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/moduleparam.h>
27 #include <linux/random.h>
28 #include <linux/slab.h>
29 #include <linux/timer.h>
30 #include <linux/vmalloc.h>
31 #include <linux/bitmap.h>
32 #include <linux/sched/signal.h>
34 #include <asm/asm-offsets.h>
35 #include <asm/lowcore.h>
37 #include <asm/pgtable.h>
40 #include <asm/switch_to.h>
43 #include <asm/cpacf.h>
44 #include <asm/timex.h>
48 #define KMSG_COMPONENT "kvm-s390"
50 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
52 #define CREATE_TRACE_POINTS
54 #include "trace-s390.h"
56 #define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */
58 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
59 (KVM_MAX_VCPUS + LOCAL_IRQS))
61 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
63 struct kvm_stats_debugfs_item debugfs_entries[] = {
64 { "userspace_handled", VCPU_STAT(exit_userspace) },
65 { "exit_null", VCPU_STAT(exit_null) },
66 { "exit_validity", VCPU_STAT(exit_validity) },
67 { "exit_stop_request", VCPU_STAT(exit_stop_request) },
68 { "exit_external_request", VCPU_STAT(exit_external_request) },
69 { "exit_external_interrupt", VCPU_STAT(exit_external_interrupt) },
70 { "exit_instruction", VCPU_STAT(exit_instruction) },
71 { "exit_pei", VCPU_STAT(exit_pei) },
72 { "exit_program_interruption", VCPU_STAT(exit_program_interruption) },
73 { "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program) },
74 { "exit_operation_exception", VCPU_STAT(exit_operation_exception) },
75 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
76 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
77 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
78 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
79 { "instruction_lctlg", VCPU_STAT(instruction_lctlg) },
80 { "instruction_lctl", VCPU_STAT(instruction_lctl) },
81 { "instruction_stctl", VCPU_STAT(instruction_stctl) },
82 { "instruction_stctg", VCPU_STAT(instruction_stctg) },
83 { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal) },
84 { "deliver_external_call", VCPU_STAT(deliver_external_call) },
85 { "deliver_service_signal", VCPU_STAT(deliver_service_signal) },
86 { "deliver_virtio_interrupt", VCPU_STAT(deliver_virtio_interrupt) },
87 { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal) },
88 { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal) },
89 { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal) },
90 { "deliver_program_interruption", VCPU_STAT(deliver_program_int) },
91 { "exit_wait_state", VCPU_STAT(exit_wait_state) },
92 { "instruction_pfmf", VCPU_STAT(instruction_pfmf) },
93 { "instruction_stidp", VCPU_STAT(instruction_stidp) },
94 { "instruction_spx", VCPU_STAT(instruction_spx) },
95 { "instruction_stpx", VCPU_STAT(instruction_stpx) },
96 { "instruction_stap", VCPU_STAT(instruction_stap) },
97 { "instruction_storage_key", VCPU_STAT(instruction_storage_key) },
98 { "instruction_ipte_interlock", VCPU_STAT(instruction_ipte_interlock) },
99 { "instruction_stsch", VCPU_STAT(instruction_stsch) },
100 { "instruction_chsc", VCPU_STAT(instruction_chsc) },
101 { "instruction_essa", VCPU_STAT(instruction_essa) },
102 { "instruction_stsi", VCPU_STAT(instruction_stsi) },
103 { "instruction_stfl", VCPU_STAT(instruction_stfl) },
104 { "instruction_tprot", VCPU_STAT(instruction_tprot) },
105 { "instruction_sthyi", VCPU_STAT(instruction_sthyi) },
106 { "instruction_sie", VCPU_STAT(instruction_sie) },
107 { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense) },
108 { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running) },
109 { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call) },
110 { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency) },
111 { "instruction_sigp_cond_emergency", VCPU_STAT(instruction_sigp_cond_emergency) },
112 { "instruction_sigp_start", VCPU_STAT(instruction_sigp_start) },
113 { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop) },
114 { "instruction_sigp_stop_store_status", VCPU_STAT(instruction_sigp_stop_store_status) },
115 { "instruction_sigp_store_status", VCPU_STAT(instruction_sigp_store_status) },
116 { "instruction_sigp_store_adtl_status", VCPU_STAT(instruction_sigp_store_adtl_status) },
117 { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch) },
118 { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix) },
119 { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart) },
120 { "instruction_sigp_cpu_reset", VCPU_STAT(instruction_sigp_cpu_reset) },
121 { "instruction_sigp_init_cpu_reset", VCPU_STAT(instruction_sigp_init_cpu_reset) },
122 { "instruction_sigp_unknown", VCPU_STAT(instruction_sigp_unknown) },
123 { "diagnose_10", VCPU_STAT(diagnose_10) },
124 { "diagnose_44", VCPU_STAT(diagnose_44) },
125 { "diagnose_9c", VCPU_STAT(diagnose_9c) },
126 { "diagnose_258", VCPU_STAT(diagnose_258) },
127 { "diagnose_308", VCPU_STAT(diagnose_308) },
128 { "diagnose_500", VCPU_STAT(diagnose_500) },
132 /* allow nested virtualization in KVM (if enabled by user space) */
134 module_param(nested, int, S_IRUGO);
135 MODULE_PARM_DESC(nested, "Nested virtualization support");
137 /* upper facilities limit for kvm */
138 unsigned long kvm_s390_fac_list_mask[16] = { FACILITIES_KVM };
140 unsigned long kvm_s390_fac_list_mask_size(void)
142 BUILD_BUG_ON(ARRAY_SIZE(kvm_s390_fac_list_mask) > S390_ARCH_FAC_MASK_SIZE_U64);
143 return ARRAY_SIZE(kvm_s390_fac_list_mask);
146 /* available cpu features supported by kvm */
147 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
148 /* available subfunctions indicated via query / "test bit" */
149 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
151 static struct gmap_notifier gmap_notifier;
152 static struct gmap_notifier vsie_gmap_notifier;
153 debug_info_t *kvm_s390_dbf;
155 /* Section: not file related */
156 int kvm_arch_hardware_enable(void)
158 /* every s390 is virtualization enabled ;-) */
162 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
166 * This callback is executed during stop_machine(). All CPUs are therefore
167 * temporarily stopped. In order not to change guest behavior, we have to
168 * disable preemption whenever we touch the epoch of kvm and the VCPUs,
169 * so a CPU won't be stopped while calculating with the epoch.
171 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
175 struct kvm_vcpu *vcpu;
177 unsigned long long *delta = v;
179 list_for_each_entry(kvm, &vm_list, vm_list) {
180 kvm->arch.epoch -= *delta;
181 kvm_for_each_vcpu(i, vcpu, kvm) {
182 vcpu->arch.sie_block->epoch -= *delta;
183 if (vcpu->arch.cputm_enabled)
184 vcpu->arch.cputm_start += *delta;
185 if (vcpu->arch.vsie_block)
186 vcpu->arch.vsie_block->epoch -= *delta;
192 static struct notifier_block kvm_clock_notifier = {
193 .notifier_call = kvm_clock_sync,
196 int kvm_arch_hardware_setup(void)
198 gmap_notifier.notifier_call = kvm_gmap_notifier;
199 gmap_register_pte_notifier(&gmap_notifier);
200 vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
201 gmap_register_pte_notifier(&vsie_gmap_notifier);
202 atomic_notifier_chain_register(&s390_epoch_delta_notifier,
203 &kvm_clock_notifier);
207 void kvm_arch_hardware_unsetup(void)
209 gmap_unregister_pte_notifier(&gmap_notifier);
210 gmap_unregister_pte_notifier(&vsie_gmap_notifier);
211 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
212 &kvm_clock_notifier);
215 static void allow_cpu_feat(unsigned long nr)
217 set_bit_inv(nr, kvm_s390_available_cpu_feat);
220 static inline int plo_test_bit(unsigned char nr)
222 register unsigned long r0 asm("0") = (unsigned long) nr | 0x100;
226 /* Parameter registers are ignored for "test bit" */
236 static void kvm_s390_cpu_feat_init(void)
240 for (i = 0; i < 256; ++i) {
242 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
245 if (test_facility(28)) /* TOD-clock steering */
246 ptff(kvm_s390_available_subfunc.ptff,
247 sizeof(kvm_s390_available_subfunc.ptff),
250 if (test_facility(17)) { /* MSA */
251 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
252 kvm_s390_available_subfunc.kmac);
253 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
254 kvm_s390_available_subfunc.kmc);
255 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
256 kvm_s390_available_subfunc.km);
257 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
258 kvm_s390_available_subfunc.kimd);
259 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
260 kvm_s390_available_subfunc.klmd);
262 if (test_facility(76)) /* MSA3 */
263 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
264 kvm_s390_available_subfunc.pckmo);
265 if (test_facility(77)) { /* MSA4 */
266 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
267 kvm_s390_available_subfunc.kmctr);
268 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
269 kvm_s390_available_subfunc.kmf);
270 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
271 kvm_s390_available_subfunc.kmo);
272 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
273 kvm_s390_available_subfunc.pcc);
275 if (test_facility(57)) /* MSA5 */
276 __cpacf_query(CPACF_PRNO, (cpacf_mask_t *)
277 kvm_s390_available_subfunc.ppno);
279 if (test_facility(146)) /* MSA8 */
280 __cpacf_query(CPACF_KMA, (cpacf_mask_t *)
281 kvm_s390_available_subfunc.kma);
283 if (MACHINE_HAS_ESOP)
284 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
286 * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
287 * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
289 if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
290 !test_facility(3) || !nested)
292 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
293 if (sclp.has_64bscao)
294 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
296 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
298 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
300 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
302 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
304 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
306 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
308 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_KSS);
310 * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
311 * all skey handling functions read/set the skey from the PGSTE
312 * instead of the real storage key.
314 * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
315 * pages being detected as preserved although they are resident.
317 * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
318 * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
320 * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
321 * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
322 * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
324 * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
325 * cannot easily shadow the SCA because of the ipte lock.
329 int kvm_arch_init(void *opaque)
331 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
335 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view)) {
336 debug_unregister(kvm_s390_dbf);
340 kvm_s390_cpu_feat_init();
342 /* Register floating interrupt controller interface. */
343 return kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
346 void kvm_arch_exit(void)
348 debug_unregister(kvm_s390_dbf);
351 /* Section: device related */
352 long kvm_arch_dev_ioctl(struct file *filp,
353 unsigned int ioctl, unsigned long arg)
355 if (ioctl == KVM_S390_ENABLE_SIE)
356 return s390_enable_sie();
360 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
365 case KVM_CAP_S390_PSW:
366 case KVM_CAP_S390_GMAP:
367 case KVM_CAP_SYNC_MMU:
368 #ifdef CONFIG_KVM_S390_UCONTROL
369 case KVM_CAP_S390_UCONTROL:
371 case KVM_CAP_ASYNC_PF:
372 case KVM_CAP_SYNC_REGS:
373 case KVM_CAP_ONE_REG:
374 case KVM_CAP_ENABLE_CAP:
375 case KVM_CAP_S390_CSS_SUPPORT:
376 case KVM_CAP_IOEVENTFD:
377 case KVM_CAP_DEVICE_CTRL:
378 case KVM_CAP_ENABLE_CAP_VM:
379 case KVM_CAP_S390_IRQCHIP:
380 case KVM_CAP_VM_ATTRIBUTES:
381 case KVM_CAP_MP_STATE:
382 case KVM_CAP_IMMEDIATE_EXIT:
383 case KVM_CAP_S390_INJECT_IRQ:
384 case KVM_CAP_S390_USER_SIGP:
385 case KVM_CAP_S390_USER_STSI:
386 case KVM_CAP_S390_SKEYS:
387 case KVM_CAP_S390_IRQ_STATE:
388 case KVM_CAP_S390_USER_INSTR0:
389 case KVM_CAP_S390_AIS:
392 case KVM_CAP_S390_MEM_OP:
395 case KVM_CAP_NR_VCPUS:
396 case KVM_CAP_MAX_VCPUS:
397 r = KVM_S390_BSCA_CPU_SLOTS;
398 if (!kvm_s390_use_sca_entries())
400 else if (sclp.has_esca && sclp.has_64bscao)
401 r = KVM_S390_ESCA_CPU_SLOTS;
403 case KVM_CAP_NR_MEMSLOTS:
404 r = KVM_USER_MEM_SLOTS;
406 case KVM_CAP_S390_COW:
407 r = MACHINE_HAS_ESOP;
409 case KVM_CAP_S390_VECTOR_REGISTERS:
412 case KVM_CAP_S390_RI:
413 r = test_facility(64);
415 case KVM_CAP_S390_GS:
416 r = test_facility(133);
424 static void kvm_s390_sync_dirty_log(struct kvm *kvm,
425 struct kvm_memory_slot *memslot)
427 gfn_t cur_gfn, last_gfn;
428 unsigned long address;
429 struct gmap *gmap = kvm->arch.gmap;
431 /* Loop over all guest pages */
432 last_gfn = memslot->base_gfn + memslot->npages;
433 for (cur_gfn = memslot->base_gfn; cur_gfn <= last_gfn; cur_gfn++) {
434 address = gfn_to_hva_memslot(memslot, cur_gfn);
436 if (test_and_clear_guest_dirty(gmap->mm, address))
437 mark_page_dirty(kvm, cur_gfn);
438 if (fatal_signal_pending(current))
444 /* Section: vm related */
445 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
448 * Get (and clear) the dirty memory log for a memory slot.
450 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
451 struct kvm_dirty_log *log)
455 struct kvm_memslots *slots;
456 struct kvm_memory_slot *memslot;
459 if (kvm_is_ucontrol(kvm))
462 mutex_lock(&kvm->slots_lock);
465 if (log->slot >= KVM_USER_MEM_SLOTS)
468 slots = kvm_memslots(kvm);
469 memslot = id_to_memslot(slots, log->slot);
471 if (!memslot->dirty_bitmap)
474 kvm_s390_sync_dirty_log(kvm, memslot);
475 r = kvm_get_dirty_log(kvm, log, &is_dirty);
479 /* Clear the dirty log */
481 n = kvm_dirty_bitmap_bytes(memslot);
482 memset(memslot->dirty_bitmap, 0, n);
486 mutex_unlock(&kvm->slots_lock);
490 static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
493 struct kvm_vcpu *vcpu;
495 kvm_for_each_vcpu(i, vcpu, kvm) {
496 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
500 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
508 case KVM_CAP_S390_IRQCHIP:
509 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
510 kvm->arch.use_irqchip = 1;
513 case KVM_CAP_S390_USER_SIGP:
514 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
515 kvm->arch.user_sigp = 1;
518 case KVM_CAP_S390_VECTOR_REGISTERS:
519 mutex_lock(&kvm->lock);
520 if (kvm->created_vcpus) {
522 } else if (MACHINE_HAS_VX) {
523 set_kvm_facility(kvm->arch.model.fac_mask, 129);
524 set_kvm_facility(kvm->arch.model.fac_list, 129);
525 if (test_facility(134)) {
526 set_kvm_facility(kvm->arch.model.fac_mask, 134);
527 set_kvm_facility(kvm->arch.model.fac_list, 134);
529 if (test_facility(135)) {
530 set_kvm_facility(kvm->arch.model.fac_mask, 135);
531 set_kvm_facility(kvm->arch.model.fac_list, 135);
536 mutex_unlock(&kvm->lock);
537 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
538 r ? "(not available)" : "(success)");
540 case KVM_CAP_S390_RI:
542 mutex_lock(&kvm->lock);
543 if (kvm->created_vcpus) {
545 } else if (test_facility(64)) {
546 set_kvm_facility(kvm->arch.model.fac_mask, 64);
547 set_kvm_facility(kvm->arch.model.fac_list, 64);
550 mutex_unlock(&kvm->lock);
551 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
552 r ? "(not available)" : "(success)");
554 case KVM_CAP_S390_AIS:
555 mutex_lock(&kvm->lock);
556 if (kvm->created_vcpus) {
559 set_kvm_facility(kvm->arch.model.fac_mask, 72);
560 set_kvm_facility(kvm->arch.model.fac_list, 72);
561 kvm->arch.float_int.ais_enabled = 1;
564 mutex_unlock(&kvm->lock);
565 VM_EVENT(kvm, 3, "ENABLE: AIS %s",
566 r ? "(not available)" : "(success)");
568 case KVM_CAP_S390_GS:
570 mutex_lock(&kvm->lock);
571 if (atomic_read(&kvm->online_vcpus)) {
573 } else if (test_facility(133)) {
574 set_kvm_facility(kvm->arch.model.fac_mask, 133);
575 set_kvm_facility(kvm->arch.model.fac_list, 133);
578 mutex_unlock(&kvm->lock);
579 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_GS %s",
580 r ? "(not available)" : "(success)");
582 case KVM_CAP_S390_USER_STSI:
583 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
584 kvm->arch.user_stsi = 1;
587 case KVM_CAP_S390_USER_INSTR0:
588 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
589 kvm->arch.user_instr0 = 1;
590 icpt_operexc_on_all_vcpus(kvm);
600 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
604 switch (attr->attr) {
605 case KVM_S390_VM_MEM_LIMIT_SIZE:
607 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
608 kvm->arch.mem_limit);
609 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
619 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
623 switch (attr->attr) {
624 case KVM_S390_VM_MEM_ENABLE_CMMA:
630 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
631 mutex_lock(&kvm->lock);
632 if (!kvm->created_vcpus) {
633 kvm->arch.use_cmma = 1;
636 mutex_unlock(&kvm->lock);
638 case KVM_S390_VM_MEM_CLR_CMMA:
643 if (!kvm->arch.use_cmma)
646 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
647 mutex_lock(&kvm->lock);
648 idx = srcu_read_lock(&kvm->srcu);
649 s390_reset_cmma(kvm->arch.gmap->mm);
650 srcu_read_unlock(&kvm->srcu, idx);
651 mutex_unlock(&kvm->lock);
654 case KVM_S390_VM_MEM_LIMIT_SIZE: {
655 unsigned long new_limit;
657 if (kvm_is_ucontrol(kvm))
660 if (get_user(new_limit, (u64 __user *)attr->addr))
663 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
664 new_limit > kvm->arch.mem_limit)
670 /* gmap_create takes last usable address */
671 if (new_limit != KVM_S390_NO_MEM_LIMIT)
675 mutex_lock(&kvm->lock);
676 if (!kvm->created_vcpus) {
677 /* gmap_create will round the limit up */
678 struct gmap *new = gmap_create(current->mm, new_limit);
683 gmap_remove(kvm->arch.gmap);
685 kvm->arch.gmap = new;
689 mutex_unlock(&kvm->lock);
690 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
691 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
692 (void *) kvm->arch.gmap->asce);
702 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
704 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
706 struct kvm_vcpu *vcpu;
709 if (!test_kvm_facility(kvm, 76))
712 mutex_lock(&kvm->lock);
713 switch (attr->attr) {
714 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
716 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
717 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
718 kvm->arch.crypto.aes_kw = 1;
719 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
721 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
723 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
724 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
725 kvm->arch.crypto.dea_kw = 1;
726 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
728 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
729 kvm->arch.crypto.aes_kw = 0;
730 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
731 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
732 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
734 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
735 kvm->arch.crypto.dea_kw = 0;
736 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
737 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
738 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
741 mutex_unlock(&kvm->lock);
745 kvm_for_each_vcpu(i, vcpu, kvm) {
746 kvm_s390_vcpu_crypto_setup(vcpu);
749 mutex_unlock(&kvm->lock);
753 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
757 if (copy_from_user(>od_high, (void __user *)attr->addr,
763 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
768 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
772 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod)))
775 kvm_s390_set_tod_clock(kvm, gtod);
776 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod);
780 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
787 switch (attr->attr) {
788 case KVM_S390_VM_TOD_HIGH:
789 ret = kvm_s390_set_tod_high(kvm, attr);
791 case KVM_S390_VM_TOD_LOW:
792 ret = kvm_s390_set_tod_low(kvm, attr);
801 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
805 if (copy_to_user((void __user *)attr->addr, >od_high,
808 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
813 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
817 gtod = kvm_s390_get_tod_clock_fast(kvm);
818 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
820 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
825 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
832 switch (attr->attr) {
833 case KVM_S390_VM_TOD_HIGH:
834 ret = kvm_s390_get_tod_high(kvm, attr);
836 case KVM_S390_VM_TOD_LOW:
837 ret = kvm_s390_get_tod_low(kvm, attr);
846 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
848 struct kvm_s390_vm_cpu_processor *proc;
849 u16 lowest_ibc, unblocked_ibc;
852 mutex_lock(&kvm->lock);
853 if (kvm->created_vcpus) {
857 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
862 if (!copy_from_user(proc, (void __user *)attr->addr,
864 kvm->arch.model.cpuid = proc->cpuid;
865 lowest_ibc = sclp.ibc >> 16 & 0xfff;
866 unblocked_ibc = sclp.ibc & 0xfff;
867 if (lowest_ibc && proc->ibc) {
868 if (proc->ibc > unblocked_ibc)
869 kvm->arch.model.ibc = unblocked_ibc;
870 else if (proc->ibc < lowest_ibc)
871 kvm->arch.model.ibc = lowest_ibc;
873 kvm->arch.model.ibc = proc->ibc;
875 memcpy(kvm->arch.model.fac_list, proc->fac_list,
876 S390_ARCH_FAC_LIST_SIZE_BYTE);
877 VM_EVENT(kvm, 3, "SET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
879 kvm->arch.model.cpuid);
880 VM_EVENT(kvm, 3, "SET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
881 kvm->arch.model.fac_list[0],
882 kvm->arch.model.fac_list[1],
883 kvm->arch.model.fac_list[2]);
888 mutex_unlock(&kvm->lock);
892 static int kvm_s390_set_processor_feat(struct kvm *kvm,
893 struct kvm_device_attr *attr)
895 struct kvm_s390_vm_cpu_feat data;
898 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
900 if (!bitmap_subset((unsigned long *) data.feat,
901 kvm_s390_available_cpu_feat,
902 KVM_S390_VM_CPU_FEAT_NR_BITS))
905 mutex_lock(&kvm->lock);
906 if (!atomic_read(&kvm->online_vcpus)) {
907 bitmap_copy(kvm->arch.cpu_feat, (unsigned long *) data.feat,
908 KVM_S390_VM_CPU_FEAT_NR_BITS);
911 mutex_unlock(&kvm->lock);
915 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
916 struct kvm_device_attr *attr)
919 * Once supported by kernel + hw, we have to store the subfunctions
920 * in kvm->arch and remember that user space configured them.
925 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
929 switch (attr->attr) {
930 case KVM_S390_VM_CPU_PROCESSOR:
931 ret = kvm_s390_set_processor(kvm, attr);
933 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
934 ret = kvm_s390_set_processor_feat(kvm, attr);
936 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
937 ret = kvm_s390_set_processor_subfunc(kvm, attr);
943 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
945 struct kvm_s390_vm_cpu_processor *proc;
948 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
953 proc->cpuid = kvm->arch.model.cpuid;
954 proc->ibc = kvm->arch.model.ibc;
955 memcpy(&proc->fac_list, kvm->arch.model.fac_list,
956 S390_ARCH_FAC_LIST_SIZE_BYTE);
957 VM_EVENT(kvm, 3, "GET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
959 kvm->arch.model.cpuid);
960 VM_EVENT(kvm, 3, "GET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
961 kvm->arch.model.fac_list[0],
962 kvm->arch.model.fac_list[1],
963 kvm->arch.model.fac_list[2]);
964 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
971 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
973 struct kvm_s390_vm_cpu_machine *mach;
976 mach = kzalloc(sizeof(*mach), GFP_KERNEL);
981 get_cpu_id((struct cpuid *) &mach->cpuid);
982 mach->ibc = sclp.ibc;
983 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
984 S390_ARCH_FAC_LIST_SIZE_BYTE);
985 memcpy((unsigned long *)&mach->fac_list, S390_lowcore.stfle_fac_list,
986 sizeof(S390_lowcore.stfle_fac_list));
987 VM_EVENT(kvm, 3, "GET: host ibc: 0x%4.4x, host cpuid: 0x%16.16llx",
989 kvm->arch.model.cpuid);
990 VM_EVENT(kvm, 3, "GET: host facmask: 0x%16.16llx.%16.16llx.%16.16llx",
994 VM_EVENT(kvm, 3, "GET: host faclist: 0x%16.16llx.%16.16llx.%16.16llx",
998 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
1005 static int kvm_s390_get_processor_feat(struct kvm *kvm,
1006 struct kvm_device_attr *attr)
1008 struct kvm_s390_vm_cpu_feat data;
1010 bitmap_copy((unsigned long *) data.feat, kvm->arch.cpu_feat,
1011 KVM_S390_VM_CPU_FEAT_NR_BITS);
1012 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1017 static int kvm_s390_get_machine_feat(struct kvm *kvm,
1018 struct kvm_device_attr *attr)
1020 struct kvm_s390_vm_cpu_feat data;
1022 bitmap_copy((unsigned long *) data.feat,
1023 kvm_s390_available_cpu_feat,
1024 KVM_S390_VM_CPU_FEAT_NR_BITS);
1025 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1030 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
1031 struct kvm_device_attr *attr)
1034 * Once we can actually configure subfunctions (kernel + hw support),
1035 * we have to check if they were already set by user space, if so copy
1036 * them from kvm->arch.
1041 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
1042 struct kvm_device_attr *attr)
1044 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
1045 sizeof(struct kvm_s390_vm_cpu_subfunc)))
1049 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1053 switch (attr->attr) {
1054 case KVM_S390_VM_CPU_PROCESSOR:
1055 ret = kvm_s390_get_processor(kvm, attr);
1057 case KVM_S390_VM_CPU_MACHINE:
1058 ret = kvm_s390_get_machine(kvm, attr);
1060 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1061 ret = kvm_s390_get_processor_feat(kvm, attr);
1063 case KVM_S390_VM_CPU_MACHINE_FEAT:
1064 ret = kvm_s390_get_machine_feat(kvm, attr);
1066 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1067 ret = kvm_s390_get_processor_subfunc(kvm, attr);
1069 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1070 ret = kvm_s390_get_machine_subfunc(kvm, attr);
1076 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1080 switch (attr->group) {
1081 case KVM_S390_VM_MEM_CTRL:
1082 ret = kvm_s390_set_mem_control(kvm, attr);
1084 case KVM_S390_VM_TOD:
1085 ret = kvm_s390_set_tod(kvm, attr);
1087 case KVM_S390_VM_CPU_MODEL:
1088 ret = kvm_s390_set_cpu_model(kvm, attr);
1090 case KVM_S390_VM_CRYPTO:
1091 ret = kvm_s390_vm_set_crypto(kvm, attr);
1101 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1105 switch (attr->group) {
1106 case KVM_S390_VM_MEM_CTRL:
1107 ret = kvm_s390_get_mem_control(kvm, attr);
1109 case KVM_S390_VM_TOD:
1110 ret = kvm_s390_get_tod(kvm, attr);
1112 case KVM_S390_VM_CPU_MODEL:
1113 ret = kvm_s390_get_cpu_model(kvm, attr);
1123 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1127 switch (attr->group) {
1128 case KVM_S390_VM_MEM_CTRL:
1129 switch (attr->attr) {
1130 case KVM_S390_VM_MEM_ENABLE_CMMA:
1131 case KVM_S390_VM_MEM_CLR_CMMA:
1132 ret = sclp.has_cmma ? 0 : -ENXIO;
1134 case KVM_S390_VM_MEM_LIMIT_SIZE:
1142 case KVM_S390_VM_TOD:
1143 switch (attr->attr) {
1144 case KVM_S390_VM_TOD_LOW:
1145 case KVM_S390_VM_TOD_HIGH:
1153 case KVM_S390_VM_CPU_MODEL:
1154 switch (attr->attr) {
1155 case KVM_S390_VM_CPU_PROCESSOR:
1156 case KVM_S390_VM_CPU_MACHINE:
1157 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1158 case KVM_S390_VM_CPU_MACHINE_FEAT:
1159 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1162 /* configuring subfunctions is not supported yet */
1163 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1169 case KVM_S390_VM_CRYPTO:
1170 switch (attr->attr) {
1171 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1172 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1173 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1174 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1190 static long kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1196 if (args->flags != 0)
1199 /* Is this guest using storage keys? */
1200 if (!mm_use_skey(current->mm))
1201 return KVM_S390_GET_SKEYS_NONE;
1203 /* Enforce sane limit on memory allocation */
1204 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1207 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL);
1211 down_read(¤t->mm->mmap_sem);
1212 for (i = 0; i < args->count; i++) {
1213 hva = gfn_to_hva(kvm, args->start_gfn + i);
1214 if (kvm_is_error_hva(hva)) {
1219 r = get_guest_storage_key(current->mm, hva, &keys[i]);
1223 up_read(¤t->mm->mmap_sem);
1226 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
1227 sizeof(uint8_t) * args->count);
1236 static long kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1242 if (args->flags != 0)
1245 /* Enforce sane limit on memory allocation */
1246 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1249 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL);
1253 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
1254 sizeof(uint8_t) * args->count);
1260 /* Enable storage key handling for the guest */
1261 r = s390_enable_skey();
1265 down_read(¤t->mm->mmap_sem);
1266 for (i = 0; i < args->count; i++) {
1267 hva = gfn_to_hva(kvm, args->start_gfn + i);
1268 if (kvm_is_error_hva(hva)) {
1273 /* Lowest order bit is reserved */
1274 if (keys[i] & 0x01) {
1279 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
1283 up_read(¤t->mm->mmap_sem);
1289 long kvm_arch_vm_ioctl(struct file *filp,
1290 unsigned int ioctl, unsigned long arg)
1292 struct kvm *kvm = filp->private_data;
1293 void __user *argp = (void __user *)arg;
1294 struct kvm_device_attr attr;
1298 case KVM_S390_INTERRUPT: {
1299 struct kvm_s390_interrupt s390int;
1302 if (copy_from_user(&s390int, argp, sizeof(s390int)))
1304 r = kvm_s390_inject_vm(kvm, &s390int);
1307 case KVM_ENABLE_CAP: {
1308 struct kvm_enable_cap cap;
1310 if (copy_from_user(&cap, argp, sizeof(cap)))
1312 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1315 case KVM_CREATE_IRQCHIP: {
1316 struct kvm_irq_routing_entry routing;
1319 if (kvm->arch.use_irqchip) {
1320 /* Set up dummy routing. */
1321 memset(&routing, 0, sizeof(routing));
1322 r = kvm_set_irq_routing(kvm, &routing, 0, 0);
1326 case KVM_SET_DEVICE_ATTR: {
1328 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1330 r = kvm_s390_vm_set_attr(kvm, &attr);
1333 case KVM_GET_DEVICE_ATTR: {
1335 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1337 r = kvm_s390_vm_get_attr(kvm, &attr);
1340 case KVM_HAS_DEVICE_ATTR: {
1342 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1344 r = kvm_s390_vm_has_attr(kvm, &attr);
1347 case KVM_S390_GET_SKEYS: {
1348 struct kvm_s390_skeys args;
1351 if (copy_from_user(&args, argp,
1352 sizeof(struct kvm_s390_skeys)))
1354 r = kvm_s390_get_skeys(kvm, &args);
1357 case KVM_S390_SET_SKEYS: {
1358 struct kvm_s390_skeys args;
1361 if (copy_from_user(&args, argp,
1362 sizeof(struct kvm_s390_skeys)))
1364 r = kvm_s390_set_skeys(kvm, &args);
1374 static int kvm_s390_query_ap_config(u8 *config)
1376 u32 fcn_code = 0x04000000UL;
1379 memset(config, 0, 128);
1383 ".long 0xb2af0000\n" /* PQAP(QCI) */
1389 : "r" (fcn_code), "r" (config)
1390 : "cc", "0", "2", "memory"
1396 static int kvm_s390_apxa_installed(void)
1401 if (test_facility(12)) {
1402 cc = kvm_s390_query_ap_config(config);
1405 pr_err("PQAP(QCI) failed with cc=%d", cc);
1407 return config[0] & 0x40;
1413 static void kvm_s390_set_crycb_format(struct kvm *kvm)
1415 kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
1417 if (kvm_s390_apxa_installed())
1418 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
1420 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
1423 static u64 kvm_s390_get_initial_cpuid(void)
1428 cpuid.version = 0xff;
1429 return *((u64 *) &cpuid);
1432 static void kvm_s390_crypto_init(struct kvm *kvm)
1434 if (!test_kvm_facility(kvm, 76))
1437 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
1438 kvm_s390_set_crycb_format(kvm);
1440 /* Enable AES/DEA protected key functions by default */
1441 kvm->arch.crypto.aes_kw = 1;
1442 kvm->arch.crypto.dea_kw = 1;
1443 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1444 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1445 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1446 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1449 static void sca_dispose(struct kvm *kvm)
1451 if (kvm->arch.use_esca)
1452 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
1454 free_page((unsigned long)(kvm->arch.sca));
1455 kvm->arch.sca = NULL;
1458 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
1460 gfp_t alloc_flags = GFP_KERNEL;
1462 char debug_name[16];
1463 static unsigned long sca_offset;
1466 #ifdef CONFIG_KVM_S390_UCONTROL
1467 if (type & ~KVM_VM_S390_UCONTROL)
1469 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
1476 rc = s390_enable_sie();
1482 ratelimit_state_init(&kvm->arch.sthyi_limit, 5 * HZ, 500);
1484 kvm->arch.use_esca = 0; /* start with basic SCA */
1485 if (!sclp.has_64bscao)
1486 alloc_flags |= GFP_DMA;
1487 rwlock_init(&kvm->arch.sca_lock);
1488 kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
1491 spin_lock(&kvm_lock);
1493 if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
1495 kvm->arch.sca = (struct bsca_block *)
1496 ((char *) kvm->arch.sca + sca_offset);
1497 spin_unlock(&kvm_lock);
1499 sprintf(debug_name, "kvm-%u", current->pid);
1501 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
1505 kvm->arch.sie_page2 =
1506 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1507 if (!kvm->arch.sie_page2)
1510 /* Populate the facility mask initially. */
1511 memcpy(kvm->arch.model.fac_mask, S390_lowcore.stfle_fac_list,
1512 sizeof(S390_lowcore.stfle_fac_list));
1513 for (i = 0; i < S390_ARCH_FAC_LIST_SIZE_U64; i++) {
1514 if (i < kvm_s390_fac_list_mask_size())
1515 kvm->arch.model.fac_mask[i] &= kvm_s390_fac_list_mask[i];
1517 kvm->arch.model.fac_mask[i] = 0UL;
1520 /* Populate the facility list initially. */
1521 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
1522 memcpy(kvm->arch.model.fac_list, kvm->arch.model.fac_mask,
1523 S390_ARCH_FAC_LIST_SIZE_BYTE);
1525 set_kvm_facility(kvm->arch.model.fac_mask, 74);
1526 set_kvm_facility(kvm->arch.model.fac_list, 74);
1528 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
1529 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
1531 kvm_s390_crypto_init(kvm);
1533 mutex_init(&kvm->arch.float_int.ais_lock);
1534 kvm->arch.float_int.simm = 0;
1535 kvm->arch.float_int.nimm = 0;
1536 kvm->arch.float_int.ais_enabled = 0;
1537 spin_lock_init(&kvm->arch.float_int.lock);
1538 for (i = 0; i < FIRQ_LIST_COUNT; i++)
1539 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
1540 init_waitqueue_head(&kvm->arch.ipte_wq);
1541 mutex_init(&kvm->arch.ipte_mutex);
1543 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
1544 VM_EVENT(kvm, 3, "vm created with type %lu", type);
1546 if (type & KVM_VM_S390_UCONTROL) {
1547 kvm->arch.gmap = NULL;
1548 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
1550 if (sclp.hamax == U64_MAX)
1551 kvm->arch.mem_limit = TASK_SIZE_MAX;
1553 kvm->arch.mem_limit = min_t(unsigned long, TASK_SIZE_MAX,
1555 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
1556 if (!kvm->arch.gmap)
1558 kvm->arch.gmap->private = kvm;
1559 kvm->arch.gmap->pfault_enabled = 0;
1562 kvm->arch.css_support = 0;
1563 kvm->arch.use_irqchip = 0;
1564 kvm->arch.epoch = 0;
1566 spin_lock_init(&kvm->arch.start_stop_lock);
1567 kvm_s390_vsie_init(kvm);
1568 KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
1572 free_page((unsigned long)kvm->arch.sie_page2);
1573 debug_unregister(kvm->arch.dbf);
1575 KVM_EVENT(3, "creation of vm failed: %d", rc);
1579 bool kvm_arch_has_vcpu_debugfs(void)
1584 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
1589 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1591 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
1592 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
1593 kvm_s390_clear_local_irqs(vcpu);
1594 kvm_clear_async_pf_completion_queue(vcpu);
1595 if (!kvm_is_ucontrol(vcpu->kvm))
1598 if (kvm_is_ucontrol(vcpu->kvm))
1599 gmap_remove(vcpu->arch.gmap);
1601 if (vcpu->kvm->arch.use_cmma)
1602 kvm_s390_vcpu_unsetup_cmma(vcpu);
1603 free_page((unsigned long)(vcpu->arch.sie_block));
1605 kvm_vcpu_uninit(vcpu);
1606 kmem_cache_free(kvm_vcpu_cache, vcpu);
1609 static void kvm_free_vcpus(struct kvm *kvm)
1612 struct kvm_vcpu *vcpu;
1614 kvm_for_each_vcpu(i, vcpu, kvm)
1615 kvm_arch_vcpu_destroy(vcpu);
1617 mutex_lock(&kvm->lock);
1618 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
1619 kvm->vcpus[i] = NULL;
1621 atomic_set(&kvm->online_vcpus, 0);
1622 mutex_unlock(&kvm->lock);
1625 void kvm_arch_destroy_vm(struct kvm *kvm)
1627 kvm_free_vcpus(kvm);
1629 debug_unregister(kvm->arch.dbf);
1630 free_page((unsigned long)kvm->arch.sie_page2);
1631 if (!kvm_is_ucontrol(kvm))
1632 gmap_remove(kvm->arch.gmap);
1633 kvm_s390_destroy_adapters(kvm);
1634 kvm_s390_clear_float_irqs(kvm);
1635 kvm_s390_vsie_destroy(kvm);
1636 KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
1639 /* Section: vcpu related */
1640 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
1642 vcpu->arch.gmap = gmap_create(current->mm, -1UL);
1643 if (!vcpu->arch.gmap)
1645 vcpu->arch.gmap->private = vcpu->kvm;
1650 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
1652 if (!kvm_s390_use_sca_entries())
1654 read_lock(&vcpu->kvm->arch.sca_lock);
1655 if (vcpu->kvm->arch.use_esca) {
1656 struct esca_block *sca = vcpu->kvm->arch.sca;
1658 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
1659 sca->cpu[vcpu->vcpu_id].sda = 0;
1661 struct bsca_block *sca = vcpu->kvm->arch.sca;
1663 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
1664 sca->cpu[vcpu->vcpu_id].sda = 0;
1666 read_unlock(&vcpu->kvm->arch.sca_lock);
1669 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
1671 if (!kvm_s390_use_sca_entries()) {
1672 struct bsca_block *sca = vcpu->kvm->arch.sca;
1674 /* we still need the basic sca for the ipte control */
1675 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1676 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
1678 read_lock(&vcpu->kvm->arch.sca_lock);
1679 if (vcpu->kvm->arch.use_esca) {
1680 struct esca_block *sca = vcpu->kvm->arch.sca;
1682 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
1683 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1684 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca & ~0x3fU;
1685 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
1686 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
1688 struct bsca_block *sca = vcpu->kvm->arch.sca;
1690 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
1691 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1692 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
1693 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
1695 read_unlock(&vcpu->kvm->arch.sca_lock);
1698 /* Basic SCA to Extended SCA data copy routines */
1699 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
1702 d->sigp_ctrl.c = s->sigp_ctrl.c;
1703 d->sigp_ctrl.scn = s->sigp_ctrl.scn;
1706 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
1710 d->ipte_control = s->ipte_control;
1712 for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
1713 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
1716 static int sca_switch_to_extended(struct kvm *kvm)
1718 struct bsca_block *old_sca = kvm->arch.sca;
1719 struct esca_block *new_sca;
1720 struct kvm_vcpu *vcpu;
1721 unsigned int vcpu_idx;
1724 new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL|__GFP_ZERO);
1728 scaoh = (u32)((u64)(new_sca) >> 32);
1729 scaol = (u32)(u64)(new_sca) & ~0x3fU;
1731 kvm_s390_vcpu_block_all(kvm);
1732 write_lock(&kvm->arch.sca_lock);
1734 sca_copy_b_to_e(new_sca, old_sca);
1736 kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
1737 vcpu->arch.sie_block->scaoh = scaoh;
1738 vcpu->arch.sie_block->scaol = scaol;
1739 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
1741 kvm->arch.sca = new_sca;
1742 kvm->arch.use_esca = 1;
1744 write_unlock(&kvm->arch.sca_lock);
1745 kvm_s390_vcpu_unblock_all(kvm);
1747 free_page((unsigned long)old_sca);
1749 VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
1750 old_sca, kvm->arch.sca);
1754 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
1758 if (!kvm_s390_use_sca_entries()) {
1759 if (id < KVM_MAX_VCPUS)
1763 if (id < KVM_S390_BSCA_CPU_SLOTS)
1765 if (!sclp.has_esca || !sclp.has_64bscao)
1768 mutex_lock(&kvm->lock);
1769 rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
1770 mutex_unlock(&kvm->lock);
1772 return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
1775 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1777 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1778 kvm_clear_async_pf_completion_queue(vcpu);
1779 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
1785 kvm_s390_set_prefix(vcpu, 0);
1786 if (test_kvm_facility(vcpu->kvm, 64))
1787 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
1788 if (test_kvm_facility(vcpu->kvm, 133))
1789 vcpu->run->kvm_valid_regs |= KVM_SYNC_GSCB;
1790 /* fprs can be synchronized via vrs, even if the guest has no vx. With
1791 * MACHINE_HAS_VX, (load|store)_fpu_regs() will work with vrs format.
1794 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
1796 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
1798 if (kvm_is_ucontrol(vcpu->kvm))
1799 return __kvm_ucontrol_vcpu_init(vcpu);
1804 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1805 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1807 WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
1808 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1809 vcpu->arch.cputm_start = get_tod_clock_fast();
1810 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1813 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1814 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1816 WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
1817 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1818 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
1819 vcpu->arch.cputm_start = 0;
1820 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1823 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1824 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1826 WARN_ON_ONCE(vcpu->arch.cputm_enabled);
1827 vcpu->arch.cputm_enabled = true;
1828 __start_cpu_timer_accounting(vcpu);
1831 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1832 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1834 WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
1835 __stop_cpu_timer_accounting(vcpu);
1836 vcpu->arch.cputm_enabled = false;
1839 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1841 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1842 __enable_cpu_timer_accounting(vcpu);
1846 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1848 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1849 __disable_cpu_timer_accounting(vcpu);
1853 /* set the cpu timer - may only be called from the VCPU thread itself */
1854 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
1856 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1857 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1858 if (vcpu->arch.cputm_enabled)
1859 vcpu->arch.cputm_start = get_tod_clock_fast();
1860 vcpu->arch.sie_block->cputm = cputm;
1861 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1865 /* update and get the cpu timer - can also be called from other VCPU threads */
1866 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
1871 if (unlikely(!vcpu->arch.cputm_enabled))
1872 return vcpu->arch.sie_block->cputm;
1874 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1876 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
1878 * If the writer would ever execute a read in the critical
1879 * section, e.g. in irq context, we have a deadlock.
1881 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
1882 value = vcpu->arch.sie_block->cputm;
1883 /* if cputm_start is 0, accounting is being started/stopped */
1884 if (likely(vcpu->arch.cputm_start))
1885 value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
1886 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
1891 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1894 gmap_enable(vcpu->arch.enabled_gmap);
1895 atomic_or(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1896 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
1897 __start_cpu_timer_accounting(vcpu);
1901 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1904 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
1905 __stop_cpu_timer_accounting(vcpu);
1906 atomic_andnot(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1907 vcpu->arch.enabled_gmap = gmap_get_enabled();
1908 gmap_disable(vcpu->arch.enabled_gmap);
1912 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
1914 /* this equals initial cpu reset in pop, but we don't switch to ESA */
1915 vcpu->arch.sie_block->gpsw.mask = 0UL;
1916 vcpu->arch.sie_block->gpsw.addr = 0UL;
1917 kvm_s390_set_prefix(vcpu, 0);
1918 kvm_s390_set_cpu_timer(vcpu, 0);
1919 vcpu->arch.sie_block->ckc = 0UL;
1920 vcpu->arch.sie_block->todpr = 0;
1921 memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64));
1922 vcpu->arch.sie_block->gcr[0] = 0xE0UL;
1923 vcpu->arch.sie_block->gcr[14] = 0xC2000000UL;
1924 /* make sure the new fpc will be lazily loaded */
1926 current->thread.fpu.fpc = 0;
1927 vcpu->arch.sie_block->gbea = 1;
1928 vcpu->arch.sie_block->pp = 0;
1929 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1930 kvm_clear_async_pf_completion_queue(vcpu);
1931 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
1932 kvm_s390_vcpu_stop(vcpu);
1933 kvm_s390_clear_local_irqs(vcpu);
1936 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1938 mutex_lock(&vcpu->kvm->lock);
1940 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
1942 mutex_unlock(&vcpu->kvm->lock);
1943 if (!kvm_is_ucontrol(vcpu->kvm)) {
1944 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
1947 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
1948 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
1949 /* make vcpu_load load the right gmap on the first trigger */
1950 vcpu->arch.enabled_gmap = vcpu->arch.gmap;
1953 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
1955 if (!test_kvm_facility(vcpu->kvm, 76))
1958 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
1960 if (vcpu->kvm->arch.crypto.aes_kw)
1961 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
1962 if (vcpu->kvm->arch.crypto.dea_kw)
1963 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
1965 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
1968 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
1970 free_page(vcpu->arch.sie_block->cbrlo);
1971 vcpu->arch.sie_block->cbrlo = 0;
1974 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
1976 vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
1977 if (!vcpu->arch.sie_block->cbrlo)
1980 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1981 vcpu->arch.sie_block->ecb2 &= ~ECB2_PFMFI;
1985 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
1987 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
1989 vcpu->arch.sie_block->ibc = model->ibc;
1990 if (test_kvm_facility(vcpu->kvm, 7))
1991 vcpu->arch.sie_block->fac = (u32)(u64) model->fac_list;
1994 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1998 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
2002 if (test_kvm_facility(vcpu->kvm, 78))
2003 atomic_or(CPUSTAT_GED2, &vcpu->arch.sie_block->cpuflags);
2004 else if (test_kvm_facility(vcpu->kvm, 8))
2005 atomic_or(CPUSTAT_GED, &vcpu->arch.sie_block->cpuflags);
2007 kvm_s390_vcpu_setup_model(vcpu);
2009 /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
2010 if (MACHINE_HAS_ESOP)
2011 vcpu->arch.sie_block->ecb |= ECB_HOSTPROTINT;
2012 if (test_kvm_facility(vcpu->kvm, 9))
2013 vcpu->arch.sie_block->ecb |= ECB_SRSI;
2014 if (test_kvm_facility(vcpu->kvm, 73))
2015 vcpu->arch.sie_block->ecb |= ECB_TE;
2017 if (test_kvm_facility(vcpu->kvm, 8) && sclp.has_pfmfi)
2018 vcpu->arch.sie_block->ecb2 |= ECB2_PFMFI;
2019 if (test_kvm_facility(vcpu->kvm, 130))
2020 vcpu->arch.sie_block->ecb2 |= ECB2_IEP;
2021 vcpu->arch.sie_block->eca = ECA_MVPGI | ECA_PROTEXCI;
2023 vcpu->arch.sie_block->eca |= ECA_CEI;
2025 vcpu->arch.sie_block->eca |= ECA_IB;
2027 vcpu->arch.sie_block->eca |= ECA_SII;
2028 if (sclp.has_sigpif)
2029 vcpu->arch.sie_block->eca |= ECA_SIGPI;
2030 if (test_kvm_facility(vcpu->kvm, 129)) {
2031 vcpu->arch.sie_block->eca |= ECA_VX;
2032 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
2034 vcpu->arch.sie_block->sdnxo = ((unsigned long) &vcpu->run->s.regs.sdnx)
2036 vcpu->arch.sie_block->riccbd = (unsigned long) &vcpu->run->s.regs.riccb;
2039 atomic_or(CPUSTAT_KSS, &vcpu->arch.sie_block->cpuflags);
2041 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
2043 if (vcpu->kvm->arch.use_cmma) {
2044 rc = kvm_s390_vcpu_setup_cmma(vcpu);
2048 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
2049 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
2051 kvm_s390_vcpu_crypto_setup(vcpu);
2056 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
2059 struct kvm_vcpu *vcpu;
2060 struct sie_page *sie_page;
2063 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
2068 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2072 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
2076 vcpu->arch.sie_block = &sie_page->sie_block;
2077 vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
2079 /* the real guest size will always be smaller than msl */
2080 vcpu->arch.sie_block->mso = 0;
2081 vcpu->arch.sie_block->msl = sclp.hamax;
2083 vcpu->arch.sie_block->icpua = id;
2084 spin_lock_init(&vcpu->arch.local_int.lock);
2085 vcpu->arch.local_int.float_int = &kvm->arch.float_int;
2086 vcpu->arch.local_int.wq = &vcpu->wq;
2087 vcpu->arch.local_int.cpuflags = &vcpu->arch.sie_block->cpuflags;
2088 seqcount_init(&vcpu->arch.cputm_seqcount);
2090 rc = kvm_vcpu_init(vcpu, kvm, id);
2092 goto out_free_sie_block;
2093 VM_EVENT(kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK", id, vcpu,
2094 vcpu->arch.sie_block);
2095 trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
2099 free_page((unsigned long)(vcpu->arch.sie_block));
2101 kmem_cache_free(kvm_vcpu_cache, vcpu);
2106 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
2108 return kvm_s390_vcpu_has_irq(vcpu, 0);
2111 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
2113 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2117 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
2119 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2122 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
2124 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2128 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
2130 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2134 * Kick a guest cpu out of SIE and wait until SIE is not running.
2135 * If the CPU is not running (e.g. waiting as idle) the function will
2136 * return immediately. */
2137 void exit_sie(struct kvm_vcpu *vcpu)
2139 atomic_or(CPUSTAT_STOP_INT, &vcpu->arch.sie_block->cpuflags);
2140 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
2144 /* Kick a guest cpu out of SIE to process a request synchronously */
2145 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
2147 kvm_make_request(req, vcpu);
2148 kvm_s390_vcpu_request(vcpu);
2151 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
2154 struct kvm *kvm = gmap->private;
2155 struct kvm_vcpu *vcpu;
2156 unsigned long prefix;
2159 if (gmap_is_shadow(gmap))
2161 if (start >= 1UL << 31)
2162 /* We are only interested in prefix pages */
2164 kvm_for_each_vcpu(i, vcpu, kvm) {
2165 /* match against both prefix pages */
2166 prefix = kvm_s390_get_prefix(vcpu);
2167 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
2168 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
2170 kvm_s390_sync_request(KVM_REQ_MMU_RELOAD, vcpu);
2175 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
2177 /* kvm common code refers to this, but never calls it */
2182 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
2183 struct kvm_one_reg *reg)
2188 case KVM_REG_S390_TODPR:
2189 r = put_user(vcpu->arch.sie_block->todpr,
2190 (u32 __user *)reg->addr);
2192 case KVM_REG_S390_EPOCHDIFF:
2193 r = put_user(vcpu->arch.sie_block->epoch,
2194 (u64 __user *)reg->addr);
2196 case KVM_REG_S390_CPU_TIMER:
2197 r = put_user(kvm_s390_get_cpu_timer(vcpu),
2198 (u64 __user *)reg->addr);
2200 case KVM_REG_S390_CLOCK_COMP:
2201 r = put_user(vcpu->arch.sie_block->ckc,
2202 (u64 __user *)reg->addr);
2204 case KVM_REG_S390_PFTOKEN:
2205 r = put_user(vcpu->arch.pfault_token,
2206 (u64 __user *)reg->addr);
2208 case KVM_REG_S390_PFCOMPARE:
2209 r = put_user(vcpu->arch.pfault_compare,
2210 (u64 __user *)reg->addr);
2212 case KVM_REG_S390_PFSELECT:
2213 r = put_user(vcpu->arch.pfault_select,
2214 (u64 __user *)reg->addr);
2216 case KVM_REG_S390_PP:
2217 r = put_user(vcpu->arch.sie_block->pp,
2218 (u64 __user *)reg->addr);
2220 case KVM_REG_S390_GBEA:
2221 r = put_user(vcpu->arch.sie_block->gbea,
2222 (u64 __user *)reg->addr);
2231 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
2232 struct kvm_one_reg *reg)
2238 case KVM_REG_S390_TODPR:
2239 r = get_user(vcpu->arch.sie_block->todpr,
2240 (u32 __user *)reg->addr);
2242 case KVM_REG_S390_EPOCHDIFF:
2243 r = get_user(vcpu->arch.sie_block->epoch,
2244 (u64 __user *)reg->addr);
2246 case KVM_REG_S390_CPU_TIMER:
2247 r = get_user(val, (u64 __user *)reg->addr);
2249 kvm_s390_set_cpu_timer(vcpu, val);
2251 case KVM_REG_S390_CLOCK_COMP:
2252 r = get_user(vcpu->arch.sie_block->ckc,
2253 (u64 __user *)reg->addr);
2255 case KVM_REG_S390_PFTOKEN:
2256 r = get_user(vcpu->arch.pfault_token,
2257 (u64 __user *)reg->addr);
2258 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2259 kvm_clear_async_pf_completion_queue(vcpu);
2261 case KVM_REG_S390_PFCOMPARE:
2262 r = get_user(vcpu->arch.pfault_compare,
2263 (u64 __user *)reg->addr);
2265 case KVM_REG_S390_PFSELECT:
2266 r = get_user(vcpu->arch.pfault_select,
2267 (u64 __user *)reg->addr);
2269 case KVM_REG_S390_PP:
2270 r = get_user(vcpu->arch.sie_block->pp,
2271 (u64 __user *)reg->addr);
2273 case KVM_REG_S390_GBEA:
2274 r = get_user(vcpu->arch.sie_block->gbea,
2275 (u64 __user *)reg->addr);
2284 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
2286 kvm_s390_vcpu_initial_reset(vcpu);
2290 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2292 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs));
2296 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2298 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
2302 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2303 struct kvm_sregs *sregs)
2305 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
2306 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
2310 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2311 struct kvm_sregs *sregs)
2313 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
2314 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
2318 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2320 if (test_fp_ctl(fpu->fpc))
2322 vcpu->run->s.regs.fpc = fpu->fpc;
2324 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs,
2325 (freg_t *) fpu->fprs);
2327 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs));
2331 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2333 /* make sure we have the latest values */
2336 convert_vx_to_fp((freg_t *) fpu->fprs,
2337 (__vector128 *) vcpu->run->s.regs.vrs);
2339 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs));
2340 fpu->fpc = vcpu->run->s.regs.fpc;
2344 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
2348 if (!is_vcpu_stopped(vcpu))
2351 vcpu->run->psw_mask = psw.mask;
2352 vcpu->run->psw_addr = psw.addr;
2357 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2358 struct kvm_translation *tr)
2360 return -EINVAL; /* not implemented yet */
2363 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
2364 KVM_GUESTDBG_USE_HW_BP | \
2365 KVM_GUESTDBG_ENABLE)
2367 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
2368 struct kvm_guest_debug *dbg)
2372 vcpu->guest_debug = 0;
2373 kvm_s390_clear_bp_data(vcpu);
2375 if (dbg->control & ~VALID_GUESTDBG_FLAGS)
2377 if (!sclp.has_gpere)
2380 if (dbg->control & KVM_GUESTDBG_ENABLE) {
2381 vcpu->guest_debug = dbg->control;
2382 /* enforce guest PER */
2383 atomic_or(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2385 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
2386 rc = kvm_s390_import_bp_data(vcpu, dbg);
2388 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2389 vcpu->arch.guestdbg.last_bp = 0;
2393 vcpu->guest_debug = 0;
2394 kvm_s390_clear_bp_data(vcpu);
2395 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2401 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2402 struct kvm_mp_state *mp_state)
2404 /* CHECK_STOP and LOAD are not supported yet */
2405 return is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
2406 KVM_MP_STATE_OPERATING;
2409 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2410 struct kvm_mp_state *mp_state)
2414 /* user space knows about this interface - let it control the state */
2415 vcpu->kvm->arch.user_cpu_state_ctrl = 1;
2417 switch (mp_state->mp_state) {
2418 case KVM_MP_STATE_STOPPED:
2419 kvm_s390_vcpu_stop(vcpu);
2421 case KVM_MP_STATE_OPERATING:
2422 kvm_s390_vcpu_start(vcpu);
2424 case KVM_MP_STATE_LOAD:
2425 case KVM_MP_STATE_CHECK_STOP:
2426 /* fall through - CHECK_STOP and LOAD are not supported yet */
2434 static bool ibs_enabled(struct kvm_vcpu *vcpu)
2436 return atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_IBS;
2439 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
2442 kvm_s390_vcpu_request_handled(vcpu);
2443 if (!vcpu->requests)
2446 * We use MMU_RELOAD just to re-arm the ipte notifier for the
2447 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
2448 * This ensures that the ipte instruction for this request has
2449 * already finished. We might race against a second unmapper that
2450 * wants to set the blocking bit. Lets just retry the request loop.
2452 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
2454 rc = gmap_mprotect_notify(vcpu->arch.gmap,
2455 kvm_s390_get_prefix(vcpu),
2456 PAGE_SIZE * 2, PROT_WRITE);
2458 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
2464 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2465 vcpu->arch.sie_block->ihcpu = 0xffff;
2469 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
2470 if (!ibs_enabled(vcpu)) {
2471 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
2472 atomic_or(CPUSTAT_IBS,
2473 &vcpu->arch.sie_block->cpuflags);
2478 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
2479 if (ibs_enabled(vcpu)) {
2480 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
2481 atomic_andnot(CPUSTAT_IBS,
2482 &vcpu->arch.sie_block->cpuflags);
2487 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
2488 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
2492 /* nothing to do, just clear the request */
2493 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
2498 void kvm_s390_set_tod_clock(struct kvm *kvm, u64 tod)
2500 struct kvm_vcpu *vcpu;
2503 mutex_lock(&kvm->lock);
2505 kvm->arch.epoch = tod - get_tod_clock();
2506 kvm_s390_vcpu_block_all(kvm);
2507 kvm_for_each_vcpu(i, vcpu, kvm)
2508 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
2509 kvm_s390_vcpu_unblock_all(kvm);
2511 mutex_unlock(&kvm->lock);
2515 * kvm_arch_fault_in_page - fault-in guest page if necessary
2516 * @vcpu: The corresponding virtual cpu
2517 * @gpa: Guest physical address
2518 * @writable: Whether the page should be writable or not
2520 * Make sure that a guest page has been faulted-in on the host.
2522 * Return: Zero on success, negative error code otherwise.
2524 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
2526 return gmap_fault(vcpu->arch.gmap, gpa,
2527 writable ? FAULT_FLAG_WRITE : 0);
2530 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
2531 unsigned long token)
2533 struct kvm_s390_interrupt inti;
2534 struct kvm_s390_irq irq;
2537 irq.u.ext.ext_params2 = token;
2538 irq.type = KVM_S390_INT_PFAULT_INIT;
2539 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
2541 inti.type = KVM_S390_INT_PFAULT_DONE;
2542 inti.parm64 = token;
2543 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
2547 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
2548 struct kvm_async_pf *work)
2550 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
2551 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
2554 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
2555 struct kvm_async_pf *work)
2557 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
2558 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
2561 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
2562 struct kvm_async_pf *work)
2564 /* s390 will always inject the page directly */
2567 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
2570 * s390 will always inject the page directly,
2571 * but we still want check_async_completion to cleanup
2576 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
2579 struct kvm_arch_async_pf arch;
2582 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2584 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
2585 vcpu->arch.pfault_compare)
2587 if (psw_extint_disabled(vcpu))
2589 if (kvm_s390_vcpu_has_irq(vcpu, 0))
2591 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
2593 if (!vcpu->arch.gmap->pfault_enabled)
2596 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
2597 hva += current->thread.gmap_addr & ~PAGE_MASK;
2598 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
2601 rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
2605 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
2610 * On s390 notifications for arriving pages will be delivered directly
2611 * to the guest but the house keeping for completed pfaults is
2612 * handled outside the worker.
2614 kvm_check_async_pf_completion(vcpu);
2616 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
2617 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
2622 if (test_cpu_flag(CIF_MCCK_PENDING))
2625 if (!kvm_is_ucontrol(vcpu->kvm)) {
2626 rc = kvm_s390_deliver_pending_interrupts(vcpu);
2631 rc = kvm_s390_handle_requests(vcpu);
2635 if (guestdbg_enabled(vcpu)) {
2636 kvm_s390_backup_guest_per_regs(vcpu);
2637 kvm_s390_patch_guest_per_regs(vcpu);
2640 vcpu->arch.sie_block->icptcode = 0;
2641 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
2642 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
2643 trace_kvm_s390_sie_enter(vcpu, cpuflags);
2648 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
2650 struct kvm_s390_pgm_info pgm_info = {
2651 .code = PGM_ADDRESSING,
2656 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
2657 trace_kvm_s390_sie_fault(vcpu);
2660 * We want to inject an addressing exception, which is defined as a
2661 * suppressing or terminating exception. However, since we came here
2662 * by a DAT access exception, the PSW still points to the faulting
2663 * instruction since DAT exceptions are nullifying. So we've got
2664 * to look up the current opcode to get the length of the instruction
2665 * to be able to forward the PSW.
2667 rc = read_guest_instr(vcpu, vcpu->arch.sie_block->gpsw.addr, &opcode, 1);
2668 ilen = insn_length(opcode);
2672 /* Instruction-Fetching Exceptions - we can't detect the ilen.
2673 * Forward by arbitrary ilc, injection will take care of
2674 * nullification if necessary.
2676 pgm_info = vcpu->arch.pgm;
2679 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
2680 kvm_s390_forward_psw(vcpu, ilen);
2681 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
2684 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
2686 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
2687 vcpu->arch.sie_block->icptcode);
2688 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
2690 if (guestdbg_enabled(vcpu))
2691 kvm_s390_restore_guest_per_regs(vcpu);
2693 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
2694 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
2696 if (vcpu->arch.sie_block->icptcode > 0) {
2697 int rc = kvm_handle_sie_intercept(vcpu);
2699 if (rc != -EOPNOTSUPP)
2701 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
2702 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
2703 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
2704 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
2706 } else if (exit_reason != -EFAULT) {
2707 vcpu->stat.exit_null++;
2709 } else if (kvm_is_ucontrol(vcpu->kvm)) {
2710 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
2711 vcpu->run->s390_ucontrol.trans_exc_code =
2712 current->thread.gmap_addr;
2713 vcpu->run->s390_ucontrol.pgm_code = 0x10;
2715 } else if (current->thread.gmap_pfault) {
2716 trace_kvm_s390_major_guest_pfault(vcpu);
2717 current->thread.gmap_pfault = 0;
2718 if (kvm_arch_setup_async_pf(vcpu))
2720 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
2722 return vcpu_post_run_fault_in_sie(vcpu);
2725 static int __vcpu_run(struct kvm_vcpu *vcpu)
2727 int rc, exit_reason;
2730 * We try to hold kvm->srcu during most of vcpu_run (except when run-
2731 * ning the guest), so that memslots (and other stuff) are protected
2733 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2736 rc = vcpu_pre_run(vcpu);
2740 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2742 * As PF_VCPU will be used in fault handler, between
2743 * guest_enter and guest_exit should be no uaccess.
2745 local_irq_disable();
2746 guest_enter_irqoff();
2747 __disable_cpu_timer_accounting(vcpu);
2749 exit_reason = sie64a(vcpu->arch.sie_block,
2750 vcpu->run->s.regs.gprs);
2751 local_irq_disable();
2752 __enable_cpu_timer_accounting(vcpu);
2753 guest_exit_irqoff();
2755 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2757 rc = vcpu_post_run(vcpu, exit_reason);
2758 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
2760 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2764 static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2766 struct runtime_instr_cb *riccb;
2769 riccb = (struct runtime_instr_cb *) &kvm_run->s.regs.riccb;
2770 gscb = (struct gs_cb *) &kvm_run->s.regs.gscb;
2771 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
2772 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
2773 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
2774 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
2775 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
2776 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
2777 /* some control register changes require a tlb flush */
2778 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2780 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
2781 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
2782 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
2783 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
2784 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
2785 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
2787 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
2788 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
2789 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
2790 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
2791 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2792 kvm_clear_async_pf_completion_queue(vcpu);
2795 * If userspace sets the riccb (e.g. after migration) to a valid state,
2796 * we should enable RI here instead of doing the lazy enablement.
2798 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) &&
2799 test_kvm_facility(vcpu->kvm, 64) &&
2801 !(vcpu->arch.sie_block->ecb3 & ECB3_RI)) {
2802 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (sync_regs)");
2803 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
2806 * If userspace sets the gscb (e.g. after migration) to non-zero,
2807 * we should enable GS here instead of doing the lazy enablement.
2809 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_GSCB) &&
2810 test_kvm_facility(vcpu->kvm, 133) &&
2812 !vcpu->arch.gs_enabled) {
2813 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (sync_regs)");
2814 vcpu->arch.sie_block->ecb |= ECB_GS;
2815 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
2816 vcpu->arch.gs_enabled = 1;
2818 save_access_regs(vcpu->arch.host_acrs);
2819 restore_access_regs(vcpu->run->s.regs.acrs);
2820 /* save host (userspace) fprs/vrs */
2822 vcpu->arch.host_fpregs.fpc = current->thread.fpu.fpc;
2823 vcpu->arch.host_fpregs.regs = current->thread.fpu.regs;
2825 current->thread.fpu.regs = vcpu->run->s.regs.vrs;
2827 current->thread.fpu.regs = vcpu->run->s.regs.fprs;
2828 current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
2829 if (test_fp_ctl(current->thread.fpu.fpc))
2830 /* User space provided an invalid FPC, let's clear it */
2831 current->thread.fpu.fpc = 0;
2832 if (MACHINE_HAS_GS) {
2834 __ctl_set_bit(2, 4);
2835 if (current->thread.gs_cb) {
2836 vcpu->arch.host_gscb = current->thread.gs_cb;
2837 save_gs_cb(vcpu->arch.host_gscb);
2839 if (vcpu->arch.gs_enabled) {
2840 current->thread.gs_cb = (struct gs_cb *)
2841 &vcpu->run->s.regs.gscb;
2842 restore_gs_cb(current->thread.gs_cb);
2847 kvm_run->kvm_dirty_regs = 0;
2850 static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2852 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
2853 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
2854 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
2855 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
2856 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
2857 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
2858 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
2859 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
2860 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
2861 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
2862 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
2863 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
2864 save_access_regs(vcpu->run->s.regs.acrs);
2865 restore_access_regs(vcpu->arch.host_acrs);
2866 /* Save guest register state */
2868 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
2869 /* Restore will be done lazily at return */
2870 current->thread.fpu.fpc = vcpu->arch.host_fpregs.fpc;
2871 current->thread.fpu.regs = vcpu->arch.host_fpregs.regs;
2872 if (MACHINE_HAS_GS) {
2873 __ctl_set_bit(2, 4);
2874 if (vcpu->arch.gs_enabled)
2875 save_gs_cb(current->thread.gs_cb);
2877 current->thread.gs_cb = vcpu->arch.host_gscb;
2878 restore_gs_cb(vcpu->arch.host_gscb);
2880 if (!vcpu->arch.host_gscb)
2881 __ctl_clear_bit(2, 4);
2882 vcpu->arch.host_gscb = NULL;
2887 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2892 if (kvm_run->immediate_exit)
2895 if (guestdbg_exit_pending(vcpu)) {
2896 kvm_s390_prepare_debug_exit(vcpu);
2900 if (vcpu->sigset_active)
2901 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2903 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
2904 kvm_s390_vcpu_start(vcpu);
2905 } else if (is_vcpu_stopped(vcpu)) {
2906 pr_err_ratelimited("can't run stopped vcpu %d\n",
2911 sync_regs(vcpu, kvm_run);
2912 enable_cpu_timer_accounting(vcpu);
2915 rc = __vcpu_run(vcpu);
2917 if (signal_pending(current) && !rc) {
2918 kvm_run->exit_reason = KVM_EXIT_INTR;
2922 if (guestdbg_exit_pending(vcpu) && !rc) {
2923 kvm_s390_prepare_debug_exit(vcpu);
2927 if (rc == -EREMOTE) {
2928 /* userspace support is needed, kvm_run has been prepared */
2932 disable_cpu_timer_accounting(vcpu);
2933 store_regs(vcpu, kvm_run);
2935 if (vcpu->sigset_active)
2936 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2938 vcpu->stat.exit_userspace++;
2943 * store status at address
2944 * we use have two special cases:
2945 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
2946 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
2948 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
2950 unsigned char archmode = 1;
2951 freg_t fprs[NUM_FPRS];
2956 px = kvm_s390_get_prefix(vcpu);
2957 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
2958 if (write_guest_abs(vcpu, 163, &archmode, 1))
2961 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
2962 if (write_guest_real(vcpu, 163, &archmode, 1))
2966 gpa -= __LC_FPREGS_SAVE_AREA;
2968 /* manually convert vector registers if necessary */
2969 if (MACHINE_HAS_VX) {
2970 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
2971 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2974 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2975 vcpu->run->s.regs.fprs, 128);
2977 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
2978 vcpu->run->s.regs.gprs, 128);
2979 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
2980 &vcpu->arch.sie_block->gpsw, 16);
2981 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
2983 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
2984 &vcpu->run->s.regs.fpc, 4);
2985 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
2986 &vcpu->arch.sie_block->todpr, 4);
2987 cputm = kvm_s390_get_cpu_timer(vcpu);
2988 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
2990 clkcomp = vcpu->arch.sie_block->ckc >> 8;
2991 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
2993 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
2994 &vcpu->run->s.regs.acrs, 64);
2995 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
2996 &vcpu->arch.sie_block->gcr, 128);
2997 return rc ? -EFAULT : 0;
3000 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
3003 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
3004 * switch in the run ioctl. Let's update our copies before we save
3005 * it into the save area
3008 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
3009 save_access_regs(vcpu->run->s.regs.acrs);
3011 return kvm_s390_store_status_unloaded(vcpu, addr);
3014 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
3016 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
3017 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
3020 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
3023 struct kvm_vcpu *vcpu;
3025 kvm_for_each_vcpu(i, vcpu, kvm) {
3026 __disable_ibs_on_vcpu(vcpu);
3030 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
3034 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
3035 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
3038 void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
3040 int i, online_vcpus, started_vcpus = 0;
3042 if (!is_vcpu_stopped(vcpu))
3045 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
3046 /* Only one cpu at a time may enter/leave the STOPPED state. */
3047 spin_lock(&vcpu->kvm->arch.start_stop_lock);
3048 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
3050 for (i = 0; i < online_vcpus; i++) {
3051 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i]))
3055 if (started_vcpus == 0) {
3056 /* we're the only active VCPU -> speed it up */
3057 __enable_ibs_on_vcpu(vcpu);
3058 } else if (started_vcpus == 1) {
3060 * As we are starting a second VCPU, we have to disable
3061 * the IBS facility on all VCPUs to remove potentially
3062 * oustanding ENABLE requests.
3064 __disable_ibs_on_all_vcpus(vcpu->kvm);
3067 atomic_andnot(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
3069 * Another VCPU might have used IBS while we were offline.
3070 * Let's play safe and flush the VCPU at startup.
3072 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
3073 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
3077 void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
3079 int i, online_vcpus, started_vcpus = 0;
3080 struct kvm_vcpu *started_vcpu = NULL;
3082 if (is_vcpu_stopped(vcpu))
3085 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
3086 /* Only one cpu at a time may enter/leave the STOPPED state. */
3087 spin_lock(&vcpu->kvm->arch.start_stop_lock);
3088 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
3090 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
3091 kvm_s390_clear_stop_irq(vcpu);
3093 atomic_or(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
3094 __disable_ibs_on_vcpu(vcpu);
3096 for (i = 0; i < online_vcpus; i++) {
3097 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) {
3099 started_vcpu = vcpu->kvm->vcpus[i];
3103 if (started_vcpus == 1) {
3105 * As we only have one VCPU left, we want to enable the
3106 * IBS facility for that VCPU to speed it up.
3108 __enable_ibs_on_vcpu(started_vcpu);
3111 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
3115 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3116 struct kvm_enable_cap *cap)
3124 case KVM_CAP_S390_CSS_SUPPORT:
3125 if (!vcpu->kvm->arch.css_support) {
3126 vcpu->kvm->arch.css_support = 1;
3127 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
3128 trace_kvm_s390_enable_css(vcpu->kvm);
3139 static long kvm_s390_guest_mem_op(struct kvm_vcpu *vcpu,
3140 struct kvm_s390_mem_op *mop)
3142 void __user *uaddr = (void __user *)mop->buf;
3143 void *tmpbuf = NULL;
3145 const u64 supported_flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION
3146 | KVM_S390_MEMOP_F_CHECK_ONLY;
3148 if (mop->flags & ~supported_flags)
3151 if (mop->size > MEM_OP_MAX_SIZE)
3154 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
3155 tmpbuf = vmalloc(mop->size);
3160 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3163 case KVM_S390_MEMOP_LOGICAL_READ:
3164 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3165 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3166 mop->size, GACC_FETCH);
3169 r = read_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3171 if (copy_to_user(uaddr, tmpbuf, mop->size))
3175 case KVM_S390_MEMOP_LOGICAL_WRITE:
3176 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3177 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3178 mop->size, GACC_STORE);
3181 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
3185 r = write_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3191 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
3193 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
3194 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
3200 long kvm_arch_vcpu_ioctl(struct file *filp,
3201 unsigned int ioctl, unsigned long arg)
3203 struct kvm_vcpu *vcpu = filp->private_data;
3204 void __user *argp = (void __user *)arg;
3209 case KVM_S390_IRQ: {
3210 struct kvm_s390_irq s390irq;
3213 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
3215 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
3218 case KVM_S390_INTERRUPT: {
3219 struct kvm_s390_interrupt s390int;
3220 struct kvm_s390_irq s390irq;
3223 if (copy_from_user(&s390int, argp, sizeof(s390int)))
3225 if (s390int_to_s390irq(&s390int, &s390irq))
3227 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
3230 case KVM_S390_STORE_STATUS:
3231 idx = srcu_read_lock(&vcpu->kvm->srcu);
3232 r = kvm_s390_vcpu_store_status(vcpu, arg);
3233 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3235 case KVM_S390_SET_INITIAL_PSW: {
3239 if (copy_from_user(&psw, argp, sizeof(psw)))
3241 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
3244 case KVM_S390_INITIAL_RESET:
3245 r = kvm_arch_vcpu_ioctl_initial_reset(vcpu);
3247 case KVM_SET_ONE_REG:
3248 case KVM_GET_ONE_REG: {
3249 struct kvm_one_reg reg;
3251 if (copy_from_user(®, argp, sizeof(reg)))
3253 if (ioctl == KVM_SET_ONE_REG)
3254 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®);
3256 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®);
3259 #ifdef CONFIG_KVM_S390_UCONTROL
3260 case KVM_S390_UCAS_MAP: {
3261 struct kvm_s390_ucas_mapping ucasmap;
3263 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3268 if (!kvm_is_ucontrol(vcpu->kvm)) {
3273 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
3274 ucasmap.vcpu_addr, ucasmap.length);
3277 case KVM_S390_UCAS_UNMAP: {
3278 struct kvm_s390_ucas_mapping ucasmap;
3280 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3285 if (!kvm_is_ucontrol(vcpu->kvm)) {
3290 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
3295 case KVM_S390_VCPU_FAULT: {
3296 r = gmap_fault(vcpu->arch.gmap, arg, 0);
3299 case KVM_ENABLE_CAP:
3301 struct kvm_enable_cap cap;
3303 if (copy_from_user(&cap, argp, sizeof(cap)))
3305 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3308 case KVM_S390_MEM_OP: {
3309 struct kvm_s390_mem_op mem_op;
3311 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
3312 r = kvm_s390_guest_mem_op(vcpu, &mem_op);
3317 case KVM_S390_SET_IRQ_STATE: {
3318 struct kvm_s390_irq_state irq_state;
3321 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3323 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
3324 irq_state.len == 0 ||
3325 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
3329 r = kvm_s390_set_irq_state(vcpu,
3330 (void __user *) irq_state.buf,
3334 case KVM_S390_GET_IRQ_STATE: {
3335 struct kvm_s390_irq_state irq_state;
3338 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3340 if (irq_state.len == 0) {
3344 r = kvm_s390_get_irq_state(vcpu,
3345 (__u8 __user *) irq_state.buf,
3355 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3357 #ifdef CONFIG_KVM_S390_UCONTROL
3358 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
3359 && (kvm_is_ucontrol(vcpu->kvm))) {
3360 vmf->page = virt_to_page(vcpu->arch.sie_block);
3361 get_page(vmf->page);
3365 return VM_FAULT_SIGBUS;
3368 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
3369 unsigned long npages)
3374 /* Section: memory related */
3375 int kvm_arch_prepare_memory_region(struct kvm *kvm,
3376 struct kvm_memory_slot *memslot,
3377 const struct kvm_userspace_memory_region *mem,
3378 enum kvm_mr_change change)
3380 /* A few sanity checks. We can have memory slots which have to be
3381 located/ended at a segment boundary (1MB). The memory in userland is
3382 ok to be fragmented into various different vmas. It is okay to mmap()
3383 and munmap() stuff in this slot after doing this call at any time */
3385 if (mem->userspace_addr & 0xffffful)
3388 if (mem->memory_size & 0xffffful)
3391 if (mem->guest_phys_addr + mem->memory_size > kvm->arch.mem_limit)
3397 void kvm_arch_commit_memory_region(struct kvm *kvm,
3398 const struct kvm_userspace_memory_region *mem,
3399 const struct kvm_memory_slot *old,
3400 const struct kvm_memory_slot *new,
3401 enum kvm_mr_change change)
3405 /* If the basics of the memslot do not change, we do not want
3406 * to update the gmap. Every update causes several unnecessary
3407 * segment translation exceptions. This is usually handled just
3408 * fine by the normal fault handler + gmap, but it will also
3409 * cause faults on the prefix page of running guest CPUs.
3411 if (old->userspace_addr == mem->userspace_addr &&
3412 old->base_gfn * PAGE_SIZE == mem->guest_phys_addr &&
3413 old->npages * PAGE_SIZE == mem->memory_size)
3416 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr,
3417 mem->guest_phys_addr, mem->memory_size);
3419 pr_warn("failed to commit memory region\n");
3423 static inline unsigned long nonhyp_mask(int i)
3425 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
3427 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
3430 void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu)
3432 vcpu->valid_wakeup = false;
3435 static int __init kvm_s390_init(void)
3439 if (!sclp.has_sief2) {
3440 pr_info("SIE not available\n");
3444 for (i = 0; i < 16; i++)
3445 kvm_s390_fac_list_mask[i] |=
3446 S390_lowcore.stfle_fac_list[i] & nonhyp_mask(i);
3448 return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
3451 static void __exit kvm_s390_exit(void)
3456 module_init(kvm_s390_init);
3457 module_exit(kvm_s390_exit);
3460 * Enable autoloading of the kvm module.
3461 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
3462 * since x86 takes a different approach.
3464 #include <linux/miscdevice.h>
3465 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3466 MODULE_ALIAS("devname:kvm");