KVM_CAP_X2APIC_API is a capability for features related to x2APIC
enablement. KVM_X2APIC_API_32BIT_FORMAT feature can be enabled to
extend APIC ID in get/set ioctl and MSI addresses to 32 bits.
Both are needed to support x2APIC.
The feature has to be enableable and disabled by default, because
get/set ioctl shifted and truncated APIC ID to 8 bits by using a
non-standard protocol inspired by xAPIC and the change is not
backward-compatible.
Changes to MSI addresses follow the format used by interrupt remapping
unit. The upper address word, that used to be 0, contains upper 24 bits
of the LAPIC address in its upper 24 bits. Lower 8 bits are reserved as
0. Using the upper address word is not backward-compatible either as we
didn't check that userspace zeroed the word. Reserved bits are still
not explicitly checked, but non-zero data will affect LAPIC addresses,
which will cause a bug.
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Specifically the change from hex to decimal helps correlating events.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.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>
Right now halt_poll_ns can be change during runtime. The
grow and shrink factors can only be set during module load.
Lets fix several aspects of grow shrink:
- make grow/shrink changeable by root
- make all variables unsigned int
- read the variables once to prevent races
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Fixes compilation with ppc64_defconfig.
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Tracepoint for dynamic halt_pool_ns, fired on every potential change.
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Common: Optional support for adding a small amount of polling on each HLT
instruction executed in the guest (or equivalent for other architectures).
This can improve latency up to 50% on some scenarios (e.g. O_DSYNC writes
or TCP_RR netperf tests). This also has to be enabled manually for now,
but the plan is to auto-tune this in the future.
ARM/ARM64: the highlights are support for GICv3 emulation and dirty page
tracking
s390: several optimizations and bugfixes. Also a first: a feature
exposed by KVM (UUID and long guest name in /proc/sysinfo) before
it is available in IBM's hypervisor! :)
MIPS: Bugfixes.
x86: Support for PML (page modification logging, a new feature in
Broadwell Xeons that speeds up dirty page tracking), nested virtualization
improvements (nested APICv---a nice optimization), usual round of emulation
fixes. There is also a new option to reduce latency of the TSC deadline
timer in the guest; this needs to be tuned manually.
Some commits are common between this pull and Catalin's; I see you
have already included his tree.
ARM has other conflicts where functions are added in the same place
by 3.19-rc and 3.20 patches. These are not large though, and entirely
within KVM.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)
iQEcBAABAgAGBQJU28rkAAoJEL/70l94x66DXqQH/1TDOfJIjW7P2kb0Sw7Fy1wi
cEX1KO/VFxAqc8R0E/0Wb55CXyPjQJM6xBXuFr5cUDaIjQ8ULSktL4pEwXyyv/s5
DBDkN65mriry2w5VuEaRLVcuX9Wy+tqLQXWNkEySfyb4uhZChWWHvKEcgw5SqCyg
NlpeHurYESIoNyov3jWqvBjr4OmaQENyv7t2c6q5ErIgG02V+iCux5QGbphM2IC9
LFtPKxoqhfeB2xFxTOIt8HJiXrZNwflsTejIlCl/NSEiDVLLxxHCxK2tWK/tUXMn
JfLD9ytXBWtNMwInvtFm4fPmDouv2VDyR0xnK2db+/axsJZnbxqjGu1um4Dqbak=
=7gdx
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Pull KVM update from Paolo Bonzini:
"Fairly small update, but there are some interesting new features.
Common:
Optional support for adding a small amount of polling on each HLT
instruction executed in the guest (or equivalent for other
architectures). This can improve latency up to 50% on some
scenarios (e.g. O_DSYNC writes or TCP_RR netperf tests). This
also has to be enabled manually for now, but the plan is to
auto-tune this in the future.
ARM/ARM64:
The highlights are support for GICv3 emulation and dirty page
tracking
s390:
Several optimizations and bugfixes. Also a first: a feature
exposed by KVM (UUID and long guest name in /proc/sysinfo) before
it is available in IBM's hypervisor! :)
MIPS:
Bugfixes.
x86:
Support for PML (page modification logging, a new feature in
Broadwell Xeons that speeds up dirty page tracking), nested
virtualization improvements (nested APICv---a nice optimization),
usual round of emulation fixes.
There is also a new option to reduce latency of the TSC deadline
timer in the guest; this needs to be tuned manually.
Some commits are common between this pull and Catalin's; I see you
have already included his tree.
Powerpc:
Nothing yet.
The KVM/PPC changes will come in through the PPC maintainers,
because I haven't received them yet and I might end up being
offline for some part of next week"
* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (130 commits)
KVM: ia64: drop kvm.h from installed user headers
KVM: x86: fix build with !CONFIG_SMP
KVM: x86: emulate: correct page fault error code for NoWrite instructions
KVM: Disable compat ioctl for s390
KVM: s390: add cpu model support
KVM: s390: use facilities and cpu_id per KVM
KVM: s390/CPACF: Choose crypto control block format
s390/kernel: Update /proc/sysinfo file with Extended Name and UUID
KVM: s390: reenable LPP facility
KVM: s390: floating irqs: fix user triggerable endless loop
kvm: add halt_poll_ns module parameter
kvm: remove KVM_MMIO_SIZE
KVM: MIPS: Don't leak FPU/DSP to guest
KVM: MIPS: Disable HTW while in guest
KVM: nVMX: Enable nested posted interrupt processing
KVM: nVMX: Enable nested virtual interrupt delivery
KVM: nVMX: Enable nested apic register virtualization
KVM: nVMX: Make nested control MSRs per-cpu
KVM: nVMX: Enable nested virtualize x2apic mode
KVM: nVMX: Prepare for using hardware MSR bitmap
...
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>
sparse complains about
include/trace/events/kvm.h:163:1: error: directive in argument list
include/trace/events/kvm.h:167:1: error: directive in argument list
include/trace/events/kvm.h:169:1: error: directive in argument list
and sparse is right. Preprocessing directives in an argument of a
macro are undefined behaviour as of C99 6.10.3p11.
Lets use an indirection to fix this.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Callbacks don't have to do extra computation to learn what the caller
(lvm_handle_hva_range()) knows very well. Useful for
debugging/tracing/printk/future.
Signed-off-by: Andres Lagar-Cavilla <andreslc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Currently, we call ioapic_service() immediately when we find the irq is still
active during eoi broadcast. But for real hardware, there's some delay between
the EOI writing and irq delivery. If we do not emulate this behavior, and
re-inject the interrupt immediately after the guest sends an EOI and re-enables
interrupts, a guest might spend all its time in the ISR if it has a broken
handler for a level-triggered interrupt.
Such livelock actually happens with Windows guests when resuming from
hibernation.
As there's no way to recognize the broken handle from new raised ones, this patch
delays an interrupt if 10.000 consecutive EOIs found that the interrupt was
still high. The guest can then make a little forward progress, until a proper
IRQ handler is set or until some detection routine in the guest (such as
Linux's note_interrupt()) recognizes the situation.
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Zhang Haoyu <zhanghy@sangfor.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Commits e4d57e1ee1 (KVM: Move irq notifier implementation into
eventfd.c, 2014-06-30) included the irq notifier code unconditionally
in eventfd.c, while it was under CONFIG_HAVE_KVM_IRQCHIP before.
Similarly, commit 297e21053a (KVM: Give IRQFD its own separate enabling
Kconfig option, 2014-06-30) moved code from CONFIG_HAVE_IRQ_ROUTING
to CONFIG_HAVE_KVM_IRQFD but forgot to move the pieces that used to be
under CONFIG_HAVE_KVM_IRQCHIP.
Together, this broke compilation without CONFIG_KVM_XICS. Fix by adding
or changing the #ifdefs so that they point at CONFIG_HAVE_KVM_IRQFD.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Page pinning is not mandatory in kvm async page fault processing since
after async page fault event is delivered to a guest it accesses page once
again and does its own GUP. Drop the FOLL_GET flag in GUP in async_pf
code, and do some simplifying in check/clear processing.
Suggested-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Gu zheng <guz.fnst@cn.fujitsu.com>
Signed-off-by: chai wen <chaiw.fnst@cn.fujitsu.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
The current irq_comm.c file contains pieces of code that are generic
across different irqchip implementations, as well as code that is
fully IOAPIC specific.
Split the generic bits out into irqchip.c.
Signed-off-by: Alexander Graf <agraf@suse.de>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Add a new capability, KVM_CAP_S390_CSS_SUPPORT, which will pass
intercepts for channel I/O instructions to userspace. Only I/O
instructions interacting with I/O interrupts need to be handled
in-kernel:
- TEST PENDING INTERRUPTION (tpi) dequeues and stores pending
interrupts entirely in-kernel.
- TEST SUBCHANNEL (tsch) dequeues pending interrupts in-kernel
and exits via KVM_EXIT_S390_TSCH to userspace for subchannel-
related processing.
Reviewed-by: Marcelo Tosatti <mtosatti@redhat.com>
Reviewed-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
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>
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>
Tracing 'async' and *pfn is useless, since 'async' is always true,
and '*pfn' is always "fault_pfn'
We can trace 'gva' and 'gfn' instead, it can help us to see the
life-cycle of an async_pf
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Send async page fault to a PV guest if it accesses swapped out memory.
Guest will choose another task to run upon receiving the fault.
Allow async page fault injection only when guest is in user mode since
otherwise guest may be in non-sleepable context and will not be able
to reschedule.
Vcpu will be halted if guest will fault on the same page again or if
vcpu executes kernel code.
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
If a guest accesses swapped out memory do not swap it in from vcpu thread
context. Schedule work to do swapping and put vcpu into halted state
instead.
Interrupts will still be delivered to the guest and if interrupt will
cause reschedule guest will continue to run another task.
[avi: remove call to get_user_pages_noio(), nacked by Linus; this
makes everything synchrnous again]
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This patch does:
- no need call tracepoint_synchronize_unregister() when kvm module
is unloaded since ftrace can handle it
- cleanup ftrace's macro
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Add tracepoint in msi/ioapic/pic set_irq() functions,
in IPI sending and in the point where IRQ is placed into
apic's IRR.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This allows use of the powerful ftrace infrastructure.
See Documentation/trace/ for usage information.
[avi, stephen: various build fixes]
[sheng: fix control register breakage]
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Sheng Yang <sheng@linux.intel.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Radim Krčmář