Merge tag 'kvm-s390-20140130' of...

Merge tag 'kvm-s390-20140130' of git://git.kernel.org/pub/scm/linux/kernel/git/kvms390/linux into HEAD

Two new features are added by this patch set:
- The floating interrupt controller (flic) that allows us to inject,
  clear and inspect non-vcpu local interrupts. This also gives us an
  opportunity to fix deficiencies in our existing interrupt definitions.
- Support for asynchronous page faults via the pfault mechanism. Testing
  show significant guest performance improvements under host swap.

Documentation/virtual/kvm/devices/s390_flic.txt

+FLIC (floating interrupt controller)
+====================================
+
+FLIC handles floating (non per-cpu) interrupts, i.e. I/O, service and some
+machine check interruptions. All interrupts are stored in a per-vm list of
+pending interrupts. FLIC performs operations on this list.
+
+Only one FLIC instance may be instantiated.
+
+FLIC provides support to
+- add interrupts (KVM_DEV_FLIC_ENQUEUE)
+- inspect currently pending interrupts (KVM_FLIC_GET_ALL_IRQS)
+- purge all pending floating interrupts (KVM_DEV_FLIC_CLEAR_IRQS)
+- enable/disable for the guest transparent async page faults
+
+Groups:
+  KVM_DEV_FLIC_ENQUEUE
+    Passes a buffer and length into the kernel which are then injected into
+    the list of pending interrupts.
+    attr->addr contains the pointer to the buffer and attr->attr contains
+    the length of the buffer.
+    The format of the data structure kvm_s390_irq as it is copied from userspace
+    is defined in usr/include/linux/kvm.h.
+
+  KVM_DEV_FLIC_GET_ALL_IRQS
+    Copies all floating interrupts into a buffer provided by userspace.
+    When the buffer is too small it returns -ENOMEM, which is the indication
+    for userspace to try again with a bigger buffer.
+    All interrupts remain pending, i.e. are not deleted from the list of
+    currently pending interrupts.
+    attr->addr contains the userspace address of the buffer into which all
+    interrupt data will be copied.
+    attr->attr contains the size of the buffer in bytes.
+
+  KVM_DEV_FLIC_CLEAR_IRQS
+    Simply deletes all elements from the list of currently pending floating
+    interrupts.  No interrupts are injected into the guest.
+
+  KVM_DEV_FLIC_APF_ENABLE
+    Enables async page faults for the guest. So in case of a major page fault
+    the host is allowed to handle this async and continues the guest.
+
+  KVM_DEV_FLIC_APF_DISABLE_WAIT
+    Disables async page faults for the guest and waits until already pending
+    async page faults are done. This is necessary to trigger a completion interrupt
+    for every init interrupt before migrating the interrupt list.

arch/s390/include/asm/kvm_host.h

@@ -16,6 +16,7 @@
 #include <linux/hrtimer.h>
 #include <linux/interrupt.h>
 #include <linux/kvm_host.h>
+#include <linux/kvm.h>
 #include <asm/debug.h>
 #include <asm/cpu.h>
 
@@ -168,18 +169,6 @@ struct kvm_vcpu_stat {
 	u32 diagnose_9c;
 };
 
-struct kvm_s390_io_info {
-	__u16        subchannel_id;            /* 0x0b8 */
-	__u16        subchannel_nr;            /* 0x0ba */
-	__u32        io_int_parm;              /* 0x0bc */
-	__u32        io_int_word;              /* 0x0c0 */
-};
-
-struct kvm_s390_ext_info {
-	__u32 ext_params;
-	__u64 ext_params2;
-};
-
 #define PGM_OPERATION            0x01
 #define PGM_PRIVILEGED_OP	 0x02
 #define PGM_EXECUTE              0x03
@@ -188,27 +177,6 @@ struct kvm_s390_ext_info {
 #define PGM_SPECIFICATION        0x06
 #define PGM_DATA                 0x07
 
-struct kvm_s390_pgm_info {
-	__u16 code;
-};
-
-struct kvm_s390_prefix_info {
-	__u32 address;
-};
-
-struct kvm_s390_extcall_info {
-	__u16 code;
-};
-
-struct kvm_s390_emerg_info {
-	__u16 code;
-};
-
-struct kvm_s390_mchk_info {
-	__u64 cr14;
-	__u64 mcic;
-};
-
 struct kvm_s390_interrupt_info {
 	struct list_head list;
 	u64	type;
@@ -246,6 +214,7 @@ struct kvm_s390_float_interrupt {
 	unsigned long idle_mask[(KVM_MAX_VCPUS + sizeof(long) - 1)
 				/ sizeof(long)];
 	struct kvm_s390_local_interrupt *local_int[KVM_MAX_VCPUS];
+	unsigned int irq_count;
 };
 
 
@@ -262,6 +231,10 @@ struct kvm_vcpu_arch {
 		u64		stidp_data;
 	};
 	struct gmap *gmap;
+#define KVM_S390_PFAULT_TOKEN_INVALID	(-1UL)
+	unsigned long pfault_token;
+	unsigned long pfault_select;
+	unsigned long pfault_compare;
 };
 
 struct kvm_vm_stat {
@@ -275,6 +248,7 @@ struct kvm_arch{
 	struct sca_block *sca;
 	debug_info_t *dbf;
 	struct kvm_s390_float_interrupt float_int;
+	struct kvm_device *flic;
 	struct gmap *gmap;
 	int css_support;
 };
@@ -287,6 +261,24 @@ static inline bool kvm_is_error_hva(unsigned long addr)
 	return IS_ERR_VALUE(addr);
 }
 
+#define ASYNC_PF_PER_VCPU	64
+struct kvm_vcpu;
+struct kvm_async_pf;
+struct kvm_arch_async_pf {
+	unsigned long pfault_token;
+};
+
+bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu);
+
+void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
+			       struct kvm_async_pf *work);
+
+void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
+				     struct kvm_async_pf *work);
+
+void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
+				 struct kvm_async_pf *work);
+
 extern int sie64a(struct kvm_s390_sie_block *, u64 *);
 extern char sie_exit;
 #endif

arch/s390/include/asm/pgtable.h

@@ -767,6 +767,7 @@ static inline void pgste_set_pte(pte_t *ptep, pte_t entry)
  * @table: pointer to the page directory
  * @asce: address space control element for gmap page table
  * @crst_list: list of all crst tables used in the guest address space
+ * @pfault_enabled: defines if pfaults are applicable for the guest
  */
 struct gmap {
 	struct list_head list;
@@ -775,6 +776,7 @@ struct gmap {
 	unsigned long asce;
 	void *private;
 	struct list_head crst_list;
+	bool pfault_enabled;
 };
 
 /**

arch/s390/include/asm/processor.h

@@ -79,6 +79,7 @@ struct thread_struct {
         unsigned long ksp;              /* kernel stack pointer             */
 	mm_segment_t mm_segment;
 	unsigned long gmap_addr;	/* address of last gmap fault. */
+	unsigned int gmap_pfault;	/* signal of a pending guest pfault */
 	struct per_regs per_user;	/* User specified PER registers */
 	struct per_event per_event;	/* Cause of the last PER trap */
 	unsigned long per_flags;	/* Flags to control debug behavior */

arch/s390/include/uapi/asm/kvm.h

@@ -16,6 +16,22 @@
 
 #define __KVM_S390
 
+/* Device control API: s390-specific devices */
+#define KVM_DEV_FLIC_GET_ALL_IRQS	1
+#define KVM_DEV_FLIC_ENQUEUE		2
+#define KVM_DEV_FLIC_CLEAR_IRQS		3
+#define KVM_DEV_FLIC_APF_ENABLE		4
+#define KVM_DEV_FLIC_APF_DISABLE_WAIT	5
+/*
+ * We can have up to 4*64k pending subchannels + 8 adapter interrupts,
+ * as well as up  to ASYNC_PF_PER_VCPU*KVM_MAX_VCPUS pfault done interrupts.
+ * There are also sclp and machine checks. This gives us
+ * sizeof(kvm_s390_irq)*(4*65536+8+64*64+1+1) = 72 * 266250 = 19170000
+ * Lets round up to 8192 pages.
+ */
+#define KVM_S390_MAX_FLOAT_IRQS	266250
+#define KVM_S390_FLIC_MAX_BUFFER	0x2000000
+
 /* for KVM_GET_REGS and KVM_SET_REGS */
 struct kvm_regs {
 	/* general purpose regs for s390 */
@@ -57,4 +73,7 @@ struct kvm_sync_regs {
 #define KVM_REG_S390_EPOCHDIFF	(KVM_REG_S390 | KVM_REG_SIZE_U64 | 0x2)
 #define KVM_REG_S390_CPU_TIMER  (KVM_REG_S390 | KVM_REG_SIZE_U64 | 0x3)
 #define KVM_REG_S390_CLOCK_COMP (KVM_REG_S390 | KVM_REG_SIZE_U64 | 0x4)
+#define KVM_REG_S390_PFTOKEN	(KVM_REG_S390 | KVM_REG_SIZE_U64 | 0x5)
+#define KVM_REG_S390_PFCOMPARE	(KVM_REG_S390 | KVM_REG_SIZE_U64 | 0x6)
+#define KVM_REG_S390_PFSELECT	(KVM_REG_S390 | KVM_REG_SIZE_U64 | 0x7)
 #endif

arch/s390/kvm/Kconfig

@@ -23,6 +23,8 @@ config KVM
 	select ANON_INODES
 	select HAVE_KVM_CPU_RELAX_INTERCEPT
 	select HAVE_KVM_EVENTFD
+	select KVM_ASYNC_PF
+	select KVM_ASYNC_PF_SYNC
 	---help---
 	  Support hosting paravirtualized guest machines using the SIE
 	  virtualization capability on the mainframe. This should work

arch/s390/kvm/Makefile

@@ -7,7 +7,7 @@
 # as published by the Free Software Foundation.
 
 KVM := ../../../virt/kvm
-common-objs = $(KVM)/kvm_main.o $(KVM)/eventfd.o
+common-objs = $(KVM)/kvm_main.o $(KVM)/eventfd.o  $(KVM)/async_pf.o
 
 ccflags-y := -Ivirt/kvm -Iarch/s390/kvm
 

arch/s390/kvm/diag.c

@@ -17,6 +17,7 @@
 #include "kvm-s390.h"
 #include "trace.h"
 #include "trace-s390.h"
+#include "gaccess.h"
 
 static int diag_release_pages(struct kvm_vcpu *vcpu)
 {
@@ -46,6 +47,87 @@ static int diag_release_pages(struct kvm_vcpu *vcpu)
 	return 0;
 }
 
+static int __diag_page_ref_service(struct kvm_vcpu *vcpu)
+{
+	struct prs_parm {
+		u16 code;
+		u16 subcode;
+		u16 parm_len;
+		u16 parm_version;
+		u64 token_addr;
+		u64 select_mask;
+		u64 compare_mask;
+		u64 zarch;
+	};
+	struct prs_parm parm;
+	int rc;
+	u16 rx = (vcpu->arch.sie_block->ipa & 0xf0) >> 4;
+	u16 ry = (vcpu->arch.sie_block->ipa & 0x0f);
+	unsigned long hva_token = KVM_HVA_ERR_BAD;
+
+	if (vcpu->run->s.regs.gprs[rx] & 7)
+		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
+	if (copy_from_guest(vcpu, &parm, vcpu->run->s.regs.gprs[rx], sizeof(parm)))
+		return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
+	if (parm.parm_version != 2 || parm.parm_len < 5 || parm.code != 0x258)
+		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
+
+	switch (parm.subcode) {
+	case 0: /* TOKEN */
+		if (vcpu->arch.pfault_token != KVM_S390_PFAULT_TOKEN_INVALID) {
+			/*
+			 * If the pagefault handshake is already activated,
+			 * the token must not be changed.  We have to return
+			 * decimal 8 instead, as mandated in SC24-6084.
+			 */
+			vcpu->run->s.regs.gprs[ry] = 8;
+			return 0;
+		}
+
+		if ((parm.compare_mask & parm.select_mask) != parm.compare_mask ||
+		    parm.token_addr & 7 || parm.zarch != 0x8000000000000000ULL)
+			return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
+
+		hva_token = gfn_to_hva(vcpu->kvm, gpa_to_gfn(parm.token_addr));
+		if (kvm_is_error_hva(hva_token))
+			return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
+
+		vcpu->arch.pfault_token = parm.token_addr;
+		vcpu->arch.pfault_select = parm.select_mask;
+		vcpu->arch.pfault_compare = parm.compare_mask;
+		vcpu->run->s.regs.gprs[ry] = 0;
+		rc = 0;
+		break;
+	case 1: /*
+		 * CANCEL
+		 * Specification allows to let already pending tokens survive
+		 * the cancel, therefore to reduce code complexity, we assume
+		 * all outstanding tokens are already pending.
+		 */
+		if (parm.token_addr || parm.select_mask ||
+		    parm.compare_mask || parm.zarch)
+			return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
+
+		vcpu->run->s.regs.gprs[ry] = 0;
+		/*
+		 * If the pfault handling was not established or is already
+		 * canceled SC24-6084 requests to return decimal 4.
+		 */
+		if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
+			vcpu->run->s.regs.gprs[ry] = 4;
+		else
+			vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
+
+		rc = 0;
+		break;
+	default:
+		rc = -EOPNOTSUPP;
+		break;
+	}
+
+	return rc;
+}
+
 static int __diag_time_slice_end(struct kvm_vcpu *vcpu)
 {
 	VCPU_EVENT(vcpu, 5, "%s", "diag time slice end");
@@ -150,6 +232,8 @@ int kvm_s390_handle_diag(struct kvm_vcpu *vcpu)
 		return __diag_time_slice_end(vcpu);
 	case 0x9c:
 		return __diag_time_slice_end_directed(vcpu);
+	case 0x258:
+		return __diag_page_ref_service(vcpu);
 	case 0x308:
 		return __diag_ipl_functions(vcpu);
 	case 0x500:

arch/s390/kvm/interrupt.c

@@ -31,7 +31,7 @@ static int is_ioint(u64 type)
 	return ((type & 0xfffe0000u) != 0xfffe0000u);
 }
 
-static int psw_extint_disabled(struct kvm_vcpu *vcpu)
+int psw_extint_disabled(struct kvm_vcpu *vcpu)
 {
 	return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_EXT);
 }
@@ -78,11 +78,8 @@ static int __interrupt_is_deliverable(struct kvm_vcpu *vcpu,
 			return 1;
 		return 0;
 	case KVM_S390_INT_SERVICE:
-		if (psw_extint_disabled(vcpu))
-			return 0;
-		if (vcpu->arch.sie_block->gcr[0] & 0x200ul)
-			return 1;
-		return 0;
+	case KVM_S390_INT_PFAULT_INIT:
+	case KVM_S390_INT_PFAULT_DONE:
 	case KVM_S390_INT_VIRTIO:
 		if (psw_extint_disabled(vcpu))
 			return 0;
@@ -150,6 +147,8 @@ static void __set_intercept_indicator(struct kvm_vcpu *vcpu,
 	case KVM_S390_INT_EXTERNAL_CALL:
 	case KVM_S390_INT_EMERGENCY:
 	case KVM_S390_INT_SERVICE:
+	case KVM_S390_INT_PFAULT_INIT:
+	case KVM_S390_INT_PFAULT_DONE:
 	case KVM_S390_INT_VIRTIO:
 		if (psw_extint_disabled(vcpu))
 			__set_cpuflag(vcpu, CPUSTAT_EXT_INT);
@@ -223,6 +222,30 @@ static void __do_deliver_interrupt(struct kvm_vcpu *vcpu,
 		rc |= put_guest(vcpu, inti->ext.ext_params,
 				(u32 __user *)__LC_EXT_PARAMS);
 		break;
+	case KVM_S390_INT_PFAULT_INIT:
+		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, inti->type, 0,
+						 inti->ext.ext_params2);
+		rc  = put_guest(vcpu, 0x2603, (u16 __user *) __LC_EXT_INT_CODE);
+		rc |= put_guest(vcpu, 0x0600, (u16 __user *) __LC_EXT_CPU_ADDR);
+		rc |= copy_to_guest(vcpu, __LC_EXT_OLD_PSW,
+				    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
+		rc |= copy_from_guest(vcpu, &vcpu->arch.sie_block->gpsw,
+				      __LC_EXT_NEW_PSW, sizeof(psw_t));
+		rc |= put_guest(vcpu, inti->ext.ext_params2,
+				(u64 __user *) __LC_EXT_PARAMS2);
+		break;
+	case KVM_S390_INT_PFAULT_DONE:
+		trace_kvm_s390_deliver_interrupt(vcpu->vcpu_id, inti->type, 0,
+						 inti->ext.ext_params2);
+		rc  = put_guest(vcpu, 0x2603, (u16 __user *) __LC_EXT_INT_CODE);
+		rc |= put_guest(vcpu, 0x0680, (u16 __user *) __LC_EXT_CPU_ADDR);
+		rc |= copy_to_guest(vcpu, __LC_EXT_OLD_PSW,
+				    &vcpu->arch.sie_block->gpsw, sizeof(psw_t));
+		rc |= copy_from_guest(vcpu, &vcpu->arch.sie_block->gpsw,
+				      __LC_EXT_NEW_PSW, sizeof(psw_t));
+		rc |= put_guest(vcpu, inti->ext.ext_params2,
+				(u64 __user *) __LC_EXT_PARAMS2);
+		break;
 	case KVM_S390_INT_VIRTIO:
 		VCPU_EVENT(vcpu, 4, "interrupt: virtio parm:%x,parm64:%llx",
 			   inti->ext.ext_params, inti->ext.ext_params2);
@@ -357,7 +380,7 @@ static int __try_deliver_ckc_interrupt(struct kvm_vcpu *vcpu)
 	return 1;
 }
 
-static int kvm_cpu_has_interrupt(struct kvm_vcpu *vcpu)
+int kvm_cpu_has_interrupt(struct kvm_vcpu *vcpu)
 {
 	struct kvm_s390_local_interrupt *li = &vcpu->arch.local_int;
 	struct kvm_s390_float_interrupt *fi = vcpu->arch.local_int.float_int;
@@ -528,6 +551,7 @@ void kvm_s390_deliver_pending_interrupts(struct kvm_vcpu *vcpu)
 			list_for_each_entry_safe(inti, n, &fi->list, list) {
 				if (__interrupt_is_deliverable(vcpu, inti)) {
 					list_del(&inti->list);
+					fi->irq_count--;
 					deliver = 1;
 					break;
 				}
@@ -583,6 +607,7 @@ void kvm_s390_deliver_pending_machine_checks(struct kvm_vcpu *vcpu)
 				if ((inti->type == KVM_S390_MCHK) &&
 				    __interrupt_is_deliverable(vcpu, inti)) {
 					list_del(&inti->list);
+					fi->irq_count--;
 					deliver = 1;
 					break;
 				}
@@ -650,8 +675,10 @@ struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
 		inti = iter;
 		break;
 	}
-	if (inti)
+	if (inti) {
 		list_del_init(&inti->list);
+		fi->irq_count--;
+	}
 	if (list_empty(&fi->list))
 		atomic_set(&fi->active, 0);
 	spin_unlock(&fi->lock);
@@ -659,53 +686,98 @@ struct kvm_s390_interrupt_info *kvm_s390_get_io_int(struct kvm *kvm,
 	return inti;
 }
 
-int kvm_s390_inject_vm(struct kvm *kvm,
-		       struct kvm_s390_interrupt *s390int)
+static int __inject_vm(struct kvm *kvm, struct kvm_s390_interrupt_info *inti)
 {
 	struct kvm_s390_local_interrupt *li;
 	struct kvm_s390_float_interrupt *fi;
-	struct kvm_s390_interrupt_info *inti, *iter;
+	struct kvm_s390_interrupt_info *iter;
 	int sigcpu;
+	int rc = 0;
+
+	mutex_lock(&kvm->lock);
+	fi = &kvm->arch.float_int;
+	spin_lock(&fi->lock);
+	if (fi->irq_count >= KVM_S390_MAX_FLOAT_IRQS) {
+		rc = -EINVAL;
+		goto unlock_fi;
+	}
+	fi->irq_count++;
+	if (!is_ioint(inti->type)) {
+		list_add_tail(&inti->list, &fi->list);
+	} else {
+		u64 isc_bits = int_word_to_isc_bits(inti->io.io_int_word);
+
+		/* Keep I/O interrupts sorted in isc order. */
+		list_for_each_entry(iter, &fi->list, list) {
+			if (!is_ioint(iter->type))
+				continue;
+			if (int_word_to_isc_bits(iter->io.io_int_word)
+			    <= isc_bits)
+				continue;
+			break;
+		}
+		list_add_tail(&inti->list, &iter->list);
+	}
+	atomic_set(&fi->active, 1);
+	sigcpu = find_first_bit(fi->idle_mask, KVM_MAX_VCPUS);
+	if (sigcpu == KVM_MAX_VCPUS) {
+		do {
+			sigcpu = fi->next_rr_cpu++;
+			if (sigcpu == KVM_MAX_VCPUS)
+				sigcpu = fi->next_rr_cpu = 0;
+		} while (fi->local_int[sigcpu] == NULL);
+	}
+	li = fi->local_int[sigcpu];
+	spin_lock_bh(&li->lock);
+	atomic_set_mask(CPUSTAT_EXT_INT, li->cpuflags);
+	if (waitqueue_active(li->wq))
+		wake_up_interruptible(li->wq);
+	spin_unlock_bh(&li->lock);
+unlock_fi:
+	spin_unlock(&fi->lock);
+	mutex_unlock(&kvm->lock);
+	return rc;
+}
+
+int kvm_s390_inject_vm(struct kvm *kvm,
+		       struct kvm_s390_interrupt *s390int)
+{
+	struct kvm_s390_interrupt_info *inti;
 
 	inti = kzalloc(sizeof(*inti), GFP_KERNEL);
 	if (!inti)
 		return -ENOMEM;
 
-	switch (s390int->type) {
+	inti->type = s390int->type;
+	switch (inti->type) {
 	case KVM_S390_INT_VIRTIO:
 		VM_EVENT(kvm, 5, "inject: virtio parm:%x,parm64:%llx",
 			 s390int->parm, s390int->parm64);
-		inti->type = s390int->type;
 		inti->ext.ext_params = s390int->parm;
 		inti->ext.ext_params2 = s390int->parm64;
 		break;
 	case KVM_S390_INT_SERVICE:
 		VM_EVENT(kvm, 5, "inject: sclp parm:%x", s390int->parm);
-		inti->type = s390int->type;
 		inti->ext.ext_params = s390int->parm;
 		break;
-	case KVM_S390_PROGRAM_INT:
-	case KVM_S390_SIGP_STOP:
-	case KVM_S390_INT_EXTERNAL_CALL:
-	case KVM_S390_INT_EMERGENCY:
-		kfree(inti);
-		return -EINVAL;
+	case KVM_S390_INT_PFAULT_DONE:
+		inti->type = s390int->type;
+		inti->ext.ext_params2 = s390int->parm64;
+		break;
 	case KVM_S390_MCHK:
 		VM_EVENT(kvm, 5, "inject: machine check parm64:%llx",
 			 s390int->parm64);
-		inti->type = s390int->type;
 		inti->mchk.cr14 = s390int->parm; /* upper bits are not used */
 		inti->mchk.mcic = s390int->parm64;
 		break;
 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
-		if (s390int->type & IOINT_AI_MASK)
+		if (inti->type & IOINT_AI_MASK)
 			VM_EVENT(kvm, 5, "%s", "inject: I/O (AI)");
 		else
 			VM_EVENT(kvm, 5, "inject: I/O css %x ss %x schid %04x",
 				 s390int->type & IOINT_CSSID_MASK,
 				 s390int->type & IOINT_SSID_MASK,
 				 s390int->type & IOINT_SCHID_MASK);
-		inti->type = s390int->type;
 		inti->io.subchannel_id = s390int->parm >> 16;
 		inti->io.subchannel_nr = s390int->parm & 0x0000ffffu;
 		inti->io.io_int_parm = s390int->parm64 >> 32;
@@ -718,43 +790,7 @@ int kvm_s390_inject_vm(struct kvm *kvm,
 	trace_kvm_s390_inject_vm(s390int->type, s390int->parm, s390int->parm64,
 				 2);
 
-	mutex_lock(&kvm->lock);
-	fi = &kvm->arch.float_int;
-	spin_lock(&fi->lock);
-	if (!is_ioint(inti->type))
-		list_add_tail(&inti->list, &fi->list);
-	else {
-		u64 isc_bits = int_word_to_isc_bits(inti->io.io_int_word);
-
-		/* Keep I/O interrupts sorted in isc order. */
-		list_for_each_entry(iter, &fi->list, list) {
-			if (!is_ioint(iter->type))
-				continue;
-			if (int_word_to_isc_bits(iter->io.io_int_word)
-			    <= isc_bits)
-				continue;
-			break;
-		}
-		list_add_tail(&inti->list, &iter->list);
-	}
-	atomic_set(&fi->active, 1);
-	sigcpu = find_first_bit(fi->idle_mask, KVM_MAX_VCPUS);
-	if (sigcpu == KVM_MAX_VCPUS) {
-		do {
-			sigcpu = fi->next_rr_cpu++;
-			if (sigcpu == KVM_MAX_VCPUS)
-				sigcpu = fi->next_rr_cpu = 0;
-		} while (fi->local_int[sigcpu] == NULL);
-	}
-	li = fi->local_int[sigcpu];
-	spin_lock_bh(&li->lock);
-	atomic_set_mask(CPUSTAT_EXT_INT, li->cpuflags);
-	if (waitqueue_active(li->wq))
-		wake_up_interruptible(li->wq);
-	spin_unlock_bh(&li->lock);
-	spin_unlock(&fi->lock);
-	mutex_unlock(&kvm->lock);
-	return 0;
+	return __inject_vm(kvm, inti);
 }
 
 int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu,
@@ -814,6 +850,10 @@ int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu,
 		inti->type = s390int->type;
 		inti->mchk.mcic = s390int->parm64;
 		break;
+	case KVM_S390_INT_PFAULT_INIT:
+		inti->type = s390int->type;
+		inti->ext.ext_params2 = s390int->parm64;
+		break;
 	case KVM_S390_INT_VIRTIO:
 	case KVM_S390_INT_SERVICE:
 	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
@@ -841,3 +881,232 @@ int kvm_s390_inject_vcpu(struct kvm_vcpu *vcpu,
 	mutex_unlock(&vcpu->kvm->lock);
 	return 0;
 }
+
+static void clear_floating_interrupts(struct kvm *kvm)
+{
+	struct kvm_s390_float_interrupt *fi;
+	struct kvm_s390_interrupt_info	*n, *inti = NULL;
+
+	mutex_lock(&kvm->lock);
+	fi = &kvm->arch.float_int;
+	spin_lock(&fi->lock);
+	list_for_each_entry_safe(inti, n, &fi->list, list) {
+		list_del(&inti->list);
+		kfree(inti);
+	}
+	fi->irq_count = 0;
+	atomic_set(&fi->active, 0);
+	spin_unlock(&fi->lock);
+	mutex_unlock(&kvm->lock);
+}
+
+static inline int copy_irq_to_user(struct kvm_s390_interrupt_info *inti,
+				   u8 *addr)
+{
+	struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
+	struct kvm_s390_irq irq = {0};
+
+	irq.type = inti->type;
+	switch (inti->type) {
+	case KVM_S390_INT_PFAULT_INIT:
+	case KVM_S390_INT_PFAULT_DONE:
+	case KVM_S390_INT_VIRTIO:
+	case KVM_S390_INT_SERVICE:
+		irq.u.ext = inti->ext;
+		break;
+	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
+		irq.u.io = inti->io;
+		break;
+	case KVM_S390_MCHK:
+		irq.u.mchk = inti->mchk;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	if (copy_to_user(uptr, &irq, sizeof(irq)))
+		return -EFAULT;
+
+	return 0;
+}
+
+static int get_all_floating_irqs(struct kvm *kvm, __u8 *buf, __u64 len)
+{
+	struct kvm_s390_interrupt_info *inti;
+	struct kvm_s390_float_interrupt *fi;
+	int ret = 0;
+	int n = 0;
+
+	mutex_lock(&kvm->lock);
+	fi = &kvm->arch.float_int;
+	spin_lock(&fi->lock);
+
+	list_for_each_entry(inti, &fi->list, list) {
+		if (len < sizeof(struct kvm_s390_irq)) {
+			/* signal userspace to try again */
+			ret = -ENOMEM;
+			break;
+		}
+		ret = copy_irq_to_user(inti, buf);
+		if (ret)
+			break;
+		buf += sizeof(struct kvm_s390_irq);
+		len -= sizeof(struct kvm_s390_irq);
+		n++;
+	}
+
+	spin_unlock(&fi->lock);
+	mutex_unlock(&kvm->lock);
+
+	return ret < 0 ? ret : n;
+}
+
+static int flic_get_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
+{
+	int r;
+
+	switch (attr->group) {
+	case KVM_DEV_FLIC_GET_ALL_IRQS:
+		r = get_all_floating_irqs(dev->kvm, (u8 *) attr->addr,
+					  attr->attr);
+		break;
+	default:
+		r = -EINVAL;
+	}
+
+	return r;
+}
+
+static inline int copy_irq_from_user(struct kvm_s390_interrupt_info *inti,
+				     u64 addr)
+{
+	struct kvm_s390_irq __user *uptr = (struct kvm_s390_irq __user *) addr;
+	void *target = NULL;
+	void __user *source;
+	u64 size;
+
+	if (get_user(inti->type, (u64 __user *)addr))
+		return -EFAULT;
+
+	switch (inti->type) {
+	case KVM_S390_INT_PFAULT_INIT:
+	case KVM_S390_INT_PFAULT_DONE:
+	case KVM_S390_INT_VIRTIO:
+	case KVM_S390_INT_SERVICE:
+		target = (void *) &inti->ext;
+		source = &uptr->u.ext;
+		size = sizeof(inti->ext);
+		break;
+	case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
+		target = (void *) &inti->io;
+		source = &uptr->u.io;
+		size = sizeof(inti->io);
+		break;
+	case KVM_S390_MCHK:
+		target = (void *) &inti->mchk;
+		source = &uptr->u.mchk;
+		size = sizeof(inti->mchk);
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	if (copy_from_user(target, source, size))
+		return -EFAULT;
+
+	return 0;
+}
+
+static int enqueue_floating_irq(struct kvm_device *dev,
+				struct kvm_device_attr *attr)
+{
+	struct kvm_s390_interrupt_info *inti = NULL;
+	int r = 0;
+	int len = attr->attr;
+
+	if (len % sizeof(struct kvm_s390_irq) != 0)
+		return -EINVAL;
+	else if (len > KVM_S390_FLIC_MAX_BUFFER)
+		return -EINVAL;
+
+	while (len >= sizeof(struct kvm_s390_irq)) {
+		inti = kzalloc(sizeof(*inti), GFP_KERNEL);
+		if (!inti)
+			return -ENOMEM;
+
+		r = copy_irq_from_user(inti, attr->addr);
+		if (r) {
+			kfree(inti);
+			return r;
+		}
+		r = __inject_vm(dev->kvm, inti);
+		if (r) {
+			kfree(inti);
+			return r;
+		}
+		len -= sizeof(struct kvm_s390_irq);
+		attr->addr += sizeof(struct kvm_s390_irq);
+	}
+
+	return r;
+}
+
+static int flic_set_attr(struct kvm_device *dev, struct kvm_device_attr *attr)
+{
+	int r = 0;
+	unsigned int i;
+	struct kvm_vcpu *vcpu;
+
+	switch (attr->group) {
+	case KVM_DEV_FLIC_ENQUEUE:
+		r = enqueue_floating_irq(dev, attr);
+		break;
+	case KVM_DEV_FLIC_CLEAR_IRQS:
+		r = 0;
+		clear_floating_interrupts(dev->kvm);
+		break;
+	case KVM_DEV_FLIC_APF_ENABLE:
+		dev->kvm->arch.gmap->pfault_enabled = 1;
+		break;
+	case KVM_DEV_FLIC_APF_DISABLE_WAIT:
+		dev->kvm->arch.gmap->pfault_enabled = 0;
+		/*
+		 * Make sure no async faults are in transition when
+		 * clearing the queues. So we don't need to worry
+		 * about late coming workers.
+		 */
+		synchronize_srcu(&dev->kvm->srcu);
+		kvm_for_each_vcpu(i, vcpu, dev->kvm)
+			kvm_clear_async_pf_completion_queue(vcpu);
+		break;
+	default:
+		r = -EINVAL;
+	}
+
+	return r;
+}
+
+static int flic_create(struct kvm_device *dev, u32 type)
+{
+	if (!dev)
+		return -EINVAL;
+	if (dev->kvm->arch.flic)
+		return -EINVAL;
+	dev->kvm->arch.flic = dev;
+	return 0;
+}
+
+static void flic_destroy(struct kvm_device *dev)
+{
+	dev->kvm->arch.flic = NULL;
+	kfree(dev);
+}
+
+/* s390 floating irq controller (flic) */
+struct kvm_device_ops kvm_flic_ops = {
+	.name = "kvm-flic",
+	.get_attr = flic_get_attr,
+	.set_attr = flic_set_attr,
+	.create = flic_create,
+	.destroy = flic_destroy,
+};

arch/s390/kvm/kvm-s390.c

@@ -152,11 +152,13 @@ int kvm_dev_ioctl_check_extension(long ext)
 #ifdef CONFIG_KVM_S390_UCONTROL
 	case KVM_CAP_S390_UCONTROL:
 #endif
+	case KVM_CAP_ASYNC_PF:
 	case KVM_CAP_SYNC_REGS:
 	case KVM_CAP_ONE_REG:
 	case KVM_CAP_ENABLE_CAP:
 	case KVM_CAP_S390_CSS_SUPPORT:
 	case KVM_CAP_IOEVENTFD:
+	case KVM_CAP_DEVICE_CTRL:
 		r = 1;
 		break;
 	case KVM_CAP_NR_VCPUS:
@@ -254,6 +256,7 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 		if (!kvm->arch.gmap)
 			goto out_nogmap;
 		kvm->arch.gmap->private = kvm;
+		kvm->arch.gmap->pfault_enabled = 0;
 	}
 
 	kvm->arch.css_support = 0;
@@ -271,6 +274,7 @@ void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 {
 	VCPU_EVENT(vcpu, 3, "%s", "free cpu");
 	trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
+	kvm_clear_async_pf_completion_queue(vcpu);
 	if (!kvm_is_ucontrol(vcpu->kvm)) {
 		clear_bit(63 - vcpu->vcpu_id,
 			  (unsigned long *) &vcpu->kvm->arch.sca->mcn);
@@ -320,6 +324,8 @@ void kvm_arch_destroy_vm(struct kvm *kvm)
 /* Section: vcpu related */
 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 {
+	vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
+	kvm_clear_async_pf_completion_queue(vcpu);
 	if (kvm_is_ucontrol(vcpu->kvm)) {
 		vcpu->arch.gmap = gmap_alloc(current->mm);
 		if (!vcpu->arch.gmap)
@@ -380,6 +386,8 @@ static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
 	vcpu->arch.guest_fpregs.fpc = 0;
 	asm volatile("lfpc %0" : : "Q" (vcpu->arch.guest_fpregs.fpc));
 	vcpu->arch.sie_block->gbea = 1;
+	vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
+	kvm_clear_async_pf_completion_queue(vcpu);
 	atomic_set_mask(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
 }
 
@@ -553,6 +561,18 @@ static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
 		r = put_user(vcpu->arch.sie_block->ckc,
 			     (u64 __user *)reg->addr);
 		break;
+	case KVM_REG_S390_PFTOKEN:
+		r = put_user(vcpu->arch.pfault_token,
+			     (u64 __user *)reg->addr);
+		break;
+	case KVM_REG_S390_PFCOMPARE:
+		r = put_user(vcpu->arch.pfault_compare,
+			     (u64 __user *)reg->addr);
+		break;
+	case KVM_REG_S390_PFSELECT:
+		r = put_user(vcpu->arch.pfault_select,
+			     (u64 __user *)reg->addr);
+		break;
 	default:
 		break;
 	}
@@ -582,6 +602,18 @@ static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
 		r = get_user(vcpu->arch.sie_block->ckc,
 			     (u64 __user *)reg->addr);
 		break;
+	case KVM_REG_S390_PFTOKEN:
+		r = get_user(vcpu->arch.pfault_token,
+			     (u64 __user *)reg->addr);
+		break;
+	case KVM_REG_S390_PFCOMPARE:
+		r = get_user(vcpu->arch.pfault_compare,
+			     (u64 __user *)reg->addr);
+		break;
+	case KVM_REG_S390_PFSELECT:
+		r = get_user(vcpu->arch.pfault_select,
+			     (u64 __user *)reg->addr);
+		break;
 	default:
 		break;
 	}
@@ -700,10 +732,100 @@ static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
 	return 0;
 }
 
+static long kvm_arch_fault_in_sync(struct kvm_vcpu *vcpu)
+{
+	long rc;
+	hva_t fault = gmap_fault(current->thread.gmap_addr, vcpu->arch.gmap);
+	struct mm_struct *mm = current->mm;
+	down_read(&mm->mmap_sem);
+	rc = get_user_pages(current, mm, fault, 1, 1, 0, NULL, NULL);
+	up_read(&mm->mmap_sem);
+	return rc;
+}
+
+static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
+				      unsigned long token)
+{
+	struct kvm_s390_interrupt inti;
+	inti.parm64 = token;
+
+	if (start_token) {
+		inti.type = KVM_S390_INT_PFAULT_INIT;
+		WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &inti));
+	} else {
+		inti.type = KVM_S390_INT_PFAULT_DONE;
+		WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
+	}
+}
+
+void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
+				     struct kvm_async_pf *work)
+{
+	trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
+	__kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
+}
+
+void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
+				 struct kvm_async_pf *work)
+{
+	trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
+	__kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
+}
+
+void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
+			       struct kvm_async_pf *work)
+{
+	/* s390 will always inject the page directly */
+}
+
+bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
+{
+	/*
+	 * s390 will always inject the page directly,
+	 * but we still want check_async_completion to cleanup
+	 */
+	return true;
+}
+
+static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
+{
+	hva_t hva;
+	struct kvm_arch_async_pf arch;
+	int rc;
+
+	if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
+		return 0;
+	if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
+	    vcpu->arch.pfault_compare)
+		return 0;
+	if (psw_extint_disabled(vcpu))
+		return 0;
+	if (kvm_cpu_has_interrupt(vcpu))
+		return 0;
+	if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
+		return 0;
+	if (!vcpu->arch.gmap->pfault_enabled)
+		return 0;
+
+	hva = gmap_fault(current->thread.gmap_addr, vcpu->arch.gmap);
+	if (copy_from_guest(vcpu, &arch.pfault_token, vcpu->arch.pfault_token, 8))
+		return 0;
+
+	rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
+	return rc;
+}
+
 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
 {
 	int rc, cpuflags;
 
+	/*
+	 * On s390 notifications for arriving pages will be delivered directly
+	 * to the guest but the house keeping for completed pfaults is
+	 * handled outside the worker.
+	 */
+	kvm_check_async_pf_completion(vcpu);
+
 	memcpy(&vcpu->arch.sie_block->gg14, &vcpu->run->s.regs.gprs[14], 16);
 
 	if (need_resched())
@@ -729,7 +851,7 @@ static int vcpu_pre_run(struct kvm_vcpu *vcpu)
 
 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
 {
-	int rc;
+	int rc = -1;
 
 	VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
 		   vcpu->arch.sie_block->icptcode);
@@ -743,7 +865,16 @@ static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
 						current->thread.gmap_addr;
 		vcpu->run->s390_ucontrol.pgm_code = 0x10;
 		rc = -EREMOTE;
-	} else {
+
+	} else if (current->thread.gmap_pfault) {
+		trace_kvm_s390_major_guest_pfault(vcpu);
+		current->thread.gmap_pfault = 0;
+		if (kvm_arch_setup_async_pf(vcpu) ||
+		    (kvm_arch_fault_in_sync(vcpu) >= 0))
+			rc = 0;
+	}
+
+	if (rc == -1) {
 		VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
 		trace_kvm_s390_sie_fault(vcpu);
 		rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);

arch/s390/kvm/kvm-s390.h

@@ -159,4 +159,8 @@ void exit_sie_sync(struct kvm_vcpu *vcpu);
 /* implemented in diag.c */
 int kvm_s390_handle_diag(struct kvm_vcpu *vcpu);
 
+/* implemented in interrupt.c */
+int kvm_cpu_has_interrupt(struct kvm_vcpu *vcpu);
+int psw_extint_disabled(struct kvm_vcpu *vcpu);
+
 #endif

arch/s390/kvm/sigp.c

@@ -224,6 +224,8 @@ static int __sigp_stop(struct kvm_vcpu *vcpu, u16 cpu_addr, int action)
 static int __sigp_set_arch(struct kvm_vcpu *vcpu, u32 parameter)
 {
 	int rc;
+	unsigned int i;
+	struct kvm_vcpu *v;
 
 	switch (parameter & 0xff) {
 	case 0:
@@ -231,6 +233,11 @@ static int __sigp_set_arch(struct kvm_vcpu *vcpu, u32 parameter)
 		break;
 	case 1:
 	case 2:
+		kvm_for_each_vcpu(i, v, vcpu->kvm) {
+			v->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
+			kvm_clear_async_pf_completion_queue(v);
+		}
+
 		rc = SIGP_CC_ORDER_CODE_ACCEPTED;
 		break;
 	default:

arch/s390/kvm/trace.h

@@ -30,6 +30,52 @@
 	TP_printk("%02d[%016lx-%016lx]: " p_str, __entry->id,		\
 		  __entry->pswmask, __entry->pswaddr, p_args)
 
+TRACE_EVENT(kvm_s390_major_guest_pfault,
+	    TP_PROTO(VCPU_PROTO_COMMON),
+	    TP_ARGS(VCPU_ARGS_COMMON),
+
+	    TP_STRUCT__entry(
+		    VCPU_FIELD_COMMON
+		    ),
+
+	    TP_fast_assign(
+		    VCPU_ASSIGN_COMMON
+		    ),
+	    VCPU_TP_PRINTK("%s", "major fault, maybe applicable for pfault")
+	);
+
+TRACE_EVENT(kvm_s390_pfault_init,
+	    TP_PROTO(VCPU_PROTO_COMMON, long pfault_token),
+	    TP_ARGS(VCPU_ARGS_COMMON, pfault_token),
+
+	    TP_STRUCT__entry(
+		    VCPU_FIELD_COMMON
+		    __field(long, pfault_token)
+		    ),
+
+	    TP_fast_assign(
+		    VCPU_ASSIGN_COMMON
+		    __entry->pfault_token = pfault_token;
+		    ),
+	    VCPU_TP_PRINTK("init pfault token %ld", __entry->pfault_token)
+	);
+
+TRACE_EVENT(kvm_s390_pfault_done,
+	    TP_PROTO(VCPU_PROTO_COMMON, long pfault_token),
+	    TP_ARGS(VCPU_ARGS_COMMON, pfault_token),
+
+	    TP_STRUCT__entry(
+		    VCPU_FIELD_COMMON
+		    __field(long, pfault_token)
+		    ),
+
+	    TP_fast_assign(
+		    VCPU_ASSIGN_COMMON
+		    __entry->pfault_token = pfault_token;
+		    ),
+	    VCPU_TP_PRINTK("done pfault token %ld", __entry->pfault_token)
+	);
+
 /*
  * Tracepoints for SIE entry and exit.
  */

arch/s390/mm/fault.c

@@ -50,6 +50,7 @@
 #define VM_FAULT_BADMAP		0x020000
 #define VM_FAULT_BADACCESS	0x040000
 #define VM_FAULT_SIGNAL		0x080000
+#define VM_FAULT_PFAULT		0x100000
 
 static unsigned long store_indication __read_mostly;
 
@@ -227,6 +228,7 @@ static noinline void do_fault_error(struct pt_regs *regs, int fault)
 			return;
 		}
 	case VM_FAULT_BADCONTEXT:
+	case VM_FAULT_PFAULT:
 		do_no_context(regs);
 		break;
 	case VM_FAULT_SIGNAL:
@@ -264,6 +266,9 @@ static noinline void do_fault_error(struct pt_regs *regs, int fault)
  */
 static inline int do_exception(struct pt_regs *regs, int access)
 {
+#ifdef CONFIG_PGSTE
+	struct gmap *gmap;
+#endif
 	struct task_struct *tsk;
 	struct mm_struct *mm;
 	struct vm_area_struct *vma;
@@ -304,9 +309,10 @@ static inline int do_exception(struct pt_regs *regs, int access)
 	down_read(&mm->mmap_sem);
 
 #ifdef CONFIG_PGSTE
-	if ((current->flags & PF_VCPU) && S390_lowcore.gmap) {
-		address = __gmap_fault(address,
-				     (struct gmap *) S390_lowcore.gmap);
+	gmap = (struct gmap *)
+		((current->flags & PF_VCPU) ? S390_lowcore.gmap : 0);
+	if (gmap) {
+		address = __gmap_fault(address, gmap);
 		if (address == -EFAULT) {
 			fault = VM_FAULT_BADMAP;
 			goto out_up;
@@ -315,6 +321,8 @@ static inline int do_exception(struct pt_regs *regs, int access)
 			fault = VM_FAULT_OOM;
 			goto out_up;
 		}
+		if (gmap->pfault_enabled)
+			flags |= FAULT_FLAG_RETRY_NOWAIT;
 	}
 #endif
 
@@ -371,9 +379,19 @@ static inline int do_exception(struct pt_regs *regs, int access)
 				      regs, address);
 		}
 		if (fault & VM_FAULT_RETRY) {
+#ifdef CONFIG_PGSTE
+			if (gmap && (flags & FAULT_FLAG_RETRY_NOWAIT)) {
+				/* FAULT_FLAG_RETRY_NOWAIT has been set,
+				 * mmap_sem has not been released */
+				current->thread.gmap_pfault = 1;
+				fault = VM_FAULT_PFAULT;
+				goto out_up;
+			}
+#endif
 			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
 			 * of starvation. */
-			flags &= ~FAULT_FLAG_ALLOW_RETRY;
+			flags &= ~(FAULT_FLAG_ALLOW_RETRY |
+				   FAULT_FLAG_RETRY_NOWAIT);
 			flags |= FAULT_FLAG_TRIED;
 			down_read(&mm->mmap_sem);
 			goto retry;

arch/x86/kvm/mmu.c

@@ -3328,7 +3328,7 @@ static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn)
 	arch.direct_map = vcpu->arch.mmu.direct_map;
 	arch.cr3 = vcpu->arch.mmu.get_cr3(vcpu);
 
-	return kvm_setup_async_pf(vcpu, gva, gfn, &arch);
+	return kvm_setup_async_pf(vcpu, gva, gfn_to_hva(vcpu->kvm, gfn), &arch);
 }
 
 static bool can_do_async_pf(struct kvm_vcpu *vcpu)

include/linux/kvm_host.h

@@ -192,7 +192,7 @@ struct kvm_async_pf {
 
 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu);
 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu);
-int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn,
+int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, unsigned long hva,
 		       struct kvm_arch_async_pf *arch);
 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu);
 #endif
@@ -1064,6 +1064,7 @@ extern struct kvm_device_ops kvm_mpic_ops;
 extern struct kvm_device_ops kvm_xics_ops;
 extern struct kvm_device_ops kvm_vfio_ops;
 extern struct kvm_device_ops kvm_arm_vgic_v2_ops;
+extern struct kvm_device_ops kvm_flic_ops;
 
 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
 

include/uapi/linux/kvm.h

@@ -413,6 +413,8 @@ struct kvm_s390_psw {
 #define KVM_S390_PROGRAM_INT		0xfffe0001u
 #define KVM_S390_SIGP_SET_PREFIX	0xfffe0002u
 #define KVM_S390_RESTART		0xfffe0003u
+#define KVM_S390_INT_PFAULT_INIT	0xfffe0004u
+#define KVM_S390_INT_PFAULT_DONE	0xfffe0005u
 #define KVM_S390_MCHK			0xfffe1000u
 #define KVM_S390_INT_VIRTIO		0xffff2603u
 #define KVM_S390_INT_SERVICE		0xffff2401u
@@ -434,6 +436,69 @@ struct kvm_s390_interrupt {
 	__u64 parm64;
 };
 
+struct kvm_s390_io_info {
+	__u16 subchannel_id;
+	__u16 subchannel_nr;
+	__u32 io_int_parm;
+	__u32 io_int_word;
+};
+
+struct kvm_s390_ext_info {
+	__u32 ext_params;
+	__u32 pad;
+	__u64 ext_params2;
+};
+
+struct kvm_s390_pgm_info {
+	__u64 trans_exc_code;
+	__u64 mon_code;
+	__u64 per_address;
+	__u32 data_exc_code;
+	__u16 code;
+	__u16 mon_class_nr;
+	__u8 per_code;
+	__u8 per_atmid;
+	__u8 exc_access_id;
+	__u8 per_access_id;
+	__u8 op_access_id;
+	__u8 pad[3];
+};
+
+struct kvm_s390_prefix_info {
+	__u32 address;
+};
+
+struct kvm_s390_extcall_info {
+	__u16 code;
+};
+
+struct kvm_s390_emerg_info {
+	__u16 code;
+};
+
+struct kvm_s390_mchk_info {
+	__u64 cr14;
+	__u64 mcic;
+	__u64 failing_storage_address;
+	__u32 ext_damage_code;
+	__u32 pad;
+	__u8 fixed_logout[16];
+};
+
+struct kvm_s390_irq {
+	__u64 type;
+	union {
+		struct kvm_s390_io_info io;
+		struct kvm_s390_ext_info ext;
+		struct kvm_s390_pgm_info pgm;
+		struct kvm_s390_emerg_info emerg;
+		struct kvm_s390_extcall_info extcall;
+		struct kvm_s390_prefix_info prefix;
+		struct kvm_s390_mchk_info mchk;
+		char reserved[64];
+	} u;
+};
+
 /* for KVM_SET_GUEST_DEBUG */
 
 #define KVM_GUESTDBG_ENABLE		0x00000001
@@ -855,6 +920,7 @@ struct kvm_device_attr {
 #define   KVM_DEV_VFIO_GROUP_ADD			1
 #define   KVM_DEV_VFIO_GROUP_DEL			2
 #define KVM_DEV_TYPE_ARM_VGIC_V2	5
+#define KVM_DEV_TYPE_FLIC		6
 
 /*
  * ioctls for VM fds

virt/kvm/Kconfig

@@ -22,6 +22,10 @@ config KVM_MMIO
 config KVM_ASYNC_PF
        bool
 
+# Toggle to switch between direct notification and batch job
+config KVM_ASYNC_PF_SYNC
+       bool
+
 config HAVE_KVM_MSI
        bool
 

virt/kvm/async_pf.c

@@ -28,6 +28,21 @@
 #include "async_pf.h"
 #include <trace/events/kvm.h>
 
+static inline void kvm_async_page_present_sync(struct kvm_vcpu *vcpu,
+					       struct kvm_async_pf *work)
+{
+#ifdef CONFIG_KVM_ASYNC_PF_SYNC
+	kvm_arch_async_page_present(vcpu, work);
+#endif
+}
+static inline void kvm_async_page_present_async(struct kvm_vcpu *vcpu,
+						struct kvm_async_pf *work)
+{
+#ifndef CONFIG_KVM_ASYNC_PF_SYNC
+	kvm_arch_async_page_present(vcpu, work);
+#endif
+}
+
 static struct kmem_cache *async_pf_cache;
 
 int kvm_async_pf_init(void)
@@ -69,6 +84,7 @@ static void async_pf_execute(struct work_struct *work)
 	down_read(&mm->mmap_sem);
 	get_user_pages(current, mm, addr, 1, 1, 0, NULL, NULL);
 	up_read(&mm->mmap_sem);
+	kvm_async_page_present_sync(vcpu, apf);
 	unuse_mm(mm);
 
 	spin_lock(&vcpu->async_pf.lock);
@@ -97,11 +113,16 @@ void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu)
 			list_entry(vcpu->async_pf.queue.next,
 				   typeof(*work), queue);
 		list_del(&work->queue);
+
+#ifdef CONFIG_KVM_ASYNC_PF_SYNC
+		flush_work(&work->work);
+#else
 		if (cancel_work_sync(&work->work)) {
 			mmdrop(work->mm);
 			kvm_put_kvm(vcpu->kvm); /* == work->vcpu->kvm */
 			kmem_cache_free(async_pf_cache, work);
 		}
+#endif
 	}
 
 	spin_lock(&vcpu->async_pf.lock);
@@ -138,7 +159,7 @@ void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
 	}
 }
 
-int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn,
+int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, unsigned long hva,
 		       struct kvm_arch_async_pf *arch)
 {
 	struct kvm_async_pf *work;
@@ -159,7 +180,7 @@ int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn,
 	work->wakeup_all = false;
 	work->vcpu = vcpu;
 	work->gva = gva;
-	work->addr = gfn_to_hva(vcpu->kvm, gfn);
+	work->addr = hva;
 	work->arch = *arch;
 	work->mm = current->mm;
 	atomic_inc(&work->mm->mm_count);

virt/kvm/kvm_main.c

@@ -2283,6 +2283,11 @@ static int kvm_ioctl_create_device(struct kvm *kvm,
 	case KVM_DEV_TYPE_ARM_VGIC_V2:
 		ops = &kvm_arm_vgic_v2_ops;
 		break;
+#endif
+#ifdef CONFIG_S390
+	case KVM_DEV_TYPE_FLIC:
+		ops = &kvm_flic_ops;
+		break;
 #endif
 	default:
 		return -ENODEV;