
KVM: s390: drop useless newline in debugging data
KVM: s390: SCA must not cross page boundaries
KVM: arm: Do not indent the arguments of DECLARE_BITMAP
...
Some times it is useful for architecture implementations of KVM to know
when the VCPU thread is about to block or when it comes back from
blocking (arm/arm64 needs to know this to properly implement timers, for
example).
Therefore provide a generic architecture callback function in line with
what we do elsewhere for KVM generic-arch interactions.
Reviewed-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <christoffer.dall@linaro.org>
Split the API and FPU type definitions into separate header files
similar to "x86/fpu: Rename fpu-internal.h to fpu/internal.h" (78f7f1e54b).
The new header files and their meaning are:
asm/fpu/types.h:
FPU related data types, needed for 'struct thread_struct' and
'struct task_struct'.
asm/fpu/api.h:
FPU related 'public' functions for other subsystems and device
drivers.
asm/fpu/internal.h:
FPU internal functions mainly used to convert
FPU register contents in signal handling.
Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
We observed some performance degradation on s390x with dynamic
halt polling. Until we can provide a proper fix, let's enable
halt_poll_ns as default only for supported architectures.
Architectures are now free to set their own halt_poll_ns
default value.
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This new statistic can help diagnosing VCPUs that, for any reason,
trigger bad behavior of halt_poll_ns autotuning.
For example, say halt_poll_ns = 480000, and wakeups are spaced exactly
like 479us, 481us, 479us, 481us. Then KVM always fails polling and wastes
10+20+40+80+160+320+480 = 1110 microseconds out of every
479+481+479+481+479+481+479 = 3359 microseconds. The VCPU then
is consuming about 30% more CPU than it would use without
polling. This would show as an abnormally high number of
attempted polling compared to the successful polls.
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com<
Reviewed-by: David Matlack <dmatlack@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Pull s390 updates from Martin Schwidefsky:
"The big one is support for fake NUMA, splitting a really large machine
in more manageable piece improves performance in some cases, e.g. for
a KVM host.
The FICON Link Incident handling has been improved, this helps the
operator to identify degraded or non-operational FICON connections.
The save and restore of floating point and vector registers has been
overhauled to allow the future use of vector registers in the kernel.
A few small enhancement, magic sys-requests for the vt220 console via
SCLP, some more assembler code has been converted to C, the PCI error
handling is improved.
And the usual cleanup and bug fixing"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux: (59 commits)
s390/jump_label: Use %*ph to print small buffers
s390/sclp_vt220: support magic sysrequests
s390/ctrlchar: improve handling of magic sysrequests
s390/numa: remove superfluous ARCH_WANT defines
s390/3270: redraw screen on unsolicited device end
s390/dcssblk: correct out of bounds array indexes
s390/mm: simplify page table alloc/free code
s390/pci: move debug messages to debugfs
s390/nmi: initialize control register 0 earlier
s390/zcrypt: use msleep() instead of mdelay()
s390/hmcdrv: fix interrupt registration
s390/setup: fix novx parameter
s390/uaccess: remove uaccess_primary kernel parameter
s390: remove unneeded sizeof(void *) comparisons
s390/facilities: remove transactional-execution bits
s390/numa: re-add DIE sched_domain_topology_level
s390/dasd: enhance CUIR scope detection
s390/dasd: fix failing path verification
s390/vdso: emit a GNU hash
s390/numa: make core to node mapping data dynamic
...
In addition to the per VM debug logs, let's provide a global
one for KVM-wide events, like new guests or fatal errors.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Sometimes kvm stat counters are the only performance metric to check
after something went wrong. Let's add additional counters for some
diagnoses.
In addition do the count for diag 10 all the time, even if we inject
a program interrupt.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: Jens Freimann <jfrei@linux.vnet.ibm.com>
Improve the save and restore behavior of FPU register contents to use the
vector extension within the kernel.
The kernel does not use floating-point or vector registers and, therefore,
saving and restoring the FPU register contents are performed for handling
signals or switching processes only. To prepare for using vector
instructions and vector registers within the kernel, enhance the save
behavior and implement a lazy restore at return to user space from a
system call or interrupt.
To implement the lazy restore, the save_fpu_regs() sets a CPU information
flag, CIF_FPU, to indicate that the FPU registers must be restored.
Saving and setting CIF_FPU is performed in an atomic fashion to be
interrupt-safe. When the kernel wants to use the vector extension or
wants to change the FPU register state for a task during signal handling,
the save_fpu_regs() must be called first. The CIF_FPU flag is also set at
process switch. At return to user space, the FPU state is restored. In
particular, the FPU state includes the floating-point or vector register
contents, as well as, vector-enablement and floating-point control. The
FPU state restore and clearing CIF_FPU is also performed in an atomic
fashion.
For KVM, the restore of the FPU register state is performed when restoring
the general-purpose guest registers before the SIE instructions is started.
Because the path towards the SIE instruction is interruptible, the CIF_FPU
flag must be checked again right before going into SIE. If set, the guest
registers must be reloaded again by re-entering the outer SIE loop. This
is the same behavior as if the SIE critical section is interrupted.
Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
exit_sie_sync is used to kick CPUs out of SIE and prevent reentering at
any point in time. This is used to reload the prefix pages and to
set the IBS stuff in a way that guarantees that after this function
returns we are no longer in SIE. All current users trigger KVM requests.
The request must be set before we block the CPUs to avoid races. Let's
make this implicit by adding the request into a new function
kvm_s390_sync_requests that replaces exit_sie_sync and split out
s390_vcpu_block and s390_vcpu_unblock, that can be used to keep
CPUs out of SIE independent of requests.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
1. Enable EDAT2 in the list of KVM facilities
2. Handle 2G frames in pfmf instruction
If we support EDAT2, we may enable handling of 2G frames if not in 24
bit mode.
3. Enable EDAT2 in sie_block
If the EDAT2 facility is available we enable GED2 mode control in the
sie_block.
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Guenther Hutzl <hutzl@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
This patch makes interrupt handling compliant to the z/Architecture
Principles of Operation with regard to interrupt priorities.
Add a bitmap for pending floating interrupts. Each bit relates to a
interrupt type and its list. A turned on bit indicates that a list
contains items (interrupts) which need to be delivered. When delivering
interrupts on a cpu we can merge the existing bitmap for cpu-local
interrupts and floating interrupts and have a single mechanism for
delivery.
Currently we have one list for all kinds of floating interrupts and a
corresponding spin lock. This patch adds a separate list per
interrupt type. An exception to this are service signal and machine check
interrupts, as there can be only one pending interrupt at a time.
Signed-off-by: Jens Freimann <jfrei@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
The patch represents capability KVM_CAP_S390_VECTOR_REGISTERS by means
of the SIMD facility bit. This allows to a) disable the use of SIMD when
used in conjunction with a not-SIMD-aware QEMU, b) to enable SIMD when
used with a SIMD-aware version of QEMU and c) finally by means of a QEMU
version using the future cpu model ioctls.
Signed-off-by: Michael Mueller <mimu@linux.vnet.ibm.com>
Reviewed-by: Eric Farman <farman@linux.vnet.ibm.com>
Tested-by: Eric Farman <farman@linux.vnet.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
The Store System Information (STSI) instruction currently collects all
information it relays to the caller in the kernel. Some information,
however, is only available in user space. An example of this is the
guest name: The kernel always sets "KVMGuest", but user space knows the
actual guest name.
This patch introduces a new exit, KVM_EXIT_S390_STSI, guarded by a
capability that can be enabled by user space if it wants to be able to
insert such data. User space will be provided with the target buffer
and the requested STSI function code.
Reviewed-by: Eric Farman <farman@linux.vnet.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Ekaterina Tumanova <tumanova@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
We finally have all the pieces in place, so let's include the
vector facility bit in the mask of available hardware facilities
for the guest to recognize. Also, enable the vector functionality
in the guest control blocks, to avoid a possible vector data
exception that would otherwise occur when a vector instruction
is issued by the guest operating system.
Signed-off-by: Eric Farman <farman@linux.vnet.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
The new SIGP order Store Additional Status at Address is totally
handled by user space, but we should still record the occurrence
of this order in the kernel code.
Signed-off-by: Eric Farman <farman@linux.vnet.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
A new exception type for vector instructions is introduced with
the new processor, but is handled exactly like a Data Exception
which is already handled by the system.
Signed-off-by: Eric Farman <farman@linux.vnet.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Define and allocate space for both the host and guest views of
the vector registers for a given vcpu. The 32 vector registers
occupy 128 bits each (512 bytes total), but architecturally are
paired with 512 additional bytes of reserved space for future
expansion.
The kvm_sync_regs structs containing the registers are union'ed
with 1024 bytes of padding in the common kvm_run struct. The
addition of 1024 bytes of new register information clearly exceeds
the existing union, so an expansion of that padding is required.
When changing environments, we need to appropriately save and
restore the vector registers viewed by both the host and guest,
into and out of the sync_regs space.
The floating point registers overlay the upper half of vector
registers 0-15, so there's a bit of data duplication here that
needs to be carefully avoided.
Signed-off-by: Eric Farman <farman@linux.vnet.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Most facility related decisions in KVM have to take into account:
- the facilities offered by the underlying run container (LPAR/VM)
- the facilities supported by the KVM code itself
- the facilities requested by a guest VM
This patch adds the KVM driver requested facilities to the test routine.
It additionally renames struct s390_model_fac to kvm_s390_fac and its field
names to be more meaningful.
The semantics of the facilities stored in the KVM architecture structure
is changed. The address arch.model.fac->list now points to the guest
facility list and arch.model.fac->mask points to the KVM facility mask.
This patch fixes the behaviour of KVM for some facilities for guests
that ignore the guest visible facility bits, e.g. guests could use
transactional memory intructions on hosts supporting them even if the
chosen cpu model would not offer them.
The userspace interface is not affected by this change.
Signed-off-by: Michael Mueller <mimu@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
This patch enables cpu model support in kvm/s390 via the vm attribute
interface.
During KVM initialization, the host properties cpuid, IBC value and the
facility list are stored in the architecture specific cpu model structure.
During vcpu setup, these properties are taken to initialize the related SIE
state. This mechanism allows to adjust the properties from user space and thus
to implement different selectable cpu models.
This patch uses the IBC functionality to block instructions that have not
been implemented at the requested CPU type and GA level compared to the
full host capability.
Userspace has to initialize the cpu model before vcpu creation. A cpu model
change of running vcpus is not possible.
Signed-off-by: Michael Mueller <mimu@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
The patch introduces facilities and cpu_ids per virtual machine.
Different virtual machines may want to expose different facilities and
cpu ids to the guest, so let's make them per-vm instead of global.
Signed-off-by: Michael Mueller <mimu@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
We need to specify a different format for the crypto control block
depending on whether the APXA facility is installed or not. Let's
test for it by executing the PQAP(QCI) function and use either a
format-1 or a format-2 crypto control block accordingly. This is a
host only change for z13 and does not affect the guest view.
Signed-off-by: Tony Krowiak <akrowiak@linux.vnet.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
This patch introduces a new module parameter for the KVM module; when it
is present, KVM attempts a bit of polling on every HLT before scheduling
itself out via kvm_vcpu_block.
This parameter helps a lot for latency-bound workloads---in particular
I tested it with O_DSYNC writes with a battery-backed disk in the host.
In this case, writes are fast (because the data doesn't have to go all
the way to the platters) but they cannot be merged by either the host or
the guest. KVM's performance here is usually around 30% of bare metal,
or 50% if you use cache=directsync or cache=writethrough (these
parameters avoid that the guest sends pointless flush requests, and
at the same time they are not slow because of the battery-backed cache).
The bad performance happens because on every halt the host CPU decides
to halt itself too. When the interrupt comes, the vCPU thread is then
migrated to a new physical CPU, and in general the latency is horrible
because the vCPU thread has to be scheduled back in.
With this patch performance reaches 60-65% of bare metal and, more
important, 99% of what you get if you use idle=poll in the guest. This
means that the tunable gets rid of this particular bottleneck, and more
work can be done to improve performance in the kernel or QEMU.
Of course there is some price to pay; every time an otherwise idle vCPUs
is interrupted by an interrupt, it will poll unnecessarily and thus
impose a little load on the host. The above results were obtained with
a mostly random value of the parameter (500000), and the load was around
1.5-2.5% CPU usage on one of the host's core for each idle guest vCPU.
The patch also adds a new stat, /sys/kernel/debug/kvm/halt_successful_poll,
that can be used to tune the parameter. It counts how many HLT
instructions received an interrupt during the polling period; each
successful poll avoids that Linux schedules the VCPU thread out and back
in, and may also avoid a likely trip to C1 and back for the physical CPU.
While the VM is idle, a Linux 4 VCPU VM halts around 10 times per second.
Of these halts, almost all are failed polls. During the benchmark,
instead, basically all halts end within the polling period, except a more
or less constant stream of 50 per second coming from vCPUs that are not
running the benchmark. The wasted time is thus very low. Things may
be slightly different for Windows VMs, which have a ~10 ms timer tick.
The effect is also visible on Marcelo's recently-introduced latency
test for the TSC deadline timer. Though of course a non-RT kernel has
awful latency bounds, the latency of the timer is around 8000-10000 clock
cycles compared to 20000-120000 without setting halt_poll_ns. For the TSC
deadline timer, thus, the effect is both a smaller average latency and
a smaller variance.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
David Hildenbrand