Currently, KVM passes PVCLOCK_TSC_STABLE_BIT to its guests when running in
so called 'masterclock' mode and this is only possible when the clocksource
on the host is TSC. When running nested on Hyper-V we're using a different
clocksource in L1 (Hyper-V TSC Page) which can actually be used for
masterclock. This series brings the required support.
Making KVM work with TSC page clocksource is relatively easy, it is done in
PATCH 5 of the series. All the rest is required to support L1 migration
when TSC frequency changes, we use a special feature from Hyper-V to do
the job.
Vitaly Kuznetsov (6):
x86/hyper-v: check for required priviliges in hyperv_init()
x86/hyper-v: add a function to read both TSC and TSC page value
simulateneously
x86/hyper-v: reenlightenment notifications support
x86/hyper-v: redirect reenlightment notifications on CPU offlining
x86/kvm: pass stable clocksource to guests when running nested on
Hyper-V
x86/kvm: support Hyper-V reenlightenment
arch/x86/entry/entry_64.S | 4 ++
arch/x86/hyperv/hv_init.c | 108 ++++++++++++++++++++++++++++++++++++-
arch/x86/include/asm/entry_arch.h | 4 ++
arch/x86/include/asm/hw_irq.h | 1 +
arch/x86/include/asm/irq_vectors.h | 7 ++-
arch/x86/include/asm/mshyperv.h | 32 +++++++++--
arch/x86/include/uapi/asm/hyperv.h | 27 ++++++++++
arch/x86/kernel/idt.c | 3 ++
arch/x86/kvm/x86.c | 104 +++++++++++++++++++++++++++++------
9 files changed, 267 insertions(+), 23 deletions(-)
--
2.14.3
In hyperv_init() we presume we always have access to VP index and hypercall
MSRs while according to the specification we should check if we're allowed
to access the corresponding MSRs before accessing them.
Signed-off-by: Vitaly Kuznetsov <[email protected]>
---
arch/x86/hyperv/hv_init.c | 9 ++++++++-
1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/arch/x86/hyperv/hv_init.c b/arch/x86/hyperv/hv_init.c
index 189a398290db..3ebc7f95e6ca 100644
--- a/arch/x86/hyperv/hv_init.c
+++ b/arch/x86/hyperv/hv_init.c
@@ -110,12 +110,19 @@ static int hv_cpu_init(unsigned int cpu)
*/
void hyperv_init(void)
{
- u64 guest_id;
+ u64 guest_id, requied_msrs;
union hv_x64_msr_hypercall_contents hypercall_msr;
if (x86_hyper_type != X86_HYPER_MS_HYPERV)
return;
+ /* Absolutely required MSRs */
+ requied_msrs = HV_X64_MSR_HYPERCALL_AVAILABLE |
+ HV_X64_MSR_VP_INDEX_AVAILABLE;
+
+ if ((ms_hyperv.features & requied_msrs) != requied_msrs)
+ return;
+
/* Allocate percpu VP index */
hv_vp_index = kmalloc_array(num_possible_cpus(), sizeof(*hv_vp_index),
GFP_KERNEL);
--
2.14.3
Hyper-V supports Live Migration notification. This is supposed to be used
in conjunction with TSC emulation: when we are migrated to a host with
different TSC frequency for some short period host emulates our accesses
to TSC and sends us an interrupt to notify about the event. When we're
done updating everything we can disable TSC emulation and everything will
start working fast again.
We didn't need these notifications before as Hyper-V guests are not
supposed to use TSC as a clocksource: in Linux we even mark it as unstable
on boot. Guests normally use 'tsc page' clocksouce and host updates its
values on migrations automatically.
Things change when we want to run nested virtualization: even when we pass
through PV clocksources (kvm-clock or tsc page) to our guests we need to
know TSC frequency and when it changes.
Hyper-V Top Level Functional Specification (as of v5.0b) wrongly specifies
EAX:BIT(12) of CPUID:0x40000009 as the feature identification bit. The
right one to check is EAX:BIT(13) of CPUID:0x40000003. I was assured that
the fix in on the way.
Signed-off-by: Vitaly Kuznetsov <[email protected]>
---
arch/x86/entry/entry_64.S | 4 +++
arch/x86/hyperv/hv_init.c | 70 ++++++++++++++++++++++++++++++++++++++
arch/x86/include/asm/entry_arch.h | 4 +++
arch/x86/include/asm/hw_irq.h | 1 +
arch/x86/include/asm/irq_vectors.h | 7 +++-
arch/x86/include/asm/mshyperv.h | 8 +++++
arch/x86/include/uapi/asm/hyperv.h | 27 +++++++++++++++
arch/x86/kernel/idt.c | 3 ++
8 files changed, 123 insertions(+), 1 deletion(-)
diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S
index a2b30ec69497..67d13b0af792 100644
--- a/arch/x86/entry/entry_64.S
+++ b/arch/x86/entry/entry_64.S
@@ -822,6 +822,10 @@ apicinterrupt SPURIOUS_APIC_VECTOR spurious_interrupt smp_spurious_interrupt
apicinterrupt IRQ_WORK_VECTOR irq_work_interrupt smp_irq_work_interrupt
#endif
+#if IS_ENABLED(CONFIG_HYPERV)
+apicinterrupt HYPERV_REENLIGHTENMENT_VECTOR hyperv_reenlightenment_intr smp_hyperv_reenlightenment_intr
+#endif
+
/*
* Exception entry points.
*/
diff --git a/arch/x86/hyperv/hv_init.c b/arch/x86/hyperv/hv_init.c
index 3ebc7f95e6ca..b6e2dc28ff05 100644
--- a/arch/x86/hyperv/hv_init.c
+++ b/arch/x86/hyperv/hv_init.c
@@ -101,6 +101,76 @@ static int hv_cpu_init(unsigned int cpu)
return 0;
}
+static void (*hv_reenlightenment_cb)(void);
+
+static void hv_reenlightenment_notify(struct work_struct *dummy)
+{
+ if (hv_reenlightenment_cb)
+ hv_reenlightenment_cb();
+}
+static DECLARE_WORK(hv_reenlightenment_work, hv_reenlightenment_notify);
+
+void hyperv_stop_tsc_emulation(void)
+{
+ u64 freq;
+ struct hv_tsc_emulation_status emu_status;
+
+ rdmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
+ emu_status.inprogress = 0;
+ wrmsrl(HV_X64_MSR_TSC_EMULATION_STATUS, *(u64 *)&emu_status);
+
+ rdmsrl(HV_X64_MSR_TSC_FREQUENCY, freq);
+ tsc_khz = freq/1000;
+}
+EXPORT_SYMBOL_GPL(hyperv_stop_tsc_emulation);
+
+void register_hv_tsc_update(void (*cb)(void))
+{
+ struct hv_reenlightenment_control re_ctrl = {
+ .vector = HYPERV_REENLIGHTENMENT_VECTOR,
+ .enabled = 1,
+ .target_vp = hv_vp_index[smp_processor_id()]
+ };
+ struct hv_tsc_emulation_control emu_ctrl = {.enabled = 1};
+
+ if (!(ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT))
+ return;
+
+ hv_reenlightenment_cb = cb;
+
+ /* Make sure callback is registered before we write to MSRs */
+ wmb();
+
+ wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
+ wrmsrl(HV_X64_MSR_TSC_EMULATION_CONTROL, *((u64 *)&emu_ctrl));
+}
+EXPORT_SYMBOL_GPL(register_hv_tsc_update);
+
+void unregister_hv_tsc_update(void)
+{
+ struct hv_reenlightenment_control re_ctrl;
+
+ if (!(ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT))
+ return;
+
+ rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
+ re_ctrl.enabled = 0;
+ wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *(u64 *)&re_ctrl);
+
+ hv_reenlightenment_cb = NULL;
+}
+EXPORT_SYMBOL_GPL(unregister_hv_tsc_update);
+
+asmlinkage __visible void
+__irq_entry smp_hyperv_reenlightenment_intr(struct pt_regs *regs)
+{
+ entering_ack_irq();
+
+ schedule_work(&hv_reenlightenment_work);
+
+ exiting_irq();
+}
+
/*
* This function is to be invoked early in the boot sequence after the
* hypervisor has been detected.
diff --git a/arch/x86/include/asm/entry_arch.h b/arch/x86/include/asm/entry_arch.h
index 416422762845..eb936cc49b62 100644
--- a/arch/x86/include/asm/entry_arch.h
+++ b/arch/x86/include/asm/entry_arch.h
@@ -54,3 +54,7 @@ BUILD_INTERRUPT(threshold_interrupt,THRESHOLD_APIC_VECTOR)
BUILD_INTERRUPT(deferred_error_interrupt, DEFERRED_ERROR_VECTOR)
#endif
#endif
+
+#if IS_ENABLED(CONFIG_HYPERV)
+BUILD_INTERRUPT(hyperv_reenlightenment_intr, HYPERV_REENLIGHTENMENT_VECTOR)
+#endif
diff --git a/arch/x86/include/asm/hw_irq.h b/arch/x86/include/asm/hw_irq.h
index b80e46733909..2f365c537d5f 100644
--- a/arch/x86/include/asm/hw_irq.h
+++ b/arch/x86/include/asm/hw_irq.h
@@ -36,6 +36,7 @@ extern asmlinkage void kvm_posted_intr_wakeup_ipi(void);
extern asmlinkage void kvm_posted_intr_nested_ipi(void);
extern asmlinkage void error_interrupt(void);
extern asmlinkage void irq_work_interrupt(void);
+extern asmlinkage void hyperv_reenlightenment_intr(void);
extern asmlinkage void spurious_interrupt(void);
extern asmlinkage void thermal_interrupt(void);
diff --git a/arch/x86/include/asm/irq_vectors.h b/arch/x86/include/asm/irq_vectors.h
index 67421f649cfa..e71c1120426b 100644
--- a/arch/x86/include/asm/irq_vectors.h
+++ b/arch/x86/include/asm/irq_vectors.h
@@ -103,7 +103,12 @@
#endif
#define MANAGED_IRQ_SHUTDOWN_VECTOR 0xef
-#define LOCAL_TIMER_VECTOR 0xee
+
+#if IS_ENABLED(CONFIG_HYPERV)
+#define HYPERV_REENLIGHTENMENT_VECTOR 0xee
+#endif
+
+#define LOCAL_TIMER_VECTOR 0xed
#define NR_VECTORS 256
diff --git a/arch/x86/include/asm/mshyperv.h b/arch/x86/include/asm/mshyperv.h
index 2611d2c49f27..6b822e116572 100644
--- a/arch/x86/include/asm/mshyperv.h
+++ b/arch/x86/include/asm/mshyperv.h
@@ -314,11 +314,19 @@ void hyper_alloc_mmu(void);
void hyperv_report_panic(struct pt_regs *regs, long err);
bool hv_is_hypercall_page_setup(void);
void hyperv_cleanup(void);
+
+asmlinkage void smp_hyperv_reenlightenment_intr(struct pt_regs *regs);
+void register_hv_tsc_update(void (*cb)(void));
+void unregister_hv_tsc_update(void);
+void hyperv_stop_tsc_emulation(void);
#else /* CONFIG_HYPERV */
static inline void hyperv_init(void) {}
static inline bool hv_is_hypercall_page_setup(void) { return false; }
static inline void hyperv_cleanup(void) {}
static inline void hyperv_setup_mmu_ops(void) {}
+static inline void register_hv_tsc_update(void (*cb)(void)) {}
+static inline void unregister_hv_tsc_update(void) {}
+static inline void hyperv_stop_tsc_emulation(void) {};
#endif /* CONFIG_HYPERV */
#ifdef CONFIG_HYPERV_TSCPAGE
diff --git a/arch/x86/include/uapi/asm/hyperv.h b/arch/x86/include/uapi/asm/hyperv.h
index 1a5bfead93b4..197c2e6c7376 100644
--- a/arch/x86/include/uapi/asm/hyperv.h
+++ b/arch/x86/include/uapi/asm/hyperv.h
@@ -40,6 +40,9 @@
*/
#define HV_X64_ACCESS_FREQUENCY_MSRS (1 << 11)
+/* AccessReenlightenmentControls privilege */
+#define HV_X64_ACCESS_REENLIGHTENMENT BIT(13)
+
/*
* Basic SynIC MSRs (HV_X64_MSR_SCONTROL through HV_X64_MSR_EOM
* and HV_X64_MSR_SINT0 through HV_X64_MSR_SINT15) available
@@ -234,6 +237,30 @@
#define HV_X64_MSR_CRASH_PARAMS \
(1 + (HV_X64_MSR_CRASH_P4 - HV_X64_MSR_CRASH_P0))
+/* TSC emulation after migration */
+#define HV_X64_MSR_REENLIGHTENMENT_CONTROL 0x40000106
+
+struct hv_reenlightenment_control {
+ u64 vector:8;
+ u64 reserved1:8;
+ u64 enabled:1;
+ u64 reserved2:15;
+ u64 target_vp:32;
+};
+
+#define HV_X64_MSR_TSC_EMULATION_CONTROL 0x40000107
+#define HV_X64_MSR_TSC_EMULATION_STATUS 0x40000108
+
+struct hv_tsc_emulation_control {
+ u64 enabled:1;
+ u64 reserved:63;
+};
+
+struct hv_tsc_emulation_status {
+ u64 inprogress:1;
+ u64 reserved:63;
+};
+
#define HV_X64_MSR_HYPERCALL_ENABLE 0x00000001
#define HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT 12
#define HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_MASK \
diff --git a/arch/x86/kernel/idt.c b/arch/x86/kernel/idt.c
index d985cef3984f..5b8512d48aa3 100644
--- a/arch/x86/kernel/idt.c
+++ b/arch/x86/kernel/idt.c
@@ -140,6 +140,9 @@ static const __initdata struct idt_data apic_idts[] = {
# ifdef CONFIG_IRQ_WORK
INTG(IRQ_WORK_VECTOR, irq_work_interrupt),
# endif
+#if IS_ENABLED(CONFIG_HYPERV)
+ INTG(HYPERV_REENLIGHTENMENT_VECTOR, hyperv_reenlightenment_intr),
+#endif
INTG(SPURIOUS_APIC_VECTOR, spurious_interrupt),
INTG(ERROR_APIC_VECTOR, error_interrupt),
#endif
--
2.14.3
When we run nested KVM on Hyper-V guests we need to update masterclocks for
all guests when L1 migrates to a host with different TSC frequency.
Implement the procedure in the following way:
- Pause all guests.
- Tell our host (Hyper-V) to stop emulating TSC accesses.
- Update our gtod copy, recompute clocks.
- Unpause all guests.
This is somewhat similar to cpufreq but we have two important differences:
we can only disable TSC emulation globally (on all CPUs) and we don't know
the new TSC frequency until we turn the emulation off so we can't
'prepare' ourselves to the event.
Signed-off-by: Vitaly Kuznetsov <[email protected]>
---
arch/x86/kvm/x86.c | 45 +++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 45 insertions(+)
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 42063b6d8e48..e4fee6db70a8 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -68,6 +68,7 @@
#include <asm/div64.h>
#include <asm/irq_remapping.h>
#include <asm/mshyperv.h>
+#include <asm/hypervisor.h>
#define CREATE_TRACE_POINTS
#include "trace.h"
@@ -5913,6 +5914,43 @@ static void tsc_khz_changed(void *data)
__this_cpu_write(cpu_tsc_khz, khz);
}
+void kvm_hyperv_tsc_notifier(void)
+{
+#ifdef CONFIG_X86_64
+ struct kvm *kvm;
+ struct kvm_vcpu *vcpu;
+ int cpu;
+
+ spin_lock(&kvm_lock);
+ list_for_each_entry(kvm, &vm_list, vm_list)
+ kvm_make_mclock_inprogress_request(kvm);
+
+ hyperv_stop_tsc_emulation();
+
+ /* TSC frequency always matches when on Hyper-V */
+ for_each_present_cpu(cpu)
+ per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
+ kvm_max_guest_tsc_khz = tsc_khz;
+
+ list_for_each_entry(kvm, &vm_list, vm_list) {
+ struct kvm_arch *ka = &kvm->arch;
+
+ spin_lock(&ka->pvclock_gtod_sync_lock);
+
+ pvclock_update_vm_gtod_copy(kvm);
+
+ kvm_for_each_vcpu(cpu, vcpu, kvm)
+ kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
+
+ kvm_for_each_vcpu(cpu, vcpu, kvm)
+ kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
+
+ spin_unlock(&ka->pvclock_gtod_sync_lock);
+ }
+ spin_unlock(&kvm_lock);
+#endif
+}
+
static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
void *data)
{
@@ -6198,6 +6236,9 @@ int kvm_arch_init(void *opaque)
kvm_lapic_init();
#ifdef CONFIG_X86_64
pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
+
+ if (x86_hyper_type == X86_HYPER_MS_HYPERV)
+ register_hv_tsc_update(kvm_hyperv_tsc_notifier);
#endif
return 0;
@@ -6210,6 +6251,10 @@ int kvm_arch_init(void *opaque)
void kvm_arch_exit(void)
{
+#ifdef CONFIG_X86_64
+ if (x86_hyper_type == X86_HYPER_MS_HYPERV)
+ unregister_hv_tsc_update();
+#endif
kvm_lapic_exit();
perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
--
2.14.3
Currently, KVM is able to work in 'masterclock' mode passing
PVCLOCK_TSC_STABLE_BIT to guests when the clocksource we use on the host
is TSC. When running nested on Hyper-V we normally use a different one:
TSC page which is resistant to TSC frequency changes on event like L1
migration. Add support for it in KVM.
The only non-trivial change in the patch is in vgettsc(): when updating
our gtod copy we now need to get both the clockread and tsc value.
Signed-off-by: Vitaly Kuznetsov <[email protected]>
---
arch/x86/kvm/x86.c | 59 +++++++++++++++++++++++++++++++++++++++---------------
1 file changed, 43 insertions(+), 16 deletions(-)
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 34c85aa2e2d1..42063b6d8e48 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -67,6 +67,7 @@
#include <asm/pvclock.h>
#include <asm/div64.h>
#include <asm/irq_remapping.h>
+#include <asm/mshyperv.h>
#define CREATE_TRACE_POINTS
#include "trace.h"
@@ -1374,6 +1375,11 @@ static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
return tsc;
}
+static inline int gtod_cs_mode_good(int mode)
+{
+ return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
+}
+
static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
{
#ifdef CONFIG_X86_64
@@ -1393,7 +1399,7 @@ static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
* perform request to enable masterclock.
*/
if (ka->use_master_clock ||
- (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
+ (gtod_cs_mode_good(gtod->clock.vclock_mode) && vcpus_matched))
kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
@@ -1456,6 +1462,19 @@ static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
vcpu->arch.tsc_offset = offset;
}
+static inline bool kvm_check_tsc_unstable(void)
+{
+#ifdef CONFIG_X86_64
+ /*
+ * TSC is marked unstable when we're running on Hyper-V,
+ * 'TSC page' clocksource is good.
+ */
+ if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
+ return false;
+#endif
+ return check_tsc_unstable();
+}
+
void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
{
struct kvm *kvm = vcpu->kvm;
@@ -1501,7 +1520,7 @@ void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
*/
if (synchronizing &&
vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
- if (!check_tsc_unstable()) {
+ if (!kvm_check_tsc_unstable()) {
offset = kvm->arch.cur_tsc_offset;
pr_debug("kvm: matched tsc offset for %llu\n", data);
} else {
@@ -1606,9 +1625,17 @@ static inline u64 vgettsc(u64 *cycle_now)
long v;
struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
- *cycle_now = read_tsc();
+ if (gtod->clock.vclock_mode == VCLOCK_HVCLOCK) {
+ u64 tsc_pg_val;
+
+ tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(), cycle_now);
+ v = (tsc_pg_val - gtod->clock.cycle_last) & gtod->clock.mask;
+ } else {
+ /* VCLOCK_TSC */
+ *cycle_now = read_tsc();
+ v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
+ }
- v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
return v * gtod->clock.mult;
}
@@ -1654,25 +1681,25 @@ static int do_realtime(struct timespec *ts, u64 *cycle_now)
return mode;
}
-/* returns true if host is using tsc clocksource */
+/* returns true if host is using TSC based clocksource */
static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
{
/* checked again under seqlock below */
- if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
+ if (!gtod_cs_mode_good(pvclock_gtod_data.clock.vclock_mode))
return false;
- return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
+ return gtod_cs_mode_good(do_monotonic_boot(kernel_ns, cycle_now));
}
-/* returns true if host is using tsc clocksource */
+/* returns true if host is using TSC based clocksource */
static bool kvm_get_walltime_and_clockread(struct timespec *ts,
u64 *cycle_now)
{
/* checked again under seqlock below */
- if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
+ if (!gtod_cs_mode_good(pvclock_gtod_data.clock.vclock_mode))
return false;
- return do_realtime(ts, cycle_now) == VCLOCK_TSC;
+ return gtod_cs_mode_good(do_realtime(ts, cycle_now));
}
#endif
@@ -2854,13 +2881,13 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
}
- if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
+ if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
rdtsc() - vcpu->arch.last_host_tsc;
if (tsc_delta < 0)
mark_tsc_unstable("KVM discovered backwards TSC");
- if (check_tsc_unstable()) {
+ if (kvm_check_tsc_unstable()) {
u64 offset = kvm_compute_tsc_offset(vcpu,
vcpu->arch.last_guest_tsc);
kvm_vcpu_write_tsc_offset(vcpu, offset);
@@ -6107,9 +6134,9 @@ static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
update_pvclock_gtod(tk);
/* disable master clock if host does not trust, or does not
- * use, TSC clocksource
+ * use, TSC based clocksource.
*/
- if (gtod->clock.vclock_mode != VCLOCK_TSC &&
+ if (!gtod_cs_mode_good(gtod->clock.vclock_mode) &&
atomic_read(&kvm_guest_has_master_clock) != 0)
queue_work(system_long_wq, &pvclock_gtod_work);
@@ -7735,7 +7762,7 @@ struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
{
struct kvm_vcpu *vcpu;
- if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
+ if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
printk_once(KERN_WARNING
"kvm: SMP vm created on host with unstable TSC; "
"guest TSC will not be reliable\n");
@@ -7889,7 +7916,7 @@ int kvm_arch_hardware_enable(void)
return ret;
local_tsc = rdtsc();
- stable = !check_tsc_unstable();
+ stable = !kvm_check_tsc_unstable();
list_for_each_entry(kvm, &vm_list, vm_list) {
kvm_for_each_vcpu(i, vcpu, kvm) {
if (!stable && vcpu->cpu == smp_processor_id())
--
2.14.3
It is very unlikely for CPUs to get offlined when we run on Hyper-V as
we have a protection in vmbus module which prevents it when we have any
VMBus devices assigned. This, however, may change in future if an option
to reassign an already active channel will be added. It is also possible
to run without any Hyper-V devices of have a CPU with no assigned channels.
Reassign reenlightenment notifications to some other active CPU when
the CPU which is assigned to get them goes offline.
Signed-off-by: Vitaly Kuznetsov <[email protected]>
---
arch/x86/hyperv/hv_init.c | 29 ++++++++++++++++++++++++++++-
1 file changed, 28 insertions(+), 1 deletion(-)
diff --git a/arch/x86/hyperv/hv_init.c b/arch/x86/hyperv/hv_init.c
index b6e2dc28ff05..6ef83d6232cb 100644
--- a/arch/x86/hyperv/hv_init.c
+++ b/arch/x86/hyperv/hv_init.c
@@ -171,6 +171,33 @@ __irq_entry smp_hyperv_reenlightenment_intr(struct pt_regs *regs)
exiting_irq();
}
+static int hv_cpu_die(unsigned int cpu)
+{
+ struct hv_reenlightenment_control re_ctrl;
+ int i;
+
+ if ((ms_hyperv.features & HV_X64_ACCESS_REENLIGHTENMENT) &&
+ hv_reenlightenment_cb != NULL) {
+ rdmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL, *((u64 *)&re_ctrl));
+ if (re_ctrl.target_vp != hv_vp_index[cpu])
+ return 0;
+
+ /* Find some other online CPU */
+ for_each_online_cpu(i) {
+ if (i == cpu)
+ continue;
+
+ re_ctrl.target_vp = hv_vp_index[i];
+ wrmsrl(HV_X64_MSR_REENLIGHTENMENT_CONTROL,
+ *((u64 *)&re_ctrl));
+
+ break;
+ }
+ }
+
+ return 0;
+}
+
/*
* This function is to be invoked early in the boot sequence after the
* hypervisor has been detected.
@@ -200,7 +227,7 @@ void hyperv_init(void)
return;
if (cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv_init:online",
- hv_cpu_init, NULL) < 0)
+ hv_cpu_init, hv_cpu_die) < 0)
goto free_vp_index;
/*
--
2.14.3
This is going to be used from KVM code were we need to get both
TSC and TSC page value.
When Hyper-V code is compiled out just return rdtsc(), this will allow us
to avoid ugly ifdefs in non-Hyper-V code.
Signed-off-by: Vitaly Kuznetsov <[email protected]>
---
arch/x86/include/asm/mshyperv.h | 24 ++++++++++++++++++++----
1 file changed, 20 insertions(+), 4 deletions(-)
diff --git a/arch/x86/include/asm/mshyperv.h b/arch/x86/include/asm/mshyperv.h
index 5400add2885b..2611d2c49f27 100644
--- a/arch/x86/include/asm/mshyperv.h
+++ b/arch/x86/include/asm/mshyperv.h
@@ -323,9 +323,10 @@ static inline void hyperv_setup_mmu_ops(void) {}
#ifdef CONFIG_HYPERV_TSCPAGE
struct ms_hyperv_tsc_page *hv_get_tsc_page(void);
-static inline u64 hv_read_tsc_page(const struct ms_hyperv_tsc_page *tsc_pg)
+static inline u64 hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg,
+ u64 *cur_tsc)
{
- u64 scale, offset, cur_tsc;
+ u64 scale, offset;
u32 sequence;
/*
@@ -356,7 +357,7 @@ static inline u64 hv_read_tsc_page(const struct ms_hyperv_tsc_page *tsc_pg)
scale = READ_ONCE(tsc_pg->tsc_scale);
offset = READ_ONCE(tsc_pg->tsc_offset);
- cur_tsc = rdtsc_ordered();
+ *cur_tsc = rdtsc_ordered();
/*
* Make sure we read sequence after we read all other values
@@ -366,7 +367,14 @@ static inline u64 hv_read_tsc_page(const struct ms_hyperv_tsc_page *tsc_pg)
} while (READ_ONCE(tsc_pg->tsc_sequence) != sequence);
- return mul_u64_u64_shr(cur_tsc, scale, 64) + offset;
+ return mul_u64_u64_shr(*cur_tsc, scale, 64) + offset;
+}
+
+static inline u64 hv_read_tsc_page(const struct ms_hyperv_tsc_page *tsc_pg)
+{
+ u64 cur_tsc;
+
+ return hv_read_tsc_page_tsc(tsc_pg, &cur_tsc);
}
#else
@@ -374,5 +382,13 @@ static inline struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
{
return NULL;
}
+
+static inline u64 hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg,
+ u64 *cur_tsc)
+{
+ *cur_tsc = rdtsc();
+
+ return cur_tsc;
+}
#endif
#endif
--
2.14.3
On 01/12/2017 14:13, Vitaly Kuznetsov wrote:
> Currently, KVM passes PVCLOCK_TSC_STABLE_BIT to its guests when running in
> so called 'masterclock' mode and this is only possible when the clocksource
> on the host is TSC. When running nested on Hyper-V we're using a different
> clocksource in L1 (Hyper-V TSC Page) which can actually be used for
> masterclock. This series brings the required support.
>
> Making KVM work with TSC page clocksource is relatively easy, it is done in
> PATCH 5 of the series. All the rest is required to support L1 migration
> when TSC frequency changes, we use a special feature from Hyper-V to do
> the job.
This looks great. If the Hyper-V folks can provide a topic branch with
patches 1-4 in it, we can merge it.
Thanks,
Paolo
On Fri, 1 Dec 2017 14:13:17 +0100
Vitaly Kuznetsov <[email protected]> wrote:
> +
> +static inline u64 hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg,
> + u64 *cur_tsc)
> +{
> + *cur_tsc = rdtsc();
> +
> + return cur_tsc;
Why do return and setting by reference. Looks like an ugly API.
2017-12-01 14:13+0100, Vitaly Kuznetsov:
> Currently, KVM is able to work in 'masterclock' mode passing
> PVCLOCK_TSC_STABLE_BIT to guests when the clocksource we use on the host
> is TSC. When running nested on Hyper-V we normally use a different one:
> TSC page which is resistant to TSC frequency changes on event like L1
> migration. Add support for it in KVM.
>
> The only non-trivial change in the patch is in vgettsc(): when updating
> our gtod copy we now need to get both the clockread and tsc value.
>
> Signed-off-by: Vitaly Kuznetsov <[email protected]>
> ---
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> @@ -1374,6 +1375,11 @@ static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
> +static inline int gtod_cs_mode_good(int mode)
"good" isn't saying much; I'd like to express that TSC is the underlying
clock ...
What about "bool gtod_is_based_on_tsc()"?
> +{
> + return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
> +}
> +
> @@ -1606,9 +1625,17 @@ static inline u64 vgettsc(u64 *cycle_now)
> long v;
> struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
>
> - *cycle_now = read_tsc();
> + if (gtod->clock.vclock_mode == VCLOCK_HVCLOCK) {
> + u64 tsc_pg_val;
> +
> + tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(), cycle_now);
This function might fail to update cycle_now and return -1.
I guess we should propagate the failure in that case.
> + v = (tsc_pg_val - gtod->clock.cycle_last) & gtod->clock.mask;
> + } else {
> + /* VCLOCK_TSC */
> + *cycle_now = read_tsc();
> + v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
cycle_now is getting pretty confusing -- it still is TSC timestamp, but
now we also have the current cycle of gtod, which might be the TSC page
timestamp. Please rename cycle_now to tsc_timestamp in the call tree,
thanks.
On 01/12/2017 18:29, Stephen Hemminger wrote:
>> +static inline u64 hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg,
>> + u64 *cur_tsc)
>> +{
>> + *cur_tsc = rdtsc();
>> +
>> + return cur_tsc;
> Why do return and setting by reference. Looks like an ugly API.
This is the fallback implementation for !CONFIG_HYPERV_TSCPAGE, which
explains why it's ugly, but why is it needed at all (or it could just
BUG())?
Thanks,
Paolo
Paolo Bonzini <[email protected]> writes:
> On 01/12/2017 18:29, Stephen Hemminger wrote:
>>> +static inline u64 hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg,
>>> + u64 *cur_tsc)
>>> +{
>>> + *cur_tsc = rdtsc();
>>> +
>>> + return cur_tsc;
>> Why do return and setting by reference. Looks like an ugly API.
>
> This is the fallback implementation for !CONFIG_HYPERV_TSCPAGE, which
> explains why it's ugly, but why is it needed at all (or it could just
> BUG())?
>
It is not needed indeed, the intention was just to avoid '#if
IS_ENABLED(CONFIG_HYPERV)' in kvm code. I can replace it with BUG() or
return U64_MAX;
--
Vitaly