[PATCH v2] KVM: halt_polling: provide a way to qualify wakeups during poll

[PATCH v2] KVM: halt_polling: provide a way to qualify wakeups during poll

[Christian Borntraeger]

I removed all Reviewed-by/Acks as the patch has changed in several places
- got rid of two new lines in Kconfig
- changed comment in interrupt.c
- added kvm_arch_vcpu_block_finish
- got rid of vcpu_reset_wakeup and vcpu_set_valid_wakeup
- add new kvm_stat 
- adopt tracing
- rebase from 4.5 to kvm/next

Better names/descriptions for
- kvm_stat "halt_poll_no_tuning"
- the trace text
- function names 
are welcome

Christian Borntraeger (1):
  KVM: halt_polling: provide a way to qualify wakeups during poll

 arch/arm/include/asm/kvm_host.h     |  2 ++
 arch/arm64/include/asm/kvm_host.h   |  2 ++
 arch/mips/include/asm/kvm_host.h    |  1 +
 arch/mips/kvm/mips.c                |  1 +
 arch/powerpc/include/asm/kvm_host.h |  2 ++
 arch/powerpc/kvm/book3s.c           |  1 +
 arch/powerpc/kvm/booke.c            |  1 +
 arch/s390/include/asm/kvm_host.h    |  3 +++
 arch/s390/kvm/Kconfig               |  1 +
 arch/s390/kvm/interrupt.c           |  5 +++++
 arch/s390/kvm/kvm-s390.c            |  6 ++++++
 arch/x86/include/asm/kvm_host.h     |  2 ++
 arch/x86/kvm/x86.c                  |  1 +
 include/linux/kvm_host.h            | 15 +++++++++++++++
 include/trace/events/kvm.h          | 11 +++++++----
 virt/kvm/Kconfig                    |  3 +++
 virt/kvm/kvm_main.c                 |  8 ++++++--
 17 files changed, 59 insertions(+), 6 deletions(-)

-- 
2.3.0

[PATCH 1/1] KVM: halt_polling: provide a way to qualify wakeups during poll

[Christian Borntraeger]

Some wakeups should not be considered a sucessful poll. For example on
s390 I/O interrupts are usually floating, which means that _ALL_ CPUs
would be considered runnable - letting all vCPUs poll all the time for
transactional like workload, even if one vCPU would be enough.
This can result in huge CPU usage for large guests.
This patch lets architectures provide a way to qualify wakeups if they
should be considered a good/bad wakeups in regard to polls.

For s390 the implementation will fence of halt polling for anything but
known good, single vCPU events. The s390 implementation for floating
interrupts does a wakeup for one vCPU, but the interrupt will be delivered
by whatever CPU checks first for a pending interrupt. We prefer the
woken up CPU by marking the poll of this CPU as "good" poll.
This code will also mark several other wakeup reasons like IPI or
expired timers as "good". This will of course also mark some events as
not sucessful. As  KVM on z runs always as a 2nd level hypervisor,
we prefer to not poll, unless we are really sure, though.

This patch successfully limits the CPU usage for cases like uperf 1byte
transactional ping pong workload or wakeup heavy workload like OLTP
while still providing a proper speedup.

This also introduced a new vcpu stat "halt_poll_no_tuning" that marks
wakeups that are considered not good for polling.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: David Matlack <dmatlack@google.com>
Cc: Wanpeng Li <kernellwp@gmail.com>
---
 arch/arm/include/asm/kvm_host.h     |  2 ++
 arch/arm64/include/asm/kvm_host.h   |  2 ++
 arch/mips/include/asm/kvm_host.h    |  1 +
 arch/mips/kvm/mips.c                |  1 +
 arch/powerpc/include/asm/kvm_host.h |  2 ++
 arch/powerpc/kvm/book3s.c           |  1 +
 arch/powerpc/kvm/booke.c            |  1 +
 arch/s390/include/asm/kvm_host.h    |  3 +++
 arch/s390/kvm/Kconfig               |  1 +
 arch/s390/kvm/interrupt.c           |  5 +++++
 arch/s390/kvm/kvm-s390.c            |  6 ++++++
 arch/x86/include/asm/kvm_host.h     |  2 ++
 arch/x86/kvm/x86.c                  |  1 +
 include/linux/kvm_host.h            | 15 +++++++++++++++
 include/trace/events/kvm.h          | 11 +++++++----
 virt/kvm/Kconfig                    |  3 +++
 virt/kvm/kvm_main.c                 |  8 ++++++--
 17 files changed, 59 insertions(+), 6 deletions(-)

diff --git a/arch/arm/include/asm/kvm_host.h b/arch/arm/include/asm/kvm_host.h
index 3850701..1db618c 100644
--- a/arch/arm/include/asm/kvm_host.h
+++ b/arch/arm/include/asm/kvm_host.h
@@ -187,6 +187,7 @@ struct kvm_vm_stat {
 struct kvm_vcpu_stat {
 	u32 halt_successful_poll;
 	u32 halt_attempted_poll;
+	u32 halt_poll_no_tuning;
 	u32 halt_wakeup;
 	u32 hvc_exit_stat;
 	u64 wfe_exit_stat;
@@ -282,6 +283,7 @@ static inline void kvm_arch_hardware_unsetup(void) {}
 static inline void kvm_arch_sync_events(struct kvm *kvm) {}
 static inline void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) {}
 static inline void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) {}
+static inline void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu) {}
 
 static inline void kvm_arm_init_debug(void) {}
 static inline void kvm_arm_setup_debug(struct kvm_vcpu *vcpu) {}
diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
index f5c6bd2..9d79394 100644
--- a/arch/arm64/include/asm/kvm_host.h
+++ b/arch/arm64/include/asm/kvm_host.h
@@ -293,6 +293,7 @@ struct kvm_vm_stat {
 struct kvm_vcpu_stat {
 	u32 halt_successful_poll;
 	u32 halt_attempted_poll;
+	u32 halt_poll_no_tuning;
 	u32 halt_wakeup;
 	u32 hvc_exit_stat;
 	u64 wfe_exit_stat;
@@ -357,6 +358,7 @@ static inline void kvm_arch_hardware_unsetup(void) {}
 static inline void kvm_arch_sync_events(struct kvm *kvm) {}
 static inline void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) {}
 static inline void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) {}
+static inline void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu) {}
 
 void kvm_arm_init_debug(void);
 void kvm_arm_setup_debug(struct kvm_vcpu *vcpu);
diff --git a/arch/mips/include/asm/kvm_host.h b/arch/mips/include/asm/kvm_host.h
index f6b1279..7f28ed2 100644
--- a/arch/mips/include/asm/kvm_host.h
+++ b/arch/mips/include/asm/kvm_host.h
@@ -812,5 +812,6 @@ static inline void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) {}
 static inline void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) {}
 static inline void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) {}
 static inline void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) {}
+static inline void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu) {}
 
 #endif /* __MIPS_KVM_HOST_H__ */
diff --git a/arch/mips/kvm/mips.c b/arch/mips/kvm/mips.c
index 70ef1a4..bc50cbc 100644
--- a/arch/mips/kvm/mips.c
+++ b/arch/mips/kvm/mips.c
@@ -56,6 +56,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
 	{ "flush_dcache", VCPU_STAT(flush_dcache_exits), KVM_STAT_VCPU },
 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll), KVM_STAT_VCPU },
 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll), KVM_STAT_VCPU },
+	{ "halt_poll_no_tuning", VCPU_STAT(halt_poll_no_tuning), KVM_STAT_VCPU },
 	{ "halt_wakeup",  VCPU_STAT(halt_wakeup),	 KVM_STAT_VCPU },
 	{NULL}
 };
diff --git a/arch/powerpc/include/asm/kvm_host.h b/arch/powerpc/include/asm/kvm_host.h
index d7b3431..e234871 100644
--- a/arch/powerpc/include/asm/kvm_host.h
+++ b/arch/powerpc/include/asm/kvm_host.h
@@ -113,6 +113,7 @@ struct kvm_vcpu_stat {
 	u32 ext_intr_exits;
 	u32 halt_successful_poll;
 	u32 halt_attempted_poll;
+	u32 halt_poll_no_tuning;
 	u32 halt_wakeup;
 	u32 dbell_exits;
 	u32 gdbell_exits;
@@ -724,5 +725,6 @@ static inline void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) {}
 static inline void kvm_arch_exit(void) {}
 static inline void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) {}
 static inline void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) {}
+static inline void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu) {}
 
 #endif /* __POWERPC_KVM_HOST_H__ */
diff --git a/arch/powerpc/kvm/book3s.c b/arch/powerpc/kvm/book3s.c
index b34220d2..2166c1f 100644
--- a/arch/powerpc/kvm/book3s.c
+++ b/arch/powerpc/kvm/book3s.c
@@ -54,6 +54,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
 	{ "queue_intr",  VCPU_STAT(queue_intr) },
 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll), },
 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll), },
+	{ "halt_poll_no_tuning", VCPU_STAT(halt_poll_no_tuning) },
 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
 	{ "pf_storage",  VCPU_STAT(pf_storage) },
 	{ "sp_storage",  VCPU_STAT(sp_storage) },
diff --git a/arch/powerpc/kvm/booke.c b/arch/powerpc/kvm/booke.c
index 4d66f44..6d28d58 100644
--- a/arch/powerpc/kvm/booke.c
+++ b/arch/powerpc/kvm/booke.c
@@ -64,6 +64,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
 	{ "ext_intr",   VCPU_STAT(ext_intr_exits) },
 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
+	{ "halt_poll_no_tuning", VCPU_STAT(halt_poll_no_tuning) },
 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
 	{ "doorbell", VCPU_STAT(dbell_exits) },
 	{ "guest doorbell", VCPU_STAT(gdbell_exits) },
diff --git a/arch/s390/include/asm/kvm_host.h b/arch/s390/include/asm/kvm_host.h
index 6da41fa..f901a24 100644
--- a/arch/s390/include/asm/kvm_host.h
+++ b/arch/s390/include/asm/kvm_host.h
@@ -247,6 +247,7 @@ struct kvm_vcpu_stat {
 	u32 exit_instruction;
 	u32 halt_successful_poll;
 	u32 halt_attempted_poll;
+	u32 halt_poll_no_tuning;
 	u32 halt_wakeup;
 	u32 instruction_lctl;
 	u32 instruction_lctlg;
@@ -700,4 +701,6 @@ static inline void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 static inline void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) {}
 static inline void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) {}
 
+void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu);
+
 #endif
diff --git a/arch/s390/kvm/Kconfig b/arch/s390/kvm/Kconfig
index 5ea5af3..ccfe6f6 100644
--- a/arch/s390/kvm/Kconfig
+++ b/arch/s390/kvm/Kconfig
@@ -28,6 +28,7 @@ config KVM
 	select HAVE_KVM_IRQCHIP
 	select HAVE_KVM_IRQFD
 	select HAVE_KVM_IRQ_ROUTING
+	select HAVE_KVM_INVALID_POLLS
 	select SRCU
 	select KVM_VFIO
 	---help---
diff --git a/arch/s390/kvm/interrupt.c b/arch/s390/kvm/interrupt.c
index e550404..5a80af7 100644
--- a/arch/s390/kvm/interrupt.c
+++ b/arch/s390/kvm/interrupt.c
@@ -977,6 +977,11 @@ no_timer:
 
 void kvm_s390_vcpu_wakeup(struct kvm_vcpu *vcpu)
 {
+	/*
+	 * We cannot move this into the if, as the CPU might be already
+	 * in kvm_vcpu_block without having the waitqueue set (polling)
+	 */
+	vcpu->valid_wakeup = true;
 	if (swait_active(&vcpu->wq)) {
 		/*
 		 * The vcpu gave up the cpu voluntarily, mark it as a good
diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
index 668c087..ce35bf0 100644
--- a/arch/s390/kvm/kvm-s390.c
+++ b/arch/s390/kvm/kvm-s390.c
@@ -65,6 +65,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
 	{ "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program) },
 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
+	{ "halt_poll_no_tuning", VCPU_STAT(halt_poll_no_tuning) },
 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
 	{ "instruction_lctlg", VCPU_STAT(instruction_lctlg) },
 	{ "instruction_lctl", VCPU_STAT(instruction_lctl) },
@@ -2971,6 +2972,11 @@ void kvm_arch_commit_memory_region(struct kvm *kvm,
 	return;
 }
 
+void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu)
+{
+	vcpu->valid_wakeup = false;
+}
+
 static int __init kvm_s390_init(void)
 {
 	if (!sclp.has_sief2) {
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index b7e3944..c8cfa2d 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -804,6 +804,7 @@ struct kvm_vcpu_stat {
 	u32 halt_exits;
 	u32 halt_successful_poll;
 	u32 halt_attempted_poll;
+	u32 halt_poll_no_tuning;
 	u32 halt_wakeup;
 	u32 request_irq_exits;
 	u32 irq_exits;
@@ -1343,5 +1344,6 @@ void kvm_set_msi_irq(struct kvm_kernel_irq_routing_entry *e,
 
 static inline void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) {}
 static inline void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) {}
+static inline void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu) {}
 
 #endif /* _ASM_X86_KVM_HOST_H */
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 9b7798c..809c40a 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -161,6 +161,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
 	{ "halt_exits", VCPU_STAT(halt_exits) },
 	{ "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
 	{ "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
+	{ "halt_poll_no_tuning", VCPU_STAT(halt_poll_no_tuning) },
 	{ "halt_wakeup", VCPU_STAT(halt_wakeup) },
 	{ "hypercalls", VCPU_STAT(hypercalls) },
 	{ "request_irq", VCPU_STAT(request_irq_exits) },
diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
index 5276fe0..0fac5d8 100644
--- a/include/linux/kvm_host.h
+++ b/include/linux/kvm_host.h
@@ -225,6 +225,7 @@ struct kvm_vcpu {
 	sigset_t sigset;
 	struct kvm_vcpu_stat stat;
 	unsigned int halt_poll_ns;
+	bool valid_wakeup;
 
 #ifdef CONFIG_HAS_IOMEM
 	int mmio_needed;
@@ -1179,4 +1180,18 @@ int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
 				  uint32_t guest_irq, bool set);
 #endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */
 
+#ifdef CONFIG_HAVE_KVM_INVALID_POLLS
+/* If we wakeup during the poll time, was it a sucessful poll? */
+static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
+{
+	return vcpu->valid_wakeup;
+}
+
+#else
+static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
+{
+	return true;
+}
+#endif /* CONFIG_HAVE_KVM_INVALID_POLLS */
+
 #endif
diff --git a/include/trace/events/kvm.h b/include/trace/events/kvm.h
index aa69253..92e6fd6 100644
--- a/include/trace/events/kvm.h
+++ b/include/trace/events/kvm.h
@@ -38,22 +38,25 @@ TRACE_EVENT(kvm_userspace_exit,
 );
 
 TRACE_EVENT(kvm_vcpu_wakeup,
-	    TP_PROTO(__u64 ns, bool waited),
-	    TP_ARGS(ns, waited),
+	    TP_PROTO(__u64 ns, bool waited, bool tuned),
+	    TP_ARGS(ns, waited, tuned),
 
 	TP_STRUCT__entry(
 		__field(	__u64,		ns		)
 		__field(	bool,		waited		)
+		__field(	bool,		tuned		)
 	),
 
 	TP_fast_assign(
 		__entry->ns		= ns;
 		__entry->waited		= waited;
+		__entry->tuned		= tuned;
 	),
 
-	TP_printk("%s time %lld ns",
+	TP_printk("%s time %lld ns, polling %s",
 		  __entry->waited ? "wait" : "poll",
-		  __entry->ns)
+		  __entry->ns,
+		  __entry->tuned ? "changed" : "unchanged")
 );
 
 #if defined(CONFIG_HAVE_KVM_IRQFD)
diff --git a/virt/kvm/Kconfig b/virt/kvm/Kconfig
index 7a79b68..4451842 100644
--- a/virt/kvm/Kconfig
+++ b/virt/kvm/Kconfig
@@ -41,6 +41,9 @@ config KVM_VFIO
 config HAVE_KVM_ARCH_TLB_FLUSH_ALL
        bool
 
+config HAVE_KVM_INVALID_POLLS
+       bool
+
 config KVM_GENERIC_DIRTYLOG_READ_PROTECT
        bool
 
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 4fd482f..096c655 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -2028,6 +2028,8 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 			 */
 			if (kvm_vcpu_check_block(vcpu) < 0) {
 				++vcpu->stat.halt_successful_poll;
+				if (!vcpu_valid_wakeup(vcpu))
+					++vcpu->stat.halt_poll_no_tuning;
 				goto out;
 			}
 			cur = ktime_get();
@@ -2057,7 +2059,8 @@ out:
 		if (block_ns <= vcpu->halt_poll_ns)
 			;
 		/* we had a long block, shrink polling */
-		else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
+		else if (!vcpu_valid_wakeup(vcpu) ||
+			(vcpu->halt_poll_ns && block_ns > halt_poll_ns))
 			shrink_halt_poll_ns(vcpu);
 		/* we had a short halt and our poll time is too small */
 		else if (vcpu->halt_poll_ns < halt_poll_ns &&
@@ -2066,7 +2069,8 @@ out:
 	} else
 		vcpu->halt_poll_ns = 0;
 
-	trace_kvm_vcpu_wakeup(block_ns, waited);
+	trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
+	kvm_arch_vcpu_block_finish(vcpu);
 }
 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
 
-- 
2.3.0

Re: [PATCH 1/1] KVM: halt_polling: provide a way to qualify wakeups during poll

[David Hildenbrand]

> Some wakeups should not be considered a sucessful poll. For example on
> s390 I/O interrupts are usually floating, which means that _ALL_ CPUs
> would be considered runnable - letting all vCPUs poll all the time for
> transactional like workload, even if one vCPU would be enough.
> This can result in huge CPU usage for large guests.
> This patch lets architectures provide a way to qualify wakeups if they
> should be considered a good/bad wakeups in regard to polls.
> 
> For s390 the implementation will fence of halt polling for anything but
> known good, single vCPU events. The s390 implementation for floating
> interrupts does a wakeup for one vCPU, but the interrupt will be delivered
> by whatever CPU checks first for a pending interrupt. We prefer the
> woken up CPU by marking the poll of this CPU as "good" poll.
> This code will also mark several other wakeup reasons like IPI or
> expired timers as "good". This will of course also mark some events as
> not sucessful. As  KVM on z runs always as a 2nd level hypervisor,
> we prefer to not poll, unless we are really sure, though.
> 
> This patch successfully limits the CPU usage for cases like uperf 1byte
> transactional ping pong workload or wakeup heavy workload like OLTP
> while still providing a proper speedup.
> 
> This also introduced a new vcpu stat "halt_poll_no_tuning" that marks
> wakeups that are considered not good for polling.
> 
> Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
> Cc: David Matlack <dmatlack@google.com>
> Cc: Wanpeng Li <kernellwp@gmail.com>
> ---

You can keep my
Acked-by: David Hildenbrand <dahi@linux.vnet.ibm.com>


David

Re: [PATCH 1/1] KVM: halt_polling: provide a way to qualify wakeups during poll

[Cornelia Huck]

On Tue,  3 May 2016 14:37:21 +0200
Christian Borntraeger <borntraeger@de.ibm.com> wrote:

> diff --git a/include/trace/events/kvm.h b/include/trace/events/kvm.h
> index aa69253..92e6fd6 100644
> --- a/include/trace/events/kvm.h
> +++ b/include/trace/events/kvm.h
> @@ -38,22 +38,25 @@ TRACE_EVENT(kvm_userspace_exit,
>  );
> 
>  TRACE_EVENT(kvm_vcpu_wakeup,
> -	    TP_PROTO(__u64 ns, bool waited),
> -	    TP_ARGS(ns, waited),
> +	    TP_PROTO(__u64 ns, bool waited, bool tuned),
> +	    TP_ARGS(ns, waited, tuned),
> 
>  	TP_STRUCT__entry(
>  		__field(	__u64,		ns		)
>  		__field(	bool,		waited		)
> +		__field(	bool,		tuned		)
>  	),
> 
>  	TP_fast_assign(
>  		__entry->ns		= ns;
>  		__entry->waited		= waited;
> +		__entry->tuned		= tuned;
>  	),
> 
> -	TP_printk("%s time %lld ns",
> +	TP_printk("%s time %lld ns, polling %s",
>  		  __entry->waited ? "wait" : "poll",
> -		  __entry->ns)
> +		  __entry->ns,
> +		  __entry->tuned ? "changed" : "unchanged")

I think "changed"/"unchanged" is a bit misleading here, as we do adjust
the intervall if we had an invalid poll... but it's hard to find a
suitable text here.

Just print "poll interval tuned" if we were (a) polling to begin with,
(b) the poll was valid and (c) the interval was actually changed and
print "invalid poll" if that's what happened? Or is that overkill?

>  );
> 
>  #if defined(CONFIG_HAVE_KVM_IRQFD)

Otherwise, looks good to me.

Re: [PATCH 1/1] KVM: halt_polling: provide a way to qualify wakeups during poll

[Radim Krčmář]

2016-05-03 14:56+0200, Cornelia Huck:
> On Tue,  3 May 2016 14:37:21 +0200
> Christian Borntraeger <borntraeger@de.ibm.com> wrote:
>> diff --git a/include/trace/events/kvm.h b/include/trace/events/kvm.h
>> index aa69253..92e6fd6 100644
>> --- a/include/trace/events/kvm.h
>> +++ b/include/trace/events/kvm.h
>> @@ -38,22 +38,25 @@ TRACE_EVENT(kvm_userspace_exit,
>>  );
>> 
>>  TRACE_EVENT(kvm_vcpu_wakeup,
>> -	    TP_PROTO(__u64 ns, bool waited),
>> -	    TP_ARGS(ns, waited),
>> +	    TP_PROTO(__u64 ns, bool waited, bool tuned),
>> +	    TP_ARGS(ns, waited, tuned),
>> 
>>  	TP_STRUCT__entry(
>>  		__field(	__u64,		ns		)
>>  		__field(	bool,		waited		)
>> +		__field(	bool,		tuned		)
>>  	),
>> 
>>  	TP_fast_assign(
>>  		__entry->ns		= ns;
>>  		__entry->waited		= waited;
>> +		__entry->tuned		= tuned;
>>  	),
>> 
>> -	TP_printk("%s time %lld ns",
>> +	TP_printk("%s time %lld ns, polling %s",
>>  		  __entry->waited ? "wait" : "poll",
>> -		  __entry->ns)
>> +		  __entry->ns,
>> +		  __entry->tuned ? "changed" : "unchanged")
> 
> I think "changed"/"unchanged" is a bit misleading here, as we do adjust
> the intervall if we had an invalid poll... but it's hard to find a
> suitable text here.
> 
> Just print "poll interval tuned" if we were (a) polling to begin with,
> (b) the poll was valid and (c) the interval was actually changed and
> print "invalid poll" if that's what happened? Or is that overkill?

Just renaming to valid/invalid is fine, IMO, the state of polling is
static and interval change can be read from other traces.

I think that having "no_tuning" counter, "unchanged" trace and "invalid"
in source names obscures the logical connection;  doesn't "invalid" fit
them all?

Re: [PATCH 1/1] KVM: halt_polling: provide a way to qualify wakeups during poll

[Radim Krčmář]

2016-05-03 14:37+0200, Christian Borntraeger:
> Some wakeups should not be considered a sucessful poll. For example on
> s390 I/O interrupts are usually floating, which means that _ALL_ CPUs
> would be considered runnable - letting all vCPUs poll all the time for
> transactional like workload, even if one vCPU would be enough.
> This can result in huge CPU usage for large guests.
> This patch lets architectures provide a way to qualify wakeups if they
> should be considered a good/bad wakeups in regard to polls.
> 
> For s390 the implementation will fence of halt polling for anything but
> known good, single vCPU events. The s390 implementation for floating
> interrupts does a wakeup for one vCPU, but the interrupt will be delivered
> by whatever CPU checks first for a pending interrupt. We prefer the
> woken up CPU by marking the poll of this CPU as "good" poll.
> This code will also mark several other wakeup reasons like IPI or
> expired timers as "good". This will of course also mark some events as
> not sucessful. As  KVM on z runs always as a 2nd level hypervisor,
> we prefer to not poll, unless we are really sure, though.
> 
> This patch successfully limits the CPU usage for cases like uperf 1byte
> transactional ping pong workload or wakeup heavy workload like OLTP
> while still providing a proper speedup.
> 
> This also introduced a new vcpu stat "halt_poll_no_tuning" that marks
> wakeups that are considered not good for polling.
> 
> Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
> Cc: David Matlack <dmatlack@google.com>
> Cc: Wanpeng Li <kernellwp@gmail.com>
> ---

Thanks for all explanations,

Acked-by: Radim Krčmář <rkrcmar@redhat.com>

Re: [PATCH 1/1] KVM: halt_polling: provide a way to qualify wakeups during poll

[Christian Borntraeger]

On 05/03/2016 05:03 PM, Radim Krčmář wrote:
> 2016-05-03 14:56+0200, Cornelia Huck:
>> On Tue,  3 May 2016 14:37:21 +0200
>> Christian Borntraeger <borntraeger@de.ibm.com> wrote:
>>> diff --git a/include/trace/events/kvm.h b/include/trace/events/kvm.h
>>> index aa69253..92e6fd6 100644
>>> --- a/include/trace/events/kvm.h
>>> +++ b/include/trace/events/kvm.h
>>> @@ -38,22 +38,25 @@ TRACE_EVENT(kvm_userspace_exit,
>>>  );
>>>
>>>  TRACE_EVENT(kvm_vcpu_wakeup,
>>> -	    TP_PROTO(__u64 ns, bool waited),
>>> -	    TP_ARGS(ns, waited),
>>> +	    TP_PROTO(__u64 ns, bool waited, bool tuned),
>>> +	    TP_ARGS(ns, waited, tuned),
>>>
>>>  	TP_STRUCT__entry(
>>>  		__field(	__u64,		ns		)
>>>  		__field(	bool,		waited		)
>>> +		__field(	bool,		tuned		)
>>>  	),
>>>
>>>  	TP_fast_assign(
>>>  		__entry->ns		= ns;
>>>  		__entry->waited		= waited;
>>> +		__entry->tuned		= tuned;
>>>  	),
>>>
>>> -	TP_printk("%s time %lld ns",
>>> +	TP_printk("%s time %lld ns, polling %s",
>>>  		  __entry->waited ? "wait" : "poll",
>>> -		  __entry->ns)
>>> +		  __entry->ns,
>>> +		  __entry->tuned ? "changed" : "unchanged")
>>
>> I think "changed"/"unchanged" is a bit misleading here, as we do adjust
>> the intervall if we had an invalid poll... but it's hard to find a
>> suitable text here.
>>
>> Just print "poll interval tuned" if we were (a) polling to begin with,
>> (b) the poll was valid and (c) the interval was actually changed and
>> print "invalid poll" if that's what happened? Or is that overkill?
> 
> Just renaming to valid/invalid is fine, IMO, the state of polling is
> static and interval change can be read from other traces.
> 
> I think that having "no_tuning" counter, "unchanged" trace and "invalid"
> in source names obscures the logical connection;  doesn't "invalid" fit
> them all?
> 

Yes, will change tracing into 
__entry->valid ? "valid" : "invalid")
and halt_poll_no_tuning --> halt_poll_invalid

That seems to be in line with the remaining parts of the patch.

Christian

Re: [PATCH 1/1] KVM: halt_polling: provide a way to qualify wakeups during poll

[Cornelia Huck]

On Tue, 3 May 2016 20:12:00 +0200
Christian Borntraeger <borntraeger@de.ibm.com> wrote:

> On 05/03/2016 05:03 PM, Radim Krčmář wrote:
> > 2016-05-03 14:56+0200, Cornelia Huck:
> >> On Tue,  3 May 2016 14:37:21 +0200
> >> Christian Borntraeger <borntraeger@de.ibm.com> wrote:
> >>> diff --git a/include/trace/events/kvm.h b/include/trace/events/kvm.h
> >>> index aa69253..92e6fd6 100644
> >>> --- a/include/trace/events/kvm.h
> >>> +++ b/include/trace/events/kvm.h
> >>> @@ -38,22 +38,25 @@ TRACE_EVENT(kvm_userspace_exit,
> >>>  );
> >>>
> >>>  TRACE_EVENT(kvm_vcpu_wakeup,
> >>> -	    TP_PROTO(__u64 ns, bool waited),
> >>> -	    TP_ARGS(ns, waited),
> >>> +	    TP_PROTO(__u64 ns, bool waited, bool tuned),
> >>> +	    TP_ARGS(ns, waited, tuned),
> >>>
> >>>  	TP_STRUCT__entry(
> >>>  		__field(	__u64,		ns		)
> >>>  		__field(	bool,		waited		)
> >>> +		__field(	bool,		tuned		)
> >>>  	),
> >>>
> >>>  	TP_fast_assign(
> >>>  		__entry->ns		= ns;
> >>>  		__entry->waited		= waited;
> >>> +		__entry->tuned		= tuned;
> >>>  	),
> >>>
> >>> -	TP_printk("%s time %lld ns",
> >>> +	TP_printk("%s time %lld ns, polling %s",
> >>>  		  __entry->waited ? "wait" : "poll",
> >>> -		  __entry->ns)
> >>> +		  __entry->ns,
> >>> +		  __entry->tuned ? "changed" : "unchanged")
> >>
> >> I think "changed"/"unchanged" is a bit misleading here, as we do adjust
> >> the intervall if we had an invalid poll... but it's hard to find a
> >> suitable text here.
> >>
> >> Just print "poll interval tuned" if we were (a) polling to begin with,
> >> (b) the poll was valid and (c) the interval was actually changed and
> >> print "invalid poll" if that's what happened? Or is that overkill?
> > 
> > Just renaming to valid/invalid is fine, IMO, the state of polling is
> > static and interval change can be read from other traces.
> > 
> > I think that having "no_tuning" counter, "unchanged" trace and "invalid"
> > in source names obscures the logical connection;  doesn't "invalid" fit
> > them all?
> > 
> 
> Yes, will change tracing into 
> __entry->valid ? "valid" : "invalid")
> and halt_poll_no_tuning --> halt_poll_invalid
> 
> That seems to be in line with the remaining parts of the patch.

It seems we agree on the colour of the bikeshed :)

Re: [PATCH 1/1] KVM: halt_polling: provide a way to qualify wakeups during poll

[Christian Borntraeger]

On 05/03/2016 05:09 PM, Radim Krčmář wrote:
> 2016-05-03 14:37+0200, Christian Borntraeger:
>> Some wakeups should not be considered a sucessful poll. For example on
>> s390 I/O interrupts are usually floating, which means that _ALL_ CPUs
>> would be considered runnable - letting all vCPUs poll all the time for
>> transactional like workload, even if one vCPU would be enough.
>> This can result in huge CPU usage for large guests.
>> This patch lets architectures provide a way to qualify wakeups if they
>> should be considered a good/bad wakeups in regard to polls.
>>
>> For s390 the implementation will fence of halt polling for anything but
>> known good, single vCPU events. The s390 implementation for floating
>> interrupts does a wakeup for one vCPU, but the interrupt will be delivered
>> by whatever CPU checks first for a pending interrupt. We prefer the
>> woken up CPU by marking the poll of this CPU as "good" poll.
>> This code will also mark several other wakeup reasons like IPI or
>> expired timers as "good". This will of course also mark some events as
>> not sucessful. As  KVM on z runs always as a 2nd level hypervisor,
>> we prefer to not poll, unless we are really sure, though.
>>
>> This patch successfully limits the CPU usage for cases like uperf 1byte
>> transactional ping pong workload or wakeup heavy workload like OLTP
>> while still providing a proper speedup.
>>
>> This also introduced a new vcpu stat "halt_poll_no_tuning" that marks
>> wakeups that are considered not good for polling.
>>
>> Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
>> Cc: David Matlack <dmatlack@google.com>
>> Cc: Wanpeng Li <kernellwp@gmail.com>
>> ---
> 
> Thanks for all explanations,
> 
> Acked-by: Radim Krčmář <rkrcmar@redhat.com>
> 


The feedback about the logic triggered some more experiments on my side.
So I was experimenting with some different workloads/heuristics and it
seems that even more aggressive shrinking (basically resetting to 0 as soon
as an invalid poll comes along) does improve the cpu usage even more.

patch on top
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index ffe0545..c168662 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -2036,12 +2036,13 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 out:
        block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
 
-       if (halt_poll_ns) {
+       if (!vcpu_valid_wakeup(vcpu))
+                shrink_halt_poll_ns(vcpu);
+       else if (halt_poll_ns) {
                if (block_ns <= vcpu->halt_poll_ns)
                        ;
                /* we had a long block, shrink polling */
-               else if (!vcpu_valid_wakeup(vcpu) ||
-                       (vcpu->halt_poll_ns && block_ns > halt_poll_ns))
+               else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
                        shrink_halt_poll_ns(vcpu);
                /* we had a short halt and our poll time is too small */
                else if (vcpu->halt_poll_ns < halt_poll_ns &&


the uperf 1byte:1byte workload seems to have all the benefits still.
I have asked the performance folks to test several other workloads if
we loose some of the benefits.
So I will defer this patch until I have a full picture which heuristics
is best. Hopefully I have some answers next week. 

(So the new diff looks like)
@@ -2034,7 +2036,9 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 out:
        block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
 
-       if (halt_poll_ns) {
+       if (!vcpu_valid_wakeup(vcpu))
+                shrink_halt_poll_ns(vcpu);
+       else if (halt_poll_ns) {
                if (block_ns <= vcpu->halt_poll_ns)
                        ;

Re: [PATCH 1/1] KVM: halt_polling: provide a way to qualify wakeups during poll

[Cornelia Huck]

On Wed, 4 May 2016 09:50:57 +0200
Christian Borntraeger <borntraeger@de.ibm.com> wrote:

> The feedback about the logic triggered some more experiments on my side.
> So I was experimenting with some different workloads/heuristics and it
> seems that even more aggressive shrinking (basically resetting to 0 as soon
> as an invalid poll comes along) does improve the cpu usage even more.

Do we still keep the shrink instead of resetting to 0 explicitly? (In
case the default shrink factor was set to != 0.) We'd lose a tuneable,
but it seems the aggressiveness is warranted.

> (So the new diff looks like)
> @@ -2034,7 +2036,9 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
>  out:
>         block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
> 
> -       if (halt_poll_ns) {
> +       if (!vcpu_valid_wakeup(vcpu))
> +                shrink_halt_poll_ns(vcpu);
> +       else if (halt_poll_ns) {
>                 if (block_ns <= vcpu->halt_poll_ns)
>                         ;

...making this

if (halt_poll_ns && vcpu_valid_wakeup(vcpu)) {

Re: [PATCH 1/1] KVM: halt_polling: provide a way to qualify wakeups during poll

[Christian Borntraeger]

On 05/04/2016 10:05 AM, Cornelia Huck wrote:
> On Wed, 4 May 2016 09:50:57 +0200
> Christian Borntraeger <borntraeger@de.ibm.com> wrote:
> 
>> The feedback about the logic triggered some more experiments on my side.
>> So I was experimenting with some different workloads/heuristics and it
>> seems that even more aggressive shrinking (basically resetting to 0 as soon
>> as an invalid poll comes along) does improve the cpu usage even more.
> 
> Do we still keep the shrink instead of resetting to 0 explicitly? (In
> case the default shrink factor was set to != 0.) We'd lose a tuneable,
> but it seems the aggressiveness is warranted.
> 
>> (So the new diff looks like)
>> @@ -2034,7 +2036,9 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
>>  out:
>>         block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
>>
>> -       if (halt_poll_ns) {
>> +       if (!vcpu_valid_wakeup(vcpu))
>> +                shrink_halt_poll_ns(vcpu);
>> +       else if (halt_poll_ns) {
>>                 if (block_ns <= vcpu->halt_poll_ns)
>>                         ;
> 
> ...making this
> 
> if (halt_poll_ns && vcpu_valid_wakeup(vcpu)) {
> 

I decided to keep my version as it should be functionally equivalent for shrink==0
but it allows the performance folks to do future testing for better heuristics.

David, Wanpeng,
I just send out the patch set but I forgot to Cc you :-/




Christian