openwrt/staging/blogic.git
5 years agoKVM: nVMX: Check GUEST_IA32_PERF_GLOBAL_CTRL on VM-Entry
Oliver Upton [Thu, 14 Nov 2019 00:17:16 +0000 (16:17 -0800)]
KVM: nVMX: Check GUEST_IA32_PERF_GLOBAL_CTRL on VM-Entry

Add condition to nested_vmx_check_guest_state() to check the validity of
GUEST_IA32_PERF_GLOBAL_CTRL. Per Intel's SDM Vol 3 26.3.1.1:

  If the "load IA32_PERF_GLOBAL_CTRL" VM-entry control is 1, bits
  reserved in the IA32_PERF_GLOBAL_CTRL MSR must be 0 in the field for that
  register.

Suggested-by: Jim Mattson <jmattson@google.com>
Co-developed-by: Krish Sadhukhan <krish.sadhukhan@oracle.com>
Signed-off-by: Krish Sadhukhan <krish.sadhukhan@oracle.com>
Signed-off-by: Oliver Upton <oupton@google.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Peter Shier <pshier@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Add helper to check reserved bits in IA32_PERF_GLOBAL_CTRL
Oliver Upton [Thu, 14 Nov 2019 00:17:15 +0000 (16:17 -0800)]
KVM: VMX: Add helper to check reserved bits in IA32_PERF_GLOBAL_CTRL

Create a helper function to check the validity of a proposed value for
IA32_PERF_GLOBAL_CTRL from the existing check in intel_pmu_set_msr().

Per Intel's SDM, the reserved bits in IA32_PERF_GLOBAL_CTRL must be
cleared for the corresponding host/guest state fields.

Suggested-by: Jim Mattson <jmattson@google.com>
Co-developed-by: Krish Sadhukhan <krish.sadhukhan@oracle.com>
Signed-off-by: Krish Sadhukhan <krish.sadhukhan@oracle.com>
Signed-off-by: Oliver Upton <oupton@google.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Reviewed-by: Peter Shier <pshier@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoselftests: kvm: Simplify loop in kvm_create_max_vcpus test
Wainer dos Santos Moschetta [Tue, 12 Nov 2019 14:21:11 +0000 (09:21 -0500)]
selftests: kvm: Simplify loop in kvm_create_max_vcpus test

On kvm_create_max_vcpus test remove unneeded local
variable in the loop that add vcpus to the VM.

Signed-off-by: Wainer dos Santos Moschetta <wainersm@redhat.com>
Reviewed-by: Krish Sadhukhan <krish.sadhukhan@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Optimization: Requst TLB flush in fast_cr3_switch() instead of do it directly
Liran Alon [Tue, 12 Nov 2019 18:33:00 +0000 (20:33 +0200)]
KVM: x86: Optimization: Requst TLB flush in fast_cr3_switch() instead of do it directly

When KVM emulates a nested VMEntry (L1->L2 VMEntry), it switches mmu root
page. If nEPT is used, this will happen from
kvm_init_shadow_ept_mmu()->__kvm_mmu_new_cr3() and otherwise it will
happpen from nested_vmx_load_cr3()->kvm_mmu_new_cr3(). Either case,
__kvm_mmu_new_cr3() will use fast_cr3_switch() in attempt to switch to a
previously cached root page.

In case fast_cr3_switch() finds a matching cached root page, it will
set it in mmu->root_hpa and request KVM_REQ_LOAD_CR3 such that on
next entry to guest, KVM will set root HPA in appropriate hardware
fields (e.g. vmcs->eptp). In addition, fast_cr3_switch() calls
kvm_x86_ops->tlb_flush() in order to flush TLB as MMU root page
was replaced.

This works as mmu->root_hpa, which vmx_flush_tlb() use, was
already replaced in cached_root_available(). However, this may
result in unnecessary INVEPT execution because a KVM_REQ_TLB_FLUSH
may have already been requested. For example, by prepare_vmcs02()
in case L1 don't use VPID.

Therefore, change fast_cr3_switch() to just request TLB flush on
next entry to guest.

Reviewed-by: Bhavesh Davda <bhavesh.davda@oracle.com>
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86/vPMU: Add lazy mechanism to release perf_event per vPMC
Like Xu [Sun, 27 Oct 2019 10:52:43 +0000 (18:52 +0800)]
KVM: x86/vPMU: Add lazy mechanism to release perf_event per vPMC

Currently, a host perf_event is created for a vPMC functionality emulation.
It’s unpredictable to determine if a disabled perf_event will be reused.
If they are disabled and are not reused for a considerable period of time,
those obsolete perf_events would increase host context switch overhead that
could have been avoided.

If the guest doesn't WRMSR any of the vPMC's MSRs during an entire vcpu
sched time slice, and its independent enable bit of the vPMC isn't set,
we can predict that the guest has finished the use of this vPMC, and then
do request KVM_REQ_PMU in kvm_arch_sched_in and release those perf_events
in the first call of kvm_pmu_handle_event() after the vcpu is scheduled in.

This lazy mechanism delays the event release time to the beginning of the
next scheduled time slice if vPMC's MSRs aren't changed during this time
slice. If guest comes back to use this vPMC in next time slice, a new perf
event would be re-created via perf_event_create_kernel_counter() as usual.

Suggested-by: Wei Wang <wei.w.wang@intel.com>
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Like Xu <like.xu@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86/vPMU: Reuse perf_event to avoid unnecessary pmc_reprogram_counter
Like Xu [Sun, 27 Oct 2019 10:52:42 +0000 (18:52 +0800)]
KVM: x86/vPMU: Reuse perf_event to avoid unnecessary pmc_reprogram_counter

The perf_event_create_kernel_counter() in the pmc_reprogram_counter() is
a heavyweight and high-frequency operation, especially when host disables
the watchdog (maximum 21000000 ns) which leads to an unacceptable latency
of the guest NMI handler. It limits the use of vPMUs in the guest.

When a vPMC is fully enabled, the legacy reprogram_*_counter() would stop
and release its existing perf_event (if any) every time EVEN in most cases
almost the same requested perf_event will be created and configured again.

For each vPMC, if the reuqested config ('u64 eventsel' for gp and 'u8 ctrl'
for fixed) is the same as its current config AND a new sample period based
on pmc->counter is accepted by host perf interface, the current event could
be reused safely as a new created one does. Otherwise, do release the
undesirable perf_event and reprogram a new one as usual.

It's light-weight to call pmc_pause_counter (disable, read and reset event)
and pmc_resume_counter (recalibrate period and re-enable event) as guest
expects instead of release-and-create again on any condition. Compared to
use the filterable event->attr or hw.config, a new 'u64 current_config'
field is added to save the last original programed config for each vPMC.

Based on this implementation, the number of calls to pmc_reprogram_counter
is reduced by ~82.5% for a gp sampling event and ~99.9% for a fixed event.
In the usage of multiplexing perf sampling mode, the average latency of the
guest NMI handler is reduced from 104923 ns to 48393 ns (~2.16x speed up).
If host disables watchdog, the minimum latecy of guest NMI handler could be
speed up at ~3413x (from 20407603 to 5979 ns) and at ~786x in the average.

Suggested-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Like Xu <like.xu@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86/vPMU: Introduce a new kvm_pmu_ops->msr_idx_to_pmc callback
Like Xu [Sun, 27 Oct 2019 10:52:41 +0000 (18:52 +0800)]
KVM: x86/vPMU: Introduce a new kvm_pmu_ops->msr_idx_to_pmc callback

Introduce a new callback msr_idx_to_pmc that returns a struct kvm_pmc*,
and change kvm_pmu_is_valid_msr to return ".msr_idx_to_pmc(vcpu, msr) ||
.is_valid_msr(vcpu, msr)" and AMD just returns false from .is_valid_msr.

Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Like Xu <like.xu@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86/vPMU: Rename pmu_ops callbacks from msr_idx to rdpmc_ecx
Like Xu [Sun, 27 Oct 2019 10:52:40 +0000 (18:52 +0800)]
KVM: x86/vPMU: Rename pmu_ops callbacks from msr_idx to rdpmc_ecx

The leagcy pmu_ops->msr_idx_to_pmc is only called in kvm_pmu_rdpmc, so
this function actually receives the contents of ECX before RDPMC, and
translates it to a kvm_pmc. Let's clarify its semantic by renaming the
existing msr_idx_to_pmc to rdpmc_ecx_to_pmc, and is_valid_msr_idx to
is_valid_rdpmc_ecx; likewise for the wrapper kvm_pmu_is_valid_msr_idx.

Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Like Xu <like.xu@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoperf/core: Provide a kernel-internal interface to pause perf_event
Like Xu [Sun, 27 Oct 2019 10:52:39 +0000 (18:52 +0800)]
perf/core: Provide a kernel-internal interface to pause perf_event

Exporting perf_event_pause() as an external accessor for kernel users (such
as KVM) who may do both disable perf_event and read count with just one
time to hold perf_event_ctx_lock. Also the value could be reset optionally.

Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Like Xu <like.xu@linux.intel.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoperf/core: Provide a kernel-internal interface to recalibrate event period
Like Xu [Sun, 27 Oct 2019 10:52:38 +0000 (18:52 +0800)]
perf/core: Provide a kernel-internal interface to recalibrate event period

Currently, perf_event_period() is used by user tools via ioctl. Based on
naming convention, exporting perf_event_period() for kernel users (such
as KVM) who may recalibrate the event period for their assigned counter
according to their requirements.

The perf_event_period() is an external accessor, just like the
perf_event_{en,dis}able() and should thus use perf_event_ctx_lock().

Suggested-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Like Xu <like.xu@linux.intel.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: nVMX: Update vmcs01 TPR_THRESHOLD if L2 changed L1 TPR
Liran Alon [Mon, 11 Nov 2019 12:30:55 +0000 (14:30 +0200)]
KVM: nVMX: Update vmcs01 TPR_THRESHOLD if L2 changed L1 TPR

When L1 don't use TPR-Shadow to run L2, L0 configures vmcs02 without
TPR-Shadow and install intercepts on CR8 access (load and store).

If L1 do not intercept L2 CR8 access, L0 intercepts on those accesses
will emulate load/store on L1's LAPIC TPR. If in this case L2 lowers
TPR such that there is now an injectable interrupt to L1,
apic_update_ppr() will request a KVM_REQ_EVENT which will trigger a call
to update_cr8_intercept() to update TPR-Threshold to highest pending IRR
priority.

However, this update to TPR-Threshold is done while active vmcs is
vmcs02 instead of vmcs01. Thus, when later at some point L0 will
emulate an exit from L2 to L1, L1 will still run with high
TPR-Threshold. This will result in every VMEntry to L1 to immediately
exit on TPR_BELOW_THRESHOLD and continue to do so infinitely until
some condition will cause KVM_REQ_EVENT to be set.
(Note that TPR_BELOW_THRESHOLD exit handler do not set KVM_REQ_EVENT
until apic_update_ppr() will notice a new injectable interrupt for PPR)

To fix this issue, change update_cr8_intercept() such that if L2 lowers
L1's TPR in a way that requires to lower L1's TPR-Threshold, save update
to TPR-Threshold and apply it to vmcs01 when L0 emulates an exit from
L2 to L1.

Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Refactor update_cr8_intercept()
Liran Alon [Mon, 11 Nov 2019 12:30:54 +0000 (14:30 +0200)]
KVM: VMX: Refactor update_cr8_intercept()

No functional changes.

Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: SVM: Remove check if APICv enabled in SVM update_cr8_intercept() handler
Liran Alon [Mon, 11 Nov 2019 12:26:21 +0000 (14:26 +0200)]
KVM: SVM: Remove check if APICv enabled in SVM update_cr8_intercept() handler

This check is unnecessary as x86 update_cr8_intercept() which calls
this VMX/SVM specific callback already performs this check.

Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: APIC: add helper func to remove duplicate code in kvm_pv_send_ipi
Miaohe Lin [Sat, 9 Nov 2019 09:46:49 +0000 (17:46 +0800)]
KVM: APIC: add helper func to remove duplicate code in kvm_pv_send_ipi

There are some duplicate code in kvm_pv_send_ipi when deal with ipi
bitmap. Add helper func to remove it, and eliminate odd out label,
get rid of unnecessary kvm_lapic_irq field init and so on.

Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: X86: avoid unused setup_syscalls_segments call when SYSCALL check failed
Miaohe Lin [Sat, 9 Nov 2019 08:58:54 +0000 (16:58 +0800)]
KVM: X86: avoid unused setup_syscalls_segments call when SYSCALL check failed

When SYSCALL/SYSENTER ability check failed, cs and ss is inited but
remain not used. Delay initializing cs and ss until SYSCALL/SYSENTER
ability check passed.

Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: MMIO: get rid of odd out_err label in kvm_coalesced_mmio_init
Miaohe Lin [Sat, 9 Nov 2019 08:08:20 +0000 (16:08 +0800)]
KVM: MMIO: get rid of odd out_err label in kvm_coalesced_mmio_init

The out_err label and var ret is unnecessary, clean them up.

Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Consume pending LAPIC INIT event when exit on INIT_SIGNAL
Liran Alon [Mon, 11 Nov 2019 12:16:05 +0000 (14:16 +0200)]
KVM: VMX: Consume pending LAPIC INIT event when exit on INIT_SIGNAL

Intel SDM section 25.2 OTHER CAUSES OF VM EXITS specifies the following
on INIT signals: "Such exits do not modify register state or clear pending
events as they would outside of VMX operation."

When commit 4b9852f4f389 ("KVM: x86: Fix INIT signal handling in various CPU states")
was applied, I interepted above Intel SDM statement such that
INIT_SIGNAL exit don’t consume the LAPIC INIT pending event.

However, when Nadav Amit run matching kvm-unit-test on a bare-metal
machine, it turned out my interpetation was wrong. i.e. INIT_SIGNAL
exit does consume the LAPIC INIT pending event.
(See: https://www.spinics.net/lists/kvm/msg196757.html)

Therefore, fix KVM code to behave as observed on bare-metal.

Fixes: 4b9852f4f389 ("KVM: x86: Fix INIT signal handling in various CPU states")
Reported-by: Nadav Amit <nadav.amit@gmail.com>
Reviewed-by: Mihai Carabas <mihai.carabas@oracle.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Prevent set vCPU into INIT/SIPI_RECEIVED state when INIT are latched
Liran Alon [Mon, 11 Nov 2019 09:16:40 +0000 (11:16 +0200)]
KVM: x86: Prevent set vCPU into INIT/SIPI_RECEIVED state when INIT are latched

Commit 4b9852f4f389 ("KVM: x86: Fix INIT signal handling in various CPU states")
fixed KVM to also latch pending LAPIC INIT event when vCPU is in VMX
operation.

However, current API of KVM_SET_MP_STATE allows userspace to put vCPU
into KVM_MP_STATE_SIPI_RECEIVED or KVM_MP_STATE_INIT_RECEIVED even when
vCPU is in VMX operation.

Fix this by introducing a util method to check if vCPU state latch INIT
signals and use it in KVM_SET_MP_STATE handler.

Fixes: 4b9852f4f389 ("KVM: x86: Fix INIT signal handling in various CPU states")
Reported-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Mihai Carabas <mihai.carabas@oracle.com>
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Evaluate latched_init in KVM_SET_VCPU_EVENTS when vCPU not in SMM
Liran Alon [Mon, 11 Nov 2019 09:16:39 +0000 (11:16 +0200)]
KVM: x86: Evaluate latched_init in KVM_SET_VCPU_EVENTS when vCPU not in SMM

Commit 4b9852f4f389 ("KVM: x86: Fix INIT signal handling in various CPU states")
fixed KVM to also latch pending LAPIC INIT event when vCPU is in VMX
operation.

However, current API of KVM_SET_VCPU_EVENTS defines this field as
part of SMM state and only set pending LAPIC INIT event if vCPU is
specified to be in SMM mode (events->smi.smm is set).

Change KVM_SET_VCPU_EVENTS handler to set pending LAPIC INIT event
by latched_init field regardless of if vCPU is in SMM mode or not.

Fixes: 4b9852f4f389 ("KVM: x86: Fix INIT signal handling in various CPU states")
Reviewed-by: Mihai Carabas <mihai.carabas@oracle.com>
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agox86: retpolines: eliminate retpoline from msr event handlers
Andrea Arcangeli [Mon, 4 Nov 2019 23:00:01 +0000 (18:00 -0500)]
x86: retpolines: eliminate retpoline from msr event handlers

It's enough to check the value and issue the direct call.

After this commit is applied, here the most common retpolines executed
under a high resolution timer workload in the guest on a VMX host:

[..]
@[
    trace_retpoline+1
    __trace_retpoline+30
    __x86_indirect_thunk_rax+33
    do_syscall_64+89
    entry_SYSCALL_64_after_hwframe+68
]: 267
@[]: 2256
@[
    trace_retpoline+1
    __trace_retpoline+30
    __x86_indirect_thunk_rax+33
    __kvm_wait_lapic_expire+284
    vmx_vcpu_run.part.97+1091
    vcpu_enter_guest+377
    kvm_arch_vcpu_ioctl_run+261
    kvm_vcpu_ioctl+559
    do_vfs_ioctl+164
    ksys_ioctl+96
    __x64_sys_ioctl+22
    do_syscall_64+89
    entry_SYSCALL_64_after_hwframe+68
]: 2390
@[]: 33410

@total: 315707

Note the highest hit above is __delay so probably not worth optimizing
even if it would be more frequent than 2k hits per sec.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: retpolines: x86: eliminate retpoline from svm.c exit handlers
Andrea Arcangeli [Mon, 4 Nov 2019 23:00:00 +0000 (18:00 -0500)]
KVM: retpolines: x86: eliminate retpoline from svm.c exit handlers

It's enough to check the exit value and issue a direct call to avoid
the retpoline for all the common vmexit reasons.

After this commit is applied, here the most common retpolines executed
under a high resolution timer workload in the guest on a SVM host:

[..]
@[
    trace_retpoline+1
    __trace_retpoline+30
    __x86_indirect_thunk_rax+33
    ktime_get_update_offsets_now+70
    hrtimer_interrupt+131
    smp_apic_timer_interrupt+106
    apic_timer_interrupt+15
    start_sw_timer+359
    restart_apic_timer+85
    kvm_set_msr_common+1497
    msr_interception+142
    vcpu_enter_guest+684
    kvm_arch_vcpu_ioctl_run+261
    kvm_vcpu_ioctl+559
    do_vfs_ioctl+164
    ksys_ioctl+96
    __x64_sys_ioctl+22
    do_syscall_64+89
    entry_SYSCALL_64_after_hwframe+68
]: 1940
@[
    trace_retpoline+1
    __trace_retpoline+30
    __x86_indirect_thunk_r12+33
    force_qs_rnp+217
    rcu_gp_kthread+1270
    kthread+268
    ret_from_fork+34
]: 4644
@[]: 25095
@[
    trace_retpoline+1
    __trace_retpoline+30
    __x86_indirect_thunk_rax+33
    lapic_next_event+28
    clockevents_program_event+148
    hrtimer_start_range_ns+528
    start_sw_timer+356
    restart_apic_timer+85
    kvm_set_msr_common+1497
    msr_interception+142
    vcpu_enter_guest+684
    kvm_arch_vcpu_ioctl_run+261
    kvm_vcpu_ioctl+559
    do_vfs_ioctl+164
    ksys_ioctl+96
    __x64_sys_ioctl+22
    do_syscall_64+89
    entry_SYSCALL_64_after_hwframe+68
]: 41474
@[
    trace_retpoline+1
    __trace_retpoline+30
    __x86_indirect_thunk_rax+33
    clockevents_program_event+148
    hrtimer_start_range_ns+528
    start_sw_timer+356
    restart_apic_timer+85
    kvm_set_msr_common+1497
    msr_interception+142
    vcpu_enter_guest+684
    kvm_arch_vcpu_ioctl_run+261
    kvm_vcpu_ioctl+559
    do_vfs_ioctl+164
    ksys_ioctl+96
    __x64_sys_ioctl+22
    do_syscall_64+89
    entry_SYSCALL_64_after_hwframe+68
]: 41474
@[
    trace_retpoline+1
    __trace_retpoline+30
    __x86_indirect_thunk_rax+33
    ktime_get+58
    clockevents_program_event+84
    hrtimer_start_range_ns+528
    start_sw_timer+356
    restart_apic_timer+85
    kvm_set_msr_common+1497
    msr_interception+142
    vcpu_enter_guest+684
    kvm_arch_vcpu_ioctl_run+261
    kvm_vcpu_ioctl+559
    do_vfs_ioctl+164
    ksys_ioctl+96
    __x64_sys_ioctl+22
    do_syscall_64+89
    entry_SYSCALL_64_after_hwframe+68
]: 41887
@[
    trace_retpoline+1
    __trace_retpoline+30
    __x86_indirect_thunk_rax+33
    lapic_next_event+28
    clockevents_program_event+148
    hrtimer_try_to_cancel+168
    hrtimer_cancel+21
    kvm_set_lapic_tscdeadline_msr+43
    kvm_set_msr_common+1497
    msr_interception+142
    vcpu_enter_guest+684
    kvm_arch_vcpu_ioctl_run+261
    kvm_vcpu_ioctl+559
    do_vfs_ioctl+164
    ksys_ioctl+96
    __x64_sys_ioctl+22
    do_syscall_64+89
    entry_SYSCALL_64_after_hwframe+68
]: 42723
@[
    trace_retpoline+1
    __trace_retpoline+30
    __x86_indirect_thunk_rax+33
    clockevents_program_event+148
    hrtimer_try_to_cancel+168
    hrtimer_cancel+21
    kvm_set_lapic_tscdeadline_msr+43
    kvm_set_msr_common+1497
    msr_interception+142
    vcpu_enter_guest+684
    kvm_arch_vcpu_ioctl_run+261
    kvm_vcpu_ioctl+559
    do_vfs_ioctl+164
    ksys_ioctl+96
    __x64_sys_ioctl+22
    do_syscall_64+89
    entry_SYSCALL_64_after_hwframe+68
]: 42766
@[
    trace_retpoline+1
    __trace_retpoline+30
    __x86_indirect_thunk_rax+33
    ktime_get+58
    clockevents_program_event+84
    hrtimer_try_to_cancel+168
    hrtimer_cancel+21
    kvm_set_lapic_tscdeadline_msr+43
    kvm_set_msr_common+1497
    msr_interception+142
    vcpu_enter_guest+684
    kvm_arch_vcpu_ioctl_run+261
    kvm_vcpu_ioctl+559
    do_vfs_ioctl+164
    ksys_ioctl+96
    __x64_sys_ioctl+22
    do_syscall_64+89
    entry_SYSCALL_64_after_hwframe+68
]: 42848
@[
    trace_retpoline+1
    __trace_retpoline+30
    __x86_indirect_thunk_rax+33
    ktime_get+58
    start_sw_timer+279
    restart_apic_timer+85
    kvm_set_msr_common+1497
    msr_interception+142
    vcpu_enter_guest+684
    kvm_arch_vcpu_ioctl_run+261
    kvm_vcpu_ioctl+559
    do_vfs_ioctl+164
    ksys_ioctl+96
    __x64_sys_ioctl+22
    do_syscall_64+89
    entry_SYSCALL_64_after_hwframe+68
]: 499845

@total: 1780243

SVM has no TSC based programmable preemption timer so it is invoking
ktime_get() frequently.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: retpolines: x86: eliminate retpoline from vmx.c exit handlers
Andrea Arcangeli [Mon, 4 Nov 2019 22:59:59 +0000 (17:59 -0500)]
KVM: retpolines: x86: eliminate retpoline from vmx.c exit handlers

It's enough to check the exit value and issue a direct call to avoid
the retpoline for all the common vmexit reasons.

Of course CONFIG_RETPOLINE already forbids gcc to use indirect jumps
while compiling all switch() statements, however switch() would still
allow the compiler to bisect the case value. It's more efficient to
prioritize the most frequent vmexits instead.

The halt may be slow paths from the point of the guest, but not
necessarily so from the point of the host if the host runs at full CPU
capacity and no host CPU is ever left idle.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: optimize more exit handlers in vmx.c
Andrea Arcangeli [Mon, 4 Nov 2019 22:59:58 +0000 (17:59 -0500)]
KVM: x86: optimize more exit handlers in vmx.c

Eliminate wasteful call/ret non RETPOLINE case and unnecessary fentry
dynamic tracing hooking points.

Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: get rid of odd out jump label in pdptrs_changed
Miaohe Lin [Fri, 25 Oct 2019 10:54:34 +0000 (18:54 +0800)]
KVM: x86: get rid of odd out jump label in pdptrs_changed

The odd out jump label is really not needed. Get rid of
it by return true directly while r < 0 as suggested by
Paolo. This further lead to var changed being unused.
Remove it too.

Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: switch KVMCLOCK base to monotonic raw clock
Marcelo Tosatti [Mon, 28 Oct 2019 14:36:22 +0000 (12:36 -0200)]
KVM: x86: switch KVMCLOCK base to monotonic raw clock

Commit 0bc48bea36d1 ("KVM: x86: update master clock before computing
kvmclock_offset")
switches the order of operations to avoid the conversion

TSC (without frequency correction) ->
system_timestamp (with frequency correction),

which might cause a time jump.

However, it leaves any other masterclock update unsafe, which includes,
at the moment:

        * HV_X64_MSR_REFERENCE_TSC MSR write.
        * TSC writes.
        * Host suspend/resume.

Avoid the time jump issue by using frequency uncorrected
CLOCK_MONOTONIC_RAW clock.

Its the guests time keeping software responsability
to track and correct a reference clock such as UTC.

This fixes forward time jump (which can result in
failure to bring up a vCPU) during vCPU hotplug:

Oct 11 14:48:33 storage kernel: CPU2 has been hot-added
Oct 11 14:48:34 storage kernel: CPU3 has been hot-added
Oct 11 14:49:22 storage kernel: smpboot: Booting Node 0 Processor 2 APIC 0x2          <-- time jump of almost 1 minute
Oct 11 14:49:22 storage kernel: smpboot: do_boot_cpu failed(-1) to wakeup CPU#2
Oct 11 14:49:23 storage kernel: smpboot: Booting Node 0 Processor 3 APIC 0x3
Oct 11 14:49:23 storage kernel: kvm-clock: cpu 3, msr 0:7ff640c1, secondary cpu clock

Which happens because:

                /*
                 * Wait 10s total for a response from AP
                 */
                boot_error = -1;
                timeout = jiffies + 10*HZ;
                while (time_before(jiffies, timeout)) {
                         ...
                }

Analyzed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoMerge tag 'kvm-ppc-next-5.5-1' of git://git.kernel.org/pub/scm/linux/kernel/git/paulu...
Paolo Bonzini [Thu, 31 Oct 2019 23:35:55 +0000 (00:35 +0100)]
Merge tag 'kvm-ppc-next-5.5-1' of git://git./linux/kernel/git/paulus/powerpc into HEAD

KVM PPC update for 5.5

* Add capability to tell userspace whether we can single-step the guest.

* Improve the allocation of XIVE virtual processor IDs, to reduce the
  risk of running out of IDs when running many VMs on POWER9.

* Rewrite interrupt synthesis code to deliver interrupts in virtual
  mode when appropriate.

* Minor cleanups and improvements.

5 years agox86/kvm: Fix -Wmissing-prototypes warnings
Yi Wang [Thu, 10 Oct 2019 06:37:25 +0000 (14:37 +0800)]
x86/kvm: Fix -Wmissing-prototypes warnings

We get two warning when build kernel with W=1:
arch/x86/kernel/kvm.c:872:6: warning: no previous prototype for ‘arch_haltpoll_enable’ [-Wmissing-prototypes]
arch/x86/kernel/kvm.c:885:6: warning: no previous prototype for ‘arch_haltpoll_disable’ [-Wmissing-prototypes]

Including the missing head file can fix this.

Signed-off-by: Yi Wang <wang.yi59@zte.com.cn>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: Add separate helper for putting borrowed reference to kvm
Sean Christopherson [Mon, 21 Oct 2019 22:58:42 +0000 (15:58 -0700)]
KVM: Add separate helper for putting borrowed reference to kvm

Add a new helper, kvm_put_kvm_no_destroy(), to handle putting a borrowed
reference[*] to the VM when installing a new file descriptor fails.  KVM
expects the refcount to remain valid in this case, as the in-progress
ioctl() has an explicit reference to the VM.  The primary motiviation
for the helper is to document that the 'kvm' pointer is still valid
after putting the borrowed reference, e.g. to document that doing
mutex(&kvm->lock) immediately after putting a ref to kvm isn't broken.

[*] When exposing a new object to userspace via a file descriptor, e.g.
    a new vcpu, KVM grabs a reference to itself (the VM) prior to making
    the object visible to userspace to avoid prematurely freeing the VM
    in the scenario where userspace immediately closes file descriptor.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agokvm: tests: Add test to verify MSR_IA32_XSS
Aaron Lewis [Mon, 21 Oct 2019 23:30:28 +0000 (16:30 -0700)]
kvm: tests: Add test to verify MSR_IA32_XSS

Ensure that IA32_XSS appears in KVM_GET_MSR_INDEX_LIST if it can be set
to a non-zero value.

Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Aaron Lewis <aaronlewis@google.com>
Change-Id: Ia2d644f69e2d6d8c27d7e0a7a45c2bf9c42bf5ff
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agokvm: svm: Update svm_xsaves_supported
Aaron Lewis [Mon, 21 Oct 2019 23:30:27 +0000 (16:30 -0700)]
kvm: svm: Update svm_xsaves_supported

AMD CPUs now support XSAVES in a limited fashion (they require IA32_XSS
to be zero).

AMD has no equivalent of Intel's "Enable XSAVES/XRSTORS" VM-execution
control. Instead, XSAVES is always available to the guest when supported
on the host.

Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Aaron Lewis <aaronlewis@google.com>
Change-Id: I40dc2c682eb0d38c2208d95d5eb7bbb6c47f6317
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agokvm: x86: Move IA32_XSS to kvm_{get,set}_msr_common
Aaron Lewis [Mon, 21 Oct 2019 23:30:26 +0000 (16:30 -0700)]
kvm: x86: Move IA32_XSS to kvm_{get,set}_msr_common

Hoist support for RDMSR/WRMSR of IA32_XSS from vmx into common code so
that it can be used for svm as well.

Right now, kvm only allows the guest IA32_XSS to be zero,
so the guest's usage of XSAVES will be exactly the same as XSAVEC.

Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Aaron Lewis <aaronlewis@google.com>
Change-Id: Ie4b0f777d71e428fbee6e82071ac2d7618e9bb40
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Move IA32_XSS-swapping on VM-entry/VM-exit to common x86 code
Aaron Lewis [Mon, 21 Oct 2019 23:30:25 +0000 (16:30 -0700)]
KVM: x86: Move IA32_XSS-swapping on VM-entry/VM-exit to common x86 code

Hoist the vendor-specific code related to loading the hardware IA32_XSS
MSR with guest/host values on VM-entry/VM-exit to common x86 code.

Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Aaron Lewis <aaronlewis@google.com>
Change-Id: Ic6e3430833955b98eb9b79ae6715cf2a3fdd6d82
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Use wrmsr for switching between guest and host IA32_XSS on Intel
Aaron Lewis [Mon, 21 Oct 2019 23:30:24 +0000 (16:30 -0700)]
KVM: VMX: Use wrmsr for switching between guest and host IA32_XSS on Intel

When the guest can execute the XSAVES/XRSTORS instructions, use wrmsr to
set the hardware IA32_XSS MSR to guest/host values on VM-entry/VM-exit,
rather than the MSR-load areas. By using the same approach as AMD, we
will be able to use a common implementation for both (in the next
patch).

Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Aaron Lewis <aaronlewis@google.com>
Change-Id: I9447d104b2615c04e39e4af0c911e1e7309bf464
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: SVM: Use wrmsr for switching between guest and host IA32_XSS on AMD
Aaron Lewis [Mon, 21 Oct 2019 23:30:23 +0000 (16:30 -0700)]
KVM: SVM: Use wrmsr for switching between guest and host IA32_XSS on AMD

When the guest can execute the XSAVES/XRSTORS instructions, set the
hardware IA32_XSS MSR to guest/host values on VM-entry/VM-exit.

Note that vcpu->arch.ia32_xss is currently guaranteed to be 0 on AMD,
since there is no way to change it.

Suggested-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Aaron Lewis <aaronlewis@google.com>
Change-Id: Id51a782462086e6d7a3ab621838e200f1c005afd
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Remove unneeded kvm_vcpu variable, guest_xcr0_loaded
Aaron Lewis [Mon, 21 Oct 2019 23:30:22 +0000 (16:30 -0700)]
KVM: x86: Remove unneeded kvm_vcpu variable, guest_xcr0_loaded

The kvm_vcpu variable, guest_xcr0_loaded, is a waste of an 'int'
and a conditional branch.  VMX and SVM are the only users, and both
unconditionally pair kvm_load_guest_xcr0() with kvm_put_guest_xcr0()
making this check unnecessary. Without this variable, the predicates in
kvm_load_guest_xcr0 and kvm_put_guest_xcr0 should match.

Suggested-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Aaron Lewis <aaronlewis@google.com>
Change-Id: I7b1eb9b62969d7bbb2850f27e42f863421641b23
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Fix conditions for guest IA32_XSS support
Aaron Lewis [Mon, 21 Oct 2019 23:30:21 +0000 (16:30 -0700)]
KVM: VMX: Fix conditions for guest IA32_XSS support

Volume 4 of the SDM says that IA32_XSS is supported
if CPUID(EAX=0DH,ECX=1):EAX.XSS[bit 3] is set, so only the
X86_FEATURE_XSAVES check is necessary (X86_FEATURE_XSAVES is the Linux
name for CPUID(EAX=0DH,ECX=1):EAX.XSS[bit 3]).

Fixes: 4d763b168e9c5 ("KVM: VMX: check CPUID before allowing read/write of IA32_XSS")
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Aaron Lewis <aaronlewis@google.com>
Change-Id: I9059b9f2e3595e4b09a4cdcf14b933b22ebad419
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Introduce vcpu->arch.xsaves_enabled
Aaron Lewis [Mon, 21 Oct 2019 23:30:20 +0000 (16:30 -0700)]
KVM: x86: Introduce vcpu->arch.xsaves_enabled

Cache whether XSAVES is enabled in the guest by adding xsaves_enabled to
vcpu->arch.

Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Aaron Lewis <aaronlewis@google.com>
Change-Id: If4638e0901c28a4494dad2e103e2c075e8ab5d68
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Rename {vmx,nested_vmx}_vcpu_setup()
Xiaoyao Li [Sun, 20 Oct 2019 09:11:01 +0000 (17:11 +0800)]
KVM: VMX: Rename {vmx,nested_vmx}_vcpu_setup()

Rename {vmx,nested_vmx}_vcpu_setup() to match what they really do.

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Initialize vmx->guest_msrs[] right after allocation
Xiaoyao Li [Sun, 20 Oct 2019 09:11:00 +0000 (17:11 +0800)]
KVM: VMX: Initialize vmx->guest_msrs[] right after allocation

Move the initialization of vmx->guest_msrs[] from vmx_vcpu_setup() to
vmx_create_vcpu(), and put it right after its allocation.

This also is the preperation for next patch.

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Remove vmx->hv_deadline_tsc initialization from vmx_vcpu_setup()
Xiaoyao Li [Sun, 20 Oct 2019 09:10:59 +0000 (17:10 +0800)]
KVM: VMX: Remove vmx->hv_deadline_tsc initialization from vmx_vcpu_setup()

... It can be removed here because the same code is called later in
vmx_vcpu_reset() as the flow:

kvm_arch_vcpu_setup()
-> kvm_vcpu_reset()
-> vmx_vcpu_reset()

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Write VPID to vmcs when creating vcpu
Xiaoyao Li [Sun, 20 Oct 2019 09:10:58 +0000 (17:10 +0800)]
KVM: VMX: Write VPID to vmcs when creating vcpu

Move the code that writes vmx->vpid to vmcs from vmx_vcpu_reset() to
vmx_vcpu_setup(), because vmx->vpid is allocated when creating vcpu and
never changed. So we don't need to update the vmcs.vpid when resetting
vcpu.

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86/vPMU: Declare kvm_pmu->reprogram_pmi field using DECLARE_BITMAP
Like Xu [Mon, 21 Oct 2019 10:55:04 +0000 (18:55 +0800)]
KVM: x86/vPMU: Declare kvm_pmu->reprogram_pmi field using DECLARE_BITMAP

Replace the explicit declaration of "u64 reprogram_pmi" with the generic
macro DECLARE_BITMAP for all possible appropriate number of bits.

Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Like Xu <like.xu@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: remove redundant code in kvm_arch_vm_ioctl
Miaohe Lin [Mon, 21 Oct 2019 02:52:56 +0000 (10:52 +0800)]
KVM: remove redundant code in kvm_arch_vm_ioctl

If we reach here with r = 0, we will reassign r = 0
unnecesarry, then do the label set_irqchip_out work.
If we reach here with r != 0, then we will do the label
work directly. So this if statement and r = 0 assignment
is redundant.

Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agokvm: x86: Modify kvm_x86_ops.get_enable_apicv() to use struct kvm parameter
Suthikulpanit, Suravee [Fri, 13 Sep 2019 19:00:49 +0000 (19:00 +0000)]
kvm: x86: Modify kvm_x86_ops.get_enable_apicv() to use struct kvm parameter

Generally, APICv for all vcpus in the VM are enable/disable in the same
manner. So, get_enable_apicv() should represent APICv status of the VM
instead of each VCPU.

Modify kvm_x86_ops.get_enable_apicv() to take struct kvm as parameter
instead of struct kvm_vcpu.

Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Fold decache_cr3() into cache_reg()
Sean Christopherson [Fri, 27 Sep 2019 21:45:23 +0000 (14:45 -0700)]
KVM: x86: Fold decache_cr3() into cache_reg()

Handle caching CR3 (from VMX's VMCS) into struct kvm_vcpu via the common
cache_reg() callback and drop the dedicated decache_cr3().  The name
decache_cr3() is somewhat confusing as the caching behavior of CR3
follows that of GPRs, RFLAGS and PDPTRs, (handled via cache_reg()), and
has nothing in common with the caching behavior of CR0/CR4 (whose
decache_cr{0,4}_guest_bits() likely provided the 'decache' verbiage).

This would effectivel adds a BUG() if KVM attempts to cache CR3 on SVM.
Change it to a WARN_ON_ONCE() -- if the cache never requires filling,
the value is already in the right place -- and opportunistically add one
in VMX to provide an equivalent check.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Add helpers to test/mark reg availability and dirtiness
Sean Christopherson [Fri, 27 Sep 2019 21:45:22 +0000 (14:45 -0700)]
KVM: x86: Add helpers to test/mark reg availability and dirtiness

Add helpers to prettify code that tests and/or marks whether or not a
register is available and/or dirty.

Suggested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Fold 'enum kvm_ex_reg' definitions into 'enum kvm_reg'
Sean Christopherson [Fri, 27 Sep 2019 21:45:21 +0000 (14:45 -0700)]
KVM: x86: Fold 'enum kvm_ex_reg' definitions into 'enum kvm_reg'

Now that indexing into arch.regs is either protected by WARN_ON_ONCE or
done with hardcoded enums, combine all definitions for registers that
are tracked by regs_avail and regs_dirty into 'enum kvm_reg'.  Having a
single enum type will simplify additional cleanup related to regs_avail
and regs_dirty.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: Add WARNs to detect out-of-bounds register indices
Sean Christopherson [Fri, 27 Sep 2019 21:45:20 +0000 (14:45 -0700)]
KVM: x86: Add WARNs to detect out-of-bounds register indices

Add WARN_ON_ONCE() checks in kvm_register_{read,write}() to detect reg
values that would cause KVM to overflow vcpu->arch.regs.  Change the reg
param to an 'int' to make it clear that the reg index is unverified.

Regarding the overhead of WARN_ON_ONCE(), now that all fixed GPR reads
and writes use dedicated accessors, e.g. kvm_rax_read(), the overhead
is limited to flows where the reg index is generated at runtime.  And
there is at least one historical bug where KVM has generated an out-of-
bounds access to arch.regs (see commit b68f3cc7d9789, "KVM: x86: Always
use 32-bit SMRAM save state for 32-bit kernels").

Adding the WARN_ON_ONCE() protection paves the way for additional
cleanup related to kvm_reg and kvm_reg_ex.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Optimize vmx_set_rflags() for unrestricted guest
Sean Christopherson [Fri, 27 Sep 2019 21:45:19 +0000 (14:45 -0700)]
KVM: VMX: Optimize vmx_set_rflags() for unrestricted guest

Rework vmx_set_rflags() to avoid the extra code need to handle emulation
of real mode and invalid state when unrestricted guest is disabled.  The
primary reason for doing so is to avoid the call to vmx_get_rflags(),
which will incur a VMREAD when RFLAGS is not already available.  When
running nested VMs, the majority of calls to vmx_set_rflags() will occur
without an associated vmx_get_rflags(), i.e. when stuffing GUEST_RFLAGS
during transitions between vmcs01 and vmcs02.

Note, vmx_get_rflags() guarantees RFLAGS is marked available.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
[Replace "else" with early "return" in the unrestricted guest branch. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Consolidate to_vmx() usage in RFLAGS accessors
Sean Christopherson [Fri, 27 Sep 2019 21:45:18 +0000 (14:45 -0700)]
KVM: VMX: Consolidate to_vmx() usage in RFLAGS accessors

Capture struct vcpu_vmx in a local variable to improve the readability
of vmx_{g,s}et_rflags().

No functional change intended.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Skip GUEST_CR3 VMREAD+VMWRITE if the VMCS is up-to-date
Sean Christopherson [Fri, 27 Sep 2019 21:45:17 +0000 (14:45 -0700)]
KVM: VMX: Skip GUEST_CR3 VMREAD+VMWRITE if the VMCS is up-to-date

Skip the VMWRITE to update GUEST_CR3 if CR3 is not available, i.e. has
not been read from the VMCS since the last VM-Enter.  If vcpu->arch.cr3
is stale, kvm_read_cr3(vcpu) will refresh vcpu->arch.cr3 from the VMCS,
meaning KVM will do a VMREAD and then VMWRITE the value it just pulled
from the VMCS.

Note, this is a purely theoretical change, no instances of skipping
the VMREAD+VMWRITE have been observed with this change.

Tested-by: Reto Buerki <reet@codelabs.ch>
Tested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: SVM: Reduce WBINVD/DF_FLUSH invocations
Tom Lendacky [Thu, 3 Oct 2019 21:17:48 +0000 (21:17 +0000)]
KVM: SVM: Reduce WBINVD/DF_FLUSH invocations

Performing a WBINVD and DF_FLUSH are expensive operations. Currently, a
WBINVD/DF_FLUSH is performed every time an SEV guest terminates. However,
the WBINVD/DF_FLUSH is only required when an ASID is being re-allocated
to a new SEV guest. Also, a single WBINVD/DF_FLUSH can enable all ASIDs
that have been disassociated from guests through DEACTIVATE.

To reduce the number of WBINVD/DF_FLUSH invocations, introduce a new ASID
bitmap to track ASIDs that need to be reclaimed. When an SEV guest is
terminated, add its ASID to the reclaim bitmap instead of clearing the
bitmap in the existing SEV ASID bitmap. This delays the need to perform a
WBINVD/DF_FLUSH invocation when an SEV guest terminates until all of the
available SEV ASIDs have been used. At that point, the WBINVD/DF_FLUSH
invocation can be performed and all ASIDs in the reclaim bitmap moved to
the available ASIDs bitmap.

The semaphore around DEACTIVATE can be changed to a read semaphore with
the semaphore taken in write mode before performing the WBINVD/DF_FLUSH.

Tested-by: David Rientjes <rientjes@google.com>
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: SVM: Remove unneeded WBINVD and DF_FLUSH when starting SEV guests
Tom Lendacky [Thu, 3 Oct 2019 21:17:46 +0000 (21:17 +0000)]
KVM: SVM: Remove unneeded WBINVD and DF_FLUSH when starting SEV guests

Performing a WBINVD and DF_FLUSH are expensive operations. The SEV support
currently performs this WBINVD/DF_FLUSH combination when an SEV guest is
terminated, so there is no need for it to be done before LAUNCH.

However, when the SEV firmware transitions the platform from UNINIT state
to INIT state, all ASIDs will be marked invalid across all threads.
Therefore, as part of transitioning the platform to INIT state, perform a
WBINVD/DF_FLUSH after a successful INIT in the PSP/SEV device driver.
Since the PSP/SEV device driver is x86 only, it can reference and use the
WBINVD related functions directly.

Cc: Gary Hook <gary.hook@amd.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Tested-by: David Rientjes <rientjes@google.com>
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: nVMX: Always write vmcs02.GUEST_CR3 during nested VM-Enter
Sean Christopherson [Fri, 27 Sep 2019 21:45:16 +0000 (14:45 -0700)]
KVM: nVMX: Always write vmcs02.GUEST_CR3 during nested VM-Enter

Write the desired L2 CR3 into vmcs02.GUEST_CR3 during nested VM-Enter
instead of deferring the VMWRITE until vmx_set_cr3().  If the VMWRITE
is deferred, then KVM can consume a stale vmcs02.GUEST_CR3 when it
refreshes vmcs12->guest_cr3 during nested_vmx_vmexit() if the emulated
VM-Exit occurs without actually entering L2, e.g. if the nested run
is squashed because nested VM-Enter (from L1) is putting L2 into HLT.

Note, the above scenario can occur regardless of whether L1 is
intercepting HLT, e.g. L1 can intercept HLT and then re-enter L2 with
vmcs.GUEST_ACTIVITY_STATE=HALTED.  But practically speaking, a VMM will
likely put a guest into HALTED if and only if it's not intercepting HLT.

In an ideal world where EPT *requires* unrestricted guest (and vice
versa), VMX could handle CR3 similar to how it handles RSP and RIP,
e.g. mark CR3 dirty and conditionally load it at vmx_vcpu_run().  But
the unrestricted guest silliness complicates the dirty tracking logic
to the point that explicitly handling vmcs02.GUEST_CR3 during nested
VM-Enter is a simpler overall implementation.

Cc: stable@vger.kernel.org
Reported-and-tested-by: Reto Buerki <reet@codelabs.ch>
Tested-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Reviewed-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: SVM: Guard against DEACTIVATE when performing WBINVD/DF_FLUSH
Tom Lendacky [Thu, 3 Oct 2019 21:17:45 +0000 (21:17 +0000)]
KVM: SVM: Guard against DEACTIVATE when performing WBINVD/DF_FLUSH

The SEV firmware DEACTIVATE command disassociates an SEV guest from an
ASID, clears the WBINVD indicator on all threads and indicates that the
SEV firmware DF_FLUSH command must be issued before the ASID can be
re-used. The SEV firmware DF_FLUSH command will return an error if a
WBINVD has not been performed on every thread before it has been invoked.
A window exists between the WBINVD and the invocation of the DF_FLUSH
command where an SEV firmware DEACTIVATE command could be invoked on
another thread, clearing the WBINVD indicator. This will cause the
subsequent SEV firmware DF_FLUSH command to fail which, in turn, results
in the SEV firmware ACTIVATE command failing for the reclaimed ASID.
This results in the SEV guest failing to start.

Use a mutex to close the WBINVD/DF_FLUSH window by obtaining the mutex
before the DEACTIVATE and releasing it after the DF_FLUSH. This ensures
that any DEACTIVATE cannot run before a DF_FLUSH has completed.

Fixes: 59414c989220 ("KVM: SVM: Add support for KVM_SEV_LAUNCH_START command")
Tested-by: David Rientjes <rientjes@google.com>
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: SVM: Serialize access to the SEV ASID bitmap
Tom Lendacky [Thu, 3 Oct 2019 21:17:43 +0000 (21:17 +0000)]
KVM: SVM: Serialize access to the SEV ASID bitmap

The SEV ASID bitmap currently is not protected against parallel SEV guest
startups. This can result in an SEV guest failing to start because another
SEV guest could have been assigned the same ASID value. Use a mutex to
serialize access to the SEV ASID bitmap.

Fixes: 1654efcbc431 ("KVM: SVM: Add KVM_SEV_INIT command")
Tested-by: David Rientjes <rientjes@google.com>
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoMerge tag 'kvm-ppc-fixes-5.4-1' of git://git.kernel.org/pub/scm/linux/kernel/git...
Paolo Bonzini [Tue, 22 Oct 2019 11:31:35 +0000 (13:31 +0200)]
Merge tag 'kvm-ppc-fixes-5.4-1' of git://git./linux/kernel/git/paulus/powerpc into HEAD

PPC KVM fix for 5.4

- Fix a bug in the XIVE code which can cause a host crash.

5 years agoMerge tag 'kvmarm-fixes-5.4-2' of git://git.kernel.org/pub/scm/linux/kernel/git/kvmar...
Paolo Bonzini [Tue, 22 Oct 2019 11:31:29 +0000 (13:31 +0200)]
Merge tag 'kvmarm-fixes-5.4-2' of git://git./linux/kernel/git/kvmarm/kvmarm into HEAD

KVM/arm fixes for 5.4, take #2

Special PMU edition:

- Fix cycle counter truncation
- Fix cycle counter overflow limit on pure 64bit system
- Allow chained events to be actually functional
- Correct sample period after overflow

5 years agokvm: clear kvmclock MSR on reset
Paolo Bonzini [Thu, 10 Oct 2019 10:49:22 +0000 (12:49 +0200)]
kvm: clear kvmclock MSR on reset

After resetting the vCPU, the kvmclock MSR keeps the previous value but it is
not enabled.  This can be confusing, so fix it.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: x86: fix bugon.cocci warnings
kbuild test robot [Sun, 29 Sep 2019 16:43:28 +0000 (18:43 +0200)]
KVM: x86: fix bugon.cocci warnings

Use BUG_ON instead of a if condition followed by BUG.

Generated by: scripts/coccinelle/misc/bugon.cocci

Fixes: 4b526de50e39 ("KVM: x86: Check kvm_rebooting in kvm_spurious_fault()")
CC: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Julia Lawall <julia.lawall@lip6.fr>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: VMX: Remove specialized handling of unexpected exit-reasons
Liran Alon [Sun, 29 Sep 2019 14:50:18 +0000 (17:50 +0300)]
KVM: VMX: Remove specialized handling of unexpected exit-reasons

Commit bf653b78f960 ("KVM: vmx: Introduce handle_unexpected_vmexit
and handle WAITPKG vmexit") introduced specialized handling of
specific exit-reasons that should not be raised by CPU because
KVM configures VMCS such that they should never be raised.

However, since commit 7396d337cfad ("KVM: x86: Return to userspace
with internal error on unexpected exit reason"), VMX & SVM
exit handlers were modified to generically handle all unexpected
exit-reasons by returning to userspace with internal error.

Therefore, there is no need for specialized handling of specific
unexpected exit-reasons (This specialized handling also introduced
inconsistency for these exit-reasons to silently skip guest instruction
instead of return to userspace on internal-error).

Fixes: bf653b78f960 ("KVM: vmx: Introduce handle_unexpected_vmexit and handle WAITPKG vmexit")
Signed-off-by: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoselftests: kvm: fix sync_regs_test with newer gccs
Vitaly Kuznetsov [Tue, 8 Oct 2019 18:08:08 +0000 (20:08 +0200)]
selftests: kvm: fix sync_regs_test with newer gccs

Commit 204c91eff798a ("KVM: selftests: do not blindly clobber registers in
 guest asm") was intended to make test more gcc-proof, however, the result
is exactly the opposite: on newer gccs (e.g. 8.2.1) the test breaks with

==== Test Assertion Failure ====
  x86_64/sync_regs_test.c:168: run->s.regs.regs.rbx == 0xBAD1DEA + 1
  pid=14170 tid=14170 - Invalid argument
     1 0x00000000004015b3: main at sync_regs_test.c:166 (discriminator 6)
     2 0x00007f413fb66412: ?? ??:0
     3 0x000000000040191d: _start at ??:?
  rbx sync regs value incorrect 0x1.

Apparently, compile is still free to play games with registers even
when they have variables attached.

Re-write guest code with 'asm volatile' by embedding ucall there and
making sure rbx is preserved.

Fixes: 204c91eff798a ("KVM: selftests: do not blindly clobber registers in guest asm")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoselftests: kvm: vmx_dirty_log_test: skip the test when VMX is not supported
Vitaly Kuznetsov [Tue, 8 Oct 2019 19:43:38 +0000 (21:43 +0200)]
selftests: kvm: vmx_dirty_log_test: skip the test when VMX is not supported

vmx_dirty_log_test fails on AMD and this is no surprise as it is VMX
specific. Bail early when nested VMX is unsupported.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoselftests: kvm: consolidate VMX support checks
Vitaly Kuznetsov [Tue, 8 Oct 2019 19:43:37 +0000 (21:43 +0200)]
selftests: kvm: consolidate VMX support checks

vmx_* tests require VMX and three of them implement the same check. Move it
to vmx library.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoselftests: kvm: vmx_set_nested_state_test: don't check for VMX support twice
Vitaly Kuznetsov [Tue, 8 Oct 2019 19:43:36 +0000 (21:43 +0200)]
selftests: kvm: vmx_set_nested_state_test: don't check for VMX support twice

vmx_set_nested_state_test() checks if VMX is supported twice: in the very
beginning (and skips the whole test if it's not) and before doing
test_vmx_nested_state(). One should be enough.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoKVM: Don't shrink/grow vCPU halt_poll_ns if host side polling is disabled
Wanpeng Li [Sun, 29 Sep 2019 01:06:56 +0000 (09:06 +0800)]
KVM: Don't shrink/grow vCPU halt_poll_ns if host side polling is disabled

Don't waste cycles to shrink/grow vCPU halt_poll_ns if host
side polling is disabled.

Acked-by: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoselftests: kvm: synchronize .gitignore to Makefile
Vitaly Kuznetsov [Mon, 7 Oct 2019 13:26:56 +0000 (15:26 +0200)]
selftests: kvm: synchronize .gitignore to Makefile

Because "Untracked files:" are annoying.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agokvm: x86: Expose RDPID in KVM_GET_SUPPORTED_CPUID
Jim Mattson [Fri, 4 Oct 2019 20:22:47 +0000 (13:22 -0700)]
kvm: x86: Expose RDPID in KVM_GET_SUPPORTED_CPUID

When the RDPID instruction is supported on the host, enumerate it in
KVM_GET_SUPPORTED_CPUID.

Signed-off-by: Jim Mattson <jmattson@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
5 years agoMerge tag 'pinctrl-v5.4-2' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw...
Linus Torvalds [Tue, 22 Oct 2019 10:40:07 +0000 (06:40 -0400)]
Merge tag 'pinctrl-v5.4-2' of git://git./linux/kernel/git/linusw/linux-pinctrl

Pull pin control fixes from Linus Walleij:
 "Here is a bunch of pin control fixes. I was lagging behind on this
  one, some fixes should have come in earlier, sorry about that.

  Anyways here it is, pretty straight-forward fixes, the Strago fix
  stand out as something serious affecting a lot of machines.

  Summary:
   - Handle multiple instances of Intel chips without complaining.
   - Restore the Intel Strago DMI workaround
   - Make the Armada 37xx handle pins over 32
   - Fix the polarity of the LED group on Armada 37xx
   - Fix an off-by-one bug in the NS2 driver
   - Fix error path for iproc's platform_get_irq()
   - Fix error path on the STMFX driver
   - Fix a typo in the Berlin AS370 driver
   - Fix up misc errors in the Aspeed 2600 BMC support
   - Fix a stray SPDX tag"

* tag 'pinctrl-v5.4-2' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl:
  pinctrl: aspeed-g6: Rename SD3 to EMMC and rework pin groups
  pinctrl: aspeed-g6: Fix UART13 group pinmux
  pinctrl: aspeed-g6: Make SIG_DESC_CLEAR() behave intuitively
  pinctrl: aspeed-g6: Fix I3C3/I3C4 pinmux configuration
  pinctrl: aspeed-g6: Fix I2C14 SDA description
  pinctrl: aspeed-g6: Sort pins for sanity
  dt-bindings: pinctrl: aspeed-g6: Rework SD3 function and groups
  pinctrl: berlin: as370: fix a typo s/spififib/spdifib
  pinctrl: armada-37xx: swap polarity on LED group
  pinctrl: stmfx: fix null pointer on remove
  pinctrl: iproc: allow for error from platform_get_irq()
  pinctrl: ns2: Fix off by one bugs in ns2_pinmux_enable()
  pinctrl: bcm-iproc: Use SPDX header
  pinctrl: armada-37xx: fix control of pins 32 and up
  pinctrl: cherryview: restore Strago DMI workaround for all versions
  pinctrl: intel: Allocate IRQ chip dynamic

5 years agoKVM: PPC: Book3S HV: Reject mflags=2 (LPCR[AIL]=2) ADDR_TRANS_MODE mode
Nicholas Piggin [Wed, 2 Oct 2019 06:00:25 +0000 (16:00 +1000)]
KVM: PPC: Book3S HV: Reject mflags=2 (LPCR[AIL]=2) ADDR_TRANS_MODE mode

AIL=2 mode has no known users, so is not well tested or supported.
Disallow guests from selecting this mode because it may become
deprecated in future versions of the architecture.

This policy decision is not left to QEMU because KVM support is
required for AIL=2 (when injecting interrupts).

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years agoKVM: PPC: Book3S HV: Implement LPCR[AIL]=3 mode for injected interrupts
Nicholas Piggin [Wed, 2 Oct 2019 06:00:24 +0000 (16:00 +1000)]
KVM: PPC: Book3S HV: Implement LPCR[AIL]=3 mode for injected interrupts

kvmppc_inject_interrupt does not implement LPCR[AIL]!=0 modes, which
can result in the guest receiving interrupts as if LPCR[AIL]=0
contrary to the ISA.

In practice, Linux guests cope with this deviation, but it should be
fixed.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years agoKVM: PPC: Book3S HV: Reuse kvmppc_inject_interrupt for async guest delivery
Nicholas Piggin [Wed, 2 Oct 2019 06:00:23 +0000 (16:00 +1000)]
KVM: PPC: Book3S HV: Reuse kvmppc_inject_interrupt for async guest delivery

This consolidates the HV interrupt delivery logic into one place.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years agoKVM: PPC: Book3S: Replace reset_msr mmu op with inject_interrupt arch op
Nicholas Piggin [Wed, 2 Oct 2019 06:00:22 +0000 (16:00 +1000)]
KVM: PPC: Book3S: Replace reset_msr mmu op with inject_interrupt arch op

reset_msr sets the MSR for interrupt injection, but it's cleaner and
more flexible to provide a single op to set both MSR and PC for the
interrupt.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years agoKVM: PPC: Book3S: Define and use SRR1_MSR_BITS
Nicholas Piggin [Wed, 2 Oct 2019 06:00:21 +0000 (16:00 +1000)]
KVM: PPC: Book3S: Define and use SRR1_MSR_BITS

Acked-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years agoKVM: PPC: Book3S HV: XIVE: Allow userspace to set the # of VPs
Greg Kurz [Fri, 27 Sep 2019 11:54:07 +0000 (13:54 +0200)]
KVM: PPC: Book3S HV: XIVE: Allow userspace to set the # of VPs

Add a new attribute to both XIVE and XICS-on-XIVE KVM devices so that
userspace can tell how many interrupt servers it needs. If a VM needs
less than the current default of KVM_MAX_VCPUS (2048), we can allocate
less VPs in OPAL. Combined with a core stride (VSMT) that matches the
number of guest threads per core, this may substantially increases the
number of VMs that can run concurrently with an in-kernel XIVE device.

Since the legacy XIVE KVM device is exposed to userspace through the
XICS KVM API, a new attribute group is added to it for this purpose.
While here, fix the syntax of the existing KVM_DEV_XICS_GRP_SOURCES
in the XICS documentation.

Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years agoKVM: PPC: Book3S HV: XIVE: Make VP block size configurable
Greg Kurz [Fri, 27 Sep 2019 11:54:01 +0000 (13:54 +0200)]
KVM: PPC: Book3S HV: XIVE: Make VP block size configurable

The XIVE VP is an internal structure which allow the XIVE interrupt
controller to maintain the interrupt context state of vCPUs non
dispatched on HW threads.

When a guest is started, the XIVE KVM device allocates a block of
XIVE VPs in OPAL, enough to accommodate the highest possible vCPU
id KVM_MAX_VCPU_ID (16384) packed down to KVM_MAX_VCPUS (2048).
With a guest's core stride of 8 and a threading mode of 1 (QEMU's
default), a VM must run at least 256 vCPUs to actually need such a
range of VPs.

A POWER9 system has a limited XIVE VP space : 512k and KVM is
currently wasting this HW resource with large VP allocations,
especially since a typical VM likely runs with a lot less vCPUs.

Make the size of the VP block configurable. Add an nr_servers
field to the XIVE structure and a function to set it for this
purpose.

Split VP allocation out of the device create function. Since the
VP block isn't used before the first vCPU connects to the XIVE KVM
device, allocation is now performed by kvmppc_xive_connect_vcpu().
This gives the opportunity to set nr_servers in between:

          kvmppc_xive_create() / kvmppc_xive_native_create()
                               .
                               .
                     kvmppc_xive_set_nr_servers()
                               .
                               .
    kvmppc_xive_connect_vcpu() / kvmppc_xive_native_connect_vcpu()

The connect_vcpu() functions check that the vCPU id is below nr_servers
and if it is the first vCPU they allocate the VP block. This is protected
against a concurrent update of nr_servers by kvmppc_xive_set_nr_servers()
with the xive->lock mutex.

Also, the block is allocated once for the device lifetime: nr_servers
should stay constant otherwise connect_vcpu() could generate a boggus
VP id and likely crash OPAL. It is thus forbidden to update nr_servers
once the block is allocated.

If the VP allocation fail, return ENOSPC which seems more appropriate to
report the depletion of system wide HW resource than ENOMEM or ENXIO.

A VM using a stride of 8 and 1 thread per core with 32 vCPUs would hence
only need 256 VPs instead of 2048. If the stride is set to match the number
of threads per core, this goes further down to 32.

This will be exposed to userspace by a subsequent patch.

Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years agoKVM: PPC: Book3S HV: XIVE: Compute the VP id in a common helper
Greg Kurz [Fri, 27 Sep 2019 11:53:55 +0000 (13:53 +0200)]
KVM: PPC: Book3S HV: XIVE: Compute the VP id in a common helper

Reduce code duplication by consolidating the checking of vCPU ids and VP
ids to a common helper used by both legacy and native XIVE KVM devices.
And explain the magic with a comment.

Signed-off-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years agoKVM: PPC: Book3S HV: XIVE: Show VP id in debugfs
Greg Kurz [Fri, 27 Sep 2019 11:53:49 +0000 (13:53 +0200)]
KVM: PPC: Book3S HV: XIVE: Show VP id in debugfs

Print out the VP id of each connected vCPU, this allow to see:
- the VP block base in which OPAL encodes information that may be
  useful when debugging
- the packed vCPU id which may differ from the raw vCPU id if the
  latter is >= KVM_MAX_VCPUS (2048)

Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years agoKVM: PPC: Book3S HV: XIVE: Set kvm->arch.xive when VPs are allocated
Greg Kurz [Fri, 27 Sep 2019 11:53:38 +0000 (13:53 +0200)]
KVM: PPC: Book3S HV: XIVE: Set kvm->arch.xive when VPs are allocated

If we cannot allocate the XIVE VPs in OPAL, the creation of a XIVE or
XICS-on-XIVE device is aborted as expected, but we leave kvm->arch.xive
set forever since the release method isn't called in this case. Any
subsequent tentative to create a XIVE or XICS-on-XIVE for this VM will
thus always fail (DoS). This is a problem for QEMU since it destroys
and re-creates these devices when the VM is reset: the VM would be
restricted to using the much slower emulated XIVE or XICS forever.

As an alternative to adding rollback, do not assign kvm->arch.xive before
making sure the XIVE VPs are allocated in OPAL.

Cc: stable@vger.kernel.org # v5.2
Fixes: 5422e95103cf ("KVM: PPC: Book3S HV: XIVE: Replace the 'destroy' method by a 'release' method")
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years agoKVM: PPC: E500: Replace current->mm by kvm->mm
Leonardo Bras [Mon, 23 Sep 2019 21:24:08 +0000 (18:24 -0300)]
KVM: PPC: E500: Replace current->mm by kvm->mm

Given that in kvm_create_vm() there is:
kvm->mm = current->mm;

And that on every kvm_*_ioctl we have:
if (kvm->mm != current->mm)
return -EIO;

I see no reason to keep using current->mm instead of kvm->mm.

By doing so, we would reduce the use of 'global' variables on code, relying
more in the contents of kvm struct.

Signed-off-by: Leonardo Bras <leonardo@linux.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years agoKVM: PPC: Reduce calls to get current->mm by storing the value locally
Leonardo Bras [Mon, 23 Sep 2019 21:30:23 +0000 (18:30 -0300)]
KVM: PPC: Reduce calls to get current->mm by storing the value locally

Reduces the number of calls to get_current() in order to get the value of
current->mm by doing it once and storing the value, since it is not
supposed to change inside the same process).

Signed-off-by: Leonardo Bras <leonardo@linux.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years agoKVM: PPC: Report single stepping capability
Fabiano Rosas [Wed, 19 Jun 2019 16:01:27 +0000 (13:01 -0300)]
KVM: PPC: Report single stepping capability

When calling the KVM_SET_GUEST_DEBUG ioctl, userspace might request
the next instruction to be single stepped via the
KVM_GUESTDBG_SINGLESTEP control bit of the kvm_guest_debug structure.

This patch adds the KVM_CAP_PPC_GUEST_DEBUG_SSTEP capability in order
to inform userspace about the state of single stepping support.

We currently don't have support for guest single stepping implemented
in Book3S HV so the capability is only present for Book3S PR and
BookE.

Signed-off-by: Fabiano Rosas <farosas@linux.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years agoLinux 5.4-rc4
Linus Torvalds [Sun, 20 Oct 2019 19:56:22 +0000 (15:56 -0400)]
Linux 5.4-rc4

5 years agoMerge tag 'kbuild-fixes-v5.4-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 20 Oct 2019 16:36:57 +0000 (12:36 -0400)]
Merge tag 'kbuild-fixes-v5.4-2' of git://git./linux/kernel/git/masahiroy/linux-kbuild

Pull more Kbuild fixes from Masahiro Yamada:

 - fix a bashism of setlocalversion

 - do not use the too new --sort option of tar

* tag 'kbuild-fixes-v5.4-2' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild:
  kheaders: substituting --sort in archive creation
  scripts: setlocalversion: fix a bashism
  kbuild: update comment about KBUILD_ALLDIRS

5 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 20 Oct 2019 10:31:14 +0000 (06:31 -0400)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Thomas Gleixner:
 "A small set of x86 fixes:

   - Prevent a NULL pointer dereference in the X2APIC code in case of a
     CPU hotplug failure.

   - Prevent boot failures on HP superdome machines by invalidating the
     level2 kernel pagetable entries outside of the kernel area as
     invalid so BIOS reserved space won't be touched unintentionally.

     Also ensure that memory holes are rounded up to the next PMD
     boundary correctly.

   - Enable X2APIC support on Hyper-V to prevent boot failures.

   - Set the paravirt name when running on Hyper-V for consistency

   - Move a function under the appropriate ifdef guard to prevent build
     warnings"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/boot/acpi: Move get_cmdline_acpi_rsdp() under #ifdef guard
  x86/hyperv: Set pv_info.name to "Hyper-V"
  x86/apic/x2apic: Fix a NULL pointer deref when handling a dying cpu
  x86/hyperv: Make vapic support x2apic mode
  x86/boot/64: Round memory hole size up to next PMD page
  x86/boot/64: Make level2_kernel_pgt pages invalid outside kernel area

5 years agoMerge branch 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 20 Oct 2019 10:27:54 +0000 (06:27 -0400)]
Merge branch 'irq-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull irq fixes from Thomas Gleixner:
 "A small set of irq chip driver fixes and updates:

   - Update the SIFIVE PLIC interrupt driver to use the fasteoi handler
     to address the shortcomings of the existing flow handling which was
     prone to lose interrupts

   - Use the proper limit for GIC interrupt line numbers

   - Add retrigger support for the recently merged Anapurna Labs Fabric
     interrupt controller to make it complete

   - Enable the ATMEL AIC5 interrupt controller driver on the new
     SAM9X60 SoC"

* 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  irqchip/sifive-plic: Switch to fasteoi flow
  irqchip/gic-v3: Fix GIC_LINE_NR accessor
  irqchip/atmel-aic5: Add support for sam9x60 irqchip
  irqchip/al-fic: Add support for irq retrigger

5 years agoMerge branch 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 20 Oct 2019 10:25:12 +0000 (06:25 -0400)]
Merge branch 'timers-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull hrtimer fixlet from Thomas Gleixner:
 "A single commit annotating the lockcless access to timer->base with
  READ_ONCE() and adding the WRITE_ONCE() counterparts for completeness"

* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  hrtimer: Annotate lockless access to timer->base

5 years agoMerge branch 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 20 Oct 2019 10:22:25 +0000 (06:22 -0400)]
Merge branch 'core-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull stop-machine fix from Thomas Gleixner:
 "A single fix, amending stop machine with WRITE/READ_ONCE() to address
  the fallout of KCSAN"

* 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  stop_machine: Avoid potential race behaviour

5 years agoKVM: arm64: pmu: Reset sample period on overflow handling
Marc Zyngier [Sun, 6 Oct 2019 09:28:50 +0000 (10:28 +0100)]
KVM: arm64: pmu: Reset sample period on overflow handling

The PMU emulation code uses the perf event sample period to trigger
the overflow detection. This works fine  for the *first* overflow
handling, but results in a huge number of interrupts on the host,
unrelated to the number of interrupts handled in the guest (a x20
factor is pretty common for the cycle counter). On a slow system
(such as a SW model), this can result in the guest only making
forward progress at a glacial pace.

It turns out that the clue is in the name. The sample period is
exactly that: a period. And once the an overflow has occured,
the following period should be the full width of the associated
counter, instead of whatever the guest had initially programed.

Reset the sample period to the architected value in the overflow
handler, which now results in a number of host interrupts that is
much closer to the number of interrupts in the guest.

Fixes: b02386eb7dac ("arm64: KVM: Add PMU overflow interrupt routing")
Reviewed-by: Andrew Murray <andrew.murray@arm.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
5 years agoKVM: arm64: pmu: Set the CHAINED attribute before creating the in-kernel event
Marc Zyngier [Tue, 8 Oct 2019 14:09:55 +0000 (15:09 +0100)]
KVM: arm64: pmu: Set the CHAINED attribute before creating the in-kernel event

The current convention for KVM to request a chained event from the
host PMU is to set bit[0] in attr.config1 (PERF_ATTR_CFG1_KVM_PMU_CHAINED).

But as it turns out, this bit gets set *after* we create the kernel
event that backs our virtual counter, meaning that we never get
a 64bit counter.

Moving the setting to an earlier point solves the problem.

Fixes: 80f393a23be6 ("KVM: arm/arm64: Support chained PMU counters")
Reviewed-by: Andrew Murray <andrew.murray@arm.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
5 years agoarm64: KVM: Handle PMCR_EL0.LC as RES1 on pure AArch64 systems
Marc Zyngier [Fri, 4 Oct 2019 10:03:09 +0000 (11:03 +0100)]
arm64: KVM: Handle PMCR_EL0.LC as RES1 on pure AArch64 systems

Of PMCR_EL0.LC, the ARMv8 ARM says:

"In an AArch64 only implementation, this field is RES 1."

So be it.

Fixes: ab9468340d2bc ("arm64: KVM: Add access handler for PMCR register")
Reviewed-by: Andrew Murray <andrew.murray@arm.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
5 years agoKVM: arm64: pmu: Fix cycle counter truncation
Marc Zyngier [Thu, 3 Oct 2019 17:02:08 +0000 (18:02 +0100)]
KVM: arm64: pmu: Fix cycle counter truncation

When a counter is disabled, its value is sampled before the event
is being disabled, and the value written back in the shadow register.

In that process, the value gets truncated to 32bit, which is adequate
for any counter but the cycle counter (defined as a 64bit counter).

This obviously results in a corrupted counter, and things like
"perf record -e cycles" not working at all when run in a guest...
A similar, but less critical bug exists in kvm_pmu_get_counter_value.

Make the truncation conditional on the counter not being the cycle
counter, which results in a minor code reorganisation.

Fixes: 80f393a23be6 ("KVM: arm/arm64: Support chained PMU counters")
Reviewed-by: Andrew Murray <andrew.murray@arm.com>
Reported-by: Julien Thierry <julien.thierry.kdev@gmail.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Sat, 19 Oct 2019 21:09:11 +0000 (17:09 -0400)]
Merge git://git./linux/kernel/git/netdev/net

Pull networking fixes from David Miller:
 "I was battling a cold after some recent trips, so quite a bit piled up
  meanwhile, sorry about that.

  Highlights:

   1) Fix fd leak in various bpf selftests, from Brian Vazquez.

   2) Fix crash in xsk when device doesn't support some methods, from
      Magnus Karlsson.

   3) Fix various leaks and use-after-free in rxrpc, from David Howells.

   4) Fix several SKB leaks due to confusion of who owns an SKB and who
      should release it in the llc code. From Eric Biggers.

   5) Kill a bunc of KCSAN warnings in TCP, from Eric Dumazet.

   6) Jumbo packets don't work after resume on r8169, as the BIOS resets
      the chip into non-jumbo mode during suspend. From Heiner Kallweit.

   7) Corrupt L2 header during MPLS push, from Davide Caratti.

   8) Prevent possible infinite loop in tc_ctl_action, from Eric
      Dumazet.

   9) Get register bits right in bcmgenet driver, based upon chip
      version. From Florian Fainelli.

  10) Fix mutex problems in microchip DSA driver, from Marek Vasut.

  11) Cure race between route lookup and invalidation in ipv4, from Wei
      Wang.

  12) Fix performance regression due to false sharing in 'net'
      structure, from Eric Dumazet"

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (145 commits)
  net: reorder 'struct net' fields to avoid false sharing
  net: dsa: fix switch tree list
  net: ethernet: dwmac-sun8i: show message only when switching to promisc
  net: aquantia: add an error handling in aq_nic_set_multicast_list
  net: netem: correct the parent's backlog when corrupted packet was dropped
  net: netem: fix error path for corrupted GSO frames
  macb: propagate errors when getting optional clocks
  xen/netback: fix error path of xenvif_connect_data()
  net: hns3: fix mis-counting IRQ vector numbers issue
  net: usb: lan78xx: Connect PHY before registering MAC
  vsock/virtio: discard packets if credit is not respected
  vsock/virtio: send a credit update when buffer size is changed
  mlxsw: spectrum_trap: Push Ethernet header before reporting trap
  net: ensure correct skb->tstamp in various fragmenters
  net: bcmgenet: reset 40nm EPHY on energy detect
  net: bcmgenet: soft reset 40nm EPHYs before MAC init
  net: phy: bcm7xxx: define soft_reset for 40nm EPHY
  net: bcmgenet: don't set phydev->link from MAC
  net: Update address for MediaTek ethernet driver in MAINTAINERS
  ipv4: fix race condition between route lookup and invalidation
  ...

5 years agonet: reorder 'struct net' fields to avoid false sharing
Eric Dumazet [Fri, 18 Oct 2019 22:20:05 +0000 (15:20 -0700)]
net: reorder 'struct net' fields to avoid false sharing

Intel test robot reported a ~7% regression on TCP_CRR tests
that they bisected to the cited commit.

Indeed, every time a new TCP socket is created or deleted,
the atomic counter net->count is touched (via get_net(net)
and put_net(net) calls)

So cpus might have to reload a contended cache line in
net_hash_mix(net) calls.

We need to reorder 'struct net' fields to move @hash_mix
in a read mostly cache line.

We move in the first cache line fields that can be
dirtied often.

We probably will have to address in a followup patch
the __randomize_layout that was added in linux-4.13,
since this might break our placement choices.

Fixes: 355b98553789 ("netns: provide pure entropy for net_hash_mix()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: kernel test robot <oliver.sang@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: fix switch tree list
Vivien Didelot [Fri, 18 Oct 2019 21:02:46 +0000 (17:02 -0400)]
net: dsa: fix switch tree list

If there are multiple switch trees on the device, only the last one
will be listed, because the arguments of list_add_tail are swapped.

Fixes: 83c0afaec7b7 ("net: dsa: Add new binding implementation")
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: dwmac-sun8i: show message only when switching to promisc
Mans Rullgard [Fri, 18 Oct 2019 16:56:58 +0000 (17:56 +0100)]
net: ethernet: dwmac-sun8i: show message only when switching to promisc

Printing the info message every time more than the max number of mac
addresses are requested generates unnecessary log spam.  Showing it only
when the hw is not already in promiscous mode is equally informative
without being annoying.

Signed-off-by: Mans Rullgard <mans@mansr.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: aquantia: add an error handling in aq_nic_set_multicast_list
Chenwandun [Fri, 18 Oct 2019 10:20:37 +0000 (18:20 +0800)]
net: aquantia: add an error handling in aq_nic_set_multicast_list

add an error handling in aq_nic_set_multicast_list, it may not
work when hw_multicast_list_set error; and at the same time
it will remove gcc Wunused-but-set-variable warning.

Signed-off-by: Chenwandun <chenwandun@huawei.com>
Reviewed-by: Igor Russkikh <igor.russkikh@aquantia.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'netem-fix-further-issues-with-packet-corruption'
David S. Miller [Sat, 19 Oct 2019 19:12:36 +0000 (12:12 -0700)]
Merge branch 'netem-fix-further-issues-with-packet-corruption'

Jakub Kicinski says:

====================
net: netem: fix further issues with packet corruption

This set is fixing two more issues with the netem packet corruption.

First patch (which was previously posted) avoids NULL pointer dereference
if the first frame gets freed due to allocation or checksum failure.
v2 improves the clarity of the code a little as requested by Cong.

Second patch ensures we don't return SUCCESS if the frame was in fact
dropped. Thanks to this commit message for patch 1 no longer needs the
"this will still break with a single-frame failure" disclaimer.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: netem: correct the parent's backlog when corrupted packet was dropped
Jakub Kicinski [Fri, 18 Oct 2019 16:16:58 +0000 (09:16 -0700)]
net: netem: correct the parent's backlog when corrupted packet was dropped

If packet corruption failed we jump to finish_segs and return
NET_XMIT_SUCCESS. Seeing success will make the parent qdisc
increment its backlog, that's incorrect - we need to return
NET_XMIT_DROP.

Fixes: 6071bd1aa13e ("netem: Segment GSO packets on enqueue")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: netem: fix error path for corrupted GSO frames
Jakub Kicinski [Fri, 18 Oct 2019 16:16:57 +0000 (09:16 -0700)]
net: netem: fix error path for corrupted GSO frames

To corrupt a GSO frame we first perform segmentation.  We then
proceed using the first segment instead of the full GSO skb and
requeue the rest of the segments as separate packets.

If there are any issues with processing the first segment we
still want to process the rest, therefore we jump to the
finish_segs label.

Commit 177b8007463c ("net: netem: fix backlog accounting for
corrupted GSO frames") started using the pointer to the first
segment in the "rest of segments processing", but as mentioned
above the first segment may had already been freed at this point.

Backlog corrections for parent qdiscs have to be adjusted.

Fixes: 177b8007463c ("net: netem: fix backlog accounting for corrupted GSO frames")
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Reported-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>