VGIC implementation.
- s390: support for trapping software breakpoints, nested virtualization
(vSIE), the STHYI opcode, initial extensions for CPU model support.
- MIPS: support for MIPS64 hosts (32-bit guests only) and lots of cleanups,
preliminary to this and the upcoming support for hardware virtualization
extensions.
- x86: support for execute-only mappings in nested EPT; reduced vmexit
latency for TSC deadline timer (by about 30%) on Intel hosts; support for
more than 255 vCPUs.
- PPC: bugfixes.
The ugly bit is the conflicts. A couple of them are simple conflicts due
to 4.7 fixes, but most of them are with other trees. There was definitely
too much reliance on Acked-by here. Some conflicts are for KVM patches
where _I_ gave my Acked-by, but the worst are for this pull request's
patches that touch files outside arch/*/kvm. KVM submaintainers should
probably learn to synchronize better with arch maintainers, with the
latter providing topic branches whenever possible instead of Acked-by.
This is what we do with arch/x86. And I should learn to refuse pull
requests when linux-next sends scary signals, even if that means that
submaintainers have to rebase their branches.
Anyhow, here's the list:
- arch/x86/kvm/vmx.c: handle_pcommit and EXIT_REASON_PCOMMIT was removed
by the nvdimm tree. This tree adds handle_preemption_timer and
EXIT_REASON_PREEMPTION_TIMER at the same place. In general all mentions
of pcommit have to go.
There is also a conflict between a stable fix and this patch, where the
stable fix removed the vmx_create_pml_buffer function and its call.
- virt/kvm/kvm_main.c: kvm_cpu_notifier was removed by the hotplug tree.
This tree adds kvm_io_bus_get_dev at the same place.
- virt/kvm/arm/vgic.c: a few final bugfixes went into 4.7 before the
file was completely removed for 4.8.
- include/linux/irqchip/arm-gic-v3.h: this one is entirely our fault;
this is a change that should have gone in through the irqchip tree and
pulled by kvm-arm. I think I would have rejected this kvm-arm pull
request. The KVM version is the right one, except that it lacks
GITS_BASER_PAGES_SHIFT.
- arch/powerpc: what a mess. For the idle_book3s.S conflict, the KVM
tree is the right one; everything else is trivial. In this case I am
not quite sure what went wrong. The commit that is causing the mess
(fd7bacbca4, "KVM: PPC: Book3S HV: Fix TB corruption in guest exit
path on HMI interrupt", 2016-05-15) touches both arch/powerpc/kernel/
and arch/powerpc/kvm/. It's large, but at 396 insertions/5 deletions
I guessed that it wasn't really possible to split it and that the 5
deletions wouldn't conflict. That wasn't the case.
- arch/s390: also messy. First is hypfs_diag.c where the KVM tree
moved some code and the s390 tree patched it. You have to reapply the
relevant part of commits 6c22c98637, plus all of e030c1125e, to
arch/s390/kernel/diag.c. Or pick the linux-next conflict
resolution from http://marc.info/?l=kvm&m=146717549531603&w=2.
Second, there is a conflict in gmap.c between a stable fix and 4.8.
The KVM version here is the correct one.
I have pushed my resolution at refs/heads/merge-20160802 (commit
3d1f53419842) at git://git.kernel.org/pub/scm/virt/kvm/kvm.git.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)
iQEcBAABAgAGBQJXoGm7AAoJEL/70l94x66DugQIAIj703ePAFepB/fCrKHkZZia
SGrsBdvAtNsOhr7FQ5qvvjLxiv/cv7CymeuJivX8H+4kuUHUllDzey+RPHYHD9X7
U6n1PdCH9F15a3IXc8tDjlDdOMNIKJixYuq1UyNZMU6NFwl00+TZf9JF8A2US65b
x/41W98ilL6nNBAsoDVmCLtPNWAqQ3lajaZELGfcqRQ9ZGKcAYOaLFXHv2YHf2XC
qIDMf+slBGSQ66UoATnYV2gAopNlWbZ7n0vO6tE2KyvhHZ1m399aBX1+k8la/0JI
69r+Tz7ZHUSFtmlmyByi5IAB87myy2WQHyAPwj+4vwJkDGPcl0TrupzbG7+T05Y=
=42ti
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Pull KVM updates from Paolo Bonzini:
- ARM: GICv3 ITS emulation and various fixes. Removal of the
old VGIC implementation.
- s390: support for trapping software breakpoints, nested
virtualization (vSIE), the STHYI opcode, initial extensions
for CPU model support.
- MIPS: support for MIPS64 hosts (32-bit guests only) and lots
of cleanups, preliminary to this and the upcoming support for
hardware virtualization extensions.
- x86: support for execute-only mappings in nested EPT; reduced
vmexit latency for TSC deadline timer (by about 30%) on Intel
hosts; support for more than 255 vCPUs.
- PPC: bugfixes.
* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (302 commits)
KVM: PPC: Introduce KVM_CAP_PPC_HTM
MIPS: Select HAVE_KVM for MIPS64_R{2,6}
MIPS: KVM: Reset CP0_PageMask during host TLB flush
MIPS: KVM: Fix ptr->int cast via KVM_GUEST_KSEGX()
MIPS: KVM: Sign extend MFC0/RDHWR results
MIPS: KVM: Fix 64-bit big endian dynamic translation
MIPS: KVM: Fail if ebase doesn't fit in CP0_EBase
MIPS: KVM: Use 64-bit CP0_EBase when appropriate
MIPS: KVM: Set CP0_Status.KX on MIPS64
MIPS: KVM: Make entry code MIPS64 friendly
MIPS: KVM: Use kmap instead of CKSEG0ADDR()
MIPS: KVM: Use virt_to_phys() to get commpage PFN
MIPS: Fix definition of KSEGX() for 64-bit
KVM: VMX: Add VMCS to CPU's loaded VMCSs before VMPTRLD
kvm: x86: nVMX: maintain internal copy of current VMCS
KVM: PPC: Book3S HV: Save/restore TM state in H_CEDE
KVM: PPC: Book3S HV: Pull out TM state save/restore into separate procedures
KVM: arm64: vgic-its: Simplify MAPI error handling
KVM: arm64: vgic-its: Make vgic_its_cmd_handle_mapi similar to other handlers
KVM: arm64: vgic-its: Turn device_id validation into generic ID validation
...
Historically a lot of these existed because we did not have
a distinction between what was modular code and what was providing
support to modules via EXPORT_SYMBOL and friends. That changed
when we forked out support for the latter into the export.h file.
This means we should be able to reduce the usage of module.h
in code that is obj-y Makefile or bool Kconfig. In the case of
kvm where it is modular, we can extend that to also include files
that are building basic support functionality but not related
to loading or registering the final module; such files also have
no need whatsoever for module.h
The advantage in removing such instances is that module.h itself
sources about 15 other headers; adding significantly to what we feed
cpp, and it can obscure what headers we are effectively using.
Since module.h was the source for init.h (for __init) and for
export.h (for EXPORT_SYMBOL) we consider each instance for the
presence of either and replace as needed.
Several instances got replaced with moduleparam.h since that was
really all that was required for those particular files.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: kvm@vger.kernel.org
Link: http://lkml.kernel.org/r/20160714001901.31603-8-paul.gortmaker@windriver.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kzalloc was replaced with kvm_kvzalloc to allow non-contiguous areas and
rcu had to be modified to cope with it.
The practical limit for KVM_MAX_VCPU_ID right now is INT_MAX, but lower
value was chosen in case there were bugs. 1023 is sufficient maximum
APIC ID for 288 VCPUs.
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK as a feature flag to
KVM_CAP_X2APIC_API.
The quirk made KVM interpret 0xff as a broadcast even in x2APIC mode.
The enableable capability is needed in order to support standard x2APIC and
remain backward compatible.
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
[Expand kvm_apic_mda comment. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
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>
LAPIC is reset in xAPIC mode and the surrounding code expects that.
KVM never resets after initialization. This patch is just for sanity.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
APIC ID should be set to the initial APIC ID when enabling LAPIC.
This only matters if the guest changes APIC ID. No sane OS does that.
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We currently always shift APIC ID as if APIC was in xAPIC mode.
x2APIC mode wants to use more bits and storing a hardware-compabible
value is the the sanest option.
KVM API to set the lapic expects that bottom 8 bits of APIC ID are in
top 8 bits of APIC_ID register, so the register needs to be shifted in
x2APIC mode.
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
x2APIC supports up to 2^32-1 LAPICs, but most guest in coming years will
probably has fewer VCPUs. Dynamic size saves memory at the cost of
turning one constant into a variable.
apic_map mutex had to be moved before allocation to avoid races with cpu
hotplug.
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Logical x2APIC IDs map injectively to physical x2APIC IDs, so we can
reuse the physical array for them. This allows us to save space by
sizing the logical maps according to the needs of xAPIC.
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
kvm_irq_delivery_to_apic_fast and kvm_intr_is_single_vcpu_fast both
compute the interrupt destination. Factor the code.
'struct kvm_lapic **dst = NULL' had to be added to silence GCC.
GCC might complain about potential NULL access in the future, because it
missed conditions that avoided uninitialized uses of dst.
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
INFO: rcu_sched detected stalls on CPUs/tasks:
1-...: (11800 GPs behind) idle=45d/140000000000000/0 softirq=0/0 fqs=21663
(detected by 0, t=65016 jiffies, g=11500, c=11499, q=719)
Task dump for CPU 1:
qemu-system-x86 R running task 0 3529 3525 0x00080808
ffff8802021791a0 ffff880212895040 0000000000000001 00007f1c2c00db40
ffff8801dd20fcd3 ffffc90002b98000 ffff8801dd20fc88 ffff8801dd20fcf8
0000000000000286 ffff8801dd2ac538 ffff8801dd20fcc0 ffffffffc06949c9
Call Trace:
? kvm_write_guest_cached+0xb9/0x160 [kvm]
? __delay+0xf/0x20
? wait_lapic_expire+0x14a/0x200 [kvm]
? kvm_arch_vcpu_ioctl_run+0xcbe/0x1b00 [kvm]
? kvm_arch_vcpu_ioctl_run+0xe34/0x1b00 [kvm]
? kvm_vcpu_ioctl+0x2d3/0x7c0 [kvm]
? __fget+0x5/0x210
? do_vfs_ioctl+0x96/0x6a0
? __fget_light+0x2a/0x90
? SyS_ioctl+0x79/0x90
? do_syscall_64+0x7c/0x1e0
? entry_SYSCALL64_slow_path+0x25/0x25
This can be reproduced readily by running a full dynticks guest(since hrtimer
in guest is heavily used) w/ lapic_timer_advance disabled.
If fail to program hardware preemption timer, we will fallback to hrtimer based
method, however, a previous programmed preemption timer miss to cancel in this
scenario which results in one hardware preemption timer and one hrtimer emulated
tsc deadline timer run simultaneously. So sometimes the target guest deadline
tsc is earlier than guest tsc, which leads to the computation in vmx_set_hv_timer
can underflow and cause delta_tsc to be set a huge value, then host soft lockup
as above.
This patch fix it by cancelling the previous programmed preemption timer if there
is once we failed to program the new preemption timer and fallback to hrtimer
based method.
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Yunhong Jiang <yunhong.jiang@intel.com>
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The host timer which emulates the guest LAPIC TSC deadline
timer has its expiration diminished by lapic_timer_advance_ns
nanoseconds. Therefore if, at wait_lapic_expire, a difference
larger than lapic_timer_advance_ns is encountered, delay at most
lapic_timer_advance_ns.
This fixes a problem where the guest can cause the host
to delay for large amounts of time.
Reported-by: Alan Jenkins <alan.christopher.jenkins@gmail.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The VMX preemption timer can be used to virtualize the TSC deadline timer.
The VMX preemption timer is armed when the vCPU is running, and a VMExit
will happen if the virtual TSC deadline timer expires.
When the vCPU thread is blocked because of HLT, KVM will switch to use
an hrtimer, and then go back to the VMX preemption timer when the vCPU
thread is unblocked.
This solution avoids the complex OS's hrtimer system, and the host
timer interrupt handling cost, replacing them with a little math
(for guest->host TSC and host TSC->preemption timer conversion)
and a cheaper VMexit. This benefits latency for isolated pCPUs.
[A word about performance... Yunhong reported a 30% reduction in average
latency from cyclictest. I made a similar test with tscdeadline_latency
from kvm-unit-tests, and measured
- ~20 clock cycles loss (out of ~3200, so less than 1% but still
statistically significant) in the worst case where the test halts
just after programming the TSC deadline timer
- ~800 clock cycles gain (25% reduction in latency) in the best case
where the test busy waits.
I removed the VMX bits from Yunhong's patch, to concentrate them in the
next patch - Paolo]
Signed-off-by: Yunhong Jiang <yunhong.jiang@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The function to start the tsc deadline timer virtualization will be used
also by the pre_block hook when we use the preemption timer; change it
to a separate function. No logic changes.
Signed-off-by: Yunhong Jiang <yunhong.jiang@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Neither APICv nor AVIC actually need the first argument of
hwapic_isr_update, but the vCPU makes more sense than passing the
pointer to the whole virtual machine! In fact in the APICv case it's
just happening that the vCPU is used implicitly, through the loaded VMCS.
The second argument instead is named differently, make it consistent.
Reviewed-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Adding kvm_x86_ops hooks to allow APICv to do post state restore.
This is required to support VM save and restore feature.
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Rename kvm_apic_get_reg to kvm_lapic_get_reg to be consistent with
the existing kvm_lapic_set_reg counterpart.
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When a vCPU runs on a nohz_full core, the hrtimer used by
the lapic emulation code can be migrated to another core.
When this happens, it's possible to observe milisecond
latency when delivering timer IRQs to KVM guests.
The huge latency is mainly due to the fact that
apic_timer_fn() expects to run during a kvm exit. It
sets KVM_REQ_PENDING_TIMER and let it be handled on kvm
entry. However, if the timer fires on a different core,
we have to wait until the next kvm exit for the guest
to see KVM_REQ_PENDING_TIMER set.
This problem became visible after commit 9642d18ee. This
commit changed the timer migration code to always attempt
to migrate timers away from nohz_full cores. While it's
discussable if this is correct/desirable (I don't think
it is), it's clear that the lapic emulation code has
a requirement on firing the hrtimer in the same core
where it was started. This is achieved by making the
hrtimer pinned.
Lastly, note that KVM has code to migrate timers when a
vCPU is scheduled to run in different core. However, this
forced migration may fail. When this happens, we can have
the same problem. If we want 100% correctness, we'll have
to modify apic_timer_fn() to cause a kvm exit when it runs
on a different core than the vCPU. Not sure if this is
possible.
Here's a reproducer for the issue being fixed:
1. Set all cores but core0 to be nohz_full cores
2. Start a guest with a single vCPU
3. Trace apic_timer_fn() and kvm_inject_apic_timer_irqs()
You'll see that apic_timer_fn() will run in core0 while
kvm_inject_apic_timer_irqs() runs in a different core. If
you get both on core0, try running a program that takes 100%
of the CPU and pin it to core0 to force the vCPU out.
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.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
...
This allows backtracking later in case the rtc irq has been
moved to another vcpu/vector.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Currently this is a bitmap which tracks which CPUs we expect
an EOI from. Move this bitmap to a struct so that we can
track additional information there.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.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>
Smatch noticed a NULL dereference in kvm_intr_is_single_vcpu_fast that
happens if VM already warned about invalid lowest-priority interrupt.
Create a function for common code while fixing it.
Fixes: 6228a0da80 ("KVM: x86: Add lowest-priority support for vt-d posted-interrupts")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Different pieces of code checked for vcpu->arch.apic being (non-)NULL,
or used kvm_vcpu_has_lapic (more optimized) or lapic_in_kernel.
Replace everything with lapic_in_kernel's name and kvm_vcpu_has_lapic's
implementation.
Reviewed-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Do for kvm_cpu_has_pending_timer and kvm_inject_pending_timer_irqs
what the other irq.c routines have been doing.
Reviewed-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Usually the in-kernel APIC's existence is checked in the caller. Do not
bother checking it again in lapic.c.
Reviewed-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use vector-hashing to deliver lowest-priority interrupts for
VT-d posted-interrupts. This patch extends kvm_intr_is_single_vcpu()
to support lowest-priority handling.
Signed-off-by: Feng Wu <feng.wu@intel.com>
Reviewed-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use vector-hashing to deliver lowest-priority interrupts, As an
example, modern Intel CPUs in server platform use this method to
handle lowest-priority interrupts.
Signed-off-by: Feng Wu <feng.wu@intel.com>
Reviewed-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
SynIC (synthetic interrupt controller) is a lapic extension,
which is controlled via MSRs and maintains for each vCPU
- 16 synthetic interrupt "lines" (SINT's); each can be configured to
trigger a specific interrupt vector optionally with auto-EOI
semantics
- a message page in the guest memory with 16 256-byte per-SINT message
slots
- an event flag page in the guest memory with 16 2048-bit per-SINT
event flag areas
The host triggers a SINT whenever it delivers a new message to the
corresponding slot or flips an event flag bit in the corresponding area.
The guest informs the host that it can try delivering a message by
explicitly asserting EOI in lapic or writing to End-Of-Message (EOM)
MSR.
The userspace (qemu) triggers interrupts and receives EOM notifications
via irqfd with resampler; for that, a GSI is allocated for each
configured SINT, and irq_routing api is extended to support GSI-SINT
mapping.
Changes v4:
* added activation of SynIC by vcpu KVM_ENABLE_CAP
* added per SynIC active flag
* added deactivation of APICv upon SynIC activation
Changes v3:
* added KVM_CAP_HYPERV_SYNIC and KVM_IRQ_ROUTING_HV_SINT notes into
docs
Changes v2:
* do not use posted interrupts for Hyper-V SynIC AutoEOI vectors
* add Hyper-V SynIC vectors into EOI exit bitmap
* Hyper-V SyniIC SINT msr write logic simplified
Signed-off-by: Andrey Smetanin <asmetanin@virtuozzo.com>
Reviewed-by: Roman Kagan <rkagan@virtuozzo.com>
Signed-off-by: Denis V. Lunev <den@openvz.org>
CC: Gleb Natapov <gleb@kernel.org>
CC: Paolo Bonzini <pbonzini@redhat.com>
CC: Roman Kagan <rkagan@virtuozzo.com>
CC: Denis V. Lunev <den@openvz.org>
CC: qemu-devel@nongnu.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The decision on whether to use hardware APIC virtualization used to be
taken globally, based on the availability of the feature in the CPU
and the value of a module parameter.
However, under certain circumstances we want to control it on per-vcpu
basis. In particular, when the userspace activates HyperV synthetic
interrupt controller (SynIC), APICv has to be disabled as it's
incompatible with SynIC auto-EOI behavior.
To achieve that, introduce 'apicv_active' flag on struct
kvm_vcpu_arch, and kvm_vcpu_deactivate_apicv() function to turn APICv
off. The flag is initialized based on the module parameter and CPU
capability, and consulted whenever an APICv-specific action is
performed.
Signed-off-by: Andrey Smetanin <asmetanin@virtuozzo.com>
Reviewed-by: Roman Kagan <rkagan@virtuozzo.com>
Signed-off-by: Denis V. Lunev <den@openvz.org>
CC: Gleb Natapov <gleb@kernel.org>
CC: Paolo Bonzini <pbonzini@redhat.com>
CC: Roman Kagan <rkagan@virtuozzo.com>
CC: Denis V. Lunev <den@openvz.org>
CC: qemu-devel@nongnu.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The function to determine if the vector is handled by ioapic used to
rely on the fact that only ioapic-handled vectors were set up to
cause vmexits when virtual apic was in use.
We're going to break this assumption when introducing Hyper-V
synthetic interrupts: they may need to cause vmexits too.
To achieve that, introduce a new bitmap dedicated specifically for
ioapic-handled vectors, and populate EOI exit bitmap from it for now.
Signed-off-by: Andrey Smetanin <asmetanin@virtuozzo.com>
Reviewed-by: Roman Kagan <rkagan@virtuozzo.com>
Signed-off-by: Denis V. Lunev <den@openvz.org>
CC: Gleb Natapov <gleb@kernel.org>
CC: Paolo Bonzini <pbonzini@redhat.com>
CC: Roman Kagan <rkagan@virtuozzo.com>
CC: Denis V. Lunev <den@openvz.org>
CC: qemu-devel@nongnu.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Both VMX and SVM scales the host TSC in the same way in call-back
read_l1_tsc(), so this patch moves the scaling logic from call-back
read_l1_tsc() to a common function kvm_read_l1_tsc().
Signed-off-by: Haozhong Zhang <haozhong.zhang@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
BSP doesn't get INIT so its apic_arb_prio isn't zeroed after reboot.
BSP won't get lowest priority interrupts until other VCPUs get enough
interrupts to match their pre-reboot apic_arb_prio.
That behavior doesn't fit into KVM's round-robin-like interpretation of
lowest priority delivery ... userspace should KVM_SET_LAPIC on reset, so
just zero apic_arb_prio there.
Reported-by: Yuki Shibuya <shibuya.yk@ncos.nec.co.jp>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
After moving PIR to IRR, the interrupt needs to be delivered manually.
Reported-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch defines a new interface kvm_intr_is_single_vcpu(),
which can returns whether the interrupt is for single-CPU or not.
It is used by VT-d PI, since now we only support single-CPU
interrupts, For lowest-priority interrupts, if user configures
it via /proc/irq or uses irqbalance to make it single-CPU, we
can use PI to deliver the interrupts to it. Full functionality
of lowest-priority support will be added later.
Signed-off-by: Feng Wu <feng.wu@intel.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
In order to support a userspace IOAPIC interacting with an in kernel
APIC, the EOI exit bitmaps need to be configurable.
If the IOAPIC is in userspace (i.e. the irqchip has been split), the
EOI exit bitmaps will be set whenever the GSI Routes are configured.
In particular, for the low MSI routes are reservable for userspace
IOAPICs. For these MSI routes, the EOI Exit bit corresponding to the
destination vector of the route will be set for the destination VCPU.
The intention is for the userspace IOAPICs to use the reservable MSI
routes to inject interrupts into the guest.
This is a slight abuse of the notion of an MSI Route, given that MSIs
classically bypass the IOAPIC. It might be worthwhile to add an
additional route type to improve clarity.
Compile tested for Intel x86.
Signed-off-by: Steve Rutherford <srutherford@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Adds KVM_EXIT_IOAPIC_EOI which allows the kernel to EOI
level-triggered IOAPIC interrupts.
Uses a per VCPU exit bitmap to decide whether or not the IOAPIC needs
to be informed (which is identical to the EOI_EXIT_BITMAP field used
by modern x86 processors, but can also be used to elide kvm IOAPIC EOI
exits on older processors).
[Note: A prototype using ResampleFDs found that decoupling the EOI
from the VCPU's thread made it possible for the VCPU to not see a
recent EOI after reentering the guest. This does not match real
hardware.]
Compile tested for Intel x86.
Signed-off-by: Steve Rutherford <srutherford@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
First patch in a series which enables the relocation of the
PIC/IOAPIC to userspace.
Adds capability KVM_CAP_SPLIT_IRQCHIP;
KVM_CAP_SPLIT_IRQCHIP enables the construction of LAPICs without the
rest of the irqchip.
Compile tested for x86.
Signed-off-by: Steve Rutherford <srutherford@google.com>
Suggested-by: Andrew Honig <ahonig@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Avoid pointer chasing and memory barriers, and simplify the code
when split irqchip (LAPIC in kernel, IOAPIC/PIC in userspace)
is introduced.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We can reuse the algorithm that computes the EOI exit bitmap to figure
out which vectors are handled by the IOAPIC. The only difference
between the two is for edge-triggered interrupts other than IRQ8
that have no notifiers active; however, the IOAPIC does not have to
do anything special for these interrupts anyway.
This again limits the interactions between the IOAPIC and the LAPIC,
making it easier to move the former to userspace.
Inspired by a patch from Steve Rutherford.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Do not compute TMR in advance. Instead, set the TMR just before the interrupt
is accepted into the IRR. This limits the coupling between IOAPIC and LAPIC.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Pull x86 asm changes from Ingo Molnar:
"The biggest changes in this cycle were:
- Revamp, simplify (and in some cases fix) Time Stamp Counter (TSC)
primitives. (Andy Lutomirski)
- Add new, comprehensible entry and exit handlers written in C.
(Andy Lutomirski)
- vm86 mode cleanups and fixes. (Brian Gerst)
- 32-bit compat code cleanups. (Brian Gerst)
The amount of simplification in low level assembly code is already
palpable:
arch/x86/entry/entry_32.S | 130 +----
arch/x86/entry/entry_64.S | 197 ++-----
but more simplifications are planned.
There's also the usual laudry mix of low level changes - see the
changelog for details"
* 'x86-asm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (83 commits)
x86/asm: Drop repeated macro of X86_EFLAGS_AC definition
x86/asm/msr: Make wrmsrl() a function
x86/asm/delay: Introduce an MWAITX-based delay with a configurable timer
x86/asm: Add MONITORX/MWAITX instruction support
x86/traps: Weaken context tracking entry assertions
x86/asm/tsc: Add rdtscll() merge helper
selftests/x86: Add syscall_nt selftest
selftests/x86: Disable sigreturn_64
x86/vdso: Emit a GNU hash
x86/entry: Remove do_notify_resume(), syscall_trace_leave(), and their TIF masks
x86/entry/32: Migrate to C exit path
x86/entry/32: Remove 32-bit syscall audit optimizations
x86/vm86: Rename vm86->v86flags and v86mask
x86/vm86: Rename vm86->vm86_info to user_vm86
x86/vm86: Clean up vm86.h includes
x86/vm86: Move the vm86 IRQ definitions to vm86.h
x86/vm86: Use the normal pt_regs area for vm86
x86/vm86: Eliminate 'struct kernel_vm86_struct'
x86/vm86: Move fields from 'struct kernel_vm86_struct' to 'struct vm86'
x86/vm86: Move vm86 fields out of 'thread_struct'
...
This fixes error handling in the function kvm_lapic_sync_from_vapic
by checking if the call to kvm_read_guest_cached has returned a
error code to signal to its caller the call to this function has
failed and due to this we must immediately return to the caller
of kvm_lapic_sync_from_vapic to avoid incorrectly call apic_set_tpc
if a error has occurred here.
Signed-off-by: Nicholas Krause <xerofoify@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Make them clearly architecture-dependent; the capability is valid for
all architectures, but the argument is not.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
