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/random.h>
26 #include <linux/slab.h>
27 #include <linux/timer.h>
28 #include <linux/vmalloc.h>
29 #include <asm/asm-offsets.h>
30 #include <asm/lowcore.h>
32 #include <asm/pgtable.h>
34 #include <asm/switch_to.h>
40 #define KMSG_COMPONENT "kvm-s390"
42 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
44 #define CREATE_TRACE_POINTS
46 #include "trace-s390.h"
48 #define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */
50 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
51 (KVM_MAX_VCPUS + LOCAL_IRQS))
53 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
55 struct kvm_stats_debugfs_item debugfs_entries[] = {
56 { "userspace_handled", VCPU_STAT(exit_userspace) },
57 { "exit_null", VCPU_STAT(exit_null) },
58 { "exit_validity", VCPU_STAT(exit_validity) },
59 { "exit_stop_request", VCPU_STAT(exit_stop_request) },
60 { "exit_external_request", VCPU_STAT(exit_external_request) },
61 { "exit_external_interrupt", VCPU_STAT(exit_external_interrupt) },
62 { "exit_instruction", VCPU_STAT(exit_instruction) },
63 { "exit_program_interruption", VCPU_STAT(exit_program_interruption) },
64 { "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program) },
65 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
66 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
67 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
68 { "instruction_lctlg", VCPU_STAT(instruction_lctlg) },
69 { "instruction_lctl", VCPU_STAT(instruction_lctl) },
70 { "instruction_stctl", VCPU_STAT(instruction_stctl) },
71 { "instruction_stctg", VCPU_STAT(instruction_stctg) },
72 { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal) },
73 { "deliver_external_call", VCPU_STAT(deliver_external_call) },
74 { "deliver_service_signal", VCPU_STAT(deliver_service_signal) },
75 { "deliver_virtio_interrupt", VCPU_STAT(deliver_virtio_interrupt) },
76 { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal) },
77 { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal) },
78 { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal) },
79 { "deliver_program_interruption", VCPU_STAT(deliver_program_int) },
80 { "exit_wait_state", VCPU_STAT(exit_wait_state) },
81 { "instruction_pfmf", VCPU_STAT(instruction_pfmf) },
82 { "instruction_stidp", VCPU_STAT(instruction_stidp) },
83 { "instruction_spx", VCPU_STAT(instruction_spx) },
84 { "instruction_stpx", VCPU_STAT(instruction_stpx) },
85 { "instruction_stap", VCPU_STAT(instruction_stap) },
86 { "instruction_storage_key", VCPU_STAT(instruction_storage_key) },
87 { "instruction_ipte_interlock", VCPU_STAT(instruction_ipte_interlock) },
88 { "instruction_stsch", VCPU_STAT(instruction_stsch) },
89 { "instruction_chsc", VCPU_STAT(instruction_chsc) },
90 { "instruction_essa", VCPU_STAT(instruction_essa) },
91 { "instruction_stsi", VCPU_STAT(instruction_stsi) },
92 { "instruction_stfl", VCPU_STAT(instruction_stfl) },
93 { "instruction_tprot", VCPU_STAT(instruction_tprot) },
94 { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense) },
95 { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running) },
96 { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call) },
97 { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency) },
98 { "instruction_sigp_cond_emergency", VCPU_STAT(instruction_sigp_cond_emergency) },
99 { "instruction_sigp_start", VCPU_STAT(instruction_sigp_start) },
100 { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop) },
101 { "instruction_sigp_stop_store_status", VCPU_STAT(instruction_sigp_stop_store_status) },
102 { "instruction_sigp_store_status", VCPU_STAT(instruction_sigp_store_status) },
103 { "instruction_sigp_store_adtl_status", VCPU_STAT(instruction_sigp_store_adtl_status) },
104 { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch) },
105 { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix) },
106 { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart) },
107 { "instruction_sigp_cpu_reset", VCPU_STAT(instruction_sigp_cpu_reset) },
108 { "instruction_sigp_init_cpu_reset", VCPU_STAT(instruction_sigp_init_cpu_reset) },
109 { "instruction_sigp_unknown", VCPU_STAT(instruction_sigp_unknown) },
110 { "diagnose_10", VCPU_STAT(diagnose_10) },
111 { "diagnose_44", VCPU_STAT(diagnose_44) },
112 { "diagnose_9c", VCPU_STAT(diagnose_9c) },
113 { "diagnose_258", VCPU_STAT(diagnose_258) },
114 { "diagnose_308", VCPU_STAT(diagnose_308) },
115 { "diagnose_500", VCPU_STAT(diagnose_500) },
119 /* upper facilities limit for kvm */
120 unsigned long kvm_s390_fac_list_mask[] = {
121 0xffe6fffbfcfdfc40UL,
122 0x005e800000000000UL,
125 unsigned long kvm_s390_fac_list_mask_size(void)
127 BUILD_BUG_ON(ARRAY_SIZE(kvm_s390_fac_list_mask) > S390_ARCH_FAC_MASK_SIZE_U64);
128 return ARRAY_SIZE(kvm_s390_fac_list_mask);
131 static struct gmap_notifier gmap_notifier;
132 debug_info_t *kvm_s390_dbf;
134 /* Section: not file related */
135 int kvm_arch_hardware_enable(void)
137 /* every s390 is virtualization enabled ;-) */
141 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long address);
144 * This callback is executed during stop_machine(). All CPUs are therefore
145 * temporarily stopped. In order not to change guest behavior, we have to
146 * disable preemption whenever we touch the epoch of kvm and the VCPUs,
147 * so a CPU won't be stopped while calculating with the epoch.
149 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
153 struct kvm_vcpu *vcpu;
155 unsigned long long *delta = v;
157 list_for_each_entry(kvm, &vm_list, vm_list) {
158 kvm->arch.epoch -= *delta;
159 kvm_for_each_vcpu(i, vcpu, kvm) {
160 vcpu->arch.sie_block->epoch -= *delta;
166 static struct notifier_block kvm_clock_notifier = {
167 .notifier_call = kvm_clock_sync,
170 int kvm_arch_hardware_setup(void)
172 gmap_notifier.notifier_call = kvm_gmap_notifier;
173 gmap_register_ipte_notifier(&gmap_notifier);
174 atomic_notifier_chain_register(&s390_epoch_delta_notifier,
175 &kvm_clock_notifier);
179 void kvm_arch_hardware_unsetup(void)
181 gmap_unregister_ipte_notifier(&gmap_notifier);
182 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
183 &kvm_clock_notifier);
186 int kvm_arch_init(void *opaque)
188 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
192 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view)) {
193 debug_unregister(kvm_s390_dbf);
197 /* Register floating interrupt controller interface. */
198 return kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
201 void kvm_arch_exit(void)
203 debug_unregister(kvm_s390_dbf);
206 /* Section: device related */
207 long kvm_arch_dev_ioctl(struct file *filp,
208 unsigned int ioctl, unsigned long arg)
210 if (ioctl == KVM_S390_ENABLE_SIE)
211 return s390_enable_sie();
215 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
220 case KVM_CAP_S390_PSW:
221 case KVM_CAP_S390_GMAP:
222 case KVM_CAP_SYNC_MMU:
223 #ifdef CONFIG_KVM_S390_UCONTROL
224 case KVM_CAP_S390_UCONTROL:
226 case KVM_CAP_ASYNC_PF:
227 case KVM_CAP_SYNC_REGS:
228 case KVM_CAP_ONE_REG:
229 case KVM_CAP_ENABLE_CAP:
230 case KVM_CAP_S390_CSS_SUPPORT:
231 case KVM_CAP_IOEVENTFD:
232 case KVM_CAP_DEVICE_CTRL:
233 case KVM_CAP_ENABLE_CAP_VM:
234 case KVM_CAP_S390_IRQCHIP:
235 case KVM_CAP_VM_ATTRIBUTES:
236 case KVM_CAP_MP_STATE:
237 case KVM_CAP_S390_INJECT_IRQ:
238 case KVM_CAP_S390_USER_SIGP:
239 case KVM_CAP_S390_USER_STSI:
240 case KVM_CAP_S390_SKEYS:
241 case KVM_CAP_S390_IRQ_STATE:
244 case KVM_CAP_S390_MEM_OP:
247 case KVM_CAP_NR_VCPUS:
248 case KVM_CAP_MAX_VCPUS:
251 case KVM_CAP_NR_MEMSLOTS:
252 r = KVM_USER_MEM_SLOTS;
254 case KVM_CAP_S390_COW:
255 r = MACHINE_HAS_ESOP;
257 case KVM_CAP_S390_VECTOR_REGISTERS:
266 static void kvm_s390_sync_dirty_log(struct kvm *kvm,
267 struct kvm_memory_slot *memslot)
269 gfn_t cur_gfn, last_gfn;
270 unsigned long address;
271 struct gmap *gmap = kvm->arch.gmap;
273 down_read(&gmap->mm->mmap_sem);
274 /* Loop over all guest pages */
275 last_gfn = memslot->base_gfn + memslot->npages;
276 for (cur_gfn = memslot->base_gfn; cur_gfn <= last_gfn; cur_gfn++) {
277 address = gfn_to_hva_memslot(memslot, cur_gfn);
279 if (gmap_test_and_clear_dirty(address, gmap))
280 mark_page_dirty(kvm, cur_gfn);
282 up_read(&gmap->mm->mmap_sem);
285 /* Section: vm related */
287 * Get (and clear) the dirty memory log for a memory slot.
289 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
290 struct kvm_dirty_log *log)
294 struct kvm_memslots *slots;
295 struct kvm_memory_slot *memslot;
298 if (kvm_is_ucontrol(kvm))
301 mutex_lock(&kvm->slots_lock);
304 if (log->slot >= KVM_USER_MEM_SLOTS)
307 slots = kvm_memslots(kvm);
308 memslot = id_to_memslot(slots, log->slot);
310 if (!memslot->dirty_bitmap)
313 kvm_s390_sync_dirty_log(kvm, memslot);
314 r = kvm_get_dirty_log(kvm, log, &is_dirty);
318 /* Clear the dirty log */
320 n = kvm_dirty_bitmap_bytes(memslot);
321 memset(memslot->dirty_bitmap, 0, n);
325 mutex_unlock(&kvm->slots_lock);
329 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
337 case KVM_CAP_S390_IRQCHIP:
338 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
339 kvm->arch.use_irqchip = 1;
342 case KVM_CAP_S390_USER_SIGP:
343 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
344 kvm->arch.user_sigp = 1;
347 case KVM_CAP_S390_VECTOR_REGISTERS:
348 mutex_lock(&kvm->lock);
349 if (atomic_read(&kvm->online_vcpus)) {
351 } else if (MACHINE_HAS_VX) {
352 set_kvm_facility(kvm->arch.model.fac->mask, 129);
353 set_kvm_facility(kvm->arch.model.fac->list, 129);
357 mutex_unlock(&kvm->lock);
358 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
359 r ? "(not available)" : "(success)");
361 case KVM_CAP_S390_USER_STSI:
362 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
363 kvm->arch.user_stsi = 1;
373 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
377 switch (attr->attr) {
378 case KVM_S390_VM_MEM_LIMIT_SIZE:
380 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
381 kvm->arch.gmap->asce_end);
382 if (put_user(kvm->arch.gmap->asce_end, (u64 __user *)attr->addr))
392 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
396 switch (attr->attr) {
397 case KVM_S390_VM_MEM_ENABLE_CMMA:
398 /* enable CMMA only for z10 and later (EDAT_1) */
400 if (!MACHINE_IS_LPAR || !MACHINE_HAS_EDAT1)
404 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
405 mutex_lock(&kvm->lock);
406 if (atomic_read(&kvm->online_vcpus) == 0) {
407 kvm->arch.use_cmma = 1;
410 mutex_unlock(&kvm->lock);
412 case KVM_S390_VM_MEM_CLR_CMMA:
414 if (!kvm->arch.use_cmma)
417 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
418 mutex_lock(&kvm->lock);
419 idx = srcu_read_lock(&kvm->srcu);
420 s390_reset_cmma(kvm->arch.gmap->mm);
421 srcu_read_unlock(&kvm->srcu, idx);
422 mutex_unlock(&kvm->lock);
425 case KVM_S390_VM_MEM_LIMIT_SIZE: {
426 unsigned long new_limit;
428 if (kvm_is_ucontrol(kvm))
431 if (get_user(new_limit, (u64 __user *)attr->addr))
434 if (new_limit > kvm->arch.gmap->asce_end)
438 mutex_lock(&kvm->lock);
439 if (atomic_read(&kvm->online_vcpus) == 0) {
440 /* gmap_alloc will round the limit up */
441 struct gmap *new = gmap_alloc(current->mm, new_limit);
446 gmap_free(kvm->arch.gmap);
448 kvm->arch.gmap = new;
452 mutex_unlock(&kvm->lock);
453 VM_EVENT(kvm, 3, "SET: max guest memory: %lu bytes", new_limit);
463 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
465 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
467 struct kvm_vcpu *vcpu;
470 if (!test_kvm_facility(kvm, 76))
473 mutex_lock(&kvm->lock);
474 switch (attr->attr) {
475 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
477 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
478 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
479 kvm->arch.crypto.aes_kw = 1;
480 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
482 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
484 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
485 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
486 kvm->arch.crypto.dea_kw = 1;
487 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
489 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
490 kvm->arch.crypto.aes_kw = 0;
491 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
492 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
493 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
495 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
496 kvm->arch.crypto.dea_kw = 0;
497 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
498 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
499 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
502 mutex_unlock(&kvm->lock);
506 kvm_for_each_vcpu(i, vcpu, kvm) {
507 kvm_s390_vcpu_crypto_setup(vcpu);
510 mutex_unlock(&kvm->lock);
514 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
518 if (copy_from_user(>od_high, (void __user *)attr->addr,
524 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
529 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
533 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod)))
536 kvm_s390_set_tod_clock(kvm, gtod);
537 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod);
541 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
548 switch (attr->attr) {
549 case KVM_S390_VM_TOD_HIGH:
550 ret = kvm_s390_set_tod_high(kvm, attr);
552 case KVM_S390_VM_TOD_LOW:
553 ret = kvm_s390_set_tod_low(kvm, attr);
562 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
566 if (copy_to_user((void __user *)attr->addr, >od_high,
569 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
574 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
578 gtod = kvm_s390_get_tod_clock_fast(kvm);
579 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
581 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
586 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
593 switch (attr->attr) {
594 case KVM_S390_VM_TOD_HIGH:
595 ret = kvm_s390_get_tod_high(kvm, attr);
597 case KVM_S390_VM_TOD_LOW:
598 ret = kvm_s390_get_tod_low(kvm, attr);
607 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
609 struct kvm_s390_vm_cpu_processor *proc;
612 mutex_lock(&kvm->lock);
613 if (atomic_read(&kvm->online_vcpus)) {
617 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
622 if (!copy_from_user(proc, (void __user *)attr->addr,
624 memcpy(&kvm->arch.model.cpu_id, &proc->cpuid,
625 sizeof(struct cpuid));
626 kvm->arch.model.ibc = proc->ibc;
627 memcpy(kvm->arch.model.fac->list, proc->fac_list,
628 S390_ARCH_FAC_LIST_SIZE_BYTE);
633 mutex_unlock(&kvm->lock);
637 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
641 switch (attr->attr) {
642 case KVM_S390_VM_CPU_PROCESSOR:
643 ret = kvm_s390_set_processor(kvm, attr);
649 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
651 struct kvm_s390_vm_cpu_processor *proc;
654 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
659 memcpy(&proc->cpuid, &kvm->arch.model.cpu_id, sizeof(struct cpuid));
660 proc->ibc = kvm->arch.model.ibc;
661 memcpy(&proc->fac_list, kvm->arch.model.fac->list, S390_ARCH_FAC_LIST_SIZE_BYTE);
662 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
669 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
671 struct kvm_s390_vm_cpu_machine *mach;
674 mach = kzalloc(sizeof(*mach), GFP_KERNEL);
679 get_cpu_id((struct cpuid *) &mach->cpuid);
680 mach->ibc = sclp.ibc;
681 memcpy(&mach->fac_mask, kvm->arch.model.fac->mask,
682 S390_ARCH_FAC_LIST_SIZE_BYTE);
683 memcpy((unsigned long *)&mach->fac_list, S390_lowcore.stfle_fac_list,
684 S390_ARCH_FAC_LIST_SIZE_BYTE);
685 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
692 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
696 switch (attr->attr) {
697 case KVM_S390_VM_CPU_PROCESSOR:
698 ret = kvm_s390_get_processor(kvm, attr);
700 case KVM_S390_VM_CPU_MACHINE:
701 ret = kvm_s390_get_machine(kvm, attr);
707 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
711 switch (attr->group) {
712 case KVM_S390_VM_MEM_CTRL:
713 ret = kvm_s390_set_mem_control(kvm, attr);
715 case KVM_S390_VM_TOD:
716 ret = kvm_s390_set_tod(kvm, attr);
718 case KVM_S390_VM_CPU_MODEL:
719 ret = kvm_s390_set_cpu_model(kvm, attr);
721 case KVM_S390_VM_CRYPTO:
722 ret = kvm_s390_vm_set_crypto(kvm, attr);
732 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
736 switch (attr->group) {
737 case KVM_S390_VM_MEM_CTRL:
738 ret = kvm_s390_get_mem_control(kvm, attr);
740 case KVM_S390_VM_TOD:
741 ret = kvm_s390_get_tod(kvm, attr);
743 case KVM_S390_VM_CPU_MODEL:
744 ret = kvm_s390_get_cpu_model(kvm, attr);
754 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
758 switch (attr->group) {
759 case KVM_S390_VM_MEM_CTRL:
760 switch (attr->attr) {
761 case KVM_S390_VM_MEM_ENABLE_CMMA:
762 case KVM_S390_VM_MEM_CLR_CMMA:
763 case KVM_S390_VM_MEM_LIMIT_SIZE:
771 case KVM_S390_VM_TOD:
772 switch (attr->attr) {
773 case KVM_S390_VM_TOD_LOW:
774 case KVM_S390_VM_TOD_HIGH:
782 case KVM_S390_VM_CPU_MODEL:
783 switch (attr->attr) {
784 case KVM_S390_VM_CPU_PROCESSOR:
785 case KVM_S390_VM_CPU_MACHINE:
793 case KVM_S390_VM_CRYPTO:
794 switch (attr->attr) {
795 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
796 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
797 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
798 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
814 static long kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
818 unsigned long curkey;
821 if (args->flags != 0)
824 /* Is this guest using storage keys? */
825 if (!mm_use_skey(current->mm))
826 return KVM_S390_GET_SKEYS_NONE;
828 /* Enforce sane limit on memory allocation */
829 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
832 keys = kmalloc_array(args->count, sizeof(uint8_t),
833 GFP_KERNEL | __GFP_NOWARN);
835 keys = vmalloc(sizeof(uint8_t) * args->count);
839 for (i = 0; i < args->count; i++) {
840 hva = gfn_to_hva(kvm, args->start_gfn + i);
841 if (kvm_is_error_hva(hva)) {
846 curkey = get_guest_storage_key(current->mm, hva);
847 if (IS_ERR_VALUE(curkey)) {
854 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
855 sizeof(uint8_t) * args->count);
863 static long kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
869 if (args->flags != 0)
872 /* Enforce sane limit on memory allocation */
873 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
876 keys = kmalloc_array(args->count, sizeof(uint8_t),
877 GFP_KERNEL | __GFP_NOWARN);
879 keys = vmalloc(sizeof(uint8_t) * args->count);
883 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
884 sizeof(uint8_t) * args->count);
890 /* Enable storage key handling for the guest */
891 r = s390_enable_skey();
895 for (i = 0; i < args->count; i++) {
896 hva = gfn_to_hva(kvm, args->start_gfn + i);
897 if (kvm_is_error_hva(hva)) {
902 /* Lowest order bit is reserved */
903 if (keys[i] & 0x01) {
908 r = set_guest_storage_key(current->mm, hva,
909 (unsigned long)keys[i], 0);
918 long kvm_arch_vm_ioctl(struct file *filp,
919 unsigned int ioctl, unsigned long arg)
921 struct kvm *kvm = filp->private_data;
922 void __user *argp = (void __user *)arg;
923 struct kvm_device_attr attr;
927 case KVM_S390_INTERRUPT: {
928 struct kvm_s390_interrupt s390int;
931 if (copy_from_user(&s390int, argp, sizeof(s390int)))
933 r = kvm_s390_inject_vm(kvm, &s390int);
936 case KVM_ENABLE_CAP: {
937 struct kvm_enable_cap cap;
939 if (copy_from_user(&cap, argp, sizeof(cap)))
941 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
944 case KVM_CREATE_IRQCHIP: {
945 struct kvm_irq_routing_entry routing;
948 if (kvm->arch.use_irqchip) {
949 /* Set up dummy routing. */
950 memset(&routing, 0, sizeof(routing));
951 r = kvm_set_irq_routing(kvm, &routing, 0, 0);
955 case KVM_SET_DEVICE_ATTR: {
957 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
959 r = kvm_s390_vm_set_attr(kvm, &attr);
962 case KVM_GET_DEVICE_ATTR: {
964 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
966 r = kvm_s390_vm_get_attr(kvm, &attr);
969 case KVM_HAS_DEVICE_ATTR: {
971 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
973 r = kvm_s390_vm_has_attr(kvm, &attr);
976 case KVM_S390_GET_SKEYS: {
977 struct kvm_s390_skeys args;
980 if (copy_from_user(&args, argp,
981 sizeof(struct kvm_s390_skeys)))
983 r = kvm_s390_get_skeys(kvm, &args);
986 case KVM_S390_SET_SKEYS: {
987 struct kvm_s390_skeys args;
990 if (copy_from_user(&args, argp,
991 sizeof(struct kvm_s390_skeys)))
993 r = kvm_s390_set_skeys(kvm, &args);
1003 static int kvm_s390_query_ap_config(u8 *config)
1005 u32 fcn_code = 0x04000000UL;
1008 memset(config, 0, 128);
1012 ".long 0xb2af0000\n" /* PQAP(QCI) */
1018 : "r" (fcn_code), "r" (config)
1019 : "cc", "0", "2", "memory"
1025 static int kvm_s390_apxa_installed(void)
1030 if (test_facility(2) && test_facility(12)) {
1031 cc = kvm_s390_query_ap_config(config);
1034 pr_err("PQAP(QCI) failed with cc=%d", cc);
1036 return config[0] & 0x40;
1042 static void kvm_s390_set_crycb_format(struct kvm *kvm)
1044 kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
1046 if (kvm_s390_apxa_installed())
1047 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
1049 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
1052 static void kvm_s390_get_cpu_id(struct cpuid *cpu_id)
1055 cpu_id->version = 0xff;
1058 static int kvm_s390_crypto_init(struct kvm *kvm)
1060 if (!test_kvm_facility(kvm, 76))
1063 kvm->arch.crypto.crycb = kzalloc(sizeof(*kvm->arch.crypto.crycb),
1064 GFP_KERNEL | GFP_DMA);
1065 if (!kvm->arch.crypto.crycb)
1068 kvm_s390_set_crycb_format(kvm);
1070 /* Enable AES/DEA protected key functions by default */
1071 kvm->arch.crypto.aes_kw = 1;
1072 kvm->arch.crypto.dea_kw = 1;
1073 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1074 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1075 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1076 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1081 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
1084 char debug_name[16];
1085 static unsigned long sca_offset;
1088 #ifdef CONFIG_KVM_S390_UCONTROL
1089 if (type & ~KVM_VM_S390_UCONTROL)
1091 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
1098 rc = s390_enable_sie();
1104 kvm->arch.sca = (struct sca_block *) get_zeroed_page(GFP_KERNEL);
1107 spin_lock(&kvm_lock);
1109 if (sca_offset + sizeof(struct sca_block) > PAGE_SIZE)
1111 kvm->arch.sca = (struct sca_block *) ((char *) kvm->arch.sca + sca_offset);
1112 spin_unlock(&kvm_lock);
1114 sprintf(debug_name, "kvm-%u", current->pid);
1116 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
1121 * The architectural maximum amount of facilities is 16 kbit. To store
1122 * this amount, 2 kbyte of memory is required. Thus we need a full
1123 * page to hold the guest facility list (arch.model.fac->list) and the
1124 * facility mask (arch.model.fac->mask). Its address size has to be
1125 * 31 bits and word aligned.
1127 kvm->arch.model.fac =
1128 (struct kvm_s390_fac *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1129 if (!kvm->arch.model.fac)
1132 /* Populate the facility mask initially. */
1133 memcpy(kvm->arch.model.fac->mask, S390_lowcore.stfle_fac_list,
1134 S390_ARCH_FAC_LIST_SIZE_BYTE);
1135 for (i = 0; i < S390_ARCH_FAC_LIST_SIZE_U64; i++) {
1136 if (i < kvm_s390_fac_list_mask_size())
1137 kvm->arch.model.fac->mask[i] &= kvm_s390_fac_list_mask[i];
1139 kvm->arch.model.fac->mask[i] = 0UL;
1142 /* Populate the facility list initially. */
1143 memcpy(kvm->arch.model.fac->list, kvm->arch.model.fac->mask,
1144 S390_ARCH_FAC_LIST_SIZE_BYTE);
1146 kvm_s390_get_cpu_id(&kvm->arch.model.cpu_id);
1147 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
1149 if (kvm_s390_crypto_init(kvm) < 0)
1152 spin_lock_init(&kvm->arch.float_int.lock);
1153 for (i = 0; i < FIRQ_LIST_COUNT; i++)
1154 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
1155 init_waitqueue_head(&kvm->arch.ipte_wq);
1156 mutex_init(&kvm->arch.ipte_mutex);
1158 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
1159 VM_EVENT(kvm, 3, "vm created with type %lu", type);
1161 if (type & KVM_VM_S390_UCONTROL) {
1162 kvm->arch.gmap = NULL;
1164 kvm->arch.gmap = gmap_alloc(current->mm, (1UL << 44) - 1);
1165 if (!kvm->arch.gmap)
1167 kvm->arch.gmap->private = kvm;
1168 kvm->arch.gmap->pfault_enabled = 0;
1171 kvm->arch.css_support = 0;
1172 kvm->arch.use_irqchip = 0;
1173 kvm->arch.epoch = 0;
1175 spin_lock_init(&kvm->arch.start_stop_lock);
1176 KVM_EVENT(3, "vm 0x%p created by pid %u", kvm, current->pid);
1180 kfree(kvm->arch.crypto.crycb);
1181 free_page((unsigned long)kvm->arch.model.fac);
1182 debug_unregister(kvm->arch.dbf);
1183 free_page((unsigned long)(kvm->arch.sca));
1184 KVM_EVENT(3, "creation of vm failed: %d", rc);
1188 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1190 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
1191 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
1192 kvm_s390_clear_local_irqs(vcpu);
1193 kvm_clear_async_pf_completion_queue(vcpu);
1194 if (!kvm_is_ucontrol(vcpu->kvm)) {
1195 clear_bit(63 - vcpu->vcpu_id,
1196 (unsigned long *) &vcpu->kvm->arch.sca->mcn);
1197 if (vcpu->kvm->arch.sca->cpu[vcpu->vcpu_id].sda ==
1198 (__u64) vcpu->arch.sie_block)
1199 vcpu->kvm->arch.sca->cpu[vcpu->vcpu_id].sda = 0;
1203 if (kvm_is_ucontrol(vcpu->kvm))
1204 gmap_free(vcpu->arch.gmap);
1206 if (vcpu->kvm->arch.use_cmma)
1207 kvm_s390_vcpu_unsetup_cmma(vcpu);
1208 free_page((unsigned long)(vcpu->arch.sie_block));
1210 kvm_vcpu_uninit(vcpu);
1211 kmem_cache_free(kvm_vcpu_cache, vcpu);
1214 static void kvm_free_vcpus(struct kvm *kvm)
1217 struct kvm_vcpu *vcpu;
1219 kvm_for_each_vcpu(i, vcpu, kvm)
1220 kvm_arch_vcpu_destroy(vcpu);
1222 mutex_lock(&kvm->lock);
1223 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
1224 kvm->vcpus[i] = NULL;
1226 atomic_set(&kvm->online_vcpus, 0);
1227 mutex_unlock(&kvm->lock);
1230 void kvm_arch_destroy_vm(struct kvm *kvm)
1232 kvm_free_vcpus(kvm);
1233 free_page((unsigned long)kvm->arch.model.fac);
1234 free_page((unsigned long)(kvm->arch.sca));
1235 debug_unregister(kvm->arch.dbf);
1236 kfree(kvm->arch.crypto.crycb);
1237 if (!kvm_is_ucontrol(kvm))
1238 gmap_free(kvm->arch.gmap);
1239 kvm_s390_destroy_adapters(kvm);
1240 kvm_s390_clear_float_irqs(kvm);
1241 KVM_EVENT(3, "vm 0x%p destroyed", kvm);
1244 /* Section: vcpu related */
1245 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
1247 vcpu->arch.gmap = gmap_alloc(current->mm, -1UL);
1248 if (!vcpu->arch.gmap)
1250 vcpu->arch.gmap->private = vcpu->kvm;
1255 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1257 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1258 kvm_clear_async_pf_completion_queue(vcpu);
1259 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
1265 if (test_kvm_facility(vcpu->kvm, 129))
1266 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
1268 if (kvm_is_ucontrol(vcpu->kvm))
1269 return __kvm_ucontrol_vcpu_init(vcpu);
1274 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1276 /* Save host register state */
1278 vcpu->arch.host_fpregs.fpc = current->thread.fpu.fpc;
1279 vcpu->arch.host_fpregs.regs = current->thread.fpu.regs;
1281 /* Depending on MACHINE_HAS_VX, data stored to vrs either
1282 * has vector register or floating point register format.
1284 current->thread.fpu.regs = vcpu->run->s.regs.vrs;
1285 current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
1286 if (test_fp_ctl(current->thread.fpu.fpc))
1287 /* User space provided an invalid FPC, let's clear it */
1288 current->thread.fpu.fpc = 0;
1290 save_access_regs(vcpu->arch.host_acrs);
1291 restore_access_regs(vcpu->run->s.regs.acrs);
1292 gmap_enable(vcpu->arch.gmap);
1293 atomic_or(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1296 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1298 atomic_andnot(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1299 gmap_disable(vcpu->arch.gmap);
1301 /* Save guest register state */
1303 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
1305 /* Restore host register state */
1306 current->thread.fpu.fpc = vcpu->arch.host_fpregs.fpc;
1307 current->thread.fpu.regs = vcpu->arch.host_fpregs.regs;
1309 save_access_regs(vcpu->run->s.regs.acrs);
1310 restore_access_regs(vcpu->arch.host_acrs);
1313 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
1315 /* this equals initial cpu reset in pop, but we don't switch to ESA */
1316 vcpu->arch.sie_block->gpsw.mask = 0UL;
1317 vcpu->arch.sie_block->gpsw.addr = 0UL;
1318 kvm_s390_set_prefix(vcpu, 0);
1319 vcpu->arch.sie_block->cputm = 0UL;
1320 vcpu->arch.sie_block->ckc = 0UL;
1321 vcpu->arch.sie_block->todpr = 0;
1322 memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64));
1323 vcpu->arch.sie_block->gcr[0] = 0xE0UL;
1324 vcpu->arch.sie_block->gcr[14] = 0xC2000000UL;
1325 /* make sure the new fpc will be lazily loaded */
1327 current->thread.fpu.fpc = 0;
1328 vcpu->arch.sie_block->gbea = 1;
1329 vcpu->arch.sie_block->pp = 0;
1330 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1331 kvm_clear_async_pf_completion_queue(vcpu);
1332 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
1333 kvm_s390_vcpu_stop(vcpu);
1334 kvm_s390_clear_local_irqs(vcpu);
1337 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1339 mutex_lock(&vcpu->kvm->lock);
1341 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
1343 mutex_unlock(&vcpu->kvm->lock);
1344 if (!kvm_is_ucontrol(vcpu->kvm))
1345 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
1348 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
1350 if (!test_kvm_facility(vcpu->kvm, 76))
1353 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
1355 if (vcpu->kvm->arch.crypto.aes_kw)
1356 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
1357 if (vcpu->kvm->arch.crypto.dea_kw)
1358 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
1360 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
1363 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
1365 free_page(vcpu->arch.sie_block->cbrlo);
1366 vcpu->arch.sie_block->cbrlo = 0;
1369 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
1371 vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
1372 if (!vcpu->arch.sie_block->cbrlo)
1375 vcpu->arch.sie_block->ecb2 |= 0x80;
1376 vcpu->arch.sie_block->ecb2 &= ~0x08;
1380 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
1382 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
1384 vcpu->arch.cpu_id = model->cpu_id;
1385 vcpu->arch.sie_block->ibc = model->ibc;
1386 vcpu->arch.sie_block->fac = (int) (long) model->fac->list;
1389 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1393 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
1397 if (test_kvm_facility(vcpu->kvm, 78))
1398 atomic_or(CPUSTAT_GED2, &vcpu->arch.sie_block->cpuflags);
1399 else if (test_kvm_facility(vcpu->kvm, 8))
1400 atomic_or(CPUSTAT_GED, &vcpu->arch.sie_block->cpuflags);
1402 kvm_s390_vcpu_setup_model(vcpu);
1404 vcpu->arch.sie_block->ecb = 6;
1405 if (test_kvm_facility(vcpu->kvm, 50) && test_kvm_facility(vcpu->kvm, 73))
1406 vcpu->arch.sie_block->ecb |= 0x10;
1408 vcpu->arch.sie_block->ecb2 = 8;
1409 vcpu->arch.sie_block->eca = 0xC1002000U;
1411 vcpu->arch.sie_block->eca |= 1;
1412 if (sclp.has_sigpif)
1413 vcpu->arch.sie_block->eca |= 0x10000000U;
1414 if (test_kvm_facility(vcpu->kvm, 129)) {
1415 vcpu->arch.sie_block->eca |= 0x00020000;
1416 vcpu->arch.sie_block->ecd |= 0x20000000;
1418 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
1420 if (vcpu->kvm->arch.use_cmma) {
1421 rc = kvm_s390_vcpu_setup_cmma(vcpu);
1425 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1426 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
1428 kvm_s390_vcpu_crypto_setup(vcpu);
1433 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1436 struct kvm_vcpu *vcpu;
1437 struct sie_page *sie_page;
1440 if (id >= KVM_MAX_VCPUS)
1445 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1449 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
1453 vcpu->arch.sie_block = &sie_page->sie_block;
1454 vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
1456 vcpu->arch.sie_block->icpua = id;
1457 if (!kvm_is_ucontrol(kvm)) {
1458 if (!kvm->arch.sca) {
1462 if (!kvm->arch.sca->cpu[id].sda)
1463 kvm->arch.sca->cpu[id].sda =
1464 (__u64) vcpu->arch.sie_block;
1465 vcpu->arch.sie_block->scaoh =
1466 (__u32)(((__u64)kvm->arch.sca) >> 32);
1467 vcpu->arch.sie_block->scaol = (__u32)(__u64)kvm->arch.sca;
1468 set_bit(63 - id, (unsigned long *) &kvm->arch.sca->mcn);
1471 spin_lock_init(&vcpu->arch.local_int.lock);
1472 vcpu->arch.local_int.float_int = &kvm->arch.float_int;
1473 vcpu->arch.local_int.wq = &vcpu->wq;
1474 vcpu->arch.local_int.cpuflags = &vcpu->arch.sie_block->cpuflags;
1476 rc = kvm_vcpu_init(vcpu, kvm, id);
1478 goto out_free_sie_block;
1479 VM_EVENT(kvm, 3, "create cpu %d at %p, sie block at %p", id, vcpu,
1480 vcpu->arch.sie_block);
1481 trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
1485 free_page((unsigned long)(vcpu->arch.sie_block));
1487 kmem_cache_free(kvm_vcpu_cache, vcpu);
1492 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
1494 return kvm_s390_vcpu_has_irq(vcpu, 0);
1497 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
1499 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
1503 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
1505 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
1508 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
1510 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
1514 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
1516 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
1520 * Kick a guest cpu out of SIE and wait until SIE is not running.
1521 * If the CPU is not running (e.g. waiting as idle) the function will
1522 * return immediately. */
1523 void exit_sie(struct kvm_vcpu *vcpu)
1525 atomic_or(CPUSTAT_STOP_INT, &vcpu->arch.sie_block->cpuflags);
1526 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
1530 /* Kick a guest cpu out of SIE to process a request synchronously */
1531 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
1533 kvm_make_request(req, vcpu);
1534 kvm_s390_vcpu_request(vcpu);
1537 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long address)
1540 struct kvm *kvm = gmap->private;
1541 struct kvm_vcpu *vcpu;
1543 kvm_for_each_vcpu(i, vcpu, kvm) {
1544 /* match against both prefix pages */
1545 if (kvm_s390_get_prefix(vcpu) == (address & ~0x1000UL)) {
1546 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx", address);
1547 kvm_s390_sync_request(KVM_REQ_MMU_RELOAD, vcpu);
1552 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
1554 /* kvm common code refers to this, but never calls it */
1559 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
1560 struct kvm_one_reg *reg)
1565 case KVM_REG_S390_TODPR:
1566 r = put_user(vcpu->arch.sie_block->todpr,
1567 (u32 __user *)reg->addr);
1569 case KVM_REG_S390_EPOCHDIFF:
1570 r = put_user(vcpu->arch.sie_block->epoch,
1571 (u64 __user *)reg->addr);
1573 case KVM_REG_S390_CPU_TIMER:
1574 r = put_user(vcpu->arch.sie_block->cputm,
1575 (u64 __user *)reg->addr);
1577 case KVM_REG_S390_CLOCK_COMP:
1578 r = put_user(vcpu->arch.sie_block->ckc,
1579 (u64 __user *)reg->addr);
1581 case KVM_REG_S390_PFTOKEN:
1582 r = put_user(vcpu->arch.pfault_token,
1583 (u64 __user *)reg->addr);
1585 case KVM_REG_S390_PFCOMPARE:
1586 r = put_user(vcpu->arch.pfault_compare,
1587 (u64 __user *)reg->addr);
1589 case KVM_REG_S390_PFSELECT:
1590 r = put_user(vcpu->arch.pfault_select,
1591 (u64 __user *)reg->addr);
1593 case KVM_REG_S390_PP:
1594 r = put_user(vcpu->arch.sie_block->pp,
1595 (u64 __user *)reg->addr);
1597 case KVM_REG_S390_GBEA:
1598 r = put_user(vcpu->arch.sie_block->gbea,
1599 (u64 __user *)reg->addr);
1608 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
1609 struct kvm_one_reg *reg)
1614 case KVM_REG_S390_TODPR:
1615 r = get_user(vcpu->arch.sie_block->todpr,
1616 (u32 __user *)reg->addr);
1618 case KVM_REG_S390_EPOCHDIFF:
1619 r = get_user(vcpu->arch.sie_block->epoch,
1620 (u64 __user *)reg->addr);
1622 case KVM_REG_S390_CPU_TIMER:
1623 r = get_user(vcpu->arch.sie_block->cputm,
1624 (u64 __user *)reg->addr);
1626 case KVM_REG_S390_CLOCK_COMP:
1627 r = get_user(vcpu->arch.sie_block->ckc,
1628 (u64 __user *)reg->addr);
1630 case KVM_REG_S390_PFTOKEN:
1631 r = get_user(vcpu->arch.pfault_token,
1632 (u64 __user *)reg->addr);
1633 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
1634 kvm_clear_async_pf_completion_queue(vcpu);
1636 case KVM_REG_S390_PFCOMPARE:
1637 r = get_user(vcpu->arch.pfault_compare,
1638 (u64 __user *)reg->addr);
1640 case KVM_REG_S390_PFSELECT:
1641 r = get_user(vcpu->arch.pfault_select,
1642 (u64 __user *)reg->addr);
1644 case KVM_REG_S390_PP:
1645 r = get_user(vcpu->arch.sie_block->pp,
1646 (u64 __user *)reg->addr);
1648 case KVM_REG_S390_GBEA:
1649 r = get_user(vcpu->arch.sie_block->gbea,
1650 (u64 __user *)reg->addr);
1659 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
1661 kvm_s390_vcpu_initial_reset(vcpu);
1665 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1667 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs));
1671 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1673 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
1677 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1678 struct kvm_sregs *sregs)
1680 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
1681 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
1682 restore_access_regs(vcpu->run->s.regs.acrs);
1686 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1687 struct kvm_sregs *sregs)
1689 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
1690 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
1694 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1696 /* make sure the new values will be lazily loaded */
1698 if (test_fp_ctl(fpu->fpc))
1700 current->thread.fpu.fpc = fpu->fpc;
1702 convert_fp_to_vx(current->thread.fpu.vxrs, (freg_t *)fpu->fprs);
1704 memcpy(current->thread.fpu.fprs, &fpu->fprs, sizeof(fpu->fprs));
1708 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1710 /* make sure we have the latest values */
1713 convert_vx_to_fp((freg_t *)fpu->fprs, current->thread.fpu.vxrs);
1715 memcpy(fpu->fprs, current->thread.fpu.fprs, sizeof(fpu->fprs));
1716 fpu->fpc = current->thread.fpu.fpc;
1720 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
1724 if (!is_vcpu_stopped(vcpu))
1727 vcpu->run->psw_mask = psw.mask;
1728 vcpu->run->psw_addr = psw.addr;
1733 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1734 struct kvm_translation *tr)
1736 return -EINVAL; /* not implemented yet */
1739 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
1740 KVM_GUESTDBG_USE_HW_BP | \
1741 KVM_GUESTDBG_ENABLE)
1743 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1744 struct kvm_guest_debug *dbg)
1748 vcpu->guest_debug = 0;
1749 kvm_s390_clear_bp_data(vcpu);
1751 if (dbg->control & ~VALID_GUESTDBG_FLAGS)
1754 if (dbg->control & KVM_GUESTDBG_ENABLE) {
1755 vcpu->guest_debug = dbg->control;
1756 /* enforce guest PER */
1757 atomic_or(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
1759 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
1760 rc = kvm_s390_import_bp_data(vcpu, dbg);
1762 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
1763 vcpu->arch.guestdbg.last_bp = 0;
1767 vcpu->guest_debug = 0;
1768 kvm_s390_clear_bp_data(vcpu);
1769 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
1775 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1776 struct kvm_mp_state *mp_state)
1778 /* CHECK_STOP and LOAD are not supported yet */
1779 return is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
1780 KVM_MP_STATE_OPERATING;
1783 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1784 struct kvm_mp_state *mp_state)
1788 /* user space knows about this interface - let it control the state */
1789 vcpu->kvm->arch.user_cpu_state_ctrl = 1;
1791 switch (mp_state->mp_state) {
1792 case KVM_MP_STATE_STOPPED:
1793 kvm_s390_vcpu_stop(vcpu);
1795 case KVM_MP_STATE_OPERATING:
1796 kvm_s390_vcpu_start(vcpu);
1798 case KVM_MP_STATE_LOAD:
1799 case KVM_MP_STATE_CHECK_STOP:
1800 /* fall through - CHECK_STOP and LOAD are not supported yet */
1808 static bool ibs_enabled(struct kvm_vcpu *vcpu)
1810 return atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_IBS;
1813 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
1816 kvm_s390_vcpu_request_handled(vcpu);
1817 if (!vcpu->requests)
1820 * We use MMU_RELOAD just to re-arm the ipte notifier for the
1821 * guest prefix page. gmap_ipte_notify will wait on the ptl lock.
1822 * This ensures that the ipte instruction for this request has
1823 * already finished. We might race against a second unmapper that
1824 * wants to set the blocking bit. Lets just retry the request loop.
1826 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
1828 rc = gmap_ipte_notify(vcpu->arch.gmap,
1829 kvm_s390_get_prefix(vcpu),
1836 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
1837 vcpu->arch.sie_block->ihcpu = 0xffff;
1841 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
1842 if (!ibs_enabled(vcpu)) {
1843 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
1844 atomic_or(CPUSTAT_IBS,
1845 &vcpu->arch.sie_block->cpuflags);
1850 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
1851 if (ibs_enabled(vcpu)) {
1852 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
1853 atomic_andnot(CPUSTAT_IBS,
1854 &vcpu->arch.sie_block->cpuflags);
1859 /* nothing to do, just clear the request */
1860 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
1865 void kvm_s390_set_tod_clock(struct kvm *kvm, u64 tod)
1867 struct kvm_vcpu *vcpu;
1870 mutex_lock(&kvm->lock);
1872 kvm->arch.epoch = tod - get_tod_clock();
1873 kvm_s390_vcpu_block_all(kvm);
1874 kvm_for_each_vcpu(i, vcpu, kvm)
1875 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
1876 kvm_s390_vcpu_unblock_all(kvm);
1878 mutex_unlock(&kvm->lock);
1882 * kvm_arch_fault_in_page - fault-in guest page if necessary
1883 * @vcpu: The corresponding virtual cpu
1884 * @gpa: Guest physical address
1885 * @writable: Whether the page should be writable or not
1887 * Make sure that a guest page has been faulted-in on the host.
1889 * Return: Zero on success, negative error code otherwise.
1891 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
1893 return gmap_fault(vcpu->arch.gmap, gpa,
1894 writable ? FAULT_FLAG_WRITE : 0);
1897 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
1898 unsigned long token)
1900 struct kvm_s390_interrupt inti;
1901 struct kvm_s390_irq irq;
1904 irq.u.ext.ext_params2 = token;
1905 irq.type = KVM_S390_INT_PFAULT_INIT;
1906 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
1908 inti.type = KVM_S390_INT_PFAULT_DONE;
1909 inti.parm64 = token;
1910 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
1914 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
1915 struct kvm_async_pf *work)
1917 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
1918 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
1921 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
1922 struct kvm_async_pf *work)
1924 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
1925 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
1928 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
1929 struct kvm_async_pf *work)
1931 /* s390 will always inject the page directly */
1934 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
1937 * s390 will always inject the page directly,
1938 * but we still want check_async_completion to cleanup
1943 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
1946 struct kvm_arch_async_pf arch;
1949 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
1951 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
1952 vcpu->arch.pfault_compare)
1954 if (psw_extint_disabled(vcpu))
1956 if (kvm_s390_vcpu_has_irq(vcpu, 0))
1958 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
1960 if (!vcpu->arch.gmap->pfault_enabled)
1963 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
1964 hva += current->thread.gmap_addr & ~PAGE_MASK;
1965 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
1968 rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
1972 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
1977 * On s390 notifications for arriving pages will be delivered directly
1978 * to the guest but the house keeping for completed pfaults is
1979 * handled outside the worker.
1981 kvm_check_async_pf_completion(vcpu);
1983 memcpy(&vcpu->arch.sie_block->gg14, &vcpu->run->s.regs.gprs[14], 16);
1988 if (test_cpu_flag(CIF_MCCK_PENDING))
1991 if (!kvm_is_ucontrol(vcpu->kvm)) {
1992 rc = kvm_s390_deliver_pending_interrupts(vcpu);
1997 rc = kvm_s390_handle_requests(vcpu);
2001 if (guestdbg_enabled(vcpu)) {
2002 kvm_s390_backup_guest_per_regs(vcpu);
2003 kvm_s390_patch_guest_per_regs(vcpu);
2006 vcpu->arch.sie_block->icptcode = 0;
2007 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
2008 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
2009 trace_kvm_s390_sie_enter(vcpu, cpuflags);
2014 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
2016 psw_t *psw = &vcpu->arch.sie_block->gpsw;
2020 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
2021 trace_kvm_s390_sie_fault(vcpu);
2024 * We want to inject an addressing exception, which is defined as a
2025 * suppressing or terminating exception. However, since we came here
2026 * by a DAT access exception, the PSW still points to the faulting
2027 * instruction since DAT exceptions are nullifying. So we've got
2028 * to look up the current opcode to get the length of the instruction
2029 * to be able to forward the PSW.
2031 rc = read_guest(vcpu, psw->addr, 0, &opcode, 1);
2033 return kvm_s390_inject_prog_cond(vcpu, rc);
2034 psw->addr = __rewind_psw(*psw, -insn_length(opcode));
2036 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
2039 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
2043 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
2044 vcpu->arch.sie_block->icptcode);
2045 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
2047 if (guestdbg_enabled(vcpu))
2048 kvm_s390_restore_guest_per_regs(vcpu);
2050 if (exit_reason >= 0) {
2052 } else if (kvm_is_ucontrol(vcpu->kvm)) {
2053 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
2054 vcpu->run->s390_ucontrol.trans_exc_code =
2055 current->thread.gmap_addr;
2056 vcpu->run->s390_ucontrol.pgm_code = 0x10;
2059 } else if (current->thread.gmap_pfault) {
2060 trace_kvm_s390_major_guest_pfault(vcpu);
2061 current->thread.gmap_pfault = 0;
2062 if (kvm_arch_setup_async_pf(vcpu)) {
2065 gpa_t gpa = current->thread.gmap_addr;
2066 rc = kvm_arch_fault_in_page(vcpu, gpa, 1);
2071 rc = vcpu_post_run_fault_in_sie(vcpu);
2073 memcpy(&vcpu->run->s.regs.gprs[14], &vcpu->arch.sie_block->gg14, 16);
2076 if (kvm_is_ucontrol(vcpu->kvm))
2077 /* Don't exit for host interrupts. */
2078 rc = vcpu->arch.sie_block->icptcode ? -EOPNOTSUPP : 0;
2080 rc = kvm_handle_sie_intercept(vcpu);
2086 static int __vcpu_run(struct kvm_vcpu *vcpu)
2088 int rc, exit_reason;
2091 * We try to hold kvm->srcu during most of vcpu_run (except when run-
2092 * ning the guest), so that memslots (and other stuff) are protected
2094 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2097 rc = vcpu_pre_run(vcpu);
2101 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2103 * As PF_VCPU will be used in fault handler, between
2104 * guest_enter and guest_exit should be no uaccess.
2106 local_irq_disable();
2107 __kvm_guest_enter();
2109 exit_reason = sie64a(vcpu->arch.sie_block,
2110 vcpu->run->s.regs.gprs);
2111 local_irq_disable();
2114 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2116 rc = vcpu_post_run(vcpu, exit_reason);
2117 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
2119 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2123 static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2125 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
2126 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
2127 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
2128 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
2129 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
2130 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
2131 /* some control register changes require a tlb flush */
2132 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2134 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
2135 vcpu->arch.sie_block->cputm = kvm_run->s.regs.cputm;
2136 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
2137 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
2138 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
2139 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
2141 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
2142 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
2143 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
2144 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
2145 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2146 kvm_clear_async_pf_completion_queue(vcpu);
2148 kvm_run->kvm_dirty_regs = 0;
2151 static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2153 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
2154 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
2155 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
2156 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
2157 kvm_run->s.regs.cputm = vcpu->arch.sie_block->cputm;
2158 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
2159 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
2160 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
2161 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
2162 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
2163 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
2164 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
2167 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2172 if (guestdbg_exit_pending(vcpu)) {
2173 kvm_s390_prepare_debug_exit(vcpu);
2177 if (vcpu->sigset_active)
2178 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2180 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
2181 kvm_s390_vcpu_start(vcpu);
2182 } else if (is_vcpu_stopped(vcpu)) {
2183 pr_err_ratelimited("can't run stopped vcpu %d\n",
2188 sync_regs(vcpu, kvm_run);
2191 rc = __vcpu_run(vcpu);
2193 if (signal_pending(current) && !rc) {
2194 kvm_run->exit_reason = KVM_EXIT_INTR;
2198 if (guestdbg_exit_pending(vcpu) && !rc) {
2199 kvm_s390_prepare_debug_exit(vcpu);
2203 if (rc == -EOPNOTSUPP) {
2204 /* intercept cannot be handled in-kernel, prepare kvm-run */
2205 kvm_run->exit_reason = KVM_EXIT_S390_SIEIC;
2206 kvm_run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
2207 kvm_run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
2208 kvm_run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
2212 if (rc == -EREMOTE) {
2213 /* intercept was handled, but userspace support is needed
2214 * kvm_run has been prepared by the handler */
2218 store_regs(vcpu, kvm_run);
2220 if (vcpu->sigset_active)
2221 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2223 vcpu->stat.exit_userspace++;
2228 * store status at address
2229 * we use have two special cases:
2230 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
2231 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
2233 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
2235 unsigned char archmode = 1;
2236 freg_t fprs[NUM_FPRS];
2241 px = kvm_s390_get_prefix(vcpu);
2242 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
2243 if (write_guest_abs(vcpu, 163, &archmode, 1))
2246 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
2247 if (write_guest_real(vcpu, 163, &archmode, 1))
2251 gpa -= __LC_FPREGS_SAVE_AREA;
2253 /* manually convert vector registers if necessary */
2254 if (MACHINE_HAS_VX) {
2255 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
2256 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2259 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2260 vcpu->run->s.regs.vrs, 128);
2262 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
2263 vcpu->run->s.regs.gprs, 128);
2264 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
2265 &vcpu->arch.sie_block->gpsw, 16);
2266 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
2268 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
2269 &vcpu->run->s.regs.fpc, 4);
2270 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
2271 &vcpu->arch.sie_block->todpr, 4);
2272 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
2273 &vcpu->arch.sie_block->cputm, 8);
2274 clkcomp = vcpu->arch.sie_block->ckc >> 8;
2275 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
2277 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
2278 &vcpu->run->s.regs.acrs, 64);
2279 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
2280 &vcpu->arch.sie_block->gcr, 128);
2281 return rc ? -EFAULT : 0;
2284 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
2287 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
2288 * copying in vcpu load/put. Lets update our copies before we save
2289 * it into the save area
2292 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
2293 save_access_regs(vcpu->run->s.regs.acrs);
2295 return kvm_s390_store_status_unloaded(vcpu, addr);
2299 * store additional status at address
2301 int kvm_s390_store_adtl_status_unloaded(struct kvm_vcpu *vcpu,
2304 /* Only bits 0-53 are used for address formation */
2305 if (!(gpa & ~0x3ff))
2308 return write_guest_abs(vcpu, gpa & ~0x3ff,
2309 (void *)&vcpu->run->s.regs.vrs, 512);
2312 int kvm_s390_vcpu_store_adtl_status(struct kvm_vcpu *vcpu, unsigned long addr)
2314 if (!test_kvm_facility(vcpu->kvm, 129))
2318 * The guest VXRS are in the host VXRs due to the lazy
2319 * copying in vcpu load/put. We can simply call save_fpu_regs()
2320 * to save the current register state because we are in the
2321 * middle of a load/put cycle.
2323 * Let's update our copies before we save it into the save area.
2327 return kvm_s390_store_adtl_status_unloaded(vcpu, addr);
2330 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2332 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
2333 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
2336 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
2339 struct kvm_vcpu *vcpu;
2341 kvm_for_each_vcpu(i, vcpu, kvm) {
2342 __disable_ibs_on_vcpu(vcpu);
2346 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2348 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
2349 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
2352 void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
2354 int i, online_vcpus, started_vcpus = 0;
2356 if (!is_vcpu_stopped(vcpu))
2359 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
2360 /* Only one cpu at a time may enter/leave the STOPPED state. */
2361 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2362 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2364 for (i = 0; i < online_vcpus; i++) {
2365 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i]))
2369 if (started_vcpus == 0) {
2370 /* we're the only active VCPU -> speed it up */
2371 __enable_ibs_on_vcpu(vcpu);
2372 } else if (started_vcpus == 1) {
2374 * As we are starting a second VCPU, we have to disable
2375 * the IBS facility on all VCPUs to remove potentially
2376 * oustanding ENABLE requests.
2378 __disable_ibs_on_all_vcpus(vcpu->kvm);
2381 atomic_andnot(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2383 * Another VCPU might have used IBS while we were offline.
2384 * Let's play safe and flush the VCPU at startup.
2386 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2387 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2391 void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
2393 int i, online_vcpus, started_vcpus = 0;
2394 struct kvm_vcpu *started_vcpu = NULL;
2396 if (is_vcpu_stopped(vcpu))
2399 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
2400 /* Only one cpu at a time may enter/leave the STOPPED state. */
2401 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2402 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2404 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
2405 kvm_s390_clear_stop_irq(vcpu);
2407 atomic_or(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2408 __disable_ibs_on_vcpu(vcpu);
2410 for (i = 0; i < online_vcpus; i++) {
2411 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) {
2413 started_vcpu = vcpu->kvm->vcpus[i];
2417 if (started_vcpus == 1) {
2419 * As we only have one VCPU left, we want to enable the
2420 * IBS facility for that VCPU to speed it up.
2422 __enable_ibs_on_vcpu(started_vcpu);
2425 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2429 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
2430 struct kvm_enable_cap *cap)
2438 case KVM_CAP_S390_CSS_SUPPORT:
2439 if (!vcpu->kvm->arch.css_support) {
2440 vcpu->kvm->arch.css_support = 1;
2441 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
2442 trace_kvm_s390_enable_css(vcpu->kvm);
2453 static long kvm_s390_guest_mem_op(struct kvm_vcpu *vcpu,
2454 struct kvm_s390_mem_op *mop)
2456 void __user *uaddr = (void __user *)mop->buf;
2457 void *tmpbuf = NULL;
2459 const u64 supported_flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION
2460 | KVM_S390_MEMOP_F_CHECK_ONLY;
2462 if (mop->flags & ~supported_flags)
2465 if (mop->size > MEM_OP_MAX_SIZE)
2468 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
2469 tmpbuf = vmalloc(mop->size);
2474 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2477 case KVM_S390_MEMOP_LOGICAL_READ:
2478 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2479 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size, false);
2482 r = read_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
2484 if (copy_to_user(uaddr, tmpbuf, mop->size))
2488 case KVM_S390_MEMOP_LOGICAL_WRITE:
2489 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2490 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size, true);
2493 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
2497 r = write_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
2503 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2505 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
2506 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
2512 long kvm_arch_vcpu_ioctl(struct file *filp,
2513 unsigned int ioctl, unsigned long arg)
2515 struct kvm_vcpu *vcpu = filp->private_data;
2516 void __user *argp = (void __user *)arg;
2521 case KVM_S390_IRQ: {
2522 struct kvm_s390_irq s390irq;
2525 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
2527 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
2530 case KVM_S390_INTERRUPT: {
2531 struct kvm_s390_interrupt s390int;
2532 struct kvm_s390_irq s390irq;
2535 if (copy_from_user(&s390int, argp, sizeof(s390int)))
2537 if (s390int_to_s390irq(&s390int, &s390irq))
2539 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
2542 case KVM_S390_STORE_STATUS:
2543 idx = srcu_read_lock(&vcpu->kvm->srcu);
2544 r = kvm_s390_vcpu_store_status(vcpu, arg);
2545 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2547 case KVM_S390_SET_INITIAL_PSW: {
2551 if (copy_from_user(&psw, argp, sizeof(psw)))
2553 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
2556 case KVM_S390_INITIAL_RESET:
2557 r = kvm_arch_vcpu_ioctl_initial_reset(vcpu);
2559 case KVM_SET_ONE_REG:
2560 case KVM_GET_ONE_REG: {
2561 struct kvm_one_reg reg;
2563 if (copy_from_user(®, argp, sizeof(reg)))
2565 if (ioctl == KVM_SET_ONE_REG)
2566 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®);
2568 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®);
2571 #ifdef CONFIG_KVM_S390_UCONTROL
2572 case KVM_S390_UCAS_MAP: {
2573 struct kvm_s390_ucas_mapping ucasmap;
2575 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
2580 if (!kvm_is_ucontrol(vcpu->kvm)) {
2585 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
2586 ucasmap.vcpu_addr, ucasmap.length);
2589 case KVM_S390_UCAS_UNMAP: {
2590 struct kvm_s390_ucas_mapping ucasmap;
2592 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
2597 if (!kvm_is_ucontrol(vcpu->kvm)) {
2602 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
2607 case KVM_S390_VCPU_FAULT: {
2608 r = gmap_fault(vcpu->arch.gmap, arg, 0);
2611 case KVM_ENABLE_CAP:
2613 struct kvm_enable_cap cap;
2615 if (copy_from_user(&cap, argp, sizeof(cap)))
2617 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
2620 case KVM_S390_MEM_OP: {
2621 struct kvm_s390_mem_op mem_op;
2623 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
2624 r = kvm_s390_guest_mem_op(vcpu, &mem_op);
2629 case KVM_S390_SET_IRQ_STATE: {
2630 struct kvm_s390_irq_state irq_state;
2633 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
2635 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
2636 irq_state.len == 0 ||
2637 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
2641 r = kvm_s390_set_irq_state(vcpu,
2642 (void __user *) irq_state.buf,
2646 case KVM_S390_GET_IRQ_STATE: {
2647 struct kvm_s390_irq_state irq_state;
2650 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
2652 if (irq_state.len == 0) {
2656 r = kvm_s390_get_irq_state(vcpu,
2657 (__u8 __user *) irq_state.buf,
2667 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2669 #ifdef CONFIG_KVM_S390_UCONTROL
2670 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
2671 && (kvm_is_ucontrol(vcpu->kvm))) {
2672 vmf->page = virt_to_page(vcpu->arch.sie_block);
2673 get_page(vmf->page);
2677 return VM_FAULT_SIGBUS;
2680 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
2681 unsigned long npages)
2686 /* Section: memory related */
2687 int kvm_arch_prepare_memory_region(struct kvm *kvm,
2688 struct kvm_memory_slot *memslot,
2689 const struct kvm_userspace_memory_region *mem,
2690 enum kvm_mr_change change)
2692 /* A few sanity checks. We can have memory slots which have to be
2693 located/ended at a segment boundary (1MB). The memory in userland is
2694 ok to be fragmented into various different vmas. It is okay to mmap()
2695 and munmap() stuff in this slot after doing this call at any time */
2697 if (mem->userspace_addr & 0xffffful)
2700 if (mem->memory_size & 0xffffful)
2706 void kvm_arch_commit_memory_region(struct kvm *kvm,
2707 const struct kvm_userspace_memory_region *mem,
2708 const struct kvm_memory_slot *old,
2709 const struct kvm_memory_slot *new,
2710 enum kvm_mr_change change)
2714 /* If the basics of the memslot do not change, we do not want
2715 * to update the gmap. Every update causes several unnecessary
2716 * segment translation exceptions. This is usually handled just
2717 * fine by the normal fault handler + gmap, but it will also
2718 * cause faults on the prefix page of running guest CPUs.
2720 if (old->userspace_addr == mem->userspace_addr &&
2721 old->base_gfn * PAGE_SIZE == mem->guest_phys_addr &&
2722 old->npages * PAGE_SIZE == mem->memory_size)
2725 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr,
2726 mem->guest_phys_addr, mem->memory_size);
2728 pr_warn("failed to commit memory region\n");
2732 static int __init kvm_s390_init(void)
2734 return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2737 static void __exit kvm_s390_exit(void)
2742 module_init(kvm_s390_init);
2743 module_exit(kvm_s390_exit);
2746 * Enable autoloading of the kvm module.
2747 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
2748 * since x86 takes a different approach.
2750 #include <linux/miscdevice.h>
2751 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2752 MODULE_ALIAS("devname:kvm");