1 // SPDX-License-Identifier: GPL-2.0
3 * hosting IBM Z kernel virtual machines (s390x)
5 * Copyright IBM Corp. 2008, 2018
7 * Author(s): Carsten Otte <cotte@de.ibm.com>
8 * Christian Borntraeger <borntraeger@de.ibm.com>
9 * Heiko Carstens <heiko.carstens@de.ibm.com>
10 * Christian Ehrhardt <ehrhardt@de.ibm.com>
11 * Jason J. Herne <jjherne@us.ibm.com>
14 #include <linux/compiler.h>
15 #include <linux/err.h>
17 #include <linux/hrtimer.h>
18 #include <linux/init.h>
19 #include <linux/kvm.h>
20 #include <linux/kvm_host.h>
21 #include <linux/mman.h>
22 #include <linux/module.h>
23 #include <linux/moduleparam.h>
24 #include <linux/random.h>
25 #include <linux/slab.h>
26 #include <linux/timer.h>
27 #include <linux/vmalloc.h>
28 #include <linux/bitmap.h>
29 #include <linux/sched/signal.h>
30 #include <linux/string.h>
32 #include <asm/asm-offsets.h>
33 #include <asm/lowcore.h>
35 #include <asm/pgtable.h>
38 #include <asm/switch_to.h>
41 #include <asm/cpacf.h>
42 #include <asm/timex.h>
46 #define KMSG_COMPONENT "kvm-s390"
48 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
50 #define CREATE_TRACE_POINTS
52 #include "trace-s390.h"
54 #define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */
56 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
57 (KVM_MAX_VCPUS + LOCAL_IRQS))
59 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
60 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
62 struct kvm_stats_debugfs_item debugfs_entries[] = {
63 { "userspace_handled", VCPU_STAT(exit_userspace) },
64 { "exit_null", VCPU_STAT(exit_null) },
65 { "exit_validity", VCPU_STAT(exit_validity) },
66 { "exit_stop_request", VCPU_STAT(exit_stop_request) },
67 { "exit_external_request", VCPU_STAT(exit_external_request) },
68 { "exit_io_request", VCPU_STAT(exit_io_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_ckc", VCPU_STAT(deliver_ckc) },
84 { "deliver_cputm", VCPU_STAT(deliver_cputm) },
85 { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal) },
86 { "deliver_external_call", VCPU_STAT(deliver_external_call) },
87 { "deliver_service_signal", VCPU_STAT(deliver_service_signal) },
88 { "deliver_virtio", VCPU_STAT(deliver_virtio) },
89 { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal) },
90 { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal) },
91 { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal) },
92 { "deliver_program", VCPU_STAT(deliver_program) },
93 { "deliver_io", VCPU_STAT(deliver_io) },
94 { "deliver_machine_check", VCPU_STAT(deliver_machine_check) },
95 { "exit_wait_state", VCPU_STAT(exit_wait_state) },
96 { "inject_ckc", VCPU_STAT(inject_ckc) },
97 { "inject_cputm", VCPU_STAT(inject_cputm) },
98 { "inject_external_call", VCPU_STAT(inject_external_call) },
99 { "inject_float_mchk", VM_STAT(inject_float_mchk) },
100 { "inject_emergency_signal", VCPU_STAT(inject_emergency_signal) },
101 { "inject_io", VM_STAT(inject_io) },
102 { "inject_mchk", VCPU_STAT(inject_mchk) },
103 { "inject_pfault_done", VM_STAT(inject_pfault_done) },
104 { "inject_program", VCPU_STAT(inject_program) },
105 { "inject_restart", VCPU_STAT(inject_restart) },
106 { "inject_service_signal", VM_STAT(inject_service_signal) },
107 { "inject_set_prefix", VCPU_STAT(inject_set_prefix) },
108 { "inject_stop_signal", VCPU_STAT(inject_stop_signal) },
109 { "inject_pfault_init", VCPU_STAT(inject_pfault_init) },
110 { "inject_virtio", VM_STAT(inject_virtio) },
111 { "instruction_epsw", VCPU_STAT(instruction_epsw) },
112 { "instruction_gs", VCPU_STAT(instruction_gs) },
113 { "instruction_io_other", VCPU_STAT(instruction_io_other) },
114 { "instruction_lpsw", VCPU_STAT(instruction_lpsw) },
115 { "instruction_lpswe", VCPU_STAT(instruction_lpswe) },
116 { "instruction_pfmf", VCPU_STAT(instruction_pfmf) },
117 { "instruction_ptff", VCPU_STAT(instruction_ptff) },
118 { "instruction_stidp", VCPU_STAT(instruction_stidp) },
119 { "instruction_sck", VCPU_STAT(instruction_sck) },
120 { "instruction_sckpf", VCPU_STAT(instruction_sckpf) },
121 { "instruction_spx", VCPU_STAT(instruction_spx) },
122 { "instruction_stpx", VCPU_STAT(instruction_stpx) },
123 { "instruction_stap", VCPU_STAT(instruction_stap) },
124 { "instruction_iske", VCPU_STAT(instruction_iske) },
125 { "instruction_ri", VCPU_STAT(instruction_ri) },
126 { "instruction_rrbe", VCPU_STAT(instruction_rrbe) },
127 { "instruction_sske", VCPU_STAT(instruction_sske) },
128 { "instruction_ipte_interlock", VCPU_STAT(instruction_ipte_interlock) },
129 { "instruction_essa", VCPU_STAT(instruction_essa) },
130 { "instruction_stsi", VCPU_STAT(instruction_stsi) },
131 { "instruction_stfl", VCPU_STAT(instruction_stfl) },
132 { "instruction_tb", VCPU_STAT(instruction_tb) },
133 { "instruction_tpi", VCPU_STAT(instruction_tpi) },
134 { "instruction_tprot", VCPU_STAT(instruction_tprot) },
135 { "instruction_tsch", VCPU_STAT(instruction_tsch) },
136 { "instruction_sthyi", VCPU_STAT(instruction_sthyi) },
137 { "instruction_sie", VCPU_STAT(instruction_sie) },
138 { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense) },
139 { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running) },
140 { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call) },
141 { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency) },
142 { "instruction_sigp_cond_emergency", VCPU_STAT(instruction_sigp_cond_emergency) },
143 { "instruction_sigp_start", VCPU_STAT(instruction_sigp_start) },
144 { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop) },
145 { "instruction_sigp_stop_store_status", VCPU_STAT(instruction_sigp_stop_store_status) },
146 { "instruction_sigp_store_status", VCPU_STAT(instruction_sigp_store_status) },
147 { "instruction_sigp_store_adtl_status", VCPU_STAT(instruction_sigp_store_adtl_status) },
148 { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch) },
149 { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix) },
150 { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart) },
151 { "instruction_sigp_cpu_reset", VCPU_STAT(instruction_sigp_cpu_reset) },
152 { "instruction_sigp_init_cpu_reset", VCPU_STAT(instruction_sigp_init_cpu_reset) },
153 { "instruction_sigp_unknown", VCPU_STAT(instruction_sigp_unknown) },
154 { "instruction_diag_10", VCPU_STAT(diagnose_10) },
155 { "instruction_diag_44", VCPU_STAT(diagnose_44) },
156 { "instruction_diag_9c", VCPU_STAT(diagnose_9c) },
157 { "instruction_diag_258", VCPU_STAT(diagnose_258) },
158 { "instruction_diag_308", VCPU_STAT(diagnose_308) },
159 { "instruction_diag_500", VCPU_STAT(diagnose_500) },
160 { "instruction_diag_other", VCPU_STAT(diagnose_other) },
164 struct kvm_s390_tod_clock_ext {
170 /* allow nested virtualization in KVM (if enabled by user space) */
172 module_param(nested, int, S_IRUGO);
173 MODULE_PARM_DESC(nested, "Nested virtualization support");
177 * For now we handle at most 16 double words as this is what the s390 base
178 * kernel handles and stores in the prefix page. If we ever need to go beyond
179 * this, this requires changes to code, but the external uapi can stay.
181 #define SIZE_INTERNAL 16
184 * Base feature mask that defines default mask for facilities. Consists of the
185 * defines in FACILITIES_KVM and the non-hypervisor managed bits.
187 static unsigned long kvm_s390_fac_base[SIZE_INTERNAL] = { FACILITIES_KVM };
189 * Extended feature mask. Consists of the defines in FACILITIES_KVM_CPUMODEL
190 * and defines the facilities that can be enabled via a cpu model.
192 static unsigned long kvm_s390_fac_ext[SIZE_INTERNAL] = { FACILITIES_KVM_CPUMODEL };
194 static unsigned long kvm_s390_fac_size(void)
196 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_MASK_SIZE_U64);
197 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_LIST_SIZE_U64);
198 BUILD_BUG_ON(SIZE_INTERNAL * sizeof(unsigned long) >
199 sizeof(S390_lowcore.stfle_fac_list));
201 return SIZE_INTERNAL;
204 /* available cpu features supported by kvm */
205 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
206 /* available subfunctions indicated via query / "test bit" */
207 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
209 static struct gmap_notifier gmap_notifier;
210 static struct gmap_notifier vsie_gmap_notifier;
211 debug_info_t *kvm_s390_dbf;
213 /* Section: not file related */
214 int kvm_arch_hardware_enable(void)
216 /* every s390 is virtualization enabled ;-) */
220 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
223 static void kvm_clock_sync_scb(struct kvm_s390_sie_block *scb, u64 delta)
228 * The TOD jumps by delta, we have to compensate this by adding
229 * -delta to the epoch.
233 /* sign-extension - we're adding to signed values below */
238 if (scb->ecd & ECD_MEF) {
239 scb->epdx += delta_idx;
240 if (scb->epoch < delta)
246 * This callback is executed during stop_machine(). All CPUs are therefore
247 * temporarily stopped. In order not to change guest behavior, we have to
248 * disable preemption whenever we touch the epoch of kvm and the VCPUs,
249 * so a CPU won't be stopped while calculating with the epoch.
251 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
255 struct kvm_vcpu *vcpu;
257 unsigned long long *delta = v;
259 list_for_each_entry(kvm, &vm_list, vm_list) {
260 kvm_for_each_vcpu(i, vcpu, kvm) {
261 kvm_clock_sync_scb(vcpu->arch.sie_block, *delta);
263 kvm->arch.epoch = vcpu->arch.sie_block->epoch;
264 kvm->arch.epdx = vcpu->arch.sie_block->epdx;
266 if (vcpu->arch.cputm_enabled)
267 vcpu->arch.cputm_start += *delta;
268 if (vcpu->arch.vsie_block)
269 kvm_clock_sync_scb(vcpu->arch.vsie_block,
276 static struct notifier_block kvm_clock_notifier = {
277 .notifier_call = kvm_clock_sync,
280 int kvm_arch_hardware_setup(void)
282 gmap_notifier.notifier_call = kvm_gmap_notifier;
283 gmap_register_pte_notifier(&gmap_notifier);
284 vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
285 gmap_register_pte_notifier(&vsie_gmap_notifier);
286 atomic_notifier_chain_register(&s390_epoch_delta_notifier,
287 &kvm_clock_notifier);
291 void kvm_arch_hardware_unsetup(void)
293 gmap_unregister_pte_notifier(&gmap_notifier);
294 gmap_unregister_pte_notifier(&vsie_gmap_notifier);
295 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
296 &kvm_clock_notifier);
299 static void allow_cpu_feat(unsigned long nr)
301 set_bit_inv(nr, kvm_s390_available_cpu_feat);
304 static inline int plo_test_bit(unsigned char nr)
306 register unsigned long r0 asm("0") = (unsigned long) nr | 0x100;
310 /* Parameter registers are ignored for "test bit" */
320 static void kvm_s390_cpu_feat_init(void)
324 for (i = 0; i < 256; ++i) {
326 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
329 if (test_facility(28)) /* TOD-clock steering */
330 ptff(kvm_s390_available_subfunc.ptff,
331 sizeof(kvm_s390_available_subfunc.ptff),
334 if (test_facility(17)) { /* MSA */
335 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
336 kvm_s390_available_subfunc.kmac);
337 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
338 kvm_s390_available_subfunc.kmc);
339 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
340 kvm_s390_available_subfunc.km);
341 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
342 kvm_s390_available_subfunc.kimd);
343 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
344 kvm_s390_available_subfunc.klmd);
346 if (test_facility(76)) /* MSA3 */
347 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
348 kvm_s390_available_subfunc.pckmo);
349 if (test_facility(77)) { /* MSA4 */
350 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
351 kvm_s390_available_subfunc.kmctr);
352 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
353 kvm_s390_available_subfunc.kmf);
354 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
355 kvm_s390_available_subfunc.kmo);
356 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
357 kvm_s390_available_subfunc.pcc);
359 if (test_facility(57)) /* MSA5 */
360 __cpacf_query(CPACF_PRNO, (cpacf_mask_t *)
361 kvm_s390_available_subfunc.ppno);
363 if (test_facility(146)) /* MSA8 */
364 __cpacf_query(CPACF_KMA, (cpacf_mask_t *)
365 kvm_s390_available_subfunc.kma);
367 if (MACHINE_HAS_ESOP)
368 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
370 * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
371 * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
373 if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
374 !test_facility(3) || !nested)
376 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
377 if (sclp.has_64bscao)
378 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
380 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
382 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
384 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
386 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
388 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
390 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
392 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_KSS);
394 * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
395 * all skey handling functions read/set the skey from the PGSTE
396 * instead of the real storage key.
398 * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
399 * pages being detected as preserved although they are resident.
401 * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
402 * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
404 * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
405 * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
406 * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
408 * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
409 * cannot easily shadow the SCA because of the ipte lock.
413 int kvm_arch_init(void *opaque)
415 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
419 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view)) {
420 debug_unregister(kvm_s390_dbf);
424 kvm_s390_cpu_feat_init();
426 /* Register floating interrupt controller interface. */
427 return kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
430 void kvm_arch_exit(void)
432 debug_unregister(kvm_s390_dbf);
435 /* Section: device related */
436 long kvm_arch_dev_ioctl(struct file *filp,
437 unsigned int ioctl, unsigned long arg)
439 if (ioctl == KVM_S390_ENABLE_SIE)
440 return s390_enable_sie();
444 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
449 case KVM_CAP_S390_PSW:
450 case KVM_CAP_S390_GMAP:
451 case KVM_CAP_SYNC_MMU:
452 #ifdef CONFIG_KVM_S390_UCONTROL
453 case KVM_CAP_S390_UCONTROL:
455 case KVM_CAP_ASYNC_PF:
456 case KVM_CAP_SYNC_REGS:
457 case KVM_CAP_ONE_REG:
458 case KVM_CAP_ENABLE_CAP:
459 case KVM_CAP_S390_CSS_SUPPORT:
460 case KVM_CAP_IOEVENTFD:
461 case KVM_CAP_DEVICE_CTRL:
462 case KVM_CAP_ENABLE_CAP_VM:
463 case KVM_CAP_S390_IRQCHIP:
464 case KVM_CAP_VM_ATTRIBUTES:
465 case KVM_CAP_MP_STATE:
466 case KVM_CAP_IMMEDIATE_EXIT:
467 case KVM_CAP_S390_INJECT_IRQ:
468 case KVM_CAP_S390_USER_SIGP:
469 case KVM_CAP_S390_USER_STSI:
470 case KVM_CAP_S390_SKEYS:
471 case KVM_CAP_S390_IRQ_STATE:
472 case KVM_CAP_S390_USER_INSTR0:
473 case KVM_CAP_S390_CMMA_MIGRATION:
474 case KVM_CAP_S390_AIS:
475 case KVM_CAP_S390_AIS_MIGRATION:
478 case KVM_CAP_S390_MEM_OP:
481 case KVM_CAP_NR_VCPUS:
482 case KVM_CAP_MAX_VCPUS:
483 r = KVM_S390_BSCA_CPU_SLOTS;
484 if (!kvm_s390_use_sca_entries())
486 else if (sclp.has_esca && sclp.has_64bscao)
487 r = KVM_S390_ESCA_CPU_SLOTS;
489 case KVM_CAP_NR_MEMSLOTS:
490 r = KVM_USER_MEM_SLOTS;
492 case KVM_CAP_S390_COW:
493 r = MACHINE_HAS_ESOP;
495 case KVM_CAP_S390_VECTOR_REGISTERS:
498 case KVM_CAP_S390_RI:
499 r = test_facility(64);
501 case KVM_CAP_S390_GS:
502 r = test_facility(133);
504 case KVM_CAP_S390_BPB:
505 r = test_facility(82);
513 static void kvm_s390_sync_dirty_log(struct kvm *kvm,
514 struct kvm_memory_slot *memslot)
516 gfn_t cur_gfn, last_gfn;
517 unsigned long address;
518 struct gmap *gmap = kvm->arch.gmap;
520 /* Loop over all guest pages */
521 last_gfn = memslot->base_gfn + memslot->npages;
522 for (cur_gfn = memslot->base_gfn; cur_gfn <= last_gfn; cur_gfn++) {
523 address = gfn_to_hva_memslot(memslot, cur_gfn);
525 if (test_and_clear_guest_dirty(gmap->mm, address))
526 mark_page_dirty(kvm, cur_gfn);
527 if (fatal_signal_pending(current))
533 /* Section: vm related */
534 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
537 * Get (and clear) the dirty memory log for a memory slot.
539 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
540 struct kvm_dirty_log *log)
544 struct kvm_memslots *slots;
545 struct kvm_memory_slot *memslot;
548 if (kvm_is_ucontrol(kvm))
551 mutex_lock(&kvm->slots_lock);
554 if (log->slot >= KVM_USER_MEM_SLOTS)
557 slots = kvm_memslots(kvm);
558 memslot = id_to_memslot(slots, log->slot);
560 if (!memslot->dirty_bitmap)
563 kvm_s390_sync_dirty_log(kvm, memslot);
564 r = kvm_get_dirty_log(kvm, log, &is_dirty);
568 /* Clear the dirty log */
570 n = kvm_dirty_bitmap_bytes(memslot);
571 memset(memslot->dirty_bitmap, 0, n);
575 mutex_unlock(&kvm->slots_lock);
579 static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
582 struct kvm_vcpu *vcpu;
584 kvm_for_each_vcpu(i, vcpu, kvm) {
585 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
589 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
597 case KVM_CAP_S390_IRQCHIP:
598 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
599 kvm->arch.use_irqchip = 1;
602 case KVM_CAP_S390_USER_SIGP:
603 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
604 kvm->arch.user_sigp = 1;
607 case KVM_CAP_S390_VECTOR_REGISTERS:
608 mutex_lock(&kvm->lock);
609 if (kvm->created_vcpus) {
611 } else if (MACHINE_HAS_VX) {
612 set_kvm_facility(kvm->arch.model.fac_mask, 129);
613 set_kvm_facility(kvm->arch.model.fac_list, 129);
614 if (test_facility(134)) {
615 set_kvm_facility(kvm->arch.model.fac_mask, 134);
616 set_kvm_facility(kvm->arch.model.fac_list, 134);
618 if (test_facility(135)) {
619 set_kvm_facility(kvm->arch.model.fac_mask, 135);
620 set_kvm_facility(kvm->arch.model.fac_list, 135);
625 mutex_unlock(&kvm->lock);
626 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
627 r ? "(not available)" : "(success)");
629 case KVM_CAP_S390_RI:
631 mutex_lock(&kvm->lock);
632 if (kvm->created_vcpus) {
634 } else if (test_facility(64)) {
635 set_kvm_facility(kvm->arch.model.fac_mask, 64);
636 set_kvm_facility(kvm->arch.model.fac_list, 64);
639 mutex_unlock(&kvm->lock);
640 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
641 r ? "(not available)" : "(success)");
643 case KVM_CAP_S390_AIS:
644 mutex_lock(&kvm->lock);
645 if (kvm->created_vcpus) {
648 set_kvm_facility(kvm->arch.model.fac_mask, 72);
649 set_kvm_facility(kvm->arch.model.fac_list, 72);
652 mutex_unlock(&kvm->lock);
653 VM_EVENT(kvm, 3, "ENABLE: AIS %s",
654 r ? "(not available)" : "(success)");
656 case KVM_CAP_S390_GS:
658 mutex_lock(&kvm->lock);
659 if (kvm->created_vcpus) {
661 } else if (test_facility(133)) {
662 set_kvm_facility(kvm->arch.model.fac_mask, 133);
663 set_kvm_facility(kvm->arch.model.fac_list, 133);
666 mutex_unlock(&kvm->lock);
667 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_GS %s",
668 r ? "(not available)" : "(success)");
670 case KVM_CAP_S390_USER_STSI:
671 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
672 kvm->arch.user_stsi = 1;
675 case KVM_CAP_S390_USER_INSTR0:
676 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
677 kvm->arch.user_instr0 = 1;
678 icpt_operexc_on_all_vcpus(kvm);
688 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
692 switch (attr->attr) {
693 case KVM_S390_VM_MEM_LIMIT_SIZE:
695 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
696 kvm->arch.mem_limit);
697 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
707 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
711 switch (attr->attr) {
712 case KVM_S390_VM_MEM_ENABLE_CMMA:
718 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
719 mutex_lock(&kvm->lock);
720 if (!kvm->created_vcpus) {
721 kvm->arch.use_cmma = 1;
722 /* Not compatible with cmma. */
723 kvm->arch.use_pfmfi = 0;
726 mutex_unlock(&kvm->lock);
728 case KVM_S390_VM_MEM_CLR_CMMA:
733 if (!kvm->arch.use_cmma)
736 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
737 mutex_lock(&kvm->lock);
738 idx = srcu_read_lock(&kvm->srcu);
739 s390_reset_cmma(kvm->arch.gmap->mm);
740 srcu_read_unlock(&kvm->srcu, idx);
741 mutex_unlock(&kvm->lock);
744 case KVM_S390_VM_MEM_LIMIT_SIZE: {
745 unsigned long new_limit;
747 if (kvm_is_ucontrol(kvm))
750 if (get_user(new_limit, (u64 __user *)attr->addr))
753 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
754 new_limit > kvm->arch.mem_limit)
760 /* gmap_create takes last usable address */
761 if (new_limit != KVM_S390_NO_MEM_LIMIT)
765 mutex_lock(&kvm->lock);
766 if (!kvm->created_vcpus) {
767 /* gmap_create will round the limit up */
768 struct gmap *new = gmap_create(current->mm, new_limit);
773 gmap_remove(kvm->arch.gmap);
775 kvm->arch.gmap = new;
779 mutex_unlock(&kvm->lock);
780 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
781 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
782 (void *) kvm->arch.gmap->asce);
792 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
794 void kvm_s390_vcpu_crypto_reset_all(struct kvm *kvm)
796 struct kvm_vcpu *vcpu;
799 kvm_s390_vcpu_block_all(kvm);
801 kvm_for_each_vcpu(i, vcpu, kvm)
802 kvm_s390_vcpu_crypto_setup(vcpu);
804 kvm_s390_vcpu_unblock_all(kvm);
807 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
809 if (!test_kvm_facility(kvm, 76))
812 mutex_lock(&kvm->lock);
813 switch (attr->attr) {
814 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
816 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
817 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
818 kvm->arch.crypto.aes_kw = 1;
819 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
821 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
823 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
824 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
825 kvm->arch.crypto.dea_kw = 1;
826 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
828 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
829 kvm->arch.crypto.aes_kw = 0;
830 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
831 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
832 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
834 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
835 kvm->arch.crypto.dea_kw = 0;
836 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
837 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
838 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
841 mutex_unlock(&kvm->lock);
845 kvm_s390_vcpu_crypto_reset_all(kvm);
846 mutex_unlock(&kvm->lock);
850 static void kvm_s390_sync_request_broadcast(struct kvm *kvm, int req)
853 struct kvm_vcpu *vcpu;
855 kvm_for_each_vcpu(cx, vcpu, kvm)
856 kvm_s390_sync_request(req, vcpu);
860 * Must be called with kvm->srcu held to avoid races on memslots, and with
861 * kvm->slots_lock to avoid races with ourselves and kvm_s390_vm_stop_migration.
863 static int kvm_s390_vm_start_migration(struct kvm *kvm)
865 struct kvm_s390_migration_state *mgs;
866 struct kvm_memory_slot *ms;
867 /* should be the only one */
868 struct kvm_memslots *slots;
869 unsigned long ram_pages;
872 /* migration mode already enabled */
873 if (kvm->arch.migration_state)
876 slots = kvm_memslots(kvm);
877 if (!slots || !slots->used_slots)
880 mgs = kzalloc(sizeof(*mgs), GFP_KERNEL);
883 kvm->arch.migration_state = mgs;
885 if (kvm->arch.use_cmma) {
887 * Get the first slot. They are reverse sorted by base_gfn, so
888 * the first slot is also the one at the end of the address
889 * space. We have verified above that at least one slot is
892 ms = slots->memslots;
893 /* round up so we only use full longs */
894 ram_pages = roundup(ms->base_gfn + ms->npages, BITS_PER_LONG);
895 /* allocate enough bytes to store all the bits */
896 mgs->pgste_bitmap = vmalloc(ram_pages / 8);
897 if (!mgs->pgste_bitmap) {
899 kvm->arch.migration_state = NULL;
903 mgs->bitmap_size = ram_pages;
904 atomic64_set(&mgs->dirty_pages, ram_pages);
905 /* mark all the pages in active slots as dirty */
906 for (slotnr = 0; slotnr < slots->used_slots; slotnr++) {
907 ms = slots->memslots + slotnr;
908 bitmap_set(mgs->pgste_bitmap, ms->base_gfn, ms->npages);
911 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_START_MIGRATION);
917 * Must be called with kvm->slots_lock to avoid races with ourselves and
918 * kvm_s390_vm_start_migration.
920 static int kvm_s390_vm_stop_migration(struct kvm *kvm)
922 struct kvm_s390_migration_state *mgs;
924 /* migration mode already disabled */
925 if (!kvm->arch.migration_state)
927 mgs = kvm->arch.migration_state;
928 kvm->arch.migration_state = NULL;
930 if (kvm->arch.use_cmma) {
931 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_STOP_MIGRATION);
932 /* We have to wait for the essa emulation to finish */
933 synchronize_srcu(&kvm->srcu);
934 vfree(mgs->pgste_bitmap);
940 static int kvm_s390_vm_set_migration(struct kvm *kvm,
941 struct kvm_device_attr *attr)
945 mutex_lock(&kvm->slots_lock);
946 switch (attr->attr) {
947 case KVM_S390_VM_MIGRATION_START:
948 res = kvm_s390_vm_start_migration(kvm);
950 case KVM_S390_VM_MIGRATION_STOP:
951 res = kvm_s390_vm_stop_migration(kvm);
956 mutex_unlock(&kvm->slots_lock);
961 static int kvm_s390_vm_get_migration(struct kvm *kvm,
962 struct kvm_device_attr *attr)
964 u64 mig = (kvm->arch.migration_state != NULL);
966 if (attr->attr != KVM_S390_VM_MIGRATION_STATUS)
969 if (copy_to_user((void __user *)attr->addr, &mig, sizeof(mig)))
974 static int kvm_s390_set_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
976 struct kvm_s390_vm_tod_clock gtod;
978 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod)))
981 if (!test_kvm_facility(kvm, 139) && gtod.epoch_idx)
983 kvm_s390_set_tod_clock(kvm, >od);
985 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x, TOD base: 0x%llx",
986 gtod.epoch_idx, gtod.tod);
991 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
995 if (copy_from_user(>od_high, (void __user *)attr->addr,
1001 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
1006 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1008 struct kvm_s390_vm_tod_clock gtod = { 0 };
1010 if (copy_from_user(>od.tod, (void __user *)attr->addr,
1014 kvm_s390_set_tod_clock(kvm, >od);
1015 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod.tod);
1019 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1026 switch (attr->attr) {
1027 case KVM_S390_VM_TOD_EXT:
1028 ret = kvm_s390_set_tod_ext(kvm, attr);
1030 case KVM_S390_VM_TOD_HIGH:
1031 ret = kvm_s390_set_tod_high(kvm, attr);
1033 case KVM_S390_VM_TOD_LOW:
1034 ret = kvm_s390_set_tod_low(kvm, attr);
1043 static void kvm_s390_get_tod_clock(struct kvm *kvm,
1044 struct kvm_s390_vm_tod_clock *gtod)
1046 struct kvm_s390_tod_clock_ext htod;
1050 get_tod_clock_ext((char *)&htod);
1052 gtod->tod = htod.tod + kvm->arch.epoch;
1053 gtod->epoch_idx = 0;
1054 if (test_kvm_facility(kvm, 139)) {
1055 gtod->epoch_idx = htod.epoch_idx + kvm->arch.epdx;
1056 if (gtod->tod < htod.tod)
1057 gtod->epoch_idx += 1;
1063 static int kvm_s390_get_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1065 struct kvm_s390_vm_tod_clock gtod;
1067 memset(>od, 0, sizeof(gtod));
1068 kvm_s390_get_tod_clock(kvm, >od);
1069 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
1072 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x, TOD base: 0x%llx",
1073 gtod.epoch_idx, gtod.tod);
1077 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1081 if (copy_to_user((void __user *)attr->addr, >od_high,
1084 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
1089 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1093 gtod = kvm_s390_get_tod_clock_fast(kvm);
1094 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
1096 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
1101 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1108 switch (attr->attr) {
1109 case KVM_S390_VM_TOD_EXT:
1110 ret = kvm_s390_get_tod_ext(kvm, attr);
1112 case KVM_S390_VM_TOD_HIGH:
1113 ret = kvm_s390_get_tod_high(kvm, attr);
1115 case KVM_S390_VM_TOD_LOW:
1116 ret = kvm_s390_get_tod_low(kvm, attr);
1125 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1127 struct kvm_s390_vm_cpu_processor *proc;
1128 u16 lowest_ibc, unblocked_ibc;
1131 mutex_lock(&kvm->lock);
1132 if (kvm->created_vcpus) {
1136 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
1141 if (!copy_from_user(proc, (void __user *)attr->addr,
1143 kvm->arch.model.cpuid = proc->cpuid;
1144 lowest_ibc = sclp.ibc >> 16 & 0xfff;
1145 unblocked_ibc = sclp.ibc & 0xfff;
1146 if (lowest_ibc && proc->ibc) {
1147 if (proc->ibc > unblocked_ibc)
1148 kvm->arch.model.ibc = unblocked_ibc;
1149 else if (proc->ibc < lowest_ibc)
1150 kvm->arch.model.ibc = lowest_ibc;
1152 kvm->arch.model.ibc = proc->ibc;
1154 memcpy(kvm->arch.model.fac_list, proc->fac_list,
1155 S390_ARCH_FAC_LIST_SIZE_BYTE);
1156 VM_EVENT(kvm, 3, "SET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1157 kvm->arch.model.ibc,
1158 kvm->arch.model.cpuid);
1159 VM_EVENT(kvm, 3, "SET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1160 kvm->arch.model.fac_list[0],
1161 kvm->arch.model.fac_list[1],
1162 kvm->arch.model.fac_list[2]);
1167 mutex_unlock(&kvm->lock);
1171 static int kvm_s390_set_processor_feat(struct kvm *kvm,
1172 struct kvm_device_attr *attr)
1174 struct kvm_s390_vm_cpu_feat data;
1176 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
1178 if (!bitmap_subset((unsigned long *) data.feat,
1179 kvm_s390_available_cpu_feat,
1180 KVM_S390_VM_CPU_FEAT_NR_BITS))
1183 mutex_lock(&kvm->lock);
1184 if (kvm->created_vcpus) {
1185 mutex_unlock(&kvm->lock);
1188 bitmap_copy(kvm->arch.cpu_feat, (unsigned long *) data.feat,
1189 KVM_S390_VM_CPU_FEAT_NR_BITS);
1190 mutex_unlock(&kvm->lock);
1191 VM_EVENT(kvm, 3, "SET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1198 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
1199 struct kvm_device_attr *attr)
1202 * Once supported by kernel + hw, we have to store the subfunctions
1203 * in kvm->arch and remember that user space configured them.
1208 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1212 switch (attr->attr) {
1213 case KVM_S390_VM_CPU_PROCESSOR:
1214 ret = kvm_s390_set_processor(kvm, attr);
1216 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1217 ret = kvm_s390_set_processor_feat(kvm, attr);
1219 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1220 ret = kvm_s390_set_processor_subfunc(kvm, attr);
1226 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1228 struct kvm_s390_vm_cpu_processor *proc;
1231 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
1236 proc->cpuid = kvm->arch.model.cpuid;
1237 proc->ibc = kvm->arch.model.ibc;
1238 memcpy(&proc->fac_list, kvm->arch.model.fac_list,
1239 S390_ARCH_FAC_LIST_SIZE_BYTE);
1240 VM_EVENT(kvm, 3, "GET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1241 kvm->arch.model.ibc,
1242 kvm->arch.model.cpuid);
1243 VM_EVENT(kvm, 3, "GET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1244 kvm->arch.model.fac_list[0],
1245 kvm->arch.model.fac_list[1],
1246 kvm->arch.model.fac_list[2]);
1247 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
1254 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
1256 struct kvm_s390_vm_cpu_machine *mach;
1259 mach = kzalloc(sizeof(*mach), GFP_KERNEL);
1264 get_cpu_id((struct cpuid *) &mach->cpuid);
1265 mach->ibc = sclp.ibc;
1266 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
1267 S390_ARCH_FAC_LIST_SIZE_BYTE);
1268 memcpy((unsigned long *)&mach->fac_list, S390_lowcore.stfle_fac_list,
1269 sizeof(S390_lowcore.stfle_fac_list));
1270 VM_EVENT(kvm, 3, "GET: host ibc: 0x%4.4x, host cpuid: 0x%16.16llx",
1271 kvm->arch.model.ibc,
1272 kvm->arch.model.cpuid);
1273 VM_EVENT(kvm, 3, "GET: host facmask: 0x%16.16llx.%16.16llx.%16.16llx",
1277 VM_EVENT(kvm, 3, "GET: host faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1281 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
1288 static int kvm_s390_get_processor_feat(struct kvm *kvm,
1289 struct kvm_device_attr *attr)
1291 struct kvm_s390_vm_cpu_feat data;
1293 bitmap_copy((unsigned long *) data.feat, kvm->arch.cpu_feat,
1294 KVM_S390_VM_CPU_FEAT_NR_BITS);
1295 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1297 VM_EVENT(kvm, 3, "GET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1304 static int kvm_s390_get_machine_feat(struct kvm *kvm,
1305 struct kvm_device_attr *attr)
1307 struct kvm_s390_vm_cpu_feat data;
1309 bitmap_copy((unsigned long *) data.feat,
1310 kvm_s390_available_cpu_feat,
1311 KVM_S390_VM_CPU_FEAT_NR_BITS);
1312 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1314 VM_EVENT(kvm, 3, "GET: host feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1321 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
1322 struct kvm_device_attr *attr)
1325 * Once we can actually configure subfunctions (kernel + hw support),
1326 * we have to check if they were already set by user space, if so copy
1327 * them from kvm->arch.
1332 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
1333 struct kvm_device_attr *attr)
1335 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
1336 sizeof(struct kvm_s390_vm_cpu_subfunc)))
1340 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1344 switch (attr->attr) {
1345 case KVM_S390_VM_CPU_PROCESSOR:
1346 ret = kvm_s390_get_processor(kvm, attr);
1348 case KVM_S390_VM_CPU_MACHINE:
1349 ret = kvm_s390_get_machine(kvm, attr);
1351 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1352 ret = kvm_s390_get_processor_feat(kvm, attr);
1354 case KVM_S390_VM_CPU_MACHINE_FEAT:
1355 ret = kvm_s390_get_machine_feat(kvm, attr);
1357 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1358 ret = kvm_s390_get_processor_subfunc(kvm, attr);
1360 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1361 ret = kvm_s390_get_machine_subfunc(kvm, attr);
1367 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1371 switch (attr->group) {
1372 case KVM_S390_VM_MEM_CTRL:
1373 ret = kvm_s390_set_mem_control(kvm, attr);
1375 case KVM_S390_VM_TOD:
1376 ret = kvm_s390_set_tod(kvm, attr);
1378 case KVM_S390_VM_CPU_MODEL:
1379 ret = kvm_s390_set_cpu_model(kvm, attr);
1381 case KVM_S390_VM_CRYPTO:
1382 ret = kvm_s390_vm_set_crypto(kvm, attr);
1384 case KVM_S390_VM_MIGRATION:
1385 ret = kvm_s390_vm_set_migration(kvm, attr);
1395 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1399 switch (attr->group) {
1400 case KVM_S390_VM_MEM_CTRL:
1401 ret = kvm_s390_get_mem_control(kvm, attr);
1403 case KVM_S390_VM_TOD:
1404 ret = kvm_s390_get_tod(kvm, attr);
1406 case KVM_S390_VM_CPU_MODEL:
1407 ret = kvm_s390_get_cpu_model(kvm, attr);
1409 case KVM_S390_VM_MIGRATION:
1410 ret = kvm_s390_vm_get_migration(kvm, attr);
1420 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1424 switch (attr->group) {
1425 case KVM_S390_VM_MEM_CTRL:
1426 switch (attr->attr) {
1427 case KVM_S390_VM_MEM_ENABLE_CMMA:
1428 case KVM_S390_VM_MEM_CLR_CMMA:
1429 ret = sclp.has_cmma ? 0 : -ENXIO;
1431 case KVM_S390_VM_MEM_LIMIT_SIZE:
1439 case KVM_S390_VM_TOD:
1440 switch (attr->attr) {
1441 case KVM_S390_VM_TOD_LOW:
1442 case KVM_S390_VM_TOD_HIGH:
1450 case KVM_S390_VM_CPU_MODEL:
1451 switch (attr->attr) {
1452 case KVM_S390_VM_CPU_PROCESSOR:
1453 case KVM_S390_VM_CPU_MACHINE:
1454 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1455 case KVM_S390_VM_CPU_MACHINE_FEAT:
1456 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1459 /* configuring subfunctions is not supported yet */
1460 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1466 case KVM_S390_VM_CRYPTO:
1467 switch (attr->attr) {
1468 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1469 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1470 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1471 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1479 case KVM_S390_VM_MIGRATION:
1490 static long kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1494 int srcu_idx, i, r = 0;
1496 if (args->flags != 0)
1499 /* Is this guest using storage keys? */
1500 if (!mm_uses_skeys(current->mm))
1501 return KVM_S390_GET_SKEYS_NONE;
1503 /* Enforce sane limit on memory allocation */
1504 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1507 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL);
1511 down_read(¤t->mm->mmap_sem);
1512 srcu_idx = srcu_read_lock(&kvm->srcu);
1513 for (i = 0; i < args->count; i++) {
1514 hva = gfn_to_hva(kvm, args->start_gfn + i);
1515 if (kvm_is_error_hva(hva)) {
1520 r = get_guest_storage_key(current->mm, hva, &keys[i]);
1524 srcu_read_unlock(&kvm->srcu, srcu_idx);
1525 up_read(¤t->mm->mmap_sem);
1528 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
1529 sizeof(uint8_t) * args->count);
1538 static long kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1542 int srcu_idx, i, r = 0;
1544 if (args->flags != 0)
1547 /* Enforce sane limit on memory allocation */
1548 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1551 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL);
1555 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
1556 sizeof(uint8_t) * args->count);
1562 /* Enable storage key handling for the guest */
1563 r = s390_enable_skey();
1567 down_read(¤t->mm->mmap_sem);
1568 srcu_idx = srcu_read_lock(&kvm->srcu);
1569 for (i = 0; i < args->count; i++) {
1570 hva = gfn_to_hva(kvm, args->start_gfn + i);
1571 if (kvm_is_error_hva(hva)) {
1576 /* Lowest order bit is reserved */
1577 if (keys[i] & 0x01) {
1582 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
1586 srcu_read_unlock(&kvm->srcu, srcu_idx);
1587 up_read(¤t->mm->mmap_sem);
1594 * Base address and length must be sent at the start of each block, therefore
1595 * it's cheaper to send some clean data, as long as it's less than the size of
1598 #define KVM_S390_MAX_BIT_DISTANCE (2 * sizeof(void *))
1599 /* for consistency */
1600 #define KVM_S390_CMMA_SIZE_MAX ((u32)KVM_S390_SKEYS_MAX)
1603 * This function searches for the next page with dirty CMMA attributes, and
1604 * saves the attributes in the buffer up to either the end of the buffer or
1605 * until a block of at least KVM_S390_MAX_BIT_DISTANCE clean bits is found;
1606 * no trailing clean bytes are saved.
1607 * In case no dirty bits were found, or if CMMA was not enabled or used, the
1608 * output buffer will indicate 0 as length.
1610 static int kvm_s390_get_cmma_bits(struct kvm *kvm,
1611 struct kvm_s390_cmma_log *args)
1613 struct kvm_s390_migration_state *s = kvm->arch.migration_state;
1614 unsigned long bufsize, hva, pgstev, i, next, cur;
1615 int srcu_idx, peek, r = 0, rr;
1618 cur = args->start_gfn;
1619 i = next = pgstev = 0;
1621 if (unlikely(!kvm->arch.use_cmma))
1623 /* Invalid/unsupported flags were specified */
1624 if (args->flags & ~KVM_S390_CMMA_PEEK)
1626 /* Migration mode query, and we are not doing a migration */
1627 peek = !!(args->flags & KVM_S390_CMMA_PEEK);
1630 /* CMMA is disabled or was not used, or the buffer has length zero */
1631 bufsize = min(args->count, KVM_S390_CMMA_SIZE_MAX);
1632 if (!bufsize || !kvm->mm->context.uses_cmm) {
1633 memset(args, 0, sizeof(*args));
1638 /* We are not peeking, and there are no dirty pages */
1639 if (!atomic64_read(&s->dirty_pages)) {
1640 memset(args, 0, sizeof(*args));
1643 cur = find_next_bit(s->pgste_bitmap, s->bitmap_size,
1645 if (cur >= s->bitmap_size) /* nothing found, loop back */
1646 cur = find_next_bit(s->pgste_bitmap, s->bitmap_size, 0);
1647 if (cur >= s->bitmap_size) { /* again! (very unlikely) */
1648 memset(args, 0, sizeof(*args));
1651 next = find_next_bit(s->pgste_bitmap, s->bitmap_size, cur + 1);
1654 res = vmalloc(bufsize);
1658 args->start_gfn = cur;
1660 down_read(&kvm->mm->mmap_sem);
1661 srcu_idx = srcu_read_lock(&kvm->srcu);
1662 while (i < bufsize) {
1663 hva = gfn_to_hva(kvm, cur);
1664 if (kvm_is_error_hva(hva)) {
1668 /* decrement only if we actually flipped the bit to 0 */
1669 if (!peek && test_and_clear_bit(cur, s->pgste_bitmap))
1670 atomic64_dec(&s->dirty_pages);
1671 r = get_pgste(kvm->mm, hva, &pgstev);
1674 /* save the value */
1675 res[i++] = (pgstev >> 24) & 0x43;
1677 * if the next bit is too far away, stop.
1678 * if we reached the previous "next", find the next one
1681 if (next > cur + KVM_S390_MAX_BIT_DISTANCE)
1684 next = find_next_bit(s->pgste_bitmap,
1685 s->bitmap_size, cur + 1);
1686 /* reached the end of the bitmap or of the buffer, stop */
1687 if ((next >= s->bitmap_size) ||
1688 (next >= args->start_gfn + bufsize))
1693 srcu_read_unlock(&kvm->srcu, srcu_idx);
1694 up_read(&kvm->mm->mmap_sem);
1696 args->remaining = s ? atomic64_read(&s->dirty_pages) : 0;
1698 rr = copy_to_user((void __user *)args->values, res, args->count);
1707 * This function sets the CMMA attributes for the given pages. If the input
1708 * buffer has zero length, no action is taken, otherwise the attributes are
1709 * set and the mm->context.uses_cmm flag is set.
1711 static int kvm_s390_set_cmma_bits(struct kvm *kvm,
1712 const struct kvm_s390_cmma_log *args)
1714 unsigned long hva, mask, pgstev, i;
1716 int srcu_idx, r = 0;
1720 if (!kvm->arch.use_cmma)
1722 /* invalid/unsupported flags */
1723 if (args->flags != 0)
1725 /* Enforce sane limit on memory allocation */
1726 if (args->count > KVM_S390_CMMA_SIZE_MAX)
1729 if (args->count == 0)
1732 bits = vmalloc(sizeof(*bits) * args->count);
1736 r = copy_from_user(bits, (void __user *)args->values, args->count);
1742 down_read(&kvm->mm->mmap_sem);
1743 srcu_idx = srcu_read_lock(&kvm->srcu);
1744 for (i = 0; i < args->count; i++) {
1745 hva = gfn_to_hva(kvm, args->start_gfn + i);
1746 if (kvm_is_error_hva(hva)) {
1752 pgstev = pgstev << 24;
1753 mask &= _PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT;
1754 set_pgste_bits(kvm->mm, hva, mask, pgstev);
1756 srcu_read_unlock(&kvm->srcu, srcu_idx);
1757 up_read(&kvm->mm->mmap_sem);
1759 if (!kvm->mm->context.uses_cmm) {
1760 down_write(&kvm->mm->mmap_sem);
1761 kvm->mm->context.uses_cmm = 1;
1762 up_write(&kvm->mm->mmap_sem);
1769 long kvm_arch_vm_ioctl(struct file *filp,
1770 unsigned int ioctl, unsigned long arg)
1772 struct kvm *kvm = filp->private_data;
1773 void __user *argp = (void __user *)arg;
1774 struct kvm_device_attr attr;
1778 case KVM_S390_INTERRUPT: {
1779 struct kvm_s390_interrupt s390int;
1782 if (copy_from_user(&s390int, argp, sizeof(s390int)))
1784 r = kvm_s390_inject_vm(kvm, &s390int);
1787 case KVM_ENABLE_CAP: {
1788 struct kvm_enable_cap cap;
1790 if (copy_from_user(&cap, argp, sizeof(cap)))
1792 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1795 case KVM_CREATE_IRQCHIP: {
1796 struct kvm_irq_routing_entry routing;
1799 if (kvm->arch.use_irqchip) {
1800 /* Set up dummy routing. */
1801 memset(&routing, 0, sizeof(routing));
1802 r = kvm_set_irq_routing(kvm, &routing, 0, 0);
1806 case KVM_SET_DEVICE_ATTR: {
1808 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1810 r = kvm_s390_vm_set_attr(kvm, &attr);
1813 case KVM_GET_DEVICE_ATTR: {
1815 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1817 r = kvm_s390_vm_get_attr(kvm, &attr);
1820 case KVM_HAS_DEVICE_ATTR: {
1822 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1824 r = kvm_s390_vm_has_attr(kvm, &attr);
1827 case KVM_S390_GET_SKEYS: {
1828 struct kvm_s390_skeys args;
1831 if (copy_from_user(&args, argp,
1832 sizeof(struct kvm_s390_skeys)))
1834 r = kvm_s390_get_skeys(kvm, &args);
1837 case KVM_S390_SET_SKEYS: {
1838 struct kvm_s390_skeys args;
1841 if (copy_from_user(&args, argp,
1842 sizeof(struct kvm_s390_skeys)))
1844 r = kvm_s390_set_skeys(kvm, &args);
1847 case KVM_S390_GET_CMMA_BITS: {
1848 struct kvm_s390_cmma_log args;
1851 if (copy_from_user(&args, argp, sizeof(args)))
1853 mutex_lock(&kvm->slots_lock);
1854 r = kvm_s390_get_cmma_bits(kvm, &args);
1855 mutex_unlock(&kvm->slots_lock);
1857 r = copy_to_user(argp, &args, sizeof(args));
1863 case KVM_S390_SET_CMMA_BITS: {
1864 struct kvm_s390_cmma_log args;
1867 if (copy_from_user(&args, argp, sizeof(args)))
1869 mutex_lock(&kvm->slots_lock);
1870 r = kvm_s390_set_cmma_bits(kvm, &args);
1871 mutex_unlock(&kvm->slots_lock);
1881 static int kvm_s390_query_ap_config(u8 *config)
1883 u32 fcn_code = 0x04000000UL;
1886 memset(config, 0, 128);
1890 ".long 0xb2af0000\n" /* PQAP(QCI) */
1896 : "r" (fcn_code), "r" (config)
1897 : "cc", "0", "2", "memory"
1903 static int kvm_s390_apxa_installed(void)
1908 if (test_facility(12)) {
1909 cc = kvm_s390_query_ap_config(config);
1912 pr_err("PQAP(QCI) failed with cc=%d", cc);
1914 return config[0] & 0x40;
1920 static void kvm_s390_set_crycb_format(struct kvm *kvm)
1922 kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
1924 if (kvm_s390_apxa_installed())
1925 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
1927 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
1930 static u64 kvm_s390_get_initial_cpuid(void)
1935 cpuid.version = 0xff;
1936 return *((u64 *) &cpuid);
1939 static void kvm_s390_crypto_init(struct kvm *kvm)
1941 if (!test_kvm_facility(kvm, 76))
1944 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
1945 kvm_s390_set_crycb_format(kvm);
1947 /* Enable AES/DEA protected key functions by default */
1948 kvm->arch.crypto.aes_kw = 1;
1949 kvm->arch.crypto.dea_kw = 1;
1950 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1951 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1952 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1953 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1956 static void sca_dispose(struct kvm *kvm)
1958 if (kvm->arch.use_esca)
1959 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
1961 free_page((unsigned long)(kvm->arch.sca));
1962 kvm->arch.sca = NULL;
1965 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
1967 gfp_t alloc_flags = GFP_KERNEL;
1969 char debug_name[16];
1970 static unsigned long sca_offset;
1973 #ifdef CONFIG_KVM_S390_UCONTROL
1974 if (type & ~KVM_VM_S390_UCONTROL)
1976 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
1983 rc = s390_enable_sie();
1989 if (!sclp.has_64bscao)
1990 alloc_flags |= GFP_DMA;
1991 rwlock_init(&kvm->arch.sca_lock);
1992 /* start with basic SCA */
1993 kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
1996 spin_lock(&kvm_lock);
1998 if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
2000 kvm->arch.sca = (struct bsca_block *)
2001 ((char *) kvm->arch.sca + sca_offset);
2002 spin_unlock(&kvm_lock);
2004 sprintf(debug_name, "kvm-%u", current->pid);
2006 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
2010 BUILD_BUG_ON(sizeof(struct sie_page2) != 4096);
2011 kvm->arch.sie_page2 =
2012 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
2013 if (!kvm->arch.sie_page2)
2016 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
2018 for (i = 0; i < kvm_s390_fac_size(); i++) {
2019 kvm->arch.model.fac_mask[i] = S390_lowcore.stfle_fac_list[i] &
2020 (kvm_s390_fac_base[i] |
2021 kvm_s390_fac_ext[i]);
2022 kvm->arch.model.fac_list[i] = S390_lowcore.stfle_fac_list[i] &
2023 kvm_s390_fac_base[i];
2026 /* we are always in czam mode - even on pre z14 machines */
2027 set_kvm_facility(kvm->arch.model.fac_mask, 138);
2028 set_kvm_facility(kvm->arch.model.fac_list, 138);
2029 /* we emulate STHYI in kvm */
2030 set_kvm_facility(kvm->arch.model.fac_mask, 74);
2031 set_kvm_facility(kvm->arch.model.fac_list, 74);
2032 if (MACHINE_HAS_TLB_GUEST) {
2033 set_kvm_facility(kvm->arch.model.fac_mask, 147);
2034 set_kvm_facility(kvm->arch.model.fac_list, 147);
2037 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
2038 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
2040 kvm_s390_crypto_init(kvm);
2042 mutex_init(&kvm->arch.float_int.ais_lock);
2043 spin_lock_init(&kvm->arch.float_int.lock);
2044 for (i = 0; i < FIRQ_LIST_COUNT; i++)
2045 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
2046 init_waitqueue_head(&kvm->arch.ipte_wq);
2047 mutex_init(&kvm->arch.ipte_mutex);
2049 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
2050 VM_EVENT(kvm, 3, "vm created with type %lu", type);
2052 if (type & KVM_VM_S390_UCONTROL) {
2053 kvm->arch.gmap = NULL;
2054 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
2056 if (sclp.hamax == U64_MAX)
2057 kvm->arch.mem_limit = TASK_SIZE_MAX;
2059 kvm->arch.mem_limit = min_t(unsigned long, TASK_SIZE_MAX,
2061 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
2062 if (!kvm->arch.gmap)
2064 kvm->arch.gmap->private = kvm;
2065 kvm->arch.gmap->pfault_enabled = 0;
2068 kvm->arch.use_pfmfi = sclp.has_pfmfi;
2069 kvm->arch.use_skf = sclp.has_skey;
2070 spin_lock_init(&kvm->arch.start_stop_lock);
2071 kvm_s390_vsie_init(kvm);
2072 kvm_s390_gisa_init(kvm);
2073 KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
2077 free_page((unsigned long)kvm->arch.sie_page2);
2078 debug_unregister(kvm->arch.dbf);
2080 KVM_EVENT(3, "creation of vm failed: %d", rc);
2084 bool kvm_arch_has_vcpu_debugfs(void)
2089 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2094 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
2096 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
2097 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
2098 kvm_s390_clear_local_irqs(vcpu);
2099 kvm_clear_async_pf_completion_queue(vcpu);
2100 if (!kvm_is_ucontrol(vcpu->kvm))
2103 if (kvm_is_ucontrol(vcpu->kvm))
2104 gmap_remove(vcpu->arch.gmap);
2106 if (vcpu->kvm->arch.use_cmma)
2107 kvm_s390_vcpu_unsetup_cmma(vcpu);
2108 free_page((unsigned long)(vcpu->arch.sie_block));
2110 kvm_vcpu_uninit(vcpu);
2111 kmem_cache_free(kvm_vcpu_cache, vcpu);
2114 static void kvm_free_vcpus(struct kvm *kvm)
2117 struct kvm_vcpu *vcpu;
2119 kvm_for_each_vcpu(i, vcpu, kvm)
2120 kvm_arch_vcpu_destroy(vcpu);
2122 mutex_lock(&kvm->lock);
2123 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
2124 kvm->vcpus[i] = NULL;
2126 atomic_set(&kvm->online_vcpus, 0);
2127 mutex_unlock(&kvm->lock);
2130 void kvm_arch_destroy_vm(struct kvm *kvm)
2132 kvm_free_vcpus(kvm);
2134 debug_unregister(kvm->arch.dbf);
2135 kvm_s390_gisa_destroy(kvm);
2136 free_page((unsigned long)kvm->arch.sie_page2);
2137 if (!kvm_is_ucontrol(kvm))
2138 gmap_remove(kvm->arch.gmap);
2139 kvm_s390_destroy_adapters(kvm);
2140 kvm_s390_clear_float_irqs(kvm);
2141 kvm_s390_vsie_destroy(kvm);
2142 if (kvm->arch.migration_state) {
2143 vfree(kvm->arch.migration_state->pgste_bitmap);
2144 kfree(kvm->arch.migration_state);
2146 KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
2149 /* Section: vcpu related */
2150 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
2152 vcpu->arch.gmap = gmap_create(current->mm, -1UL);
2153 if (!vcpu->arch.gmap)
2155 vcpu->arch.gmap->private = vcpu->kvm;
2160 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
2162 if (!kvm_s390_use_sca_entries())
2164 read_lock(&vcpu->kvm->arch.sca_lock);
2165 if (vcpu->kvm->arch.use_esca) {
2166 struct esca_block *sca = vcpu->kvm->arch.sca;
2168 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
2169 sca->cpu[vcpu->vcpu_id].sda = 0;
2171 struct bsca_block *sca = vcpu->kvm->arch.sca;
2173 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
2174 sca->cpu[vcpu->vcpu_id].sda = 0;
2176 read_unlock(&vcpu->kvm->arch.sca_lock);
2179 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
2181 if (!kvm_s390_use_sca_entries()) {
2182 struct bsca_block *sca = vcpu->kvm->arch.sca;
2184 /* we still need the basic sca for the ipte control */
2185 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
2186 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
2189 read_lock(&vcpu->kvm->arch.sca_lock);
2190 if (vcpu->kvm->arch.use_esca) {
2191 struct esca_block *sca = vcpu->kvm->arch.sca;
2193 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
2194 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
2195 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca & ~0x3fU;
2196 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
2197 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
2199 struct bsca_block *sca = vcpu->kvm->arch.sca;
2201 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
2202 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
2203 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
2204 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
2206 read_unlock(&vcpu->kvm->arch.sca_lock);
2209 /* Basic SCA to Extended SCA data copy routines */
2210 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
2213 d->sigp_ctrl.c = s->sigp_ctrl.c;
2214 d->sigp_ctrl.scn = s->sigp_ctrl.scn;
2217 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
2221 d->ipte_control = s->ipte_control;
2223 for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
2224 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
2227 static int sca_switch_to_extended(struct kvm *kvm)
2229 struct bsca_block *old_sca = kvm->arch.sca;
2230 struct esca_block *new_sca;
2231 struct kvm_vcpu *vcpu;
2232 unsigned int vcpu_idx;
2235 new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL|__GFP_ZERO);
2239 scaoh = (u32)((u64)(new_sca) >> 32);
2240 scaol = (u32)(u64)(new_sca) & ~0x3fU;
2242 kvm_s390_vcpu_block_all(kvm);
2243 write_lock(&kvm->arch.sca_lock);
2245 sca_copy_b_to_e(new_sca, old_sca);
2247 kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
2248 vcpu->arch.sie_block->scaoh = scaoh;
2249 vcpu->arch.sie_block->scaol = scaol;
2250 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
2252 kvm->arch.sca = new_sca;
2253 kvm->arch.use_esca = 1;
2255 write_unlock(&kvm->arch.sca_lock);
2256 kvm_s390_vcpu_unblock_all(kvm);
2258 free_page((unsigned long)old_sca);
2260 VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
2261 old_sca, kvm->arch.sca);
2265 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
2269 if (!kvm_s390_use_sca_entries()) {
2270 if (id < KVM_MAX_VCPUS)
2274 if (id < KVM_S390_BSCA_CPU_SLOTS)
2276 if (!sclp.has_esca || !sclp.has_64bscao)
2279 mutex_lock(&kvm->lock);
2280 rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
2281 mutex_unlock(&kvm->lock);
2283 return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
2286 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
2288 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
2289 kvm_clear_async_pf_completion_queue(vcpu);
2290 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
2296 kvm_s390_set_prefix(vcpu, 0);
2297 if (test_kvm_facility(vcpu->kvm, 64))
2298 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
2299 if (test_kvm_facility(vcpu->kvm, 82))
2300 vcpu->run->kvm_valid_regs |= KVM_SYNC_BPBC;
2301 if (test_kvm_facility(vcpu->kvm, 133))
2302 vcpu->run->kvm_valid_regs |= KVM_SYNC_GSCB;
2303 /* fprs can be synchronized via vrs, even if the guest has no vx. With
2304 * MACHINE_HAS_VX, (load|store)_fpu_regs() will work with vrs format.
2307 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
2309 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
2311 if (kvm_is_ucontrol(vcpu->kvm))
2312 return __kvm_ucontrol_vcpu_init(vcpu);
2317 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2318 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2320 WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
2321 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
2322 vcpu->arch.cputm_start = get_tod_clock_fast();
2323 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
2326 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2327 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2329 WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
2330 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
2331 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
2332 vcpu->arch.cputm_start = 0;
2333 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
2336 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2337 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2339 WARN_ON_ONCE(vcpu->arch.cputm_enabled);
2340 vcpu->arch.cputm_enabled = true;
2341 __start_cpu_timer_accounting(vcpu);
2344 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2345 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2347 WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
2348 __stop_cpu_timer_accounting(vcpu);
2349 vcpu->arch.cputm_enabled = false;
2352 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2354 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2355 __enable_cpu_timer_accounting(vcpu);
2359 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2361 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2362 __disable_cpu_timer_accounting(vcpu);
2366 /* set the cpu timer - may only be called from the VCPU thread itself */
2367 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
2369 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2370 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
2371 if (vcpu->arch.cputm_enabled)
2372 vcpu->arch.cputm_start = get_tod_clock_fast();
2373 vcpu->arch.sie_block->cputm = cputm;
2374 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
2378 /* update and get the cpu timer - can also be called from other VCPU threads */
2379 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
2384 if (unlikely(!vcpu->arch.cputm_enabled))
2385 return vcpu->arch.sie_block->cputm;
2387 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2389 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
2391 * If the writer would ever execute a read in the critical
2392 * section, e.g. in irq context, we have a deadlock.
2394 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
2395 value = vcpu->arch.sie_block->cputm;
2396 /* if cputm_start is 0, accounting is being started/stopped */
2397 if (likely(vcpu->arch.cputm_start))
2398 value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
2399 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
2404 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2407 gmap_enable(vcpu->arch.enabled_gmap);
2408 kvm_s390_set_cpuflags(vcpu, CPUSTAT_RUNNING);
2409 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
2410 __start_cpu_timer_accounting(vcpu);
2414 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2417 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
2418 __stop_cpu_timer_accounting(vcpu);
2419 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_RUNNING);
2420 vcpu->arch.enabled_gmap = gmap_get_enabled();
2421 gmap_disable(vcpu->arch.enabled_gmap);
2425 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
2427 /* this equals initial cpu reset in pop, but we don't switch to ESA */
2428 vcpu->arch.sie_block->gpsw.mask = 0UL;
2429 vcpu->arch.sie_block->gpsw.addr = 0UL;
2430 kvm_s390_set_prefix(vcpu, 0);
2431 kvm_s390_set_cpu_timer(vcpu, 0);
2432 vcpu->arch.sie_block->ckc = 0UL;
2433 vcpu->arch.sie_block->todpr = 0;
2434 memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64));
2435 vcpu->arch.sie_block->gcr[0] = CR0_UNUSED_56 |
2436 CR0_INTERRUPT_KEY_SUBMASK |
2437 CR0_MEASUREMENT_ALERT_SUBMASK;
2438 vcpu->arch.sie_block->gcr[14] = CR14_UNUSED_32 |
2440 CR14_EXTERNAL_DAMAGE_SUBMASK;
2441 /* make sure the new fpc will be lazily loaded */
2443 current->thread.fpu.fpc = 0;
2444 vcpu->arch.sie_block->gbea = 1;
2445 vcpu->arch.sie_block->pp = 0;
2446 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
2447 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
2448 kvm_clear_async_pf_completion_queue(vcpu);
2449 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
2450 kvm_s390_vcpu_stop(vcpu);
2451 kvm_s390_clear_local_irqs(vcpu);
2454 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
2456 mutex_lock(&vcpu->kvm->lock);
2458 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
2459 vcpu->arch.sie_block->epdx = vcpu->kvm->arch.epdx;
2461 mutex_unlock(&vcpu->kvm->lock);
2462 if (!kvm_is_ucontrol(vcpu->kvm)) {
2463 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
2466 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
2467 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
2468 /* make vcpu_load load the right gmap on the first trigger */
2469 vcpu->arch.enabled_gmap = vcpu->arch.gmap;
2472 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
2474 if (!test_kvm_facility(vcpu->kvm, 76))
2477 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
2479 if (vcpu->kvm->arch.crypto.aes_kw)
2480 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
2481 if (vcpu->kvm->arch.crypto.dea_kw)
2482 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
2484 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
2487 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
2489 free_page(vcpu->arch.sie_block->cbrlo);
2490 vcpu->arch.sie_block->cbrlo = 0;
2493 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
2495 vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
2496 if (!vcpu->arch.sie_block->cbrlo)
2501 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
2503 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
2505 vcpu->arch.sie_block->ibc = model->ibc;
2506 if (test_kvm_facility(vcpu->kvm, 7))
2507 vcpu->arch.sie_block->fac = (u32)(u64) model->fac_list;
2510 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
2514 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
2518 if (test_kvm_facility(vcpu->kvm, 78))
2519 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED2);
2520 else if (test_kvm_facility(vcpu->kvm, 8))
2521 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED);
2523 kvm_s390_vcpu_setup_model(vcpu);
2525 /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
2526 if (MACHINE_HAS_ESOP)
2527 vcpu->arch.sie_block->ecb |= ECB_HOSTPROTINT;
2528 if (test_kvm_facility(vcpu->kvm, 9))
2529 vcpu->arch.sie_block->ecb |= ECB_SRSI;
2530 if (test_kvm_facility(vcpu->kvm, 73))
2531 vcpu->arch.sie_block->ecb |= ECB_TE;
2533 if (test_kvm_facility(vcpu->kvm, 8) && vcpu->kvm->arch.use_pfmfi)
2534 vcpu->arch.sie_block->ecb2 |= ECB2_PFMFI;
2535 if (test_kvm_facility(vcpu->kvm, 130))
2536 vcpu->arch.sie_block->ecb2 |= ECB2_IEP;
2537 vcpu->arch.sie_block->eca = ECA_MVPGI | ECA_PROTEXCI;
2539 vcpu->arch.sie_block->eca |= ECA_CEI;
2541 vcpu->arch.sie_block->eca |= ECA_IB;
2543 vcpu->arch.sie_block->eca |= ECA_SII;
2544 if (sclp.has_sigpif)
2545 vcpu->arch.sie_block->eca |= ECA_SIGPI;
2546 if (test_kvm_facility(vcpu->kvm, 129)) {
2547 vcpu->arch.sie_block->eca |= ECA_VX;
2548 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
2550 if (test_kvm_facility(vcpu->kvm, 139))
2551 vcpu->arch.sie_block->ecd |= ECD_MEF;
2553 if (vcpu->arch.sie_block->gd) {
2554 vcpu->arch.sie_block->eca |= ECA_AIV;
2555 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
2556 vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
2558 vcpu->arch.sie_block->sdnxo = ((unsigned long) &vcpu->run->s.regs.sdnx)
2560 vcpu->arch.sie_block->riccbd = (unsigned long) &vcpu->run->s.regs.riccb;
2563 kvm_s390_set_cpuflags(vcpu, CPUSTAT_KSS);
2565 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
2567 if (vcpu->kvm->arch.use_cmma) {
2568 rc = kvm_s390_vcpu_setup_cmma(vcpu);
2572 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
2573 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
2575 kvm_s390_vcpu_crypto_setup(vcpu);
2580 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
2583 struct kvm_vcpu *vcpu;
2584 struct sie_page *sie_page;
2587 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
2592 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2596 BUILD_BUG_ON(sizeof(struct sie_page) != 4096);
2597 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
2601 vcpu->arch.sie_block = &sie_page->sie_block;
2602 vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
2604 /* the real guest size will always be smaller than msl */
2605 vcpu->arch.sie_block->mso = 0;
2606 vcpu->arch.sie_block->msl = sclp.hamax;
2608 vcpu->arch.sie_block->icpua = id;
2609 spin_lock_init(&vcpu->arch.local_int.lock);
2610 vcpu->arch.sie_block->gd = (u32)(u64)kvm->arch.gisa;
2611 if (vcpu->arch.sie_block->gd && sclp.has_gisaf)
2612 vcpu->arch.sie_block->gd |= GISA_FORMAT1;
2613 seqcount_init(&vcpu->arch.cputm_seqcount);
2615 rc = kvm_vcpu_init(vcpu, kvm, id);
2617 goto out_free_sie_block;
2618 VM_EVENT(kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK", id, vcpu,
2619 vcpu->arch.sie_block);
2620 trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
2624 free_page((unsigned long)(vcpu->arch.sie_block));
2626 kmem_cache_free(kvm_vcpu_cache, vcpu);
2631 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
2633 return kvm_s390_vcpu_has_irq(vcpu, 0);
2636 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
2638 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE);
2641 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
2643 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2647 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
2649 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2652 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
2654 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2658 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
2660 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2664 * Kick a guest cpu out of SIE and wait until SIE is not running.
2665 * If the CPU is not running (e.g. waiting as idle) the function will
2666 * return immediately. */
2667 void exit_sie(struct kvm_vcpu *vcpu)
2669 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
2670 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
2674 /* Kick a guest cpu out of SIE to process a request synchronously */
2675 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
2677 kvm_make_request(req, vcpu);
2678 kvm_s390_vcpu_request(vcpu);
2681 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
2684 struct kvm *kvm = gmap->private;
2685 struct kvm_vcpu *vcpu;
2686 unsigned long prefix;
2689 if (gmap_is_shadow(gmap))
2691 if (start >= 1UL << 31)
2692 /* We are only interested in prefix pages */
2694 kvm_for_each_vcpu(i, vcpu, kvm) {
2695 /* match against both prefix pages */
2696 prefix = kvm_s390_get_prefix(vcpu);
2697 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
2698 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
2700 kvm_s390_sync_request(KVM_REQ_MMU_RELOAD, vcpu);
2705 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
2707 /* kvm common code refers to this, but never calls it */
2712 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
2713 struct kvm_one_reg *reg)
2718 case KVM_REG_S390_TODPR:
2719 r = put_user(vcpu->arch.sie_block->todpr,
2720 (u32 __user *)reg->addr);
2722 case KVM_REG_S390_EPOCHDIFF:
2723 r = put_user(vcpu->arch.sie_block->epoch,
2724 (u64 __user *)reg->addr);
2726 case KVM_REG_S390_CPU_TIMER:
2727 r = put_user(kvm_s390_get_cpu_timer(vcpu),
2728 (u64 __user *)reg->addr);
2730 case KVM_REG_S390_CLOCK_COMP:
2731 r = put_user(vcpu->arch.sie_block->ckc,
2732 (u64 __user *)reg->addr);
2734 case KVM_REG_S390_PFTOKEN:
2735 r = put_user(vcpu->arch.pfault_token,
2736 (u64 __user *)reg->addr);
2738 case KVM_REG_S390_PFCOMPARE:
2739 r = put_user(vcpu->arch.pfault_compare,
2740 (u64 __user *)reg->addr);
2742 case KVM_REG_S390_PFSELECT:
2743 r = put_user(vcpu->arch.pfault_select,
2744 (u64 __user *)reg->addr);
2746 case KVM_REG_S390_PP:
2747 r = put_user(vcpu->arch.sie_block->pp,
2748 (u64 __user *)reg->addr);
2750 case KVM_REG_S390_GBEA:
2751 r = put_user(vcpu->arch.sie_block->gbea,
2752 (u64 __user *)reg->addr);
2761 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
2762 struct kvm_one_reg *reg)
2768 case KVM_REG_S390_TODPR:
2769 r = get_user(vcpu->arch.sie_block->todpr,
2770 (u32 __user *)reg->addr);
2772 case KVM_REG_S390_EPOCHDIFF:
2773 r = get_user(vcpu->arch.sie_block->epoch,
2774 (u64 __user *)reg->addr);
2776 case KVM_REG_S390_CPU_TIMER:
2777 r = get_user(val, (u64 __user *)reg->addr);
2779 kvm_s390_set_cpu_timer(vcpu, val);
2781 case KVM_REG_S390_CLOCK_COMP:
2782 r = get_user(vcpu->arch.sie_block->ckc,
2783 (u64 __user *)reg->addr);
2785 case KVM_REG_S390_PFTOKEN:
2786 r = get_user(vcpu->arch.pfault_token,
2787 (u64 __user *)reg->addr);
2788 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2789 kvm_clear_async_pf_completion_queue(vcpu);
2791 case KVM_REG_S390_PFCOMPARE:
2792 r = get_user(vcpu->arch.pfault_compare,
2793 (u64 __user *)reg->addr);
2795 case KVM_REG_S390_PFSELECT:
2796 r = get_user(vcpu->arch.pfault_select,
2797 (u64 __user *)reg->addr);
2799 case KVM_REG_S390_PP:
2800 r = get_user(vcpu->arch.sie_block->pp,
2801 (u64 __user *)reg->addr);
2803 case KVM_REG_S390_GBEA:
2804 r = get_user(vcpu->arch.sie_block->gbea,
2805 (u64 __user *)reg->addr);
2814 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
2816 kvm_s390_vcpu_initial_reset(vcpu);
2820 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2823 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs));
2828 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2831 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
2836 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2837 struct kvm_sregs *sregs)
2841 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
2842 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
2848 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2849 struct kvm_sregs *sregs)
2853 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
2854 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
2860 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2866 if (test_fp_ctl(fpu->fpc)) {
2870 vcpu->run->s.regs.fpc = fpu->fpc;
2872 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs,
2873 (freg_t *) fpu->fprs);
2875 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs));
2882 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2886 /* make sure we have the latest values */
2889 convert_vx_to_fp((freg_t *) fpu->fprs,
2890 (__vector128 *) vcpu->run->s.regs.vrs);
2892 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs));
2893 fpu->fpc = vcpu->run->s.regs.fpc;
2899 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
2903 if (!is_vcpu_stopped(vcpu))
2906 vcpu->run->psw_mask = psw.mask;
2907 vcpu->run->psw_addr = psw.addr;
2912 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2913 struct kvm_translation *tr)
2915 return -EINVAL; /* not implemented yet */
2918 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
2919 KVM_GUESTDBG_USE_HW_BP | \
2920 KVM_GUESTDBG_ENABLE)
2922 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
2923 struct kvm_guest_debug *dbg)
2929 vcpu->guest_debug = 0;
2930 kvm_s390_clear_bp_data(vcpu);
2932 if (dbg->control & ~VALID_GUESTDBG_FLAGS) {
2936 if (!sclp.has_gpere) {
2941 if (dbg->control & KVM_GUESTDBG_ENABLE) {
2942 vcpu->guest_debug = dbg->control;
2943 /* enforce guest PER */
2944 kvm_s390_set_cpuflags(vcpu, CPUSTAT_P);
2946 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
2947 rc = kvm_s390_import_bp_data(vcpu, dbg);
2949 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
2950 vcpu->arch.guestdbg.last_bp = 0;
2954 vcpu->guest_debug = 0;
2955 kvm_s390_clear_bp_data(vcpu);
2956 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
2964 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2965 struct kvm_mp_state *mp_state)
2971 /* CHECK_STOP and LOAD are not supported yet */
2972 ret = is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
2973 KVM_MP_STATE_OPERATING;
2979 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2980 struct kvm_mp_state *mp_state)
2986 /* user space knows about this interface - let it control the state */
2987 vcpu->kvm->arch.user_cpu_state_ctrl = 1;
2989 switch (mp_state->mp_state) {
2990 case KVM_MP_STATE_STOPPED:
2991 kvm_s390_vcpu_stop(vcpu);
2993 case KVM_MP_STATE_OPERATING:
2994 kvm_s390_vcpu_start(vcpu);
2996 case KVM_MP_STATE_LOAD:
2997 case KVM_MP_STATE_CHECK_STOP:
2998 /* fall through - CHECK_STOP and LOAD are not supported yet */
3007 static bool ibs_enabled(struct kvm_vcpu *vcpu)
3009 return kvm_s390_test_cpuflags(vcpu, CPUSTAT_IBS);
3012 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
3015 kvm_s390_vcpu_request_handled(vcpu);
3016 if (!kvm_request_pending(vcpu))
3019 * We use MMU_RELOAD just to re-arm the ipte notifier for the
3020 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
3021 * This ensures that the ipte instruction for this request has
3022 * already finished. We might race against a second unmapper that
3023 * wants to set the blocking bit. Lets just retry the request loop.
3025 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
3027 rc = gmap_mprotect_notify(vcpu->arch.gmap,
3028 kvm_s390_get_prefix(vcpu),
3029 PAGE_SIZE * 2, PROT_WRITE);
3031 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
3037 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
3038 vcpu->arch.sie_block->ihcpu = 0xffff;
3042 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
3043 if (!ibs_enabled(vcpu)) {
3044 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
3045 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IBS);
3050 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
3051 if (ibs_enabled(vcpu)) {
3052 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
3053 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IBS);
3058 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
3059 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
3063 if (kvm_check_request(KVM_REQ_START_MIGRATION, vcpu)) {
3065 * Disable CMM virtualization; we will emulate the ESSA
3066 * instruction manually, in order to provide additional
3067 * functionalities needed for live migration.
3069 vcpu->arch.sie_block->ecb2 &= ~ECB2_CMMA;
3073 if (kvm_check_request(KVM_REQ_STOP_MIGRATION, vcpu)) {
3075 * Re-enable CMM virtualization if CMMA is available and
3076 * CMM has been used.
3078 if ((vcpu->kvm->arch.use_cmma) &&
3079 (vcpu->kvm->mm->context.uses_cmm))
3080 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
3084 /* nothing to do, just clear the request */
3085 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
3090 void kvm_s390_set_tod_clock(struct kvm *kvm,
3091 const struct kvm_s390_vm_tod_clock *gtod)
3093 struct kvm_vcpu *vcpu;
3094 struct kvm_s390_tod_clock_ext htod;
3097 mutex_lock(&kvm->lock);
3100 get_tod_clock_ext((char *)&htod);
3102 kvm->arch.epoch = gtod->tod - htod.tod;
3104 if (test_kvm_facility(kvm, 139)) {
3105 kvm->arch.epdx = gtod->epoch_idx - htod.epoch_idx;
3106 if (kvm->arch.epoch > gtod->tod)
3107 kvm->arch.epdx -= 1;
3110 kvm_s390_vcpu_block_all(kvm);
3111 kvm_for_each_vcpu(i, vcpu, kvm) {
3112 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
3113 vcpu->arch.sie_block->epdx = kvm->arch.epdx;
3116 kvm_s390_vcpu_unblock_all(kvm);
3118 mutex_unlock(&kvm->lock);
3122 * kvm_arch_fault_in_page - fault-in guest page if necessary
3123 * @vcpu: The corresponding virtual cpu
3124 * @gpa: Guest physical address
3125 * @writable: Whether the page should be writable or not
3127 * Make sure that a guest page has been faulted-in on the host.
3129 * Return: Zero on success, negative error code otherwise.
3131 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
3133 return gmap_fault(vcpu->arch.gmap, gpa,
3134 writable ? FAULT_FLAG_WRITE : 0);
3137 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
3138 unsigned long token)
3140 struct kvm_s390_interrupt inti;
3141 struct kvm_s390_irq irq;
3144 irq.u.ext.ext_params2 = token;
3145 irq.type = KVM_S390_INT_PFAULT_INIT;
3146 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
3148 inti.type = KVM_S390_INT_PFAULT_DONE;
3149 inti.parm64 = token;
3150 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
3154 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
3155 struct kvm_async_pf *work)
3157 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
3158 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
3161 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
3162 struct kvm_async_pf *work)
3164 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
3165 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
3168 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
3169 struct kvm_async_pf *work)
3171 /* s390 will always inject the page directly */
3174 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
3177 * s390 will always inject the page directly,
3178 * but we still want check_async_completion to cleanup
3183 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
3186 struct kvm_arch_async_pf arch;
3189 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
3191 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
3192 vcpu->arch.pfault_compare)
3194 if (psw_extint_disabled(vcpu))
3196 if (kvm_s390_vcpu_has_irq(vcpu, 0))
3198 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK))
3200 if (!vcpu->arch.gmap->pfault_enabled)
3203 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
3204 hva += current->thread.gmap_addr & ~PAGE_MASK;
3205 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
3208 rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
3212 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
3217 * On s390 notifications for arriving pages will be delivered directly
3218 * to the guest but the house keeping for completed pfaults is
3219 * handled outside the worker.
3221 kvm_check_async_pf_completion(vcpu);
3223 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
3224 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
3229 if (test_cpu_flag(CIF_MCCK_PENDING))
3232 if (!kvm_is_ucontrol(vcpu->kvm)) {
3233 rc = kvm_s390_deliver_pending_interrupts(vcpu);
3238 rc = kvm_s390_handle_requests(vcpu);
3242 if (guestdbg_enabled(vcpu)) {
3243 kvm_s390_backup_guest_per_regs(vcpu);
3244 kvm_s390_patch_guest_per_regs(vcpu);
3247 vcpu->arch.sie_block->icptcode = 0;
3248 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
3249 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
3250 trace_kvm_s390_sie_enter(vcpu, cpuflags);
3255 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
3257 struct kvm_s390_pgm_info pgm_info = {
3258 .code = PGM_ADDRESSING,
3263 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
3264 trace_kvm_s390_sie_fault(vcpu);
3267 * We want to inject an addressing exception, which is defined as a
3268 * suppressing or terminating exception. However, since we came here
3269 * by a DAT access exception, the PSW still points to the faulting
3270 * instruction since DAT exceptions are nullifying. So we've got
3271 * to look up the current opcode to get the length of the instruction
3272 * to be able to forward the PSW.
3274 rc = read_guest_instr(vcpu, vcpu->arch.sie_block->gpsw.addr, &opcode, 1);
3275 ilen = insn_length(opcode);
3279 /* Instruction-Fetching Exceptions - we can't detect the ilen.
3280 * Forward by arbitrary ilc, injection will take care of
3281 * nullification if necessary.
3283 pgm_info = vcpu->arch.pgm;
3286 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
3287 kvm_s390_forward_psw(vcpu, ilen);
3288 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
3291 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
3293 struct mcck_volatile_info *mcck_info;
3294 struct sie_page *sie_page;
3296 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
3297 vcpu->arch.sie_block->icptcode);
3298 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
3300 if (guestdbg_enabled(vcpu))
3301 kvm_s390_restore_guest_per_regs(vcpu);
3303 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
3304 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
3306 if (exit_reason == -EINTR) {
3307 VCPU_EVENT(vcpu, 3, "%s", "machine check");
3308 sie_page = container_of(vcpu->arch.sie_block,
3309 struct sie_page, sie_block);
3310 mcck_info = &sie_page->mcck_info;
3311 kvm_s390_reinject_machine_check(vcpu, mcck_info);
3315 if (vcpu->arch.sie_block->icptcode > 0) {
3316 int rc = kvm_handle_sie_intercept(vcpu);
3318 if (rc != -EOPNOTSUPP)
3320 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
3321 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
3322 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
3323 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
3325 } else if (exit_reason != -EFAULT) {
3326 vcpu->stat.exit_null++;
3328 } else if (kvm_is_ucontrol(vcpu->kvm)) {
3329 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
3330 vcpu->run->s390_ucontrol.trans_exc_code =
3331 current->thread.gmap_addr;
3332 vcpu->run->s390_ucontrol.pgm_code = 0x10;
3334 } else if (current->thread.gmap_pfault) {
3335 trace_kvm_s390_major_guest_pfault(vcpu);
3336 current->thread.gmap_pfault = 0;
3337 if (kvm_arch_setup_async_pf(vcpu))
3339 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
3341 return vcpu_post_run_fault_in_sie(vcpu);
3344 static int __vcpu_run(struct kvm_vcpu *vcpu)
3346 int rc, exit_reason;
3349 * We try to hold kvm->srcu during most of vcpu_run (except when run-
3350 * ning the guest), so that memslots (and other stuff) are protected
3352 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3355 rc = vcpu_pre_run(vcpu);
3359 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
3361 * As PF_VCPU will be used in fault handler, between
3362 * guest_enter and guest_exit should be no uaccess.
3364 local_irq_disable();
3365 guest_enter_irqoff();
3366 __disable_cpu_timer_accounting(vcpu);
3368 exit_reason = sie64a(vcpu->arch.sie_block,
3369 vcpu->run->s.regs.gprs);
3370 local_irq_disable();
3371 __enable_cpu_timer_accounting(vcpu);
3372 guest_exit_irqoff();
3374 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3376 rc = vcpu_post_run(vcpu, exit_reason);
3377 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
3379 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
3383 static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3385 struct runtime_instr_cb *riccb;
3388 riccb = (struct runtime_instr_cb *) &kvm_run->s.regs.riccb;
3389 gscb = (struct gs_cb *) &kvm_run->s.regs.gscb;
3390 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
3391 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
3392 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
3393 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
3394 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
3395 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
3396 /* some control register changes require a tlb flush */
3397 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
3399 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
3400 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
3401 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
3402 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
3403 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
3404 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
3406 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
3407 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
3408 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
3409 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
3410 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
3411 kvm_clear_async_pf_completion_queue(vcpu);
3414 * If userspace sets the riccb (e.g. after migration) to a valid state,
3415 * we should enable RI here instead of doing the lazy enablement.
3417 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) &&
3418 test_kvm_facility(vcpu->kvm, 64) &&
3420 !(vcpu->arch.sie_block->ecb3 & ECB3_RI)) {
3421 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (sync_regs)");
3422 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
3425 * If userspace sets the gscb (e.g. after migration) to non-zero,
3426 * we should enable GS here instead of doing the lazy enablement.
3428 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_GSCB) &&
3429 test_kvm_facility(vcpu->kvm, 133) &&
3431 !vcpu->arch.gs_enabled) {
3432 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (sync_regs)");
3433 vcpu->arch.sie_block->ecb |= ECB_GS;
3434 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
3435 vcpu->arch.gs_enabled = 1;
3437 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_BPBC) &&
3438 test_kvm_facility(vcpu->kvm, 82)) {
3439 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
3440 vcpu->arch.sie_block->fpf |= kvm_run->s.regs.bpbc ? FPF_BPBC : 0;
3442 save_access_regs(vcpu->arch.host_acrs);
3443 restore_access_regs(vcpu->run->s.regs.acrs);
3444 /* save host (userspace) fprs/vrs */
3446 vcpu->arch.host_fpregs.fpc = current->thread.fpu.fpc;
3447 vcpu->arch.host_fpregs.regs = current->thread.fpu.regs;
3449 current->thread.fpu.regs = vcpu->run->s.regs.vrs;
3451 current->thread.fpu.regs = vcpu->run->s.regs.fprs;
3452 current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
3453 if (test_fp_ctl(current->thread.fpu.fpc))
3454 /* User space provided an invalid FPC, let's clear it */
3455 current->thread.fpu.fpc = 0;
3456 if (MACHINE_HAS_GS) {
3458 __ctl_set_bit(2, 4);
3459 if (current->thread.gs_cb) {
3460 vcpu->arch.host_gscb = current->thread.gs_cb;
3461 save_gs_cb(vcpu->arch.host_gscb);
3463 if (vcpu->arch.gs_enabled) {
3464 current->thread.gs_cb = (struct gs_cb *)
3465 &vcpu->run->s.regs.gscb;
3466 restore_gs_cb(current->thread.gs_cb);
3471 kvm_run->kvm_dirty_regs = 0;
3474 static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3476 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
3477 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
3478 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
3479 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
3480 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
3481 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
3482 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
3483 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
3484 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
3485 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
3486 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
3487 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
3488 kvm_run->s.regs.bpbc = (vcpu->arch.sie_block->fpf & FPF_BPBC) == FPF_BPBC;
3489 save_access_regs(vcpu->run->s.regs.acrs);
3490 restore_access_regs(vcpu->arch.host_acrs);
3491 /* Save guest register state */
3493 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
3494 /* Restore will be done lazily at return */
3495 current->thread.fpu.fpc = vcpu->arch.host_fpregs.fpc;
3496 current->thread.fpu.regs = vcpu->arch.host_fpregs.regs;
3497 if (MACHINE_HAS_GS) {
3498 __ctl_set_bit(2, 4);
3499 if (vcpu->arch.gs_enabled)
3500 save_gs_cb(current->thread.gs_cb);
3502 current->thread.gs_cb = vcpu->arch.host_gscb;
3503 restore_gs_cb(vcpu->arch.host_gscb);
3505 if (!vcpu->arch.host_gscb)
3506 __ctl_clear_bit(2, 4);
3507 vcpu->arch.host_gscb = NULL;
3512 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3516 if (kvm_run->immediate_exit)
3521 if (guestdbg_exit_pending(vcpu)) {
3522 kvm_s390_prepare_debug_exit(vcpu);
3527 kvm_sigset_activate(vcpu);
3529 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
3530 kvm_s390_vcpu_start(vcpu);
3531 } else if (is_vcpu_stopped(vcpu)) {
3532 pr_err_ratelimited("can't run stopped vcpu %d\n",
3538 sync_regs(vcpu, kvm_run);
3539 enable_cpu_timer_accounting(vcpu);
3542 rc = __vcpu_run(vcpu);
3544 if (signal_pending(current) && !rc) {
3545 kvm_run->exit_reason = KVM_EXIT_INTR;
3549 if (guestdbg_exit_pending(vcpu) && !rc) {
3550 kvm_s390_prepare_debug_exit(vcpu);
3554 if (rc == -EREMOTE) {
3555 /* userspace support is needed, kvm_run has been prepared */
3559 disable_cpu_timer_accounting(vcpu);
3560 store_regs(vcpu, kvm_run);
3562 kvm_sigset_deactivate(vcpu);
3564 vcpu->stat.exit_userspace++;
3571 * store status at address
3572 * we use have two special cases:
3573 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
3574 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
3576 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
3578 unsigned char archmode = 1;
3579 freg_t fprs[NUM_FPRS];
3584 px = kvm_s390_get_prefix(vcpu);
3585 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
3586 if (write_guest_abs(vcpu, 163, &archmode, 1))
3589 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
3590 if (write_guest_real(vcpu, 163, &archmode, 1))
3594 gpa -= __LC_FPREGS_SAVE_AREA;
3596 /* manually convert vector registers if necessary */
3597 if (MACHINE_HAS_VX) {
3598 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
3599 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
3602 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
3603 vcpu->run->s.regs.fprs, 128);
3605 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
3606 vcpu->run->s.regs.gprs, 128);
3607 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
3608 &vcpu->arch.sie_block->gpsw, 16);
3609 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
3611 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
3612 &vcpu->run->s.regs.fpc, 4);
3613 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
3614 &vcpu->arch.sie_block->todpr, 4);
3615 cputm = kvm_s390_get_cpu_timer(vcpu);
3616 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
3618 clkcomp = vcpu->arch.sie_block->ckc >> 8;
3619 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
3621 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
3622 &vcpu->run->s.regs.acrs, 64);
3623 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
3624 &vcpu->arch.sie_block->gcr, 128);
3625 return rc ? -EFAULT : 0;
3628 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
3631 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
3632 * switch in the run ioctl. Let's update our copies before we save
3633 * it into the save area
3636 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
3637 save_access_regs(vcpu->run->s.regs.acrs);
3639 return kvm_s390_store_status_unloaded(vcpu, addr);
3642 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
3644 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
3645 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
3648 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
3651 struct kvm_vcpu *vcpu;
3653 kvm_for_each_vcpu(i, vcpu, kvm) {
3654 __disable_ibs_on_vcpu(vcpu);
3658 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
3662 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
3663 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
3666 void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
3668 int i, online_vcpus, started_vcpus = 0;
3670 if (!is_vcpu_stopped(vcpu))
3673 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
3674 /* Only one cpu at a time may enter/leave the STOPPED state. */
3675 spin_lock(&vcpu->kvm->arch.start_stop_lock);
3676 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
3678 for (i = 0; i < online_vcpus; i++) {
3679 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i]))
3683 if (started_vcpus == 0) {
3684 /* we're the only active VCPU -> speed it up */
3685 __enable_ibs_on_vcpu(vcpu);
3686 } else if (started_vcpus == 1) {
3688 * As we are starting a second VCPU, we have to disable
3689 * the IBS facility on all VCPUs to remove potentially
3690 * oustanding ENABLE requests.
3692 __disable_ibs_on_all_vcpus(vcpu->kvm);
3695 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_STOPPED);
3697 * Another VCPU might have used IBS while we were offline.
3698 * Let's play safe and flush the VCPU at startup.
3700 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
3701 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
3705 void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
3707 int i, online_vcpus, started_vcpus = 0;
3708 struct kvm_vcpu *started_vcpu = NULL;
3710 if (is_vcpu_stopped(vcpu))
3713 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
3714 /* Only one cpu at a time may enter/leave the STOPPED state. */
3715 spin_lock(&vcpu->kvm->arch.start_stop_lock);
3716 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
3718 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
3719 kvm_s390_clear_stop_irq(vcpu);
3721 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOPPED);
3722 __disable_ibs_on_vcpu(vcpu);
3724 for (i = 0; i < online_vcpus; i++) {
3725 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) {
3727 started_vcpu = vcpu->kvm->vcpus[i];
3731 if (started_vcpus == 1) {
3733 * As we only have one VCPU left, we want to enable the
3734 * IBS facility for that VCPU to speed it up.
3736 __enable_ibs_on_vcpu(started_vcpu);
3739 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
3743 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3744 struct kvm_enable_cap *cap)
3752 case KVM_CAP_S390_CSS_SUPPORT:
3753 if (!vcpu->kvm->arch.css_support) {
3754 vcpu->kvm->arch.css_support = 1;
3755 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
3756 trace_kvm_s390_enable_css(vcpu->kvm);
3767 static long kvm_s390_guest_mem_op(struct kvm_vcpu *vcpu,
3768 struct kvm_s390_mem_op *mop)
3770 void __user *uaddr = (void __user *)mop->buf;
3771 void *tmpbuf = NULL;
3773 const u64 supported_flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION
3774 | KVM_S390_MEMOP_F_CHECK_ONLY;
3776 if (mop->flags & ~supported_flags)
3779 if (mop->size > MEM_OP_MAX_SIZE)
3782 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
3783 tmpbuf = vmalloc(mop->size);
3788 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3791 case KVM_S390_MEMOP_LOGICAL_READ:
3792 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3793 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3794 mop->size, GACC_FETCH);
3797 r = read_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3799 if (copy_to_user(uaddr, tmpbuf, mop->size))
3803 case KVM_S390_MEMOP_LOGICAL_WRITE:
3804 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3805 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3806 mop->size, GACC_STORE);
3809 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
3813 r = write_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3819 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
3821 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
3822 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
3828 long kvm_arch_vcpu_async_ioctl(struct file *filp,
3829 unsigned int ioctl, unsigned long arg)
3831 struct kvm_vcpu *vcpu = filp->private_data;
3832 void __user *argp = (void __user *)arg;
3835 case KVM_S390_IRQ: {
3836 struct kvm_s390_irq s390irq;
3838 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
3840 return kvm_s390_inject_vcpu(vcpu, &s390irq);
3842 case KVM_S390_INTERRUPT: {
3843 struct kvm_s390_interrupt s390int;
3844 struct kvm_s390_irq s390irq;
3846 if (copy_from_user(&s390int, argp, sizeof(s390int)))
3848 if (s390int_to_s390irq(&s390int, &s390irq))
3850 return kvm_s390_inject_vcpu(vcpu, &s390irq);
3853 return -ENOIOCTLCMD;
3856 long kvm_arch_vcpu_ioctl(struct file *filp,
3857 unsigned int ioctl, unsigned long arg)
3859 struct kvm_vcpu *vcpu = filp->private_data;
3860 void __user *argp = (void __user *)arg;
3867 case KVM_S390_STORE_STATUS:
3868 idx = srcu_read_lock(&vcpu->kvm->srcu);
3869 r = kvm_s390_vcpu_store_status(vcpu, arg);
3870 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3872 case KVM_S390_SET_INITIAL_PSW: {
3876 if (copy_from_user(&psw, argp, sizeof(psw)))
3878 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
3881 case KVM_S390_INITIAL_RESET:
3882 r = kvm_arch_vcpu_ioctl_initial_reset(vcpu);
3884 case KVM_SET_ONE_REG:
3885 case KVM_GET_ONE_REG: {
3886 struct kvm_one_reg reg;
3888 if (copy_from_user(®, argp, sizeof(reg)))
3890 if (ioctl == KVM_SET_ONE_REG)
3891 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®);
3893 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®);
3896 #ifdef CONFIG_KVM_S390_UCONTROL
3897 case KVM_S390_UCAS_MAP: {
3898 struct kvm_s390_ucas_mapping ucasmap;
3900 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3905 if (!kvm_is_ucontrol(vcpu->kvm)) {
3910 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
3911 ucasmap.vcpu_addr, ucasmap.length);
3914 case KVM_S390_UCAS_UNMAP: {
3915 struct kvm_s390_ucas_mapping ucasmap;
3917 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3922 if (!kvm_is_ucontrol(vcpu->kvm)) {
3927 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
3932 case KVM_S390_VCPU_FAULT: {
3933 r = gmap_fault(vcpu->arch.gmap, arg, 0);
3936 case KVM_ENABLE_CAP:
3938 struct kvm_enable_cap cap;
3940 if (copy_from_user(&cap, argp, sizeof(cap)))
3942 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3945 case KVM_S390_MEM_OP: {
3946 struct kvm_s390_mem_op mem_op;
3948 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
3949 r = kvm_s390_guest_mem_op(vcpu, &mem_op);
3954 case KVM_S390_SET_IRQ_STATE: {
3955 struct kvm_s390_irq_state irq_state;
3958 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3960 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
3961 irq_state.len == 0 ||
3962 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
3966 /* do not use irq_state.flags, it will break old QEMUs */
3967 r = kvm_s390_set_irq_state(vcpu,
3968 (void __user *) irq_state.buf,
3972 case KVM_S390_GET_IRQ_STATE: {
3973 struct kvm_s390_irq_state irq_state;
3976 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3978 if (irq_state.len == 0) {
3982 /* do not use irq_state.flags, it will break old QEMUs */
3983 r = kvm_s390_get_irq_state(vcpu,
3984 (__u8 __user *) irq_state.buf,
3996 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3998 #ifdef CONFIG_KVM_S390_UCONTROL
3999 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
4000 && (kvm_is_ucontrol(vcpu->kvm))) {
4001 vmf->page = virt_to_page(vcpu->arch.sie_block);
4002 get_page(vmf->page);
4006 return VM_FAULT_SIGBUS;
4009 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
4010 unsigned long npages)
4015 /* Section: memory related */
4016 int kvm_arch_prepare_memory_region(struct kvm *kvm,
4017 struct kvm_memory_slot *memslot,
4018 const struct kvm_userspace_memory_region *mem,
4019 enum kvm_mr_change change)
4021 /* A few sanity checks. We can have memory slots which have to be
4022 located/ended at a segment boundary (1MB). The memory in userland is
4023 ok to be fragmented into various different vmas. It is okay to mmap()
4024 and munmap() stuff in this slot after doing this call at any time */
4026 if (mem->userspace_addr & 0xffffful)
4029 if (mem->memory_size & 0xffffful)
4032 if (mem->guest_phys_addr + mem->memory_size > kvm->arch.mem_limit)
4038 void kvm_arch_commit_memory_region(struct kvm *kvm,
4039 const struct kvm_userspace_memory_region *mem,
4040 const struct kvm_memory_slot *old,
4041 const struct kvm_memory_slot *new,
4042 enum kvm_mr_change change)
4046 /* If the basics of the memslot do not change, we do not want
4047 * to update the gmap. Every update causes several unnecessary
4048 * segment translation exceptions. This is usually handled just
4049 * fine by the normal fault handler + gmap, but it will also
4050 * cause faults on the prefix page of running guest CPUs.
4052 if (old->userspace_addr == mem->userspace_addr &&
4053 old->base_gfn * PAGE_SIZE == mem->guest_phys_addr &&
4054 old->npages * PAGE_SIZE == mem->memory_size)
4057 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr,
4058 mem->guest_phys_addr, mem->memory_size);
4060 pr_warn("failed to commit memory region\n");
4064 static inline unsigned long nonhyp_mask(int i)
4066 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
4068 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
4071 void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu)
4073 vcpu->valid_wakeup = false;
4076 static int __init kvm_s390_init(void)
4080 if (!sclp.has_sief2) {
4081 pr_info("SIE not available\n");
4085 for (i = 0; i < 16; i++)
4086 kvm_s390_fac_base[i] |=
4087 S390_lowcore.stfle_fac_list[i] & nonhyp_mask(i);
4089 return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
4092 static void __exit kvm_s390_exit(void)
4097 module_init(kvm_s390_init);
4098 module_exit(kvm_s390_exit);
4101 * Enable autoloading of the kvm module.
4102 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
4103 * since x86 takes a different approach.
4105 #include <linux/miscdevice.h>
4106 MODULE_ALIAS_MISCDEV(KVM_MINOR);
4107 MODULE_ALIAS("devname:kvm");