2 * hosting zSeries kernel virtual machines
4 * Copyright IBM Corp. 2008, 2009
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License (version 2 only)
8 * as published by the Free Software Foundation.
10 * Author(s): Carsten Otte <cotte@de.ibm.com>
11 * Christian Borntraeger <borntraeger@de.ibm.com>
12 * Heiko Carstens <heiko.carstens@de.ibm.com>
13 * Christian Ehrhardt <ehrhardt@de.ibm.com>
14 * Jason J. Herne <jjherne@us.ibm.com>
17 #include <linux/compiler.h>
18 #include <linux/err.h>
20 #include <linux/hrtimer.h>
21 #include <linux/init.h>
22 #include <linux/kvm.h>
23 #include <linux/kvm_host.h>
24 #include <linux/module.h>
25 #include <linux/slab.h>
26 #include <linux/timer.h>
27 #include <asm/asm-offsets.h>
28 #include <asm/lowcore.h>
29 #include <asm/pgtable.h>
31 #include <asm/switch_to.h>
32 #include <asm/facility.h>
37 #define CREATE_TRACE_POINTS
39 #include "trace-s390.h"
41 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
43 struct kvm_stats_debugfs_item debugfs_entries[] = {
44 { "userspace_handled", VCPU_STAT(exit_userspace) },
45 { "exit_null", VCPU_STAT(exit_null) },
46 { "exit_validity", VCPU_STAT(exit_validity) },
47 { "exit_stop_request", VCPU_STAT(exit_stop_request) },
48 { "exit_external_request", VCPU_STAT(exit_external_request) },
49 { "exit_external_interrupt", VCPU_STAT(exit_external_interrupt) },
50 { "exit_instruction", VCPU_STAT(exit_instruction) },
51 { "exit_program_interruption", VCPU_STAT(exit_program_interruption) },
52 { "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program) },
53 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
54 { "instruction_lctlg", VCPU_STAT(instruction_lctlg) },
55 { "instruction_lctl", VCPU_STAT(instruction_lctl) },
56 { "instruction_stctl", VCPU_STAT(instruction_stctl) },
57 { "instruction_stctg", VCPU_STAT(instruction_stctg) },
58 { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal) },
59 { "deliver_external_call", VCPU_STAT(deliver_external_call) },
60 { "deliver_service_signal", VCPU_STAT(deliver_service_signal) },
61 { "deliver_virtio_interrupt", VCPU_STAT(deliver_virtio_interrupt) },
62 { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal) },
63 { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal) },
64 { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal) },
65 { "deliver_program_interruption", VCPU_STAT(deliver_program_int) },
66 { "exit_wait_state", VCPU_STAT(exit_wait_state) },
67 { "instruction_pfmf", VCPU_STAT(instruction_pfmf) },
68 { "instruction_stidp", VCPU_STAT(instruction_stidp) },
69 { "instruction_spx", VCPU_STAT(instruction_spx) },
70 { "instruction_stpx", VCPU_STAT(instruction_stpx) },
71 { "instruction_stap", VCPU_STAT(instruction_stap) },
72 { "instruction_storage_key", VCPU_STAT(instruction_storage_key) },
73 { "instruction_ipte_interlock", VCPU_STAT(instruction_ipte_interlock) },
74 { "instruction_stsch", VCPU_STAT(instruction_stsch) },
75 { "instruction_chsc", VCPU_STAT(instruction_chsc) },
76 { "instruction_essa", VCPU_STAT(instruction_essa) },
77 { "instruction_stsi", VCPU_STAT(instruction_stsi) },
78 { "instruction_stfl", VCPU_STAT(instruction_stfl) },
79 { "instruction_tprot", VCPU_STAT(instruction_tprot) },
80 { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense) },
81 { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running) },
82 { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call) },
83 { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency) },
84 { "instruction_sigp_cond_emergency", VCPU_STAT(instruction_sigp_cond_emergency) },
85 { "instruction_sigp_start", VCPU_STAT(instruction_sigp_start) },
86 { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop) },
87 { "instruction_sigp_stop_store_status", VCPU_STAT(instruction_sigp_stop_store_status) },
88 { "instruction_sigp_store_status", VCPU_STAT(instruction_sigp_store_status) },
89 { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch) },
90 { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix) },
91 { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart) },
92 { "instruction_sigp_cpu_reset", VCPU_STAT(instruction_sigp_cpu_reset) },
93 { "instruction_sigp_init_cpu_reset", VCPU_STAT(instruction_sigp_init_cpu_reset) },
94 { "instruction_sigp_unknown", VCPU_STAT(instruction_sigp_unknown) },
95 { "diagnose_10", VCPU_STAT(diagnose_10) },
96 { "diagnose_44", VCPU_STAT(diagnose_44) },
97 { "diagnose_9c", VCPU_STAT(diagnose_9c) },
101 unsigned long *vfacilities;
102 static struct gmap_notifier gmap_notifier;
104 /* test availability of vfacility */
105 int test_vfacility(unsigned long nr)
107 return __test_facility(nr, (void *) vfacilities);
110 /* Section: not file related */
111 int kvm_arch_hardware_enable(void)
113 /* every s390 is virtualization enabled ;-) */
117 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long address);
119 int kvm_arch_hardware_setup(void)
121 gmap_notifier.notifier_call = kvm_gmap_notifier;
122 gmap_register_ipte_notifier(&gmap_notifier);
126 void kvm_arch_hardware_unsetup(void)
128 gmap_unregister_ipte_notifier(&gmap_notifier);
131 int kvm_arch_init(void *opaque)
133 /* Register floating interrupt controller interface. */
134 return kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
137 /* Section: device related */
138 long kvm_arch_dev_ioctl(struct file *filp,
139 unsigned int ioctl, unsigned long arg)
141 if (ioctl == KVM_S390_ENABLE_SIE)
142 return s390_enable_sie();
146 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
151 case KVM_CAP_S390_PSW:
152 case KVM_CAP_S390_GMAP:
153 case KVM_CAP_SYNC_MMU:
154 #ifdef CONFIG_KVM_S390_UCONTROL
155 case KVM_CAP_S390_UCONTROL:
157 case KVM_CAP_ASYNC_PF:
158 case KVM_CAP_SYNC_REGS:
159 case KVM_CAP_ONE_REG:
160 case KVM_CAP_ENABLE_CAP:
161 case KVM_CAP_S390_CSS_SUPPORT:
163 case KVM_CAP_IOEVENTFD:
164 case KVM_CAP_DEVICE_CTRL:
165 case KVM_CAP_ENABLE_CAP_VM:
166 case KVM_CAP_S390_IRQCHIP:
167 case KVM_CAP_VM_ATTRIBUTES:
168 case KVM_CAP_MP_STATE:
171 case KVM_CAP_NR_VCPUS:
172 case KVM_CAP_MAX_VCPUS:
175 case KVM_CAP_NR_MEMSLOTS:
176 r = KVM_USER_MEM_SLOTS;
178 case KVM_CAP_S390_COW:
179 r = MACHINE_HAS_ESOP;
187 static void kvm_s390_sync_dirty_log(struct kvm *kvm,
188 struct kvm_memory_slot *memslot)
190 gfn_t cur_gfn, last_gfn;
191 unsigned long address;
192 struct gmap *gmap = kvm->arch.gmap;
194 down_read(&gmap->mm->mmap_sem);
195 /* Loop over all guest pages */
196 last_gfn = memslot->base_gfn + memslot->npages;
197 for (cur_gfn = memslot->base_gfn; cur_gfn <= last_gfn; cur_gfn++) {
198 address = gfn_to_hva_memslot(memslot, cur_gfn);
200 if (gmap_test_and_clear_dirty(address, gmap))
201 mark_page_dirty(kvm, cur_gfn);
203 up_read(&gmap->mm->mmap_sem);
206 /* Section: vm related */
208 * Get (and clear) the dirty memory log for a memory slot.
210 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
211 struct kvm_dirty_log *log)
215 struct kvm_memory_slot *memslot;
218 mutex_lock(&kvm->slots_lock);
221 if (log->slot >= KVM_USER_MEM_SLOTS)
224 memslot = id_to_memslot(kvm->memslots, log->slot);
226 if (!memslot->dirty_bitmap)
229 kvm_s390_sync_dirty_log(kvm, memslot);
230 r = kvm_get_dirty_log(kvm, log, &is_dirty);
234 /* Clear the dirty log */
236 n = kvm_dirty_bitmap_bytes(memslot);
237 memset(memslot->dirty_bitmap, 0, n);
241 mutex_unlock(&kvm->slots_lock);
245 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
253 case KVM_CAP_S390_IRQCHIP:
254 kvm->arch.use_irqchip = 1;
264 static int kvm_s390_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
268 switch (attr->attr) {
269 case KVM_S390_VM_MEM_ENABLE_CMMA:
271 mutex_lock(&kvm->lock);
272 if (atomic_read(&kvm->online_vcpus) == 0) {
273 kvm->arch.use_cmma = 1;
276 mutex_unlock(&kvm->lock);
278 case KVM_S390_VM_MEM_CLR_CMMA:
279 mutex_lock(&kvm->lock);
280 idx = srcu_read_lock(&kvm->srcu);
281 s390_reset_cmma(kvm->arch.gmap->mm);
282 srcu_read_unlock(&kvm->srcu, idx);
283 mutex_unlock(&kvm->lock);
293 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
297 switch (attr->group) {
298 case KVM_S390_VM_MEM_CTRL:
299 ret = kvm_s390_mem_control(kvm, attr);
309 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
314 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
318 switch (attr->group) {
319 case KVM_S390_VM_MEM_CTRL:
320 switch (attr->attr) {
321 case KVM_S390_VM_MEM_ENABLE_CMMA:
322 case KVM_S390_VM_MEM_CLR_CMMA:
338 long kvm_arch_vm_ioctl(struct file *filp,
339 unsigned int ioctl, unsigned long arg)
341 struct kvm *kvm = filp->private_data;
342 void __user *argp = (void __user *)arg;
343 struct kvm_device_attr attr;
347 case KVM_S390_INTERRUPT: {
348 struct kvm_s390_interrupt s390int;
351 if (copy_from_user(&s390int, argp, sizeof(s390int)))
353 r = kvm_s390_inject_vm(kvm, &s390int);
356 case KVM_ENABLE_CAP: {
357 struct kvm_enable_cap cap;
359 if (copy_from_user(&cap, argp, sizeof(cap)))
361 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
364 case KVM_CREATE_IRQCHIP: {
365 struct kvm_irq_routing_entry routing;
368 if (kvm->arch.use_irqchip) {
369 /* Set up dummy routing. */
370 memset(&routing, 0, sizeof(routing));
371 kvm_set_irq_routing(kvm, &routing, 0, 0);
376 case KVM_SET_DEVICE_ATTR: {
378 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
380 r = kvm_s390_vm_set_attr(kvm, &attr);
383 case KVM_GET_DEVICE_ATTR: {
385 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
387 r = kvm_s390_vm_get_attr(kvm, &attr);
390 case KVM_HAS_DEVICE_ATTR: {
392 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
394 r = kvm_s390_vm_has_attr(kvm, &attr);
404 static int kvm_s390_crypto_init(struct kvm *kvm)
406 if (!test_vfacility(76))
409 kvm->arch.crypto.crycb = kzalloc(sizeof(*kvm->arch.crypto.crycb),
410 GFP_KERNEL | GFP_DMA);
411 if (!kvm->arch.crypto.crycb)
414 kvm->arch.crypto.crycbd = (__u32) (unsigned long) kvm->arch.crypto.crycb |
420 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
424 static unsigned long sca_offset;
427 #ifdef CONFIG_KVM_S390_UCONTROL
428 if (type & ~KVM_VM_S390_UCONTROL)
430 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
437 rc = s390_enable_sie();
443 kvm->arch.sca = (struct sca_block *) get_zeroed_page(GFP_KERNEL);
446 spin_lock(&kvm_lock);
447 sca_offset = (sca_offset + 16) & 0x7f0;
448 kvm->arch.sca = (struct sca_block *) ((char *) kvm->arch.sca + sca_offset);
449 spin_unlock(&kvm_lock);
451 sprintf(debug_name, "kvm-%u", current->pid);
453 kvm->arch.dbf = debug_register(debug_name, 8, 2, 8 * sizeof(long));
457 if (kvm_s390_crypto_init(kvm) < 0)
460 spin_lock_init(&kvm->arch.float_int.lock);
461 INIT_LIST_HEAD(&kvm->arch.float_int.list);
462 init_waitqueue_head(&kvm->arch.ipte_wq);
463 mutex_init(&kvm->arch.ipte_mutex);
465 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
466 VM_EVENT(kvm, 3, "%s", "vm created");
468 if (type & KVM_VM_S390_UCONTROL) {
469 kvm->arch.gmap = NULL;
471 kvm->arch.gmap = gmap_alloc(current->mm, (1UL << 44) - 1);
474 kvm->arch.gmap->private = kvm;
475 kvm->arch.gmap->pfault_enabled = 0;
478 kvm->arch.css_support = 0;
479 kvm->arch.use_irqchip = 0;
481 spin_lock_init(&kvm->arch.start_stop_lock);
485 kfree(kvm->arch.crypto.crycb);
487 debug_unregister(kvm->arch.dbf);
489 free_page((unsigned long)(kvm->arch.sca));
494 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
496 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
497 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
498 kvm_s390_clear_local_irqs(vcpu);
499 kvm_clear_async_pf_completion_queue(vcpu);
500 if (!kvm_is_ucontrol(vcpu->kvm)) {
501 clear_bit(63 - vcpu->vcpu_id,
502 (unsigned long *) &vcpu->kvm->arch.sca->mcn);
503 if (vcpu->kvm->arch.sca->cpu[vcpu->vcpu_id].sda ==
504 (__u64) vcpu->arch.sie_block)
505 vcpu->kvm->arch.sca->cpu[vcpu->vcpu_id].sda = 0;
509 if (kvm_is_ucontrol(vcpu->kvm))
510 gmap_free(vcpu->arch.gmap);
512 if (kvm_s390_cmma_enabled(vcpu->kvm))
513 kvm_s390_vcpu_unsetup_cmma(vcpu);
514 free_page((unsigned long)(vcpu->arch.sie_block));
516 kvm_vcpu_uninit(vcpu);
517 kmem_cache_free(kvm_vcpu_cache, vcpu);
520 static void kvm_free_vcpus(struct kvm *kvm)
523 struct kvm_vcpu *vcpu;
525 kvm_for_each_vcpu(i, vcpu, kvm)
526 kvm_arch_vcpu_destroy(vcpu);
528 mutex_lock(&kvm->lock);
529 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
530 kvm->vcpus[i] = NULL;
532 atomic_set(&kvm->online_vcpus, 0);
533 mutex_unlock(&kvm->lock);
536 void kvm_arch_destroy_vm(struct kvm *kvm)
539 free_page((unsigned long)(kvm->arch.sca));
540 debug_unregister(kvm->arch.dbf);
541 kfree(kvm->arch.crypto.crycb);
542 if (!kvm_is_ucontrol(kvm))
543 gmap_free(kvm->arch.gmap);
544 kvm_s390_destroy_adapters(kvm);
545 kvm_s390_clear_float_irqs(kvm);
548 /* Section: vcpu related */
549 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
551 vcpu->arch.gmap = gmap_alloc(current->mm, -1UL);
552 if (!vcpu->arch.gmap)
554 vcpu->arch.gmap->private = vcpu->kvm;
559 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
561 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
562 kvm_clear_async_pf_completion_queue(vcpu);
563 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
570 if (kvm_is_ucontrol(vcpu->kvm))
571 return __kvm_ucontrol_vcpu_init(vcpu);
576 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
578 save_fp_ctl(&vcpu->arch.host_fpregs.fpc);
579 save_fp_regs(vcpu->arch.host_fpregs.fprs);
580 save_access_regs(vcpu->arch.host_acrs);
581 restore_fp_ctl(&vcpu->arch.guest_fpregs.fpc);
582 restore_fp_regs(vcpu->arch.guest_fpregs.fprs);
583 restore_access_regs(vcpu->run->s.regs.acrs);
584 gmap_enable(vcpu->arch.gmap);
585 atomic_set_mask(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
588 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
590 atomic_clear_mask(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
591 gmap_disable(vcpu->arch.gmap);
592 save_fp_ctl(&vcpu->arch.guest_fpregs.fpc);
593 save_fp_regs(vcpu->arch.guest_fpregs.fprs);
594 save_access_regs(vcpu->run->s.regs.acrs);
595 restore_fp_ctl(&vcpu->arch.host_fpregs.fpc);
596 restore_fp_regs(vcpu->arch.host_fpregs.fprs);
597 restore_access_regs(vcpu->arch.host_acrs);
600 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
602 /* this equals initial cpu reset in pop, but we don't switch to ESA */
603 vcpu->arch.sie_block->gpsw.mask = 0UL;
604 vcpu->arch.sie_block->gpsw.addr = 0UL;
605 kvm_s390_set_prefix(vcpu, 0);
606 vcpu->arch.sie_block->cputm = 0UL;
607 vcpu->arch.sie_block->ckc = 0UL;
608 vcpu->arch.sie_block->todpr = 0;
609 memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64));
610 vcpu->arch.sie_block->gcr[0] = 0xE0UL;
611 vcpu->arch.sie_block->gcr[14] = 0xC2000000UL;
612 vcpu->arch.guest_fpregs.fpc = 0;
613 asm volatile("lfpc %0" : : "Q" (vcpu->arch.guest_fpregs.fpc));
614 vcpu->arch.sie_block->gbea = 1;
615 vcpu->arch.sie_block->pp = 0;
616 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
617 kvm_clear_async_pf_completion_queue(vcpu);
618 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
619 kvm_s390_vcpu_stop(vcpu);
620 kvm_s390_clear_local_irqs(vcpu);
623 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
625 if (!kvm_is_ucontrol(vcpu->kvm))
626 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
629 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
631 if (!test_vfacility(76))
634 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
637 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
639 free_page(vcpu->arch.sie_block->cbrlo);
640 vcpu->arch.sie_block->cbrlo = 0;
643 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
645 vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
646 if (!vcpu->arch.sie_block->cbrlo)
649 vcpu->arch.sie_block->ecb2 |= 0x80;
650 vcpu->arch.sie_block->ecb2 &= ~0x08;
654 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
658 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
662 vcpu->arch.sie_block->ecb = 6;
663 if (test_vfacility(50) && test_vfacility(73))
664 vcpu->arch.sie_block->ecb |= 0x10;
666 vcpu->arch.sie_block->ecb2 = 8;
667 vcpu->arch.sie_block->eca = 0xD1002000U;
669 vcpu->arch.sie_block->eca |= 1;
670 vcpu->arch.sie_block->fac = (int) (long) vfacilities;
671 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE |
674 if (kvm_s390_cmma_enabled(vcpu->kvm)) {
675 rc = kvm_s390_vcpu_setup_cmma(vcpu);
679 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
680 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
681 get_cpu_id(&vcpu->arch.cpu_id);
682 vcpu->arch.cpu_id.version = 0xff;
684 kvm_s390_vcpu_crypto_setup(vcpu);
689 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
692 struct kvm_vcpu *vcpu;
693 struct sie_page *sie_page;
696 if (id >= KVM_MAX_VCPUS)
701 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
705 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
709 vcpu->arch.sie_block = &sie_page->sie_block;
710 vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
712 vcpu->arch.sie_block->icpua = id;
713 if (!kvm_is_ucontrol(kvm)) {
714 if (!kvm->arch.sca) {
718 if (!kvm->arch.sca->cpu[id].sda)
719 kvm->arch.sca->cpu[id].sda =
720 (__u64) vcpu->arch.sie_block;
721 vcpu->arch.sie_block->scaoh =
722 (__u32)(((__u64)kvm->arch.sca) >> 32);
723 vcpu->arch.sie_block->scaol = (__u32)(__u64)kvm->arch.sca;
724 set_bit(63 - id, (unsigned long *) &kvm->arch.sca->mcn);
727 spin_lock_init(&vcpu->arch.local_int.lock);
728 vcpu->arch.local_int.float_int = &kvm->arch.float_int;
729 vcpu->arch.local_int.wq = &vcpu->wq;
730 vcpu->arch.local_int.cpuflags = &vcpu->arch.sie_block->cpuflags;
732 rc = kvm_vcpu_init(vcpu, kvm, id);
734 goto out_free_sie_block;
735 VM_EVENT(kvm, 3, "create cpu %d at %p, sie block at %p", id, vcpu,
736 vcpu->arch.sie_block);
737 trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
741 free_page((unsigned long)(vcpu->arch.sie_block));
743 kmem_cache_free(kvm_vcpu_cache, vcpu);
748 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
750 return kvm_cpu_has_interrupt(vcpu);
753 void s390_vcpu_block(struct kvm_vcpu *vcpu)
755 atomic_set_mask(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
758 void s390_vcpu_unblock(struct kvm_vcpu *vcpu)
760 atomic_clear_mask(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
764 * Kick a guest cpu out of SIE and wait until SIE is not running.
765 * If the CPU is not running (e.g. waiting as idle) the function will
766 * return immediately. */
767 void exit_sie(struct kvm_vcpu *vcpu)
769 atomic_set_mask(CPUSTAT_STOP_INT, &vcpu->arch.sie_block->cpuflags);
770 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
774 /* Kick a guest cpu out of SIE and prevent SIE-reentry */
775 void exit_sie_sync(struct kvm_vcpu *vcpu)
777 s390_vcpu_block(vcpu);
781 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long address)
784 struct kvm *kvm = gmap->private;
785 struct kvm_vcpu *vcpu;
787 kvm_for_each_vcpu(i, vcpu, kvm) {
788 /* match against both prefix pages */
789 if (kvm_s390_get_prefix(vcpu) == (address & ~0x1000UL)) {
790 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx", address);
791 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
797 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
799 /* kvm common code refers to this, but never calls it */
804 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
805 struct kvm_one_reg *reg)
810 case KVM_REG_S390_TODPR:
811 r = put_user(vcpu->arch.sie_block->todpr,
812 (u32 __user *)reg->addr);
814 case KVM_REG_S390_EPOCHDIFF:
815 r = put_user(vcpu->arch.sie_block->epoch,
816 (u64 __user *)reg->addr);
818 case KVM_REG_S390_CPU_TIMER:
819 r = put_user(vcpu->arch.sie_block->cputm,
820 (u64 __user *)reg->addr);
822 case KVM_REG_S390_CLOCK_COMP:
823 r = put_user(vcpu->arch.sie_block->ckc,
824 (u64 __user *)reg->addr);
826 case KVM_REG_S390_PFTOKEN:
827 r = put_user(vcpu->arch.pfault_token,
828 (u64 __user *)reg->addr);
830 case KVM_REG_S390_PFCOMPARE:
831 r = put_user(vcpu->arch.pfault_compare,
832 (u64 __user *)reg->addr);
834 case KVM_REG_S390_PFSELECT:
835 r = put_user(vcpu->arch.pfault_select,
836 (u64 __user *)reg->addr);
838 case KVM_REG_S390_PP:
839 r = put_user(vcpu->arch.sie_block->pp,
840 (u64 __user *)reg->addr);
842 case KVM_REG_S390_GBEA:
843 r = put_user(vcpu->arch.sie_block->gbea,
844 (u64 __user *)reg->addr);
853 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
854 struct kvm_one_reg *reg)
859 case KVM_REG_S390_TODPR:
860 r = get_user(vcpu->arch.sie_block->todpr,
861 (u32 __user *)reg->addr);
863 case KVM_REG_S390_EPOCHDIFF:
864 r = get_user(vcpu->arch.sie_block->epoch,
865 (u64 __user *)reg->addr);
867 case KVM_REG_S390_CPU_TIMER:
868 r = get_user(vcpu->arch.sie_block->cputm,
869 (u64 __user *)reg->addr);
871 case KVM_REG_S390_CLOCK_COMP:
872 r = get_user(vcpu->arch.sie_block->ckc,
873 (u64 __user *)reg->addr);
875 case KVM_REG_S390_PFTOKEN:
876 r = get_user(vcpu->arch.pfault_token,
877 (u64 __user *)reg->addr);
879 case KVM_REG_S390_PFCOMPARE:
880 r = get_user(vcpu->arch.pfault_compare,
881 (u64 __user *)reg->addr);
883 case KVM_REG_S390_PFSELECT:
884 r = get_user(vcpu->arch.pfault_select,
885 (u64 __user *)reg->addr);
887 case KVM_REG_S390_PP:
888 r = get_user(vcpu->arch.sie_block->pp,
889 (u64 __user *)reg->addr);
891 case KVM_REG_S390_GBEA:
892 r = get_user(vcpu->arch.sie_block->gbea,
893 (u64 __user *)reg->addr);
902 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
904 kvm_s390_vcpu_initial_reset(vcpu);
908 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
910 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs));
914 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
916 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
920 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
921 struct kvm_sregs *sregs)
923 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
924 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
925 restore_access_regs(vcpu->run->s.regs.acrs);
929 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
930 struct kvm_sregs *sregs)
932 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
933 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
937 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
939 if (test_fp_ctl(fpu->fpc))
941 memcpy(&vcpu->arch.guest_fpregs.fprs, &fpu->fprs, sizeof(fpu->fprs));
942 vcpu->arch.guest_fpregs.fpc = fpu->fpc;
943 restore_fp_ctl(&vcpu->arch.guest_fpregs.fpc);
944 restore_fp_regs(vcpu->arch.guest_fpregs.fprs);
948 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
950 memcpy(&fpu->fprs, &vcpu->arch.guest_fpregs.fprs, sizeof(fpu->fprs));
951 fpu->fpc = vcpu->arch.guest_fpregs.fpc;
955 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
959 if (!is_vcpu_stopped(vcpu))
962 vcpu->run->psw_mask = psw.mask;
963 vcpu->run->psw_addr = psw.addr;
968 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
969 struct kvm_translation *tr)
971 return -EINVAL; /* not implemented yet */
974 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
975 KVM_GUESTDBG_USE_HW_BP | \
978 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
979 struct kvm_guest_debug *dbg)
983 vcpu->guest_debug = 0;
984 kvm_s390_clear_bp_data(vcpu);
986 if (dbg->control & ~VALID_GUESTDBG_FLAGS)
989 if (dbg->control & KVM_GUESTDBG_ENABLE) {
990 vcpu->guest_debug = dbg->control;
991 /* enforce guest PER */
992 atomic_set_mask(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
994 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
995 rc = kvm_s390_import_bp_data(vcpu, dbg);
997 atomic_clear_mask(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
998 vcpu->arch.guestdbg.last_bp = 0;
1002 vcpu->guest_debug = 0;
1003 kvm_s390_clear_bp_data(vcpu);
1004 atomic_clear_mask(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
1010 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1011 struct kvm_mp_state *mp_state)
1013 /* CHECK_STOP and LOAD are not supported yet */
1014 return is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
1015 KVM_MP_STATE_OPERATING;
1018 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1019 struct kvm_mp_state *mp_state)
1023 /* user space knows about this interface - let it control the state */
1024 vcpu->kvm->arch.user_cpu_state_ctrl = 1;
1026 switch (mp_state->mp_state) {
1027 case KVM_MP_STATE_STOPPED:
1028 kvm_s390_vcpu_stop(vcpu);
1030 case KVM_MP_STATE_OPERATING:
1031 kvm_s390_vcpu_start(vcpu);
1033 case KVM_MP_STATE_LOAD:
1034 case KVM_MP_STATE_CHECK_STOP:
1035 /* fall through - CHECK_STOP and LOAD are not supported yet */
1043 bool kvm_s390_cmma_enabled(struct kvm *kvm)
1045 if (!MACHINE_IS_LPAR)
1047 /* only enable for z10 and later */
1048 if (!MACHINE_HAS_EDAT1)
1050 if (!kvm->arch.use_cmma)
1055 static bool ibs_enabled(struct kvm_vcpu *vcpu)
1057 return atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_IBS;
1060 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
1063 s390_vcpu_unblock(vcpu);
1065 * We use MMU_RELOAD just to re-arm the ipte notifier for the
1066 * guest prefix page. gmap_ipte_notify will wait on the ptl lock.
1067 * This ensures that the ipte instruction for this request has
1068 * already finished. We might race against a second unmapper that
1069 * wants to set the blocking bit. Lets just retry the request loop.
1071 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
1073 rc = gmap_ipte_notify(vcpu->arch.gmap,
1074 kvm_s390_get_prefix(vcpu),
1081 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
1082 vcpu->arch.sie_block->ihcpu = 0xffff;
1086 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
1087 if (!ibs_enabled(vcpu)) {
1088 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
1089 atomic_set_mask(CPUSTAT_IBS,
1090 &vcpu->arch.sie_block->cpuflags);
1095 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
1096 if (ibs_enabled(vcpu)) {
1097 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
1098 atomic_clear_mask(CPUSTAT_IBS,
1099 &vcpu->arch.sie_block->cpuflags);
1104 /* nothing to do, just clear the request */
1105 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
1111 * kvm_arch_fault_in_page - fault-in guest page if necessary
1112 * @vcpu: The corresponding virtual cpu
1113 * @gpa: Guest physical address
1114 * @writable: Whether the page should be writable or not
1116 * Make sure that a guest page has been faulted-in on the host.
1118 * Return: Zero on success, negative error code otherwise.
1120 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
1122 return gmap_fault(vcpu->arch.gmap, gpa,
1123 writable ? FAULT_FLAG_WRITE : 0);
1126 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
1127 unsigned long token)
1129 struct kvm_s390_interrupt inti;
1130 struct kvm_s390_irq irq;
1133 irq.u.ext.ext_params2 = token;
1134 irq.type = KVM_S390_INT_PFAULT_INIT;
1135 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
1137 inti.type = KVM_S390_INT_PFAULT_DONE;
1138 inti.parm64 = token;
1139 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
1143 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
1144 struct kvm_async_pf *work)
1146 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
1147 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
1150 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
1151 struct kvm_async_pf *work)
1153 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
1154 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
1157 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
1158 struct kvm_async_pf *work)
1160 /* s390 will always inject the page directly */
1163 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
1166 * s390 will always inject the page directly,
1167 * but we still want check_async_completion to cleanup
1172 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
1175 struct kvm_arch_async_pf arch;
1178 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
1180 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
1181 vcpu->arch.pfault_compare)
1183 if (psw_extint_disabled(vcpu))
1185 if (kvm_cpu_has_interrupt(vcpu))
1187 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
1189 if (!vcpu->arch.gmap->pfault_enabled)
1192 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
1193 hva += current->thread.gmap_addr & ~PAGE_MASK;
1194 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
1197 rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
1201 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
1206 * On s390 notifications for arriving pages will be delivered directly
1207 * to the guest but the house keeping for completed pfaults is
1208 * handled outside the worker.
1210 kvm_check_async_pf_completion(vcpu);
1212 memcpy(&vcpu->arch.sie_block->gg14, &vcpu->run->s.regs.gprs[14], 16);
1217 if (test_cpu_flag(CIF_MCCK_PENDING))
1220 if (!kvm_is_ucontrol(vcpu->kvm)) {
1221 rc = kvm_s390_deliver_pending_interrupts(vcpu);
1226 rc = kvm_s390_handle_requests(vcpu);
1230 if (guestdbg_enabled(vcpu)) {
1231 kvm_s390_backup_guest_per_regs(vcpu);
1232 kvm_s390_patch_guest_per_regs(vcpu);
1235 vcpu->arch.sie_block->icptcode = 0;
1236 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
1237 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
1238 trace_kvm_s390_sie_enter(vcpu, cpuflags);
1243 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
1247 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
1248 vcpu->arch.sie_block->icptcode);
1249 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
1251 if (guestdbg_enabled(vcpu))
1252 kvm_s390_restore_guest_per_regs(vcpu);
1254 if (exit_reason >= 0) {
1256 } else if (kvm_is_ucontrol(vcpu->kvm)) {
1257 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
1258 vcpu->run->s390_ucontrol.trans_exc_code =
1259 current->thread.gmap_addr;
1260 vcpu->run->s390_ucontrol.pgm_code = 0x10;
1263 } else if (current->thread.gmap_pfault) {
1264 trace_kvm_s390_major_guest_pfault(vcpu);
1265 current->thread.gmap_pfault = 0;
1266 if (kvm_arch_setup_async_pf(vcpu)) {
1269 gpa_t gpa = current->thread.gmap_addr;
1270 rc = kvm_arch_fault_in_page(vcpu, gpa, 1);
1275 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
1276 trace_kvm_s390_sie_fault(vcpu);
1277 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1280 memcpy(&vcpu->run->s.regs.gprs[14], &vcpu->arch.sie_block->gg14, 16);
1283 if (kvm_is_ucontrol(vcpu->kvm))
1284 /* Don't exit for host interrupts. */
1285 rc = vcpu->arch.sie_block->icptcode ? -EOPNOTSUPP : 0;
1287 rc = kvm_handle_sie_intercept(vcpu);
1293 static int __vcpu_run(struct kvm_vcpu *vcpu)
1295 int rc, exit_reason;
1298 * We try to hold kvm->srcu during most of vcpu_run (except when run-
1299 * ning the guest), so that memslots (and other stuff) are protected
1301 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1304 rc = vcpu_pre_run(vcpu);
1308 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1310 * As PF_VCPU will be used in fault handler, between
1311 * guest_enter and guest_exit should be no uaccess.
1316 exit_reason = sie64a(vcpu->arch.sie_block,
1317 vcpu->run->s.regs.gprs);
1319 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1321 rc = vcpu_post_run(vcpu, exit_reason);
1322 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
1324 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1328 static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1330 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
1331 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
1332 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
1333 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
1334 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
1335 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
1336 /* some control register changes require a tlb flush */
1337 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1339 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
1340 vcpu->arch.sie_block->cputm = kvm_run->s.regs.cputm;
1341 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
1342 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
1343 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
1344 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
1346 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
1347 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
1348 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
1349 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
1351 kvm_run->kvm_dirty_regs = 0;
1354 static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1356 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
1357 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
1358 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
1359 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
1360 kvm_run->s.regs.cputm = vcpu->arch.sie_block->cputm;
1361 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
1362 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
1363 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
1364 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
1365 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
1366 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
1367 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
1370 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1375 if (guestdbg_exit_pending(vcpu)) {
1376 kvm_s390_prepare_debug_exit(vcpu);
1380 if (vcpu->sigset_active)
1381 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1383 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
1384 kvm_s390_vcpu_start(vcpu);
1385 } else if (is_vcpu_stopped(vcpu)) {
1386 pr_err_ratelimited("kvm-s390: can't run stopped vcpu %d\n",
1391 sync_regs(vcpu, kvm_run);
1394 rc = __vcpu_run(vcpu);
1396 if (signal_pending(current) && !rc) {
1397 kvm_run->exit_reason = KVM_EXIT_INTR;
1401 if (guestdbg_exit_pending(vcpu) && !rc) {
1402 kvm_s390_prepare_debug_exit(vcpu);
1406 if (rc == -EOPNOTSUPP) {
1407 /* intercept cannot be handled in-kernel, prepare kvm-run */
1408 kvm_run->exit_reason = KVM_EXIT_S390_SIEIC;
1409 kvm_run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
1410 kvm_run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
1411 kvm_run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
1415 if (rc == -EREMOTE) {
1416 /* intercept was handled, but userspace support is needed
1417 * kvm_run has been prepared by the handler */
1421 store_regs(vcpu, kvm_run);
1423 if (vcpu->sigset_active)
1424 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1426 vcpu->stat.exit_userspace++;
1431 * store status at address
1432 * we use have two special cases:
1433 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
1434 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
1436 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
1438 unsigned char archmode = 1;
1443 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
1444 if (write_guest_abs(vcpu, 163, &archmode, 1))
1446 gpa = SAVE_AREA_BASE;
1447 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
1448 if (write_guest_real(vcpu, 163, &archmode, 1))
1450 gpa = kvm_s390_real_to_abs(vcpu, SAVE_AREA_BASE);
1452 rc = write_guest_abs(vcpu, gpa + offsetof(struct save_area, fp_regs),
1453 vcpu->arch.guest_fpregs.fprs, 128);
1454 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, gp_regs),
1455 vcpu->run->s.regs.gprs, 128);
1456 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, psw),
1457 &vcpu->arch.sie_block->gpsw, 16);
1458 px = kvm_s390_get_prefix(vcpu);
1459 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, pref_reg),
1461 rc |= write_guest_abs(vcpu,
1462 gpa + offsetof(struct save_area, fp_ctrl_reg),
1463 &vcpu->arch.guest_fpregs.fpc, 4);
1464 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, tod_reg),
1465 &vcpu->arch.sie_block->todpr, 4);
1466 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, timer),
1467 &vcpu->arch.sie_block->cputm, 8);
1468 clkcomp = vcpu->arch.sie_block->ckc >> 8;
1469 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, clk_cmp),
1471 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, acc_regs),
1472 &vcpu->run->s.regs.acrs, 64);
1473 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, ctrl_regs),
1474 &vcpu->arch.sie_block->gcr, 128);
1475 return rc ? -EFAULT : 0;
1478 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
1481 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
1482 * copying in vcpu load/put. Lets update our copies before we save
1483 * it into the save area
1485 save_fp_ctl(&vcpu->arch.guest_fpregs.fpc);
1486 save_fp_regs(vcpu->arch.guest_fpregs.fprs);
1487 save_access_regs(vcpu->run->s.regs.acrs);
1489 return kvm_s390_store_status_unloaded(vcpu, addr);
1492 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
1494 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
1495 kvm_make_request(KVM_REQ_DISABLE_IBS, vcpu);
1496 exit_sie_sync(vcpu);
1499 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
1502 struct kvm_vcpu *vcpu;
1504 kvm_for_each_vcpu(i, vcpu, kvm) {
1505 __disable_ibs_on_vcpu(vcpu);
1509 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
1511 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
1512 kvm_make_request(KVM_REQ_ENABLE_IBS, vcpu);
1513 exit_sie_sync(vcpu);
1516 void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
1518 int i, online_vcpus, started_vcpus = 0;
1520 if (!is_vcpu_stopped(vcpu))
1523 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
1524 /* Only one cpu at a time may enter/leave the STOPPED state. */
1525 spin_lock(&vcpu->kvm->arch.start_stop_lock);
1526 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
1528 for (i = 0; i < online_vcpus; i++) {
1529 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i]))
1533 if (started_vcpus == 0) {
1534 /* we're the only active VCPU -> speed it up */
1535 __enable_ibs_on_vcpu(vcpu);
1536 } else if (started_vcpus == 1) {
1538 * As we are starting a second VCPU, we have to disable
1539 * the IBS facility on all VCPUs to remove potentially
1540 * oustanding ENABLE requests.
1542 __disable_ibs_on_all_vcpus(vcpu->kvm);
1545 atomic_clear_mask(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
1547 * Another VCPU might have used IBS while we were offline.
1548 * Let's play safe and flush the VCPU at startup.
1550 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1551 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
1555 void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
1557 int i, online_vcpus, started_vcpus = 0;
1558 struct kvm_vcpu *started_vcpu = NULL;
1560 if (is_vcpu_stopped(vcpu))
1563 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
1564 /* Only one cpu at a time may enter/leave the STOPPED state. */
1565 spin_lock(&vcpu->kvm->arch.start_stop_lock);
1566 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
1568 /* Need to lock access to action_bits to avoid a SIGP race condition */
1569 spin_lock(&vcpu->arch.local_int.lock);
1570 atomic_set_mask(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
1572 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
1573 vcpu->arch.local_int.action_bits &=
1574 ~(ACTION_STOP_ON_STOP | ACTION_STORE_ON_STOP);
1575 spin_unlock(&vcpu->arch.local_int.lock);
1577 __disable_ibs_on_vcpu(vcpu);
1579 for (i = 0; i < online_vcpus; i++) {
1580 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) {
1582 started_vcpu = vcpu->kvm->vcpus[i];
1586 if (started_vcpus == 1) {
1588 * As we only have one VCPU left, we want to enable the
1589 * IBS facility for that VCPU to speed it up.
1591 __enable_ibs_on_vcpu(started_vcpu);
1594 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
1598 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
1599 struct kvm_enable_cap *cap)
1607 case KVM_CAP_S390_CSS_SUPPORT:
1608 if (!vcpu->kvm->arch.css_support) {
1609 vcpu->kvm->arch.css_support = 1;
1610 trace_kvm_s390_enable_css(vcpu->kvm);
1621 long kvm_arch_vcpu_ioctl(struct file *filp,
1622 unsigned int ioctl, unsigned long arg)
1624 struct kvm_vcpu *vcpu = filp->private_data;
1625 void __user *argp = (void __user *)arg;
1630 case KVM_S390_INTERRUPT: {
1631 struct kvm_s390_interrupt s390int;
1632 struct kvm_s390_irq s390irq;
1635 if (copy_from_user(&s390int, argp, sizeof(s390int)))
1637 if (s390int_to_s390irq(&s390int, &s390irq))
1639 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
1642 case KVM_S390_STORE_STATUS:
1643 idx = srcu_read_lock(&vcpu->kvm->srcu);
1644 r = kvm_s390_vcpu_store_status(vcpu, arg);
1645 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1647 case KVM_S390_SET_INITIAL_PSW: {
1651 if (copy_from_user(&psw, argp, sizeof(psw)))
1653 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
1656 case KVM_S390_INITIAL_RESET:
1657 r = kvm_arch_vcpu_ioctl_initial_reset(vcpu);
1659 case KVM_SET_ONE_REG:
1660 case KVM_GET_ONE_REG: {
1661 struct kvm_one_reg reg;
1663 if (copy_from_user(®, argp, sizeof(reg)))
1665 if (ioctl == KVM_SET_ONE_REG)
1666 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®);
1668 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®);
1671 #ifdef CONFIG_KVM_S390_UCONTROL
1672 case KVM_S390_UCAS_MAP: {
1673 struct kvm_s390_ucas_mapping ucasmap;
1675 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
1680 if (!kvm_is_ucontrol(vcpu->kvm)) {
1685 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
1686 ucasmap.vcpu_addr, ucasmap.length);
1689 case KVM_S390_UCAS_UNMAP: {
1690 struct kvm_s390_ucas_mapping ucasmap;
1692 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
1697 if (!kvm_is_ucontrol(vcpu->kvm)) {
1702 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
1707 case KVM_S390_VCPU_FAULT: {
1708 r = gmap_fault(vcpu->arch.gmap, arg, 0);
1711 case KVM_ENABLE_CAP:
1713 struct kvm_enable_cap cap;
1715 if (copy_from_user(&cap, argp, sizeof(cap)))
1717 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
1726 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
1728 #ifdef CONFIG_KVM_S390_UCONTROL
1729 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
1730 && (kvm_is_ucontrol(vcpu->kvm))) {
1731 vmf->page = virt_to_page(vcpu->arch.sie_block);
1732 get_page(vmf->page);
1736 return VM_FAULT_SIGBUS;
1739 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
1740 unsigned long npages)
1745 /* Section: memory related */
1746 int kvm_arch_prepare_memory_region(struct kvm *kvm,
1747 struct kvm_memory_slot *memslot,
1748 struct kvm_userspace_memory_region *mem,
1749 enum kvm_mr_change change)
1751 /* A few sanity checks. We can have memory slots which have to be
1752 located/ended at a segment boundary (1MB). The memory in userland is
1753 ok to be fragmented into various different vmas. It is okay to mmap()
1754 and munmap() stuff in this slot after doing this call at any time */
1756 if (mem->userspace_addr & 0xffffful)
1759 if (mem->memory_size & 0xffffful)
1765 void kvm_arch_commit_memory_region(struct kvm *kvm,
1766 struct kvm_userspace_memory_region *mem,
1767 const struct kvm_memory_slot *old,
1768 enum kvm_mr_change change)
1772 /* If the basics of the memslot do not change, we do not want
1773 * to update the gmap. Every update causes several unnecessary
1774 * segment translation exceptions. This is usually handled just
1775 * fine by the normal fault handler + gmap, but it will also
1776 * cause faults on the prefix page of running guest CPUs.
1778 if (old->userspace_addr == mem->userspace_addr &&
1779 old->base_gfn * PAGE_SIZE == mem->guest_phys_addr &&
1780 old->npages * PAGE_SIZE == mem->memory_size)
1783 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr,
1784 mem->guest_phys_addr, mem->memory_size);
1786 printk(KERN_WARNING "kvm-s390: failed to commit memory region\n");
1790 static int __init kvm_s390_init(void)
1793 ret = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1798 * guests can ask for up to 255+1 double words, we need a full page
1799 * to hold the maximum amount of facilities. On the other hand, we
1800 * only set facilities that are known to work in KVM.
1802 vfacilities = (unsigned long *) get_zeroed_page(GFP_KERNEL|GFP_DMA);
1807 memcpy(vfacilities, S390_lowcore.stfle_fac_list, 16);
1808 vfacilities[0] &= 0xff82fffbf47c2000UL;
1809 vfacilities[1] &= 0x005c000000000000UL;
1813 static void __exit kvm_s390_exit(void)
1815 free_page((unsigned long) vfacilities);
1819 module_init(kvm_s390_init);
1820 module_exit(kvm_s390_exit);
1823 * Enable autoloading of the kvm module.
1824 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
1825 * since x86 takes a different approach.
1827 #include <linux/miscdevice.h>
1828 MODULE_ALIAS_MISCDEV(KVM_MINOR);
1829 MODULE_ALIAS("devname:kvm");