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/mman.h>
25 #include <linux/module.h>
26 #include <linux/random.h>
27 #include <linux/slab.h>
28 #include <linux/timer.h>
29 #include <linux/vmalloc.h>
30 #include <linux/bitmap.h>
31 #include <asm/asm-offsets.h>
32 #include <asm/lowcore.h>
34 #include <asm/pgtable.h>
37 #include <asm/switch_to.h>
40 #include <asm/cpacf.h>
41 #include <asm/timex.h>
45 #define KMSG_COMPONENT "kvm-s390"
47 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
49 #define CREATE_TRACE_POINTS
51 #include "trace-s390.h"
53 #define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */
55 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
56 (KVM_MAX_VCPUS + LOCAL_IRQS))
58 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
60 struct kvm_stats_debugfs_item debugfs_entries[] = {
61 { "userspace_handled", VCPU_STAT(exit_userspace) },
62 { "exit_null", VCPU_STAT(exit_null) },
63 { "exit_validity", VCPU_STAT(exit_validity) },
64 { "exit_stop_request", VCPU_STAT(exit_stop_request) },
65 { "exit_external_request", VCPU_STAT(exit_external_request) },
66 { "exit_external_interrupt", VCPU_STAT(exit_external_interrupt) },
67 { "exit_instruction", VCPU_STAT(exit_instruction) },
68 { "exit_pei", VCPU_STAT(exit_pei) },
69 { "exit_program_interruption", VCPU_STAT(exit_program_interruption) },
70 { "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program) },
71 { "exit_operation_exception", VCPU_STAT(exit_operation_exception) },
72 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
73 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
74 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
75 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
76 { "instruction_lctlg", VCPU_STAT(instruction_lctlg) },
77 { "instruction_lctl", VCPU_STAT(instruction_lctl) },
78 { "instruction_stctl", VCPU_STAT(instruction_stctl) },
79 { "instruction_stctg", VCPU_STAT(instruction_stctg) },
80 { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal) },
81 { "deliver_external_call", VCPU_STAT(deliver_external_call) },
82 { "deliver_service_signal", VCPU_STAT(deliver_service_signal) },
83 { "deliver_virtio_interrupt", VCPU_STAT(deliver_virtio_interrupt) },
84 { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal) },
85 { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal) },
86 { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal) },
87 { "deliver_program_interruption", VCPU_STAT(deliver_program_int) },
88 { "exit_wait_state", VCPU_STAT(exit_wait_state) },
89 { "instruction_pfmf", VCPU_STAT(instruction_pfmf) },
90 { "instruction_stidp", VCPU_STAT(instruction_stidp) },
91 { "instruction_spx", VCPU_STAT(instruction_spx) },
92 { "instruction_stpx", VCPU_STAT(instruction_stpx) },
93 { "instruction_stap", VCPU_STAT(instruction_stap) },
94 { "instruction_storage_key", VCPU_STAT(instruction_storage_key) },
95 { "instruction_ipte_interlock", VCPU_STAT(instruction_ipte_interlock) },
96 { "instruction_stsch", VCPU_STAT(instruction_stsch) },
97 { "instruction_chsc", VCPU_STAT(instruction_chsc) },
98 { "instruction_essa", VCPU_STAT(instruction_essa) },
99 { "instruction_stsi", VCPU_STAT(instruction_stsi) },
100 { "instruction_stfl", VCPU_STAT(instruction_stfl) },
101 { "instruction_tprot", VCPU_STAT(instruction_tprot) },
102 { "instruction_sthyi", VCPU_STAT(instruction_sthyi) },
103 { "instruction_sie", VCPU_STAT(instruction_sie) },
104 { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense) },
105 { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running) },
106 { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call) },
107 { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency) },
108 { "instruction_sigp_cond_emergency", VCPU_STAT(instruction_sigp_cond_emergency) },
109 { "instruction_sigp_start", VCPU_STAT(instruction_sigp_start) },
110 { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop) },
111 { "instruction_sigp_stop_store_status", VCPU_STAT(instruction_sigp_stop_store_status) },
112 { "instruction_sigp_store_status", VCPU_STAT(instruction_sigp_store_status) },
113 { "instruction_sigp_store_adtl_status", VCPU_STAT(instruction_sigp_store_adtl_status) },
114 { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch) },
115 { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix) },
116 { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart) },
117 { "instruction_sigp_cpu_reset", VCPU_STAT(instruction_sigp_cpu_reset) },
118 { "instruction_sigp_init_cpu_reset", VCPU_STAT(instruction_sigp_init_cpu_reset) },
119 { "instruction_sigp_unknown", VCPU_STAT(instruction_sigp_unknown) },
120 { "diagnose_10", VCPU_STAT(diagnose_10) },
121 { "diagnose_44", VCPU_STAT(diagnose_44) },
122 { "diagnose_9c", VCPU_STAT(diagnose_9c) },
123 { "diagnose_258", VCPU_STAT(diagnose_258) },
124 { "diagnose_308", VCPU_STAT(diagnose_308) },
125 { "diagnose_500", VCPU_STAT(diagnose_500) },
129 /* allow nested virtualization in KVM (if enabled by user space) */
131 module_param(nested, int, S_IRUGO);
132 MODULE_PARM_DESC(nested, "Nested virtualization support");
134 /* upper facilities limit for kvm */
135 unsigned long kvm_s390_fac_list_mask[16] = { FACILITIES_KVM };
137 unsigned long kvm_s390_fac_list_mask_size(void)
139 BUILD_BUG_ON(ARRAY_SIZE(kvm_s390_fac_list_mask) > S390_ARCH_FAC_MASK_SIZE_U64);
140 return ARRAY_SIZE(kvm_s390_fac_list_mask);
143 /* available cpu features supported by kvm */
144 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
145 /* available subfunctions indicated via query / "test bit" */
146 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
148 static struct gmap_notifier gmap_notifier;
149 static struct gmap_notifier vsie_gmap_notifier;
150 debug_info_t *kvm_s390_dbf;
152 /* Section: not file related */
153 int kvm_arch_hardware_enable(void)
155 /* every s390 is virtualization enabled ;-) */
159 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
163 * This callback is executed during stop_machine(). All CPUs are therefore
164 * temporarily stopped. In order not to change guest behavior, we have to
165 * disable preemption whenever we touch the epoch of kvm and the VCPUs,
166 * so a CPU won't be stopped while calculating with the epoch.
168 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
172 struct kvm_vcpu *vcpu;
174 unsigned long long *delta = v;
176 list_for_each_entry(kvm, &vm_list, vm_list) {
177 kvm->arch.epoch -= *delta;
178 kvm_for_each_vcpu(i, vcpu, kvm) {
179 vcpu->arch.sie_block->epoch -= *delta;
180 if (vcpu->arch.cputm_enabled)
181 vcpu->arch.cputm_start += *delta;
182 if (vcpu->arch.vsie_block)
183 vcpu->arch.vsie_block->epoch -= *delta;
189 static struct notifier_block kvm_clock_notifier = {
190 .notifier_call = kvm_clock_sync,
193 int kvm_arch_hardware_setup(void)
195 gmap_notifier.notifier_call = kvm_gmap_notifier;
196 gmap_register_pte_notifier(&gmap_notifier);
197 vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
198 gmap_register_pte_notifier(&vsie_gmap_notifier);
199 atomic_notifier_chain_register(&s390_epoch_delta_notifier,
200 &kvm_clock_notifier);
204 void kvm_arch_hardware_unsetup(void)
206 gmap_unregister_pte_notifier(&gmap_notifier);
207 gmap_unregister_pte_notifier(&vsie_gmap_notifier);
208 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
209 &kvm_clock_notifier);
212 static void allow_cpu_feat(unsigned long nr)
214 set_bit_inv(nr, kvm_s390_available_cpu_feat);
217 static inline int plo_test_bit(unsigned char nr)
219 register unsigned long r0 asm("0") = (unsigned long) nr | 0x100;
220 int cc = 3; /* subfunction not available */
223 /* Parameter registers are ignored for "test bit" */
233 static void kvm_s390_cpu_feat_init(void)
237 for (i = 0; i < 256; ++i) {
239 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
242 if (test_facility(28)) /* TOD-clock steering */
243 ptff(kvm_s390_available_subfunc.ptff,
244 sizeof(kvm_s390_available_subfunc.ptff),
247 if (test_facility(17)) { /* MSA */
248 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
249 kvm_s390_available_subfunc.kmac);
250 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
251 kvm_s390_available_subfunc.kmc);
252 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
253 kvm_s390_available_subfunc.km);
254 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
255 kvm_s390_available_subfunc.kimd);
256 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
257 kvm_s390_available_subfunc.klmd);
259 if (test_facility(76)) /* MSA3 */
260 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
261 kvm_s390_available_subfunc.pckmo);
262 if (test_facility(77)) { /* MSA4 */
263 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
264 kvm_s390_available_subfunc.kmctr);
265 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
266 kvm_s390_available_subfunc.kmf);
267 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
268 kvm_s390_available_subfunc.kmo);
269 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
270 kvm_s390_available_subfunc.pcc);
272 if (test_facility(57)) /* MSA5 */
273 __cpacf_query(CPACF_PPNO, (cpacf_mask_t *)
274 kvm_s390_available_subfunc.ppno);
276 if (MACHINE_HAS_ESOP)
277 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
279 * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
280 * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
282 if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
283 !test_facility(3) || !nested)
285 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
286 if (sclp.has_64bscao)
287 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
289 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
291 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
293 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
295 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
297 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
299 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
301 * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
302 * all skey handling functions read/set the skey from the PGSTE
303 * instead of the real storage key.
305 * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
306 * pages being detected as preserved although they are resident.
308 * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
309 * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
311 * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
312 * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
313 * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
315 * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
316 * cannot easily shadow the SCA because of the ipte lock.
320 int kvm_arch_init(void *opaque)
322 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
326 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view)) {
327 debug_unregister(kvm_s390_dbf);
331 kvm_s390_cpu_feat_init();
333 /* Register floating interrupt controller interface. */
334 return kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
337 void kvm_arch_exit(void)
339 debug_unregister(kvm_s390_dbf);
342 /* Section: device related */
343 long kvm_arch_dev_ioctl(struct file *filp,
344 unsigned int ioctl, unsigned long arg)
346 if (ioctl == KVM_S390_ENABLE_SIE)
347 return s390_enable_sie();
351 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
356 case KVM_CAP_S390_PSW:
357 case KVM_CAP_S390_GMAP:
358 case KVM_CAP_SYNC_MMU:
359 #ifdef CONFIG_KVM_S390_UCONTROL
360 case KVM_CAP_S390_UCONTROL:
362 case KVM_CAP_ASYNC_PF:
363 case KVM_CAP_SYNC_REGS:
364 case KVM_CAP_ONE_REG:
365 case KVM_CAP_ENABLE_CAP:
366 case KVM_CAP_S390_CSS_SUPPORT:
367 case KVM_CAP_IOEVENTFD:
368 case KVM_CAP_DEVICE_CTRL:
369 case KVM_CAP_ENABLE_CAP_VM:
370 case KVM_CAP_S390_IRQCHIP:
371 case KVM_CAP_VM_ATTRIBUTES:
372 case KVM_CAP_MP_STATE:
373 case KVM_CAP_S390_INJECT_IRQ:
374 case KVM_CAP_S390_USER_SIGP:
375 case KVM_CAP_S390_USER_STSI:
376 case KVM_CAP_S390_SKEYS:
377 case KVM_CAP_S390_IRQ_STATE:
378 case KVM_CAP_S390_USER_INSTR0:
381 case KVM_CAP_S390_MEM_OP:
384 case KVM_CAP_NR_VCPUS:
385 case KVM_CAP_MAX_VCPUS:
386 r = KVM_S390_BSCA_CPU_SLOTS;
387 if (!kvm_s390_use_sca_entries())
389 else if (sclp.has_esca && sclp.has_64bscao)
390 r = KVM_S390_ESCA_CPU_SLOTS;
392 case KVM_CAP_NR_MEMSLOTS:
393 r = KVM_USER_MEM_SLOTS;
395 case KVM_CAP_S390_COW:
396 r = MACHINE_HAS_ESOP;
398 case KVM_CAP_S390_VECTOR_REGISTERS:
401 case KVM_CAP_S390_RI:
402 r = test_facility(64);
410 static void kvm_s390_sync_dirty_log(struct kvm *kvm,
411 struct kvm_memory_slot *memslot)
413 gfn_t cur_gfn, last_gfn;
414 unsigned long address;
415 struct gmap *gmap = kvm->arch.gmap;
417 /* Loop over all guest pages */
418 last_gfn = memslot->base_gfn + memslot->npages;
419 for (cur_gfn = memslot->base_gfn; cur_gfn <= last_gfn; cur_gfn++) {
420 address = gfn_to_hva_memslot(memslot, cur_gfn);
422 if (test_and_clear_guest_dirty(gmap->mm, address))
423 mark_page_dirty(kvm, cur_gfn);
424 if (fatal_signal_pending(current))
430 /* Section: vm related */
431 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
434 * Get (and clear) the dirty memory log for a memory slot.
436 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
437 struct kvm_dirty_log *log)
441 struct kvm_memslots *slots;
442 struct kvm_memory_slot *memslot;
445 mutex_lock(&kvm->slots_lock);
448 if (log->slot >= KVM_USER_MEM_SLOTS)
451 slots = kvm_memslots(kvm);
452 memslot = id_to_memslot(slots, log->slot);
454 if (!memslot->dirty_bitmap)
457 kvm_s390_sync_dirty_log(kvm, memslot);
458 r = kvm_get_dirty_log(kvm, log, &is_dirty);
462 /* Clear the dirty log */
464 n = kvm_dirty_bitmap_bytes(memslot);
465 memset(memslot->dirty_bitmap, 0, n);
469 mutex_unlock(&kvm->slots_lock);
473 static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
476 struct kvm_vcpu *vcpu;
478 kvm_for_each_vcpu(i, vcpu, kvm) {
479 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
483 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
491 case KVM_CAP_S390_IRQCHIP:
492 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
493 kvm->arch.use_irqchip = 1;
496 case KVM_CAP_S390_USER_SIGP:
497 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
498 kvm->arch.user_sigp = 1;
501 case KVM_CAP_S390_VECTOR_REGISTERS:
502 mutex_lock(&kvm->lock);
503 if (kvm->created_vcpus) {
505 } else if (MACHINE_HAS_VX) {
506 set_kvm_facility(kvm->arch.model.fac_mask, 129);
507 set_kvm_facility(kvm->arch.model.fac_list, 129);
511 mutex_unlock(&kvm->lock);
512 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
513 r ? "(not available)" : "(success)");
515 case KVM_CAP_S390_RI:
517 mutex_lock(&kvm->lock);
518 if (kvm->created_vcpus) {
520 } else if (test_facility(64)) {
521 set_kvm_facility(kvm->arch.model.fac_mask, 64);
522 set_kvm_facility(kvm->arch.model.fac_list, 64);
525 mutex_unlock(&kvm->lock);
526 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
527 r ? "(not available)" : "(success)");
529 case KVM_CAP_S390_USER_STSI:
530 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
531 kvm->arch.user_stsi = 1;
534 case KVM_CAP_S390_USER_INSTR0:
535 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
536 kvm->arch.user_instr0 = 1;
537 icpt_operexc_on_all_vcpus(kvm);
547 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
551 switch (attr->attr) {
552 case KVM_S390_VM_MEM_LIMIT_SIZE:
554 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
555 kvm->arch.mem_limit);
556 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
566 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
570 switch (attr->attr) {
571 case KVM_S390_VM_MEM_ENABLE_CMMA:
577 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
578 mutex_lock(&kvm->lock);
579 if (!kvm->created_vcpus) {
580 kvm->arch.use_cmma = 1;
583 mutex_unlock(&kvm->lock);
585 case KVM_S390_VM_MEM_CLR_CMMA:
590 if (!kvm->arch.use_cmma)
593 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
594 mutex_lock(&kvm->lock);
595 idx = srcu_read_lock(&kvm->srcu);
596 s390_reset_cmma(kvm->arch.gmap->mm);
597 srcu_read_unlock(&kvm->srcu, idx);
598 mutex_unlock(&kvm->lock);
601 case KVM_S390_VM_MEM_LIMIT_SIZE: {
602 unsigned long new_limit;
604 if (kvm_is_ucontrol(kvm))
607 if (get_user(new_limit, (u64 __user *)attr->addr))
610 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
611 new_limit > kvm->arch.mem_limit)
617 /* gmap_create takes last usable address */
618 if (new_limit != KVM_S390_NO_MEM_LIMIT)
622 mutex_lock(&kvm->lock);
623 if (!kvm->created_vcpus) {
624 /* gmap_create will round the limit up */
625 struct gmap *new = gmap_create(current->mm, new_limit);
630 gmap_remove(kvm->arch.gmap);
632 kvm->arch.gmap = new;
636 mutex_unlock(&kvm->lock);
637 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
638 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
639 (void *) kvm->arch.gmap->asce);
649 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
651 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
653 struct kvm_vcpu *vcpu;
656 if (!test_kvm_facility(kvm, 76))
659 mutex_lock(&kvm->lock);
660 switch (attr->attr) {
661 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
663 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
664 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
665 kvm->arch.crypto.aes_kw = 1;
666 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
668 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
670 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
671 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
672 kvm->arch.crypto.dea_kw = 1;
673 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
675 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
676 kvm->arch.crypto.aes_kw = 0;
677 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
678 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
679 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
681 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
682 kvm->arch.crypto.dea_kw = 0;
683 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
684 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
685 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
688 mutex_unlock(&kvm->lock);
692 kvm_for_each_vcpu(i, vcpu, kvm) {
693 kvm_s390_vcpu_crypto_setup(vcpu);
696 mutex_unlock(&kvm->lock);
700 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
704 if (copy_from_user(>od_high, (void __user *)attr->addr,
710 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
715 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
719 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod)))
722 kvm_s390_set_tod_clock(kvm, gtod);
723 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod);
727 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
734 switch (attr->attr) {
735 case KVM_S390_VM_TOD_HIGH:
736 ret = kvm_s390_set_tod_high(kvm, attr);
738 case KVM_S390_VM_TOD_LOW:
739 ret = kvm_s390_set_tod_low(kvm, attr);
748 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
752 if (copy_to_user((void __user *)attr->addr, >od_high,
755 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
760 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
764 gtod = kvm_s390_get_tod_clock_fast(kvm);
765 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
767 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
772 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
779 switch (attr->attr) {
780 case KVM_S390_VM_TOD_HIGH:
781 ret = kvm_s390_get_tod_high(kvm, attr);
783 case KVM_S390_VM_TOD_LOW:
784 ret = kvm_s390_get_tod_low(kvm, attr);
793 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
795 struct kvm_s390_vm_cpu_processor *proc;
796 u16 lowest_ibc, unblocked_ibc;
799 mutex_lock(&kvm->lock);
800 if (kvm->created_vcpus) {
804 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
809 if (!copy_from_user(proc, (void __user *)attr->addr,
811 kvm->arch.model.cpuid = proc->cpuid;
812 lowest_ibc = sclp.ibc >> 16 & 0xfff;
813 unblocked_ibc = sclp.ibc & 0xfff;
814 if (lowest_ibc && proc->ibc) {
815 if (proc->ibc > unblocked_ibc)
816 kvm->arch.model.ibc = unblocked_ibc;
817 else if (proc->ibc < lowest_ibc)
818 kvm->arch.model.ibc = lowest_ibc;
820 kvm->arch.model.ibc = proc->ibc;
822 memcpy(kvm->arch.model.fac_list, proc->fac_list,
823 S390_ARCH_FAC_LIST_SIZE_BYTE);
828 mutex_unlock(&kvm->lock);
832 static int kvm_s390_set_processor_feat(struct kvm *kvm,
833 struct kvm_device_attr *attr)
835 struct kvm_s390_vm_cpu_feat data;
838 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
840 if (!bitmap_subset((unsigned long *) data.feat,
841 kvm_s390_available_cpu_feat,
842 KVM_S390_VM_CPU_FEAT_NR_BITS))
845 mutex_lock(&kvm->lock);
846 if (!atomic_read(&kvm->online_vcpus)) {
847 bitmap_copy(kvm->arch.cpu_feat, (unsigned long *) data.feat,
848 KVM_S390_VM_CPU_FEAT_NR_BITS);
851 mutex_unlock(&kvm->lock);
855 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
856 struct kvm_device_attr *attr)
859 * Once supported by kernel + hw, we have to store the subfunctions
860 * in kvm->arch and remember that user space configured them.
865 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
869 switch (attr->attr) {
870 case KVM_S390_VM_CPU_PROCESSOR:
871 ret = kvm_s390_set_processor(kvm, attr);
873 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
874 ret = kvm_s390_set_processor_feat(kvm, attr);
876 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
877 ret = kvm_s390_set_processor_subfunc(kvm, attr);
883 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
885 struct kvm_s390_vm_cpu_processor *proc;
888 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
893 proc->cpuid = kvm->arch.model.cpuid;
894 proc->ibc = kvm->arch.model.ibc;
895 memcpy(&proc->fac_list, kvm->arch.model.fac_list,
896 S390_ARCH_FAC_LIST_SIZE_BYTE);
897 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
904 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
906 struct kvm_s390_vm_cpu_machine *mach;
909 mach = kzalloc(sizeof(*mach), GFP_KERNEL);
914 get_cpu_id((struct cpuid *) &mach->cpuid);
915 mach->ibc = sclp.ibc;
916 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
917 S390_ARCH_FAC_LIST_SIZE_BYTE);
918 memcpy((unsigned long *)&mach->fac_list, S390_lowcore.stfle_fac_list,
919 S390_ARCH_FAC_LIST_SIZE_BYTE);
920 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
927 static int kvm_s390_get_processor_feat(struct kvm *kvm,
928 struct kvm_device_attr *attr)
930 struct kvm_s390_vm_cpu_feat data;
932 bitmap_copy((unsigned long *) data.feat, kvm->arch.cpu_feat,
933 KVM_S390_VM_CPU_FEAT_NR_BITS);
934 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
939 static int kvm_s390_get_machine_feat(struct kvm *kvm,
940 struct kvm_device_attr *attr)
942 struct kvm_s390_vm_cpu_feat data;
944 bitmap_copy((unsigned long *) data.feat,
945 kvm_s390_available_cpu_feat,
946 KVM_S390_VM_CPU_FEAT_NR_BITS);
947 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
952 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
953 struct kvm_device_attr *attr)
956 * Once we can actually configure subfunctions (kernel + hw support),
957 * we have to check if they were already set by user space, if so copy
958 * them from kvm->arch.
963 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
964 struct kvm_device_attr *attr)
966 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
967 sizeof(struct kvm_s390_vm_cpu_subfunc)))
971 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
975 switch (attr->attr) {
976 case KVM_S390_VM_CPU_PROCESSOR:
977 ret = kvm_s390_get_processor(kvm, attr);
979 case KVM_S390_VM_CPU_MACHINE:
980 ret = kvm_s390_get_machine(kvm, attr);
982 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
983 ret = kvm_s390_get_processor_feat(kvm, attr);
985 case KVM_S390_VM_CPU_MACHINE_FEAT:
986 ret = kvm_s390_get_machine_feat(kvm, attr);
988 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
989 ret = kvm_s390_get_processor_subfunc(kvm, attr);
991 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
992 ret = kvm_s390_get_machine_subfunc(kvm, attr);
998 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1002 switch (attr->group) {
1003 case KVM_S390_VM_MEM_CTRL:
1004 ret = kvm_s390_set_mem_control(kvm, attr);
1006 case KVM_S390_VM_TOD:
1007 ret = kvm_s390_set_tod(kvm, attr);
1009 case KVM_S390_VM_CPU_MODEL:
1010 ret = kvm_s390_set_cpu_model(kvm, attr);
1012 case KVM_S390_VM_CRYPTO:
1013 ret = kvm_s390_vm_set_crypto(kvm, attr);
1023 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1027 switch (attr->group) {
1028 case KVM_S390_VM_MEM_CTRL:
1029 ret = kvm_s390_get_mem_control(kvm, attr);
1031 case KVM_S390_VM_TOD:
1032 ret = kvm_s390_get_tod(kvm, attr);
1034 case KVM_S390_VM_CPU_MODEL:
1035 ret = kvm_s390_get_cpu_model(kvm, attr);
1045 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1049 switch (attr->group) {
1050 case KVM_S390_VM_MEM_CTRL:
1051 switch (attr->attr) {
1052 case KVM_S390_VM_MEM_ENABLE_CMMA:
1053 case KVM_S390_VM_MEM_CLR_CMMA:
1054 ret = sclp.has_cmma ? 0 : -ENXIO;
1056 case KVM_S390_VM_MEM_LIMIT_SIZE:
1064 case KVM_S390_VM_TOD:
1065 switch (attr->attr) {
1066 case KVM_S390_VM_TOD_LOW:
1067 case KVM_S390_VM_TOD_HIGH:
1075 case KVM_S390_VM_CPU_MODEL:
1076 switch (attr->attr) {
1077 case KVM_S390_VM_CPU_PROCESSOR:
1078 case KVM_S390_VM_CPU_MACHINE:
1079 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1080 case KVM_S390_VM_CPU_MACHINE_FEAT:
1081 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1084 /* configuring subfunctions is not supported yet */
1085 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1091 case KVM_S390_VM_CRYPTO:
1092 switch (attr->attr) {
1093 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1094 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1095 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1096 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1112 static long kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1118 if (args->flags != 0)
1121 /* Is this guest using storage keys? */
1122 if (!mm_use_skey(current->mm))
1123 return KVM_S390_GET_SKEYS_NONE;
1125 /* Enforce sane limit on memory allocation */
1126 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1129 keys = kmalloc_array(args->count, sizeof(uint8_t),
1130 GFP_KERNEL | __GFP_NOWARN);
1132 keys = vmalloc(sizeof(uint8_t) * args->count);
1136 down_read(¤t->mm->mmap_sem);
1137 for (i = 0; i < args->count; i++) {
1138 hva = gfn_to_hva(kvm, args->start_gfn + i);
1139 if (kvm_is_error_hva(hva)) {
1144 r = get_guest_storage_key(current->mm, hva, &keys[i]);
1148 up_read(¤t->mm->mmap_sem);
1151 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
1152 sizeof(uint8_t) * args->count);
1161 static long kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1167 if (args->flags != 0)
1170 /* Enforce sane limit on memory allocation */
1171 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1174 keys = kmalloc_array(args->count, sizeof(uint8_t),
1175 GFP_KERNEL | __GFP_NOWARN);
1177 keys = vmalloc(sizeof(uint8_t) * args->count);
1181 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
1182 sizeof(uint8_t) * args->count);
1188 /* Enable storage key handling for the guest */
1189 r = s390_enable_skey();
1193 down_read(¤t->mm->mmap_sem);
1194 for (i = 0; i < args->count; i++) {
1195 hva = gfn_to_hva(kvm, args->start_gfn + i);
1196 if (kvm_is_error_hva(hva)) {
1201 /* Lowest order bit is reserved */
1202 if (keys[i] & 0x01) {
1207 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
1211 up_read(¤t->mm->mmap_sem);
1217 long kvm_arch_vm_ioctl(struct file *filp,
1218 unsigned int ioctl, unsigned long arg)
1220 struct kvm *kvm = filp->private_data;
1221 void __user *argp = (void __user *)arg;
1222 struct kvm_device_attr attr;
1226 case KVM_S390_INTERRUPT: {
1227 struct kvm_s390_interrupt s390int;
1230 if (copy_from_user(&s390int, argp, sizeof(s390int)))
1232 r = kvm_s390_inject_vm(kvm, &s390int);
1235 case KVM_ENABLE_CAP: {
1236 struct kvm_enable_cap cap;
1238 if (copy_from_user(&cap, argp, sizeof(cap)))
1240 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1243 case KVM_CREATE_IRQCHIP: {
1244 struct kvm_irq_routing_entry routing;
1247 if (kvm->arch.use_irqchip) {
1248 /* Set up dummy routing. */
1249 memset(&routing, 0, sizeof(routing));
1250 r = kvm_set_irq_routing(kvm, &routing, 0, 0);
1254 case KVM_SET_DEVICE_ATTR: {
1256 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1258 r = kvm_s390_vm_set_attr(kvm, &attr);
1261 case KVM_GET_DEVICE_ATTR: {
1263 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1265 r = kvm_s390_vm_get_attr(kvm, &attr);
1268 case KVM_HAS_DEVICE_ATTR: {
1270 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1272 r = kvm_s390_vm_has_attr(kvm, &attr);
1275 case KVM_S390_GET_SKEYS: {
1276 struct kvm_s390_skeys args;
1279 if (copy_from_user(&args, argp,
1280 sizeof(struct kvm_s390_skeys)))
1282 r = kvm_s390_get_skeys(kvm, &args);
1285 case KVM_S390_SET_SKEYS: {
1286 struct kvm_s390_skeys args;
1289 if (copy_from_user(&args, argp,
1290 sizeof(struct kvm_s390_skeys)))
1292 r = kvm_s390_set_skeys(kvm, &args);
1302 static int kvm_s390_query_ap_config(u8 *config)
1304 u32 fcn_code = 0x04000000UL;
1307 memset(config, 0, 128);
1311 ".long 0xb2af0000\n" /* PQAP(QCI) */
1317 : "r" (fcn_code), "r" (config)
1318 : "cc", "0", "2", "memory"
1324 static int kvm_s390_apxa_installed(void)
1329 if (test_facility(12)) {
1330 cc = kvm_s390_query_ap_config(config);
1333 pr_err("PQAP(QCI) failed with cc=%d", cc);
1335 return config[0] & 0x40;
1341 static void kvm_s390_set_crycb_format(struct kvm *kvm)
1343 kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
1345 if (kvm_s390_apxa_installed())
1346 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
1348 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
1351 static u64 kvm_s390_get_initial_cpuid(void)
1356 cpuid.version = 0xff;
1357 return *((u64 *) &cpuid);
1360 static void kvm_s390_crypto_init(struct kvm *kvm)
1362 if (!test_kvm_facility(kvm, 76))
1365 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
1366 kvm_s390_set_crycb_format(kvm);
1368 /* Enable AES/DEA protected key functions by default */
1369 kvm->arch.crypto.aes_kw = 1;
1370 kvm->arch.crypto.dea_kw = 1;
1371 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1372 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1373 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1374 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1377 static void sca_dispose(struct kvm *kvm)
1379 if (kvm->arch.use_esca)
1380 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
1382 free_page((unsigned long)(kvm->arch.sca));
1383 kvm->arch.sca = NULL;
1386 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
1388 gfp_t alloc_flags = GFP_KERNEL;
1390 char debug_name[16];
1391 static unsigned long sca_offset;
1394 #ifdef CONFIG_KVM_S390_UCONTROL
1395 if (type & ~KVM_VM_S390_UCONTROL)
1397 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
1404 rc = s390_enable_sie();
1410 ratelimit_state_init(&kvm->arch.sthyi_limit, 5 * HZ, 500);
1412 kvm->arch.use_esca = 0; /* start with basic SCA */
1413 if (!sclp.has_64bscao)
1414 alloc_flags |= GFP_DMA;
1415 rwlock_init(&kvm->arch.sca_lock);
1416 kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
1419 spin_lock(&kvm_lock);
1421 if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
1423 kvm->arch.sca = (struct bsca_block *)
1424 ((char *) kvm->arch.sca + sca_offset);
1425 spin_unlock(&kvm_lock);
1427 sprintf(debug_name, "kvm-%u", current->pid);
1429 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
1433 kvm->arch.sie_page2 =
1434 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1435 if (!kvm->arch.sie_page2)
1438 /* Populate the facility mask initially. */
1439 memcpy(kvm->arch.model.fac_mask, S390_lowcore.stfle_fac_list,
1440 S390_ARCH_FAC_LIST_SIZE_BYTE);
1441 for (i = 0; i < S390_ARCH_FAC_LIST_SIZE_U64; i++) {
1442 if (i < kvm_s390_fac_list_mask_size())
1443 kvm->arch.model.fac_mask[i] &= kvm_s390_fac_list_mask[i];
1445 kvm->arch.model.fac_mask[i] = 0UL;
1448 /* Populate the facility list initially. */
1449 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
1450 memcpy(kvm->arch.model.fac_list, kvm->arch.model.fac_mask,
1451 S390_ARCH_FAC_LIST_SIZE_BYTE);
1453 set_kvm_facility(kvm->arch.model.fac_mask, 74);
1454 set_kvm_facility(kvm->arch.model.fac_list, 74);
1456 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
1457 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
1459 kvm_s390_crypto_init(kvm);
1461 spin_lock_init(&kvm->arch.float_int.lock);
1462 for (i = 0; i < FIRQ_LIST_COUNT; i++)
1463 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
1464 init_waitqueue_head(&kvm->arch.ipte_wq);
1465 mutex_init(&kvm->arch.ipte_mutex);
1467 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
1468 VM_EVENT(kvm, 3, "vm created with type %lu", type);
1470 if (type & KVM_VM_S390_UCONTROL) {
1471 kvm->arch.gmap = NULL;
1472 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
1474 if (sclp.hamax == U64_MAX)
1475 kvm->arch.mem_limit = TASK_MAX_SIZE;
1477 kvm->arch.mem_limit = min_t(unsigned long, TASK_MAX_SIZE,
1479 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
1480 if (!kvm->arch.gmap)
1482 kvm->arch.gmap->private = kvm;
1483 kvm->arch.gmap->pfault_enabled = 0;
1486 kvm->arch.css_support = 0;
1487 kvm->arch.use_irqchip = 0;
1488 kvm->arch.epoch = 0;
1490 spin_lock_init(&kvm->arch.start_stop_lock);
1491 kvm_s390_vsie_init(kvm);
1492 KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
1496 free_page((unsigned long)kvm->arch.sie_page2);
1497 debug_unregister(kvm->arch.dbf);
1499 KVM_EVENT(3, "creation of vm failed: %d", rc);
1503 bool kvm_arch_has_vcpu_debugfs(void)
1508 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
1513 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1515 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
1516 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
1517 kvm_s390_clear_local_irqs(vcpu);
1518 kvm_clear_async_pf_completion_queue(vcpu);
1519 if (!kvm_is_ucontrol(vcpu->kvm))
1522 if (kvm_is_ucontrol(vcpu->kvm))
1523 gmap_remove(vcpu->arch.gmap);
1525 if (vcpu->kvm->arch.use_cmma)
1526 kvm_s390_vcpu_unsetup_cmma(vcpu);
1527 free_page((unsigned long)(vcpu->arch.sie_block));
1529 kvm_vcpu_uninit(vcpu);
1530 kmem_cache_free(kvm_vcpu_cache, vcpu);
1533 static void kvm_free_vcpus(struct kvm *kvm)
1536 struct kvm_vcpu *vcpu;
1538 kvm_for_each_vcpu(i, vcpu, kvm)
1539 kvm_arch_vcpu_destroy(vcpu);
1541 mutex_lock(&kvm->lock);
1542 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
1543 kvm->vcpus[i] = NULL;
1545 atomic_set(&kvm->online_vcpus, 0);
1546 mutex_unlock(&kvm->lock);
1549 void kvm_arch_destroy_vm(struct kvm *kvm)
1551 kvm_free_vcpus(kvm);
1553 debug_unregister(kvm->arch.dbf);
1554 free_page((unsigned long)kvm->arch.sie_page2);
1555 if (!kvm_is_ucontrol(kvm))
1556 gmap_remove(kvm->arch.gmap);
1557 kvm_s390_destroy_adapters(kvm);
1558 kvm_s390_clear_float_irqs(kvm);
1559 kvm_s390_vsie_destroy(kvm);
1560 KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
1563 /* Section: vcpu related */
1564 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
1566 vcpu->arch.gmap = gmap_create(current->mm, -1UL);
1567 if (!vcpu->arch.gmap)
1569 vcpu->arch.gmap->private = vcpu->kvm;
1574 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
1576 if (!kvm_s390_use_sca_entries())
1578 read_lock(&vcpu->kvm->arch.sca_lock);
1579 if (vcpu->kvm->arch.use_esca) {
1580 struct esca_block *sca = vcpu->kvm->arch.sca;
1582 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
1583 sca->cpu[vcpu->vcpu_id].sda = 0;
1585 struct bsca_block *sca = vcpu->kvm->arch.sca;
1587 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
1588 sca->cpu[vcpu->vcpu_id].sda = 0;
1590 read_unlock(&vcpu->kvm->arch.sca_lock);
1593 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
1595 if (!kvm_s390_use_sca_entries()) {
1596 struct bsca_block *sca = vcpu->kvm->arch.sca;
1598 /* we still need the basic sca for the ipte control */
1599 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1600 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
1602 read_lock(&vcpu->kvm->arch.sca_lock);
1603 if (vcpu->kvm->arch.use_esca) {
1604 struct esca_block *sca = vcpu->kvm->arch.sca;
1606 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
1607 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1608 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca & ~0x3fU;
1609 vcpu->arch.sie_block->ecb2 |= 0x04U;
1610 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
1612 struct bsca_block *sca = vcpu->kvm->arch.sca;
1614 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
1615 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1616 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
1617 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
1619 read_unlock(&vcpu->kvm->arch.sca_lock);
1622 /* Basic SCA to Extended SCA data copy routines */
1623 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
1626 d->sigp_ctrl.c = s->sigp_ctrl.c;
1627 d->sigp_ctrl.scn = s->sigp_ctrl.scn;
1630 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
1634 d->ipte_control = s->ipte_control;
1636 for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
1637 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
1640 static int sca_switch_to_extended(struct kvm *kvm)
1642 struct bsca_block *old_sca = kvm->arch.sca;
1643 struct esca_block *new_sca;
1644 struct kvm_vcpu *vcpu;
1645 unsigned int vcpu_idx;
1648 new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL|__GFP_ZERO);
1652 scaoh = (u32)((u64)(new_sca) >> 32);
1653 scaol = (u32)(u64)(new_sca) & ~0x3fU;
1655 kvm_s390_vcpu_block_all(kvm);
1656 write_lock(&kvm->arch.sca_lock);
1658 sca_copy_b_to_e(new_sca, old_sca);
1660 kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
1661 vcpu->arch.sie_block->scaoh = scaoh;
1662 vcpu->arch.sie_block->scaol = scaol;
1663 vcpu->arch.sie_block->ecb2 |= 0x04U;
1665 kvm->arch.sca = new_sca;
1666 kvm->arch.use_esca = 1;
1668 write_unlock(&kvm->arch.sca_lock);
1669 kvm_s390_vcpu_unblock_all(kvm);
1671 free_page((unsigned long)old_sca);
1673 VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
1674 old_sca, kvm->arch.sca);
1678 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
1682 if (!kvm_s390_use_sca_entries()) {
1683 if (id < KVM_MAX_VCPUS)
1687 if (id < KVM_S390_BSCA_CPU_SLOTS)
1689 if (!sclp.has_esca || !sclp.has_64bscao)
1692 mutex_lock(&kvm->lock);
1693 rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
1694 mutex_unlock(&kvm->lock);
1696 return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
1699 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1701 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1702 kvm_clear_async_pf_completion_queue(vcpu);
1703 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
1709 kvm_s390_set_prefix(vcpu, 0);
1710 if (test_kvm_facility(vcpu->kvm, 64))
1711 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
1712 /* fprs can be synchronized via vrs, even if the guest has no vx. With
1713 * MACHINE_HAS_VX, (load|store)_fpu_regs() will work with vrs format.
1716 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
1718 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
1720 if (kvm_is_ucontrol(vcpu->kvm))
1721 return __kvm_ucontrol_vcpu_init(vcpu);
1726 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1727 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1729 WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
1730 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1731 vcpu->arch.cputm_start = get_tod_clock_fast();
1732 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1735 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1736 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1738 WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
1739 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1740 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
1741 vcpu->arch.cputm_start = 0;
1742 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1745 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1746 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1748 WARN_ON_ONCE(vcpu->arch.cputm_enabled);
1749 vcpu->arch.cputm_enabled = true;
1750 __start_cpu_timer_accounting(vcpu);
1753 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1754 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1756 WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
1757 __stop_cpu_timer_accounting(vcpu);
1758 vcpu->arch.cputm_enabled = false;
1761 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1763 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1764 __enable_cpu_timer_accounting(vcpu);
1768 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1770 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1771 __disable_cpu_timer_accounting(vcpu);
1775 /* set the cpu timer - may only be called from the VCPU thread itself */
1776 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
1778 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1779 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1780 if (vcpu->arch.cputm_enabled)
1781 vcpu->arch.cputm_start = get_tod_clock_fast();
1782 vcpu->arch.sie_block->cputm = cputm;
1783 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1787 /* update and get the cpu timer - can also be called from other VCPU threads */
1788 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
1793 if (unlikely(!vcpu->arch.cputm_enabled))
1794 return vcpu->arch.sie_block->cputm;
1796 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1798 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
1800 * If the writer would ever execute a read in the critical
1801 * section, e.g. in irq context, we have a deadlock.
1803 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
1804 value = vcpu->arch.sie_block->cputm;
1805 /* if cputm_start is 0, accounting is being started/stopped */
1806 if (likely(vcpu->arch.cputm_start))
1807 value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
1808 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
1813 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1815 /* Save host register state */
1817 vcpu->arch.host_fpregs.fpc = current->thread.fpu.fpc;
1818 vcpu->arch.host_fpregs.regs = current->thread.fpu.regs;
1821 current->thread.fpu.regs = vcpu->run->s.regs.vrs;
1823 current->thread.fpu.regs = vcpu->run->s.regs.fprs;
1824 current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
1825 if (test_fp_ctl(current->thread.fpu.fpc))
1826 /* User space provided an invalid FPC, let's clear it */
1827 current->thread.fpu.fpc = 0;
1829 save_access_regs(vcpu->arch.host_acrs);
1830 restore_access_regs(vcpu->run->s.regs.acrs);
1831 gmap_enable(vcpu->arch.enabled_gmap);
1832 atomic_or(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1833 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
1834 __start_cpu_timer_accounting(vcpu);
1838 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1841 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
1842 __stop_cpu_timer_accounting(vcpu);
1843 atomic_andnot(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1844 vcpu->arch.enabled_gmap = gmap_get_enabled();
1845 gmap_disable(vcpu->arch.enabled_gmap);
1847 /* Save guest register state */
1849 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
1851 /* Restore host register state */
1852 current->thread.fpu.fpc = vcpu->arch.host_fpregs.fpc;
1853 current->thread.fpu.regs = vcpu->arch.host_fpregs.regs;
1855 save_access_regs(vcpu->run->s.regs.acrs);
1856 restore_access_regs(vcpu->arch.host_acrs);
1859 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
1861 /* this equals initial cpu reset in pop, but we don't switch to ESA */
1862 vcpu->arch.sie_block->gpsw.mask = 0UL;
1863 vcpu->arch.sie_block->gpsw.addr = 0UL;
1864 kvm_s390_set_prefix(vcpu, 0);
1865 kvm_s390_set_cpu_timer(vcpu, 0);
1866 vcpu->arch.sie_block->ckc = 0UL;
1867 vcpu->arch.sie_block->todpr = 0;
1868 memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64));
1869 vcpu->arch.sie_block->gcr[0] = 0xE0UL;
1870 vcpu->arch.sie_block->gcr[14] = 0xC2000000UL;
1871 /* make sure the new fpc will be lazily loaded */
1873 current->thread.fpu.fpc = 0;
1874 vcpu->arch.sie_block->gbea = 1;
1875 vcpu->arch.sie_block->pp = 0;
1876 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1877 kvm_clear_async_pf_completion_queue(vcpu);
1878 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
1879 kvm_s390_vcpu_stop(vcpu);
1880 kvm_s390_clear_local_irqs(vcpu);
1883 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1885 mutex_lock(&vcpu->kvm->lock);
1887 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
1889 mutex_unlock(&vcpu->kvm->lock);
1890 if (!kvm_is_ucontrol(vcpu->kvm)) {
1891 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
1894 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
1895 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
1896 /* make vcpu_load load the right gmap on the first trigger */
1897 vcpu->arch.enabled_gmap = vcpu->arch.gmap;
1900 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
1902 if (!test_kvm_facility(vcpu->kvm, 76))
1905 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
1907 if (vcpu->kvm->arch.crypto.aes_kw)
1908 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
1909 if (vcpu->kvm->arch.crypto.dea_kw)
1910 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
1912 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
1915 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
1917 free_page(vcpu->arch.sie_block->cbrlo);
1918 vcpu->arch.sie_block->cbrlo = 0;
1921 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
1923 vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
1924 if (!vcpu->arch.sie_block->cbrlo)
1927 vcpu->arch.sie_block->ecb2 |= 0x80;
1928 vcpu->arch.sie_block->ecb2 &= ~0x08;
1932 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
1934 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
1936 vcpu->arch.sie_block->ibc = model->ibc;
1937 if (test_kvm_facility(vcpu->kvm, 7))
1938 vcpu->arch.sie_block->fac = (u32)(u64) model->fac_list;
1941 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1945 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
1949 if (test_kvm_facility(vcpu->kvm, 78))
1950 atomic_or(CPUSTAT_GED2, &vcpu->arch.sie_block->cpuflags);
1951 else if (test_kvm_facility(vcpu->kvm, 8))
1952 atomic_or(CPUSTAT_GED, &vcpu->arch.sie_block->cpuflags);
1954 kvm_s390_vcpu_setup_model(vcpu);
1956 /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
1957 if (MACHINE_HAS_ESOP)
1958 vcpu->arch.sie_block->ecb |= 0x02;
1959 if (test_kvm_facility(vcpu->kvm, 9))
1960 vcpu->arch.sie_block->ecb |= 0x04;
1961 if (test_kvm_facility(vcpu->kvm, 73))
1962 vcpu->arch.sie_block->ecb |= 0x10;
1964 if (test_kvm_facility(vcpu->kvm, 8) && sclp.has_pfmfi)
1965 vcpu->arch.sie_block->ecb2 |= 0x08;
1966 vcpu->arch.sie_block->eca = 0x1002000U;
1968 vcpu->arch.sie_block->eca |= 0x80000000U;
1970 vcpu->arch.sie_block->eca |= 0x40000000U;
1972 vcpu->arch.sie_block->eca |= 1;
1973 if (sclp.has_sigpif)
1974 vcpu->arch.sie_block->eca |= 0x10000000U;
1975 if (test_kvm_facility(vcpu->kvm, 129)) {
1976 vcpu->arch.sie_block->eca |= 0x00020000;
1977 vcpu->arch.sie_block->ecd |= 0x20000000;
1979 vcpu->arch.sie_block->riccbd = (unsigned long) &vcpu->run->s.regs.riccb;
1980 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
1982 if (vcpu->kvm->arch.use_cmma) {
1983 rc = kvm_s390_vcpu_setup_cmma(vcpu);
1987 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1988 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
1990 kvm_s390_vcpu_crypto_setup(vcpu);
1995 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1998 struct kvm_vcpu *vcpu;
1999 struct sie_page *sie_page;
2002 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
2007 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2011 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
2015 vcpu->arch.sie_block = &sie_page->sie_block;
2016 vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
2018 /* the real guest size will always be smaller than msl */
2019 vcpu->arch.sie_block->mso = 0;
2020 vcpu->arch.sie_block->msl = sclp.hamax;
2022 vcpu->arch.sie_block->icpua = id;
2023 spin_lock_init(&vcpu->arch.local_int.lock);
2024 vcpu->arch.local_int.float_int = &kvm->arch.float_int;
2025 vcpu->arch.local_int.wq = &vcpu->wq;
2026 vcpu->arch.local_int.cpuflags = &vcpu->arch.sie_block->cpuflags;
2027 seqcount_init(&vcpu->arch.cputm_seqcount);
2029 rc = kvm_vcpu_init(vcpu, kvm, id);
2031 goto out_free_sie_block;
2032 VM_EVENT(kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK", id, vcpu,
2033 vcpu->arch.sie_block);
2034 trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
2038 free_page((unsigned long)(vcpu->arch.sie_block));
2040 kmem_cache_free(kvm_vcpu_cache, vcpu);
2045 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
2047 return kvm_s390_vcpu_has_irq(vcpu, 0);
2050 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
2052 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2056 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
2058 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2061 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
2063 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2067 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
2069 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2073 * Kick a guest cpu out of SIE and wait until SIE is not running.
2074 * If the CPU is not running (e.g. waiting as idle) the function will
2075 * return immediately. */
2076 void exit_sie(struct kvm_vcpu *vcpu)
2078 atomic_or(CPUSTAT_STOP_INT, &vcpu->arch.sie_block->cpuflags);
2079 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
2083 /* Kick a guest cpu out of SIE to process a request synchronously */
2084 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
2086 kvm_make_request(req, vcpu);
2087 kvm_s390_vcpu_request(vcpu);
2090 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
2093 struct kvm *kvm = gmap->private;
2094 struct kvm_vcpu *vcpu;
2095 unsigned long prefix;
2098 if (gmap_is_shadow(gmap))
2100 if (start >= 1UL << 31)
2101 /* We are only interested in prefix pages */
2103 kvm_for_each_vcpu(i, vcpu, kvm) {
2104 /* match against both prefix pages */
2105 prefix = kvm_s390_get_prefix(vcpu);
2106 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
2107 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
2109 kvm_s390_sync_request(KVM_REQ_MMU_RELOAD, vcpu);
2114 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
2116 /* kvm common code refers to this, but never calls it */
2121 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
2122 struct kvm_one_reg *reg)
2127 case KVM_REG_S390_TODPR:
2128 r = put_user(vcpu->arch.sie_block->todpr,
2129 (u32 __user *)reg->addr);
2131 case KVM_REG_S390_EPOCHDIFF:
2132 r = put_user(vcpu->arch.sie_block->epoch,
2133 (u64 __user *)reg->addr);
2135 case KVM_REG_S390_CPU_TIMER:
2136 r = put_user(kvm_s390_get_cpu_timer(vcpu),
2137 (u64 __user *)reg->addr);
2139 case KVM_REG_S390_CLOCK_COMP:
2140 r = put_user(vcpu->arch.sie_block->ckc,
2141 (u64 __user *)reg->addr);
2143 case KVM_REG_S390_PFTOKEN:
2144 r = put_user(vcpu->arch.pfault_token,
2145 (u64 __user *)reg->addr);
2147 case KVM_REG_S390_PFCOMPARE:
2148 r = put_user(vcpu->arch.pfault_compare,
2149 (u64 __user *)reg->addr);
2151 case KVM_REG_S390_PFSELECT:
2152 r = put_user(vcpu->arch.pfault_select,
2153 (u64 __user *)reg->addr);
2155 case KVM_REG_S390_PP:
2156 r = put_user(vcpu->arch.sie_block->pp,
2157 (u64 __user *)reg->addr);
2159 case KVM_REG_S390_GBEA:
2160 r = put_user(vcpu->arch.sie_block->gbea,
2161 (u64 __user *)reg->addr);
2170 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
2171 struct kvm_one_reg *reg)
2177 case KVM_REG_S390_TODPR:
2178 r = get_user(vcpu->arch.sie_block->todpr,
2179 (u32 __user *)reg->addr);
2181 case KVM_REG_S390_EPOCHDIFF:
2182 r = get_user(vcpu->arch.sie_block->epoch,
2183 (u64 __user *)reg->addr);
2185 case KVM_REG_S390_CPU_TIMER:
2186 r = get_user(val, (u64 __user *)reg->addr);
2188 kvm_s390_set_cpu_timer(vcpu, val);
2190 case KVM_REG_S390_CLOCK_COMP:
2191 r = get_user(vcpu->arch.sie_block->ckc,
2192 (u64 __user *)reg->addr);
2194 case KVM_REG_S390_PFTOKEN:
2195 r = get_user(vcpu->arch.pfault_token,
2196 (u64 __user *)reg->addr);
2197 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2198 kvm_clear_async_pf_completion_queue(vcpu);
2200 case KVM_REG_S390_PFCOMPARE:
2201 r = get_user(vcpu->arch.pfault_compare,
2202 (u64 __user *)reg->addr);
2204 case KVM_REG_S390_PFSELECT:
2205 r = get_user(vcpu->arch.pfault_select,
2206 (u64 __user *)reg->addr);
2208 case KVM_REG_S390_PP:
2209 r = get_user(vcpu->arch.sie_block->pp,
2210 (u64 __user *)reg->addr);
2212 case KVM_REG_S390_GBEA:
2213 r = get_user(vcpu->arch.sie_block->gbea,
2214 (u64 __user *)reg->addr);
2223 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
2225 kvm_s390_vcpu_initial_reset(vcpu);
2229 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2231 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs));
2235 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2237 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
2241 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2242 struct kvm_sregs *sregs)
2244 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
2245 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
2246 restore_access_regs(vcpu->run->s.regs.acrs);
2250 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2251 struct kvm_sregs *sregs)
2253 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
2254 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
2258 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2260 /* make sure the new values will be lazily loaded */
2262 if (test_fp_ctl(fpu->fpc))
2264 current->thread.fpu.fpc = fpu->fpc;
2266 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs,
2267 (freg_t *) fpu->fprs);
2269 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs));
2273 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2275 /* make sure we have the latest values */
2278 convert_vx_to_fp((freg_t *) fpu->fprs,
2279 (__vector128 *) vcpu->run->s.regs.vrs);
2281 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs));
2282 fpu->fpc = current->thread.fpu.fpc;
2286 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
2290 if (!is_vcpu_stopped(vcpu))
2293 vcpu->run->psw_mask = psw.mask;
2294 vcpu->run->psw_addr = psw.addr;
2299 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2300 struct kvm_translation *tr)
2302 return -EINVAL; /* not implemented yet */
2305 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
2306 KVM_GUESTDBG_USE_HW_BP | \
2307 KVM_GUESTDBG_ENABLE)
2309 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
2310 struct kvm_guest_debug *dbg)
2314 vcpu->guest_debug = 0;
2315 kvm_s390_clear_bp_data(vcpu);
2317 if (dbg->control & ~VALID_GUESTDBG_FLAGS)
2319 if (!sclp.has_gpere)
2322 if (dbg->control & KVM_GUESTDBG_ENABLE) {
2323 vcpu->guest_debug = dbg->control;
2324 /* enforce guest PER */
2325 atomic_or(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2327 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
2328 rc = kvm_s390_import_bp_data(vcpu, dbg);
2330 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2331 vcpu->arch.guestdbg.last_bp = 0;
2335 vcpu->guest_debug = 0;
2336 kvm_s390_clear_bp_data(vcpu);
2337 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2343 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2344 struct kvm_mp_state *mp_state)
2346 /* CHECK_STOP and LOAD are not supported yet */
2347 return is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
2348 KVM_MP_STATE_OPERATING;
2351 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2352 struct kvm_mp_state *mp_state)
2356 /* user space knows about this interface - let it control the state */
2357 vcpu->kvm->arch.user_cpu_state_ctrl = 1;
2359 switch (mp_state->mp_state) {
2360 case KVM_MP_STATE_STOPPED:
2361 kvm_s390_vcpu_stop(vcpu);
2363 case KVM_MP_STATE_OPERATING:
2364 kvm_s390_vcpu_start(vcpu);
2366 case KVM_MP_STATE_LOAD:
2367 case KVM_MP_STATE_CHECK_STOP:
2368 /* fall through - CHECK_STOP and LOAD are not supported yet */
2376 static bool ibs_enabled(struct kvm_vcpu *vcpu)
2378 return atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_IBS;
2381 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
2384 kvm_s390_vcpu_request_handled(vcpu);
2385 if (!vcpu->requests)
2388 * We use MMU_RELOAD just to re-arm the ipte notifier for the
2389 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
2390 * This ensures that the ipte instruction for this request has
2391 * already finished. We might race against a second unmapper that
2392 * wants to set the blocking bit. Lets just retry the request loop.
2394 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
2396 rc = gmap_mprotect_notify(vcpu->arch.gmap,
2397 kvm_s390_get_prefix(vcpu),
2398 PAGE_SIZE * 2, PROT_WRITE);
2400 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
2406 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2407 vcpu->arch.sie_block->ihcpu = 0xffff;
2411 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
2412 if (!ibs_enabled(vcpu)) {
2413 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
2414 atomic_or(CPUSTAT_IBS,
2415 &vcpu->arch.sie_block->cpuflags);
2420 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
2421 if (ibs_enabled(vcpu)) {
2422 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
2423 atomic_andnot(CPUSTAT_IBS,
2424 &vcpu->arch.sie_block->cpuflags);
2429 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
2430 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
2434 /* nothing to do, just clear the request */
2435 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
2440 void kvm_s390_set_tod_clock(struct kvm *kvm, u64 tod)
2442 struct kvm_vcpu *vcpu;
2445 mutex_lock(&kvm->lock);
2447 kvm->arch.epoch = tod - get_tod_clock();
2448 kvm_s390_vcpu_block_all(kvm);
2449 kvm_for_each_vcpu(i, vcpu, kvm)
2450 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
2451 kvm_s390_vcpu_unblock_all(kvm);
2453 mutex_unlock(&kvm->lock);
2457 * kvm_arch_fault_in_page - fault-in guest page if necessary
2458 * @vcpu: The corresponding virtual cpu
2459 * @gpa: Guest physical address
2460 * @writable: Whether the page should be writable or not
2462 * Make sure that a guest page has been faulted-in on the host.
2464 * Return: Zero on success, negative error code otherwise.
2466 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
2468 return gmap_fault(vcpu->arch.gmap, gpa,
2469 writable ? FAULT_FLAG_WRITE : 0);
2472 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
2473 unsigned long token)
2475 struct kvm_s390_interrupt inti;
2476 struct kvm_s390_irq irq;
2479 irq.u.ext.ext_params2 = token;
2480 irq.type = KVM_S390_INT_PFAULT_INIT;
2481 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
2483 inti.type = KVM_S390_INT_PFAULT_DONE;
2484 inti.parm64 = token;
2485 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
2489 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
2490 struct kvm_async_pf *work)
2492 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
2493 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
2496 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
2497 struct kvm_async_pf *work)
2499 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
2500 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
2503 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
2504 struct kvm_async_pf *work)
2506 /* s390 will always inject the page directly */
2509 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
2512 * s390 will always inject the page directly,
2513 * but we still want check_async_completion to cleanup
2518 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
2521 struct kvm_arch_async_pf arch;
2524 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2526 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
2527 vcpu->arch.pfault_compare)
2529 if (psw_extint_disabled(vcpu))
2531 if (kvm_s390_vcpu_has_irq(vcpu, 0))
2533 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
2535 if (!vcpu->arch.gmap->pfault_enabled)
2538 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
2539 hva += current->thread.gmap_addr & ~PAGE_MASK;
2540 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
2543 rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
2547 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
2552 * On s390 notifications for arriving pages will be delivered directly
2553 * to the guest but the house keeping for completed pfaults is
2554 * handled outside the worker.
2556 kvm_check_async_pf_completion(vcpu);
2558 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
2559 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
2564 if (test_cpu_flag(CIF_MCCK_PENDING))
2567 if (!kvm_is_ucontrol(vcpu->kvm)) {
2568 rc = kvm_s390_deliver_pending_interrupts(vcpu);
2573 rc = kvm_s390_handle_requests(vcpu);
2577 if (guestdbg_enabled(vcpu)) {
2578 kvm_s390_backup_guest_per_regs(vcpu);
2579 kvm_s390_patch_guest_per_regs(vcpu);
2582 vcpu->arch.sie_block->icptcode = 0;
2583 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
2584 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
2585 trace_kvm_s390_sie_enter(vcpu, cpuflags);
2590 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
2592 struct kvm_s390_pgm_info pgm_info = {
2593 .code = PGM_ADDRESSING,
2598 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
2599 trace_kvm_s390_sie_fault(vcpu);
2602 * We want to inject an addressing exception, which is defined as a
2603 * suppressing or terminating exception. However, since we came here
2604 * by a DAT access exception, the PSW still points to the faulting
2605 * instruction since DAT exceptions are nullifying. So we've got
2606 * to look up the current opcode to get the length of the instruction
2607 * to be able to forward the PSW.
2609 rc = read_guest_instr(vcpu, &opcode, 1);
2610 ilen = insn_length(opcode);
2614 /* Instruction-Fetching Exceptions - we can't detect the ilen.
2615 * Forward by arbitrary ilc, injection will take care of
2616 * nullification if necessary.
2618 pgm_info = vcpu->arch.pgm;
2621 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
2622 kvm_s390_forward_psw(vcpu, ilen);
2623 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
2626 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
2628 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
2629 vcpu->arch.sie_block->icptcode);
2630 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
2632 if (guestdbg_enabled(vcpu))
2633 kvm_s390_restore_guest_per_regs(vcpu);
2635 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
2636 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
2638 if (vcpu->arch.sie_block->icptcode > 0) {
2639 int rc = kvm_handle_sie_intercept(vcpu);
2641 if (rc != -EOPNOTSUPP)
2643 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
2644 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
2645 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
2646 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
2648 } else if (exit_reason != -EFAULT) {
2649 vcpu->stat.exit_null++;
2651 } else if (kvm_is_ucontrol(vcpu->kvm)) {
2652 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
2653 vcpu->run->s390_ucontrol.trans_exc_code =
2654 current->thread.gmap_addr;
2655 vcpu->run->s390_ucontrol.pgm_code = 0x10;
2657 } else if (current->thread.gmap_pfault) {
2658 trace_kvm_s390_major_guest_pfault(vcpu);
2659 current->thread.gmap_pfault = 0;
2660 if (kvm_arch_setup_async_pf(vcpu))
2662 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
2664 return vcpu_post_run_fault_in_sie(vcpu);
2667 static int __vcpu_run(struct kvm_vcpu *vcpu)
2669 int rc, exit_reason;
2672 * We try to hold kvm->srcu during most of vcpu_run (except when run-
2673 * ning the guest), so that memslots (and other stuff) are protected
2675 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2678 rc = vcpu_pre_run(vcpu);
2682 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2684 * As PF_VCPU will be used in fault handler, between
2685 * guest_enter and guest_exit should be no uaccess.
2687 local_irq_disable();
2688 guest_enter_irqoff();
2689 __disable_cpu_timer_accounting(vcpu);
2691 exit_reason = sie64a(vcpu->arch.sie_block,
2692 vcpu->run->s.regs.gprs);
2693 local_irq_disable();
2694 __enable_cpu_timer_accounting(vcpu);
2695 guest_exit_irqoff();
2697 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2699 rc = vcpu_post_run(vcpu, exit_reason);
2700 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
2702 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2706 static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2708 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
2709 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
2710 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
2711 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
2712 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
2713 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
2714 /* some control register changes require a tlb flush */
2715 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2717 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
2718 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
2719 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
2720 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
2721 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
2722 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
2724 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
2725 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
2726 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
2727 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
2728 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2729 kvm_clear_async_pf_completion_queue(vcpu);
2732 * If userspace sets the riccb (e.g. after migration) to a valid state,
2733 * we should enable RI here instead of doing the lazy enablement.
2735 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) &&
2736 test_kvm_facility(vcpu->kvm, 64)) {
2737 struct runtime_instr_cb *riccb =
2738 (struct runtime_instr_cb *) &kvm_run->s.regs.riccb;
2741 vcpu->arch.sie_block->ecb3 |= 0x01;
2744 kvm_run->kvm_dirty_regs = 0;
2747 static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2749 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
2750 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
2751 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
2752 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
2753 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
2754 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
2755 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
2756 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
2757 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
2758 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
2759 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
2760 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
2763 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2768 if (guestdbg_exit_pending(vcpu)) {
2769 kvm_s390_prepare_debug_exit(vcpu);
2773 if (vcpu->sigset_active)
2774 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2776 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
2777 kvm_s390_vcpu_start(vcpu);
2778 } else if (is_vcpu_stopped(vcpu)) {
2779 pr_err_ratelimited("can't run stopped vcpu %d\n",
2784 sync_regs(vcpu, kvm_run);
2785 enable_cpu_timer_accounting(vcpu);
2788 rc = __vcpu_run(vcpu);
2790 if (signal_pending(current) && !rc) {
2791 kvm_run->exit_reason = KVM_EXIT_INTR;
2795 if (guestdbg_exit_pending(vcpu) && !rc) {
2796 kvm_s390_prepare_debug_exit(vcpu);
2800 if (rc == -EREMOTE) {
2801 /* userspace support is needed, kvm_run has been prepared */
2805 disable_cpu_timer_accounting(vcpu);
2806 store_regs(vcpu, kvm_run);
2808 if (vcpu->sigset_active)
2809 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2811 vcpu->stat.exit_userspace++;
2816 * store status at address
2817 * we use have two special cases:
2818 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
2819 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
2821 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
2823 unsigned char archmode = 1;
2824 freg_t fprs[NUM_FPRS];
2829 px = kvm_s390_get_prefix(vcpu);
2830 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
2831 if (write_guest_abs(vcpu, 163, &archmode, 1))
2834 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
2835 if (write_guest_real(vcpu, 163, &archmode, 1))
2839 gpa -= __LC_FPREGS_SAVE_AREA;
2841 /* manually convert vector registers if necessary */
2842 if (MACHINE_HAS_VX) {
2843 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
2844 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2847 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2848 vcpu->run->s.regs.fprs, 128);
2850 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
2851 vcpu->run->s.regs.gprs, 128);
2852 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
2853 &vcpu->arch.sie_block->gpsw, 16);
2854 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
2856 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
2857 &vcpu->run->s.regs.fpc, 4);
2858 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
2859 &vcpu->arch.sie_block->todpr, 4);
2860 cputm = kvm_s390_get_cpu_timer(vcpu);
2861 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
2863 clkcomp = vcpu->arch.sie_block->ckc >> 8;
2864 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
2866 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
2867 &vcpu->run->s.regs.acrs, 64);
2868 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
2869 &vcpu->arch.sie_block->gcr, 128);
2870 return rc ? -EFAULT : 0;
2873 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
2876 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
2877 * copying in vcpu load/put. Lets update our copies before we save
2878 * it into the save area
2881 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
2882 save_access_regs(vcpu->run->s.regs.acrs);
2884 return kvm_s390_store_status_unloaded(vcpu, addr);
2887 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2889 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
2890 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
2893 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
2896 struct kvm_vcpu *vcpu;
2898 kvm_for_each_vcpu(i, vcpu, kvm) {
2899 __disable_ibs_on_vcpu(vcpu);
2903 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2907 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
2908 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
2911 void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
2913 int i, online_vcpus, started_vcpus = 0;
2915 if (!is_vcpu_stopped(vcpu))
2918 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
2919 /* Only one cpu at a time may enter/leave the STOPPED state. */
2920 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2921 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2923 for (i = 0; i < online_vcpus; i++) {
2924 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i]))
2928 if (started_vcpus == 0) {
2929 /* we're the only active VCPU -> speed it up */
2930 __enable_ibs_on_vcpu(vcpu);
2931 } else if (started_vcpus == 1) {
2933 * As we are starting a second VCPU, we have to disable
2934 * the IBS facility on all VCPUs to remove potentially
2935 * oustanding ENABLE requests.
2937 __disable_ibs_on_all_vcpus(vcpu->kvm);
2940 atomic_andnot(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2942 * Another VCPU might have used IBS while we were offline.
2943 * Let's play safe and flush the VCPU at startup.
2945 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2946 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2950 void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
2952 int i, online_vcpus, started_vcpus = 0;
2953 struct kvm_vcpu *started_vcpu = NULL;
2955 if (is_vcpu_stopped(vcpu))
2958 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
2959 /* Only one cpu at a time may enter/leave the STOPPED state. */
2960 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2961 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2963 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
2964 kvm_s390_clear_stop_irq(vcpu);
2966 atomic_or(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2967 __disable_ibs_on_vcpu(vcpu);
2969 for (i = 0; i < online_vcpus; i++) {
2970 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) {
2972 started_vcpu = vcpu->kvm->vcpus[i];
2976 if (started_vcpus == 1) {
2978 * As we only have one VCPU left, we want to enable the
2979 * IBS facility for that VCPU to speed it up.
2981 __enable_ibs_on_vcpu(started_vcpu);
2984 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2988 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
2989 struct kvm_enable_cap *cap)
2997 case KVM_CAP_S390_CSS_SUPPORT:
2998 if (!vcpu->kvm->arch.css_support) {
2999 vcpu->kvm->arch.css_support = 1;
3000 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
3001 trace_kvm_s390_enable_css(vcpu->kvm);
3012 static long kvm_s390_guest_mem_op(struct kvm_vcpu *vcpu,
3013 struct kvm_s390_mem_op *mop)
3015 void __user *uaddr = (void __user *)mop->buf;
3016 void *tmpbuf = NULL;
3018 const u64 supported_flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION
3019 | KVM_S390_MEMOP_F_CHECK_ONLY;
3021 if (mop->flags & ~supported_flags)
3024 if (mop->size > MEM_OP_MAX_SIZE)
3027 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
3028 tmpbuf = vmalloc(mop->size);
3033 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3036 case KVM_S390_MEMOP_LOGICAL_READ:
3037 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3038 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3039 mop->size, GACC_FETCH);
3042 r = read_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3044 if (copy_to_user(uaddr, tmpbuf, mop->size))
3048 case KVM_S390_MEMOP_LOGICAL_WRITE:
3049 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3050 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3051 mop->size, GACC_STORE);
3054 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
3058 r = write_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3064 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
3066 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
3067 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
3073 long kvm_arch_vcpu_ioctl(struct file *filp,
3074 unsigned int ioctl, unsigned long arg)
3076 struct kvm_vcpu *vcpu = filp->private_data;
3077 void __user *argp = (void __user *)arg;
3082 case KVM_S390_IRQ: {
3083 struct kvm_s390_irq s390irq;
3086 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
3088 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
3091 case KVM_S390_INTERRUPT: {
3092 struct kvm_s390_interrupt s390int;
3093 struct kvm_s390_irq s390irq;
3096 if (copy_from_user(&s390int, argp, sizeof(s390int)))
3098 if (s390int_to_s390irq(&s390int, &s390irq))
3100 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
3103 case KVM_S390_STORE_STATUS:
3104 idx = srcu_read_lock(&vcpu->kvm->srcu);
3105 r = kvm_s390_vcpu_store_status(vcpu, arg);
3106 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3108 case KVM_S390_SET_INITIAL_PSW: {
3112 if (copy_from_user(&psw, argp, sizeof(psw)))
3114 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
3117 case KVM_S390_INITIAL_RESET:
3118 r = kvm_arch_vcpu_ioctl_initial_reset(vcpu);
3120 case KVM_SET_ONE_REG:
3121 case KVM_GET_ONE_REG: {
3122 struct kvm_one_reg reg;
3124 if (copy_from_user(®, argp, sizeof(reg)))
3126 if (ioctl == KVM_SET_ONE_REG)
3127 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®);
3129 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®);
3132 #ifdef CONFIG_KVM_S390_UCONTROL
3133 case KVM_S390_UCAS_MAP: {
3134 struct kvm_s390_ucas_mapping ucasmap;
3136 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3141 if (!kvm_is_ucontrol(vcpu->kvm)) {
3146 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
3147 ucasmap.vcpu_addr, ucasmap.length);
3150 case KVM_S390_UCAS_UNMAP: {
3151 struct kvm_s390_ucas_mapping ucasmap;
3153 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3158 if (!kvm_is_ucontrol(vcpu->kvm)) {
3163 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
3168 case KVM_S390_VCPU_FAULT: {
3169 r = gmap_fault(vcpu->arch.gmap, arg, 0);
3172 case KVM_ENABLE_CAP:
3174 struct kvm_enable_cap cap;
3176 if (copy_from_user(&cap, argp, sizeof(cap)))
3178 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3181 case KVM_S390_MEM_OP: {
3182 struct kvm_s390_mem_op mem_op;
3184 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
3185 r = kvm_s390_guest_mem_op(vcpu, &mem_op);
3190 case KVM_S390_SET_IRQ_STATE: {
3191 struct kvm_s390_irq_state irq_state;
3194 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3196 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
3197 irq_state.len == 0 ||
3198 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
3202 r = kvm_s390_set_irq_state(vcpu,
3203 (void __user *) irq_state.buf,
3207 case KVM_S390_GET_IRQ_STATE: {
3208 struct kvm_s390_irq_state irq_state;
3211 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3213 if (irq_state.len == 0) {
3217 r = kvm_s390_get_irq_state(vcpu,
3218 (__u8 __user *) irq_state.buf,
3228 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3230 #ifdef CONFIG_KVM_S390_UCONTROL
3231 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
3232 && (kvm_is_ucontrol(vcpu->kvm))) {
3233 vmf->page = virt_to_page(vcpu->arch.sie_block);
3234 get_page(vmf->page);
3238 return VM_FAULT_SIGBUS;
3241 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
3242 unsigned long npages)
3247 /* Section: memory related */
3248 int kvm_arch_prepare_memory_region(struct kvm *kvm,
3249 struct kvm_memory_slot *memslot,
3250 const struct kvm_userspace_memory_region *mem,
3251 enum kvm_mr_change change)
3253 /* A few sanity checks. We can have memory slots which have to be
3254 located/ended at a segment boundary (1MB). The memory in userland is
3255 ok to be fragmented into various different vmas. It is okay to mmap()
3256 and munmap() stuff in this slot after doing this call at any time */
3258 if (mem->userspace_addr & 0xffffful)
3261 if (mem->memory_size & 0xffffful)
3264 if (mem->guest_phys_addr + mem->memory_size > kvm->arch.mem_limit)
3270 void kvm_arch_commit_memory_region(struct kvm *kvm,
3271 const struct kvm_userspace_memory_region *mem,
3272 const struct kvm_memory_slot *old,
3273 const struct kvm_memory_slot *new,
3274 enum kvm_mr_change change)
3278 /* If the basics of the memslot do not change, we do not want
3279 * to update the gmap. Every update causes several unnecessary
3280 * segment translation exceptions. This is usually handled just
3281 * fine by the normal fault handler + gmap, but it will also
3282 * cause faults on the prefix page of running guest CPUs.
3284 if (old->userspace_addr == mem->userspace_addr &&
3285 old->base_gfn * PAGE_SIZE == mem->guest_phys_addr &&
3286 old->npages * PAGE_SIZE == mem->memory_size)
3289 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr,
3290 mem->guest_phys_addr, mem->memory_size);
3292 pr_warn("failed to commit memory region\n");
3296 static inline unsigned long nonhyp_mask(int i)
3298 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
3300 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
3303 void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu)
3305 vcpu->valid_wakeup = false;
3308 static int __init kvm_s390_init(void)
3312 if (!sclp.has_sief2) {
3313 pr_info("SIE not available\n");
3317 for (i = 0; i < 16; i++)
3318 kvm_s390_fac_list_mask[i] |=
3319 S390_lowcore.stfle_fac_list[i] & nonhyp_mask(i);
3321 return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
3324 static void __exit kvm_s390_exit(void)
3329 module_init(kvm_s390_init);
3330 module_exit(kvm_s390_exit);
3333 * Enable autoloading of the kvm module.
3334 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
3335 * since x86 takes a different approach.
3337 #include <linux/miscdevice.h>
3338 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3339 MODULE_ALIAS("devname:kvm");