Skip to content

cloud-kernel
cloud-kernel

openanolis / cloud-kernel
1 年多 前同步成功

通知 160
Star 36
Fork 7
cloud-kernel
cloud-kernel
切换分支/标签
  1. 09 2月, 2015 3 次提交
  3. 06 2月, 2015 1 次提交
    • P
      kvm: add halt_poll_ns module parameter · f7819512
      由 Paolo Bonzini 提交于 
      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: NPaolo Bonzini <pbonzini@redhat.com>
      f7819512
  5. 05 2月, 2015 2 次提交
    • T
      kvm: remove KVM_MMIO_SIZE · 1c2b364b
      由 Tiejun Chen 提交于 
      After f78146b0, "KVM: Fix page-crossing MMIO", and
87da7e66, "KVM: x86: fix vcpu->mmio_fragments overflow",
actually KVM_MMIO_SIZE is gone.
Signed-off-by: NTiejun Chen <tiejun.chen@intel.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
      1c2b364b
    • J
      KVM: MIPS: Don't leak FPU/DSP to guest · f798217d
      由 James Hogan 提交于 
      The FPU and DSP are enabled via the CP0 Status CU1 and MX bits by
kvm_mips_set_c0_status() on a guest exit, presumably in case there is
active state that needs saving if pre-emption occurs. However neither of
these bits are cleared again when returning to the guest.

This effectively gives the guest access to the FPU/DSP hardware after
the first guest exit even though it is not aware of its presence,
allowing FP instructions in guest user code to intermittently actually
execute instead of trapping into the guest OS for emulation. It will
then read & manipulate the hardware FP registers which technically
belong to the user process (e.g. QEMU), or are stale from another user
process. It can also crash the guest OS by causing an FP exception, for
which a guest exception handler won't have been registered.

First lets save and disable the FPU (and MSA) state with lose_fpu(1)
before entering the guest. This simplifies the problem, especially for
when guest FPU/MSA support is added in the future, and prevents FR=1 FPU
state being live when the FR bit gets cleared for the guest, which
according to the architecture causes the contents of the FPU and vector
registers to become UNPREDICTABLE.

We can then safely remove the enabling of the FPU in
kvm_mips_set_c0_status(), since there should never be any active FPU or
MSA state to save at pre-emption, which should plug the FPU leak.

DSP state is always live rather than being lazily restored, so for that
it is simpler to just clear the MX bit again when re-entering the guest.
Signed-off-by: NJames Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Sanjay Lal <sanjayl@kymasys.com>
Cc: Gleb Natapov <gleb@kernel.org>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Cc: <stable@vger.kernel.org> # v3.10+: 044f0f03: MIPS: KVM: Deliver guest interrupts
Cc: <stable@vger.kernel.org> # v3.10+
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
      f798217d
  7. 04 2月, 2015 7 次提交
    • J
      KVM: MIPS: Disable HTW while in guest · c4c6f2ca
      由 James Hogan 提交于 
      Ensure any hardware page table walker (HTW) is disabled while in KVM
guest mode, as KVM doesn't yet set up hardware page table walking for
guest mappings so the wrong mappings would get loaded, resulting in the
guest hanging or crashing once it reaches userland.

The HTW is disabled and re-enabled around the call to
__kvm_mips_vcpu_run() which does the initial switch into guest mode and
the final switch out of guest context. Additionally it is enabled for
the duration of guest exits (i.e. kvm_mips_handle_exit()), getting
disabled again before returning back to guest or host.

In all cases the HTW is only disabled in normal kernel mode while
interrupts are disabled, so that the HTW doesn't get left disabled if
the process is preempted.
Signed-off-by: NJames Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Markos Chandras <markos.chandras@imgtec.com>
Cc: Gleb Natapov <gleb@kernel.org>
Cc: kvm@vger.kernel.org
Cc: linux-mips@linux-mips.org
Cc: <stable@vger.kernel.org> # v3.17+
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
      c4c6f2ca
    • W
      KVM: nVMX: Enable nested posted interrupt processing · 705699a1
      由 Wincy Van 提交于 
      If vcpu has a interrupt in vmx non-root mode, injecting that interrupt
requires a vmexit.  With posted interrupt processing, the vmexit
is not needed, and interrupts are fully taken care of by hardware.
In nested vmx, this feature avoids much more vmexits than non-nested vmx.

When L1 asks L0 to deliver L1's posted interrupt vector, and the target
VCPU is in non-root mode, we use a physical ipi to deliver POSTED_INTR_NV
to the target vCPU.  Using POSTED_INTR_NV avoids unexpected interrupts
if a concurrent vmexit happens and L1's vector is different with L0's.
The IPI triggers posted interrupt processing in the target physical CPU.

In case the target vCPU was not in guest mode, complete the posted
interrupt delivery on the next entry to L2.
Signed-off-by: NWincy Van <fanwenyi0529@gmail.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
      705699a1
    • W
      KVM: nVMX: Enable nested virtual interrupt delivery · 608406e2
      由 Wincy Van 提交于 
      With virtual interrupt delivery, the hardware lets KVM use a more
efficient mechanism for interrupt injection. This is an important feature
for nested VMX, because it reduces vmexits substantially and they are
much more expensive with nested virtualization.  This is especially
important for throughput-bound scenarios.
Signed-off-by: NWincy Van <fanwenyi0529@gmail.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
      608406e2
    • W
      KVM: nVMX: Enable nested apic register virtualization · 82f0dd4b
      由 Wincy Van 提交于 
      We can reduce apic register virtualization cost with this feature,
it is also a requirement for virtual interrupt delivery and posted
interrupt processing.
Signed-off-by: NWincy Van <fanwenyi0529@gmail.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
      82f0dd4b
    • W
      KVM: nVMX: Make nested control MSRs per-cpu · b9c237bb
      由 Wincy Van 提交于 
      To enable nested apicv support, we need per-cpu vmx
control MSRs:
  1. If in-kernel irqchip is enabled, we can enable nested
     posted interrupt, we should set posted intr bit in
     the nested_vmx_pinbased_ctls_high.
  2. If in-kernel irqchip is disabled, we can not enable
     nested posted interrupt, the posted intr bit
     in the nested_vmx_pinbased_ctls_high will be cleared.

Since there would be different settings about in-kernel
irqchip between VMs, different nested control MSRs
are needed.
Signed-off-by: NWincy Van <fanwenyi0529@gmail.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
      b9c237bb
    • W
      KVM: nVMX: Enable nested virtualize x2apic mode · f2b93280
      由 Wincy Van 提交于 
      When L2 is using x2apic, we can use virtualize x2apic mode to
gain higher performance, especially in apicv case.

This patch also introduces nested_vmx_check_apicv_controls
for the nested apicv patches.
Signed-off-by: NWincy Van <fanwenyi0529@gmail.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
      f2b93280
    • W
      KVM: nVMX: Prepare for using hardware MSR bitmap · 3af18d9c
      由 Wincy Van 提交于 
      Currently, if L1 enables MSR_BITMAP, we will emulate this feature, all
of L2's msr access is intercepted by L0.  Features like "virtualize
x2apic mode" require that the MSR bitmap is enabled, or the hardware
will exit and for example not virtualize the x2apic MSRs.  In order to
let L1 use these features, we need to build a merged bitmap that only
not cause a VMEXIT if 1) L1 requires that 2) the bit is not required by
the processor for APIC virtualization.

For now the guests are still run with MSR bitmap disabled, but this
patch already introduces nested_vmx_merge_msr_bitmap for future use.
Signed-off-by: NWincy Van <fanwenyi0529@gmail.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
      3af18d9c
  9. 03 2月, 2015 2 次提交
  11. 30 1月, 2015 6 次提交
  13. 29 1月, 2015 5 次提交
  15. 28 1月, 2015 1 次提交
  17. 26 1月, 2015 7 次提交
  19. 23 1月, 2015 6 次提交