[RFC PATCH v4 0/2] s390x: Improvements to SIGP handling [KVM]

[RFC PATCH v4 0/2] s390x: Improvements to SIGP handling [KVM]

[Eric Farman]

Here is an update to the handling of SIGP between kernel and userspace.

As before, I'm looking at problems encountered when a SIGP order that is
processed in the kernel (for example, SIGP SENSE) is run concurrently
with another one is processed in userspace (for example, SIGP STOP).
Being able to provide an honest answer in the SIGP SENSE as to whether
the targeted VCPU is/not stopped is important to provide a consistent
answer while a guest OS is bringing its configuration online.
Furthermore, there is a desire to be architecturally correct in this
space, where the Principles of Operation states lists of orders
that should return a CC2 (BUSY) when any of another list of orders
is being processed on the targeted CPU.

This version goes back to something similar to v2, in that there's
only a single IOCTL created. But unlike v2, userspace handles both the
"set" and "reset" sides of the equation, specifying the operations in
the payload sent via the IOCTL.

With this, I started considering that the kernel itself could mark the
cpu busy/ready as it relates to SIGP, such that a !USER_SIGP userspace
would be happy with matters too, but haven't quite gotten that working
to my liking.

I'll send the QEMU series shortly, which takes advantage of this.

Thoughts?

Previous RFCs:
v1: https://lore.kernel.org/r/20211008203112.1979843-1-farman@linux.ibm.com/
v2: https://lore.kernel.org/r/20211102194652.2685098-1-farman@linux.ibm.com/
v3: https://lore.kernel.org/r/20211110203322.1374925-1-farman@linux.ibm.com/

Eric Farman (2):
  Capability/IOCTL/Documentation
  KVM: s390: Introduce a USER_BUSY capability and IOCTL

 Documentation/virt/kvm/api.rst   | 22 ++++++++++++++++++
 arch/s390/include/asm/kvm_host.h |  2 ++
 arch/s390/kvm/kvm-s390.c         | 40 ++++++++++++++++++++++++++++++++
 arch/s390/kvm/kvm-s390.h         | 15 ++++++++++++
 arch/s390/kvm/sigp.c             |  3 +++
 include/uapi/linux/kvm.h         | 16 +++++++++++++
 6 files changed, 98 insertions(+)

-- 
2.25.1

[RFC PATCH v4 1/2] Capability/IOCTL/Documentation

[Eric Farman]

(This should be squashed with the next patch; it's just broken
out for ease-of-future rebase.)

Signed-off-by: Eric Farman <farman@linux.ibm.com>
---
 Documentation/virt/kvm/api.rst | 22 ++++++++++++++++++++++
 include/uapi/linux/kvm.h       | 16 ++++++++++++++++
 2 files changed, 38 insertions(+)

diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst
index a6729c8cf063..012167775419 100644
--- a/Documentation/virt/kvm/api.rst
+++ b/Documentation/virt/kvm/api.rst
@@ -5317,6 +5317,18 @@ the trailing ``'\0'``, is indicated by ``name_size`` in the header.
 The Stats Data block contains an array of 64-bit values in the same order
 as the descriptors in Descriptors block.
 
+4.134 KVM_S390_USER_BUSY
+---------------------------------
+
+:Capability: KVM_CAP_S390_USER_BUSY
+:Architectures: s390
+:Type: vcpu ioctl
+:Parameters: struct kvm_s390_busy_info
+:Returns: 0, or -EBUSY if VCPU is already busy
+
+This ioctl sets the VCPU's indicator that it is busy processing an item
+of work, and is thus unavailable for additional work of that type.
+
 5. The kvm_run structure
 ========================
 
@@ -6706,6 +6718,16 @@ MAP_SHARED mmap will result in an -EINVAL return.
 When enabled the VMM may make use of the ``KVM_ARM_MTE_COPY_TAGS`` ioctl to
 perform a bulk copy of tags to/from the guest.
 
+7.29 KVM_CAP_S390_USER_BUSY
+--------------------------------
+
+:Architectures: s390
+:Parameters: none
+
+This capability indicates that KVM should indicate when a vcpu is busy
+performing some work, and conflicting work directed at the same vcpu should
+return CC2 (BUSY) until it has completed.
+
 8. Other capabilities.
 ======================
 
diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
index a067410ebea5..7f16f9fb4ae8 100644
--- a/include/uapi/linux/kvm.h
+++ b/include/uapi/linux/kvm.h
@@ -1112,6 +1112,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_BINARY_STATS_FD 203
 #define KVM_CAP_EXIT_ON_EMULATION_FAILURE 204
 #define KVM_CAP_ARM_MTE 205
+#define KVM_CAP_S390_USER_BUSY 206
 
 #ifdef KVM_CAP_IRQ_ROUTING
 
@@ -2007,4 +2008,19 @@ struct kvm_stats_desc {
 
 #define KVM_GET_STATS_FD  _IO(KVMIO,  0xce)
 
+/* Available with KVM_CAP_S390_USER_BUSY */
+#define KVM_S390_USER_BUSY	_IOW(KVMIO, 0xcf, struct kvm_s390_user_busy_info)
+
+#define KVM_S390_USER_BUSY_REASON_SIGP		1
+
+#define KVM_S390_USER_BUSY_FUNCTION_RESET	0
+#define KVM_S390_USER_BUSY_FUNCTION_SET		1
+
+/* FIXME struct description */
+struct kvm_s390_user_busy_info {
+	__u32 reason;
+	__u32 function;
+	__u32 payload;
+};
+
 #endif /* __LINUX_KVM_H */
-- 
2.25.1

[RFC PATCH v4 2/2] KVM: s390: Introduce a USER_BUSY capability and IOCTL

[Eric Farman]

With commit 2444b352c3ac ("KVM: s390: forward most SIGP orders to user
space") we have a capability that allows the "fast" SIGP orders (as
defined by the Programming Notes for the SIGNAL PROCESSOR instruction in
the Principles of Operation) to be handled in-kernel, while all others are
sent to userspace for processing.

This works fine but it creates a situation when, for example, a SIGP SENSE
might return CC1 (STATUS STORED, and status bits indicating the vcpu is
stopped), when in actuality userspace is still processing a SIGP STOP AND
STORE STATUS order, and the vcpu is not yet actually stopped. Thus, the
SIGP SENSE should actually be returning CC2 (busy) instead of CC1.

To fix this, add another CPU capability and an associated IOCTL.
The IOCTL can be used by userspace to mark a vcpu "busy" processing a
SIGP order, and cause concurrent orders handled in-kernel to be returned
with CC2 (busy). Another invocation of the IOCTL with a different payload
can be used to mark the SIGP completed, and thus the vcpu is free to
process additional orders.

Signed-off-by: Eric Farman <farman@linux.ibm.com>
---
 arch/s390/include/asm/kvm_host.h |  2 ++
 arch/s390/kvm/kvm-s390.c         | 40 ++++++++++++++++++++++++++++++++
 arch/s390/kvm/kvm-s390.h         | 15 ++++++++++++
 arch/s390/kvm/sigp.c             |  3 +++
 4 files changed, 60 insertions(+)

diff --git a/arch/s390/include/asm/kvm_host.h b/arch/s390/include/asm/kvm_host.h
index a604d51acfc8..d05cb4d2e1d5 100644
--- a/arch/s390/include/asm/kvm_host.h
+++ b/arch/s390/include/asm/kvm_host.h
@@ -748,6 +748,7 @@ struct kvm_vcpu_arch {
 	bool skey_enabled;
 	struct kvm_s390_pv_vcpu pv;
 	union diag318_info diag318_info;
+	atomic_t busy;
 };
 
 struct kvm_vm_stat {
@@ -941,6 +942,7 @@ struct kvm_arch{
 	int user_sigp;
 	int user_stsi;
 	int user_instr0;
+	int user_busy;
 	struct s390_io_adapter *adapters[MAX_S390_IO_ADAPTERS];
 	wait_queue_head_t ipte_wq;
 	int ipte_lock_count;
diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
index 5f52e7eec02f..c9f61777abea 100644
--- a/arch/s390/kvm/kvm-s390.c
+++ b/arch/s390/kvm/kvm-s390.c
@@ -564,6 +564,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 	case KVM_CAP_S390_VCPU_RESETS:
 	case KVM_CAP_SET_GUEST_DEBUG:
 	case KVM_CAP_S390_DIAG318:
+	case KVM_CAP_S390_USER_BUSY:
 		r = 1;
 		break;
 	case KVM_CAP_SET_GUEST_DEBUG2:
@@ -706,6 +707,11 @@ int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
 		kvm->arch.user_sigp = 1;
 		r = 0;
 		break;
+	case KVM_CAP_S390_USER_BUSY:
+		VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_BUSY");
+		kvm->arch.user_busy = 1;
+		r = 0;
+		break;
 	case KVM_CAP_S390_VECTOR_REGISTERS:
 		mutex_lock(&kvm->lock);
 		if (kvm->created_vcpus) {
@@ -4825,6 +4831,40 @@ long kvm_arch_vcpu_async_ioctl(struct file *filp,
 			return -EINVAL;
 		return kvm_s390_inject_vcpu(vcpu, &s390irq);
 	}
+	case KVM_S390_USER_BUSY: {
+		struct kvm_s390_user_busy_info busy;
+
+		if (!vcpu->kvm->arch.user_busy)
+			return -EINVAL;
+
+		if (copy_from_user(&busy, argp, sizeof(busy)))
+			return -EFAULT;
+
+		switch (busy.reason) {
+		case KVM_S390_USER_BUSY_REASON_SIGP:
+			break;
+		default:
+			return -EINVAL;
+		}
+
+		switch (busy.function) {
+		case KVM_S390_USER_BUSY_FUNCTION_RESET:
+			kvm_s390_vcpu_clear_busy(vcpu);
+			break;
+		case KVM_S390_USER_BUSY_FUNCTION_SET:
+			if (!kvm_s390_vcpu_set_busy(vcpu))
+				return -EBUSY;
+			break;
+		default:
+			return -EINVAL;
+		}
+
+		VCPU_EVENT(vcpu, 3, "BUSY: CPU %x %x reason %x payload %x",
+			   vcpu->vcpu_id,
+			   busy.function, busy.reason, busy.payload);
+
+		return 0;
+	}
 	}
 	return -ENOIOCTLCMD;
 }
diff --git a/arch/s390/kvm/kvm-s390.h b/arch/s390/kvm/kvm-s390.h
index c07a050d757d..92497c23e5a4 100644
--- a/arch/s390/kvm/kvm-s390.h
+++ b/arch/s390/kvm/kvm-s390.h
@@ -82,6 +82,21 @@ static inline int is_vcpu_idle(struct kvm_vcpu *vcpu)
 	return test_bit(vcpu->vcpu_idx, vcpu->kvm->arch.idle_mask);
 }
 
+static inline bool kvm_s390_vcpu_is_busy(struct kvm_vcpu *vcpu)
+{
+	return atomic_read(&vcpu->arch.busy) == 1;
+}
+
+static inline bool kvm_s390_vcpu_set_busy(struct kvm_vcpu *vcpu)
+{
+	return atomic_cmpxchg(&vcpu->arch.busy, 0, 1) == 0;
+}
+
+static inline void kvm_s390_vcpu_clear_busy(struct kvm_vcpu *vcpu)
+{
+	atomic_set(&vcpu->arch.busy, 0);
+}
+
 static inline int kvm_is_ucontrol(struct kvm *kvm)
 {
 #ifdef CONFIG_KVM_S390_UCONTROL
diff --git a/arch/s390/kvm/sigp.c b/arch/s390/kvm/sigp.c
index 5ad3fb4619f1..c3066b134e50 100644
--- a/arch/s390/kvm/sigp.c
+++ b/arch/s390/kvm/sigp.c
@@ -276,6 +276,9 @@ static int handle_sigp_dst(struct kvm_vcpu *vcpu, u8 order_code,
 	if (!dst_vcpu)
 		return SIGP_CC_NOT_OPERATIONAL;
 
+	if (kvm_s390_vcpu_is_busy(dst_vcpu))
+		return SIGP_CC_BUSY;
+
 	switch (order_code) {
 	case SIGP_SENSE:
 		vcpu->stat.instruction_sigp_sense++;
-- 
2.25.1