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->regs = current->thread.fpu.regs;
1299 * Switches the FP/VX register save area from which to lazy
1300 * restore register contents.
1302 static inline void load_fpu_from(struct fpu *from)
1304 current->thread.fpu.fpc = from->fpc;
1305 current->thread.fpu.regs = from->regs;
1308 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1310 /* Save host register state */
1312 save_fpu_to(&vcpu->arch.host_fpregs);
1314 if (test_kvm_facility(vcpu->kvm, 129)) {
1315 current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
1317 * Use the register save area in the SIE-control block
1318 * for register restore and save in kvm_arch_vcpu_put()
1320 current->thread.fpu.vxrs =
1321 (__vector128 *)&vcpu->run->s.regs.vrs;
1323 load_fpu_from(&vcpu->arch.guest_fpregs);
1325 if (test_fp_ctl(current->thread.fpu.fpc))
1326 /* User space provided an invalid FPC, let's clear it */
1327 current->thread.fpu.fpc = 0;
1329 save_access_regs(vcpu->arch.host_acrs);
1330 restore_access_regs(vcpu->run->s.regs.acrs);
1331 gmap_enable(vcpu->arch.gmap);
1332 atomic_or(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1335 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1337 atomic_andnot(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1338 gmap_disable(vcpu->arch.gmap);
1342 if (test_kvm_facility(vcpu->kvm, 129))
1344 * kvm_arch_vcpu_load() set up the register save area to
1345 * the &vcpu->run->s.regs.vrs and, thus, the vector registers
1346 * are already saved. Only the floating-point control must be
1349 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
1351 save_fpu_to(&vcpu->arch.guest_fpregs);
1352 load_fpu_from(&vcpu->arch.host_fpregs);
1354 save_access_regs(vcpu->run->s.regs.acrs);
1355 restore_access_regs(vcpu->arch.host_acrs);
1358 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
1360 /* this equals initial cpu reset in pop, but we don't switch to ESA */
1361 vcpu->arch.sie_block->gpsw.mask = 0UL;
1362 vcpu->arch.sie_block->gpsw.addr = 0UL;
1363 kvm_s390_set_prefix(vcpu, 0);
1364 vcpu->arch.sie_block->cputm = 0UL;
1365 vcpu->arch.sie_block->ckc = 0UL;
1366 vcpu->arch.sie_block->todpr = 0;
1367 memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64));
1368 vcpu->arch.sie_block->gcr[0] = 0xE0UL;
1369 vcpu->arch.sie_block->gcr[14] = 0xC2000000UL;
1370 vcpu->arch.guest_fpregs.fpc = 0;
1371 asm volatile("lfpc %0" : : "Q" (vcpu->arch.guest_fpregs.fpc));
1372 vcpu->arch.sie_block->gbea = 1;
1373 vcpu->arch.sie_block->pp = 0;
1374 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1375 kvm_clear_async_pf_completion_queue(vcpu);
1376 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
1377 kvm_s390_vcpu_stop(vcpu);
1378 kvm_s390_clear_local_irqs(vcpu);
1381 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1383 mutex_lock(&vcpu->kvm->lock);
1385 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
1387 mutex_unlock(&vcpu->kvm->lock);
1388 if (!kvm_is_ucontrol(vcpu->kvm))
1389 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
1392 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
1394 if (!test_kvm_facility(vcpu->kvm, 76))
1397 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
1399 if (vcpu->kvm->arch.crypto.aes_kw)
1400 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
1401 if (vcpu->kvm->arch.crypto.dea_kw)
1402 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
1404 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
1407 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
1409 free_page(vcpu->arch.sie_block->cbrlo);
1410 vcpu->arch.sie_block->cbrlo = 0;
1413 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
1415 vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
1416 if (!vcpu->arch.sie_block->cbrlo)
1419 vcpu->arch.sie_block->ecb2 |= 0x80;
1420 vcpu->arch.sie_block->ecb2 &= ~0x08;
1424 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
1426 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
1428 vcpu->arch.cpu_id = model->cpu_id;
1429 vcpu->arch.sie_block->ibc = model->ibc;
1430 vcpu->arch.sie_block->fac = (int) (long) model->fac->list;
1433 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1437 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
1441 if (test_kvm_facility(vcpu->kvm, 78))
1442 atomic_or(CPUSTAT_GED2, &vcpu->arch.sie_block->cpuflags);
1443 else if (test_kvm_facility(vcpu->kvm, 8))
1444 atomic_or(CPUSTAT_GED, &vcpu->arch.sie_block->cpuflags);
1446 kvm_s390_vcpu_setup_model(vcpu);
1448 vcpu->arch.sie_block->ecb = 6;
1449 if (test_kvm_facility(vcpu->kvm, 50) && test_kvm_facility(vcpu->kvm, 73))
1450 vcpu->arch.sie_block->ecb |= 0x10;
1452 vcpu->arch.sie_block->ecb2 = 8;
1453 vcpu->arch.sie_block->eca = 0xC1002000U;
1455 vcpu->arch.sie_block->eca |= 1;
1456 if (sclp.has_sigpif)
1457 vcpu->arch.sie_block->eca |= 0x10000000U;
1458 if (test_kvm_facility(vcpu->kvm, 129)) {
1459 vcpu->arch.sie_block->eca |= 0x00020000;
1460 vcpu->arch.sie_block->ecd |= 0x20000000;
1462 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
1464 if (vcpu->kvm->arch.use_cmma) {
1465 rc = kvm_s390_vcpu_setup_cmma(vcpu);
1469 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1470 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
1472 kvm_s390_vcpu_crypto_setup(vcpu);
1477 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1480 struct kvm_vcpu *vcpu;
1481 struct sie_page *sie_page;
1484 if (id >= KVM_MAX_VCPUS)
1489 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1493 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
1497 vcpu->arch.sie_block = &sie_page->sie_block;
1498 vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
1500 vcpu->arch.sie_block->icpua = id;
1501 if (!kvm_is_ucontrol(kvm)) {
1502 if (!kvm->arch.sca) {
1506 if (!kvm->arch.sca->cpu[id].sda)
1507 kvm->arch.sca->cpu[id].sda =
1508 (__u64) vcpu->arch.sie_block;
1509 vcpu->arch.sie_block->scaoh =
1510 (__u32)(((__u64)kvm->arch.sca) >> 32);
1511 vcpu->arch.sie_block->scaol = (__u32)(__u64)kvm->arch.sca;
1512 set_bit(63 - id, (unsigned long *) &kvm->arch.sca->mcn);
1515 spin_lock_init(&vcpu->arch.local_int.lock);
1516 vcpu->arch.local_int.float_int = &kvm->arch.float_int;
1517 vcpu->arch.local_int.wq = &vcpu->wq;
1518 vcpu->arch.local_int.cpuflags = &vcpu->arch.sie_block->cpuflags;
1521 * Allocate a save area for floating-point registers. If the vector
1522 * extension is available, register contents are saved in the SIE
1523 * control block. The allocated save area is still required in
1524 * particular places, for example, in kvm_s390_vcpu_store_status().
1526 vcpu->arch.guest_fpregs.fprs = kzalloc(sizeof(freg_t) * __NUM_FPRS,
1528 if (!vcpu->arch.guest_fpregs.fprs) {
1530 goto out_free_sie_block;
1533 rc = kvm_vcpu_init(vcpu, kvm, id);
1535 goto out_free_sie_block;
1536 VM_EVENT(kvm, 3, "create cpu %d at %p, sie block at %p", id, vcpu,
1537 vcpu->arch.sie_block);
1538 trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
1542 free_page((unsigned long)(vcpu->arch.sie_block));
1544 kmem_cache_free(kvm_vcpu_cache, vcpu);
1549 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
1551 return kvm_s390_vcpu_has_irq(vcpu, 0);
1554 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
1556 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
1560 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
1562 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
1565 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
1567 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
1571 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
1573 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
1577 * Kick a guest cpu out of SIE and wait until SIE is not running.
1578 * If the CPU is not running (e.g. waiting as idle) the function will
1579 * return immediately. */
1580 void exit_sie(struct kvm_vcpu *vcpu)
1582 atomic_or(CPUSTAT_STOP_INT, &vcpu->arch.sie_block->cpuflags);
1583 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
1587 /* Kick a guest cpu out of SIE to process a request synchronously */
1588 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
1590 kvm_make_request(req, vcpu);
1591 kvm_s390_vcpu_request(vcpu);
1594 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long address)
1597 struct kvm *kvm = gmap->private;
1598 struct kvm_vcpu *vcpu;
1600 kvm_for_each_vcpu(i, vcpu, kvm) {
1601 /* match against both prefix pages */
1602 if (kvm_s390_get_prefix(vcpu) == (address & ~0x1000UL)) {
1603 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx", address);
1604 kvm_s390_sync_request(KVM_REQ_MMU_RELOAD, vcpu);
1609 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
1611 /* kvm common code refers to this, but never calls it */
1616 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
1617 struct kvm_one_reg *reg)
1622 case KVM_REG_S390_TODPR:
1623 r = put_user(vcpu->arch.sie_block->todpr,
1624 (u32 __user *)reg->addr);
1626 case KVM_REG_S390_EPOCHDIFF:
1627 r = put_user(vcpu->arch.sie_block->epoch,
1628 (u64 __user *)reg->addr);
1630 case KVM_REG_S390_CPU_TIMER:
1631 r = put_user(vcpu->arch.sie_block->cputm,
1632 (u64 __user *)reg->addr);
1634 case KVM_REG_S390_CLOCK_COMP:
1635 r = put_user(vcpu->arch.sie_block->ckc,
1636 (u64 __user *)reg->addr);
1638 case KVM_REG_S390_PFTOKEN:
1639 r = put_user(vcpu->arch.pfault_token,
1640 (u64 __user *)reg->addr);
1642 case KVM_REG_S390_PFCOMPARE:
1643 r = put_user(vcpu->arch.pfault_compare,
1644 (u64 __user *)reg->addr);
1646 case KVM_REG_S390_PFSELECT:
1647 r = put_user(vcpu->arch.pfault_select,
1648 (u64 __user *)reg->addr);
1650 case KVM_REG_S390_PP:
1651 r = put_user(vcpu->arch.sie_block->pp,
1652 (u64 __user *)reg->addr);
1654 case KVM_REG_S390_GBEA:
1655 r = put_user(vcpu->arch.sie_block->gbea,
1656 (u64 __user *)reg->addr);
1665 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
1666 struct kvm_one_reg *reg)
1671 case KVM_REG_S390_TODPR:
1672 r = get_user(vcpu->arch.sie_block->todpr,
1673 (u32 __user *)reg->addr);
1675 case KVM_REG_S390_EPOCHDIFF:
1676 r = get_user(vcpu->arch.sie_block->epoch,
1677 (u64 __user *)reg->addr);
1679 case KVM_REG_S390_CPU_TIMER:
1680 r = get_user(vcpu->arch.sie_block->cputm,
1681 (u64 __user *)reg->addr);
1683 case KVM_REG_S390_CLOCK_COMP:
1684 r = get_user(vcpu->arch.sie_block->ckc,
1685 (u64 __user *)reg->addr);
1687 case KVM_REG_S390_PFTOKEN:
1688 r = get_user(vcpu->arch.pfault_token,
1689 (u64 __user *)reg->addr);
1690 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
1691 kvm_clear_async_pf_completion_queue(vcpu);
1693 case KVM_REG_S390_PFCOMPARE:
1694 r = get_user(vcpu->arch.pfault_compare,
1695 (u64 __user *)reg->addr);
1697 case KVM_REG_S390_PFSELECT:
1698 r = get_user(vcpu->arch.pfault_select,
1699 (u64 __user *)reg->addr);
1701 case KVM_REG_S390_PP:
1702 r = get_user(vcpu->arch.sie_block->pp,
1703 (u64 __user *)reg->addr);
1705 case KVM_REG_S390_GBEA:
1706 r = get_user(vcpu->arch.sie_block->gbea,
1707 (u64 __user *)reg->addr);
1716 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
1718 kvm_s390_vcpu_initial_reset(vcpu);
1722 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1724 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs));
1728 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1730 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
1734 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1735 struct kvm_sregs *sregs)
1737 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
1738 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
1739 restore_access_regs(vcpu->run->s.regs.acrs);
1743 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1744 struct kvm_sregs *sregs)
1746 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
1747 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
1751 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1753 if (test_fp_ctl(fpu->fpc))
1755 memcpy(vcpu->arch.guest_fpregs.fprs, &fpu->fprs, sizeof(fpu->fprs));
1756 vcpu->arch.guest_fpregs.fpc = fpu->fpc;
1758 load_fpu_from(&vcpu->arch.guest_fpregs);
1762 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
1764 memcpy(&fpu->fprs, vcpu->arch.guest_fpregs.fprs, sizeof(fpu->fprs));
1765 fpu->fpc = vcpu->arch.guest_fpregs.fpc;
1769 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
1773 if (!is_vcpu_stopped(vcpu))
1776 vcpu->run->psw_mask = psw.mask;
1777 vcpu->run->psw_addr = psw.addr;
1782 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1783 struct kvm_translation *tr)
1785 return -EINVAL; /* not implemented yet */
1788 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
1789 KVM_GUESTDBG_USE_HW_BP | \
1790 KVM_GUESTDBG_ENABLE)
1792 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
1793 struct kvm_guest_debug *dbg)
1797 vcpu->guest_debug = 0;
1798 kvm_s390_clear_bp_data(vcpu);
1800 if (dbg->control & ~VALID_GUESTDBG_FLAGS)
1803 if (dbg->control & KVM_GUESTDBG_ENABLE) {
1804 vcpu->guest_debug = dbg->control;
1805 /* enforce guest PER */
1806 atomic_or(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
1808 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
1809 rc = kvm_s390_import_bp_data(vcpu, dbg);
1811 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
1812 vcpu->arch.guestdbg.last_bp = 0;
1816 vcpu->guest_debug = 0;
1817 kvm_s390_clear_bp_data(vcpu);
1818 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
1824 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
1825 struct kvm_mp_state *mp_state)
1827 /* CHECK_STOP and LOAD are not supported yet */
1828 return is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
1829 KVM_MP_STATE_OPERATING;
1832 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
1833 struct kvm_mp_state *mp_state)
1837 /* user space knows about this interface - let it control the state */
1838 vcpu->kvm->arch.user_cpu_state_ctrl = 1;
1840 switch (mp_state->mp_state) {
1841 case KVM_MP_STATE_STOPPED:
1842 kvm_s390_vcpu_stop(vcpu);
1844 case KVM_MP_STATE_OPERATING:
1845 kvm_s390_vcpu_start(vcpu);
1847 case KVM_MP_STATE_LOAD:
1848 case KVM_MP_STATE_CHECK_STOP:
1849 /* fall through - CHECK_STOP and LOAD are not supported yet */
1857 static bool ibs_enabled(struct kvm_vcpu *vcpu)
1859 return atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_IBS;
1862 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
1865 kvm_s390_vcpu_request_handled(vcpu);
1866 if (!vcpu->requests)
1869 * We use MMU_RELOAD just to re-arm the ipte notifier for the
1870 * guest prefix page. gmap_ipte_notify will wait on the ptl lock.
1871 * This ensures that the ipte instruction for this request has
1872 * already finished. We might race against a second unmapper that
1873 * wants to set the blocking bit. Lets just retry the request loop.
1875 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
1877 rc = gmap_ipte_notify(vcpu->arch.gmap,
1878 kvm_s390_get_prefix(vcpu),
1885 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
1886 vcpu->arch.sie_block->ihcpu = 0xffff;
1890 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
1891 if (!ibs_enabled(vcpu)) {
1892 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
1893 atomic_or(CPUSTAT_IBS,
1894 &vcpu->arch.sie_block->cpuflags);
1899 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
1900 if (ibs_enabled(vcpu)) {
1901 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
1902 atomic_andnot(CPUSTAT_IBS,
1903 &vcpu->arch.sie_block->cpuflags);
1908 /* nothing to do, just clear the request */
1909 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
1915 * kvm_arch_fault_in_page - fault-in guest page if necessary
1916 * @vcpu: The corresponding virtual cpu
1917 * @gpa: Guest physical address
1918 * @writable: Whether the page should be writable or not
1920 * Make sure that a guest page has been faulted-in on the host.
1922 * Return: Zero on success, negative error code otherwise.
1924 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
1926 return gmap_fault(vcpu->arch.gmap, gpa,
1927 writable ? FAULT_FLAG_WRITE : 0);
1930 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
1931 unsigned long token)
1933 struct kvm_s390_interrupt inti;
1934 struct kvm_s390_irq irq;
1937 irq.u.ext.ext_params2 = token;
1938 irq.type = KVM_S390_INT_PFAULT_INIT;
1939 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
1941 inti.type = KVM_S390_INT_PFAULT_DONE;
1942 inti.parm64 = token;
1943 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
1947 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
1948 struct kvm_async_pf *work)
1950 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
1951 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
1954 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
1955 struct kvm_async_pf *work)
1957 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
1958 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
1961 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
1962 struct kvm_async_pf *work)
1964 /* s390 will always inject the page directly */
1967 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
1970 * s390 will always inject the page directly,
1971 * but we still want check_async_completion to cleanup
1976 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
1979 struct kvm_arch_async_pf arch;
1982 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
1984 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
1985 vcpu->arch.pfault_compare)
1987 if (psw_extint_disabled(vcpu))
1989 if (kvm_s390_vcpu_has_irq(vcpu, 0))
1991 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
1993 if (!vcpu->arch.gmap->pfault_enabled)
1996 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
1997 hva += current->thread.gmap_addr & ~PAGE_MASK;
1998 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
2001 rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
2005 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
2010 * On s390 notifications for arriving pages will be delivered directly
2011 * to the guest but the house keeping for completed pfaults is
2012 * handled outside the worker.
2014 kvm_check_async_pf_completion(vcpu);
2016 memcpy(&vcpu->arch.sie_block->gg14, &vcpu->run->s.regs.gprs[14], 16);
2021 if (test_cpu_flag(CIF_MCCK_PENDING))
2024 if (!kvm_is_ucontrol(vcpu->kvm)) {
2025 rc = kvm_s390_deliver_pending_interrupts(vcpu);
2030 rc = kvm_s390_handle_requests(vcpu);
2034 if (guestdbg_enabled(vcpu)) {
2035 kvm_s390_backup_guest_per_regs(vcpu);
2036 kvm_s390_patch_guest_per_regs(vcpu);
2039 vcpu->arch.sie_block->icptcode = 0;
2040 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
2041 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
2042 trace_kvm_s390_sie_enter(vcpu, cpuflags);
2047 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
2049 psw_t *psw = &vcpu->arch.sie_block->gpsw;
2053 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
2054 trace_kvm_s390_sie_fault(vcpu);
2057 * We want to inject an addressing exception, which is defined as a
2058 * suppressing or terminating exception. However, since we came here
2059 * by a DAT access exception, the PSW still points to the faulting
2060 * instruction since DAT exceptions are nullifying. So we've got
2061 * to look up the current opcode to get the length of the instruction
2062 * to be able to forward the PSW.
2064 rc = read_guest(vcpu, psw->addr, 0, &opcode, 1);
2066 return kvm_s390_inject_prog_cond(vcpu, rc);
2067 psw->addr = __rewind_psw(*psw, -insn_length(opcode));
2069 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
2072 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
2076 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
2077 vcpu->arch.sie_block->icptcode);
2078 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
2080 if (guestdbg_enabled(vcpu))
2081 kvm_s390_restore_guest_per_regs(vcpu);
2083 if (exit_reason >= 0) {
2085 } else if (kvm_is_ucontrol(vcpu->kvm)) {
2086 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
2087 vcpu->run->s390_ucontrol.trans_exc_code =
2088 current->thread.gmap_addr;
2089 vcpu->run->s390_ucontrol.pgm_code = 0x10;
2092 } else if (current->thread.gmap_pfault) {
2093 trace_kvm_s390_major_guest_pfault(vcpu);
2094 current->thread.gmap_pfault = 0;
2095 if (kvm_arch_setup_async_pf(vcpu)) {
2098 gpa_t gpa = current->thread.gmap_addr;
2099 rc = kvm_arch_fault_in_page(vcpu, gpa, 1);
2104 rc = vcpu_post_run_fault_in_sie(vcpu);
2106 memcpy(&vcpu->run->s.regs.gprs[14], &vcpu->arch.sie_block->gg14, 16);
2109 if (kvm_is_ucontrol(vcpu->kvm))
2110 /* Don't exit for host interrupts. */
2111 rc = vcpu->arch.sie_block->icptcode ? -EOPNOTSUPP : 0;
2113 rc = kvm_handle_sie_intercept(vcpu);
2119 static int __vcpu_run(struct kvm_vcpu *vcpu)
2121 int rc, exit_reason;
2124 * We try to hold kvm->srcu during most of vcpu_run (except when run-
2125 * ning the guest), so that memslots (and other stuff) are protected
2127 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2130 rc = vcpu_pre_run(vcpu);
2134 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2136 * As PF_VCPU will be used in fault handler, between
2137 * guest_enter and guest_exit should be no uaccess.
2139 local_irq_disable();
2140 __kvm_guest_enter();
2142 exit_reason = sie64a(vcpu->arch.sie_block,
2143 vcpu->run->s.regs.gprs);
2144 local_irq_disable();
2147 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2149 rc = vcpu_post_run(vcpu, exit_reason);
2150 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
2152 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2156 static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2158 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
2159 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
2160 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
2161 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
2162 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
2163 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
2164 /* some control register changes require a tlb flush */
2165 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2167 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
2168 vcpu->arch.sie_block->cputm = kvm_run->s.regs.cputm;
2169 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
2170 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
2171 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
2172 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
2174 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
2175 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
2176 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
2177 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
2178 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2179 kvm_clear_async_pf_completion_queue(vcpu);
2181 kvm_run->kvm_dirty_regs = 0;
2184 static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2186 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
2187 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
2188 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
2189 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
2190 kvm_run->s.regs.cputm = vcpu->arch.sie_block->cputm;
2191 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
2192 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
2193 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
2194 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
2195 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
2196 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
2197 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
2200 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2205 if (guestdbg_exit_pending(vcpu)) {
2206 kvm_s390_prepare_debug_exit(vcpu);
2210 if (vcpu->sigset_active)
2211 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2213 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
2214 kvm_s390_vcpu_start(vcpu);
2215 } else if (is_vcpu_stopped(vcpu)) {
2216 pr_err_ratelimited("can't run stopped vcpu %d\n",
2221 sync_regs(vcpu, kvm_run);
2224 rc = __vcpu_run(vcpu);
2226 if (signal_pending(current) && !rc) {
2227 kvm_run->exit_reason = KVM_EXIT_INTR;
2231 if (guestdbg_exit_pending(vcpu) && !rc) {
2232 kvm_s390_prepare_debug_exit(vcpu);
2236 if (rc == -EOPNOTSUPP) {
2237 /* intercept cannot be handled in-kernel, prepare kvm-run */
2238 kvm_run->exit_reason = KVM_EXIT_S390_SIEIC;
2239 kvm_run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
2240 kvm_run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
2241 kvm_run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
2245 if (rc == -EREMOTE) {
2246 /* intercept was handled, but userspace support is needed
2247 * kvm_run has been prepared by the handler */
2251 store_regs(vcpu, kvm_run);
2253 if (vcpu->sigset_active)
2254 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2256 vcpu->stat.exit_userspace++;
2261 * store status at address
2262 * we use have two special cases:
2263 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
2264 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
2266 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
2268 unsigned char archmode = 1;
2273 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
2274 if (write_guest_abs(vcpu, 163, &archmode, 1))
2276 gpa = SAVE_AREA_BASE;
2277 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
2278 if (write_guest_real(vcpu, 163, &archmode, 1))
2280 gpa = kvm_s390_real_to_abs(vcpu, SAVE_AREA_BASE);
2282 rc = write_guest_abs(vcpu, gpa + offsetof(struct save_area, fp_regs),
2283 vcpu->arch.guest_fpregs.fprs, 128);
2284 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, gp_regs),
2285 vcpu->run->s.regs.gprs, 128);
2286 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, psw),
2287 &vcpu->arch.sie_block->gpsw, 16);
2288 px = kvm_s390_get_prefix(vcpu);
2289 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, pref_reg),
2291 rc |= write_guest_abs(vcpu,
2292 gpa + offsetof(struct save_area, fp_ctrl_reg),
2293 &vcpu->arch.guest_fpregs.fpc, 4);
2294 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, tod_reg),
2295 &vcpu->arch.sie_block->todpr, 4);
2296 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, timer),
2297 &vcpu->arch.sie_block->cputm, 8);
2298 clkcomp = vcpu->arch.sie_block->ckc >> 8;
2299 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, clk_cmp),
2301 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, acc_regs),
2302 &vcpu->run->s.regs.acrs, 64);
2303 rc |= write_guest_abs(vcpu, gpa + offsetof(struct save_area, ctrl_regs),
2304 &vcpu->arch.sie_block->gcr, 128);
2305 return rc ? -EFAULT : 0;
2308 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
2311 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
2312 * copying in vcpu load/put. Lets update our copies before we save
2313 * it into the save area
2316 if (test_kvm_facility(vcpu->kvm, 129)) {
2318 * If the vector extension is available, the vector registers
2319 * which overlaps with floating-point registers are saved in
2320 * the SIE-control block. Hence, extract the floating-point
2321 * registers and the FPC value and store them in the
2322 * guest_fpregs structure.
2324 vcpu->arch.guest_fpregs.fpc = current->thread.fpu.fpc;
2325 convert_vx_to_fp(vcpu->arch.guest_fpregs.fprs,
2326 current->thread.fpu.vxrs);
2328 save_fpu_to(&vcpu->arch.guest_fpregs);
2329 save_access_regs(vcpu->run->s.regs.acrs);
2331 return kvm_s390_store_status_unloaded(vcpu, addr);
2335 * store additional status at address
2337 int kvm_s390_store_adtl_status_unloaded(struct kvm_vcpu *vcpu,
2340 /* Only bits 0-53 are used for address formation */
2341 if (!(gpa & ~0x3ff))
2344 return write_guest_abs(vcpu, gpa & ~0x3ff,
2345 (void *)&vcpu->run->s.regs.vrs, 512);
2348 int kvm_s390_vcpu_store_adtl_status(struct kvm_vcpu *vcpu, unsigned long addr)
2350 if (!test_kvm_facility(vcpu->kvm, 129))
2354 * The guest VXRS are in the host VXRs due to the lazy
2355 * copying in vcpu load/put. We can simply call save_fpu_regs()
2356 * to save the current register state because we are in the
2357 * middle of a load/put cycle.
2359 * Let's update our copies before we save it into the save area.
2363 return kvm_s390_store_adtl_status_unloaded(vcpu, addr);
2366 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2368 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
2369 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
2372 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
2375 struct kvm_vcpu *vcpu;
2377 kvm_for_each_vcpu(i, vcpu, kvm) {
2378 __disable_ibs_on_vcpu(vcpu);
2382 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2384 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
2385 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
2388 void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
2390 int i, online_vcpus, started_vcpus = 0;
2392 if (!is_vcpu_stopped(vcpu))
2395 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
2396 /* Only one cpu at a time may enter/leave the STOPPED state. */
2397 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2398 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2400 for (i = 0; i < online_vcpus; i++) {
2401 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i]))
2405 if (started_vcpus == 0) {
2406 /* we're the only active VCPU -> speed it up */
2407 __enable_ibs_on_vcpu(vcpu);
2408 } else if (started_vcpus == 1) {
2410 * As we are starting a second VCPU, we have to disable
2411 * the IBS facility on all VCPUs to remove potentially
2412 * oustanding ENABLE requests.
2414 __disable_ibs_on_all_vcpus(vcpu->kvm);
2417 atomic_andnot(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2419 * Another VCPU might have used IBS while we were offline.
2420 * Let's play safe and flush the VCPU at startup.
2422 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2423 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2427 void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
2429 int i, online_vcpus, started_vcpus = 0;
2430 struct kvm_vcpu *started_vcpu = NULL;
2432 if (is_vcpu_stopped(vcpu))
2435 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
2436 /* Only one cpu at a time may enter/leave the STOPPED state. */
2437 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2438 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2440 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
2441 kvm_s390_clear_stop_irq(vcpu);
2443 atomic_or(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2444 __disable_ibs_on_vcpu(vcpu);
2446 for (i = 0; i < online_vcpus; i++) {
2447 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) {
2449 started_vcpu = vcpu->kvm->vcpus[i];
2453 if (started_vcpus == 1) {
2455 * As we only have one VCPU left, we want to enable the
2456 * IBS facility for that VCPU to speed it up.
2458 __enable_ibs_on_vcpu(started_vcpu);
2461 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2465 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
2466 struct kvm_enable_cap *cap)
2474 case KVM_CAP_S390_CSS_SUPPORT:
2475 if (!vcpu->kvm->arch.css_support) {
2476 vcpu->kvm->arch.css_support = 1;
2477 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
2478 trace_kvm_s390_enable_css(vcpu->kvm);
2489 static long kvm_s390_guest_mem_op(struct kvm_vcpu *vcpu,
2490 struct kvm_s390_mem_op *mop)
2492 void __user *uaddr = (void __user *)mop->buf;
2493 void *tmpbuf = NULL;
2495 const u64 supported_flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION
2496 | KVM_S390_MEMOP_F_CHECK_ONLY;
2498 if (mop->flags & ~supported_flags)
2501 if (mop->size > MEM_OP_MAX_SIZE)
2504 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
2505 tmpbuf = vmalloc(mop->size);
2510 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2513 case KVM_S390_MEMOP_LOGICAL_READ:
2514 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2515 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size, false);
2518 r = read_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
2520 if (copy_to_user(uaddr, tmpbuf, mop->size))
2524 case KVM_S390_MEMOP_LOGICAL_WRITE:
2525 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2526 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size, true);
2529 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
2533 r = write_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
2539 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2541 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
2542 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
2548 long kvm_arch_vcpu_ioctl(struct file *filp,
2549 unsigned int ioctl, unsigned long arg)
2551 struct kvm_vcpu *vcpu = filp->private_data;
2552 void __user *argp = (void __user *)arg;
2557 case KVM_S390_IRQ: {
2558 struct kvm_s390_irq s390irq;
2561 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
2563 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
2566 case KVM_S390_INTERRUPT: {
2567 struct kvm_s390_interrupt s390int;
2568 struct kvm_s390_irq s390irq;
2571 if (copy_from_user(&s390int, argp, sizeof(s390int)))
2573 if (s390int_to_s390irq(&s390int, &s390irq))
2575 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
2578 case KVM_S390_STORE_STATUS:
2579 idx = srcu_read_lock(&vcpu->kvm->srcu);
2580 r = kvm_s390_vcpu_store_status(vcpu, arg);
2581 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2583 case KVM_S390_SET_INITIAL_PSW: {
2587 if (copy_from_user(&psw, argp, sizeof(psw)))
2589 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
2592 case KVM_S390_INITIAL_RESET:
2593 r = kvm_arch_vcpu_ioctl_initial_reset(vcpu);
2595 case KVM_SET_ONE_REG:
2596 case KVM_GET_ONE_REG: {
2597 struct kvm_one_reg reg;
2599 if (copy_from_user(®, argp, sizeof(reg)))
2601 if (ioctl == KVM_SET_ONE_REG)
2602 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®);
2604 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®);
2607 #ifdef CONFIG_KVM_S390_UCONTROL
2608 case KVM_S390_UCAS_MAP: {
2609 struct kvm_s390_ucas_mapping ucasmap;
2611 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
2616 if (!kvm_is_ucontrol(vcpu->kvm)) {
2621 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
2622 ucasmap.vcpu_addr, ucasmap.length);
2625 case KVM_S390_UCAS_UNMAP: {
2626 struct kvm_s390_ucas_mapping ucasmap;
2628 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
2633 if (!kvm_is_ucontrol(vcpu->kvm)) {
2638 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
2643 case KVM_S390_VCPU_FAULT: {
2644 r = gmap_fault(vcpu->arch.gmap, arg, 0);
2647 case KVM_ENABLE_CAP:
2649 struct kvm_enable_cap cap;
2651 if (copy_from_user(&cap, argp, sizeof(cap)))
2653 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
2656 case KVM_S390_MEM_OP: {
2657 struct kvm_s390_mem_op mem_op;
2659 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
2660 r = kvm_s390_guest_mem_op(vcpu, &mem_op);
2665 case KVM_S390_SET_IRQ_STATE: {
2666 struct kvm_s390_irq_state irq_state;
2669 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
2671 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
2672 irq_state.len == 0 ||
2673 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
2677 r = kvm_s390_set_irq_state(vcpu,
2678 (void __user *) irq_state.buf,
2682 case KVM_S390_GET_IRQ_STATE: {
2683 struct kvm_s390_irq_state irq_state;
2686 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
2688 if (irq_state.len == 0) {
2692 r = kvm_s390_get_irq_state(vcpu,
2693 (__u8 __user *) irq_state.buf,
2703 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2705 #ifdef CONFIG_KVM_S390_UCONTROL
2706 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
2707 && (kvm_is_ucontrol(vcpu->kvm))) {
2708 vmf->page = virt_to_page(vcpu->arch.sie_block);
2709 get_page(vmf->page);
2713 return VM_FAULT_SIGBUS;
2716 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
2717 unsigned long npages)
2722 /* Section: memory related */
2723 int kvm_arch_prepare_memory_region(struct kvm *kvm,
2724 struct kvm_memory_slot *memslot,
2725 const struct kvm_userspace_memory_region *mem,
2726 enum kvm_mr_change change)
2728 /* A few sanity checks. We can have memory slots which have to be
2729 located/ended at a segment boundary (1MB). The memory in userland is
2730 ok to be fragmented into various different vmas. It is okay to mmap()
2731 and munmap() stuff in this slot after doing this call at any time */
2733 if (mem->userspace_addr & 0xffffful)
2736 if (mem->memory_size & 0xffffful)
2742 void kvm_arch_commit_memory_region(struct kvm *kvm,
2743 const struct kvm_userspace_memory_region *mem,
2744 const struct kvm_memory_slot *old,
2745 const struct kvm_memory_slot *new,
2746 enum kvm_mr_change change)
2750 /* If the basics of the memslot do not change, we do not want
2751 * to update the gmap. Every update causes several unnecessary
2752 * segment translation exceptions. This is usually handled just
2753 * fine by the normal fault handler + gmap, but it will also
2754 * cause faults on the prefix page of running guest CPUs.
2756 if (old->userspace_addr == mem->userspace_addr &&
2757 old->base_gfn * PAGE_SIZE == mem->guest_phys_addr &&
2758 old->npages * PAGE_SIZE == mem->memory_size)
2761 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr,
2762 mem->guest_phys_addr, mem->memory_size);
2764 pr_warn("failed to commit memory region\n");
2768 static int __init kvm_s390_init(void)
2770 return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2773 static void __exit kvm_s390_exit(void)
2778 module_init(kvm_s390_init);
2779 module_exit(kvm_s390_exit);
2782 * Enable autoloading of the kvm module.
2783 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
2784 * since x86 takes a different approach.
2786 #include <linux/miscdevice.h>
2787 MODULE_ALIAS_MISCDEV(KVM_MINOR);
2788 MODULE_ALIAS("devname:kvm");