Skip to contentSkip to conversation

KVM: s390: diag9c (directed yield) forwarding

PATCH In review Pierre Morel 3 participants 5 messages Feb 22, 2021 – Feb 23, 2021

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

v4 cover KVM: s390: diag9c (directed yield) forwarding

This patch will forward the yield_to hypercall (diag9c) if in the host the target CPU is also not running. As we do not yet have performance data (and recommendations) the default is turned off, but this can be changed during runtime.

The documentation notation for the titles is kept coherent with the other DIAGNOSE entries using the zVM notation DIAGNOSE 'X'9C, While in the explanation part we use the standard Linux notation DIAGNOSE 0x9c.

Regards,
Pierre

Pierre Morel (1):
KVM: s390: diag9c (directed yield) forwarding

 Documentation/virt/kvm/s390-diag.rst | 33 ++++++++++++++++++++++++++++
 arch/s390/include/asm/kvm_host.h     |  1 +
 arch/s390/include/asm/smp.h          |  1 +
 arch/s390/kernel/smp.c               |  1 +
 arch/s390/kvm/diag.c                 | 31 +++++++++++++++++++++++---
 arch/s390/kvm/kvm-s390.c             |  6 +++++
 arch/s390/kvm/kvm-s390.h             |  8 +++++++
 7 files changed, 78 insertions(+), 3 deletions(-)
Signature · 30 lines
2.17.1

Changelog:

From v3:

- rewording of the commit message
  (Connie)

- great rewording of the documentation
  (Connie)

- Keep smp_yield_cpu symbol GPL
  (Connie)

- More explicit diag9c_forwarding_hz parameter description
  (Connie)

From v2:

- update overrun comparison, avoid comparison on unsigned
  (Pierre)

From v1:

- more precise comments
  (Connie)

- Documentation
  (Janosch)
#1 lore raw

PATCH v4 1/1 KVM: s390: diag9c (directed yield) forwarding

When we intercept a DIAG_9C from the guest we verify that the target real CPU associated with the virtual CPU designated by the guest is running and if not we forward the DIAG_9C to the target real CPU.

To avoid a diag9c storm we allow a maximal rate of diag9c forwarding.

The rate is calculated as a count per second defined as a new parameter of the s390 kvm module: diag9c_forwarding_hz .

The default value of 0 is to not forward diag9c.

---
 Documentation/virt/kvm/s390-diag.rst | 33 ++++++++++++++++++++++++++++
 arch/s390/include/asm/kvm_host.h     |  1 +
 arch/s390/include/asm/smp.h          |  1 +
 arch/s390/kernel/smp.c               |  1 +
 arch/s390/kvm/diag.c                 | 31 +++++++++++++++++++++++---
 arch/s390/kvm/kvm-s390.c             |  6 +++++
 arch/s390/kvm/kvm-s390.h             |  8 +++++++
 7 files changed, 78 insertions(+), 3 deletions(-)
Documentation/virt/kvm/s390-diag.rst +33 -0
index eaac4864d3d6..ca85f030eb0b 100644
@@ -84,3 +84,36 @@ If the function code specifies 0x501, breakpoint functions may be performed.
 This function code is handled by userspace.
 
 This diagnose function code has no subfunctions and uses no parameters.
+
+
+DIAGNOSE function code 'X'9C - Voluntary Time Slice Yield
+---------------------------------------------------------
+
+General register 1 contains the target CPU address.
+
+In a guest of a hypervisor like LPAR, KVM or z/VM using shared host CPUs,
+DIAGNOSE with function code 0x9c may improve system performance by
+yielding the host CPU on which the guest CPU is running to be assigned
+to another guest CPU, preferably the logical CPU containing the specified
+target CPU.
+
+
+DIAG 'X'9C forwarding
++++++++++++++++++++++
+
+The guest may send a DIAGNOSE 0x9c in order to yield to a certain
+other vcpu. An example is a Linux guest that tries to yield to the vcpu
+that is currently holding a spinlock, but not running.
+
+However, on the host the real cpu backing the vcpu may itself not be
+running.
+Forwarding the DIAGNOSE 0x9c initially sent by the guest to yield to
+the backing cpu will hopefully cause that cpu, and thus subsequently
+the guest's vcpu, to be scheduled.
+
+
+diag9c_forwarding_hz
+    KVM kernel parameter allowing to specify the maximum number of DIAGNOSE
+    0x9c forwarding per second in the purpose of avoiding a DIAGNOSE 0x9c
+    forwarding storm.
+    A value of 0 turns the forwarding off.
arch/s390/include/asm/kvm_host.h +1 -0
index 527776a1f076..cb19508c22fb 100644
@@ -456,6 +456,7 @@ struct kvm_vcpu_stat {
 	u64 diagnose_44;
 	u64 diagnose_9c;
 	u64 diagnose_9c_ignored;
+	u64 diagnose_9c_forward;
 	u64 diagnose_258;
 	u64 diagnose_308;
 	u64 diagnose_500;
arch/s390/include/asm/smp.h +1 -0
index 01e360004481..e317fd4866c1 100644
@@ -63,5 +63,6 @@ extern void __noreturn cpu_die(void);
 extern void __cpu_die(unsigned int cpu);
 extern int __cpu_disable(void);
 extern void schedule_mcck_handler(void);
+void notrace smp_yield_cpu(int cpu);
 
 #endif /* __ASM_SMP_H */
arch/s390/kernel/smp.c +1 -0
index 27c763014114..cecdecd33a58 100644
@@ -422,6 +422,7 @@ void notrace smp_yield_cpu(int cpu)
 	asm volatile("diag %0,0,0x9c"
 		     : : "d" (pcpu_devices[cpu].address));
 }
+EXPORT_SYMBOL_GPL(smp_yield_cpu);
 
 /*
  * Send cpus emergency shutdown signal. This gives the cpus the
arch/s390/kvm/diag.c +28 -3
index 5b8ec1c447e1..02c146f9e5cd 100644
@@ -150,6 +150,19 @@ static int __diag_time_slice_end(struct kvm_vcpu *vcpu)
 	return 0;
 }
 
+static int forward_cnt;
+static unsigned long cur_slice;
+
+static int diag9c_forwarding_overrun(void)
+{
+	/* Reset the count on a new slice */
+	if (time_after(jiffies, cur_slice)) {
+		cur_slice = jiffies;
+		forward_cnt = diag9c_forwarding_hz / HZ;
+	}
+	return forward_cnt-- <= 0 ? 1 : 0;
+}
+
 static int __diag_time_slice_end_directed(struct kvm_vcpu *vcpu)
 {
 	struct kvm_vcpu *tcpu;
@@ -167,9 +180,21 @@ static int __diag_time_slice_end_directed(struct kvm_vcpu *vcpu)
 	if (!tcpu)
 		goto no_yield;
 
-	/* target already running */
-	if (READ_ONCE(tcpu->cpu) >= 0)
-		goto no_yield;
+	/* target guest VCPU already running */
+	if (READ_ONCE(tcpu->cpu) >= 0) {
+		if (!diag9c_forwarding_hz || diag9c_forwarding_overrun())
+			goto no_yield;
+
+		/* target host CPU already running */
+		if (!vcpu_is_preempted(tcpu->cpu))
+			goto no_yield;
+		smp_yield_cpu(tcpu->cpu);
+		VCPU_EVENT(vcpu, 5,
+			   "diag time slice end directed to %d: yield forwarded",
+			   tid);
+		vcpu->stat.diagnose_9c_forward++;
+		return 0;
+	}
 
 	if (kvm_vcpu_yield_to(tcpu) <= 0)
 		goto no_yield;
arch/s390/kvm/kvm-s390.c +6 -0
index 053ef36784e4..7b113d14ed7c 100644
@@ -157,6 +157,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
 	VCPU_STAT("instruction_diag_44", diagnose_44),
 	VCPU_STAT("instruction_diag_9c", diagnose_9c),
 	VCPU_STAT("diag_9c_ignored", diagnose_9c_ignored),
+	VCPU_STAT("diag_9c_forward", diagnose_9c_forward),
 	VCPU_STAT("instruction_diag_258", diagnose_258),
 	VCPU_STAT("instruction_diag_308", diagnose_308),
 	VCPU_STAT("instruction_diag_500", diagnose_500),
@@ -190,6 +191,11 @@ static bool use_gisa  = true;
 module_param(use_gisa, bool, 0644);
 MODULE_PARM_DESC(use_gisa, "Use the GISA if the host supports it.");
 
+/* maximum diag9c forwarding per second */
+unsigned int diag9c_forwarding_hz;
+module_param(diag9c_forwarding_hz, uint, 0644);
+MODULE_PARM_DESC(diag9c_forwarding_hz, "Maximum diag9c forwarding per second, 0 to turn off");
+
 /*
  * For now we handle at most 16 double words as this is what the s390 base
  * kernel handles and stores in the prefix page. If we ever need to go beyond
arch/s390/kvm/kvm-s390.h +8 -0
index 79dcd647b378..9fad25109b0d 100644
@@ -471,4 +471,12 @@ void kvm_s390_reinject_machine_check(struct kvm_vcpu *vcpu,
  * @kvm: the KVM guest
  */
 void kvm_s390_vcpu_crypto_reset_all(struct kvm *kvm);
+
+/**
+ * diag9c_forwarding_hz
+ *
+ * Set the maximum number of diag9c forwarding per second
+ */
+extern unsigned int diag9c_forwarding_hz;
+
 #endif
Signature · 1 line
2.17.1
Reviewed-by on v4 1/1
#2 lore raw
Quoted 13 lines · On Mon, 22 Feb 2021 13:41:01 +0100 Pierre Morel <[email protected]> wrote:

When we intercept a DIAG_9C from the guest we verify that the target real CPU associated with the virtual CPU designated by the guest is running and if not we forward the DIAG_9C to the target real CPU.

To avoid a diag9c storm we allow a maximal rate of diag9c forwarding.

The rate is calculated as a count per second defined as a new parameter of the s390 kvm module: diag9c_forwarding_hz .

The default value of 0 is to not forward diag9c.

Quoted 9 lines
---
 Documentation/virt/kvm/s390-diag.rst | 33 ++++++++++++++++++++++++++++
 arch/s390/include/asm/kvm_host.h     |  1 +
 arch/s390/include/asm/smp.h          |  1 +
 arch/s390/kernel/smp.c               |  1 +
 arch/s390/kvm/diag.c                 | 31 +++++++++++++++++++++++---
 arch/s390/kvm/kvm-s390.c             |  6 +++++
 arch/s390/kvm/kvm-s390.h             |  8 +++++++
 7 files changed, 78 insertions(+), 3 deletions(-)
#3 lore raw
Quoted 26 lines · Cornelia Huck
Quoted 13 lines · On Mon, 22 Feb 2021 13:41:01 +0100 Pierre Morel <[email protected]> wrote:

When we intercept a DIAG_9C from the guest we verify that the target real CPU associated with the virtual CPU designated by the guest is running and if not we forward the DIAG_9C to the target real CPU.

To avoid a diag9c storm we allow a maximal rate of diag9c forwarding.

The rate is calculated as a count per second defined as a new parameter of the s390 kvm module: diag9c_forwarding_hz .

The default value of 0 is to not forward diag9c.

Quoted 9 lines
---
  Documentation/virt/kvm/s390-diag.rst | 33 ++++++++++++++++++++++++++++
  arch/s390/include/asm/kvm_host.h     |  1 +
  arch/s390/include/asm/smp.h          |  1 +
  arch/s390/kernel/smp.c               |  1 +
  arch/s390/kvm/diag.c                 | 31 +++++++++++++++++++++++---
  arch/s390/kvm/kvm-s390.c             |  6 +++++
  arch/s390/kvm/kvm-s390.h             |  8 +++++++
  7 files changed, 78 insertions(+), 3 deletions(-)
Quoted 2 lines

Thanks,
Pierre

Signature · 2 lines
Pierre Morel
IBM Lab Boeblingen
#4 lore raw
In reply to Pierre Morel
Quoted 5 lines · Pierre Morel

When we intercept a DIAG_9C from the guest we verify that the target real CPU associated with the virtual CPU designated by the guest is running and if not we forward the DIAG_9C to the target real CPU.

Quoted 8 lines

To avoid a diag9c storm we allow a maximal rate of diag9c forwarding.

The rate is calculated as a count per second defined as a new parameter of the s390 kvm module: diag9c_forwarding_hz .

The default value of 0 is to not forward diag9c.

I will carry that via the s390kvm tree, thanks.