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 <linux/bitmap.h>
30 #include <asm/asm-offsets.h>
31 #include <asm/lowcore.h>
33 #include <asm/pgtable.h>
36 #include <asm/switch_to.h>
39 #include <asm/cpacf.h>
44 #define KMSG_COMPONENT "kvm-s390"
46 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
48 #define CREATE_TRACE_POINTS
50 #include "trace-s390.h"
52 #define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */
54 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
55 (KVM_MAX_VCPUS + LOCAL_IRQS))
57 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
59 struct kvm_stats_debugfs_item debugfs_entries[] = {
60 { "userspace_handled", VCPU_STAT(exit_userspace) },
61 { "exit_null", VCPU_STAT(exit_null) },
62 { "exit_validity", VCPU_STAT(exit_validity) },
63 { "exit_stop_request", VCPU_STAT(exit_stop_request) },
64 { "exit_external_request", VCPU_STAT(exit_external_request) },
65 { "exit_external_interrupt", VCPU_STAT(exit_external_interrupt) },
66 { "exit_instruction", VCPU_STAT(exit_instruction) },
67 { "exit_program_interruption", VCPU_STAT(exit_program_interruption) },
68 { "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program) },
69 { "exit_operation_exception", VCPU_STAT(exit_operation_exception) },
70 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
71 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
72 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
73 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
74 { "instruction_lctlg", VCPU_STAT(instruction_lctlg) },
75 { "instruction_lctl", VCPU_STAT(instruction_lctl) },
76 { "instruction_stctl", VCPU_STAT(instruction_stctl) },
77 { "instruction_stctg", VCPU_STAT(instruction_stctg) },
78 { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal) },
79 { "deliver_external_call", VCPU_STAT(deliver_external_call) },
80 { "deliver_service_signal", VCPU_STAT(deliver_service_signal) },
81 { "deliver_virtio_interrupt", VCPU_STAT(deliver_virtio_interrupt) },
82 { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal) },
83 { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal) },
84 { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal) },
85 { "deliver_program_interruption", VCPU_STAT(deliver_program_int) },
86 { "exit_wait_state", VCPU_STAT(exit_wait_state) },
87 { "instruction_pfmf", VCPU_STAT(instruction_pfmf) },
88 { "instruction_stidp", VCPU_STAT(instruction_stidp) },
89 { "instruction_spx", VCPU_STAT(instruction_spx) },
90 { "instruction_stpx", VCPU_STAT(instruction_stpx) },
91 { "instruction_stap", VCPU_STAT(instruction_stap) },
92 { "instruction_storage_key", VCPU_STAT(instruction_storage_key) },
93 { "instruction_ipte_interlock", VCPU_STAT(instruction_ipte_interlock) },
94 { "instruction_stsch", VCPU_STAT(instruction_stsch) },
95 { "instruction_chsc", VCPU_STAT(instruction_chsc) },
96 { "instruction_essa", VCPU_STAT(instruction_essa) },
97 { "instruction_stsi", VCPU_STAT(instruction_stsi) },
98 { "instruction_stfl", VCPU_STAT(instruction_stfl) },
99 { "instruction_tprot", VCPU_STAT(instruction_tprot) },
100 { "instruction_sthyi", VCPU_STAT(instruction_sthyi) },
101 { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense) },
102 { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running) },
103 { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call) },
104 { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency) },
105 { "instruction_sigp_cond_emergency", VCPU_STAT(instruction_sigp_cond_emergency) },
106 { "instruction_sigp_start", VCPU_STAT(instruction_sigp_start) },
107 { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop) },
108 { "instruction_sigp_stop_store_status", VCPU_STAT(instruction_sigp_stop_store_status) },
109 { "instruction_sigp_store_status", VCPU_STAT(instruction_sigp_store_status) },
110 { "instruction_sigp_store_adtl_status", VCPU_STAT(instruction_sigp_store_adtl_status) },
111 { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch) },
112 { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix) },
113 { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart) },
114 { "instruction_sigp_cpu_reset", VCPU_STAT(instruction_sigp_cpu_reset) },
115 { "instruction_sigp_init_cpu_reset", VCPU_STAT(instruction_sigp_init_cpu_reset) },
116 { "instruction_sigp_unknown", VCPU_STAT(instruction_sigp_unknown) },
117 { "diagnose_10", VCPU_STAT(diagnose_10) },
118 { "diagnose_44", VCPU_STAT(diagnose_44) },
119 { "diagnose_9c", VCPU_STAT(diagnose_9c) },
120 { "diagnose_258", VCPU_STAT(diagnose_258) },
121 { "diagnose_308", VCPU_STAT(diagnose_308) },
122 { "diagnose_500", VCPU_STAT(diagnose_500) },
126 /* upper facilities limit for kvm */
127 unsigned long kvm_s390_fac_list_mask[16] = {
128 0xffe6000000000000UL,
129 0x005e000000000000UL,
132 unsigned long kvm_s390_fac_list_mask_size(void)
134 BUILD_BUG_ON(ARRAY_SIZE(kvm_s390_fac_list_mask) > S390_ARCH_FAC_MASK_SIZE_U64);
135 return ARRAY_SIZE(kvm_s390_fac_list_mask);
138 /* available cpu features supported by kvm */
139 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
140 /* available subfunctions indicated via query / "test bit" */
141 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
143 static struct gmap_notifier gmap_notifier;
144 debug_info_t *kvm_s390_dbf;
146 /* Section: not file related */
147 int kvm_arch_hardware_enable(void)
149 /* every s390 is virtualization enabled ;-) */
153 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long address);
156 * This callback is executed during stop_machine(). All CPUs are therefore
157 * temporarily stopped. In order not to change guest behavior, we have to
158 * disable preemption whenever we touch the epoch of kvm and the VCPUs,
159 * so a CPU won't be stopped while calculating with the epoch.
161 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
165 struct kvm_vcpu *vcpu;
167 unsigned long long *delta = v;
169 list_for_each_entry(kvm, &vm_list, vm_list) {
170 kvm->arch.epoch -= *delta;
171 kvm_for_each_vcpu(i, vcpu, kvm) {
172 vcpu->arch.sie_block->epoch -= *delta;
173 if (vcpu->arch.cputm_enabled)
174 vcpu->arch.cputm_start += *delta;
180 static struct notifier_block kvm_clock_notifier = {
181 .notifier_call = kvm_clock_sync,
184 int kvm_arch_hardware_setup(void)
186 gmap_notifier.notifier_call = kvm_gmap_notifier;
187 gmap_register_ipte_notifier(&gmap_notifier);
188 atomic_notifier_chain_register(&s390_epoch_delta_notifier,
189 &kvm_clock_notifier);
193 void kvm_arch_hardware_unsetup(void)
195 gmap_unregister_ipte_notifier(&gmap_notifier);
196 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
197 &kvm_clock_notifier);
200 static void allow_cpu_feat(unsigned long nr)
202 set_bit_inv(nr, kvm_s390_available_cpu_feat);
205 static inline int plo_test_bit(unsigned char nr)
207 register unsigned long r0 asm("0") = (unsigned long) nr | 0x100;
208 int cc = 3; /* subfunction not available */
211 /* Parameter registers are ignored for "test bit" */
221 static void kvm_s390_cpu_feat_init(void)
225 for (i = 0; i < 256; ++i) {
227 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
230 if (test_facility(28)) /* TOD-clock steering */
231 etr_ptff(kvm_s390_available_subfunc.ptff, ETR_PTFF_QAF);
233 if (test_facility(17)) { /* MSA */
234 __cpacf_query(CPACF_KMAC, kvm_s390_available_subfunc.kmac);
235 __cpacf_query(CPACF_KMC, kvm_s390_available_subfunc.kmc);
236 __cpacf_query(CPACF_KM, kvm_s390_available_subfunc.km);
237 __cpacf_query(CPACF_KIMD, kvm_s390_available_subfunc.kimd);
238 __cpacf_query(CPACF_KLMD, kvm_s390_available_subfunc.klmd);
240 if (test_facility(76)) /* MSA3 */
241 __cpacf_query(CPACF_PCKMO, kvm_s390_available_subfunc.pckmo);
242 if (test_facility(77)) { /* MSA4 */
243 __cpacf_query(CPACF_KMCTR, kvm_s390_available_subfunc.kmctr);
244 __cpacf_query(CPACF_KMF, kvm_s390_available_subfunc.kmf);
245 __cpacf_query(CPACF_KMO, kvm_s390_available_subfunc.kmo);
246 __cpacf_query(CPACF_PCC, kvm_s390_available_subfunc.pcc);
248 if (test_facility(57)) /* MSA5 */
249 __cpacf_query(CPACF_PPNO, kvm_s390_available_subfunc.ppno);
251 if (MACHINE_HAS_ESOP)
252 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
255 int kvm_arch_init(void *opaque)
257 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
261 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view)) {
262 debug_unregister(kvm_s390_dbf);
266 kvm_s390_cpu_feat_init();
268 /* Register floating interrupt controller interface. */
269 return kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
272 void kvm_arch_exit(void)
274 debug_unregister(kvm_s390_dbf);
277 /* Section: device related */
278 long kvm_arch_dev_ioctl(struct file *filp,
279 unsigned int ioctl, unsigned long arg)
281 if (ioctl == KVM_S390_ENABLE_SIE)
282 return s390_enable_sie();
286 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
291 case KVM_CAP_S390_PSW:
292 case KVM_CAP_S390_GMAP:
293 case KVM_CAP_SYNC_MMU:
294 #ifdef CONFIG_KVM_S390_UCONTROL
295 case KVM_CAP_S390_UCONTROL:
297 case KVM_CAP_ASYNC_PF:
298 case KVM_CAP_SYNC_REGS:
299 case KVM_CAP_ONE_REG:
300 case KVM_CAP_ENABLE_CAP:
301 case KVM_CAP_S390_CSS_SUPPORT:
302 case KVM_CAP_IOEVENTFD:
303 case KVM_CAP_DEVICE_CTRL:
304 case KVM_CAP_ENABLE_CAP_VM:
305 case KVM_CAP_S390_IRQCHIP:
306 case KVM_CAP_VM_ATTRIBUTES:
307 case KVM_CAP_MP_STATE:
308 case KVM_CAP_S390_INJECT_IRQ:
309 case KVM_CAP_S390_USER_SIGP:
310 case KVM_CAP_S390_USER_STSI:
311 case KVM_CAP_S390_SKEYS:
312 case KVM_CAP_S390_IRQ_STATE:
315 case KVM_CAP_S390_MEM_OP:
318 case KVM_CAP_NR_VCPUS:
319 case KVM_CAP_MAX_VCPUS:
320 r = KVM_S390_BSCA_CPU_SLOTS;
321 if (sclp.has_esca && sclp.has_64bscao)
322 r = KVM_S390_ESCA_CPU_SLOTS;
324 case KVM_CAP_NR_MEMSLOTS:
325 r = KVM_USER_MEM_SLOTS;
327 case KVM_CAP_S390_COW:
328 r = MACHINE_HAS_ESOP;
330 case KVM_CAP_S390_VECTOR_REGISTERS:
333 case KVM_CAP_S390_RI:
334 r = test_facility(64);
342 static void kvm_s390_sync_dirty_log(struct kvm *kvm,
343 struct kvm_memory_slot *memslot)
345 gfn_t cur_gfn, last_gfn;
346 unsigned long address;
347 struct gmap *gmap = kvm->arch.gmap;
349 /* Loop over all guest pages */
350 last_gfn = memslot->base_gfn + memslot->npages;
351 for (cur_gfn = memslot->base_gfn; cur_gfn <= last_gfn; cur_gfn++) {
352 address = gfn_to_hva_memslot(memslot, cur_gfn);
354 if (test_and_clear_guest_dirty(gmap->mm, address))
355 mark_page_dirty(kvm, cur_gfn);
356 if (fatal_signal_pending(current))
362 /* Section: vm related */
363 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
366 * Get (and clear) the dirty memory log for a memory slot.
368 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
369 struct kvm_dirty_log *log)
373 struct kvm_memslots *slots;
374 struct kvm_memory_slot *memslot;
377 mutex_lock(&kvm->slots_lock);
380 if (log->slot >= KVM_USER_MEM_SLOTS)
383 slots = kvm_memslots(kvm);
384 memslot = id_to_memslot(slots, log->slot);
386 if (!memslot->dirty_bitmap)
389 kvm_s390_sync_dirty_log(kvm, memslot);
390 r = kvm_get_dirty_log(kvm, log, &is_dirty);
394 /* Clear the dirty log */
396 n = kvm_dirty_bitmap_bytes(memslot);
397 memset(memslot->dirty_bitmap, 0, n);
401 mutex_unlock(&kvm->slots_lock);
405 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
413 case KVM_CAP_S390_IRQCHIP:
414 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
415 kvm->arch.use_irqchip = 1;
418 case KVM_CAP_S390_USER_SIGP:
419 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
420 kvm->arch.user_sigp = 1;
423 case KVM_CAP_S390_VECTOR_REGISTERS:
424 mutex_lock(&kvm->lock);
425 if (atomic_read(&kvm->online_vcpus)) {
427 } else if (MACHINE_HAS_VX) {
428 set_kvm_facility(kvm->arch.model.fac_mask, 129);
429 set_kvm_facility(kvm->arch.model.fac_list, 129);
433 mutex_unlock(&kvm->lock);
434 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
435 r ? "(not available)" : "(success)");
437 case KVM_CAP_S390_RI:
439 mutex_lock(&kvm->lock);
440 if (atomic_read(&kvm->online_vcpus)) {
442 } else if (test_facility(64)) {
443 set_kvm_facility(kvm->arch.model.fac_mask, 64);
444 set_kvm_facility(kvm->arch.model.fac_list, 64);
447 mutex_unlock(&kvm->lock);
448 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
449 r ? "(not available)" : "(success)");
451 case KVM_CAP_S390_USER_STSI:
452 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
453 kvm->arch.user_stsi = 1;
463 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
467 switch (attr->attr) {
468 case KVM_S390_VM_MEM_LIMIT_SIZE:
470 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
471 kvm->arch.mem_limit);
472 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
482 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
486 switch (attr->attr) {
487 case KVM_S390_VM_MEM_ENABLE_CMMA:
493 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
494 mutex_lock(&kvm->lock);
495 if (atomic_read(&kvm->online_vcpus) == 0) {
496 kvm->arch.use_cmma = 1;
499 mutex_unlock(&kvm->lock);
501 case KVM_S390_VM_MEM_CLR_CMMA:
506 if (!kvm->arch.use_cmma)
509 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
510 mutex_lock(&kvm->lock);
511 idx = srcu_read_lock(&kvm->srcu);
512 s390_reset_cmma(kvm->arch.gmap->mm);
513 srcu_read_unlock(&kvm->srcu, idx);
514 mutex_unlock(&kvm->lock);
517 case KVM_S390_VM_MEM_LIMIT_SIZE: {
518 unsigned long new_limit;
520 if (kvm_is_ucontrol(kvm))
523 if (get_user(new_limit, (u64 __user *)attr->addr))
526 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
527 new_limit > kvm->arch.mem_limit)
533 /* gmap_alloc takes last usable address */
534 if (new_limit != KVM_S390_NO_MEM_LIMIT)
538 mutex_lock(&kvm->lock);
539 if (atomic_read(&kvm->online_vcpus) == 0) {
540 /* gmap_alloc will round the limit up */
541 struct gmap *new = gmap_alloc(current->mm, new_limit);
546 gmap_free(kvm->arch.gmap);
548 kvm->arch.gmap = new;
552 mutex_unlock(&kvm->lock);
553 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
554 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
555 (void *) kvm->arch.gmap->asce);
565 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
567 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
569 struct kvm_vcpu *vcpu;
572 if (!test_kvm_facility(kvm, 76))
575 mutex_lock(&kvm->lock);
576 switch (attr->attr) {
577 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
579 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
580 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
581 kvm->arch.crypto.aes_kw = 1;
582 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
584 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
586 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
587 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
588 kvm->arch.crypto.dea_kw = 1;
589 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
591 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
592 kvm->arch.crypto.aes_kw = 0;
593 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
594 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
595 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
597 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
598 kvm->arch.crypto.dea_kw = 0;
599 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
600 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
601 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
604 mutex_unlock(&kvm->lock);
608 kvm_for_each_vcpu(i, vcpu, kvm) {
609 kvm_s390_vcpu_crypto_setup(vcpu);
612 mutex_unlock(&kvm->lock);
616 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
620 if (copy_from_user(>od_high, (void __user *)attr->addr,
626 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
631 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
635 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod)))
638 kvm_s390_set_tod_clock(kvm, gtod);
639 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod);
643 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
650 switch (attr->attr) {
651 case KVM_S390_VM_TOD_HIGH:
652 ret = kvm_s390_set_tod_high(kvm, attr);
654 case KVM_S390_VM_TOD_LOW:
655 ret = kvm_s390_set_tod_low(kvm, attr);
664 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
668 if (copy_to_user((void __user *)attr->addr, >od_high,
671 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
676 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
680 gtod = kvm_s390_get_tod_clock_fast(kvm);
681 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
683 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
688 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
695 switch (attr->attr) {
696 case KVM_S390_VM_TOD_HIGH:
697 ret = kvm_s390_get_tod_high(kvm, attr);
699 case KVM_S390_VM_TOD_LOW:
700 ret = kvm_s390_get_tod_low(kvm, attr);
709 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
711 struct kvm_s390_vm_cpu_processor *proc;
712 u16 lowest_ibc, unblocked_ibc;
715 mutex_lock(&kvm->lock);
716 if (atomic_read(&kvm->online_vcpus)) {
720 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
725 if (!copy_from_user(proc, (void __user *)attr->addr,
727 kvm->arch.model.cpuid = proc->cpuid;
728 lowest_ibc = sclp.ibc >> 16 & 0xfff;
729 unblocked_ibc = sclp.ibc & 0xfff;
731 if (proc->ibc > unblocked_ibc)
732 kvm->arch.model.ibc = unblocked_ibc;
733 else if (proc->ibc < lowest_ibc)
734 kvm->arch.model.ibc = lowest_ibc;
736 kvm->arch.model.ibc = proc->ibc;
738 memcpy(kvm->arch.model.fac_list, proc->fac_list,
739 S390_ARCH_FAC_LIST_SIZE_BYTE);
744 mutex_unlock(&kvm->lock);
748 static int kvm_s390_set_processor_feat(struct kvm *kvm,
749 struct kvm_device_attr *attr)
751 struct kvm_s390_vm_cpu_feat data;
754 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
756 if (!bitmap_subset((unsigned long *) data.feat,
757 kvm_s390_available_cpu_feat,
758 KVM_S390_VM_CPU_FEAT_NR_BITS))
761 mutex_lock(&kvm->lock);
762 if (!atomic_read(&kvm->online_vcpus)) {
763 bitmap_copy(kvm->arch.cpu_feat, (unsigned long *) data.feat,
764 KVM_S390_VM_CPU_FEAT_NR_BITS);
767 mutex_unlock(&kvm->lock);
771 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
772 struct kvm_device_attr *attr)
775 * Once supported by kernel + hw, we have to store the subfunctions
776 * in kvm->arch and remember that user space configured them.
781 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
785 switch (attr->attr) {
786 case KVM_S390_VM_CPU_PROCESSOR:
787 ret = kvm_s390_set_processor(kvm, attr);
789 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
790 ret = kvm_s390_set_processor_feat(kvm, attr);
792 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
793 ret = kvm_s390_set_processor_subfunc(kvm, attr);
799 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
801 struct kvm_s390_vm_cpu_processor *proc;
804 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
809 proc->cpuid = kvm->arch.model.cpuid;
810 proc->ibc = kvm->arch.model.ibc;
811 memcpy(&proc->fac_list, kvm->arch.model.fac_list,
812 S390_ARCH_FAC_LIST_SIZE_BYTE);
813 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
820 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
822 struct kvm_s390_vm_cpu_machine *mach;
825 mach = kzalloc(sizeof(*mach), GFP_KERNEL);
830 get_cpu_id((struct cpuid *) &mach->cpuid);
831 mach->ibc = sclp.ibc;
832 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
833 S390_ARCH_FAC_LIST_SIZE_BYTE);
834 memcpy((unsigned long *)&mach->fac_list, S390_lowcore.stfle_fac_list,
835 S390_ARCH_FAC_LIST_SIZE_BYTE);
836 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
843 static int kvm_s390_get_processor_feat(struct kvm *kvm,
844 struct kvm_device_attr *attr)
846 struct kvm_s390_vm_cpu_feat data;
848 bitmap_copy((unsigned long *) data.feat, kvm->arch.cpu_feat,
849 KVM_S390_VM_CPU_FEAT_NR_BITS);
850 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
855 static int kvm_s390_get_machine_feat(struct kvm *kvm,
856 struct kvm_device_attr *attr)
858 struct kvm_s390_vm_cpu_feat data;
860 bitmap_copy((unsigned long *) data.feat,
861 kvm_s390_available_cpu_feat,
862 KVM_S390_VM_CPU_FEAT_NR_BITS);
863 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
868 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
869 struct kvm_device_attr *attr)
872 * Once we can actually configure subfunctions (kernel + hw support),
873 * we have to check if they were already set by user space, if so copy
874 * them from kvm->arch.
879 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
880 struct kvm_device_attr *attr)
882 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
883 sizeof(struct kvm_s390_vm_cpu_subfunc)))
887 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
891 switch (attr->attr) {
892 case KVM_S390_VM_CPU_PROCESSOR:
893 ret = kvm_s390_get_processor(kvm, attr);
895 case KVM_S390_VM_CPU_MACHINE:
896 ret = kvm_s390_get_machine(kvm, attr);
898 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
899 ret = kvm_s390_get_processor_feat(kvm, attr);
901 case KVM_S390_VM_CPU_MACHINE_FEAT:
902 ret = kvm_s390_get_machine_feat(kvm, attr);
904 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
905 ret = kvm_s390_get_processor_subfunc(kvm, attr);
907 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
908 ret = kvm_s390_get_machine_subfunc(kvm, attr);
914 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
918 switch (attr->group) {
919 case KVM_S390_VM_MEM_CTRL:
920 ret = kvm_s390_set_mem_control(kvm, attr);
922 case KVM_S390_VM_TOD:
923 ret = kvm_s390_set_tod(kvm, attr);
925 case KVM_S390_VM_CPU_MODEL:
926 ret = kvm_s390_set_cpu_model(kvm, attr);
928 case KVM_S390_VM_CRYPTO:
929 ret = kvm_s390_vm_set_crypto(kvm, attr);
939 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
943 switch (attr->group) {
944 case KVM_S390_VM_MEM_CTRL:
945 ret = kvm_s390_get_mem_control(kvm, attr);
947 case KVM_S390_VM_TOD:
948 ret = kvm_s390_get_tod(kvm, attr);
950 case KVM_S390_VM_CPU_MODEL:
951 ret = kvm_s390_get_cpu_model(kvm, attr);
961 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
965 switch (attr->group) {
966 case KVM_S390_VM_MEM_CTRL:
967 switch (attr->attr) {
968 case KVM_S390_VM_MEM_ENABLE_CMMA:
969 case KVM_S390_VM_MEM_CLR_CMMA:
970 ret = sclp.has_cmma ? 0 : -ENXIO;
972 case KVM_S390_VM_MEM_LIMIT_SIZE:
980 case KVM_S390_VM_TOD:
981 switch (attr->attr) {
982 case KVM_S390_VM_TOD_LOW:
983 case KVM_S390_VM_TOD_HIGH:
991 case KVM_S390_VM_CPU_MODEL:
992 switch (attr->attr) {
993 case KVM_S390_VM_CPU_PROCESSOR:
994 case KVM_S390_VM_CPU_MACHINE:
995 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
996 case KVM_S390_VM_CPU_MACHINE_FEAT:
997 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1000 /* configuring subfunctions is not supported yet */
1001 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1007 case KVM_S390_VM_CRYPTO:
1008 switch (attr->attr) {
1009 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1010 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1011 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1012 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1028 static long kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1034 if (args->flags != 0)
1037 /* Is this guest using storage keys? */
1038 if (!mm_use_skey(current->mm))
1039 return KVM_S390_GET_SKEYS_NONE;
1041 /* Enforce sane limit on memory allocation */
1042 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1045 keys = kmalloc_array(args->count, sizeof(uint8_t),
1046 GFP_KERNEL | __GFP_NOWARN);
1048 keys = vmalloc(sizeof(uint8_t) * args->count);
1052 down_read(¤t->mm->mmap_sem);
1053 for (i = 0; i < args->count; i++) {
1054 hva = gfn_to_hva(kvm, args->start_gfn + i);
1055 if (kvm_is_error_hva(hva)) {
1060 r = get_guest_storage_key(current->mm, hva, &keys[i]);
1064 up_read(¤t->mm->mmap_sem);
1067 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
1068 sizeof(uint8_t) * args->count);
1077 static long kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1083 if (args->flags != 0)
1086 /* Enforce sane limit on memory allocation */
1087 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1090 keys = kmalloc_array(args->count, sizeof(uint8_t),
1091 GFP_KERNEL | __GFP_NOWARN);
1093 keys = vmalloc(sizeof(uint8_t) * args->count);
1097 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
1098 sizeof(uint8_t) * args->count);
1104 /* Enable storage key handling for the guest */
1105 r = s390_enable_skey();
1109 down_read(¤t->mm->mmap_sem);
1110 for (i = 0; i < args->count; i++) {
1111 hva = gfn_to_hva(kvm, args->start_gfn + i);
1112 if (kvm_is_error_hva(hva)) {
1117 /* Lowest order bit is reserved */
1118 if (keys[i] & 0x01) {
1123 r = set_guest_storage_key(current->mm, hva,
1124 (unsigned long)keys[i], 0);
1128 up_read(¤t->mm->mmap_sem);
1134 long kvm_arch_vm_ioctl(struct file *filp,
1135 unsigned int ioctl, unsigned long arg)
1137 struct kvm *kvm = filp->private_data;
1138 void __user *argp = (void __user *)arg;
1139 struct kvm_device_attr attr;
1143 case KVM_S390_INTERRUPT: {
1144 struct kvm_s390_interrupt s390int;
1147 if (copy_from_user(&s390int, argp, sizeof(s390int)))
1149 r = kvm_s390_inject_vm(kvm, &s390int);
1152 case KVM_ENABLE_CAP: {
1153 struct kvm_enable_cap cap;
1155 if (copy_from_user(&cap, argp, sizeof(cap)))
1157 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1160 case KVM_CREATE_IRQCHIP: {
1161 struct kvm_irq_routing_entry routing;
1164 if (kvm->arch.use_irqchip) {
1165 /* Set up dummy routing. */
1166 memset(&routing, 0, sizeof(routing));
1167 r = kvm_set_irq_routing(kvm, &routing, 0, 0);
1171 case KVM_SET_DEVICE_ATTR: {
1173 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1175 r = kvm_s390_vm_set_attr(kvm, &attr);
1178 case KVM_GET_DEVICE_ATTR: {
1180 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1182 r = kvm_s390_vm_get_attr(kvm, &attr);
1185 case KVM_HAS_DEVICE_ATTR: {
1187 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1189 r = kvm_s390_vm_has_attr(kvm, &attr);
1192 case KVM_S390_GET_SKEYS: {
1193 struct kvm_s390_skeys args;
1196 if (copy_from_user(&args, argp,
1197 sizeof(struct kvm_s390_skeys)))
1199 r = kvm_s390_get_skeys(kvm, &args);
1202 case KVM_S390_SET_SKEYS: {
1203 struct kvm_s390_skeys args;
1206 if (copy_from_user(&args, argp,
1207 sizeof(struct kvm_s390_skeys)))
1209 r = kvm_s390_set_skeys(kvm, &args);
1219 static int kvm_s390_query_ap_config(u8 *config)
1221 u32 fcn_code = 0x04000000UL;
1224 memset(config, 0, 128);
1228 ".long 0xb2af0000\n" /* PQAP(QCI) */
1234 : "r" (fcn_code), "r" (config)
1235 : "cc", "0", "2", "memory"
1241 static int kvm_s390_apxa_installed(void)
1246 if (test_facility(12)) {
1247 cc = kvm_s390_query_ap_config(config);
1250 pr_err("PQAP(QCI) failed with cc=%d", cc);
1252 return config[0] & 0x40;
1258 static void kvm_s390_set_crycb_format(struct kvm *kvm)
1260 kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
1262 if (kvm_s390_apxa_installed())
1263 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
1265 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
1268 static u64 kvm_s390_get_initial_cpuid(void)
1273 cpuid.version = 0xff;
1274 return *((u64 *) &cpuid);
1277 static void kvm_s390_crypto_init(struct kvm *kvm)
1279 if (!test_kvm_facility(kvm, 76))
1282 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
1283 kvm_s390_set_crycb_format(kvm);
1285 /* Enable AES/DEA protected key functions by default */
1286 kvm->arch.crypto.aes_kw = 1;
1287 kvm->arch.crypto.dea_kw = 1;
1288 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1289 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1290 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1291 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1294 static void sca_dispose(struct kvm *kvm)
1296 if (kvm->arch.use_esca)
1297 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
1299 free_page((unsigned long)(kvm->arch.sca));
1300 kvm->arch.sca = NULL;
1303 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
1305 gfp_t alloc_flags = GFP_KERNEL;
1307 char debug_name[16];
1308 static unsigned long sca_offset;
1311 #ifdef CONFIG_KVM_S390_UCONTROL
1312 if (type & ~KVM_VM_S390_UCONTROL)
1314 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
1321 rc = s390_enable_sie();
1327 ratelimit_state_init(&kvm->arch.sthyi_limit, 5 * HZ, 500);
1329 kvm->arch.use_esca = 0; /* start with basic SCA */
1330 if (!sclp.has_64bscao)
1331 alloc_flags |= GFP_DMA;
1332 rwlock_init(&kvm->arch.sca_lock);
1333 kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
1336 spin_lock(&kvm_lock);
1338 if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
1340 kvm->arch.sca = (struct bsca_block *)
1341 ((char *) kvm->arch.sca + sca_offset);
1342 spin_unlock(&kvm_lock);
1344 sprintf(debug_name, "kvm-%u", current->pid);
1346 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
1350 kvm->arch.sie_page2 =
1351 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1352 if (!kvm->arch.sie_page2)
1355 /* Populate the facility mask initially. */
1356 memcpy(kvm->arch.model.fac_mask, S390_lowcore.stfle_fac_list,
1357 S390_ARCH_FAC_LIST_SIZE_BYTE);
1358 for (i = 0; i < S390_ARCH_FAC_LIST_SIZE_U64; i++) {
1359 if (i < kvm_s390_fac_list_mask_size())
1360 kvm->arch.model.fac_mask[i] &= kvm_s390_fac_list_mask[i];
1362 kvm->arch.model.fac_mask[i] = 0UL;
1365 /* Populate the facility list initially. */
1366 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
1367 memcpy(kvm->arch.model.fac_list, kvm->arch.model.fac_mask,
1368 S390_ARCH_FAC_LIST_SIZE_BYTE);
1370 set_kvm_facility(kvm->arch.model.fac_mask, 74);
1371 set_kvm_facility(kvm->arch.model.fac_list, 74);
1373 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
1374 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
1376 kvm_s390_crypto_init(kvm);
1378 spin_lock_init(&kvm->arch.float_int.lock);
1379 for (i = 0; i < FIRQ_LIST_COUNT; i++)
1380 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
1381 init_waitqueue_head(&kvm->arch.ipte_wq);
1382 mutex_init(&kvm->arch.ipte_mutex);
1384 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
1385 VM_EVENT(kvm, 3, "vm created with type %lu", type);
1387 if (type & KVM_VM_S390_UCONTROL) {
1388 kvm->arch.gmap = NULL;
1389 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
1391 if (sclp.hamax == U64_MAX)
1392 kvm->arch.mem_limit = TASK_MAX_SIZE;
1394 kvm->arch.mem_limit = min_t(unsigned long, TASK_MAX_SIZE,
1396 kvm->arch.gmap = gmap_alloc(current->mm, kvm->arch.mem_limit - 1);
1397 if (!kvm->arch.gmap)
1399 kvm->arch.gmap->private = kvm;
1400 kvm->arch.gmap->pfault_enabled = 0;
1403 kvm->arch.css_support = 0;
1404 kvm->arch.use_irqchip = 0;
1405 kvm->arch.epoch = 0;
1407 spin_lock_init(&kvm->arch.start_stop_lock);
1408 KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
1412 free_page((unsigned long)kvm->arch.sie_page2);
1413 debug_unregister(kvm->arch.dbf);
1415 KVM_EVENT(3, "creation of vm failed: %d", rc);
1419 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1421 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
1422 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
1423 kvm_s390_clear_local_irqs(vcpu);
1424 kvm_clear_async_pf_completion_queue(vcpu);
1425 if (!kvm_is_ucontrol(vcpu->kvm))
1428 if (kvm_is_ucontrol(vcpu->kvm))
1429 gmap_free(vcpu->arch.gmap);
1431 if (vcpu->kvm->arch.use_cmma)
1432 kvm_s390_vcpu_unsetup_cmma(vcpu);
1433 free_page((unsigned long)(vcpu->arch.sie_block));
1435 kvm_vcpu_uninit(vcpu);
1436 kmem_cache_free(kvm_vcpu_cache, vcpu);
1439 static void kvm_free_vcpus(struct kvm *kvm)
1442 struct kvm_vcpu *vcpu;
1444 kvm_for_each_vcpu(i, vcpu, kvm)
1445 kvm_arch_vcpu_destroy(vcpu);
1447 mutex_lock(&kvm->lock);
1448 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
1449 kvm->vcpus[i] = NULL;
1451 atomic_set(&kvm->online_vcpus, 0);
1452 mutex_unlock(&kvm->lock);
1455 void kvm_arch_destroy_vm(struct kvm *kvm)
1457 kvm_free_vcpus(kvm);
1459 debug_unregister(kvm->arch.dbf);
1460 free_page((unsigned long)kvm->arch.sie_page2);
1461 if (!kvm_is_ucontrol(kvm))
1462 gmap_free(kvm->arch.gmap);
1463 kvm_s390_destroy_adapters(kvm);
1464 kvm_s390_clear_float_irqs(kvm);
1465 KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
1468 /* Section: vcpu related */
1469 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
1471 vcpu->arch.gmap = gmap_alloc(current->mm, -1UL);
1472 if (!vcpu->arch.gmap)
1474 vcpu->arch.gmap->private = vcpu->kvm;
1479 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
1481 read_lock(&vcpu->kvm->arch.sca_lock);
1482 if (vcpu->kvm->arch.use_esca) {
1483 struct esca_block *sca = vcpu->kvm->arch.sca;
1485 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
1486 sca->cpu[vcpu->vcpu_id].sda = 0;
1488 struct bsca_block *sca = vcpu->kvm->arch.sca;
1490 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
1491 sca->cpu[vcpu->vcpu_id].sda = 0;
1493 read_unlock(&vcpu->kvm->arch.sca_lock);
1496 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
1498 read_lock(&vcpu->kvm->arch.sca_lock);
1499 if (vcpu->kvm->arch.use_esca) {
1500 struct esca_block *sca = vcpu->kvm->arch.sca;
1502 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
1503 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1504 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca & ~0x3fU;
1505 vcpu->arch.sie_block->ecb2 |= 0x04U;
1506 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
1508 struct bsca_block *sca = vcpu->kvm->arch.sca;
1510 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
1511 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1512 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
1513 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
1515 read_unlock(&vcpu->kvm->arch.sca_lock);
1518 /* Basic SCA to Extended SCA data copy routines */
1519 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
1522 d->sigp_ctrl.c = s->sigp_ctrl.c;
1523 d->sigp_ctrl.scn = s->sigp_ctrl.scn;
1526 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
1530 d->ipte_control = s->ipte_control;
1532 for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
1533 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
1536 static int sca_switch_to_extended(struct kvm *kvm)
1538 struct bsca_block *old_sca = kvm->arch.sca;
1539 struct esca_block *new_sca;
1540 struct kvm_vcpu *vcpu;
1541 unsigned int vcpu_idx;
1544 new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL|__GFP_ZERO);
1548 scaoh = (u32)((u64)(new_sca) >> 32);
1549 scaol = (u32)(u64)(new_sca) & ~0x3fU;
1551 kvm_s390_vcpu_block_all(kvm);
1552 write_lock(&kvm->arch.sca_lock);
1554 sca_copy_b_to_e(new_sca, old_sca);
1556 kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
1557 vcpu->arch.sie_block->scaoh = scaoh;
1558 vcpu->arch.sie_block->scaol = scaol;
1559 vcpu->arch.sie_block->ecb2 |= 0x04U;
1561 kvm->arch.sca = new_sca;
1562 kvm->arch.use_esca = 1;
1564 write_unlock(&kvm->arch.sca_lock);
1565 kvm_s390_vcpu_unblock_all(kvm);
1567 free_page((unsigned long)old_sca);
1569 VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
1570 old_sca, kvm->arch.sca);
1574 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
1578 if (id < KVM_S390_BSCA_CPU_SLOTS)
1580 if (!sclp.has_esca || !sclp.has_64bscao)
1583 mutex_lock(&kvm->lock);
1584 rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
1585 mutex_unlock(&kvm->lock);
1587 return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
1590 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1592 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1593 kvm_clear_async_pf_completion_queue(vcpu);
1594 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
1600 if (test_kvm_facility(vcpu->kvm, 64))
1601 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
1602 /* fprs can be synchronized via vrs, even if the guest has no vx. With
1603 * MACHINE_HAS_VX, (load|store)_fpu_regs() will work with vrs format.
1606 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
1608 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
1610 if (kvm_is_ucontrol(vcpu->kvm))
1611 return __kvm_ucontrol_vcpu_init(vcpu);
1616 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1617 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1619 WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
1620 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1621 vcpu->arch.cputm_start = get_tod_clock_fast();
1622 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1625 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1626 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1628 WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
1629 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1630 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
1631 vcpu->arch.cputm_start = 0;
1632 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1635 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1636 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1638 WARN_ON_ONCE(vcpu->arch.cputm_enabled);
1639 vcpu->arch.cputm_enabled = true;
1640 __start_cpu_timer_accounting(vcpu);
1643 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1644 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1646 WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
1647 __stop_cpu_timer_accounting(vcpu);
1648 vcpu->arch.cputm_enabled = false;
1651 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1653 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1654 __enable_cpu_timer_accounting(vcpu);
1658 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1660 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1661 __disable_cpu_timer_accounting(vcpu);
1665 /* set the cpu timer - may only be called from the VCPU thread itself */
1666 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
1668 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1669 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1670 if (vcpu->arch.cputm_enabled)
1671 vcpu->arch.cputm_start = get_tod_clock_fast();
1672 vcpu->arch.sie_block->cputm = cputm;
1673 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1677 /* update and get the cpu timer - can also be called from other VCPU threads */
1678 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
1683 if (unlikely(!vcpu->arch.cputm_enabled))
1684 return vcpu->arch.sie_block->cputm;
1686 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1688 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
1690 * If the writer would ever execute a read in the critical
1691 * section, e.g. in irq context, we have a deadlock.
1693 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
1694 value = vcpu->arch.sie_block->cputm;
1695 /* if cputm_start is 0, accounting is being started/stopped */
1696 if (likely(vcpu->arch.cputm_start))
1697 value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
1698 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
1703 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1705 /* Save host register state */
1707 vcpu->arch.host_fpregs.fpc = current->thread.fpu.fpc;
1708 vcpu->arch.host_fpregs.regs = current->thread.fpu.regs;
1711 current->thread.fpu.regs = vcpu->run->s.regs.vrs;
1713 current->thread.fpu.regs = vcpu->run->s.regs.fprs;
1714 current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
1715 if (test_fp_ctl(current->thread.fpu.fpc))
1716 /* User space provided an invalid FPC, let's clear it */
1717 current->thread.fpu.fpc = 0;
1719 save_access_regs(vcpu->arch.host_acrs);
1720 restore_access_regs(vcpu->run->s.regs.acrs);
1721 gmap_enable(vcpu->arch.gmap);
1722 atomic_or(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1723 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
1724 __start_cpu_timer_accounting(vcpu);
1728 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1731 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
1732 __stop_cpu_timer_accounting(vcpu);
1733 atomic_andnot(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1734 gmap_disable(vcpu->arch.gmap);
1736 /* Save guest register state */
1738 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
1740 /* Restore host register state */
1741 current->thread.fpu.fpc = vcpu->arch.host_fpregs.fpc;
1742 current->thread.fpu.regs = vcpu->arch.host_fpregs.regs;
1744 save_access_regs(vcpu->run->s.regs.acrs);
1745 restore_access_regs(vcpu->arch.host_acrs);
1748 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
1750 /* this equals initial cpu reset in pop, but we don't switch to ESA */
1751 vcpu->arch.sie_block->gpsw.mask = 0UL;
1752 vcpu->arch.sie_block->gpsw.addr = 0UL;
1753 kvm_s390_set_prefix(vcpu, 0);
1754 kvm_s390_set_cpu_timer(vcpu, 0);
1755 vcpu->arch.sie_block->ckc = 0UL;
1756 vcpu->arch.sie_block->todpr = 0;
1757 memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64));
1758 vcpu->arch.sie_block->gcr[0] = 0xE0UL;
1759 vcpu->arch.sie_block->gcr[14] = 0xC2000000UL;
1760 /* make sure the new fpc will be lazily loaded */
1762 current->thread.fpu.fpc = 0;
1763 vcpu->arch.sie_block->gbea = 1;
1764 vcpu->arch.sie_block->pp = 0;
1765 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1766 kvm_clear_async_pf_completion_queue(vcpu);
1767 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
1768 kvm_s390_vcpu_stop(vcpu);
1769 kvm_s390_clear_local_irqs(vcpu);
1772 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1774 mutex_lock(&vcpu->kvm->lock);
1776 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
1778 mutex_unlock(&vcpu->kvm->lock);
1779 if (!kvm_is_ucontrol(vcpu->kvm)) {
1780 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
1786 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
1788 if (!test_kvm_facility(vcpu->kvm, 76))
1791 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
1793 if (vcpu->kvm->arch.crypto.aes_kw)
1794 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
1795 if (vcpu->kvm->arch.crypto.dea_kw)
1796 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
1798 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
1801 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
1803 free_page(vcpu->arch.sie_block->cbrlo);
1804 vcpu->arch.sie_block->cbrlo = 0;
1807 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
1809 vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
1810 if (!vcpu->arch.sie_block->cbrlo)
1813 vcpu->arch.sie_block->ecb2 |= 0x80;
1814 vcpu->arch.sie_block->ecb2 &= ~0x08;
1818 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
1820 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
1822 vcpu->arch.sie_block->ibc = model->ibc;
1823 if (test_kvm_facility(vcpu->kvm, 7))
1824 vcpu->arch.sie_block->fac = (u32)(u64) model->fac_list;
1827 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1831 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
1835 if (test_kvm_facility(vcpu->kvm, 78))
1836 atomic_or(CPUSTAT_GED2, &vcpu->arch.sie_block->cpuflags);
1837 else if (test_kvm_facility(vcpu->kvm, 8))
1838 atomic_or(CPUSTAT_GED, &vcpu->arch.sie_block->cpuflags);
1840 kvm_s390_vcpu_setup_model(vcpu);
1842 /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
1843 if (MACHINE_HAS_ESOP)
1844 vcpu->arch.sie_block->ecb |= 0x02;
1845 if (test_kvm_facility(vcpu->kvm, 9))
1846 vcpu->arch.sie_block->ecb |= 0x04;
1847 if (test_kvm_facility(vcpu->kvm, 73))
1848 vcpu->arch.sie_block->ecb |= 0x10;
1850 if (test_kvm_facility(vcpu->kvm, 8) && sclp.has_pfmfi)
1851 vcpu->arch.sie_block->ecb2 |= 0x08;
1852 vcpu->arch.sie_block->eca = 0x1002000U;
1854 vcpu->arch.sie_block->eca |= 0x80000000U;
1856 vcpu->arch.sie_block->eca |= 0x40000000U;
1858 vcpu->arch.sie_block->eca |= 1;
1859 if (sclp.has_sigpif)
1860 vcpu->arch.sie_block->eca |= 0x10000000U;
1861 if (test_kvm_facility(vcpu->kvm, 64))
1862 vcpu->arch.sie_block->ecb3 |= 0x01;
1863 if (test_kvm_facility(vcpu->kvm, 129)) {
1864 vcpu->arch.sie_block->eca |= 0x00020000;
1865 vcpu->arch.sie_block->ecd |= 0x20000000;
1867 vcpu->arch.sie_block->riccbd = (unsigned long) &vcpu->run->s.regs.riccb;
1868 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
1869 if (test_kvm_facility(vcpu->kvm, 74))
1870 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
1872 if (vcpu->kvm->arch.use_cmma) {
1873 rc = kvm_s390_vcpu_setup_cmma(vcpu);
1877 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1878 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
1880 kvm_s390_vcpu_crypto_setup(vcpu);
1885 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1888 struct kvm_vcpu *vcpu;
1889 struct sie_page *sie_page;
1892 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
1897 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1901 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
1905 vcpu->arch.sie_block = &sie_page->sie_block;
1906 vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
1908 /* the real guest size will always be smaller than msl */
1909 vcpu->arch.sie_block->mso = 0;
1910 vcpu->arch.sie_block->msl = sclp.hamax;
1912 vcpu->arch.sie_block->icpua = id;
1913 spin_lock_init(&vcpu->arch.local_int.lock);
1914 vcpu->arch.local_int.float_int = &kvm->arch.float_int;
1915 vcpu->arch.local_int.wq = &vcpu->wq;
1916 vcpu->arch.local_int.cpuflags = &vcpu->arch.sie_block->cpuflags;
1917 seqcount_init(&vcpu->arch.cputm_seqcount);
1919 rc = kvm_vcpu_init(vcpu, kvm, id);
1921 goto out_free_sie_block;
1922 VM_EVENT(kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK", id, vcpu,
1923 vcpu->arch.sie_block);
1924 trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
1928 free_page((unsigned long)(vcpu->arch.sie_block));
1930 kmem_cache_free(kvm_vcpu_cache, vcpu);
1935 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
1937 return kvm_s390_vcpu_has_irq(vcpu, 0);
1940 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
1942 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
1946 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
1948 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
1951 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
1953 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
1957 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
1959 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
1963 * Kick a guest cpu out of SIE and wait until SIE is not running.
1964 * If the CPU is not running (e.g. waiting as idle) the function will
1965 * return immediately. */
1966 void exit_sie(struct kvm_vcpu *vcpu)
1968 atomic_or(CPUSTAT_STOP_INT, &vcpu->arch.sie_block->cpuflags);
1969 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
1973 /* Kick a guest cpu out of SIE to process a request synchronously */
1974 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
1976 kvm_make_request(req, vcpu);
1977 kvm_s390_vcpu_request(vcpu);
1980 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long address)
1983 struct kvm *kvm = gmap->private;
1984 struct kvm_vcpu *vcpu;
1986 kvm_for_each_vcpu(i, vcpu, kvm) {
1987 /* match against both prefix pages */
1988 if (kvm_s390_get_prefix(vcpu) == (address & ~0x1000UL)) {
1989 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx", address);
1990 kvm_s390_sync_request(KVM_REQ_MMU_RELOAD, vcpu);
1995 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
1997 /* kvm common code refers to this, but never calls it */
2002 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
2003 struct kvm_one_reg *reg)
2008 case KVM_REG_S390_TODPR:
2009 r = put_user(vcpu->arch.sie_block->todpr,
2010 (u32 __user *)reg->addr);
2012 case KVM_REG_S390_EPOCHDIFF:
2013 r = put_user(vcpu->arch.sie_block->epoch,
2014 (u64 __user *)reg->addr);
2016 case KVM_REG_S390_CPU_TIMER:
2017 r = put_user(kvm_s390_get_cpu_timer(vcpu),
2018 (u64 __user *)reg->addr);
2020 case KVM_REG_S390_CLOCK_COMP:
2021 r = put_user(vcpu->arch.sie_block->ckc,
2022 (u64 __user *)reg->addr);
2024 case KVM_REG_S390_PFTOKEN:
2025 r = put_user(vcpu->arch.pfault_token,
2026 (u64 __user *)reg->addr);
2028 case KVM_REG_S390_PFCOMPARE:
2029 r = put_user(vcpu->arch.pfault_compare,
2030 (u64 __user *)reg->addr);
2032 case KVM_REG_S390_PFSELECT:
2033 r = put_user(vcpu->arch.pfault_select,
2034 (u64 __user *)reg->addr);
2036 case KVM_REG_S390_PP:
2037 r = put_user(vcpu->arch.sie_block->pp,
2038 (u64 __user *)reg->addr);
2040 case KVM_REG_S390_GBEA:
2041 r = put_user(vcpu->arch.sie_block->gbea,
2042 (u64 __user *)reg->addr);
2051 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
2052 struct kvm_one_reg *reg)
2058 case KVM_REG_S390_TODPR:
2059 r = get_user(vcpu->arch.sie_block->todpr,
2060 (u32 __user *)reg->addr);
2062 case KVM_REG_S390_EPOCHDIFF:
2063 r = get_user(vcpu->arch.sie_block->epoch,
2064 (u64 __user *)reg->addr);
2066 case KVM_REG_S390_CPU_TIMER:
2067 r = get_user(val, (u64 __user *)reg->addr);
2069 kvm_s390_set_cpu_timer(vcpu, val);
2071 case KVM_REG_S390_CLOCK_COMP:
2072 r = get_user(vcpu->arch.sie_block->ckc,
2073 (u64 __user *)reg->addr);
2075 case KVM_REG_S390_PFTOKEN:
2076 r = get_user(vcpu->arch.pfault_token,
2077 (u64 __user *)reg->addr);
2078 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2079 kvm_clear_async_pf_completion_queue(vcpu);
2081 case KVM_REG_S390_PFCOMPARE:
2082 r = get_user(vcpu->arch.pfault_compare,
2083 (u64 __user *)reg->addr);
2085 case KVM_REG_S390_PFSELECT:
2086 r = get_user(vcpu->arch.pfault_select,
2087 (u64 __user *)reg->addr);
2089 case KVM_REG_S390_PP:
2090 r = get_user(vcpu->arch.sie_block->pp,
2091 (u64 __user *)reg->addr);
2093 case KVM_REG_S390_GBEA:
2094 r = get_user(vcpu->arch.sie_block->gbea,
2095 (u64 __user *)reg->addr);
2104 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
2106 kvm_s390_vcpu_initial_reset(vcpu);
2110 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2112 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs));
2116 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2118 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
2122 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2123 struct kvm_sregs *sregs)
2125 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
2126 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
2127 restore_access_regs(vcpu->run->s.regs.acrs);
2131 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2132 struct kvm_sregs *sregs)
2134 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
2135 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
2139 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2141 /* make sure the new values will be lazily loaded */
2143 if (test_fp_ctl(fpu->fpc))
2145 current->thread.fpu.fpc = fpu->fpc;
2147 convert_fp_to_vx(current->thread.fpu.vxrs, (freg_t *)fpu->fprs);
2149 memcpy(current->thread.fpu.fprs, &fpu->fprs, sizeof(fpu->fprs));
2153 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2155 /* make sure we have the latest values */
2158 convert_vx_to_fp((freg_t *)fpu->fprs, current->thread.fpu.vxrs);
2160 memcpy(fpu->fprs, current->thread.fpu.fprs, sizeof(fpu->fprs));
2161 fpu->fpc = current->thread.fpu.fpc;
2165 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
2169 if (!is_vcpu_stopped(vcpu))
2172 vcpu->run->psw_mask = psw.mask;
2173 vcpu->run->psw_addr = psw.addr;
2178 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2179 struct kvm_translation *tr)
2181 return -EINVAL; /* not implemented yet */
2184 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
2185 KVM_GUESTDBG_USE_HW_BP | \
2186 KVM_GUESTDBG_ENABLE)
2188 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
2189 struct kvm_guest_debug *dbg)
2193 vcpu->guest_debug = 0;
2194 kvm_s390_clear_bp_data(vcpu);
2196 if (dbg->control & ~VALID_GUESTDBG_FLAGS)
2198 if (!sclp.has_gpere)
2201 if (dbg->control & KVM_GUESTDBG_ENABLE) {
2202 vcpu->guest_debug = dbg->control;
2203 /* enforce guest PER */
2204 atomic_or(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2206 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
2207 rc = kvm_s390_import_bp_data(vcpu, dbg);
2209 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2210 vcpu->arch.guestdbg.last_bp = 0;
2214 vcpu->guest_debug = 0;
2215 kvm_s390_clear_bp_data(vcpu);
2216 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2222 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2223 struct kvm_mp_state *mp_state)
2225 /* CHECK_STOP and LOAD are not supported yet */
2226 return is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
2227 KVM_MP_STATE_OPERATING;
2230 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2231 struct kvm_mp_state *mp_state)
2235 /* user space knows about this interface - let it control the state */
2236 vcpu->kvm->arch.user_cpu_state_ctrl = 1;
2238 switch (mp_state->mp_state) {
2239 case KVM_MP_STATE_STOPPED:
2240 kvm_s390_vcpu_stop(vcpu);
2242 case KVM_MP_STATE_OPERATING:
2243 kvm_s390_vcpu_start(vcpu);
2245 case KVM_MP_STATE_LOAD:
2246 case KVM_MP_STATE_CHECK_STOP:
2247 /* fall through - CHECK_STOP and LOAD are not supported yet */
2255 static bool ibs_enabled(struct kvm_vcpu *vcpu)
2257 return atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_IBS;
2260 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
2263 kvm_s390_vcpu_request_handled(vcpu);
2264 if (!vcpu->requests)
2267 * We use MMU_RELOAD just to re-arm the ipte notifier for the
2268 * guest prefix page. gmap_ipte_notify will wait on the ptl lock.
2269 * This ensures that the ipte instruction for this request has
2270 * already finished. We might race against a second unmapper that
2271 * wants to set the blocking bit. Lets just retry the request loop.
2273 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
2275 rc = gmap_ipte_notify(vcpu->arch.gmap,
2276 kvm_s390_get_prefix(vcpu),
2283 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2284 vcpu->arch.sie_block->ihcpu = 0xffff;
2288 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
2289 if (!ibs_enabled(vcpu)) {
2290 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
2291 atomic_or(CPUSTAT_IBS,
2292 &vcpu->arch.sie_block->cpuflags);
2297 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
2298 if (ibs_enabled(vcpu)) {
2299 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
2300 atomic_andnot(CPUSTAT_IBS,
2301 &vcpu->arch.sie_block->cpuflags);
2306 /* nothing to do, just clear the request */
2307 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
2312 void kvm_s390_set_tod_clock(struct kvm *kvm, u64 tod)
2314 struct kvm_vcpu *vcpu;
2317 mutex_lock(&kvm->lock);
2319 kvm->arch.epoch = tod - get_tod_clock();
2320 kvm_s390_vcpu_block_all(kvm);
2321 kvm_for_each_vcpu(i, vcpu, kvm)
2322 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
2323 kvm_s390_vcpu_unblock_all(kvm);
2325 mutex_unlock(&kvm->lock);
2329 * kvm_arch_fault_in_page - fault-in guest page if necessary
2330 * @vcpu: The corresponding virtual cpu
2331 * @gpa: Guest physical address
2332 * @writable: Whether the page should be writable or not
2334 * Make sure that a guest page has been faulted-in on the host.
2336 * Return: Zero on success, negative error code otherwise.
2338 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
2340 return gmap_fault(vcpu->arch.gmap, gpa,
2341 writable ? FAULT_FLAG_WRITE : 0);
2344 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
2345 unsigned long token)
2347 struct kvm_s390_interrupt inti;
2348 struct kvm_s390_irq irq;
2351 irq.u.ext.ext_params2 = token;
2352 irq.type = KVM_S390_INT_PFAULT_INIT;
2353 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
2355 inti.type = KVM_S390_INT_PFAULT_DONE;
2356 inti.parm64 = token;
2357 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
2361 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
2362 struct kvm_async_pf *work)
2364 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
2365 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
2368 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
2369 struct kvm_async_pf *work)
2371 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
2372 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
2375 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
2376 struct kvm_async_pf *work)
2378 /* s390 will always inject the page directly */
2381 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
2384 * s390 will always inject the page directly,
2385 * but we still want check_async_completion to cleanup
2390 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
2393 struct kvm_arch_async_pf arch;
2396 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2398 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
2399 vcpu->arch.pfault_compare)
2401 if (psw_extint_disabled(vcpu))
2403 if (kvm_s390_vcpu_has_irq(vcpu, 0))
2405 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
2407 if (!vcpu->arch.gmap->pfault_enabled)
2410 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
2411 hva += current->thread.gmap_addr & ~PAGE_MASK;
2412 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
2415 rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
2419 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
2424 * On s390 notifications for arriving pages will be delivered directly
2425 * to the guest but the house keeping for completed pfaults is
2426 * handled outside the worker.
2428 kvm_check_async_pf_completion(vcpu);
2430 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
2431 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
2436 if (test_cpu_flag(CIF_MCCK_PENDING))
2439 if (!kvm_is_ucontrol(vcpu->kvm)) {
2440 rc = kvm_s390_deliver_pending_interrupts(vcpu);
2445 rc = kvm_s390_handle_requests(vcpu);
2449 if (guestdbg_enabled(vcpu)) {
2450 kvm_s390_backup_guest_per_regs(vcpu);
2451 kvm_s390_patch_guest_per_regs(vcpu);
2454 vcpu->arch.sie_block->icptcode = 0;
2455 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
2456 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
2457 trace_kvm_s390_sie_enter(vcpu, cpuflags);
2462 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
2464 struct kvm_s390_pgm_info pgm_info = {
2465 .code = PGM_ADDRESSING,
2470 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
2471 trace_kvm_s390_sie_fault(vcpu);
2474 * We want to inject an addressing exception, which is defined as a
2475 * suppressing or terminating exception. However, since we came here
2476 * by a DAT access exception, the PSW still points to the faulting
2477 * instruction since DAT exceptions are nullifying. So we've got
2478 * to look up the current opcode to get the length of the instruction
2479 * to be able to forward the PSW.
2481 rc = read_guest_instr(vcpu, &opcode, 1);
2482 ilen = insn_length(opcode);
2486 /* Instruction-Fetching Exceptions - we can't detect the ilen.
2487 * Forward by arbitrary ilc, injection will take care of
2488 * nullification if necessary.
2490 pgm_info = vcpu->arch.pgm;
2493 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
2494 kvm_s390_forward_psw(vcpu, ilen);
2495 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
2498 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
2500 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
2501 vcpu->arch.sie_block->icptcode);
2502 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
2504 if (guestdbg_enabled(vcpu))
2505 kvm_s390_restore_guest_per_regs(vcpu);
2507 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
2508 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
2510 if (vcpu->arch.sie_block->icptcode > 0) {
2511 int rc = kvm_handle_sie_intercept(vcpu);
2513 if (rc != -EOPNOTSUPP)
2515 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
2516 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
2517 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
2518 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
2520 } else if (exit_reason != -EFAULT) {
2521 vcpu->stat.exit_null++;
2523 } else if (kvm_is_ucontrol(vcpu->kvm)) {
2524 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
2525 vcpu->run->s390_ucontrol.trans_exc_code =
2526 current->thread.gmap_addr;
2527 vcpu->run->s390_ucontrol.pgm_code = 0x10;
2529 } else if (current->thread.gmap_pfault) {
2530 trace_kvm_s390_major_guest_pfault(vcpu);
2531 current->thread.gmap_pfault = 0;
2532 if (kvm_arch_setup_async_pf(vcpu))
2534 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
2536 return vcpu_post_run_fault_in_sie(vcpu);
2539 static int __vcpu_run(struct kvm_vcpu *vcpu)
2541 int rc, exit_reason;
2544 * We try to hold kvm->srcu during most of vcpu_run (except when run-
2545 * ning the guest), so that memslots (and other stuff) are protected
2547 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2550 rc = vcpu_pre_run(vcpu);
2554 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2556 * As PF_VCPU will be used in fault handler, between
2557 * guest_enter and guest_exit should be no uaccess.
2559 local_irq_disable();
2560 __kvm_guest_enter();
2561 __disable_cpu_timer_accounting(vcpu);
2563 exit_reason = sie64a(vcpu->arch.sie_block,
2564 vcpu->run->s.regs.gprs);
2565 local_irq_disable();
2566 __enable_cpu_timer_accounting(vcpu);
2569 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2571 rc = vcpu_post_run(vcpu, exit_reason);
2572 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
2574 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2578 static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2580 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
2581 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
2582 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
2583 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
2584 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
2585 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
2586 /* some control register changes require a tlb flush */
2587 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2589 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
2590 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
2591 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
2592 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
2593 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
2594 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
2596 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
2597 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
2598 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
2599 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
2600 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2601 kvm_clear_async_pf_completion_queue(vcpu);
2603 kvm_run->kvm_dirty_regs = 0;
2606 static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2608 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
2609 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
2610 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
2611 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
2612 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
2613 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
2614 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
2615 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
2616 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
2617 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
2618 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
2619 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
2622 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2627 if (guestdbg_exit_pending(vcpu)) {
2628 kvm_s390_prepare_debug_exit(vcpu);
2632 if (vcpu->sigset_active)
2633 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2635 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
2636 kvm_s390_vcpu_start(vcpu);
2637 } else if (is_vcpu_stopped(vcpu)) {
2638 pr_err_ratelimited("can't run stopped vcpu %d\n",
2643 sync_regs(vcpu, kvm_run);
2644 enable_cpu_timer_accounting(vcpu);
2647 rc = __vcpu_run(vcpu);
2649 if (signal_pending(current) && !rc) {
2650 kvm_run->exit_reason = KVM_EXIT_INTR;
2654 if (guestdbg_exit_pending(vcpu) && !rc) {
2655 kvm_s390_prepare_debug_exit(vcpu);
2659 if (rc == -EREMOTE) {
2660 /* userspace support is needed, kvm_run has been prepared */
2664 disable_cpu_timer_accounting(vcpu);
2665 store_regs(vcpu, kvm_run);
2667 if (vcpu->sigset_active)
2668 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2670 vcpu->stat.exit_userspace++;
2675 * store status at address
2676 * we use have two special cases:
2677 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
2678 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
2680 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
2682 unsigned char archmode = 1;
2683 freg_t fprs[NUM_FPRS];
2688 px = kvm_s390_get_prefix(vcpu);
2689 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
2690 if (write_guest_abs(vcpu, 163, &archmode, 1))
2693 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
2694 if (write_guest_real(vcpu, 163, &archmode, 1))
2698 gpa -= __LC_FPREGS_SAVE_AREA;
2700 /* manually convert vector registers if necessary */
2701 if (MACHINE_HAS_VX) {
2702 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
2703 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2706 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2707 vcpu->run->s.regs.fprs, 128);
2709 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
2710 vcpu->run->s.regs.gprs, 128);
2711 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
2712 &vcpu->arch.sie_block->gpsw, 16);
2713 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
2715 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
2716 &vcpu->run->s.regs.fpc, 4);
2717 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
2718 &vcpu->arch.sie_block->todpr, 4);
2719 cputm = kvm_s390_get_cpu_timer(vcpu);
2720 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
2722 clkcomp = vcpu->arch.sie_block->ckc >> 8;
2723 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
2725 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
2726 &vcpu->run->s.regs.acrs, 64);
2727 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
2728 &vcpu->arch.sie_block->gcr, 128);
2729 return rc ? -EFAULT : 0;
2732 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
2735 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
2736 * copying in vcpu load/put. Lets update our copies before we save
2737 * it into the save area
2740 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
2741 save_access_regs(vcpu->run->s.regs.acrs);
2743 return kvm_s390_store_status_unloaded(vcpu, addr);
2747 * store additional status at address
2749 int kvm_s390_store_adtl_status_unloaded(struct kvm_vcpu *vcpu,
2752 /* Only bits 0-53 are used for address formation */
2753 if (!(gpa & ~0x3ff))
2756 return write_guest_abs(vcpu, gpa & ~0x3ff,
2757 (void *)&vcpu->run->s.regs.vrs, 512);
2760 int kvm_s390_vcpu_store_adtl_status(struct kvm_vcpu *vcpu, unsigned long addr)
2762 if (!test_kvm_facility(vcpu->kvm, 129))
2766 * The guest VXRS are in the host VXRs due to the lazy
2767 * copying in vcpu load/put. We can simply call save_fpu_regs()
2768 * to save the current register state because we are in the
2769 * middle of a load/put cycle.
2771 * Let's update our copies before we save it into the save area.
2775 return kvm_s390_store_adtl_status_unloaded(vcpu, addr);
2778 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2780 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
2781 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
2784 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
2787 struct kvm_vcpu *vcpu;
2789 kvm_for_each_vcpu(i, vcpu, kvm) {
2790 __disable_ibs_on_vcpu(vcpu);
2794 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2798 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
2799 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
2802 void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
2804 int i, online_vcpus, started_vcpus = 0;
2806 if (!is_vcpu_stopped(vcpu))
2809 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
2810 /* Only one cpu at a time may enter/leave the STOPPED state. */
2811 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2812 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2814 for (i = 0; i < online_vcpus; i++) {
2815 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i]))
2819 if (started_vcpus == 0) {
2820 /* we're the only active VCPU -> speed it up */
2821 __enable_ibs_on_vcpu(vcpu);
2822 } else if (started_vcpus == 1) {
2824 * As we are starting a second VCPU, we have to disable
2825 * the IBS facility on all VCPUs to remove potentially
2826 * oustanding ENABLE requests.
2828 __disable_ibs_on_all_vcpus(vcpu->kvm);
2831 atomic_andnot(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2833 * Another VCPU might have used IBS while we were offline.
2834 * Let's play safe and flush the VCPU at startup.
2836 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2837 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2841 void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
2843 int i, online_vcpus, started_vcpus = 0;
2844 struct kvm_vcpu *started_vcpu = NULL;
2846 if (is_vcpu_stopped(vcpu))
2849 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
2850 /* Only one cpu at a time may enter/leave the STOPPED state. */
2851 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2852 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2854 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
2855 kvm_s390_clear_stop_irq(vcpu);
2857 atomic_or(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2858 __disable_ibs_on_vcpu(vcpu);
2860 for (i = 0; i < online_vcpus; i++) {
2861 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) {
2863 started_vcpu = vcpu->kvm->vcpus[i];
2867 if (started_vcpus == 1) {
2869 * As we only have one VCPU left, we want to enable the
2870 * IBS facility for that VCPU to speed it up.
2872 __enable_ibs_on_vcpu(started_vcpu);
2875 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2879 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
2880 struct kvm_enable_cap *cap)
2888 case KVM_CAP_S390_CSS_SUPPORT:
2889 if (!vcpu->kvm->arch.css_support) {
2890 vcpu->kvm->arch.css_support = 1;
2891 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
2892 trace_kvm_s390_enable_css(vcpu->kvm);
2903 static long kvm_s390_guest_mem_op(struct kvm_vcpu *vcpu,
2904 struct kvm_s390_mem_op *mop)
2906 void __user *uaddr = (void __user *)mop->buf;
2907 void *tmpbuf = NULL;
2909 const u64 supported_flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION
2910 | KVM_S390_MEMOP_F_CHECK_ONLY;
2912 if (mop->flags & ~supported_flags)
2915 if (mop->size > MEM_OP_MAX_SIZE)
2918 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
2919 tmpbuf = vmalloc(mop->size);
2924 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2927 case KVM_S390_MEMOP_LOGICAL_READ:
2928 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2929 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
2930 mop->size, GACC_FETCH);
2933 r = read_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
2935 if (copy_to_user(uaddr, tmpbuf, mop->size))
2939 case KVM_S390_MEMOP_LOGICAL_WRITE:
2940 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2941 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
2942 mop->size, GACC_STORE);
2945 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
2949 r = write_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
2955 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
2957 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
2958 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
2964 long kvm_arch_vcpu_ioctl(struct file *filp,
2965 unsigned int ioctl, unsigned long arg)
2967 struct kvm_vcpu *vcpu = filp->private_data;
2968 void __user *argp = (void __user *)arg;
2973 case KVM_S390_IRQ: {
2974 struct kvm_s390_irq s390irq;
2977 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
2979 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
2982 case KVM_S390_INTERRUPT: {
2983 struct kvm_s390_interrupt s390int;
2984 struct kvm_s390_irq s390irq;
2987 if (copy_from_user(&s390int, argp, sizeof(s390int)))
2989 if (s390int_to_s390irq(&s390int, &s390irq))
2991 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
2994 case KVM_S390_STORE_STATUS:
2995 idx = srcu_read_lock(&vcpu->kvm->srcu);
2996 r = kvm_s390_vcpu_store_status(vcpu, arg);
2997 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2999 case KVM_S390_SET_INITIAL_PSW: {
3003 if (copy_from_user(&psw, argp, sizeof(psw)))
3005 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
3008 case KVM_S390_INITIAL_RESET:
3009 r = kvm_arch_vcpu_ioctl_initial_reset(vcpu);
3011 case KVM_SET_ONE_REG:
3012 case KVM_GET_ONE_REG: {
3013 struct kvm_one_reg reg;
3015 if (copy_from_user(®, argp, sizeof(reg)))
3017 if (ioctl == KVM_SET_ONE_REG)
3018 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®);
3020 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®);
3023 #ifdef CONFIG_KVM_S390_UCONTROL
3024 case KVM_S390_UCAS_MAP: {
3025 struct kvm_s390_ucas_mapping ucasmap;
3027 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3032 if (!kvm_is_ucontrol(vcpu->kvm)) {
3037 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
3038 ucasmap.vcpu_addr, ucasmap.length);
3041 case KVM_S390_UCAS_UNMAP: {
3042 struct kvm_s390_ucas_mapping ucasmap;
3044 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3049 if (!kvm_is_ucontrol(vcpu->kvm)) {
3054 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
3059 case KVM_S390_VCPU_FAULT: {
3060 r = gmap_fault(vcpu->arch.gmap, arg, 0);
3063 case KVM_ENABLE_CAP:
3065 struct kvm_enable_cap cap;
3067 if (copy_from_user(&cap, argp, sizeof(cap)))
3069 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3072 case KVM_S390_MEM_OP: {
3073 struct kvm_s390_mem_op mem_op;
3075 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
3076 r = kvm_s390_guest_mem_op(vcpu, &mem_op);
3081 case KVM_S390_SET_IRQ_STATE: {
3082 struct kvm_s390_irq_state irq_state;
3085 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3087 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
3088 irq_state.len == 0 ||
3089 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
3093 r = kvm_s390_set_irq_state(vcpu,
3094 (void __user *) irq_state.buf,
3098 case KVM_S390_GET_IRQ_STATE: {
3099 struct kvm_s390_irq_state irq_state;
3102 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3104 if (irq_state.len == 0) {
3108 r = kvm_s390_get_irq_state(vcpu,
3109 (__u8 __user *) irq_state.buf,
3119 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3121 #ifdef CONFIG_KVM_S390_UCONTROL
3122 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
3123 && (kvm_is_ucontrol(vcpu->kvm))) {
3124 vmf->page = virt_to_page(vcpu->arch.sie_block);
3125 get_page(vmf->page);
3129 return VM_FAULT_SIGBUS;
3132 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
3133 unsigned long npages)
3138 /* Section: memory related */
3139 int kvm_arch_prepare_memory_region(struct kvm *kvm,
3140 struct kvm_memory_slot *memslot,
3141 const struct kvm_userspace_memory_region *mem,
3142 enum kvm_mr_change change)
3144 /* A few sanity checks. We can have memory slots which have to be
3145 located/ended at a segment boundary (1MB). The memory in userland is
3146 ok to be fragmented into various different vmas. It is okay to mmap()
3147 and munmap() stuff in this slot after doing this call at any time */
3149 if (mem->userspace_addr & 0xffffful)
3152 if (mem->memory_size & 0xffffful)
3155 if (mem->guest_phys_addr + mem->memory_size > kvm->arch.mem_limit)
3161 void kvm_arch_commit_memory_region(struct kvm *kvm,
3162 const struct kvm_userspace_memory_region *mem,
3163 const struct kvm_memory_slot *old,
3164 const struct kvm_memory_slot *new,
3165 enum kvm_mr_change change)
3169 /* If the basics of the memslot do not change, we do not want
3170 * to update the gmap. Every update causes several unnecessary
3171 * segment translation exceptions. This is usually handled just
3172 * fine by the normal fault handler + gmap, but it will also
3173 * cause faults on the prefix page of running guest CPUs.
3175 if (old->userspace_addr == mem->userspace_addr &&
3176 old->base_gfn * PAGE_SIZE == mem->guest_phys_addr &&
3177 old->npages * PAGE_SIZE == mem->memory_size)
3180 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr,
3181 mem->guest_phys_addr, mem->memory_size);
3183 pr_warn("failed to commit memory region\n");
3187 static inline unsigned long nonhyp_mask(int i)
3189 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
3191 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
3194 void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu)
3196 vcpu->valid_wakeup = false;
3199 static int __init kvm_s390_init(void)
3203 if (!sclp.has_sief2) {
3204 pr_info("SIE not available\n");
3208 for (i = 0; i < 16; i++)
3209 kvm_s390_fac_list_mask[i] |=
3210 S390_lowcore.stfle_fac_list[i] & nonhyp_mask(i);
3212 return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
3215 static void __exit kvm_s390_exit(void)
3220 module_init(kvm_s390_init);
3221 module_exit(kvm_s390_exit);
3224 * Enable autoloading of the kvm module.
3225 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
3226 * since x86 takes a different approach.
3228 #include <linux/miscdevice.h>
3229 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3230 MODULE_ALIAS("devname:kvm");