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 mutex_lock(&kvm->slots_lock);
301 if (log->slot >= KVM_USER_MEM_SLOTS)
304 slots = kvm_memslots(kvm);
305 memslot = id_to_memslot(slots, log->slot);
307 if (!memslot->dirty_bitmap)
310 kvm_s390_sync_dirty_log(kvm, memslot);
311 r = kvm_get_dirty_log(kvm, log, &is_dirty);
315 /* Clear the dirty log */
317 n = kvm_dirty_bitmap_bytes(memslot);
318 memset(memslot->dirty_bitmap, 0, n);
322 mutex_unlock(&kvm->slots_lock);
326 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
334 case KVM_CAP_S390_IRQCHIP:
335 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
336 kvm->arch.use_irqchip = 1;
339 case KVM_CAP_S390_USER_SIGP:
340 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
341 kvm->arch.user_sigp = 1;
344 case KVM_CAP_S390_VECTOR_REGISTERS:
345 if (MACHINE_HAS_VX) {
346 set_kvm_facility(kvm->arch.model.fac->mask, 129);
347 set_kvm_facility(kvm->arch.model.fac->list, 129);
351 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
352 r ? "(not available)" : "(success)");
354 case KVM_CAP_S390_USER_STSI:
355 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
356 kvm->arch.user_stsi = 1;
366 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
370 switch (attr->attr) {
371 case KVM_S390_VM_MEM_LIMIT_SIZE:
373 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
374 kvm->arch.gmap->asce_end);
375 if (put_user(kvm->arch.gmap->asce_end, (u64 __user *)attr->addr))
385 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
389 switch (attr->attr) {
390 case KVM_S390_VM_MEM_ENABLE_CMMA:
391 /* enable CMMA only for z10 and later (EDAT_1) */
393 if (!MACHINE_IS_LPAR || !MACHINE_HAS_EDAT1)
397 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
398 mutex_lock(&kvm->lock);
399 if (atomic_read(&kvm->online_vcpus) == 0) {
400 kvm->arch.use_cmma = 1;
403 mutex_unlock(&kvm->lock);
405 case KVM_S390_VM_MEM_CLR_CMMA:
407 if (!kvm->arch.use_cmma)
410 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
411 mutex_lock(&kvm->lock);
412 idx = srcu_read_lock(&kvm->srcu);
413 s390_reset_cmma(kvm->arch.gmap->mm);
414 srcu_read_unlock(&kvm->srcu, idx);
415 mutex_unlock(&kvm->lock);
418 case KVM_S390_VM_MEM_LIMIT_SIZE: {
419 unsigned long new_limit;
421 if (kvm_is_ucontrol(kvm))
424 if (get_user(new_limit, (u64 __user *)attr->addr))
427 if (new_limit > kvm->arch.gmap->asce_end)
431 mutex_lock(&kvm->lock);
432 if (atomic_read(&kvm->online_vcpus) == 0) {
433 /* gmap_alloc will round the limit up */
434 struct gmap *new = gmap_alloc(current->mm, new_limit);
439 gmap_free(kvm->arch.gmap);
441 kvm->arch.gmap = new;
445 mutex_unlock(&kvm->lock);
446 VM_EVENT(kvm, 3, "SET: max guest memory: %lu bytes", new_limit);
456 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
458 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
460 struct kvm_vcpu *vcpu;
463 if (!test_kvm_facility(kvm, 76))
466 mutex_lock(&kvm->lock);
467 switch (attr->attr) {
468 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
470 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
471 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
472 kvm->arch.crypto.aes_kw = 1;
473 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
475 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
477 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
478 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
479 kvm->arch.crypto.dea_kw = 1;
480 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
482 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
483 kvm->arch.crypto.aes_kw = 0;
484 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
485 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
486 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
488 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
489 kvm->arch.crypto.dea_kw = 0;
490 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
491 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
492 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
495 mutex_unlock(&kvm->lock);
499 kvm_for_each_vcpu(i, vcpu, kvm) {
500 kvm_s390_vcpu_crypto_setup(vcpu);
503 mutex_unlock(&kvm->lock);
507 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
511 if (copy_from_user(>od_high, (void __user *)attr->addr,
517 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x\n", gtod_high);
522 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
524 struct kvm_vcpu *cur_vcpu;
525 unsigned int vcpu_idx;
529 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod)))
532 r = store_tod_clock(&host_tod);
536 mutex_lock(&kvm->lock);
538 kvm->arch.epoch = gtod - host_tod;
539 kvm_s390_vcpu_block_all(kvm);
540 kvm_for_each_vcpu(vcpu_idx, cur_vcpu, kvm)
541 cur_vcpu->arch.sie_block->epoch = kvm->arch.epoch;
542 kvm_s390_vcpu_unblock_all(kvm);
544 mutex_unlock(&kvm->lock);
545 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx\n", gtod);
549 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
556 switch (attr->attr) {
557 case KVM_S390_VM_TOD_HIGH:
558 ret = kvm_s390_set_tod_high(kvm, attr);
560 case KVM_S390_VM_TOD_LOW:
561 ret = kvm_s390_set_tod_low(kvm, attr);
570 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
574 if (copy_to_user((void __user *)attr->addr, >od_high,
577 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x\n", gtod_high);
582 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
587 r = store_tod_clock(&host_tod);
592 gtod = host_tod + kvm->arch.epoch;
594 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
596 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx\n", gtod);
601 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
608 switch (attr->attr) {
609 case KVM_S390_VM_TOD_HIGH:
610 ret = kvm_s390_get_tod_high(kvm, attr);
612 case KVM_S390_VM_TOD_LOW:
613 ret = kvm_s390_get_tod_low(kvm, attr);
622 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
624 struct kvm_s390_vm_cpu_processor *proc;
627 mutex_lock(&kvm->lock);
628 if (atomic_read(&kvm->online_vcpus)) {
632 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
637 if (!copy_from_user(proc, (void __user *)attr->addr,
639 memcpy(&kvm->arch.model.cpu_id, &proc->cpuid,
640 sizeof(struct cpuid));
641 kvm->arch.model.ibc = proc->ibc;
642 memcpy(kvm->arch.model.fac->list, proc->fac_list,
643 S390_ARCH_FAC_LIST_SIZE_BYTE);
648 mutex_unlock(&kvm->lock);
652 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
656 switch (attr->attr) {
657 case KVM_S390_VM_CPU_PROCESSOR:
658 ret = kvm_s390_set_processor(kvm, attr);
664 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
666 struct kvm_s390_vm_cpu_processor *proc;
669 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
674 memcpy(&proc->cpuid, &kvm->arch.model.cpu_id, sizeof(struct cpuid));
675 proc->ibc = kvm->arch.model.ibc;
676 memcpy(&proc->fac_list, kvm->arch.model.fac->list, S390_ARCH_FAC_LIST_SIZE_BYTE);
677 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
684 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
686 struct kvm_s390_vm_cpu_machine *mach;
689 mach = kzalloc(sizeof(*mach), GFP_KERNEL);
694 get_cpu_id((struct cpuid *) &mach->cpuid);
695 mach->ibc = sclp.ibc;
696 memcpy(&mach->fac_mask, kvm->arch.model.fac->mask,
697 S390_ARCH_FAC_LIST_SIZE_BYTE);
698 memcpy((unsigned long *)&mach->fac_list, S390_lowcore.stfle_fac_list,
699 S390_ARCH_FAC_LIST_SIZE_BYTE);
700 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
707 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
711 switch (attr->attr) {
712 case KVM_S390_VM_CPU_PROCESSOR:
713 ret = kvm_s390_get_processor(kvm, attr);
715 case KVM_S390_VM_CPU_MACHINE:
716 ret = kvm_s390_get_machine(kvm, attr);
722 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
726 switch (attr->group) {
727 case KVM_S390_VM_MEM_CTRL:
728 ret = kvm_s390_set_mem_control(kvm, attr);
730 case KVM_S390_VM_TOD:
731 ret = kvm_s390_set_tod(kvm, attr);
733 case KVM_S390_VM_CPU_MODEL:
734 ret = kvm_s390_set_cpu_model(kvm, attr);
736 case KVM_S390_VM_CRYPTO:
737 ret = kvm_s390_vm_set_crypto(kvm, attr);
747 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
751 switch (attr->group) {
752 case KVM_S390_VM_MEM_CTRL:
753 ret = kvm_s390_get_mem_control(kvm, attr);
755 case KVM_S390_VM_TOD:
756 ret = kvm_s390_get_tod(kvm, attr);
758 case KVM_S390_VM_CPU_MODEL:
759 ret = kvm_s390_get_cpu_model(kvm, attr);
769 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
773 switch (attr->group) {
774 case KVM_S390_VM_MEM_CTRL:
775 switch (attr->attr) {
776 case KVM_S390_VM_MEM_ENABLE_CMMA:
777 case KVM_S390_VM_MEM_CLR_CMMA:
778 case KVM_S390_VM_MEM_LIMIT_SIZE:
786 case KVM_S390_VM_TOD:
787 switch (attr->attr) {
788 case KVM_S390_VM_TOD_LOW:
789 case KVM_S390_VM_TOD_HIGH:
797 case KVM_S390_VM_CPU_MODEL:
798 switch (attr->attr) {
799 case KVM_S390_VM_CPU_PROCESSOR:
800 case KVM_S390_VM_CPU_MACHINE:
808 case KVM_S390_VM_CRYPTO:
809 switch (attr->attr) {
810 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
811 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
812 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
813 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
829 static long kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
833 unsigned long curkey;
836 if (args->flags != 0)
839 /* Is this guest using storage keys? */
840 if (!mm_use_skey(current->mm))
841 return KVM_S390_GET_SKEYS_NONE;
843 /* Enforce sane limit on memory allocation */
844 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
847 keys = kmalloc_array(args->count, sizeof(uint8_t),
848 GFP_KERNEL | __GFP_NOWARN);
850 keys = vmalloc(sizeof(uint8_t) * args->count);
854 for (i = 0; i < args->count; i++) {
855 hva = gfn_to_hva(kvm, args->start_gfn + i);
856 if (kvm_is_error_hva(hva)) {
861 curkey = get_guest_storage_key(current->mm, hva);
862 if (IS_ERR_VALUE(curkey)) {
869 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
870 sizeof(uint8_t) * args->count);
878 static long kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
884 if (args->flags != 0)
887 /* Enforce sane limit on memory allocation */
888 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
891 keys = kmalloc_array(args->count, sizeof(uint8_t),
892 GFP_KERNEL | __GFP_NOWARN);
894 keys = vmalloc(sizeof(uint8_t) * args->count);
898 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
899 sizeof(uint8_t) * args->count);
905 /* Enable storage key handling for the guest */
906 r = s390_enable_skey();
910 for (i = 0; i < args->count; i++) {
911 hva = gfn_to_hva(kvm, args->start_gfn + i);
912 if (kvm_is_error_hva(hva)) {
917 /* Lowest order bit is reserved */
918 if (keys[i] & 0x01) {
923 r = set_guest_storage_key(current->mm, hva,
924 (unsigned long)keys[i], 0);
933 long kvm_arch_vm_ioctl(struct file *filp,
934 unsigned int ioctl, unsigned long arg)
936 struct kvm *kvm = filp->private_data;
937 void __user *argp = (void __user *)arg;
938 struct kvm_device_attr attr;
942 case KVM_S390_INTERRUPT: {
943 struct kvm_s390_interrupt s390int;
946 if (copy_from_user(&s390int, argp, sizeof(s390int)))
948 r = kvm_s390_inject_vm(kvm, &s390int);
951 case KVM_ENABLE_CAP: {
952 struct kvm_enable_cap cap;
954 if (copy_from_user(&cap, argp, sizeof(cap)))
956 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
959 case KVM_CREATE_IRQCHIP: {
960 struct kvm_irq_routing_entry routing;
963 if (kvm->arch.use_irqchip) {
964 /* Set up dummy routing. */
965 memset(&routing, 0, sizeof(routing));
966 r = kvm_set_irq_routing(kvm, &routing, 0, 0);
970 case KVM_SET_DEVICE_ATTR: {
972 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
974 r = kvm_s390_vm_set_attr(kvm, &attr);
977 case KVM_GET_DEVICE_ATTR: {
979 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
981 r = kvm_s390_vm_get_attr(kvm, &attr);
984 case KVM_HAS_DEVICE_ATTR: {
986 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
988 r = kvm_s390_vm_has_attr(kvm, &attr);
991 case KVM_S390_GET_SKEYS: {
992 struct kvm_s390_skeys args;
995 if (copy_from_user(&args, argp,
996 sizeof(struct kvm_s390_skeys)))
998 r = kvm_s390_get_skeys(kvm, &args);
1001 case KVM_S390_SET_SKEYS: {
1002 struct kvm_s390_skeys args;
1005 if (copy_from_user(&args, argp,
1006 sizeof(struct kvm_s390_skeys)))
1008 r = kvm_s390_set_skeys(kvm, &args);
1018 static int kvm_s390_query_ap_config(u8 *config)
1020 u32 fcn_code = 0x04000000UL;
1023 memset(config, 0, 128);
1027 ".long 0xb2af0000\n" /* PQAP(QCI) */
1033 : "r" (fcn_code), "r" (config)
1034 : "cc", "0", "2", "memory"
1040 static int kvm_s390_apxa_installed(void)
1045 if (test_facility(2) && test_facility(12)) {
1046 cc = kvm_s390_query_ap_config(config);
1049 pr_err("PQAP(QCI) failed with cc=%d", cc);
1051 return config[0] & 0x40;
1057 static void kvm_s390_set_crycb_format(struct kvm *kvm)
1059 kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
1061 if (kvm_s390_apxa_installed())
1062 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
1064 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
1067 static void kvm_s390_get_cpu_id(struct cpuid *cpu_id)
1070 cpu_id->version = 0xff;
1073 static int kvm_s390_crypto_init(struct kvm *kvm)
1075 if (!test_kvm_facility(kvm, 76))
1078 kvm->arch.crypto.crycb = kzalloc(sizeof(*kvm->arch.crypto.crycb),
1079 GFP_KERNEL | GFP_DMA);
1080 if (!kvm->arch.crypto.crycb)
1083 kvm_s390_set_crycb_format(kvm);
1085 /* Enable AES/DEA protected key functions by default */
1086 kvm->arch.crypto.aes_kw = 1;
1087 kvm->arch.crypto.dea_kw = 1;
1088 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1089 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1090 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1091 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1096 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
1099 char debug_name[16];
1100 static unsigned long sca_offset;
1103 #ifdef CONFIG_KVM_S390_UCONTROL
1104 if (type & ~KVM_VM_S390_UCONTROL)
1106 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
1113 rc = s390_enable_sie();
1119 kvm->arch.sca = (struct sca_block *) get_zeroed_page(GFP_KERNEL);
1122 spin_lock(&kvm_lock);
1123 sca_offset = (sca_offset + 16) & 0x7f0;
1124 kvm->arch.sca = (struct sca_block *) ((char *) kvm->arch.sca + sca_offset);
1125 spin_unlock(&kvm_lock);
1127 sprintf(debug_name, "kvm-%u", current->pid);
1129 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
1134 * The architectural maximum amount of facilities is 16 kbit. To store
1135 * this amount, 2 kbyte of memory is required. Thus we need a full
1136 * page to hold the guest facility list (arch.model.fac->list) and the
1137 * facility mask (arch.model.fac->mask). Its address size has to be
1138 * 31 bits and word aligned.
1140 kvm->arch.model.fac =
1141 (struct kvm_s390_fac *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1142 if (!kvm->arch.model.fac)
1145 /* Populate the facility mask initially. */
1146 memcpy(kvm->arch.model.fac->mask, S390_lowcore.stfle_fac_list,
1147 S390_ARCH_FAC_LIST_SIZE_BYTE);
1148 for (i = 0; i < S390_ARCH_FAC_LIST_SIZE_U64; i++) {
1149 if (i < kvm_s390_fac_list_mask_size())
1150 kvm->arch.model.fac->mask[i] &= kvm_s390_fac_list_mask[i];
1152 kvm->arch.model.fac->mask[i] = 0UL;
1155 /* Populate the facility list initially. */
1156 memcpy(kvm->arch.model.fac->list, kvm->arch.model.fac->mask,
1157 S390_ARCH_FAC_LIST_SIZE_BYTE);
1159 kvm_s390_get_cpu_id(&kvm->arch.model.cpu_id);
1160 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
1162 if (kvm_s390_crypto_init(kvm) < 0)
1165 spin_lock_init(&kvm->arch.float_int.lock);
1166 for (i = 0; i < FIRQ_LIST_COUNT; i++)
1167 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
1168 init_waitqueue_head(&kvm->arch.ipte_wq);
1169 mutex_init(&kvm->arch.ipte_mutex);
1171 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
1172 VM_EVENT(kvm, 3, "vm created with type %lu", type);
1174 if (type & KVM_VM_S390_UCONTROL) {
1175 kvm->arch.gmap = NULL;
1177 kvm->arch.gmap = gmap_alloc(current->mm, (1UL << 44) - 1);
1178 if (!kvm->arch.gmap)
1180 kvm->arch.gmap->private = kvm;
1181 kvm->arch.gmap->pfault_enabled = 0;
1184 kvm->arch.css_support = 0;
1185 kvm->arch.use_irqchip = 0;
1186 kvm->arch.epoch = 0;
1188 spin_lock_init(&kvm->arch.start_stop_lock);
1189 KVM_EVENT(3, "vm 0x%p created by pid %u", kvm, current->pid);
1193 kfree(kvm->arch.crypto.crycb);
1194 free_page((unsigned long)kvm->arch.model.fac);
1195 debug_unregister(kvm->arch.dbf);
1196 free_page((unsigned long)(kvm->arch.sca));
1197 KVM_EVENT(3, "creation of vm failed: %d", rc);
1201 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1203 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
1204 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
1205 kvm_s390_clear_local_irqs(vcpu);
1206 kvm_clear_async_pf_completion_queue(vcpu);
1207 if (!kvm_is_ucontrol(vcpu->kvm)) {
1208 clear_bit(63 - vcpu->vcpu_id,
1209 (unsigned long *) &vcpu->kvm->arch.sca->mcn);
1210 if (vcpu->kvm->arch.sca->cpu[vcpu->vcpu_id].sda ==
1211 (__u64) vcpu->arch.sie_block)
1212 vcpu->kvm->arch.sca->cpu[vcpu->vcpu_id].sda = 0;
1216 if (kvm_is_ucontrol(vcpu->kvm))
1217 gmap_free(vcpu->arch.gmap);
1219 if (vcpu->kvm->arch.use_cmma)
1220 kvm_s390_vcpu_unsetup_cmma(vcpu);
1221 free_page((unsigned long)(vcpu->arch.sie_block));
1223 kvm_vcpu_uninit(vcpu);
1224 kmem_cache_free(kvm_vcpu_cache, vcpu);
1227 static void kvm_free_vcpus(struct kvm *kvm)
1230 struct kvm_vcpu *vcpu;
1232 kvm_for_each_vcpu(i, vcpu, kvm)
1233 kvm_arch_vcpu_destroy(vcpu);
1235 mutex_lock(&kvm->lock);
1236 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
1237 kvm->vcpus[i] = NULL;
1239 atomic_set(&kvm->online_vcpus, 0);
1240 mutex_unlock(&kvm->lock);
1243 void kvm_arch_destroy_vm(struct kvm *kvm)
1245 kvm_free_vcpus(kvm);
1246 free_page((unsigned long)kvm->arch.model.fac);
1247 free_page((unsigned long)(kvm->arch.sca));
1248 debug_unregister(kvm->arch.dbf);
1249 kfree(kvm->arch.crypto.crycb);
1250 if (!kvm_is_ucontrol(kvm))
1251 gmap_free(kvm->arch.gmap);
1252 kvm_s390_destroy_adapters(kvm);
1253 kvm_s390_clear_float_irqs(kvm);
1254 KVM_EVENT(3, "vm 0x%p destroyed", kvm);
1257 /* Section: vcpu related */
1258 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
1260 vcpu->arch.gmap = gmap_alloc(current->mm, -1UL);
1261 if (!vcpu->arch.gmap)
1263 vcpu->arch.gmap->private = vcpu->kvm;
1268 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1270 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1271 kvm_clear_async_pf_completion_queue(vcpu);
1272 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
1278 if (test_kvm_facility(vcpu->kvm, 129))
1279 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
1281 if (kvm_is_ucontrol(vcpu->kvm))
1282 return __kvm_ucontrol_vcpu_init(vcpu);
1288 * Backs up the current FP/VX register save area on a particular
1289 * destination. Used to switch between different register save
1292 static inline void save_fpu_to(struct fpu *dst)
1294 dst->fpc = current->thread.fpu.fpc;
1295 dst->flags = current->thread.fpu.flags;
1296 dst->regs = current->thread.fpu.regs;
1300 * Switches the FP/VX register save area from which to lazy
1301 * restore register contents.
1303 static inline void load_fpu_from(struct fpu *from)
1305 current->thread.fpu.fpc = from->fpc;
1306 current->thread.fpu.flags = from->flags;
1307 current->thread.fpu.regs = from->regs;
1310 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1312 /* Save host register state */
1314 save_fpu_to(&vcpu->arch.host_fpregs);
1316 if (test_kvm_facility(vcpu->kvm, 129)) {
1317 current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
1318 current->thread.fpu.flags = FPU_USE_VX;
1320 * Use the register save area in the SIE-control block
1321 * for register restore and save in kvm_arch_vcpu_put()
1323 current->thread.fpu.vxrs =
1324 (__vector128 *)&vcpu->run->s.regs.vrs;
1325 /* Always enable the vector extension for KVM */
1328 load_fpu_from(&vcpu->arch.guest_fpregs);
1330 if (test_fp_ctl(current->thread.fpu.fpc))
1331 /* User space provided an invalid FPC, let's clear it */
1332 current->thread.fpu.fpc = 0;
1334 save_access_regs(vcpu->arch.host_acrs);
1335 restore_access_regs(vcpu->run->s.regs.acrs);
1336 gmap_enable(vcpu->arch.gmap);
1337 atomic_or(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1340 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1342 atomic_andnot(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1343 gmap_disable(vcpu->arch.gmap);
1347 if (test_kvm_facility(vcpu->kvm, 129))
1349 * kvm_arch_vcpu_load() set up the register save area to
1350 * the &vcpu->run->s.regs.vrs and, thus, the vector registers
1351 * are already saved. Only the floating-point control must be
1354 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
1356 save_fpu_to(&vcpu->arch.guest_fpregs);
1357 load_fpu_from(&vcpu->arch.host_fpregs);
1359 save_access_regs(vcpu->run->s.regs.acrs);
1360 restore_access_regs(vcpu->arch.host_acrs);
1363 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
1365 /* this equals initial cpu reset in pop, but we don't switch to ESA */
1366 vcpu->arch.sie_block->gpsw.mask = 0UL;
1367 vcpu->arch.sie_block->gpsw.addr = 0UL;
1368 kvm_s390_set_prefix(vcpu, 0);
1369 vcpu->arch.sie_block->cputm = 0UL;
1370 vcpu->arch.sie_block->ckc = 0UL;
1371 vcpu->arch.sie_block->todpr = 0;
1372 memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64));
1373 vcpu->arch.sie_block->gcr[0] = 0xE0UL;
1374 vcpu->arch.sie_block->gcr[14] = 0xC2000000UL;
1375 vcpu->arch.guest_fpregs.fpc = 0;
1376 asm volatile("lfpc %0" : : "Q" (vcpu->arch.guest_fpregs.fpc));
1377 vcpu->arch.sie_block->gbea = 1;
1378 vcpu->arch.sie_block->pp = 0;
1379 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1380 kvm_clear_async_pf_completion_queue(vcpu);
1381 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
1382 kvm_s390_vcpu_stop(vcpu);
1383 kvm_s390_clear_local_irqs(vcpu);
1386 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1388 mutex_lock(&vcpu->kvm->lock);
1390 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
1392 mutex_unlock(&vcpu->kvm->lock);
1393 if (!kvm_is_ucontrol(vcpu->kvm))
1394 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
1397 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
1399 if (!test_kvm_facility(vcpu->kvm, 76))
1402 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
1404 if (vcpu->kvm->arch.crypto.aes_kw)
1405 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
1406 if (vcpu->kvm->arch.crypto.dea_kw)
1407 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
1409 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
1412 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
1414 free_page(vcpu->arch.sie_block->cbrlo);
1415 vcpu->arch.sie_block->cbrlo = 0;
1418 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
1420 vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
1421 if (!vcpu->arch.sie_block->cbrlo)
1424 vcpu->arch.sie_block->ecb2 |= 0x80;
1425 vcpu->arch.sie_block->ecb2 &= ~0x08;
1429 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
1431 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
1433 vcpu->arch.cpu_id = model->cpu_id;
1434 vcpu->arch.sie_block->ibc = model->ibc;
1435 vcpu->arch.sie_block->fac = (int) (long) model->fac->list;
1438 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1442 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
1446 if (test_kvm_facility(vcpu->kvm, 78))
1447 atomic_or(CPUSTAT_GED2, &vcpu->arch.sie_block->cpuflags);
1448 else if (test_kvm_facility(vcpu->kvm, 8))
1449 atomic_or(CPUSTAT_GED, &vcpu->arch.sie_block->cpuflags);
1451 kvm_s390_vcpu_setup_model(vcpu);
1453 vcpu->arch.sie_block->ecb = 6;
1454 if (test_kvm_facility(vcpu->kvm, 50) && test_kvm_facility(vcpu->kvm, 73))
1455 vcpu->arch.sie_block->ecb |= 0x10;
1457 vcpu->arch.sie_block->ecb2 = 8;
1458 vcpu->arch.sie_block->eca = 0xC1002000U;
1460 vcpu->arch.sie_block->eca |= 1;
1461 if (sclp.has_sigpif)
1462 vcpu->arch.sie_block->eca |= 0x10000000U;
1463 if (test_kvm_facility(vcpu->kvm, 129)) {
1464 vcpu->arch.sie_block->eca |= 0x00020000;
1465 vcpu->arch.sie_block->ecd |= 0x20000000;
1467 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
1469 if (vcpu->kvm->arch.use_cmma) {
1470 rc = kvm_s390_vcpu_setup_cmma(vcpu);
1474 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1475 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
1477 kvm_s390_vcpu_crypto_setup(vcpu);
1482 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1485 struct kvm_vcpu *vcpu;
1486 struct sie_page *sie_page;
1489 if (id >= KVM_MAX_VCPUS)
1494 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1498 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
1502 vcpu->arch.sie_block = &sie_page->sie_block;
1503 vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
1505 vcpu->arch.sie_block->icpua = id;
1506 if (!kvm_is_ucontrol(kvm)) {
1507 if (!kvm->arch.sca) {
1511 if (!kvm->arch.sca->cpu[id].sda)
1512 kvm->arch.sca->cpu[id].sda =
1513 (__u64) vcpu->arch.sie_block;
1514 vcpu->arch.sie_block->scaoh =
1515 (__u32)(((__u64)kvm->arch.sca) >> 32);
1516 vcpu->arch.sie_block->scaol = (__u32)(__u64)kvm->arch.sca;
1517 set_bit(63 - id, (unsigned long *) &kvm->arch.sca->mcn);
1520 spin_lock_init(&vcpu->arch.local_int.lock);
1521 vcpu->arch.local_int.float_int = &kvm->arch.float_int;
1522 vcpu->arch.local_int.wq = &vcpu->wq;
1523 vcpu->arch.local_int.cpuflags = &vcpu->arch.sie_block->cpuflags;
1526 * Allocate a save area for floating-point registers. If the vector
1527 * extension is available, register contents are saved in the SIE
1528 * control block. The allocated save area is still required in
1529 * particular places, for example, in kvm_s390_vcpu_store_status().
1531 vcpu->arch.guest_fpregs.fprs = kzalloc(sizeof(freg_t) * __NUM_FPRS,
1533 if (!vcpu->arch.guest_fpregs.fprs) {
1535 goto out_free_sie_block;
1538 rc = kvm_vcpu_init(vcpu, kvm, id);
1540 goto out_free_sie_block;
1541 VM_EVENT(kvm, 3, "create cpu %d at %p, sie block at %p", id, vcpu,
1542 vcpu->arch.sie_block);
1543 trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
1547 free_page((unsigned long)(vcpu->arch.sie_block));
1549 kmem_cache_free(kvm_vcpu_cache, vcpu);
1554 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
1556 return kvm_s390_vcpu_has_irq(vcpu, 0);
1559 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
1561 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
1565 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
1567 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
1570 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
1572 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
1576 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
1578 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
1582 * Kick a guest cpu out of SIE and wait until SIE is not running.
1583 * If the CPU is not running (e.g. waiting as idle) the function will
1584 * return immediately. */
1585 void exit_sie(struct kvm_vcpu *vcpu)
1587 atomic_or(CPUSTAT_STOP_INT, &vcpu->arch.sie_block->cpuflags);
1588 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
1592 /* Kick a guest cpu out of SIE to process a request synchronously */
1593 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
1595 kvm_make_request(req, vcpu);
1596 kvm_s390_vcpu_request(vcpu);
1599 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long address)
1602 struct kvm *kvm = gmap->private;
1603 struct kvm_vcpu *vcpu;
1605 kvm_for_each_vcpu(i, vcpu, kvm) {
1606 /* match against both prefix pages */
1607 if (kvm_s390_get_prefix(vcpu) == (address & ~0x1000UL)) {
1608 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx", address);
1609 kvm_s390_sync_request(KVM_REQ_MMU_RELOAD, vcpu);
1614 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
1616 /* kvm common code refers to this, but never calls it */
1621 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
1622 struct kvm_one_reg *reg)
1627 case KVM_REG_S390_TODPR:
1628 r = put_user(vcpu->arch.sie_block->todpr,
1629 (u32 __user *)reg->addr);
1631 case KVM_REG_S390_EPOCHDIFF:
1632 r = put_user(vcpu->arch.sie_block->epoch,
1633 (u64 __user *)reg->addr);
1635 case KVM_REG_S390_CPU_TIMER:
1636 r = put_user(vcpu->arch.sie_block->cputm,
1637 (u64 __user *)reg->addr);
1639 case KVM_REG_S390_CLOCK_COMP:
1640 r = put_user(vcpu->arch.sie_block->ckc,
1641 (u64 __user *)reg->addr);
1643 case KVM_REG_S390_PFTOKEN:
1644 r = put_user(vcpu->arch.pfault_token,
1645 (u64 __user *)reg->addr);
1647 case KVM_REG_S390_PFCOMPARE:
1648 r = put_user(vcpu->arch.pfault_compare,
1649 (u64 __user *)reg->addr);
1651 case KVM_REG_S390_PFSELECT:
1652 r = put_user(vcpu->arch.pfault_select,
1653 (u64 __user *)reg->addr);
1655 case KVM_REG_S390_PP:
1656 r = put_user(vcpu->arch.sie_block->pp,
1657 (u64 __user *)reg->addr);
1659 case KVM_REG_S390_GBEA:
1660 r = put_user(vcpu->arch.sie_block->gbea,
1661 (u64 __user *)reg->addr);
1670 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
1671 struct kvm_one_reg *reg)
1676 case KVM_REG_S390_TODPR:
1677 r = get_user(vcpu->arch.sie_block->todpr,
1678 (u32 __user *)reg->addr);
1680 case KVM_REG_S390_EPOCHDIFF:
1681 r = get_user(vcpu->arch.sie_block->epoch,
1682 (u64 __user *)reg->addr);
1684 case KVM_REG_S390_CPU_TIMER:
1685 r = get_user(vcpu->arch.sie_block->cputm,
1686 (u64 __user *)reg->addr);
1688 case KVM_REG_S390_CLOCK_COMP:
1689 r = get_user(vcpu->arch.sie_block->ckc,
1690 (u64 __user *)reg->addr);
1692 case KVM_REG_S390_PFTOKEN:
1693 r = get_user(vcpu->arch.pfault_token,
1694 (u64 __user *)reg->addr);
1695 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
1696 kvm_clear_async_pf_completion_queue(vcpu);
1698 case KVM_REG_S390_PFCOMPARE:
1699 r = get_user(vcpu->arch.pfault_compare,
1700 (u64 __user *)reg->addr);
1702 case KVM_REG_S390_PFSELECT:
1703 r = get_user(vcpu->arch.pfault_select,
1704 (u64 __user *)reg->addr);
1706 case KVM_REG_S390_PP:
1707 r = get_user(vcpu->arch.sie_block->pp,
1708 (u64 __user *)reg->addr);
1710 case KVM_REG_S390_GBEA:
1711 r = get_user(vcpu->arch.sie_block->gbea,
1712 (u64 __user *)reg->addr);
1721 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
1723 kvm_s390_vcpu_initial_reset(vcpu);
1727 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1729 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs));
1733 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1735 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
1739 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1740 struct kvm_sregs *sregs)
1742 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
1743 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
1744 restore_access_regs(vcpu->run->s.regs.acrs);
1748 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1749 struct kvm_sregs *sregs)
1751 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
1752 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
1756 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1758 if (test_fp_ctl(fpu->fpc))
1760 memcpy(vcpu->arch.guest_fpregs.fprs, &fpu->fprs, sizeof(fpu->fprs));
1761 vcpu->arch.guest_fpregs.fpc = fpu->fpc;
1763 load_fpu_from(&vcpu->arch.guest_fpregs);
1767 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1769 memcpy(&fpu->fprs, vcpu->arch.guest_fpregs.fprs, sizeof(fpu->fprs));
1770 fpu->fpc = vcpu->arch.guest_fpregs.fpc;
1774 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
1778 if (!is_vcpu_stopped(vcpu))
1781 vcpu->run->psw_mask = psw.mask;
1782 vcpu->run->psw_addr = psw.addr;
1787 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1788 struct kvm_translation *tr)
1790 return -EINVAL; /* not implemented yet */
1793 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
1794 KVM_GUESTDBG_USE_HW_BP | \
1795 KVM_GUESTDBG_ENABLE)
1797 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1798 struct kvm_guest_debug *dbg)
1802 vcpu->guest_debug = 0;
1803 kvm_s390_clear_bp_data(vcpu);
1805 if (dbg->control & ~VALID_GUESTDBG_FLAGS)
1808 if (dbg->control & KVM_GUESTDBG_ENABLE) {
1809 vcpu->guest_debug = dbg->control;
1810 /* enforce guest PER */
1811 atomic_or(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
1813 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
1814 rc = kvm_s390_import_bp_data(vcpu, dbg);
1816 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
1817 vcpu->arch.guestdbg.last_bp = 0;
1821 vcpu->guest_debug = 0;
1822 kvm_s390_clear_bp_data(vcpu);
1823 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
1829 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1830 struct kvm_mp_state *mp_state)
1832 /* CHECK_STOP and LOAD are not supported yet */
1833 return is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
1834 KVM_MP_STATE_OPERATING;
1837 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1838 struct kvm_mp_state *mp_state)
1842 /* user space knows about this interface - let it control the state */
1843 vcpu->kvm->arch.user_cpu_state_ctrl = 1;
1845 switch (mp_state->mp_state) {
1846 case KVM_MP_STATE_STOPPED:
1847 kvm_s390_vcpu_stop(vcpu);
1849 case KVM_MP_STATE_OPERATING:
1850 kvm_s390_vcpu_start(vcpu);
1852 case KVM_MP_STATE_LOAD:
1853 case KVM_MP_STATE_CHECK_STOP:
1854 /* fall through - CHECK_STOP and LOAD are not supported yet */
1862 static bool ibs_enabled(struct kvm_vcpu *vcpu)
1864 return atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_IBS;
1867 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
1870 kvm_s390_vcpu_request_handled(vcpu);
1871 if (!vcpu->requests)
1874 * We use MMU_RELOAD just to re-arm the ipte notifier for the
1875 * guest prefix page. gmap_ipte_notify will wait on the ptl lock.
1876 * This ensures that the ipte instruction for this request has
1877 * already finished. We might race against a second unmapper that
1878 * wants to set the blocking bit. Lets just retry the request loop.
1880 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
1882 rc = gmap_ipte_notify(vcpu->arch.gmap,
1883 kvm_s390_get_prefix(vcpu),
1890 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
1891 vcpu->arch.sie_block->ihcpu = 0xffff;
1895 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
1896 if (!ibs_enabled(vcpu)) {
1897 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
1898 atomic_or(CPUSTAT_IBS,
1899 &vcpu->arch.sie_block->cpuflags);
1904 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
1905 if (ibs_enabled(vcpu)) {
1906 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
1907 atomic_andnot(CPUSTAT_IBS,
1908 &vcpu->arch.sie_block->cpuflags);
1913 /* nothing to do, just clear the request */
1914 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
1920 * kvm_arch_fault_in_page - fault-in guest page if necessary
1921 * @vcpu: The corresponding virtual cpu
1922 * @gpa: Guest physical address
1923 * @writable: Whether the page should be writable or not
1925 * Make sure that a guest page has been faulted-in on the host.
1927 * Return: Zero on success, negative error code otherwise.
1929 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
1931 return gmap_fault(vcpu->arch.gmap, gpa,
1932 writable ? FAULT_FLAG_WRITE : 0);
1935 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
1936 unsigned long token)
1938 struct kvm_s390_interrupt inti;
1939 struct kvm_s390_irq irq;
1942 irq.u.ext.ext_params2 = token;
1943 irq.type = KVM_S390_INT_PFAULT_INIT;
1944 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
1946 inti.type = KVM_S390_INT_PFAULT_DONE;
1947 inti.parm64 = token;
1948 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
1952 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
1953 struct kvm_async_pf *work)
1955 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
1956 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
1959 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
1960 struct kvm_async_pf *work)
1962 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
1963 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
1966 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
1967 struct kvm_async_pf *work)
1969 /* s390 will always inject the page directly */
1972 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
1975 * s390 will always inject the page directly,
1976 * but we still want check_async_completion to cleanup
1981 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
1984 struct kvm_arch_async_pf arch;
1987 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
1989 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
1990 vcpu->arch.pfault_compare)
1992 if (psw_extint_disabled(vcpu))
1994 if (kvm_s390_vcpu_has_irq(vcpu, 0))
1996 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
1998 if (!vcpu->arch.gmap->pfault_enabled)
2001 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
2002 hva += current->thread.gmap_addr & ~PAGE_MASK;
2003 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
2006 rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
2010 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
2015 * On s390 notifications for arriving pages will be delivered directly
2016 * to the guest but the house keeping for completed pfaults is
2017 * handled outside the worker.
2019 kvm_check_async_pf_completion(vcpu);
2021 memcpy(&vcpu->arch.sie_block->gg14, &vcpu->run->s.regs.gprs[14], 16);
2026 if (test_cpu_flag(CIF_MCCK_PENDING))
2029 if (!kvm_is_ucontrol(vcpu->kvm)) {
2030 rc = kvm_s390_deliver_pending_interrupts(vcpu);
2035 rc = kvm_s390_handle_requests(vcpu);
2039 if (guestdbg_enabled(vcpu)) {
2040 kvm_s390_backup_guest_per_regs(vcpu);
2041 kvm_s390_patch_guest_per_regs(vcpu);
2044 vcpu->arch.sie_block->icptcode = 0;
2045 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
2046 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
2047 trace_kvm_s390_sie_enter(vcpu, cpuflags);
2052 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
2054 psw_t *psw = &vcpu->arch.sie_block->gpsw;
2058 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
2059 trace_kvm_s390_sie_fault(vcpu);
2062 * We want to inject an addressing exception, which is defined as a
2063 * suppressing or terminating exception. However, since we came here
2064 * by a DAT access exception, the PSW still points to the faulting
2065 * instruction since DAT exceptions are nullifying. So we've got
2066 * to look up the current opcode to get the length of the instruction
2067 * to be able to forward the PSW.
2069 rc = read_guest(vcpu, psw->addr, 0, &opcode, 1);
2071 return kvm_s390_inject_prog_cond(vcpu, rc);
2072 psw->addr = __rewind_psw(*psw, -insn_length(opcode));
2074 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
2077 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
2081 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
2082 vcpu->arch.sie_block->icptcode);
2083 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
2085 if (guestdbg_enabled(vcpu))
2086 kvm_s390_restore_guest_per_regs(vcpu);
2088 if (exit_reason >= 0) {
2090 } else if (kvm_is_ucontrol(vcpu->kvm)) {
2091 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
2092 vcpu->run->s390_ucontrol.trans_exc_code =
2093 current->thread.gmap_addr;
2094 vcpu->run->s390_ucontrol.pgm_code = 0x10;
2097 } else if (current->thread.gmap_pfault) {
2098 trace_kvm_s390_major_guest_pfault(vcpu);
2099 current->thread.gmap_pfault = 0;
2100 if (kvm_arch_setup_async_pf(vcpu)) {
2103 gpa_t gpa = current->thread.gmap_addr;
2104 rc = kvm_arch_fault_in_page(vcpu, gpa, 1);
2109 rc = vcpu_post_run_fault_in_sie(vcpu);
2111 memcpy(&vcpu->run->s.regs.gprs[14], &vcpu->arch.sie_block->gg14, 16);
2114 if (kvm_is_ucontrol(vcpu->kvm))
2115 /* Don't exit for host interrupts. */
2116 rc = vcpu->arch.sie_block->icptcode ? -EOPNOTSUPP : 0;
2118 rc = kvm_handle_sie_intercept(vcpu);
2124 static int __vcpu_run(struct kvm_vcpu *vcpu)
2126 int rc, exit_reason;
2129 * We try to hold kvm->srcu during most of vcpu_run (except when run-
2130 * ning the guest), so that memslots (and other stuff) are protected
2132 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2135 rc = vcpu_pre_run(vcpu);
2139 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2141 * As PF_VCPU will be used in fault handler, between
2142 * guest_enter and guest_exit should be no uaccess.
2144 local_irq_disable();
2145 __kvm_guest_enter();
2147 exit_reason = sie64a(vcpu->arch.sie_block,
2148 vcpu->run->s.regs.gprs);
2149 local_irq_disable();
2152 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2154 rc = vcpu_post_run(vcpu, exit_reason);
2155 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
2157 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2161 static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2163 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
2164 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
2165 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
2166 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
2167 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
2168 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
2169 /* some control register changes require a tlb flush */
2170 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2172 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
2173 vcpu->arch.sie_block->cputm = kvm_run->s.regs.cputm;
2174 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
2175 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
2176 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
2177 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
2179 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
2180 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
2181 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
2182 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
2183 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2184 kvm_clear_async_pf_completion_queue(vcpu);
2186 kvm_run->kvm_dirty_regs = 0;
2189 static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2191 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
2192 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
2193 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
2194 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
2195 kvm_run->s.regs.cputm = vcpu->arch.sie_block->cputm;
2196 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
2197 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
2198 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
2199 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
2200 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
2201 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
2202 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
2205 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2210 if (guestdbg_exit_pending(vcpu)) {
2211 kvm_s390_prepare_debug_exit(vcpu);
2215 if (vcpu->sigset_active)
2216 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2218 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
2219 kvm_s390_vcpu_start(vcpu);
2220 } else if (is_vcpu_stopped(vcpu)) {
2221 pr_err_ratelimited("can't run stopped vcpu %d\n",
2226 sync_regs(vcpu, kvm_run);
2229 rc = __vcpu_run(vcpu);
2231 if (signal_pending(current) && !rc) {
2232 kvm_run->exit_reason = KVM_EXIT_INTR;
2236 if (guestdbg_exit_pending(vcpu) && !rc) {
2237 kvm_s390_prepare_debug_exit(vcpu);
2241 if (rc == -EOPNOTSUPP) {
2242 /* intercept cannot be handled in-kernel, prepare kvm-run */
2243 kvm_run->exit_reason = KVM_EXIT_S390_SIEIC;
2244 kvm_run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
2245 kvm_run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
2246 kvm_run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
2250 if (rc == -EREMOTE) {
2251 /* intercept was handled, but userspace support is needed
2252 * kvm_run has been prepared by the handler */
2256 store_regs(vcpu, kvm_run);
2258 if (vcpu->sigset_active)
2259 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2261 vcpu->stat.exit_userspace++;
2266 * store status at address
2267 * we use have two special cases:
2268 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
2269 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
2271 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
2273 unsigned char archmode = 1;
2278 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
2279 if (write_guest_abs(vcpu, 163, &archmode, 1))
2281 gpa = SAVE_AREA_BASE;
2282 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
2283 if (write_guest_real(vcpu, 163, &archmode, 1))
2285 gpa = kvm_s390_real_to_abs(vcpu, SAVE_AREA_BASE);
2287 rc = write_guest_abs(vcpu, gpa + offsetof(struct save_area, fp_regs),
2288 vcpu->arch.guest_fpregs.fprs, 128);
2289 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, gp_regs),
2290 vcpu->run->s.regs.gprs, 128);
2291 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, psw),
2292 &vcpu->arch.sie_block->gpsw, 16);
2293 px = kvm_s390_get_prefix(vcpu);
2294 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, pref_reg),
2296 rc |= write_guest_abs(vcpu,
2297 gpa + offsetof(struct save_area, fp_ctrl_reg),
2298 &vcpu->arch.guest_fpregs.fpc, 4);
2299 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, tod_reg),
2300 &vcpu->arch.sie_block->todpr, 4);
2301 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, timer),
2302 &vcpu->arch.sie_block->cputm, 8);
2303 clkcomp = vcpu->arch.sie_block->ckc >> 8;
2304 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, clk_cmp),
2306 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, acc_regs),
2307 &vcpu->run->s.regs.acrs, 64);
2308 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, ctrl_regs),
2309 &vcpu->arch.sie_block->gcr, 128);
2310 return rc ? -EFAULT : 0;
2313 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
2316 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
2317 * copying in vcpu load/put. Lets update our copies before we save
2318 * it into the save area
2321 if (test_kvm_facility(vcpu->kvm, 129)) {
2323 * If the vector extension is available, the vector registers
2324 * which overlaps with floating-point registers are saved in
2325 * the SIE-control block. Hence, extract the floating-point
2326 * registers and the FPC value and store them in the
2327 * guest_fpregs structure.
2329 WARN_ON(!is_vx_task(current)); /* XXX remove later */
2330 vcpu->arch.guest_fpregs.fpc = current->thread.fpu.fpc;
2331 convert_vx_to_fp(vcpu->arch.guest_fpregs.fprs,
2332 current->thread.fpu.vxrs);
2334 save_fpu_to(&vcpu->arch.guest_fpregs);
2335 save_access_regs(vcpu->run->s.regs.acrs);
2337 return kvm_s390_store_status_unloaded(vcpu, addr);
2341 * store additional status at address
2343 int kvm_s390_store_adtl_status_unloaded(struct kvm_vcpu *vcpu,
2346 /* Only bits 0-53 are used for address formation */
2347 if (!(gpa & ~0x3ff))
2350 return write_guest_abs(vcpu, gpa & ~0x3ff,
2351 (void *)&vcpu->run->s.regs.vrs, 512);
2354 int kvm_s390_vcpu_store_adtl_status(struct kvm_vcpu *vcpu, unsigned long addr)
2356 if (!test_kvm_facility(vcpu->kvm, 129))
2360 * The guest VXRS are in the host VXRs due to the lazy
2361 * copying in vcpu load/put. We can simply call save_fpu_regs()
2362 * to save the current register state because we are in the
2363 * middle of a load/put cycle.
2365 * Let's update our copies before we save it into the save area.
2369 return kvm_s390_store_adtl_status_unloaded(vcpu, addr);
2372 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2374 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
2375 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
2378 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
2381 struct kvm_vcpu *vcpu;
2383 kvm_for_each_vcpu(i, vcpu, kvm) {
2384 __disable_ibs_on_vcpu(vcpu);
2388 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2390 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
2391 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
2394 void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
2396 int i, online_vcpus, started_vcpus = 0;
2398 if (!is_vcpu_stopped(vcpu))
2401 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
2402 /* Only one cpu at a time may enter/leave the STOPPED state. */
2403 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2404 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2406 for (i = 0; i < online_vcpus; i++) {
2407 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i]))
2411 if (started_vcpus == 0) {
2412 /* we're the only active VCPU -> speed it up */
2413 __enable_ibs_on_vcpu(vcpu);
2414 } else if (started_vcpus == 1) {
2416 * As we are starting a second VCPU, we have to disable
2417 * the IBS facility on all VCPUs to remove potentially
2418 * oustanding ENABLE requests.
2420 __disable_ibs_on_all_vcpus(vcpu->kvm);
2423 atomic_andnot(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2425 * Another VCPU might have used IBS while we were offline.
2426 * Let's play safe and flush the VCPU at startup.
2428 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2429 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2433 void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
2435 int i, online_vcpus, started_vcpus = 0;
2436 struct kvm_vcpu *started_vcpu = NULL;
2438 if (is_vcpu_stopped(vcpu))
2441 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
2442 /* Only one cpu at a time may enter/leave the STOPPED state. */
2443 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2444 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2446 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
2447 kvm_s390_clear_stop_irq(vcpu);
2449 atomic_or(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2450 __disable_ibs_on_vcpu(vcpu);
2452 for (i = 0; i < online_vcpus; i++) {
2453 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) {
2455 started_vcpu = vcpu->kvm->vcpus[i];
2459 if (started_vcpus == 1) {
2461 * As we only have one VCPU left, we want to enable the
2462 * IBS facility for that VCPU to speed it up.
2464 __enable_ibs_on_vcpu(started_vcpu);
2467 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2471 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
2472 struct kvm_enable_cap *cap)
2480 case KVM_CAP_S390_CSS_SUPPORT:
2481 if (!vcpu->kvm->arch.css_support) {
2482 vcpu->kvm->arch.css_support = 1;
2483 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
2484 trace_kvm_s390_enable_css(vcpu->kvm);
2495 static long kvm_s390_guest_mem_op(struct kvm_vcpu *vcpu,
2496 struct kvm_s390_mem_op *mop)
2498 void __user *uaddr = (void __user *)mop->buf;
2499 void *tmpbuf = NULL;
2501 const u64 supported_flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION
2502 | KVM_S390_MEMOP_F_CHECK_ONLY;
2504 if (mop->flags & ~supported_flags)
2507 if (mop->size > MEM_OP_MAX_SIZE)
2510 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
2511 tmpbuf = vmalloc(mop->size);
2516 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2519 case KVM_S390_MEMOP_LOGICAL_READ:
2520 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2521 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size, false);
2524 r = read_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
2526 if (copy_to_user(uaddr, tmpbuf, mop->size))
2530 case KVM_S390_MEMOP_LOGICAL_WRITE:
2531 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2532 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size, true);
2535 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
2539 r = write_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
2545 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2547 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
2548 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
2554 long kvm_arch_vcpu_ioctl(struct file *filp,
2555 unsigned int ioctl, unsigned long arg)
2557 struct kvm_vcpu *vcpu = filp->private_data;
2558 void __user *argp = (void __user *)arg;
2563 case KVM_S390_IRQ: {
2564 struct kvm_s390_irq s390irq;
2567 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
2569 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
2572 case KVM_S390_INTERRUPT: {
2573 struct kvm_s390_interrupt s390int;
2574 struct kvm_s390_irq s390irq;
2577 if (copy_from_user(&s390int, argp, sizeof(s390int)))
2579 if (s390int_to_s390irq(&s390int, &s390irq))
2581 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
2584 case KVM_S390_STORE_STATUS:
2585 idx = srcu_read_lock(&vcpu->kvm->srcu);
2586 r = kvm_s390_vcpu_store_status(vcpu, arg);
2587 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2589 case KVM_S390_SET_INITIAL_PSW: {
2593 if (copy_from_user(&psw, argp, sizeof(psw)))
2595 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
2598 case KVM_S390_INITIAL_RESET:
2599 r = kvm_arch_vcpu_ioctl_initial_reset(vcpu);
2601 case KVM_SET_ONE_REG:
2602 case KVM_GET_ONE_REG: {
2603 struct kvm_one_reg reg;
2605 if (copy_from_user(®, argp, sizeof(reg)))
2607 if (ioctl == KVM_SET_ONE_REG)
2608 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®);
2610 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®);
2613 #ifdef CONFIG_KVM_S390_UCONTROL
2614 case KVM_S390_UCAS_MAP: {
2615 struct kvm_s390_ucas_mapping ucasmap;
2617 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
2622 if (!kvm_is_ucontrol(vcpu->kvm)) {
2627 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
2628 ucasmap.vcpu_addr, ucasmap.length);
2631 case KVM_S390_UCAS_UNMAP: {
2632 struct kvm_s390_ucas_mapping ucasmap;
2634 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
2639 if (!kvm_is_ucontrol(vcpu->kvm)) {
2644 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
2649 case KVM_S390_VCPU_FAULT: {
2650 r = gmap_fault(vcpu->arch.gmap, arg, 0);
2653 case KVM_ENABLE_CAP:
2655 struct kvm_enable_cap cap;
2657 if (copy_from_user(&cap, argp, sizeof(cap)))
2659 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
2662 case KVM_S390_MEM_OP: {
2663 struct kvm_s390_mem_op mem_op;
2665 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
2666 r = kvm_s390_guest_mem_op(vcpu, &mem_op);
2671 case KVM_S390_SET_IRQ_STATE: {
2672 struct kvm_s390_irq_state irq_state;
2675 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
2677 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
2678 irq_state.len == 0 ||
2679 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
2683 r = kvm_s390_set_irq_state(vcpu,
2684 (void __user *) irq_state.buf,
2688 case KVM_S390_GET_IRQ_STATE: {
2689 struct kvm_s390_irq_state irq_state;
2692 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
2694 if (irq_state.len == 0) {
2698 r = kvm_s390_get_irq_state(vcpu,
2699 (__u8 __user *) irq_state.buf,
2709 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2711 #ifdef CONFIG_KVM_S390_UCONTROL
2712 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
2713 && (kvm_is_ucontrol(vcpu->kvm))) {
2714 vmf->page = virt_to_page(vcpu->arch.sie_block);
2715 get_page(vmf->page);
2719 return VM_FAULT_SIGBUS;
2722 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
2723 unsigned long npages)
2728 /* Section: memory related */
2729 int kvm_arch_prepare_memory_region(struct kvm *kvm,
2730 struct kvm_memory_slot *memslot,
2731 const struct kvm_userspace_memory_region *mem,
2732 enum kvm_mr_change change)
2734 /* A few sanity checks. We can have memory slots which have to be
2735 located/ended at a segment boundary (1MB). The memory in userland is
2736 ok to be fragmented into various different vmas. It is okay to mmap()
2737 and munmap() stuff in this slot after doing this call at any time */
2739 if (mem->userspace_addr & 0xffffful)
2742 if (mem->memory_size & 0xffffful)
2748 void kvm_arch_commit_memory_region(struct kvm *kvm,
2749 const struct kvm_userspace_memory_region *mem,
2750 const struct kvm_memory_slot *old,
2751 const struct kvm_memory_slot *new,
2752 enum kvm_mr_change change)
2756 /* If the basics of the memslot do not change, we do not want
2757 * to update the gmap. Every update causes several unnecessary
2758 * segment translation exceptions. This is usually handled just
2759 * fine by the normal fault handler + gmap, but it will also
2760 * cause faults on the prefix page of running guest CPUs.
2762 if (old->userspace_addr == mem->userspace_addr &&
2763 old->base_gfn * PAGE_SIZE == mem->guest_phys_addr &&
2764 old->npages * PAGE_SIZE == mem->memory_size)
2767 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr,
2768 mem->guest_phys_addr, mem->memory_size);
2770 pr_warn("failed to commit memory region\n");
2774 static int __init kvm_s390_init(void)
2776 return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2779 static void __exit kvm_s390_exit(void)
2784 module_init(kvm_s390_init);
2785 module_exit(kvm_s390_exit);
2788 * Enable autoloading of the kvm module.
2789 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
2790 * since x86 takes a different approach.
2792 #include <linux/miscdevice.h>
2793 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2794 MODULE_ALIAS("devname:kvm");