1 // SPDX-License-Identifier: GPL-2.0
3 * hosting IBM Z kernel virtual machines (s390x)
5 * Copyright IBM Corp. 2008, 2020
7 * Author(s): Carsten Otte <cotte@de.ibm.com>
8 * Christian Borntraeger <borntraeger@de.ibm.com>
9 * Christian Ehrhardt <ehrhardt@de.ibm.com>
10 * Jason J. Herne <jjherne@us.ibm.com>
13 #define KMSG_COMPONENT "kvm-s390"
14 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
16 #include <linux/compiler.h>
17 #include <linux/err.h>
19 #include <linux/hrtimer.h>
20 #include <linux/init.h>
21 #include <linux/kvm.h>
22 #include <linux/kvm_host.h>
23 #include <linux/mman.h>
24 #include <linux/module.h>
25 #include <linux/moduleparam.h>
26 #include <linux/random.h>
27 #include <linux/slab.h>
28 #include <linux/timer.h>
29 #include <linux/vmalloc.h>
30 #include <linux/bitmap.h>
31 #include <linux/sched/signal.h>
32 #include <linux/string.h>
33 #include <linux/pgtable.h>
34 #include <linux/mmu_notifier.h>
36 #include <asm/asm-offsets.h>
37 #include <asm/lowcore.h>
41 #include <asm/switch_to.h>
44 #include <asm/cpacf.h>
45 #include <asm/timex.h>
48 #include <asm/fpu/api.h>
53 #define CREATE_TRACE_POINTS
55 #include "trace-s390.h"
57 #define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */
59 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
60 (KVM_MAX_VCPUS + LOCAL_IRQS))
62 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
63 KVM_GENERIC_VM_STATS(),
64 STATS_DESC_COUNTER(VM, inject_io),
65 STATS_DESC_COUNTER(VM, inject_float_mchk),
66 STATS_DESC_COUNTER(VM, inject_pfault_done),
67 STATS_DESC_COUNTER(VM, inject_service_signal),
68 STATS_DESC_COUNTER(VM, inject_virtio),
69 STATS_DESC_COUNTER(VM, aen_forward),
70 STATS_DESC_COUNTER(VM, gmap_shadow_reuse),
71 STATS_DESC_COUNTER(VM, gmap_shadow_create),
72 STATS_DESC_COUNTER(VM, gmap_shadow_r1_entry),
73 STATS_DESC_COUNTER(VM, gmap_shadow_r2_entry),
74 STATS_DESC_COUNTER(VM, gmap_shadow_r3_entry),
75 STATS_DESC_COUNTER(VM, gmap_shadow_sg_entry),
76 STATS_DESC_COUNTER(VM, gmap_shadow_pg_entry),
79 const struct kvm_stats_header kvm_vm_stats_header = {
80 .name_size = KVM_STATS_NAME_SIZE,
81 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
82 .id_offset = sizeof(struct kvm_stats_header),
83 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
84 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
85 sizeof(kvm_vm_stats_desc),
88 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
89 KVM_GENERIC_VCPU_STATS(),
90 STATS_DESC_COUNTER(VCPU, exit_userspace),
91 STATS_DESC_COUNTER(VCPU, exit_null),
92 STATS_DESC_COUNTER(VCPU, exit_external_request),
93 STATS_DESC_COUNTER(VCPU, exit_io_request),
94 STATS_DESC_COUNTER(VCPU, exit_external_interrupt),
95 STATS_DESC_COUNTER(VCPU, exit_stop_request),
96 STATS_DESC_COUNTER(VCPU, exit_validity),
97 STATS_DESC_COUNTER(VCPU, exit_instruction),
98 STATS_DESC_COUNTER(VCPU, exit_pei),
99 STATS_DESC_COUNTER(VCPU, halt_no_poll_steal),
100 STATS_DESC_COUNTER(VCPU, instruction_lctl),
101 STATS_DESC_COUNTER(VCPU, instruction_lctlg),
102 STATS_DESC_COUNTER(VCPU, instruction_stctl),
103 STATS_DESC_COUNTER(VCPU, instruction_stctg),
104 STATS_DESC_COUNTER(VCPU, exit_program_interruption),
105 STATS_DESC_COUNTER(VCPU, exit_instr_and_program),
106 STATS_DESC_COUNTER(VCPU, exit_operation_exception),
107 STATS_DESC_COUNTER(VCPU, deliver_ckc),
108 STATS_DESC_COUNTER(VCPU, deliver_cputm),
109 STATS_DESC_COUNTER(VCPU, deliver_external_call),
110 STATS_DESC_COUNTER(VCPU, deliver_emergency_signal),
111 STATS_DESC_COUNTER(VCPU, deliver_service_signal),
112 STATS_DESC_COUNTER(VCPU, deliver_virtio),
113 STATS_DESC_COUNTER(VCPU, deliver_stop_signal),
114 STATS_DESC_COUNTER(VCPU, deliver_prefix_signal),
115 STATS_DESC_COUNTER(VCPU, deliver_restart_signal),
116 STATS_DESC_COUNTER(VCPU, deliver_program),
117 STATS_DESC_COUNTER(VCPU, deliver_io),
118 STATS_DESC_COUNTER(VCPU, deliver_machine_check),
119 STATS_DESC_COUNTER(VCPU, exit_wait_state),
120 STATS_DESC_COUNTER(VCPU, inject_ckc),
121 STATS_DESC_COUNTER(VCPU, inject_cputm),
122 STATS_DESC_COUNTER(VCPU, inject_external_call),
123 STATS_DESC_COUNTER(VCPU, inject_emergency_signal),
124 STATS_DESC_COUNTER(VCPU, inject_mchk),
125 STATS_DESC_COUNTER(VCPU, inject_pfault_init),
126 STATS_DESC_COUNTER(VCPU, inject_program),
127 STATS_DESC_COUNTER(VCPU, inject_restart),
128 STATS_DESC_COUNTER(VCPU, inject_set_prefix),
129 STATS_DESC_COUNTER(VCPU, inject_stop_signal),
130 STATS_DESC_COUNTER(VCPU, instruction_epsw),
131 STATS_DESC_COUNTER(VCPU, instruction_gs),
132 STATS_DESC_COUNTER(VCPU, instruction_io_other),
133 STATS_DESC_COUNTER(VCPU, instruction_lpsw),
134 STATS_DESC_COUNTER(VCPU, instruction_lpswe),
135 STATS_DESC_COUNTER(VCPU, instruction_pfmf),
136 STATS_DESC_COUNTER(VCPU, instruction_ptff),
137 STATS_DESC_COUNTER(VCPU, instruction_sck),
138 STATS_DESC_COUNTER(VCPU, instruction_sckpf),
139 STATS_DESC_COUNTER(VCPU, instruction_stidp),
140 STATS_DESC_COUNTER(VCPU, instruction_spx),
141 STATS_DESC_COUNTER(VCPU, instruction_stpx),
142 STATS_DESC_COUNTER(VCPU, instruction_stap),
143 STATS_DESC_COUNTER(VCPU, instruction_iske),
144 STATS_DESC_COUNTER(VCPU, instruction_ri),
145 STATS_DESC_COUNTER(VCPU, instruction_rrbe),
146 STATS_DESC_COUNTER(VCPU, instruction_sske),
147 STATS_DESC_COUNTER(VCPU, instruction_ipte_interlock),
148 STATS_DESC_COUNTER(VCPU, instruction_stsi),
149 STATS_DESC_COUNTER(VCPU, instruction_stfl),
150 STATS_DESC_COUNTER(VCPU, instruction_tb),
151 STATS_DESC_COUNTER(VCPU, instruction_tpi),
152 STATS_DESC_COUNTER(VCPU, instruction_tprot),
153 STATS_DESC_COUNTER(VCPU, instruction_tsch),
154 STATS_DESC_COUNTER(VCPU, instruction_sie),
155 STATS_DESC_COUNTER(VCPU, instruction_essa),
156 STATS_DESC_COUNTER(VCPU, instruction_sthyi),
157 STATS_DESC_COUNTER(VCPU, instruction_sigp_sense),
158 STATS_DESC_COUNTER(VCPU, instruction_sigp_sense_running),
159 STATS_DESC_COUNTER(VCPU, instruction_sigp_external_call),
160 STATS_DESC_COUNTER(VCPU, instruction_sigp_emergency),
161 STATS_DESC_COUNTER(VCPU, instruction_sigp_cond_emergency),
162 STATS_DESC_COUNTER(VCPU, instruction_sigp_start),
163 STATS_DESC_COUNTER(VCPU, instruction_sigp_stop),
164 STATS_DESC_COUNTER(VCPU, instruction_sigp_stop_store_status),
165 STATS_DESC_COUNTER(VCPU, instruction_sigp_store_status),
166 STATS_DESC_COUNTER(VCPU, instruction_sigp_store_adtl_status),
167 STATS_DESC_COUNTER(VCPU, instruction_sigp_arch),
168 STATS_DESC_COUNTER(VCPU, instruction_sigp_prefix),
169 STATS_DESC_COUNTER(VCPU, instruction_sigp_restart),
170 STATS_DESC_COUNTER(VCPU, instruction_sigp_init_cpu_reset),
171 STATS_DESC_COUNTER(VCPU, instruction_sigp_cpu_reset),
172 STATS_DESC_COUNTER(VCPU, instruction_sigp_unknown),
173 STATS_DESC_COUNTER(VCPU, instruction_diagnose_10),
174 STATS_DESC_COUNTER(VCPU, instruction_diagnose_44),
175 STATS_DESC_COUNTER(VCPU, instruction_diagnose_9c),
176 STATS_DESC_COUNTER(VCPU, diag_9c_ignored),
177 STATS_DESC_COUNTER(VCPU, diag_9c_forward),
178 STATS_DESC_COUNTER(VCPU, instruction_diagnose_258),
179 STATS_DESC_COUNTER(VCPU, instruction_diagnose_308),
180 STATS_DESC_COUNTER(VCPU, instruction_diagnose_500),
181 STATS_DESC_COUNTER(VCPU, instruction_diagnose_other),
182 STATS_DESC_COUNTER(VCPU, pfault_sync)
185 const struct kvm_stats_header kvm_vcpu_stats_header = {
186 .name_size = KVM_STATS_NAME_SIZE,
187 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
188 .id_offset = sizeof(struct kvm_stats_header),
189 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
190 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
191 sizeof(kvm_vcpu_stats_desc),
194 /* allow nested virtualization in KVM (if enabled by user space) */
196 module_param(nested, int, S_IRUGO);
197 MODULE_PARM_DESC(nested, "Nested virtualization support");
199 /* allow 1m huge page guest backing, if !nested */
201 module_param(hpage, int, 0444);
202 MODULE_PARM_DESC(hpage, "1m huge page backing support");
204 /* maximum percentage of steal time for polling. >100 is treated like 100 */
205 static u8 halt_poll_max_steal = 10;
206 module_param(halt_poll_max_steal, byte, 0644);
207 MODULE_PARM_DESC(halt_poll_max_steal, "Maximum percentage of steal time to allow polling");
209 /* if set to true, the GISA will be initialized and used if available */
210 static bool use_gisa = true;
211 module_param(use_gisa, bool, 0644);
212 MODULE_PARM_DESC(use_gisa, "Use the GISA if the host supports it.");
214 /* maximum diag9c forwarding per second */
215 unsigned int diag9c_forwarding_hz;
216 module_param(diag9c_forwarding_hz, uint, 0644);
217 MODULE_PARM_DESC(diag9c_forwarding_hz, "Maximum diag9c forwarding per second, 0 to turn off");
220 * allow asynchronous deinit for protected guests; enable by default since
221 * the feature is opt-in anyway
223 static int async_destroy = 1;
224 module_param(async_destroy, int, 0444);
225 MODULE_PARM_DESC(async_destroy, "Asynchronous destroy for protected guests");
228 * For now we handle at most 16 double words as this is what the s390 base
229 * kernel handles and stores in the prefix page. If we ever need to go beyond
230 * this, this requires changes to code, but the external uapi can stay.
232 #define SIZE_INTERNAL 16
235 * Base feature mask that defines default mask for facilities. Consists of the
236 * defines in FACILITIES_KVM and the non-hypervisor managed bits.
238 static unsigned long kvm_s390_fac_base[SIZE_INTERNAL] = { FACILITIES_KVM };
240 * Extended feature mask. Consists of the defines in FACILITIES_KVM_CPUMODEL
241 * and defines the facilities that can be enabled via a cpu model.
243 static unsigned long kvm_s390_fac_ext[SIZE_INTERNAL] = { FACILITIES_KVM_CPUMODEL };
245 static unsigned long kvm_s390_fac_size(void)
247 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_MASK_SIZE_U64);
248 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_LIST_SIZE_U64);
249 BUILD_BUG_ON(SIZE_INTERNAL * sizeof(unsigned long) >
250 sizeof(stfle_fac_list));
252 return SIZE_INTERNAL;
255 /* available cpu features supported by kvm */
256 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
257 /* available subfunctions indicated via query / "test bit" */
258 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
260 static struct gmap_notifier gmap_notifier;
261 static struct gmap_notifier vsie_gmap_notifier;
262 debug_info_t *kvm_s390_dbf;
263 debug_info_t *kvm_s390_dbf_uv;
265 /* Section: not file related */
266 /* forward declarations */
267 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
269 static int sca_switch_to_extended(struct kvm *kvm);
271 static void kvm_clock_sync_scb(struct kvm_s390_sie_block *scb, u64 delta)
276 * The TOD jumps by delta, we have to compensate this by adding
277 * -delta to the epoch.
281 /* sign-extension - we're adding to signed values below */
286 if (scb->ecd & ECD_MEF) {
287 scb->epdx += delta_idx;
288 if (scb->epoch < delta)
294 * This callback is executed during stop_machine(). All CPUs are therefore
295 * temporarily stopped. In order not to change guest behavior, we have to
296 * disable preemption whenever we touch the epoch of kvm and the VCPUs,
297 * so a CPU won't be stopped while calculating with the epoch.
299 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
303 struct kvm_vcpu *vcpu;
305 unsigned long long *delta = v;
307 list_for_each_entry(kvm, &vm_list, vm_list) {
308 kvm_for_each_vcpu(i, vcpu, kvm) {
309 kvm_clock_sync_scb(vcpu->arch.sie_block, *delta);
311 kvm->arch.epoch = vcpu->arch.sie_block->epoch;
312 kvm->arch.epdx = vcpu->arch.sie_block->epdx;
314 if (vcpu->arch.cputm_enabled)
315 vcpu->arch.cputm_start += *delta;
316 if (vcpu->arch.vsie_block)
317 kvm_clock_sync_scb(vcpu->arch.vsie_block,
324 static struct notifier_block kvm_clock_notifier = {
325 .notifier_call = kvm_clock_sync,
328 static void allow_cpu_feat(unsigned long nr)
330 set_bit_inv(nr, kvm_s390_available_cpu_feat);
333 static inline int plo_test_bit(unsigned char nr)
335 unsigned long function = (unsigned long)nr | 0x100;
339 " lgr 0,%[function]\n"
340 /* Parameter registers are ignored for "test bit" */
345 : [function] "d" (function)
350 static __always_inline void __insn32_query(unsigned int opcode, u8 *query)
355 /* Parameter registers are ignored */
356 " .insn rrf,%[opc] << 16,2,4,6,0\n"
358 : [query] "d" ((unsigned long)query), [opc] "i" (opcode)
359 : "cc", "memory", "0", "1");
362 #define INSN_SORTL 0xb938
363 #define INSN_DFLTCC 0xb939
365 static void __init kvm_s390_cpu_feat_init(void)
369 for (i = 0; i < 256; ++i) {
371 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
374 if (test_facility(28)) /* TOD-clock steering */
375 ptff(kvm_s390_available_subfunc.ptff,
376 sizeof(kvm_s390_available_subfunc.ptff),
379 if (test_facility(17)) { /* MSA */
380 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
381 kvm_s390_available_subfunc.kmac);
382 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
383 kvm_s390_available_subfunc.kmc);
384 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
385 kvm_s390_available_subfunc.km);
386 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
387 kvm_s390_available_subfunc.kimd);
388 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
389 kvm_s390_available_subfunc.klmd);
391 if (test_facility(76)) /* MSA3 */
392 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
393 kvm_s390_available_subfunc.pckmo);
394 if (test_facility(77)) { /* MSA4 */
395 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
396 kvm_s390_available_subfunc.kmctr);
397 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
398 kvm_s390_available_subfunc.kmf);
399 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
400 kvm_s390_available_subfunc.kmo);
401 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
402 kvm_s390_available_subfunc.pcc);
404 if (test_facility(57)) /* MSA5 */
405 __cpacf_query(CPACF_PRNO, (cpacf_mask_t *)
406 kvm_s390_available_subfunc.ppno);
408 if (test_facility(146)) /* MSA8 */
409 __cpacf_query(CPACF_KMA, (cpacf_mask_t *)
410 kvm_s390_available_subfunc.kma);
412 if (test_facility(155)) /* MSA9 */
413 __cpacf_query(CPACF_KDSA, (cpacf_mask_t *)
414 kvm_s390_available_subfunc.kdsa);
416 if (test_facility(150)) /* SORTL */
417 __insn32_query(INSN_SORTL, kvm_s390_available_subfunc.sortl);
419 if (test_facility(151)) /* DFLTCC */
420 __insn32_query(INSN_DFLTCC, kvm_s390_available_subfunc.dfltcc);
422 if (MACHINE_HAS_ESOP)
423 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
425 * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
426 * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
428 if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
429 !test_facility(3) || !nested)
431 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
432 if (sclp.has_64bscao)
433 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
435 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
437 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
439 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
441 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
443 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
445 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
447 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_KSS);
449 * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
450 * all skey handling functions read/set the skey from the PGSTE
451 * instead of the real storage key.
453 * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
454 * pages being detected as preserved although they are resident.
456 * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
457 * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
459 * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
460 * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
461 * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
463 * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
464 * cannot easily shadow the SCA because of the ipte lock.
468 static int __init __kvm_s390_init(void)
472 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
476 kvm_s390_dbf_uv = debug_register("kvm-uv", 32, 1, 7 * sizeof(long));
477 if (!kvm_s390_dbf_uv)
480 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view) ||
481 debug_register_view(kvm_s390_dbf_uv, &debug_sprintf_view))
484 kvm_s390_cpu_feat_init();
486 /* Register floating interrupt controller interface. */
487 rc = kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
489 pr_err("A FLIC registration call failed with rc=%d\n", rc);
493 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
494 rc = kvm_s390_pci_init();
496 pr_err("Unable to allocate AIFT for PCI\n");
501 rc = kvm_s390_gib_init(GAL_ISC);
505 gmap_notifier.notifier_call = kvm_gmap_notifier;
506 gmap_register_pte_notifier(&gmap_notifier);
507 vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
508 gmap_register_pte_notifier(&vsie_gmap_notifier);
509 atomic_notifier_chain_register(&s390_epoch_delta_notifier,
510 &kvm_clock_notifier);
515 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
520 debug_unregister(kvm_s390_dbf_uv);
522 debug_unregister(kvm_s390_dbf);
526 static void __kvm_s390_exit(void)
528 gmap_unregister_pte_notifier(&gmap_notifier);
529 gmap_unregister_pte_notifier(&vsie_gmap_notifier);
530 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
531 &kvm_clock_notifier);
533 kvm_s390_gib_destroy();
534 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
536 debug_unregister(kvm_s390_dbf);
537 debug_unregister(kvm_s390_dbf_uv);
540 /* Section: device related */
541 long kvm_arch_dev_ioctl(struct file *filp,
542 unsigned int ioctl, unsigned long arg)
544 if (ioctl == KVM_S390_ENABLE_SIE)
545 return s390_enable_sie();
549 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
554 case KVM_CAP_S390_PSW:
555 case KVM_CAP_S390_GMAP:
556 case KVM_CAP_SYNC_MMU:
557 #ifdef CONFIG_KVM_S390_UCONTROL
558 case KVM_CAP_S390_UCONTROL:
560 case KVM_CAP_ASYNC_PF:
561 case KVM_CAP_SYNC_REGS:
562 case KVM_CAP_ONE_REG:
563 case KVM_CAP_ENABLE_CAP:
564 case KVM_CAP_S390_CSS_SUPPORT:
565 case KVM_CAP_IOEVENTFD:
566 case KVM_CAP_DEVICE_CTRL:
567 case KVM_CAP_S390_IRQCHIP:
568 case KVM_CAP_VM_ATTRIBUTES:
569 case KVM_CAP_MP_STATE:
570 case KVM_CAP_IMMEDIATE_EXIT:
571 case KVM_CAP_S390_INJECT_IRQ:
572 case KVM_CAP_S390_USER_SIGP:
573 case KVM_CAP_S390_USER_STSI:
574 case KVM_CAP_S390_SKEYS:
575 case KVM_CAP_S390_IRQ_STATE:
576 case KVM_CAP_S390_USER_INSTR0:
577 case KVM_CAP_S390_CMMA_MIGRATION:
578 case KVM_CAP_S390_AIS:
579 case KVM_CAP_S390_AIS_MIGRATION:
580 case KVM_CAP_S390_VCPU_RESETS:
581 case KVM_CAP_SET_GUEST_DEBUG:
582 case KVM_CAP_S390_DIAG318:
583 case KVM_CAP_IRQFD_RESAMPLE:
586 case KVM_CAP_SET_GUEST_DEBUG2:
587 r = KVM_GUESTDBG_VALID_MASK;
589 case KVM_CAP_S390_HPAGE_1M:
591 if (hpage && !kvm_is_ucontrol(kvm))
594 case KVM_CAP_S390_MEM_OP:
597 case KVM_CAP_S390_MEM_OP_EXTENSION:
599 * Flag bits indicating which extensions are supported.
600 * If r > 0, the base extension must also be supported/indicated,
601 * in order to maintain backwards compatibility.
603 r = KVM_S390_MEMOP_EXTENSION_CAP_BASE |
604 KVM_S390_MEMOP_EXTENSION_CAP_CMPXCHG;
606 case KVM_CAP_NR_VCPUS:
607 case KVM_CAP_MAX_VCPUS:
608 case KVM_CAP_MAX_VCPU_ID:
609 r = KVM_S390_BSCA_CPU_SLOTS;
610 if (!kvm_s390_use_sca_entries())
612 else if (sclp.has_esca && sclp.has_64bscao)
613 r = KVM_S390_ESCA_CPU_SLOTS;
614 if (ext == KVM_CAP_NR_VCPUS)
615 r = min_t(unsigned int, num_online_cpus(), r);
617 case KVM_CAP_S390_COW:
618 r = MACHINE_HAS_ESOP;
620 case KVM_CAP_S390_VECTOR_REGISTERS:
623 case KVM_CAP_S390_RI:
624 r = test_facility(64);
626 case KVM_CAP_S390_GS:
627 r = test_facility(133);
629 case KVM_CAP_S390_BPB:
630 r = test_facility(82);
632 case KVM_CAP_S390_PROTECTED_ASYNC_DISABLE:
633 r = async_destroy && is_prot_virt_host();
635 case KVM_CAP_S390_PROTECTED:
636 r = is_prot_virt_host();
638 case KVM_CAP_S390_PROTECTED_DUMP: {
639 u64 pv_cmds_dump[] = {
640 BIT_UVC_CMD_DUMP_INIT,
641 BIT_UVC_CMD_DUMP_CONFIG_STOR_STATE,
642 BIT_UVC_CMD_DUMP_CPU,
643 BIT_UVC_CMD_DUMP_COMPLETE,
647 r = is_prot_virt_host();
649 for (i = 0; i < ARRAY_SIZE(pv_cmds_dump); i++) {
650 if (!test_bit_inv(pv_cmds_dump[i],
651 (unsigned long *)&uv_info.inst_calls_list)) {
658 case KVM_CAP_S390_ZPCI_OP:
659 r = kvm_s390_pci_interp_allowed();
661 case KVM_CAP_S390_CPU_TOPOLOGY:
662 r = test_facility(11);
670 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
673 gfn_t cur_gfn, last_gfn;
674 unsigned long gaddr, vmaddr;
675 struct gmap *gmap = kvm->arch.gmap;
676 DECLARE_BITMAP(bitmap, _PAGE_ENTRIES);
678 /* Loop over all guest segments */
679 cur_gfn = memslot->base_gfn;
680 last_gfn = memslot->base_gfn + memslot->npages;
681 for (; cur_gfn <= last_gfn; cur_gfn += _PAGE_ENTRIES) {
682 gaddr = gfn_to_gpa(cur_gfn);
683 vmaddr = gfn_to_hva_memslot(memslot, cur_gfn);
684 if (kvm_is_error_hva(vmaddr))
687 bitmap_zero(bitmap, _PAGE_ENTRIES);
688 gmap_sync_dirty_log_pmd(gmap, bitmap, gaddr, vmaddr);
689 for (i = 0; i < _PAGE_ENTRIES; i++) {
690 if (test_bit(i, bitmap))
691 mark_page_dirty(kvm, cur_gfn + i);
694 if (fatal_signal_pending(current))
700 /* Section: vm related */
701 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
704 * Get (and clear) the dirty memory log for a memory slot.
706 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
707 struct kvm_dirty_log *log)
711 struct kvm_memory_slot *memslot;
714 if (kvm_is_ucontrol(kvm))
717 mutex_lock(&kvm->slots_lock);
720 if (log->slot >= KVM_USER_MEM_SLOTS)
723 r = kvm_get_dirty_log(kvm, log, &is_dirty, &memslot);
727 /* Clear the dirty log */
729 n = kvm_dirty_bitmap_bytes(memslot);
730 memset(memslot->dirty_bitmap, 0, n);
734 mutex_unlock(&kvm->slots_lock);
738 static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
741 struct kvm_vcpu *vcpu;
743 kvm_for_each_vcpu(i, vcpu, kvm) {
744 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
748 int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
756 case KVM_CAP_S390_IRQCHIP:
757 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
758 kvm->arch.use_irqchip = 1;
761 case KVM_CAP_S390_USER_SIGP:
762 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
763 kvm->arch.user_sigp = 1;
766 case KVM_CAP_S390_VECTOR_REGISTERS:
767 mutex_lock(&kvm->lock);
768 if (kvm->created_vcpus) {
770 } else if (MACHINE_HAS_VX) {
771 set_kvm_facility(kvm->arch.model.fac_mask, 129);
772 set_kvm_facility(kvm->arch.model.fac_list, 129);
773 if (test_facility(134)) {
774 set_kvm_facility(kvm->arch.model.fac_mask, 134);
775 set_kvm_facility(kvm->arch.model.fac_list, 134);
777 if (test_facility(135)) {
778 set_kvm_facility(kvm->arch.model.fac_mask, 135);
779 set_kvm_facility(kvm->arch.model.fac_list, 135);
781 if (test_facility(148)) {
782 set_kvm_facility(kvm->arch.model.fac_mask, 148);
783 set_kvm_facility(kvm->arch.model.fac_list, 148);
785 if (test_facility(152)) {
786 set_kvm_facility(kvm->arch.model.fac_mask, 152);
787 set_kvm_facility(kvm->arch.model.fac_list, 152);
789 if (test_facility(192)) {
790 set_kvm_facility(kvm->arch.model.fac_mask, 192);
791 set_kvm_facility(kvm->arch.model.fac_list, 192);
796 mutex_unlock(&kvm->lock);
797 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
798 r ? "(not available)" : "(success)");
800 case KVM_CAP_S390_RI:
802 mutex_lock(&kvm->lock);
803 if (kvm->created_vcpus) {
805 } else if (test_facility(64)) {
806 set_kvm_facility(kvm->arch.model.fac_mask, 64);
807 set_kvm_facility(kvm->arch.model.fac_list, 64);
810 mutex_unlock(&kvm->lock);
811 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
812 r ? "(not available)" : "(success)");
814 case KVM_CAP_S390_AIS:
815 mutex_lock(&kvm->lock);
816 if (kvm->created_vcpus) {
819 set_kvm_facility(kvm->arch.model.fac_mask, 72);
820 set_kvm_facility(kvm->arch.model.fac_list, 72);
823 mutex_unlock(&kvm->lock);
824 VM_EVENT(kvm, 3, "ENABLE: AIS %s",
825 r ? "(not available)" : "(success)");
827 case KVM_CAP_S390_GS:
829 mutex_lock(&kvm->lock);
830 if (kvm->created_vcpus) {
832 } else if (test_facility(133)) {
833 set_kvm_facility(kvm->arch.model.fac_mask, 133);
834 set_kvm_facility(kvm->arch.model.fac_list, 133);
837 mutex_unlock(&kvm->lock);
838 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_GS %s",
839 r ? "(not available)" : "(success)");
841 case KVM_CAP_S390_HPAGE_1M:
842 mutex_lock(&kvm->lock);
843 if (kvm->created_vcpus)
845 else if (!hpage || kvm->arch.use_cmma || kvm_is_ucontrol(kvm))
849 mmap_write_lock(kvm->mm);
850 kvm->mm->context.allow_gmap_hpage_1m = 1;
851 mmap_write_unlock(kvm->mm);
853 * We might have to create fake 4k page
854 * tables. To avoid that the hardware works on
855 * stale PGSTEs, we emulate these instructions.
857 kvm->arch.use_skf = 0;
858 kvm->arch.use_pfmfi = 0;
860 mutex_unlock(&kvm->lock);
861 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_HPAGE %s",
862 r ? "(not available)" : "(success)");
864 case KVM_CAP_S390_USER_STSI:
865 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
866 kvm->arch.user_stsi = 1;
869 case KVM_CAP_S390_USER_INSTR0:
870 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
871 kvm->arch.user_instr0 = 1;
872 icpt_operexc_on_all_vcpus(kvm);
875 case KVM_CAP_S390_CPU_TOPOLOGY:
877 mutex_lock(&kvm->lock);
878 if (kvm->created_vcpus) {
880 } else if (test_facility(11)) {
881 set_kvm_facility(kvm->arch.model.fac_mask, 11);
882 set_kvm_facility(kvm->arch.model.fac_list, 11);
885 mutex_unlock(&kvm->lock);
886 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_CPU_TOPOLOGY %s",
887 r ? "(not available)" : "(success)");
896 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
900 switch (attr->attr) {
901 case KVM_S390_VM_MEM_LIMIT_SIZE:
903 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
904 kvm->arch.mem_limit);
905 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
915 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
919 switch (attr->attr) {
920 case KVM_S390_VM_MEM_ENABLE_CMMA:
925 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
926 mutex_lock(&kvm->lock);
927 if (kvm->created_vcpus)
929 else if (kvm->mm->context.allow_gmap_hpage_1m)
932 kvm->arch.use_cmma = 1;
933 /* Not compatible with cmma. */
934 kvm->arch.use_pfmfi = 0;
937 mutex_unlock(&kvm->lock);
939 case KVM_S390_VM_MEM_CLR_CMMA:
944 if (!kvm->arch.use_cmma)
947 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
948 mutex_lock(&kvm->lock);
949 idx = srcu_read_lock(&kvm->srcu);
950 s390_reset_cmma(kvm->arch.gmap->mm);
951 srcu_read_unlock(&kvm->srcu, idx);
952 mutex_unlock(&kvm->lock);
955 case KVM_S390_VM_MEM_LIMIT_SIZE: {
956 unsigned long new_limit;
958 if (kvm_is_ucontrol(kvm))
961 if (get_user(new_limit, (u64 __user *)attr->addr))
964 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
965 new_limit > kvm->arch.mem_limit)
971 /* gmap_create takes last usable address */
972 if (new_limit != KVM_S390_NO_MEM_LIMIT)
976 mutex_lock(&kvm->lock);
977 if (!kvm->created_vcpus) {
978 /* gmap_create will round the limit up */
979 struct gmap *new = gmap_create(current->mm, new_limit);
984 gmap_remove(kvm->arch.gmap);
986 kvm->arch.gmap = new;
990 mutex_unlock(&kvm->lock);
991 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
992 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
993 (void *) kvm->arch.gmap->asce);
1003 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
1005 void kvm_s390_vcpu_crypto_reset_all(struct kvm *kvm)
1007 struct kvm_vcpu *vcpu;
1010 kvm_s390_vcpu_block_all(kvm);
1012 kvm_for_each_vcpu(i, vcpu, kvm) {
1013 kvm_s390_vcpu_crypto_setup(vcpu);
1014 /* recreate the shadow crycb by leaving the VSIE handler */
1015 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu);
1018 kvm_s390_vcpu_unblock_all(kvm);
1021 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
1023 mutex_lock(&kvm->lock);
1024 switch (attr->attr) {
1025 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1026 if (!test_kvm_facility(kvm, 76)) {
1027 mutex_unlock(&kvm->lock);
1031 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1032 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1033 kvm->arch.crypto.aes_kw = 1;
1034 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
1036 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1037 if (!test_kvm_facility(kvm, 76)) {
1038 mutex_unlock(&kvm->lock);
1042 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1043 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1044 kvm->arch.crypto.dea_kw = 1;
1045 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
1047 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1048 if (!test_kvm_facility(kvm, 76)) {
1049 mutex_unlock(&kvm->lock);
1052 kvm->arch.crypto.aes_kw = 0;
1053 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
1054 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1055 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
1057 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1058 if (!test_kvm_facility(kvm, 76)) {
1059 mutex_unlock(&kvm->lock);
1062 kvm->arch.crypto.dea_kw = 0;
1063 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
1064 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1065 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
1067 case KVM_S390_VM_CRYPTO_ENABLE_APIE:
1068 if (!ap_instructions_available()) {
1069 mutex_unlock(&kvm->lock);
1072 kvm->arch.crypto.apie = 1;
1074 case KVM_S390_VM_CRYPTO_DISABLE_APIE:
1075 if (!ap_instructions_available()) {
1076 mutex_unlock(&kvm->lock);
1079 kvm->arch.crypto.apie = 0;
1082 mutex_unlock(&kvm->lock);
1086 kvm_s390_vcpu_crypto_reset_all(kvm);
1087 mutex_unlock(&kvm->lock);
1091 static void kvm_s390_vcpu_pci_setup(struct kvm_vcpu *vcpu)
1093 /* Only set the ECB bits after guest requests zPCI interpretation */
1094 if (!vcpu->kvm->arch.use_zpci_interp)
1097 vcpu->arch.sie_block->ecb2 |= ECB2_ZPCI_LSI;
1098 vcpu->arch.sie_block->ecb3 |= ECB3_AISII + ECB3_AISI;
1101 void kvm_s390_vcpu_pci_enable_interp(struct kvm *kvm)
1103 struct kvm_vcpu *vcpu;
1106 lockdep_assert_held(&kvm->lock);
1108 if (!kvm_s390_pci_interp_allowed())
1112 * If host is configured for PCI and the necessary facilities are
1113 * available, turn on interpretation for the life of this guest
1115 kvm->arch.use_zpci_interp = 1;
1117 kvm_s390_vcpu_block_all(kvm);
1119 kvm_for_each_vcpu(i, vcpu, kvm) {
1120 kvm_s390_vcpu_pci_setup(vcpu);
1121 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu);
1124 kvm_s390_vcpu_unblock_all(kvm);
1127 static void kvm_s390_sync_request_broadcast(struct kvm *kvm, int req)
1130 struct kvm_vcpu *vcpu;
1132 kvm_for_each_vcpu(cx, vcpu, kvm)
1133 kvm_s390_sync_request(req, vcpu);
1137 * Must be called with kvm->srcu held to avoid races on memslots, and with
1138 * kvm->slots_lock to avoid races with ourselves and kvm_s390_vm_stop_migration.
1140 static int kvm_s390_vm_start_migration(struct kvm *kvm)
1142 struct kvm_memory_slot *ms;
1143 struct kvm_memslots *slots;
1144 unsigned long ram_pages = 0;
1147 /* migration mode already enabled */
1148 if (kvm->arch.migration_mode)
1150 slots = kvm_memslots(kvm);
1151 if (!slots || kvm_memslots_empty(slots))
1154 if (!kvm->arch.use_cmma) {
1155 kvm->arch.migration_mode = 1;
1158 /* mark all the pages in active slots as dirty */
1159 kvm_for_each_memslot(ms, bkt, slots) {
1160 if (!ms->dirty_bitmap)
1163 * The second half of the bitmap is only used on x86,
1164 * and would be wasted otherwise, so we put it to good
1165 * use here to keep track of the state of the storage
1168 memset(kvm_second_dirty_bitmap(ms), 0xff, kvm_dirty_bitmap_bytes(ms));
1169 ram_pages += ms->npages;
1171 atomic64_set(&kvm->arch.cmma_dirty_pages, ram_pages);
1172 kvm->arch.migration_mode = 1;
1173 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_START_MIGRATION);
1178 * Must be called with kvm->slots_lock to avoid races with ourselves and
1179 * kvm_s390_vm_start_migration.
1181 static int kvm_s390_vm_stop_migration(struct kvm *kvm)
1183 /* migration mode already disabled */
1184 if (!kvm->arch.migration_mode)
1186 kvm->arch.migration_mode = 0;
1187 if (kvm->arch.use_cmma)
1188 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_STOP_MIGRATION);
1192 static int kvm_s390_vm_set_migration(struct kvm *kvm,
1193 struct kvm_device_attr *attr)
1197 mutex_lock(&kvm->slots_lock);
1198 switch (attr->attr) {
1199 case KVM_S390_VM_MIGRATION_START:
1200 res = kvm_s390_vm_start_migration(kvm);
1202 case KVM_S390_VM_MIGRATION_STOP:
1203 res = kvm_s390_vm_stop_migration(kvm);
1208 mutex_unlock(&kvm->slots_lock);
1213 static int kvm_s390_vm_get_migration(struct kvm *kvm,
1214 struct kvm_device_attr *attr)
1216 u64 mig = kvm->arch.migration_mode;
1218 if (attr->attr != KVM_S390_VM_MIGRATION_STATUS)
1221 if (copy_to_user((void __user *)attr->addr, &mig, sizeof(mig)))
1226 static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod);
1228 static int kvm_s390_set_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1230 struct kvm_s390_vm_tod_clock gtod;
1232 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod)))
1235 if (!test_kvm_facility(kvm, 139) && gtod.epoch_idx)
1237 __kvm_s390_set_tod_clock(kvm, >od);
1239 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x, TOD base: 0x%llx",
1240 gtod.epoch_idx, gtod.tod);
1245 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1249 if (copy_from_user(>od_high, (void __user *)attr->addr,
1255 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
1260 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1262 struct kvm_s390_vm_tod_clock gtod = { 0 };
1264 if (copy_from_user(>od.tod, (void __user *)attr->addr,
1268 __kvm_s390_set_tod_clock(kvm, >od);
1269 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod.tod);
1273 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1280 mutex_lock(&kvm->lock);
1282 * For protected guests, the TOD is managed by the ultravisor, so trying
1283 * to change it will never bring the expected results.
1285 if (kvm_s390_pv_is_protected(kvm)) {
1290 switch (attr->attr) {
1291 case KVM_S390_VM_TOD_EXT:
1292 ret = kvm_s390_set_tod_ext(kvm, attr);
1294 case KVM_S390_VM_TOD_HIGH:
1295 ret = kvm_s390_set_tod_high(kvm, attr);
1297 case KVM_S390_VM_TOD_LOW:
1298 ret = kvm_s390_set_tod_low(kvm, attr);
1306 mutex_unlock(&kvm->lock);
1310 static void kvm_s390_get_tod_clock(struct kvm *kvm,
1311 struct kvm_s390_vm_tod_clock *gtod)
1313 union tod_clock clk;
1317 store_tod_clock_ext(&clk);
1319 gtod->tod = clk.tod + kvm->arch.epoch;
1320 gtod->epoch_idx = 0;
1321 if (test_kvm_facility(kvm, 139)) {
1322 gtod->epoch_idx = clk.ei + kvm->arch.epdx;
1323 if (gtod->tod < clk.tod)
1324 gtod->epoch_idx += 1;
1330 static int kvm_s390_get_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1332 struct kvm_s390_vm_tod_clock gtod;
1334 memset(>od, 0, sizeof(gtod));
1335 kvm_s390_get_tod_clock(kvm, >od);
1336 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
1339 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x, TOD base: 0x%llx",
1340 gtod.epoch_idx, gtod.tod);
1344 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1348 if (copy_to_user((void __user *)attr->addr, >od_high,
1351 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
1356 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1360 gtod = kvm_s390_get_tod_clock_fast(kvm);
1361 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
1363 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
1368 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1375 switch (attr->attr) {
1376 case KVM_S390_VM_TOD_EXT:
1377 ret = kvm_s390_get_tod_ext(kvm, attr);
1379 case KVM_S390_VM_TOD_HIGH:
1380 ret = kvm_s390_get_tod_high(kvm, attr);
1382 case KVM_S390_VM_TOD_LOW:
1383 ret = kvm_s390_get_tod_low(kvm, attr);
1392 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1394 struct kvm_s390_vm_cpu_processor *proc;
1395 u16 lowest_ibc, unblocked_ibc;
1398 mutex_lock(&kvm->lock);
1399 if (kvm->created_vcpus) {
1403 proc = kzalloc(sizeof(*proc), GFP_KERNEL_ACCOUNT);
1408 if (!copy_from_user(proc, (void __user *)attr->addr,
1410 kvm->arch.model.cpuid = proc->cpuid;
1411 lowest_ibc = sclp.ibc >> 16 & 0xfff;
1412 unblocked_ibc = sclp.ibc & 0xfff;
1413 if (lowest_ibc && proc->ibc) {
1414 if (proc->ibc > unblocked_ibc)
1415 kvm->arch.model.ibc = unblocked_ibc;
1416 else if (proc->ibc < lowest_ibc)
1417 kvm->arch.model.ibc = lowest_ibc;
1419 kvm->arch.model.ibc = proc->ibc;
1421 memcpy(kvm->arch.model.fac_list, proc->fac_list,
1422 S390_ARCH_FAC_LIST_SIZE_BYTE);
1423 VM_EVENT(kvm, 3, "SET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1424 kvm->arch.model.ibc,
1425 kvm->arch.model.cpuid);
1426 VM_EVENT(kvm, 3, "SET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1427 kvm->arch.model.fac_list[0],
1428 kvm->arch.model.fac_list[1],
1429 kvm->arch.model.fac_list[2]);
1434 mutex_unlock(&kvm->lock);
1438 static int kvm_s390_set_processor_feat(struct kvm *kvm,
1439 struct kvm_device_attr *attr)
1441 struct kvm_s390_vm_cpu_feat data;
1443 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
1445 if (!bitmap_subset((unsigned long *) data.feat,
1446 kvm_s390_available_cpu_feat,
1447 KVM_S390_VM_CPU_FEAT_NR_BITS))
1450 mutex_lock(&kvm->lock);
1451 if (kvm->created_vcpus) {
1452 mutex_unlock(&kvm->lock);
1455 bitmap_from_arr64(kvm->arch.cpu_feat, data.feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1456 mutex_unlock(&kvm->lock);
1457 VM_EVENT(kvm, 3, "SET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1464 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
1465 struct kvm_device_attr *attr)
1467 mutex_lock(&kvm->lock);
1468 if (kvm->created_vcpus) {
1469 mutex_unlock(&kvm->lock);
1473 if (copy_from_user(&kvm->arch.model.subfuncs, (void __user *)attr->addr,
1474 sizeof(struct kvm_s390_vm_cpu_subfunc))) {
1475 mutex_unlock(&kvm->lock);
1478 mutex_unlock(&kvm->lock);
1480 VM_EVENT(kvm, 3, "SET: guest PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1481 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0],
1482 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1],
1483 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2],
1484 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]);
1485 VM_EVENT(kvm, 3, "SET: guest PTFF subfunc 0x%16.16lx.%16.16lx",
1486 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0],
1487 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]);
1488 VM_EVENT(kvm, 3, "SET: guest KMAC subfunc 0x%16.16lx.%16.16lx",
1489 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0],
1490 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]);
1491 VM_EVENT(kvm, 3, "SET: guest KMC subfunc 0x%16.16lx.%16.16lx",
1492 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0],
1493 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]);
1494 VM_EVENT(kvm, 3, "SET: guest KM subfunc 0x%16.16lx.%16.16lx",
1495 ((unsigned long *) &kvm->arch.model.subfuncs.km)[0],
1496 ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]);
1497 VM_EVENT(kvm, 3, "SET: guest KIMD subfunc 0x%16.16lx.%16.16lx",
1498 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0],
1499 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]);
1500 VM_EVENT(kvm, 3, "SET: guest KLMD subfunc 0x%16.16lx.%16.16lx",
1501 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0],
1502 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]);
1503 VM_EVENT(kvm, 3, "SET: guest PCKMO subfunc 0x%16.16lx.%16.16lx",
1504 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0],
1505 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]);
1506 VM_EVENT(kvm, 3, "SET: guest KMCTR subfunc 0x%16.16lx.%16.16lx",
1507 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0],
1508 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]);
1509 VM_EVENT(kvm, 3, "SET: guest KMF subfunc 0x%16.16lx.%16.16lx",
1510 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0],
1511 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]);
1512 VM_EVENT(kvm, 3, "SET: guest KMO subfunc 0x%16.16lx.%16.16lx",
1513 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0],
1514 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]);
1515 VM_EVENT(kvm, 3, "SET: guest PCC subfunc 0x%16.16lx.%16.16lx",
1516 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0],
1517 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]);
1518 VM_EVENT(kvm, 3, "SET: guest PPNO subfunc 0x%16.16lx.%16.16lx",
1519 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0],
1520 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]);
1521 VM_EVENT(kvm, 3, "SET: guest KMA subfunc 0x%16.16lx.%16.16lx",
1522 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0],
1523 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]);
1524 VM_EVENT(kvm, 3, "SET: guest KDSA subfunc 0x%16.16lx.%16.16lx",
1525 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0],
1526 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]);
1527 VM_EVENT(kvm, 3, "SET: guest SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1528 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0],
1529 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1],
1530 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2],
1531 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]);
1532 VM_EVENT(kvm, 3, "SET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1533 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0],
1534 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1],
1535 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2],
1536 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]);
1541 #define KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK \
1543 ((struct kvm_s390_vm_cpu_uv_feat){ \
1550 static int kvm_s390_set_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1552 struct kvm_s390_vm_cpu_uv_feat __user *ptr = (void __user *)attr->addr;
1553 unsigned long data, filter;
1555 filter = uv_info.uv_feature_indications & KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK;
1556 if (get_user(data, &ptr->feat))
1558 if (!bitmap_subset(&data, &filter, KVM_S390_VM_CPU_UV_FEAT_NR_BITS))
1561 mutex_lock(&kvm->lock);
1562 if (kvm->created_vcpus) {
1563 mutex_unlock(&kvm->lock);
1566 kvm->arch.model.uv_feat_guest.feat = data;
1567 mutex_unlock(&kvm->lock);
1569 VM_EVENT(kvm, 3, "SET: guest UV-feat: 0x%16.16lx", data);
1574 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1578 switch (attr->attr) {
1579 case KVM_S390_VM_CPU_PROCESSOR:
1580 ret = kvm_s390_set_processor(kvm, attr);
1582 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1583 ret = kvm_s390_set_processor_feat(kvm, attr);
1585 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1586 ret = kvm_s390_set_processor_subfunc(kvm, attr);
1588 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
1589 ret = kvm_s390_set_uv_feat(kvm, attr);
1595 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1597 struct kvm_s390_vm_cpu_processor *proc;
1600 proc = kzalloc(sizeof(*proc), GFP_KERNEL_ACCOUNT);
1605 proc->cpuid = kvm->arch.model.cpuid;
1606 proc->ibc = kvm->arch.model.ibc;
1607 memcpy(&proc->fac_list, kvm->arch.model.fac_list,
1608 S390_ARCH_FAC_LIST_SIZE_BYTE);
1609 VM_EVENT(kvm, 3, "GET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1610 kvm->arch.model.ibc,
1611 kvm->arch.model.cpuid);
1612 VM_EVENT(kvm, 3, "GET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1613 kvm->arch.model.fac_list[0],
1614 kvm->arch.model.fac_list[1],
1615 kvm->arch.model.fac_list[2]);
1616 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
1623 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
1625 struct kvm_s390_vm_cpu_machine *mach;
1628 mach = kzalloc(sizeof(*mach), GFP_KERNEL_ACCOUNT);
1633 get_cpu_id((struct cpuid *) &mach->cpuid);
1634 mach->ibc = sclp.ibc;
1635 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
1636 S390_ARCH_FAC_LIST_SIZE_BYTE);
1637 memcpy((unsigned long *)&mach->fac_list, stfle_fac_list,
1638 sizeof(stfle_fac_list));
1639 VM_EVENT(kvm, 3, "GET: host ibc: 0x%4.4x, host cpuid: 0x%16.16llx",
1640 kvm->arch.model.ibc,
1641 kvm->arch.model.cpuid);
1642 VM_EVENT(kvm, 3, "GET: host facmask: 0x%16.16llx.%16.16llx.%16.16llx",
1646 VM_EVENT(kvm, 3, "GET: host faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1650 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
1657 static int kvm_s390_get_processor_feat(struct kvm *kvm,
1658 struct kvm_device_attr *attr)
1660 struct kvm_s390_vm_cpu_feat data;
1662 bitmap_to_arr64(data.feat, kvm->arch.cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1663 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1665 VM_EVENT(kvm, 3, "GET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1672 static int kvm_s390_get_machine_feat(struct kvm *kvm,
1673 struct kvm_device_attr *attr)
1675 struct kvm_s390_vm_cpu_feat data;
1677 bitmap_to_arr64(data.feat, kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1678 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1680 VM_EVENT(kvm, 3, "GET: host feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1687 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
1688 struct kvm_device_attr *attr)
1690 if (copy_to_user((void __user *)attr->addr, &kvm->arch.model.subfuncs,
1691 sizeof(struct kvm_s390_vm_cpu_subfunc)))
1694 VM_EVENT(kvm, 3, "GET: guest PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1695 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0],
1696 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1],
1697 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2],
1698 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]);
1699 VM_EVENT(kvm, 3, "GET: guest PTFF subfunc 0x%16.16lx.%16.16lx",
1700 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0],
1701 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]);
1702 VM_EVENT(kvm, 3, "GET: guest KMAC subfunc 0x%16.16lx.%16.16lx",
1703 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0],
1704 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]);
1705 VM_EVENT(kvm, 3, "GET: guest KMC subfunc 0x%16.16lx.%16.16lx",
1706 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0],
1707 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]);
1708 VM_EVENT(kvm, 3, "GET: guest KM subfunc 0x%16.16lx.%16.16lx",
1709 ((unsigned long *) &kvm->arch.model.subfuncs.km)[0],
1710 ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]);
1711 VM_EVENT(kvm, 3, "GET: guest KIMD subfunc 0x%16.16lx.%16.16lx",
1712 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0],
1713 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]);
1714 VM_EVENT(kvm, 3, "GET: guest KLMD subfunc 0x%16.16lx.%16.16lx",
1715 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0],
1716 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]);
1717 VM_EVENT(kvm, 3, "GET: guest PCKMO subfunc 0x%16.16lx.%16.16lx",
1718 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0],
1719 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]);
1720 VM_EVENT(kvm, 3, "GET: guest KMCTR subfunc 0x%16.16lx.%16.16lx",
1721 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0],
1722 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]);
1723 VM_EVENT(kvm, 3, "GET: guest KMF subfunc 0x%16.16lx.%16.16lx",
1724 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0],
1725 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]);
1726 VM_EVENT(kvm, 3, "GET: guest KMO subfunc 0x%16.16lx.%16.16lx",
1727 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0],
1728 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]);
1729 VM_EVENT(kvm, 3, "GET: guest PCC subfunc 0x%16.16lx.%16.16lx",
1730 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0],
1731 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]);
1732 VM_EVENT(kvm, 3, "GET: guest PPNO subfunc 0x%16.16lx.%16.16lx",
1733 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0],
1734 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]);
1735 VM_EVENT(kvm, 3, "GET: guest KMA subfunc 0x%16.16lx.%16.16lx",
1736 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0],
1737 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]);
1738 VM_EVENT(kvm, 3, "GET: guest KDSA subfunc 0x%16.16lx.%16.16lx",
1739 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0],
1740 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]);
1741 VM_EVENT(kvm, 3, "GET: guest SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1742 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0],
1743 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1],
1744 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2],
1745 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]);
1746 VM_EVENT(kvm, 3, "GET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1747 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0],
1748 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1],
1749 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2],
1750 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]);
1755 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
1756 struct kvm_device_attr *attr)
1758 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
1759 sizeof(struct kvm_s390_vm_cpu_subfunc)))
1762 VM_EVENT(kvm, 3, "GET: host PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1763 ((unsigned long *) &kvm_s390_available_subfunc.plo)[0],
1764 ((unsigned long *) &kvm_s390_available_subfunc.plo)[1],
1765 ((unsigned long *) &kvm_s390_available_subfunc.plo)[2],
1766 ((unsigned long *) &kvm_s390_available_subfunc.plo)[3]);
1767 VM_EVENT(kvm, 3, "GET: host PTFF subfunc 0x%16.16lx.%16.16lx",
1768 ((unsigned long *) &kvm_s390_available_subfunc.ptff)[0],
1769 ((unsigned long *) &kvm_s390_available_subfunc.ptff)[1]);
1770 VM_EVENT(kvm, 3, "GET: host KMAC subfunc 0x%16.16lx.%16.16lx",
1771 ((unsigned long *) &kvm_s390_available_subfunc.kmac)[0],
1772 ((unsigned long *) &kvm_s390_available_subfunc.kmac)[1]);
1773 VM_EVENT(kvm, 3, "GET: host KMC subfunc 0x%16.16lx.%16.16lx",
1774 ((unsigned long *) &kvm_s390_available_subfunc.kmc)[0],
1775 ((unsigned long *) &kvm_s390_available_subfunc.kmc)[1]);
1776 VM_EVENT(kvm, 3, "GET: host KM subfunc 0x%16.16lx.%16.16lx",
1777 ((unsigned long *) &kvm_s390_available_subfunc.km)[0],
1778 ((unsigned long *) &kvm_s390_available_subfunc.km)[1]);
1779 VM_EVENT(kvm, 3, "GET: host KIMD subfunc 0x%16.16lx.%16.16lx",
1780 ((unsigned long *) &kvm_s390_available_subfunc.kimd)[0],
1781 ((unsigned long *) &kvm_s390_available_subfunc.kimd)[1]);
1782 VM_EVENT(kvm, 3, "GET: host KLMD subfunc 0x%16.16lx.%16.16lx",
1783 ((unsigned long *) &kvm_s390_available_subfunc.klmd)[0],
1784 ((unsigned long *) &kvm_s390_available_subfunc.klmd)[1]);
1785 VM_EVENT(kvm, 3, "GET: host PCKMO subfunc 0x%16.16lx.%16.16lx",
1786 ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[0],
1787 ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[1]);
1788 VM_EVENT(kvm, 3, "GET: host KMCTR subfunc 0x%16.16lx.%16.16lx",
1789 ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[0],
1790 ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[1]);
1791 VM_EVENT(kvm, 3, "GET: host KMF subfunc 0x%16.16lx.%16.16lx",
1792 ((unsigned long *) &kvm_s390_available_subfunc.kmf)[0],
1793 ((unsigned long *) &kvm_s390_available_subfunc.kmf)[1]);
1794 VM_EVENT(kvm, 3, "GET: host KMO subfunc 0x%16.16lx.%16.16lx",
1795 ((unsigned long *) &kvm_s390_available_subfunc.kmo)[0],
1796 ((unsigned long *) &kvm_s390_available_subfunc.kmo)[1]);
1797 VM_EVENT(kvm, 3, "GET: host PCC subfunc 0x%16.16lx.%16.16lx",
1798 ((unsigned long *) &kvm_s390_available_subfunc.pcc)[0],
1799 ((unsigned long *) &kvm_s390_available_subfunc.pcc)[1]);
1800 VM_EVENT(kvm, 3, "GET: host PPNO subfunc 0x%16.16lx.%16.16lx",
1801 ((unsigned long *) &kvm_s390_available_subfunc.ppno)[0],
1802 ((unsigned long *) &kvm_s390_available_subfunc.ppno)[1]);
1803 VM_EVENT(kvm, 3, "GET: host KMA subfunc 0x%16.16lx.%16.16lx",
1804 ((unsigned long *) &kvm_s390_available_subfunc.kma)[0],
1805 ((unsigned long *) &kvm_s390_available_subfunc.kma)[1]);
1806 VM_EVENT(kvm, 3, "GET: host KDSA subfunc 0x%16.16lx.%16.16lx",
1807 ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[0],
1808 ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[1]);
1809 VM_EVENT(kvm, 3, "GET: host SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1810 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[0],
1811 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[1],
1812 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[2],
1813 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[3]);
1814 VM_EVENT(kvm, 3, "GET: host DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1815 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[0],
1816 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[1],
1817 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[2],
1818 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[3]);
1823 static int kvm_s390_get_processor_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1825 struct kvm_s390_vm_cpu_uv_feat __user *dst = (void __user *)attr->addr;
1826 unsigned long feat = kvm->arch.model.uv_feat_guest.feat;
1828 if (put_user(feat, &dst->feat))
1830 VM_EVENT(kvm, 3, "GET: guest UV-feat: 0x%16.16lx", feat);
1835 static int kvm_s390_get_machine_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1837 struct kvm_s390_vm_cpu_uv_feat __user *dst = (void __user *)attr->addr;
1840 BUILD_BUG_ON(sizeof(*dst) != sizeof(uv_info.uv_feature_indications));
1842 feat = uv_info.uv_feature_indications & KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK;
1843 if (put_user(feat, &dst->feat))
1845 VM_EVENT(kvm, 3, "GET: guest UV-feat: 0x%16.16lx", feat);
1850 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1854 switch (attr->attr) {
1855 case KVM_S390_VM_CPU_PROCESSOR:
1856 ret = kvm_s390_get_processor(kvm, attr);
1858 case KVM_S390_VM_CPU_MACHINE:
1859 ret = kvm_s390_get_machine(kvm, attr);
1861 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1862 ret = kvm_s390_get_processor_feat(kvm, attr);
1864 case KVM_S390_VM_CPU_MACHINE_FEAT:
1865 ret = kvm_s390_get_machine_feat(kvm, attr);
1867 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1868 ret = kvm_s390_get_processor_subfunc(kvm, attr);
1870 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1871 ret = kvm_s390_get_machine_subfunc(kvm, attr);
1873 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
1874 ret = kvm_s390_get_processor_uv_feat(kvm, attr);
1876 case KVM_S390_VM_CPU_MACHINE_UV_FEAT_GUEST:
1877 ret = kvm_s390_get_machine_uv_feat(kvm, attr);
1884 * kvm_s390_update_topology_change_report - update CPU topology change report
1885 * @kvm: guest KVM description
1886 * @val: set or clear the MTCR bit
1888 * Updates the Multiprocessor Topology-Change-Report bit to signal
1889 * the guest with a topology change.
1890 * This is only relevant if the topology facility is present.
1892 * The SCA version, bsca or esca, doesn't matter as offset is the same.
1894 static void kvm_s390_update_topology_change_report(struct kvm *kvm, bool val)
1896 union sca_utility new, old;
1897 struct bsca_block *sca;
1899 read_lock(&kvm->arch.sca_lock);
1900 sca = kvm->arch.sca;
1902 old = READ_ONCE(sca->utility);
1905 } while (cmpxchg(&sca->utility.val, old.val, new.val) != old.val);
1906 read_unlock(&kvm->arch.sca_lock);
1909 static int kvm_s390_set_topo_change_indication(struct kvm *kvm,
1910 struct kvm_device_attr *attr)
1912 if (!test_kvm_facility(kvm, 11))
1915 kvm_s390_update_topology_change_report(kvm, !!attr->attr);
1919 static int kvm_s390_get_topo_change_indication(struct kvm *kvm,
1920 struct kvm_device_attr *attr)
1924 if (!test_kvm_facility(kvm, 11))
1927 read_lock(&kvm->arch.sca_lock);
1928 topo = ((struct bsca_block *)kvm->arch.sca)->utility.mtcr;
1929 read_unlock(&kvm->arch.sca_lock);
1931 return put_user(topo, (u8 __user *)attr->addr);
1934 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1938 switch (attr->group) {
1939 case KVM_S390_VM_MEM_CTRL:
1940 ret = kvm_s390_set_mem_control(kvm, attr);
1942 case KVM_S390_VM_TOD:
1943 ret = kvm_s390_set_tod(kvm, attr);
1945 case KVM_S390_VM_CPU_MODEL:
1946 ret = kvm_s390_set_cpu_model(kvm, attr);
1948 case KVM_S390_VM_CRYPTO:
1949 ret = kvm_s390_vm_set_crypto(kvm, attr);
1951 case KVM_S390_VM_MIGRATION:
1952 ret = kvm_s390_vm_set_migration(kvm, attr);
1954 case KVM_S390_VM_CPU_TOPOLOGY:
1955 ret = kvm_s390_set_topo_change_indication(kvm, attr);
1965 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1969 switch (attr->group) {
1970 case KVM_S390_VM_MEM_CTRL:
1971 ret = kvm_s390_get_mem_control(kvm, attr);
1973 case KVM_S390_VM_TOD:
1974 ret = kvm_s390_get_tod(kvm, attr);
1976 case KVM_S390_VM_CPU_MODEL:
1977 ret = kvm_s390_get_cpu_model(kvm, attr);
1979 case KVM_S390_VM_MIGRATION:
1980 ret = kvm_s390_vm_get_migration(kvm, attr);
1982 case KVM_S390_VM_CPU_TOPOLOGY:
1983 ret = kvm_s390_get_topo_change_indication(kvm, attr);
1993 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1997 switch (attr->group) {
1998 case KVM_S390_VM_MEM_CTRL:
1999 switch (attr->attr) {
2000 case KVM_S390_VM_MEM_ENABLE_CMMA:
2001 case KVM_S390_VM_MEM_CLR_CMMA:
2002 ret = sclp.has_cmma ? 0 : -ENXIO;
2004 case KVM_S390_VM_MEM_LIMIT_SIZE:
2012 case KVM_S390_VM_TOD:
2013 switch (attr->attr) {
2014 case KVM_S390_VM_TOD_LOW:
2015 case KVM_S390_VM_TOD_HIGH:
2023 case KVM_S390_VM_CPU_MODEL:
2024 switch (attr->attr) {
2025 case KVM_S390_VM_CPU_PROCESSOR:
2026 case KVM_S390_VM_CPU_MACHINE:
2027 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
2028 case KVM_S390_VM_CPU_MACHINE_FEAT:
2029 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
2030 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
2031 case KVM_S390_VM_CPU_MACHINE_UV_FEAT_GUEST:
2032 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
2040 case KVM_S390_VM_CRYPTO:
2041 switch (attr->attr) {
2042 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
2043 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
2044 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
2045 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
2048 case KVM_S390_VM_CRYPTO_ENABLE_APIE:
2049 case KVM_S390_VM_CRYPTO_DISABLE_APIE:
2050 ret = ap_instructions_available() ? 0 : -ENXIO;
2057 case KVM_S390_VM_MIGRATION:
2060 case KVM_S390_VM_CPU_TOPOLOGY:
2061 ret = test_kvm_facility(kvm, 11) ? 0 : -ENXIO;
2071 static int kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
2075 int srcu_idx, i, r = 0;
2077 if (args->flags != 0)
2080 /* Is this guest using storage keys? */
2081 if (!mm_uses_skeys(current->mm))
2082 return KVM_S390_GET_SKEYS_NONE;
2084 /* Enforce sane limit on memory allocation */
2085 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
2088 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL_ACCOUNT);
2092 mmap_read_lock(current->mm);
2093 srcu_idx = srcu_read_lock(&kvm->srcu);
2094 for (i = 0; i < args->count; i++) {
2095 hva = gfn_to_hva(kvm, args->start_gfn + i);
2096 if (kvm_is_error_hva(hva)) {
2101 r = get_guest_storage_key(current->mm, hva, &keys[i]);
2105 srcu_read_unlock(&kvm->srcu, srcu_idx);
2106 mmap_read_unlock(current->mm);
2109 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
2110 sizeof(uint8_t) * args->count);
2119 static int kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
2123 int srcu_idx, i, r = 0;
2126 if (args->flags != 0)
2129 /* Enforce sane limit on memory allocation */
2130 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
2133 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL_ACCOUNT);
2137 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
2138 sizeof(uint8_t) * args->count);
2144 /* Enable storage key handling for the guest */
2145 r = s390_enable_skey();
2150 mmap_read_lock(current->mm);
2151 srcu_idx = srcu_read_lock(&kvm->srcu);
2152 while (i < args->count) {
2154 hva = gfn_to_hva(kvm, args->start_gfn + i);
2155 if (kvm_is_error_hva(hva)) {
2160 /* Lowest order bit is reserved */
2161 if (keys[i] & 0x01) {
2166 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
2168 r = fixup_user_fault(current->mm, hva,
2169 FAULT_FLAG_WRITE, &unlocked);
2176 srcu_read_unlock(&kvm->srcu, srcu_idx);
2177 mmap_read_unlock(current->mm);
2184 * Base address and length must be sent at the start of each block, therefore
2185 * it's cheaper to send some clean data, as long as it's less than the size of
2188 #define KVM_S390_MAX_BIT_DISTANCE (2 * sizeof(void *))
2189 /* for consistency */
2190 #define KVM_S390_CMMA_SIZE_MAX ((u32)KVM_S390_SKEYS_MAX)
2192 static int kvm_s390_peek_cmma(struct kvm *kvm, struct kvm_s390_cmma_log *args,
2193 u8 *res, unsigned long bufsize)
2195 unsigned long pgstev, hva, cur_gfn = args->start_gfn;
2198 while (args->count < bufsize) {
2199 hva = gfn_to_hva(kvm, cur_gfn);
2201 * We return an error if the first value was invalid, but we
2202 * return successfully if at least one value was copied.
2204 if (kvm_is_error_hva(hva))
2205 return args->count ? 0 : -EFAULT;
2206 if (get_pgste(kvm->mm, hva, &pgstev) < 0)
2208 res[args->count++] = (pgstev >> 24) & 0x43;
2215 static struct kvm_memory_slot *gfn_to_memslot_approx(struct kvm_memslots *slots,
2218 return ____gfn_to_memslot(slots, gfn, true);
2221 static unsigned long kvm_s390_next_dirty_cmma(struct kvm_memslots *slots,
2222 unsigned long cur_gfn)
2224 struct kvm_memory_slot *ms = gfn_to_memslot_approx(slots, cur_gfn);
2225 unsigned long ofs = cur_gfn - ms->base_gfn;
2226 struct rb_node *mnode = &ms->gfn_node[slots->node_idx];
2228 if (ms->base_gfn + ms->npages <= cur_gfn) {
2229 mnode = rb_next(mnode);
2230 /* If we are above the highest slot, wrap around */
2232 mnode = rb_first(&slots->gfn_tree);
2234 ms = container_of(mnode, struct kvm_memory_slot, gfn_node[slots->node_idx]);
2238 if (cur_gfn < ms->base_gfn)
2241 ofs = find_next_bit(kvm_second_dirty_bitmap(ms), ms->npages, ofs);
2242 while (ofs >= ms->npages && (mnode = rb_next(mnode))) {
2243 ms = container_of(mnode, struct kvm_memory_slot, gfn_node[slots->node_idx]);
2244 ofs = find_first_bit(kvm_second_dirty_bitmap(ms), ms->npages);
2246 return ms->base_gfn + ofs;
2249 static int kvm_s390_get_cmma(struct kvm *kvm, struct kvm_s390_cmma_log *args,
2250 u8 *res, unsigned long bufsize)
2252 unsigned long mem_end, cur_gfn, next_gfn, hva, pgstev;
2253 struct kvm_memslots *slots = kvm_memslots(kvm);
2254 struct kvm_memory_slot *ms;
2256 if (unlikely(kvm_memslots_empty(slots)))
2259 cur_gfn = kvm_s390_next_dirty_cmma(slots, args->start_gfn);
2260 ms = gfn_to_memslot(kvm, cur_gfn);
2262 args->start_gfn = cur_gfn;
2265 next_gfn = kvm_s390_next_dirty_cmma(slots, cur_gfn + 1);
2266 mem_end = kvm_s390_get_gfn_end(slots);
2268 while (args->count < bufsize) {
2269 hva = gfn_to_hva(kvm, cur_gfn);
2270 if (kvm_is_error_hva(hva))
2272 /* Decrement only if we actually flipped the bit to 0 */
2273 if (test_and_clear_bit(cur_gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
2274 atomic64_dec(&kvm->arch.cmma_dirty_pages);
2275 if (get_pgste(kvm->mm, hva, &pgstev) < 0)
2277 /* Save the value */
2278 res[args->count++] = (pgstev >> 24) & 0x43;
2279 /* If the next bit is too far away, stop. */
2280 if (next_gfn > cur_gfn + KVM_S390_MAX_BIT_DISTANCE)
2282 /* If we reached the previous "next", find the next one */
2283 if (cur_gfn == next_gfn)
2284 next_gfn = kvm_s390_next_dirty_cmma(slots, cur_gfn + 1);
2285 /* Reached the end of memory or of the buffer, stop */
2286 if ((next_gfn >= mem_end) ||
2287 (next_gfn - args->start_gfn >= bufsize))
2290 /* Reached the end of the current memslot, take the next one. */
2291 if (cur_gfn - ms->base_gfn >= ms->npages) {
2292 ms = gfn_to_memslot(kvm, cur_gfn);
2301 * This function searches for the next page with dirty CMMA attributes, and
2302 * saves the attributes in the buffer up to either the end of the buffer or
2303 * until a block of at least KVM_S390_MAX_BIT_DISTANCE clean bits is found;
2304 * no trailing clean bytes are saved.
2305 * In case no dirty bits were found, or if CMMA was not enabled or used, the
2306 * output buffer will indicate 0 as length.
2308 static int kvm_s390_get_cmma_bits(struct kvm *kvm,
2309 struct kvm_s390_cmma_log *args)
2311 unsigned long bufsize;
2312 int srcu_idx, peek, ret;
2315 if (!kvm->arch.use_cmma)
2317 /* Invalid/unsupported flags were specified */
2318 if (args->flags & ~KVM_S390_CMMA_PEEK)
2320 /* Migration mode query, and we are not doing a migration */
2321 peek = !!(args->flags & KVM_S390_CMMA_PEEK);
2322 if (!peek && !kvm->arch.migration_mode)
2324 /* CMMA is disabled or was not used, or the buffer has length zero */
2325 bufsize = min(args->count, KVM_S390_CMMA_SIZE_MAX);
2326 if (!bufsize || !kvm->mm->context.uses_cmm) {
2327 memset(args, 0, sizeof(*args));
2330 /* We are not peeking, and there are no dirty pages */
2331 if (!peek && !atomic64_read(&kvm->arch.cmma_dirty_pages)) {
2332 memset(args, 0, sizeof(*args));
2336 values = vmalloc(bufsize);
2340 mmap_read_lock(kvm->mm);
2341 srcu_idx = srcu_read_lock(&kvm->srcu);
2343 ret = kvm_s390_peek_cmma(kvm, args, values, bufsize);
2345 ret = kvm_s390_get_cmma(kvm, args, values, bufsize);
2346 srcu_read_unlock(&kvm->srcu, srcu_idx);
2347 mmap_read_unlock(kvm->mm);
2349 if (kvm->arch.migration_mode)
2350 args->remaining = atomic64_read(&kvm->arch.cmma_dirty_pages);
2352 args->remaining = 0;
2354 if (copy_to_user((void __user *)args->values, values, args->count))
2362 * This function sets the CMMA attributes for the given pages. If the input
2363 * buffer has zero length, no action is taken, otherwise the attributes are
2364 * set and the mm->context.uses_cmm flag is set.
2366 static int kvm_s390_set_cmma_bits(struct kvm *kvm,
2367 const struct kvm_s390_cmma_log *args)
2369 unsigned long hva, mask, pgstev, i;
2371 int srcu_idx, r = 0;
2375 if (!kvm->arch.use_cmma)
2377 /* invalid/unsupported flags */
2378 if (args->flags != 0)
2380 /* Enforce sane limit on memory allocation */
2381 if (args->count > KVM_S390_CMMA_SIZE_MAX)
2384 if (args->count == 0)
2387 bits = vmalloc(array_size(sizeof(*bits), args->count));
2391 r = copy_from_user(bits, (void __user *)args->values, args->count);
2397 mmap_read_lock(kvm->mm);
2398 srcu_idx = srcu_read_lock(&kvm->srcu);
2399 for (i = 0; i < args->count; i++) {
2400 hva = gfn_to_hva(kvm, args->start_gfn + i);
2401 if (kvm_is_error_hva(hva)) {
2407 pgstev = pgstev << 24;
2408 mask &= _PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT;
2409 set_pgste_bits(kvm->mm, hva, mask, pgstev);
2411 srcu_read_unlock(&kvm->srcu, srcu_idx);
2412 mmap_read_unlock(kvm->mm);
2414 if (!kvm->mm->context.uses_cmm) {
2415 mmap_write_lock(kvm->mm);
2416 kvm->mm->context.uses_cmm = 1;
2417 mmap_write_unlock(kvm->mm);
2425 * kvm_s390_cpus_from_pv - Convert all protected vCPUs in a protected VM to
2427 * @kvm: the VM whose protected vCPUs are to be converted
2428 * @rc: return value for the RC field of the UVC (in case of error)
2429 * @rrc: return value for the RRC field of the UVC (in case of error)
2431 * Does not stop in case of error, tries to convert as many
2432 * CPUs as possible. In case of error, the RC and RRC of the last error are
2435 * Return: 0 in case of success, otherwise -EIO
2437 int kvm_s390_cpus_from_pv(struct kvm *kvm, u16 *rc, u16 *rrc)
2439 struct kvm_vcpu *vcpu;
2445 * We ignore failures and try to destroy as many CPUs as possible.
2446 * At the same time we must not free the assigned resources when
2447 * this fails, as the ultravisor has still access to that memory.
2448 * So kvm_s390_pv_destroy_cpu can leave a "wanted" memory leak
2450 * We want to return the first failure rc and rrc, though.
2452 kvm_for_each_vcpu(i, vcpu, kvm) {
2453 mutex_lock(&vcpu->mutex);
2454 if (kvm_s390_pv_destroy_cpu(vcpu, &_rc, &_rrc) && !ret) {
2459 mutex_unlock(&vcpu->mutex);
2461 /* Ensure that we re-enable gisa if the non-PV guest used it but the PV guest did not. */
2463 kvm_s390_gisa_enable(kvm);
2468 * kvm_s390_cpus_to_pv - Convert all non-protected vCPUs in a protected VM
2470 * @kvm: the VM whose protected vCPUs are to be converted
2471 * @rc: return value for the RC field of the UVC (in case of error)
2472 * @rrc: return value for the RRC field of the UVC (in case of error)
2474 * Tries to undo the conversion in case of error.
2476 * Return: 0 in case of success, otherwise -EIO
2478 static int kvm_s390_cpus_to_pv(struct kvm *kvm, u16 *rc, u16 *rrc)
2484 struct kvm_vcpu *vcpu;
2486 /* Disable the GISA if the ultravisor does not support AIV. */
2487 if (!uv_has_feature(BIT_UV_FEAT_AIV))
2488 kvm_s390_gisa_disable(kvm);
2490 kvm_for_each_vcpu(i, vcpu, kvm) {
2491 mutex_lock(&vcpu->mutex);
2492 r = kvm_s390_pv_create_cpu(vcpu, rc, rrc);
2493 mutex_unlock(&vcpu->mutex);
2498 kvm_s390_cpus_from_pv(kvm, &dummy, &dummy);
2503 * Here we provide user space with a direct interface to query UV
2504 * related data like UV maxima and available features as well as
2505 * feature specific data.
2507 * To facilitate future extension of the data structures we'll try to
2508 * write data up to the maximum requested length.
2510 static ssize_t kvm_s390_handle_pv_info(struct kvm_s390_pv_info *info)
2514 switch (info->header.id) {
2515 case KVM_PV_INFO_VM: {
2516 len_min = sizeof(info->header) + sizeof(info->vm);
2518 if (info->header.len_max < len_min)
2521 memcpy(info->vm.inst_calls_list,
2522 uv_info.inst_calls_list,
2523 sizeof(uv_info.inst_calls_list));
2525 /* It's max cpuid not max cpus, so it's off by one */
2526 info->vm.max_cpus = uv_info.max_guest_cpu_id + 1;
2527 info->vm.max_guests = uv_info.max_num_sec_conf;
2528 info->vm.max_guest_addr = uv_info.max_sec_stor_addr;
2529 info->vm.feature_indication = uv_info.uv_feature_indications;
2533 case KVM_PV_INFO_DUMP: {
2534 len_min = sizeof(info->header) + sizeof(info->dump);
2536 if (info->header.len_max < len_min)
2539 info->dump.dump_cpu_buffer_len = uv_info.guest_cpu_stor_len;
2540 info->dump.dump_config_mem_buffer_per_1m = uv_info.conf_dump_storage_state_len;
2541 info->dump.dump_config_finalize_len = uv_info.conf_dump_finalize_len;
2549 static int kvm_s390_pv_dmp(struct kvm *kvm, struct kvm_pv_cmd *cmd,
2550 struct kvm_s390_pv_dmp dmp)
2553 void __user *result_buff = (void __user *)dmp.buff_addr;
2555 switch (dmp.subcmd) {
2556 case KVM_PV_DUMP_INIT: {
2557 if (kvm->arch.pv.dumping)
2561 * Block SIE entry as concurrent dump UVCs could lead
2564 kvm_s390_vcpu_block_all(kvm);
2566 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2567 UVC_CMD_DUMP_INIT, &cmd->rc, &cmd->rrc);
2568 KVM_UV_EVENT(kvm, 3, "PROTVIRT DUMP INIT: rc %x rrc %x",
2571 kvm->arch.pv.dumping = true;
2573 kvm_s390_vcpu_unblock_all(kvm);
2578 case KVM_PV_DUMP_CONFIG_STOR_STATE: {
2579 if (!kvm->arch.pv.dumping)
2583 * gaddr is an output parameter since we might stop
2584 * early. As dmp will be copied back in our caller, we
2585 * don't need to do it ourselves.
2587 r = kvm_s390_pv_dump_stor_state(kvm, result_buff, &dmp.gaddr, dmp.buff_len,
2588 &cmd->rc, &cmd->rrc);
2591 case KVM_PV_DUMP_COMPLETE: {
2592 if (!kvm->arch.pv.dumping)
2596 if (dmp.buff_len < uv_info.conf_dump_finalize_len)
2599 r = kvm_s390_pv_dump_complete(kvm, result_buff,
2600 &cmd->rc, &cmd->rrc);
2611 static int kvm_s390_handle_pv(struct kvm *kvm, struct kvm_pv_cmd *cmd)
2613 const bool need_lock = (cmd->cmd != KVM_PV_ASYNC_CLEANUP_PERFORM);
2614 void __user *argp = (void __user *)cmd->data;
2619 mutex_lock(&kvm->lock);
2622 case KVM_PV_ENABLE: {
2624 if (kvm_s390_pv_is_protected(kvm))
2628 * FMT 4 SIE needs esca. As we never switch back to bsca from
2629 * esca, we need no cleanup in the error cases below
2631 r = sca_switch_to_extended(kvm);
2635 mmap_write_lock(current->mm);
2636 r = gmap_mark_unmergeable();
2637 mmap_write_unlock(current->mm);
2641 r = kvm_s390_pv_init_vm(kvm, &cmd->rc, &cmd->rrc);
2645 r = kvm_s390_cpus_to_pv(kvm, &cmd->rc, &cmd->rrc);
2647 kvm_s390_pv_deinit_vm(kvm, &dummy, &dummy);
2649 /* we need to block service interrupts from now on */
2650 set_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2653 case KVM_PV_ASYNC_CLEANUP_PREPARE:
2655 if (!kvm_s390_pv_is_protected(kvm) || !async_destroy)
2658 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc);
2660 * If a CPU could not be destroyed, destroy VM will also fail.
2661 * There is no point in trying to destroy it. Instead return
2662 * the rc and rrc from the first CPU that failed destroying.
2666 r = kvm_s390_pv_set_aside(kvm, &cmd->rc, &cmd->rrc);
2668 /* no need to block service interrupts any more */
2669 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2671 case KVM_PV_ASYNC_CLEANUP_PERFORM:
2675 /* kvm->lock must not be held; this is asserted inside the function. */
2676 r = kvm_s390_pv_deinit_aside_vm(kvm, &cmd->rc, &cmd->rrc);
2678 case KVM_PV_DISABLE: {
2680 if (!kvm_s390_pv_is_protected(kvm))
2683 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc);
2685 * If a CPU could not be destroyed, destroy VM will also fail.
2686 * There is no point in trying to destroy it. Instead return
2687 * the rc and rrc from the first CPU that failed destroying.
2691 r = kvm_s390_pv_deinit_cleanup_all(kvm, &cmd->rc, &cmd->rrc);
2693 /* no need to block service interrupts any more */
2694 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2697 case KVM_PV_SET_SEC_PARMS: {
2698 struct kvm_s390_pv_sec_parm parms = {};
2702 if (!kvm_s390_pv_is_protected(kvm))
2706 if (copy_from_user(&parms, argp, sizeof(parms)))
2709 /* Currently restricted to 8KB */
2711 if (parms.length > PAGE_SIZE * 2)
2715 hdr = vmalloc(parms.length);
2720 if (!copy_from_user(hdr, (void __user *)parms.origin,
2722 r = kvm_s390_pv_set_sec_parms(kvm, hdr, parms.length,
2723 &cmd->rc, &cmd->rrc);
2728 case KVM_PV_UNPACK: {
2729 struct kvm_s390_pv_unp unp = {};
2732 if (!kvm_s390_pv_is_protected(kvm) || !mm_is_protected(kvm->mm))
2736 if (copy_from_user(&unp, argp, sizeof(unp)))
2739 r = kvm_s390_pv_unpack(kvm, unp.addr, unp.size, unp.tweak,
2740 &cmd->rc, &cmd->rrc);
2743 case KVM_PV_VERIFY: {
2745 if (!kvm_s390_pv_is_protected(kvm))
2748 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2749 UVC_CMD_VERIFY_IMG, &cmd->rc, &cmd->rrc);
2750 KVM_UV_EVENT(kvm, 3, "PROTVIRT VERIFY: rc %x rrc %x", cmd->rc,
2754 case KVM_PV_PREP_RESET: {
2756 if (!kvm_s390_pv_is_protected(kvm))
2759 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2760 UVC_CMD_PREPARE_RESET, &cmd->rc, &cmd->rrc);
2761 KVM_UV_EVENT(kvm, 3, "PROTVIRT PREP RESET: rc %x rrc %x",
2765 case KVM_PV_UNSHARE_ALL: {
2767 if (!kvm_s390_pv_is_protected(kvm))
2770 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2771 UVC_CMD_SET_UNSHARE_ALL, &cmd->rc, &cmd->rrc);
2772 KVM_UV_EVENT(kvm, 3, "PROTVIRT UNSHARE: rc %x rrc %x",
2777 struct kvm_s390_pv_info info = {};
2781 * No need to check the VM protection here.
2783 * Maybe user space wants to query some of the data
2784 * when the VM is still unprotected. If we see the
2785 * need to fence a new data command we can still
2786 * return an error in the info handler.
2790 if (copy_from_user(&info, argp, sizeof(info.header)))
2794 if (info.header.len_max < sizeof(info.header))
2797 data_len = kvm_s390_handle_pv_info(&info);
2803 * If a data command struct is extended (multiple
2804 * times) this can be used to determine how much of it
2807 info.header.len_written = data_len;
2810 if (copy_to_user(argp, &info, data_len))
2817 struct kvm_s390_pv_dmp dmp;
2820 if (!kvm_s390_pv_is_protected(kvm))
2824 if (copy_from_user(&dmp, argp, sizeof(dmp)))
2827 r = kvm_s390_pv_dmp(kvm, cmd, dmp);
2831 if (copy_to_user(argp, &dmp, sizeof(dmp))) {
2842 mutex_unlock(&kvm->lock);
2847 static int mem_op_validate_common(struct kvm_s390_mem_op *mop, u64 supported_flags)
2849 if (mop->flags & ~supported_flags || !mop->size)
2851 if (mop->size > MEM_OP_MAX_SIZE)
2853 if (mop->flags & KVM_S390_MEMOP_F_SKEY_PROTECTION) {
2862 static int kvm_s390_vm_mem_op_abs(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2864 void __user *uaddr = (void __user *)mop->buf;
2865 enum gacc_mode acc_mode;
2866 void *tmpbuf = NULL;
2869 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_SKEY_PROTECTION |
2870 KVM_S390_MEMOP_F_CHECK_ONLY);
2874 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
2875 tmpbuf = vmalloc(mop->size);
2880 srcu_idx = srcu_read_lock(&kvm->srcu);
2882 if (kvm_is_error_gpa(kvm, mop->gaddr)) {
2887 acc_mode = mop->op == KVM_S390_MEMOP_ABSOLUTE_READ ? GACC_FETCH : GACC_STORE;
2888 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2889 r = check_gpa_range(kvm, mop->gaddr, mop->size, acc_mode, mop->key);
2892 if (acc_mode == GACC_FETCH) {
2893 r = access_guest_abs_with_key(kvm, mop->gaddr, tmpbuf,
2894 mop->size, GACC_FETCH, mop->key);
2897 if (copy_to_user(uaddr, tmpbuf, mop->size))
2900 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
2904 r = access_guest_abs_with_key(kvm, mop->gaddr, tmpbuf,
2905 mop->size, GACC_STORE, mop->key);
2909 srcu_read_unlock(&kvm->srcu, srcu_idx);
2915 static int kvm_s390_vm_mem_op_cmpxchg(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2917 void __user *uaddr = (void __user *)mop->buf;
2918 void __user *old_addr = (void __user *)mop->old_addr;
2921 char raw[sizeof(__uint128_t)];
2922 } old = { .quad = 0}, new = { .quad = 0 };
2923 unsigned int off_in_quad = sizeof(new) - mop->size;
2927 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_SKEY_PROTECTION);
2931 * This validates off_in_quad. Checking that size is a power
2932 * of two is not necessary, as cmpxchg_guest_abs_with_key
2933 * takes care of that
2935 if (mop->size > sizeof(new))
2937 if (copy_from_user(&new.raw[off_in_quad], uaddr, mop->size))
2939 if (copy_from_user(&old.raw[off_in_quad], old_addr, mop->size))
2942 srcu_idx = srcu_read_lock(&kvm->srcu);
2944 if (kvm_is_error_gpa(kvm, mop->gaddr)) {
2949 r = cmpxchg_guest_abs_with_key(kvm, mop->gaddr, mop->size, &old.quad,
2950 new.quad, mop->key, &success);
2951 if (!success && copy_to_user(old_addr, &old.raw[off_in_quad], mop->size))
2955 srcu_read_unlock(&kvm->srcu, srcu_idx);
2959 static int kvm_s390_vm_mem_op(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2962 * This is technically a heuristic only, if the kvm->lock is not
2963 * taken, it is not guaranteed that the vm is/remains non-protected.
2964 * This is ok from a kernel perspective, wrongdoing is detected
2965 * on the access, -EFAULT is returned and the vm may crash the
2966 * next time it accesses the memory in question.
2967 * There is no sane usecase to do switching and a memop on two
2968 * different CPUs at the same time.
2970 if (kvm_s390_pv_get_handle(kvm))
2974 case KVM_S390_MEMOP_ABSOLUTE_READ:
2975 case KVM_S390_MEMOP_ABSOLUTE_WRITE:
2976 return kvm_s390_vm_mem_op_abs(kvm, mop);
2977 case KVM_S390_MEMOP_ABSOLUTE_CMPXCHG:
2978 return kvm_s390_vm_mem_op_cmpxchg(kvm, mop);
2984 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
2986 struct kvm *kvm = filp->private_data;
2987 void __user *argp = (void __user *)arg;
2988 struct kvm_device_attr attr;
2992 case KVM_S390_INTERRUPT: {
2993 struct kvm_s390_interrupt s390int;
2996 if (copy_from_user(&s390int, argp, sizeof(s390int)))
2998 r = kvm_s390_inject_vm(kvm, &s390int);
3001 case KVM_CREATE_IRQCHIP: {
3002 struct kvm_irq_routing_entry routing;
3005 if (kvm->arch.use_irqchip) {
3006 /* Set up dummy routing. */
3007 memset(&routing, 0, sizeof(routing));
3008 r = kvm_set_irq_routing(kvm, &routing, 0, 0);
3012 case KVM_SET_DEVICE_ATTR: {
3014 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3016 r = kvm_s390_vm_set_attr(kvm, &attr);
3019 case KVM_GET_DEVICE_ATTR: {
3021 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3023 r = kvm_s390_vm_get_attr(kvm, &attr);
3026 case KVM_HAS_DEVICE_ATTR: {
3028 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3030 r = kvm_s390_vm_has_attr(kvm, &attr);
3033 case KVM_S390_GET_SKEYS: {
3034 struct kvm_s390_skeys args;
3037 if (copy_from_user(&args, argp,
3038 sizeof(struct kvm_s390_skeys)))
3040 r = kvm_s390_get_skeys(kvm, &args);
3043 case KVM_S390_SET_SKEYS: {
3044 struct kvm_s390_skeys args;
3047 if (copy_from_user(&args, argp,
3048 sizeof(struct kvm_s390_skeys)))
3050 r = kvm_s390_set_skeys(kvm, &args);
3053 case KVM_S390_GET_CMMA_BITS: {
3054 struct kvm_s390_cmma_log args;
3057 if (copy_from_user(&args, argp, sizeof(args)))
3059 mutex_lock(&kvm->slots_lock);
3060 r = kvm_s390_get_cmma_bits(kvm, &args);
3061 mutex_unlock(&kvm->slots_lock);
3063 r = copy_to_user(argp, &args, sizeof(args));
3069 case KVM_S390_SET_CMMA_BITS: {
3070 struct kvm_s390_cmma_log args;
3073 if (copy_from_user(&args, argp, sizeof(args)))
3075 mutex_lock(&kvm->slots_lock);
3076 r = kvm_s390_set_cmma_bits(kvm, &args);
3077 mutex_unlock(&kvm->slots_lock);
3080 case KVM_S390_PV_COMMAND: {
3081 struct kvm_pv_cmd args;
3083 /* protvirt means user cpu state */
3084 kvm_s390_set_user_cpu_state_ctrl(kvm);
3086 if (!is_prot_virt_host()) {
3090 if (copy_from_user(&args, argp, sizeof(args))) {
3098 /* must be called without kvm->lock */
3099 r = kvm_s390_handle_pv(kvm, &args);
3100 if (copy_to_user(argp, &args, sizeof(args))) {
3106 case KVM_S390_MEM_OP: {
3107 struct kvm_s390_mem_op mem_op;
3109 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
3110 r = kvm_s390_vm_mem_op(kvm, &mem_op);
3115 case KVM_S390_ZPCI_OP: {
3116 struct kvm_s390_zpci_op args;
3119 if (!IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
3121 if (copy_from_user(&args, argp, sizeof(args))) {
3125 r = kvm_s390_pci_zpci_op(kvm, &args);
3135 static int kvm_s390_apxa_installed(void)
3137 struct ap_config_info info;
3139 if (ap_instructions_available()) {
3140 if (ap_qci(&info) == 0)
3148 * The format of the crypto control block (CRYCB) is specified in the 3 low
3149 * order bits of the CRYCB designation (CRYCBD) field as follows:
3150 * Format 0: Neither the message security assist extension 3 (MSAX3) nor the
3151 * AP extended addressing (APXA) facility are installed.
3152 * Format 1: The APXA facility is not installed but the MSAX3 facility is.
3153 * Format 2: Both the APXA and MSAX3 facilities are installed
3155 static void kvm_s390_set_crycb_format(struct kvm *kvm)
3157 kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
3159 /* Clear the CRYCB format bits - i.e., set format 0 by default */
3160 kvm->arch.crypto.crycbd &= ~(CRYCB_FORMAT_MASK);
3162 /* Check whether MSAX3 is installed */
3163 if (!test_kvm_facility(kvm, 76))
3166 if (kvm_s390_apxa_installed())
3167 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
3169 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
3173 * kvm_arch_crypto_set_masks
3175 * @kvm: pointer to the target guest's KVM struct containing the crypto masks
3177 * @apm: the mask identifying the accessible AP adapters
3178 * @aqm: the mask identifying the accessible AP domains
3179 * @adm: the mask identifying the accessible AP control domains
3181 * Set the masks that identify the adapters, domains and control domains to
3182 * which the KVM guest is granted access.
3184 * Note: The kvm->lock mutex must be locked by the caller before invoking this
3187 void kvm_arch_crypto_set_masks(struct kvm *kvm, unsigned long *apm,
3188 unsigned long *aqm, unsigned long *adm)
3190 struct kvm_s390_crypto_cb *crycb = kvm->arch.crypto.crycb;
3192 kvm_s390_vcpu_block_all(kvm);
3194 switch (kvm->arch.crypto.crycbd & CRYCB_FORMAT_MASK) {
3195 case CRYCB_FORMAT2: /* APCB1 use 256 bits */
3196 memcpy(crycb->apcb1.apm, apm, 32);
3197 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx %016lx %016lx %016lx",
3198 apm[0], apm[1], apm[2], apm[3]);
3199 memcpy(crycb->apcb1.aqm, aqm, 32);
3200 VM_EVENT(kvm, 3, "SET CRYCB: aqm %016lx %016lx %016lx %016lx",
3201 aqm[0], aqm[1], aqm[2], aqm[3]);
3202 memcpy(crycb->apcb1.adm, adm, 32);
3203 VM_EVENT(kvm, 3, "SET CRYCB: adm %016lx %016lx %016lx %016lx",
3204 adm[0], adm[1], adm[2], adm[3]);
3207 case CRYCB_FORMAT0: /* Fall through both use APCB0 */
3208 memcpy(crycb->apcb0.apm, apm, 8);
3209 memcpy(crycb->apcb0.aqm, aqm, 2);
3210 memcpy(crycb->apcb0.adm, adm, 2);
3211 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx aqm %04x adm %04x",
3212 apm[0], *((unsigned short *)aqm),
3213 *((unsigned short *)adm));
3215 default: /* Can not happen */
3219 /* recreate the shadow crycb for each vcpu */
3220 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART);
3221 kvm_s390_vcpu_unblock_all(kvm);
3223 EXPORT_SYMBOL_GPL(kvm_arch_crypto_set_masks);
3226 * kvm_arch_crypto_clear_masks
3228 * @kvm: pointer to the target guest's KVM struct containing the crypto masks
3231 * Clear the masks that identify the adapters, domains and control domains to
3232 * which the KVM guest is granted access.
3234 * Note: The kvm->lock mutex must be locked by the caller before invoking this
3237 void kvm_arch_crypto_clear_masks(struct kvm *kvm)
3239 kvm_s390_vcpu_block_all(kvm);
3241 memset(&kvm->arch.crypto.crycb->apcb0, 0,
3242 sizeof(kvm->arch.crypto.crycb->apcb0));
3243 memset(&kvm->arch.crypto.crycb->apcb1, 0,
3244 sizeof(kvm->arch.crypto.crycb->apcb1));
3246 VM_EVENT(kvm, 3, "%s", "CLR CRYCB:");
3247 /* recreate the shadow crycb for each vcpu */
3248 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART);
3249 kvm_s390_vcpu_unblock_all(kvm);
3251 EXPORT_SYMBOL_GPL(kvm_arch_crypto_clear_masks);
3253 static u64 kvm_s390_get_initial_cpuid(void)
3258 cpuid.version = 0xff;
3259 return *((u64 *) &cpuid);
3262 static void kvm_s390_crypto_init(struct kvm *kvm)
3264 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
3265 kvm_s390_set_crycb_format(kvm);
3266 init_rwsem(&kvm->arch.crypto.pqap_hook_rwsem);
3268 if (!test_kvm_facility(kvm, 76))
3271 /* Enable AES/DEA protected key functions by default */
3272 kvm->arch.crypto.aes_kw = 1;
3273 kvm->arch.crypto.dea_kw = 1;
3274 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
3275 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
3276 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
3277 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
3280 static void sca_dispose(struct kvm *kvm)
3282 if (kvm->arch.use_esca)
3283 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
3285 free_page((unsigned long)(kvm->arch.sca));
3286 kvm->arch.sca = NULL;
3289 void kvm_arch_free_vm(struct kvm *kvm)
3291 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
3292 kvm_s390_pci_clear_list(kvm);
3294 __kvm_arch_free_vm(kvm);
3297 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
3299 gfp_t alloc_flags = GFP_KERNEL_ACCOUNT;
3301 char debug_name[16];
3302 static unsigned long sca_offset;
3305 #ifdef CONFIG_KVM_S390_UCONTROL
3306 if (type & ~KVM_VM_S390_UCONTROL)
3308 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
3315 rc = s390_enable_sie();
3321 if (!sclp.has_64bscao)
3322 alloc_flags |= GFP_DMA;
3323 rwlock_init(&kvm->arch.sca_lock);
3324 /* start with basic SCA */
3325 kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
3328 mutex_lock(&kvm_lock);
3330 if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
3332 kvm->arch.sca = (struct bsca_block *)
3333 ((char *) kvm->arch.sca + sca_offset);
3334 mutex_unlock(&kvm_lock);
3336 sprintf(debug_name, "kvm-%u", current->pid);
3338 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
3342 BUILD_BUG_ON(sizeof(struct sie_page2) != 4096);
3343 kvm->arch.sie_page2 =
3344 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3345 if (!kvm->arch.sie_page2)
3348 kvm->arch.sie_page2->kvm = kvm;
3349 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
3351 for (i = 0; i < kvm_s390_fac_size(); i++) {
3352 kvm->arch.model.fac_mask[i] = stfle_fac_list[i] &
3353 (kvm_s390_fac_base[i] |
3354 kvm_s390_fac_ext[i]);
3355 kvm->arch.model.fac_list[i] = stfle_fac_list[i] &
3356 kvm_s390_fac_base[i];
3358 kvm->arch.model.subfuncs = kvm_s390_available_subfunc;
3360 /* we are always in czam mode - even on pre z14 machines */
3361 set_kvm_facility(kvm->arch.model.fac_mask, 138);
3362 set_kvm_facility(kvm->arch.model.fac_list, 138);
3363 /* we emulate STHYI in kvm */
3364 set_kvm_facility(kvm->arch.model.fac_mask, 74);
3365 set_kvm_facility(kvm->arch.model.fac_list, 74);
3366 if (MACHINE_HAS_TLB_GUEST) {
3367 set_kvm_facility(kvm->arch.model.fac_mask, 147);
3368 set_kvm_facility(kvm->arch.model.fac_list, 147);
3371 if (css_general_characteristics.aiv && test_facility(65))
3372 set_kvm_facility(kvm->arch.model.fac_mask, 65);
3374 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
3375 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
3377 kvm->arch.model.uv_feat_guest.feat = 0;
3379 kvm_s390_crypto_init(kvm);
3381 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
3382 mutex_lock(&kvm->lock);
3383 kvm_s390_pci_init_list(kvm);
3384 kvm_s390_vcpu_pci_enable_interp(kvm);
3385 mutex_unlock(&kvm->lock);
3388 mutex_init(&kvm->arch.float_int.ais_lock);
3389 spin_lock_init(&kvm->arch.float_int.lock);
3390 for (i = 0; i < FIRQ_LIST_COUNT; i++)
3391 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
3392 init_waitqueue_head(&kvm->arch.ipte_wq);
3393 mutex_init(&kvm->arch.ipte_mutex);
3395 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
3396 VM_EVENT(kvm, 3, "vm created with type %lu", type);
3398 if (type & KVM_VM_S390_UCONTROL) {
3399 kvm->arch.gmap = NULL;
3400 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
3402 if (sclp.hamax == U64_MAX)
3403 kvm->arch.mem_limit = TASK_SIZE_MAX;
3405 kvm->arch.mem_limit = min_t(unsigned long, TASK_SIZE_MAX,
3407 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
3408 if (!kvm->arch.gmap)
3410 kvm->arch.gmap->private = kvm;
3411 kvm->arch.gmap->pfault_enabled = 0;
3414 kvm->arch.use_pfmfi = sclp.has_pfmfi;
3415 kvm->arch.use_skf = sclp.has_skey;
3416 spin_lock_init(&kvm->arch.start_stop_lock);
3417 kvm_s390_vsie_init(kvm);
3419 kvm_s390_gisa_init(kvm);
3420 INIT_LIST_HEAD(&kvm->arch.pv.need_cleanup);
3421 kvm->arch.pv.set_aside = NULL;
3422 KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
3426 free_page((unsigned long)kvm->arch.sie_page2);
3427 debug_unregister(kvm->arch.dbf);
3429 KVM_EVENT(3, "creation of vm failed: %d", rc);
3433 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3437 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
3438 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
3439 kvm_s390_clear_local_irqs(vcpu);
3440 kvm_clear_async_pf_completion_queue(vcpu);
3441 if (!kvm_is_ucontrol(vcpu->kvm))
3443 kvm_s390_update_topology_change_report(vcpu->kvm, 1);
3445 if (kvm_is_ucontrol(vcpu->kvm))
3446 gmap_remove(vcpu->arch.gmap);
3448 if (vcpu->kvm->arch.use_cmma)
3449 kvm_s390_vcpu_unsetup_cmma(vcpu);
3450 /* We can not hold the vcpu mutex here, we are already dying */
3451 if (kvm_s390_pv_cpu_get_handle(vcpu))
3452 kvm_s390_pv_destroy_cpu(vcpu, &rc, &rrc);
3453 free_page((unsigned long)(vcpu->arch.sie_block));
3456 void kvm_arch_destroy_vm(struct kvm *kvm)
3460 kvm_destroy_vcpus(kvm);
3462 kvm_s390_gisa_destroy(kvm);
3464 * We are already at the end of life and kvm->lock is not taken.
3465 * This is ok as the file descriptor is closed by now and nobody
3466 * can mess with the pv state.
3468 kvm_s390_pv_deinit_cleanup_all(kvm, &rc, &rrc);
3470 * Remove the mmu notifier only when the whole KVM VM is torn down,
3471 * and only if one was registered to begin with. If the VM is
3472 * currently not protected, but has been previously been protected,
3473 * then it's possible that the notifier is still registered.
3475 if (kvm->arch.pv.mmu_notifier.ops)
3476 mmu_notifier_unregister(&kvm->arch.pv.mmu_notifier, kvm->mm);
3478 debug_unregister(kvm->arch.dbf);
3479 free_page((unsigned long)kvm->arch.sie_page2);
3480 if (!kvm_is_ucontrol(kvm))
3481 gmap_remove(kvm->arch.gmap);
3482 kvm_s390_destroy_adapters(kvm);
3483 kvm_s390_clear_float_irqs(kvm);
3484 kvm_s390_vsie_destroy(kvm);
3485 KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
3488 /* Section: vcpu related */
3489 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
3491 vcpu->arch.gmap = gmap_create(current->mm, -1UL);
3492 if (!vcpu->arch.gmap)
3494 vcpu->arch.gmap->private = vcpu->kvm;
3499 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
3501 if (!kvm_s390_use_sca_entries())
3503 read_lock(&vcpu->kvm->arch.sca_lock);
3504 if (vcpu->kvm->arch.use_esca) {
3505 struct esca_block *sca = vcpu->kvm->arch.sca;
3507 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
3508 sca->cpu[vcpu->vcpu_id].sda = 0;
3510 struct bsca_block *sca = vcpu->kvm->arch.sca;
3512 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
3513 sca->cpu[vcpu->vcpu_id].sda = 0;
3515 read_unlock(&vcpu->kvm->arch.sca_lock);
3518 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
3520 if (!kvm_s390_use_sca_entries()) {
3521 phys_addr_t sca_phys = virt_to_phys(vcpu->kvm->arch.sca);
3523 /* we still need the basic sca for the ipte control */
3524 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3525 vcpu->arch.sie_block->scaol = sca_phys;
3528 read_lock(&vcpu->kvm->arch.sca_lock);
3529 if (vcpu->kvm->arch.use_esca) {
3530 struct esca_block *sca = vcpu->kvm->arch.sca;
3531 phys_addr_t sca_phys = virt_to_phys(sca);
3533 sca->cpu[vcpu->vcpu_id].sda = virt_to_phys(vcpu->arch.sie_block);
3534 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3535 vcpu->arch.sie_block->scaol = sca_phys & ESCA_SCAOL_MASK;
3536 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
3537 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
3539 struct bsca_block *sca = vcpu->kvm->arch.sca;
3540 phys_addr_t sca_phys = virt_to_phys(sca);
3542 sca->cpu[vcpu->vcpu_id].sda = virt_to_phys(vcpu->arch.sie_block);
3543 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3544 vcpu->arch.sie_block->scaol = sca_phys;
3545 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
3547 read_unlock(&vcpu->kvm->arch.sca_lock);
3550 /* Basic SCA to Extended SCA data copy routines */
3551 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
3554 d->sigp_ctrl.c = s->sigp_ctrl.c;
3555 d->sigp_ctrl.scn = s->sigp_ctrl.scn;
3558 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
3562 d->ipte_control = s->ipte_control;
3564 for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
3565 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
3568 static int sca_switch_to_extended(struct kvm *kvm)
3570 struct bsca_block *old_sca = kvm->arch.sca;
3571 struct esca_block *new_sca;
3572 struct kvm_vcpu *vcpu;
3573 unsigned long vcpu_idx;
3575 phys_addr_t new_sca_phys;
3577 if (kvm->arch.use_esca)
3580 new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3584 new_sca_phys = virt_to_phys(new_sca);
3585 scaoh = new_sca_phys >> 32;
3586 scaol = new_sca_phys & ESCA_SCAOL_MASK;
3588 kvm_s390_vcpu_block_all(kvm);
3589 write_lock(&kvm->arch.sca_lock);
3591 sca_copy_b_to_e(new_sca, old_sca);
3593 kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
3594 vcpu->arch.sie_block->scaoh = scaoh;
3595 vcpu->arch.sie_block->scaol = scaol;
3596 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
3598 kvm->arch.sca = new_sca;
3599 kvm->arch.use_esca = 1;
3601 write_unlock(&kvm->arch.sca_lock);
3602 kvm_s390_vcpu_unblock_all(kvm);
3604 free_page((unsigned long)old_sca);
3606 VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
3607 old_sca, kvm->arch.sca);
3611 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
3615 if (!kvm_s390_use_sca_entries()) {
3616 if (id < KVM_MAX_VCPUS)
3620 if (id < KVM_S390_BSCA_CPU_SLOTS)
3622 if (!sclp.has_esca || !sclp.has_64bscao)
3625 rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
3627 return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
3630 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3631 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3633 WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
3634 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3635 vcpu->arch.cputm_start = get_tod_clock_fast();
3636 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3639 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3640 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3642 WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
3643 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3644 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
3645 vcpu->arch.cputm_start = 0;
3646 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3649 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3650 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3652 WARN_ON_ONCE(vcpu->arch.cputm_enabled);
3653 vcpu->arch.cputm_enabled = true;
3654 __start_cpu_timer_accounting(vcpu);
3657 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3658 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3660 WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
3661 __stop_cpu_timer_accounting(vcpu);
3662 vcpu->arch.cputm_enabled = false;
3665 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3667 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3668 __enable_cpu_timer_accounting(vcpu);
3672 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3674 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3675 __disable_cpu_timer_accounting(vcpu);
3679 /* set the cpu timer - may only be called from the VCPU thread itself */
3680 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
3682 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3683 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3684 if (vcpu->arch.cputm_enabled)
3685 vcpu->arch.cputm_start = get_tod_clock_fast();
3686 vcpu->arch.sie_block->cputm = cputm;
3687 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3691 /* update and get the cpu timer - can also be called from other VCPU threads */
3692 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
3697 if (unlikely(!vcpu->arch.cputm_enabled))
3698 return vcpu->arch.sie_block->cputm;
3700 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3702 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
3704 * If the writer would ever execute a read in the critical
3705 * section, e.g. in irq context, we have a deadlock.
3707 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
3708 value = vcpu->arch.sie_block->cputm;
3709 /* if cputm_start is 0, accounting is being started/stopped */
3710 if (likely(vcpu->arch.cputm_start))
3711 value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
3712 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
3717 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3720 gmap_enable(vcpu->arch.enabled_gmap);
3721 kvm_s390_set_cpuflags(vcpu, CPUSTAT_RUNNING);
3722 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
3723 __start_cpu_timer_accounting(vcpu);
3727 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3730 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
3731 __stop_cpu_timer_accounting(vcpu);
3732 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_RUNNING);
3733 vcpu->arch.enabled_gmap = gmap_get_enabled();
3734 gmap_disable(vcpu->arch.enabled_gmap);
3738 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
3740 mutex_lock(&vcpu->kvm->lock);
3742 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
3743 vcpu->arch.sie_block->epdx = vcpu->kvm->arch.epdx;
3745 mutex_unlock(&vcpu->kvm->lock);
3746 if (!kvm_is_ucontrol(vcpu->kvm)) {
3747 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
3750 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
3751 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
3752 /* make vcpu_load load the right gmap on the first trigger */
3753 vcpu->arch.enabled_gmap = vcpu->arch.gmap;
3756 static bool kvm_has_pckmo_subfunc(struct kvm *kvm, unsigned long nr)
3758 if (test_bit_inv(nr, (unsigned long *)&kvm->arch.model.subfuncs.pckmo) &&
3759 test_bit_inv(nr, (unsigned long *)&kvm_s390_available_subfunc.pckmo))
3764 static bool kvm_has_pckmo_ecc(struct kvm *kvm)
3766 /* At least one ECC subfunction must be present */
3767 return kvm_has_pckmo_subfunc(kvm, 32) ||
3768 kvm_has_pckmo_subfunc(kvm, 33) ||
3769 kvm_has_pckmo_subfunc(kvm, 34) ||
3770 kvm_has_pckmo_subfunc(kvm, 40) ||
3771 kvm_has_pckmo_subfunc(kvm, 41);
3775 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
3778 * If the AP instructions are not being interpreted and the MSAX3
3779 * facility is not configured for the guest, there is nothing to set up.
3781 if (!vcpu->kvm->arch.crypto.apie && !test_kvm_facility(vcpu->kvm, 76))
3784 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
3785 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
3786 vcpu->arch.sie_block->eca &= ~ECA_APIE;
3787 vcpu->arch.sie_block->ecd &= ~ECD_ECC;
3789 if (vcpu->kvm->arch.crypto.apie)
3790 vcpu->arch.sie_block->eca |= ECA_APIE;
3792 /* Set up protected key support */
3793 if (vcpu->kvm->arch.crypto.aes_kw) {
3794 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
3795 /* ecc is also wrapped with AES key */
3796 if (kvm_has_pckmo_ecc(vcpu->kvm))
3797 vcpu->arch.sie_block->ecd |= ECD_ECC;
3800 if (vcpu->kvm->arch.crypto.dea_kw)
3801 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
3804 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
3806 free_page((unsigned long)phys_to_virt(vcpu->arch.sie_block->cbrlo));
3807 vcpu->arch.sie_block->cbrlo = 0;
3810 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
3812 void *cbrlo_page = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
3817 vcpu->arch.sie_block->cbrlo = virt_to_phys(cbrlo_page);
3821 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
3823 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
3825 vcpu->arch.sie_block->ibc = model->ibc;
3826 if (test_kvm_facility(vcpu->kvm, 7))
3827 vcpu->arch.sie_block->fac = virt_to_phys(model->fac_list);
3830 static int kvm_s390_vcpu_setup(struct kvm_vcpu *vcpu)
3835 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
3839 if (test_kvm_facility(vcpu->kvm, 78))
3840 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED2);
3841 else if (test_kvm_facility(vcpu->kvm, 8))
3842 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED);
3844 kvm_s390_vcpu_setup_model(vcpu);
3846 /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
3847 if (MACHINE_HAS_ESOP)
3848 vcpu->arch.sie_block->ecb |= ECB_HOSTPROTINT;
3849 if (test_kvm_facility(vcpu->kvm, 9))
3850 vcpu->arch.sie_block->ecb |= ECB_SRSI;
3851 if (test_kvm_facility(vcpu->kvm, 11))
3852 vcpu->arch.sie_block->ecb |= ECB_PTF;
3853 if (test_kvm_facility(vcpu->kvm, 73))
3854 vcpu->arch.sie_block->ecb |= ECB_TE;
3855 if (!kvm_is_ucontrol(vcpu->kvm))
3856 vcpu->arch.sie_block->ecb |= ECB_SPECI;
3858 if (test_kvm_facility(vcpu->kvm, 8) && vcpu->kvm->arch.use_pfmfi)
3859 vcpu->arch.sie_block->ecb2 |= ECB2_PFMFI;
3860 if (test_kvm_facility(vcpu->kvm, 130))
3861 vcpu->arch.sie_block->ecb2 |= ECB2_IEP;
3862 vcpu->arch.sie_block->eca = ECA_MVPGI | ECA_PROTEXCI;
3864 vcpu->arch.sie_block->eca |= ECA_CEI;
3866 vcpu->arch.sie_block->eca |= ECA_IB;
3868 vcpu->arch.sie_block->eca |= ECA_SII;
3869 if (sclp.has_sigpif)
3870 vcpu->arch.sie_block->eca |= ECA_SIGPI;
3871 if (test_kvm_facility(vcpu->kvm, 129)) {
3872 vcpu->arch.sie_block->eca |= ECA_VX;
3873 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
3875 if (test_kvm_facility(vcpu->kvm, 139))
3876 vcpu->arch.sie_block->ecd |= ECD_MEF;
3877 if (test_kvm_facility(vcpu->kvm, 156))
3878 vcpu->arch.sie_block->ecd |= ECD_ETOKENF;
3879 if (vcpu->arch.sie_block->gd) {
3880 vcpu->arch.sie_block->eca |= ECA_AIV;
3881 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
3882 vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
3884 vcpu->arch.sie_block->sdnxo = virt_to_phys(&vcpu->run->s.regs.sdnx) | SDNXC;
3885 vcpu->arch.sie_block->riccbd = virt_to_phys(&vcpu->run->s.regs.riccb);
3888 kvm_s390_set_cpuflags(vcpu, CPUSTAT_KSS);
3890 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
3892 if (vcpu->kvm->arch.use_cmma) {
3893 rc = kvm_s390_vcpu_setup_cmma(vcpu);
3897 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3898 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
3900 vcpu->arch.sie_block->hpid = HPID_KVM;
3902 kvm_s390_vcpu_crypto_setup(vcpu);
3904 kvm_s390_vcpu_pci_setup(vcpu);
3906 mutex_lock(&vcpu->kvm->lock);
3907 if (kvm_s390_pv_is_protected(vcpu->kvm)) {
3908 rc = kvm_s390_pv_create_cpu(vcpu, &uvrc, &uvrrc);
3910 kvm_s390_vcpu_unsetup_cmma(vcpu);
3912 mutex_unlock(&vcpu->kvm->lock);
3917 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
3919 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
3924 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
3926 struct sie_page *sie_page;
3929 BUILD_BUG_ON(sizeof(struct sie_page) != 4096);
3930 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL_ACCOUNT);
3934 vcpu->arch.sie_block = &sie_page->sie_block;
3935 vcpu->arch.sie_block->itdba = virt_to_phys(&sie_page->itdb);
3937 /* the real guest size will always be smaller than msl */
3938 vcpu->arch.sie_block->mso = 0;
3939 vcpu->arch.sie_block->msl = sclp.hamax;
3941 vcpu->arch.sie_block->icpua = vcpu->vcpu_id;
3942 spin_lock_init(&vcpu->arch.local_int.lock);
3943 vcpu->arch.sie_block->gd = kvm_s390_get_gisa_desc(vcpu->kvm);
3944 seqcount_init(&vcpu->arch.cputm_seqcount);
3946 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
3947 kvm_clear_async_pf_completion_queue(vcpu);
3948 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
3955 kvm_s390_set_prefix(vcpu, 0);
3956 if (test_kvm_facility(vcpu->kvm, 64))
3957 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
3958 if (test_kvm_facility(vcpu->kvm, 82))
3959 vcpu->run->kvm_valid_regs |= KVM_SYNC_BPBC;
3960 if (test_kvm_facility(vcpu->kvm, 133))
3961 vcpu->run->kvm_valid_regs |= KVM_SYNC_GSCB;
3962 if (test_kvm_facility(vcpu->kvm, 156))
3963 vcpu->run->kvm_valid_regs |= KVM_SYNC_ETOKEN;
3964 /* fprs can be synchronized via vrs, even if the guest has no vx. With
3965 * MACHINE_HAS_VX, (load|store)_fpu_regs() will work with vrs format.
3968 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
3970 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
3972 if (kvm_is_ucontrol(vcpu->kvm)) {
3973 rc = __kvm_ucontrol_vcpu_init(vcpu);
3975 goto out_free_sie_block;
3978 VM_EVENT(vcpu->kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK",
3979 vcpu->vcpu_id, vcpu, vcpu->arch.sie_block);
3980 trace_kvm_s390_create_vcpu(vcpu->vcpu_id, vcpu, vcpu->arch.sie_block);
3982 rc = kvm_s390_vcpu_setup(vcpu);
3984 goto out_ucontrol_uninit;
3986 kvm_s390_update_topology_change_report(vcpu->kvm, 1);
3989 out_ucontrol_uninit:
3990 if (kvm_is_ucontrol(vcpu->kvm))
3991 gmap_remove(vcpu->arch.gmap);
3993 free_page((unsigned long)(vcpu->arch.sie_block));
3997 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
3999 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask);
4000 return kvm_s390_vcpu_has_irq(vcpu, 0);
4003 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
4005 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE);
4008 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
4010 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
4014 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
4016 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
4019 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
4021 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
4025 bool kvm_s390_vcpu_sie_inhibited(struct kvm_vcpu *vcpu)
4027 return atomic_read(&vcpu->arch.sie_block->prog20) &
4028 (PROG_BLOCK_SIE | PROG_REQUEST);
4031 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
4033 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
4037 * Kick a guest cpu out of (v)SIE and wait until (v)SIE is not running.
4038 * If the CPU is not running (e.g. waiting as idle) the function will
4039 * return immediately. */
4040 void exit_sie(struct kvm_vcpu *vcpu)
4042 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
4043 kvm_s390_vsie_kick(vcpu);
4044 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
4048 /* Kick a guest cpu out of SIE to process a request synchronously */
4049 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
4051 __kvm_make_request(req, vcpu);
4052 kvm_s390_vcpu_request(vcpu);
4055 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
4058 struct kvm *kvm = gmap->private;
4059 struct kvm_vcpu *vcpu;
4060 unsigned long prefix;
4063 trace_kvm_s390_gmap_notifier(start, end, gmap_is_shadow(gmap));
4065 if (gmap_is_shadow(gmap))
4067 if (start >= 1UL << 31)
4068 /* We are only interested in prefix pages */
4070 kvm_for_each_vcpu(i, vcpu, kvm) {
4071 /* match against both prefix pages */
4072 prefix = kvm_s390_get_prefix(vcpu);
4073 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
4074 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
4076 kvm_s390_sync_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu);
4081 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
4083 /* do not poll with more than halt_poll_max_steal percent of steal time */
4084 if (S390_lowcore.avg_steal_timer * 100 / (TICK_USEC << 12) >=
4085 READ_ONCE(halt_poll_max_steal)) {
4086 vcpu->stat.halt_no_poll_steal++;
4092 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
4094 /* kvm common code refers to this, but never calls it */
4099 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
4100 struct kvm_one_reg *reg)
4105 case KVM_REG_S390_TODPR:
4106 r = put_user(vcpu->arch.sie_block->todpr,
4107 (u32 __user *)reg->addr);
4109 case KVM_REG_S390_EPOCHDIFF:
4110 r = put_user(vcpu->arch.sie_block->epoch,
4111 (u64 __user *)reg->addr);
4113 case KVM_REG_S390_CPU_TIMER:
4114 r = put_user(kvm_s390_get_cpu_timer(vcpu),
4115 (u64 __user *)reg->addr);
4117 case KVM_REG_S390_CLOCK_COMP:
4118 r = put_user(vcpu->arch.sie_block->ckc,
4119 (u64 __user *)reg->addr);
4121 case KVM_REG_S390_PFTOKEN:
4122 r = put_user(vcpu->arch.pfault_token,
4123 (u64 __user *)reg->addr);
4125 case KVM_REG_S390_PFCOMPARE:
4126 r = put_user(vcpu->arch.pfault_compare,
4127 (u64 __user *)reg->addr);
4129 case KVM_REG_S390_PFSELECT:
4130 r = put_user(vcpu->arch.pfault_select,
4131 (u64 __user *)reg->addr);
4133 case KVM_REG_S390_PP:
4134 r = put_user(vcpu->arch.sie_block->pp,
4135 (u64 __user *)reg->addr);
4137 case KVM_REG_S390_GBEA:
4138 r = put_user(vcpu->arch.sie_block->gbea,
4139 (u64 __user *)reg->addr);
4148 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
4149 struct kvm_one_reg *reg)
4155 case KVM_REG_S390_TODPR:
4156 r = get_user(vcpu->arch.sie_block->todpr,
4157 (u32 __user *)reg->addr);
4159 case KVM_REG_S390_EPOCHDIFF:
4160 r = get_user(vcpu->arch.sie_block->epoch,
4161 (u64 __user *)reg->addr);
4163 case KVM_REG_S390_CPU_TIMER:
4164 r = get_user(val, (u64 __user *)reg->addr);
4166 kvm_s390_set_cpu_timer(vcpu, val);
4168 case KVM_REG_S390_CLOCK_COMP:
4169 r = get_user(vcpu->arch.sie_block->ckc,
4170 (u64 __user *)reg->addr);
4172 case KVM_REG_S390_PFTOKEN:
4173 r = get_user(vcpu->arch.pfault_token,
4174 (u64 __user *)reg->addr);
4175 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4176 kvm_clear_async_pf_completion_queue(vcpu);
4178 case KVM_REG_S390_PFCOMPARE:
4179 r = get_user(vcpu->arch.pfault_compare,
4180 (u64 __user *)reg->addr);
4182 case KVM_REG_S390_PFSELECT:
4183 r = get_user(vcpu->arch.pfault_select,
4184 (u64 __user *)reg->addr);
4186 case KVM_REG_S390_PP:
4187 r = get_user(vcpu->arch.sie_block->pp,
4188 (u64 __user *)reg->addr);
4190 case KVM_REG_S390_GBEA:
4191 r = get_user(vcpu->arch.sie_block->gbea,
4192 (u64 __user *)reg->addr);
4201 static void kvm_arch_vcpu_ioctl_normal_reset(struct kvm_vcpu *vcpu)
4203 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_RI;
4204 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
4205 memset(vcpu->run->s.regs.riccb, 0, sizeof(vcpu->run->s.regs.riccb));
4207 kvm_clear_async_pf_completion_queue(vcpu);
4208 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
4209 kvm_s390_vcpu_stop(vcpu);
4210 kvm_s390_clear_local_irqs(vcpu);
4213 static void kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
4215 /* Initial reset is a superset of the normal reset */
4216 kvm_arch_vcpu_ioctl_normal_reset(vcpu);
4219 * This equals initial cpu reset in pop, but we don't switch to ESA.
4220 * We do not only reset the internal data, but also ...
4222 vcpu->arch.sie_block->gpsw.mask = 0;
4223 vcpu->arch.sie_block->gpsw.addr = 0;
4224 kvm_s390_set_prefix(vcpu, 0);
4225 kvm_s390_set_cpu_timer(vcpu, 0);
4226 vcpu->arch.sie_block->ckc = 0;
4227 memset(vcpu->arch.sie_block->gcr, 0, sizeof(vcpu->arch.sie_block->gcr));
4228 vcpu->arch.sie_block->gcr[0] = CR0_INITIAL_MASK;
4229 vcpu->arch.sie_block->gcr[14] = CR14_INITIAL_MASK;
4231 /* ... the data in sync regs */
4232 memset(vcpu->run->s.regs.crs, 0, sizeof(vcpu->run->s.regs.crs));
4233 vcpu->run->s.regs.ckc = 0;
4234 vcpu->run->s.regs.crs[0] = CR0_INITIAL_MASK;
4235 vcpu->run->s.regs.crs[14] = CR14_INITIAL_MASK;
4236 vcpu->run->psw_addr = 0;
4237 vcpu->run->psw_mask = 0;
4238 vcpu->run->s.regs.todpr = 0;
4239 vcpu->run->s.regs.cputm = 0;
4240 vcpu->run->s.regs.ckc = 0;
4241 vcpu->run->s.regs.pp = 0;
4242 vcpu->run->s.regs.gbea = 1;
4243 vcpu->run->s.regs.fpc = 0;
4245 * Do not reset these registers in the protected case, as some of
4246 * them are overlaid and they are not accessible in this case
4249 if (!kvm_s390_pv_cpu_is_protected(vcpu)) {
4250 vcpu->arch.sie_block->gbea = 1;
4251 vcpu->arch.sie_block->pp = 0;
4252 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
4253 vcpu->arch.sie_block->todpr = 0;
4257 static void kvm_arch_vcpu_ioctl_clear_reset(struct kvm_vcpu *vcpu)
4259 struct kvm_sync_regs *regs = &vcpu->run->s.regs;
4261 /* Clear reset is a superset of the initial reset */
4262 kvm_arch_vcpu_ioctl_initial_reset(vcpu);
4264 memset(®s->gprs, 0, sizeof(regs->gprs));
4265 memset(®s->vrs, 0, sizeof(regs->vrs));
4266 memset(®s->acrs, 0, sizeof(regs->acrs));
4267 memset(®s->gscb, 0, sizeof(regs->gscb));
4270 regs->etoken_extension = 0;
4273 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4276 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs));
4281 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4284 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
4289 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4290 struct kvm_sregs *sregs)
4294 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
4295 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
4301 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4302 struct kvm_sregs *sregs)
4306 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
4307 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
4313 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4319 if (test_fp_ctl(fpu->fpc)) {
4323 vcpu->run->s.regs.fpc = fpu->fpc;
4325 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs,
4326 (freg_t *) fpu->fprs);
4328 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs));
4335 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4339 /* make sure we have the latest values */
4342 convert_vx_to_fp((freg_t *) fpu->fprs,
4343 (__vector128 *) vcpu->run->s.regs.vrs);
4345 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs));
4346 fpu->fpc = vcpu->run->s.regs.fpc;
4352 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
4356 if (!is_vcpu_stopped(vcpu))
4359 vcpu->run->psw_mask = psw.mask;
4360 vcpu->run->psw_addr = psw.addr;
4365 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4366 struct kvm_translation *tr)
4368 return -EINVAL; /* not implemented yet */
4371 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
4372 KVM_GUESTDBG_USE_HW_BP | \
4373 KVM_GUESTDBG_ENABLE)
4375 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4376 struct kvm_guest_debug *dbg)
4382 vcpu->guest_debug = 0;
4383 kvm_s390_clear_bp_data(vcpu);
4385 if (dbg->control & ~VALID_GUESTDBG_FLAGS) {
4389 if (!sclp.has_gpere) {
4394 if (dbg->control & KVM_GUESTDBG_ENABLE) {
4395 vcpu->guest_debug = dbg->control;
4396 /* enforce guest PER */
4397 kvm_s390_set_cpuflags(vcpu, CPUSTAT_P);
4399 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
4400 rc = kvm_s390_import_bp_data(vcpu, dbg);
4402 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
4403 vcpu->arch.guestdbg.last_bp = 0;
4407 vcpu->guest_debug = 0;
4408 kvm_s390_clear_bp_data(vcpu);
4409 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
4417 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4418 struct kvm_mp_state *mp_state)
4424 /* CHECK_STOP and LOAD are not supported yet */
4425 ret = is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
4426 KVM_MP_STATE_OPERATING;
4432 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4433 struct kvm_mp_state *mp_state)
4439 /* user space knows about this interface - let it control the state */
4440 kvm_s390_set_user_cpu_state_ctrl(vcpu->kvm);
4442 switch (mp_state->mp_state) {
4443 case KVM_MP_STATE_STOPPED:
4444 rc = kvm_s390_vcpu_stop(vcpu);
4446 case KVM_MP_STATE_OPERATING:
4447 rc = kvm_s390_vcpu_start(vcpu);
4449 case KVM_MP_STATE_LOAD:
4450 if (!kvm_s390_pv_cpu_is_protected(vcpu)) {
4454 rc = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR_LOAD);
4456 case KVM_MP_STATE_CHECK_STOP:
4457 fallthrough; /* CHECK_STOP and LOAD are not supported yet */
4466 static bool ibs_enabled(struct kvm_vcpu *vcpu)
4468 return kvm_s390_test_cpuflags(vcpu, CPUSTAT_IBS);
4471 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
4474 kvm_s390_vcpu_request_handled(vcpu);
4475 if (!kvm_request_pending(vcpu))
4478 * If the guest prefix changed, re-arm the ipte notifier for the
4479 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
4480 * This ensures that the ipte instruction for this request has
4481 * already finished. We might race against a second unmapper that
4482 * wants to set the blocking bit. Lets just retry the request loop.
4484 if (kvm_check_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu)) {
4486 rc = gmap_mprotect_notify(vcpu->arch.gmap,
4487 kvm_s390_get_prefix(vcpu),
4488 PAGE_SIZE * 2, PROT_WRITE);
4490 kvm_make_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu);
4496 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
4497 vcpu->arch.sie_block->ihcpu = 0xffff;
4501 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
4502 if (!ibs_enabled(vcpu)) {
4503 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
4504 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IBS);
4509 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
4510 if (ibs_enabled(vcpu)) {
4511 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
4512 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IBS);
4517 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
4518 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
4522 if (kvm_check_request(KVM_REQ_START_MIGRATION, vcpu)) {
4524 * Disable CMM virtualization; we will emulate the ESSA
4525 * instruction manually, in order to provide additional
4526 * functionalities needed for live migration.
4528 vcpu->arch.sie_block->ecb2 &= ~ECB2_CMMA;
4532 if (kvm_check_request(KVM_REQ_STOP_MIGRATION, vcpu)) {
4534 * Re-enable CMM virtualization if CMMA is available and
4535 * CMM has been used.
4537 if ((vcpu->kvm->arch.use_cmma) &&
4538 (vcpu->kvm->mm->context.uses_cmm))
4539 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
4543 /* we left the vsie handler, nothing to do, just clear the request */
4544 kvm_clear_request(KVM_REQ_VSIE_RESTART, vcpu);
4549 static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod)
4551 struct kvm_vcpu *vcpu;
4552 union tod_clock clk;
4557 store_tod_clock_ext(&clk);
4559 kvm->arch.epoch = gtod->tod - clk.tod;
4561 if (test_kvm_facility(kvm, 139)) {
4562 kvm->arch.epdx = gtod->epoch_idx - clk.ei;
4563 if (kvm->arch.epoch > gtod->tod)
4564 kvm->arch.epdx -= 1;
4567 kvm_s390_vcpu_block_all(kvm);
4568 kvm_for_each_vcpu(i, vcpu, kvm) {
4569 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
4570 vcpu->arch.sie_block->epdx = kvm->arch.epdx;
4573 kvm_s390_vcpu_unblock_all(kvm);
4577 int kvm_s390_try_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod)
4579 if (!mutex_trylock(&kvm->lock))
4581 __kvm_s390_set_tod_clock(kvm, gtod);
4582 mutex_unlock(&kvm->lock);
4587 * kvm_arch_fault_in_page - fault-in guest page if necessary
4588 * @vcpu: The corresponding virtual cpu
4589 * @gpa: Guest physical address
4590 * @writable: Whether the page should be writable or not
4592 * Make sure that a guest page has been faulted-in on the host.
4594 * Return: Zero on success, negative error code otherwise.
4596 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
4598 return gmap_fault(vcpu->arch.gmap, gpa,
4599 writable ? FAULT_FLAG_WRITE : 0);
4602 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
4603 unsigned long token)
4605 struct kvm_s390_interrupt inti;
4606 struct kvm_s390_irq irq;
4609 irq.u.ext.ext_params2 = token;
4610 irq.type = KVM_S390_INT_PFAULT_INIT;
4611 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
4613 inti.type = KVM_S390_INT_PFAULT_DONE;
4614 inti.parm64 = token;
4615 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
4619 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
4620 struct kvm_async_pf *work)
4622 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
4623 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
4628 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
4629 struct kvm_async_pf *work)
4631 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
4632 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
4635 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
4636 struct kvm_async_pf *work)
4638 /* s390 will always inject the page directly */
4641 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
4644 * s390 will always inject the page directly,
4645 * but we still want check_async_completion to cleanup
4650 static bool kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
4653 struct kvm_arch_async_pf arch;
4655 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4657 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
4658 vcpu->arch.pfault_compare)
4660 if (psw_extint_disabled(vcpu))
4662 if (kvm_s390_vcpu_has_irq(vcpu, 0))
4664 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK))
4666 if (!vcpu->arch.gmap->pfault_enabled)
4669 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
4670 hva += current->thread.gmap_addr & ~PAGE_MASK;
4671 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
4674 return kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
4677 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
4682 * On s390 notifications for arriving pages will be delivered directly
4683 * to the guest but the house keeping for completed pfaults is
4684 * handled outside the worker.
4686 kvm_check_async_pf_completion(vcpu);
4688 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
4689 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
4694 if (!kvm_is_ucontrol(vcpu->kvm)) {
4695 rc = kvm_s390_deliver_pending_interrupts(vcpu);
4696 if (rc || guestdbg_exit_pending(vcpu))
4700 rc = kvm_s390_handle_requests(vcpu);
4704 if (guestdbg_enabled(vcpu)) {
4705 kvm_s390_backup_guest_per_regs(vcpu);
4706 kvm_s390_patch_guest_per_regs(vcpu);
4709 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask);
4711 vcpu->arch.sie_block->icptcode = 0;
4712 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
4713 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
4714 trace_kvm_s390_sie_enter(vcpu, cpuflags);
4719 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
4721 struct kvm_s390_pgm_info pgm_info = {
4722 .code = PGM_ADDRESSING,
4727 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
4728 trace_kvm_s390_sie_fault(vcpu);
4731 * We want to inject an addressing exception, which is defined as a
4732 * suppressing or terminating exception. However, since we came here
4733 * by a DAT access exception, the PSW still points to the faulting
4734 * instruction since DAT exceptions are nullifying. So we've got
4735 * to look up the current opcode to get the length of the instruction
4736 * to be able to forward the PSW.
4738 rc = read_guest_instr(vcpu, vcpu->arch.sie_block->gpsw.addr, &opcode, 1);
4739 ilen = insn_length(opcode);
4743 /* Instruction-Fetching Exceptions - we can't detect the ilen.
4744 * Forward by arbitrary ilc, injection will take care of
4745 * nullification if necessary.
4747 pgm_info = vcpu->arch.pgm;
4750 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
4751 kvm_s390_forward_psw(vcpu, ilen);
4752 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
4755 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
4757 struct mcck_volatile_info *mcck_info;
4758 struct sie_page *sie_page;
4760 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
4761 vcpu->arch.sie_block->icptcode);
4762 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
4764 if (guestdbg_enabled(vcpu))
4765 kvm_s390_restore_guest_per_regs(vcpu);
4767 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
4768 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
4770 if (exit_reason == -EINTR) {
4771 VCPU_EVENT(vcpu, 3, "%s", "machine check");
4772 sie_page = container_of(vcpu->arch.sie_block,
4773 struct sie_page, sie_block);
4774 mcck_info = &sie_page->mcck_info;
4775 kvm_s390_reinject_machine_check(vcpu, mcck_info);
4779 if (vcpu->arch.sie_block->icptcode > 0) {
4780 int rc = kvm_handle_sie_intercept(vcpu);
4782 if (rc != -EOPNOTSUPP)
4784 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
4785 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
4786 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
4787 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
4789 } else if (exit_reason != -EFAULT) {
4790 vcpu->stat.exit_null++;
4792 } else if (kvm_is_ucontrol(vcpu->kvm)) {
4793 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
4794 vcpu->run->s390_ucontrol.trans_exc_code =
4795 current->thread.gmap_addr;
4796 vcpu->run->s390_ucontrol.pgm_code = 0x10;
4798 } else if (current->thread.gmap_pfault) {
4799 trace_kvm_s390_major_guest_pfault(vcpu);
4800 current->thread.gmap_pfault = 0;
4801 if (kvm_arch_setup_async_pf(vcpu))
4803 vcpu->stat.pfault_sync++;
4804 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
4806 return vcpu_post_run_fault_in_sie(vcpu);
4809 #define PSW_INT_MASK (PSW_MASK_EXT | PSW_MASK_IO | PSW_MASK_MCHECK)
4810 static int __vcpu_run(struct kvm_vcpu *vcpu)
4812 int rc, exit_reason;
4813 struct sie_page *sie_page = (struct sie_page *)vcpu->arch.sie_block;
4816 * We try to hold kvm->srcu during most of vcpu_run (except when run-
4817 * ning the guest), so that memslots (and other stuff) are protected
4819 kvm_vcpu_srcu_read_lock(vcpu);
4822 rc = vcpu_pre_run(vcpu);
4823 if (rc || guestdbg_exit_pending(vcpu))
4826 kvm_vcpu_srcu_read_unlock(vcpu);
4828 * As PF_VCPU will be used in fault handler, between
4829 * guest_enter and guest_exit should be no uaccess.
4831 local_irq_disable();
4832 guest_enter_irqoff();
4833 __disable_cpu_timer_accounting(vcpu);
4835 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
4836 memcpy(sie_page->pv_grregs,
4837 vcpu->run->s.regs.gprs,
4838 sizeof(sie_page->pv_grregs));
4840 if (test_cpu_flag(CIF_FPU))
4842 exit_reason = sie64a(vcpu->arch.sie_block,
4843 vcpu->run->s.regs.gprs);
4844 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
4845 memcpy(vcpu->run->s.regs.gprs,
4846 sie_page->pv_grregs,
4847 sizeof(sie_page->pv_grregs));
4849 * We're not allowed to inject interrupts on intercepts
4850 * that leave the guest state in an "in-between" state
4851 * where the next SIE entry will do a continuation.
4852 * Fence interrupts in our "internal" PSW.
4854 if (vcpu->arch.sie_block->icptcode == ICPT_PV_INSTR ||
4855 vcpu->arch.sie_block->icptcode == ICPT_PV_PREF) {
4856 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK;
4859 local_irq_disable();
4860 __enable_cpu_timer_accounting(vcpu);
4861 guest_exit_irqoff();
4863 kvm_vcpu_srcu_read_lock(vcpu);
4865 rc = vcpu_post_run(vcpu, exit_reason);
4866 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
4868 kvm_vcpu_srcu_read_unlock(vcpu);
4872 static void sync_regs_fmt2(struct kvm_vcpu *vcpu)
4874 struct kvm_run *kvm_run = vcpu->run;
4875 struct runtime_instr_cb *riccb;
4878 riccb = (struct runtime_instr_cb *) &kvm_run->s.regs.riccb;
4879 gscb = (struct gs_cb *) &kvm_run->s.regs.gscb;
4880 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
4881 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
4882 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
4883 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
4884 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
4885 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
4887 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
4888 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
4889 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
4890 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
4891 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4892 kvm_clear_async_pf_completion_queue(vcpu);
4894 if (kvm_run->kvm_dirty_regs & KVM_SYNC_DIAG318) {
4895 vcpu->arch.diag318_info.val = kvm_run->s.regs.diag318;
4896 vcpu->arch.sie_block->cpnc = vcpu->arch.diag318_info.cpnc;
4897 VCPU_EVENT(vcpu, 3, "setting cpnc to %d", vcpu->arch.diag318_info.cpnc);
4900 * If userspace sets the riccb (e.g. after migration) to a valid state,
4901 * we should enable RI here instead of doing the lazy enablement.
4903 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) &&
4904 test_kvm_facility(vcpu->kvm, 64) &&
4906 !(vcpu->arch.sie_block->ecb3 & ECB3_RI)) {
4907 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (sync_regs)");
4908 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
4911 * If userspace sets the gscb (e.g. after migration) to non-zero,
4912 * we should enable GS here instead of doing the lazy enablement.
4914 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_GSCB) &&
4915 test_kvm_facility(vcpu->kvm, 133) &&
4917 !vcpu->arch.gs_enabled) {
4918 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (sync_regs)");
4919 vcpu->arch.sie_block->ecb |= ECB_GS;
4920 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
4921 vcpu->arch.gs_enabled = 1;
4923 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_BPBC) &&
4924 test_kvm_facility(vcpu->kvm, 82)) {
4925 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
4926 vcpu->arch.sie_block->fpf |= kvm_run->s.regs.bpbc ? FPF_BPBC : 0;
4928 if (MACHINE_HAS_GS) {
4930 __ctl_set_bit(2, 4);
4931 if (current->thread.gs_cb) {
4932 vcpu->arch.host_gscb = current->thread.gs_cb;
4933 save_gs_cb(vcpu->arch.host_gscb);
4935 if (vcpu->arch.gs_enabled) {
4936 current->thread.gs_cb = (struct gs_cb *)
4937 &vcpu->run->s.regs.gscb;
4938 restore_gs_cb(current->thread.gs_cb);
4942 /* SIE will load etoken directly from SDNX and therefore kvm_run */
4945 static void sync_regs(struct kvm_vcpu *vcpu)
4947 struct kvm_run *kvm_run = vcpu->run;
4949 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
4950 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
4951 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
4952 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
4953 /* some control register changes require a tlb flush */
4954 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
4956 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
4957 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
4958 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
4960 save_access_regs(vcpu->arch.host_acrs);
4961 restore_access_regs(vcpu->run->s.regs.acrs);
4962 /* save host (userspace) fprs/vrs */
4964 vcpu->arch.host_fpregs.fpc = current->thread.fpu.fpc;
4965 vcpu->arch.host_fpregs.regs = current->thread.fpu.regs;
4967 current->thread.fpu.regs = vcpu->run->s.regs.vrs;
4969 current->thread.fpu.regs = vcpu->run->s.regs.fprs;
4970 current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
4971 if (test_fp_ctl(current->thread.fpu.fpc))
4972 /* User space provided an invalid FPC, let's clear it */
4973 current->thread.fpu.fpc = 0;
4975 /* Sync fmt2 only data */
4976 if (likely(!kvm_s390_pv_cpu_is_protected(vcpu))) {
4977 sync_regs_fmt2(vcpu);
4980 * In several places we have to modify our internal view to
4981 * not do things that are disallowed by the ultravisor. For
4982 * example we must not inject interrupts after specific exits
4983 * (e.g. 112 prefix page not secure). We do this by turning
4984 * off the machine check, external and I/O interrupt bits
4985 * of our PSW copy. To avoid getting validity intercepts, we
4986 * do only accept the condition code from userspace.
4988 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_CC;
4989 vcpu->arch.sie_block->gpsw.mask |= kvm_run->psw_mask &
4993 kvm_run->kvm_dirty_regs = 0;
4996 static void store_regs_fmt2(struct kvm_vcpu *vcpu)
4998 struct kvm_run *kvm_run = vcpu->run;
5000 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
5001 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
5002 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
5003 kvm_run->s.regs.bpbc = (vcpu->arch.sie_block->fpf & FPF_BPBC) == FPF_BPBC;
5004 kvm_run->s.regs.diag318 = vcpu->arch.diag318_info.val;
5005 if (MACHINE_HAS_GS) {
5007 __ctl_set_bit(2, 4);
5008 if (vcpu->arch.gs_enabled)
5009 save_gs_cb(current->thread.gs_cb);
5010 current->thread.gs_cb = vcpu->arch.host_gscb;
5011 restore_gs_cb(vcpu->arch.host_gscb);
5012 if (!vcpu->arch.host_gscb)
5013 __ctl_clear_bit(2, 4);
5014 vcpu->arch.host_gscb = NULL;
5017 /* SIE will save etoken directly into SDNX and therefore kvm_run */
5020 static void store_regs(struct kvm_vcpu *vcpu)
5022 struct kvm_run *kvm_run = vcpu->run;
5024 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
5025 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
5026 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
5027 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
5028 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
5029 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
5030 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
5031 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
5032 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
5033 save_access_regs(vcpu->run->s.regs.acrs);
5034 restore_access_regs(vcpu->arch.host_acrs);
5035 /* Save guest register state */
5037 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
5038 /* Restore will be done lazily at return */
5039 current->thread.fpu.fpc = vcpu->arch.host_fpregs.fpc;
5040 current->thread.fpu.regs = vcpu->arch.host_fpregs.regs;
5041 if (likely(!kvm_s390_pv_cpu_is_protected(vcpu)))
5042 store_regs_fmt2(vcpu);
5045 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
5047 struct kvm_run *kvm_run = vcpu->run;
5051 * Running a VM while dumping always has the potential to
5052 * produce inconsistent dump data. But for PV vcpus a SIE
5053 * entry while dumping could also lead to a fatal validity
5054 * intercept which we absolutely want to avoid.
5056 if (vcpu->kvm->arch.pv.dumping)
5059 if (kvm_run->immediate_exit)
5062 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_S390_VALID_FIELDS ||
5063 kvm_run->kvm_dirty_regs & ~KVM_SYNC_S390_VALID_FIELDS)
5068 if (guestdbg_exit_pending(vcpu)) {
5069 kvm_s390_prepare_debug_exit(vcpu);
5074 kvm_sigset_activate(vcpu);
5077 * no need to check the return value of vcpu_start as it can only have
5078 * an error for protvirt, but protvirt means user cpu state
5080 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
5081 kvm_s390_vcpu_start(vcpu);
5082 } else if (is_vcpu_stopped(vcpu)) {
5083 pr_err_ratelimited("can't run stopped vcpu %d\n",
5090 enable_cpu_timer_accounting(vcpu);
5093 rc = __vcpu_run(vcpu);
5095 if (signal_pending(current) && !rc) {
5096 kvm_run->exit_reason = KVM_EXIT_INTR;
5100 if (guestdbg_exit_pending(vcpu) && !rc) {
5101 kvm_s390_prepare_debug_exit(vcpu);
5105 if (rc == -EREMOTE) {
5106 /* userspace support is needed, kvm_run has been prepared */
5110 disable_cpu_timer_accounting(vcpu);
5113 kvm_sigset_deactivate(vcpu);
5115 vcpu->stat.exit_userspace++;
5122 * store status at address
5123 * we use have two special cases:
5124 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
5125 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
5127 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
5129 unsigned char archmode = 1;
5130 freg_t fprs[NUM_FPRS];
5135 px = kvm_s390_get_prefix(vcpu);
5136 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
5137 if (write_guest_abs(vcpu, 163, &archmode, 1))
5140 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
5141 if (write_guest_real(vcpu, 163, &archmode, 1))
5145 gpa -= __LC_FPREGS_SAVE_AREA;
5147 /* manually convert vector registers if necessary */
5148 if (MACHINE_HAS_VX) {
5149 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
5150 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
5153 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
5154 vcpu->run->s.regs.fprs, 128);
5156 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
5157 vcpu->run->s.regs.gprs, 128);
5158 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
5159 &vcpu->arch.sie_block->gpsw, 16);
5160 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
5162 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
5163 &vcpu->run->s.regs.fpc, 4);
5164 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
5165 &vcpu->arch.sie_block->todpr, 4);
5166 cputm = kvm_s390_get_cpu_timer(vcpu);
5167 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
5169 clkcomp = vcpu->arch.sie_block->ckc >> 8;
5170 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
5172 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
5173 &vcpu->run->s.regs.acrs, 64);
5174 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
5175 &vcpu->arch.sie_block->gcr, 128);
5176 return rc ? -EFAULT : 0;
5179 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
5182 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
5183 * switch in the run ioctl. Let's update our copies before we save
5184 * it into the save area
5187 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
5188 save_access_regs(vcpu->run->s.regs.acrs);
5190 return kvm_s390_store_status_unloaded(vcpu, addr);
5193 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
5195 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
5196 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
5199 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
5202 struct kvm_vcpu *vcpu;
5204 kvm_for_each_vcpu(i, vcpu, kvm) {
5205 __disable_ibs_on_vcpu(vcpu);
5209 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
5213 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
5214 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
5217 int kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
5219 int i, online_vcpus, r = 0, started_vcpus = 0;
5221 if (!is_vcpu_stopped(vcpu))
5224 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
5225 /* Only one cpu at a time may enter/leave the STOPPED state. */
5226 spin_lock(&vcpu->kvm->arch.start_stop_lock);
5227 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
5229 /* Let's tell the UV that we want to change into the operating state */
5230 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5231 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR);
5233 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5238 for (i = 0; i < online_vcpus; i++) {
5239 if (!is_vcpu_stopped(kvm_get_vcpu(vcpu->kvm, i)))
5243 if (started_vcpus == 0) {
5244 /* we're the only active VCPU -> speed it up */
5245 __enable_ibs_on_vcpu(vcpu);
5246 } else if (started_vcpus == 1) {
5248 * As we are starting a second VCPU, we have to disable
5249 * the IBS facility on all VCPUs to remove potentially
5250 * outstanding ENABLE requests.
5252 __disable_ibs_on_all_vcpus(vcpu->kvm);
5255 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_STOPPED);
5257 * The real PSW might have changed due to a RESTART interpreted by the
5258 * ultravisor. We block all interrupts and let the next sie exit
5261 if (kvm_s390_pv_cpu_is_protected(vcpu))
5262 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK;
5264 * Another VCPU might have used IBS while we were offline.
5265 * Let's play safe and flush the VCPU at startup.
5267 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5268 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5272 int kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
5274 int i, online_vcpus, r = 0, started_vcpus = 0;
5275 struct kvm_vcpu *started_vcpu = NULL;
5277 if (is_vcpu_stopped(vcpu))
5280 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
5281 /* Only one cpu at a time may enter/leave the STOPPED state. */
5282 spin_lock(&vcpu->kvm->arch.start_stop_lock);
5283 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
5285 /* Let's tell the UV that we want to change into the stopped state */
5286 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5287 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_STP);
5289 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5295 * Set the VCPU to STOPPED and THEN clear the interrupt flag,
5296 * now that the SIGP STOP and SIGP STOP AND STORE STATUS orders
5297 * have been fully processed. This will ensure that the VCPU
5298 * is kept BUSY if another VCPU is inquiring with SIGP SENSE.
5300 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOPPED);
5301 kvm_s390_clear_stop_irq(vcpu);
5303 __disable_ibs_on_vcpu(vcpu);
5305 for (i = 0; i < online_vcpus; i++) {
5306 struct kvm_vcpu *tmp = kvm_get_vcpu(vcpu->kvm, i);
5308 if (!is_vcpu_stopped(tmp)) {
5314 if (started_vcpus == 1) {
5316 * As we only have one VCPU left, we want to enable the
5317 * IBS facility for that VCPU to speed it up.
5319 __enable_ibs_on_vcpu(started_vcpu);
5322 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5326 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5327 struct kvm_enable_cap *cap)
5335 case KVM_CAP_S390_CSS_SUPPORT:
5336 if (!vcpu->kvm->arch.css_support) {
5337 vcpu->kvm->arch.css_support = 1;
5338 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
5339 trace_kvm_s390_enable_css(vcpu->kvm);
5350 static long kvm_s390_vcpu_sida_op(struct kvm_vcpu *vcpu,
5351 struct kvm_s390_mem_op *mop)
5353 void __user *uaddr = (void __user *)mop->buf;
5357 if (mop->flags || !mop->size)
5359 if (mop->size + mop->sida_offset < mop->size)
5361 if (mop->size + mop->sida_offset > sida_size(vcpu->arch.sie_block))
5363 if (!kvm_s390_pv_cpu_is_protected(vcpu))
5366 sida_addr = (char *)sida_addr(vcpu->arch.sie_block) + mop->sida_offset;
5369 case KVM_S390_MEMOP_SIDA_READ:
5370 if (copy_to_user(uaddr, sida_addr, mop->size))
5374 case KVM_S390_MEMOP_SIDA_WRITE:
5375 if (copy_from_user(sida_addr, uaddr, mop->size))
5382 static long kvm_s390_vcpu_mem_op(struct kvm_vcpu *vcpu,
5383 struct kvm_s390_mem_op *mop)
5385 void __user *uaddr = (void __user *)mop->buf;
5386 enum gacc_mode acc_mode;
5387 void *tmpbuf = NULL;
5390 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_INJECT_EXCEPTION |
5391 KVM_S390_MEMOP_F_CHECK_ONLY |
5392 KVM_S390_MEMOP_F_SKEY_PROTECTION);
5395 if (mop->ar >= NUM_ACRS)
5397 if (kvm_s390_pv_cpu_is_protected(vcpu))
5399 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
5400 tmpbuf = vmalloc(mop->size);
5405 acc_mode = mop->op == KVM_S390_MEMOP_LOGICAL_READ ? GACC_FETCH : GACC_STORE;
5406 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
5407 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size,
5408 acc_mode, mop->key);
5411 if (acc_mode == GACC_FETCH) {
5412 r = read_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf,
5413 mop->size, mop->key);
5416 if (copy_to_user(uaddr, tmpbuf, mop->size)) {
5421 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
5425 r = write_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf,
5426 mop->size, mop->key);
5430 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
5431 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
5438 static long kvm_s390_vcpu_memsida_op(struct kvm_vcpu *vcpu,
5439 struct kvm_s390_mem_op *mop)
5443 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5446 case KVM_S390_MEMOP_LOGICAL_READ:
5447 case KVM_S390_MEMOP_LOGICAL_WRITE:
5448 r = kvm_s390_vcpu_mem_op(vcpu, mop);
5450 case KVM_S390_MEMOP_SIDA_READ:
5451 case KVM_S390_MEMOP_SIDA_WRITE:
5452 /* we are locked against sida going away by the vcpu->mutex */
5453 r = kvm_s390_vcpu_sida_op(vcpu, mop);
5459 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
5463 long kvm_arch_vcpu_async_ioctl(struct file *filp,
5464 unsigned int ioctl, unsigned long arg)
5466 struct kvm_vcpu *vcpu = filp->private_data;
5467 void __user *argp = (void __user *)arg;
5471 case KVM_S390_IRQ: {
5472 struct kvm_s390_irq s390irq;
5474 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
5476 rc = kvm_s390_inject_vcpu(vcpu, &s390irq);
5479 case KVM_S390_INTERRUPT: {
5480 struct kvm_s390_interrupt s390int;
5481 struct kvm_s390_irq s390irq = {};
5483 if (copy_from_user(&s390int, argp, sizeof(s390int)))
5485 if (s390int_to_s390irq(&s390int, &s390irq))
5487 rc = kvm_s390_inject_vcpu(vcpu, &s390irq);
5496 * To simplify single stepping of userspace-emulated instructions,
5497 * KVM_EXIT_S390_SIEIC exit sets KVM_GUESTDBG_EXIT_PENDING (see
5498 * should_handle_per_ifetch()). However, if userspace emulation injects
5499 * an interrupt, it needs to be cleared, so that KVM_EXIT_DEBUG happens
5500 * after (and not before) the interrupt delivery.
5503 vcpu->guest_debug &= ~KVM_GUESTDBG_EXIT_PENDING;
5508 static int kvm_s390_handle_pv_vcpu_dump(struct kvm_vcpu *vcpu,
5509 struct kvm_pv_cmd *cmd)
5511 struct kvm_s390_pv_dmp dmp;
5515 /* Dump initialization is a prerequisite */
5516 if (!vcpu->kvm->arch.pv.dumping)
5519 if (copy_from_user(&dmp, (__u8 __user *)cmd->data, sizeof(dmp)))
5522 /* We only handle this subcmd right now */
5523 if (dmp.subcmd != KVM_PV_DUMP_CPU)
5526 /* CPU dump length is the same as create cpu storage donation. */
5527 if (dmp.buff_len != uv_info.guest_cpu_stor_len)
5530 data = kvzalloc(uv_info.guest_cpu_stor_len, GFP_KERNEL);
5534 ret = kvm_s390_pv_dump_cpu(vcpu, data, &cmd->rc, &cmd->rrc);
5536 VCPU_EVENT(vcpu, 3, "PROTVIRT DUMP CPU %d rc %x rrc %x",
5537 vcpu->vcpu_id, cmd->rc, cmd->rrc);
5542 /* On success copy over the dump data */
5543 if (!ret && copy_to_user((__u8 __user *)dmp.buff_addr, data, uv_info.guest_cpu_stor_len))
5550 long kvm_arch_vcpu_ioctl(struct file *filp,
5551 unsigned int ioctl, unsigned long arg)
5553 struct kvm_vcpu *vcpu = filp->private_data;
5554 void __user *argp = (void __user *)arg;
5562 case KVM_S390_STORE_STATUS:
5563 idx = srcu_read_lock(&vcpu->kvm->srcu);
5564 r = kvm_s390_store_status_unloaded(vcpu, arg);
5565 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5567 case KVM_S390_SET_INITIAL_PSW: {
5571 if (copy_from_user(&psw, argp, sizeof(psw)))
5573 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
5576 case KVM_S390_CLEAR_RESET:
5578 kvm_arch_vcpu_ioctl_clear_reset(vcpu);
5579 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5580 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5581 UVC_CMD_CPU_RESET_CLEAR, &rc, &rrc);
5582 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET CLEAR VCPU: rc %x rrc %x",
5586 case KVM_S390_INITIAL_RESET:
5588 kvm_arch_vcpu_ioctl_initial_reset(vcpu);
5589 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5590 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5591 UVC_CMD_CPU_RESET_INITIAL,
5593 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET INITIAL VCPU: rc %x rrc %x",
5597 case KVM_S390_NORMAL_RESET:
5599 kvm_arch_vcpu_ioctl_normal_reset(vcpu);
5600 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5601 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5602 UVC_CMD_CPU_RESET, &rc, &rrc);
5603 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET NORMAL VCPU: rc %x rrc %x",
5607 case KVM_SET_ONE_REG:
5608 case KVM_GET_ONE_REG: {
5609 struct kvm_one_reg reg;
5611 if (kvm_s390_pv_cpu_is_protected(vcpu))
5614 if (copy_from_user(®, argp, sizeof(reg)))
5616 if (ioctl == KVM_SET_ONE_REG)
5617 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®);
5619 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®);
5622 #ifdef CONFIG_KVM_S390_UCONTROL
5623 case KVM_S390_UCAS_MAP: {
5624 struct kvm_s390_ucas_mapping ucasmap;
5626 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
5631 if (!kvm_is_ucontrol(vcpu->kvm)) {
5636 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
5637 ucasmap.vcpu_addr, ucasmap.length);
5640 case KVM_S390_UCAS_UNMAP: {
5641 struct kvm_s390_ucas_mapping ucasmap;
5643 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
5648 if (!kvm_is_ucontrol(vcpu->kvm)) {
5653 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
5658 case KVM_S390_VCPU_FAULT: {
5659 r = gmap_fault(vcpu->arch.gmap, arg, 0);
5662 case KVM_ENABLE_CAP:
5664 struct kvm_enable_cap cap;
5666 if (copy_from_user(&cap, argp, sizeof(cap)))
5668 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5671 case KVM_S390_MEM_OP: {
5672 struct kvm_s390_mem_op mem_op;
5674 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
5675 r = kvm_s390_vcpu_memsida_op(vcpu, &mem_op);
5680 case KVM_S390_SET_IRQ_STATE: {
5681 struct kvm_s390_irq_state irq_state;
5684 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
5686 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
5687 irq_state.len == 0 ||
5688 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
5692 /* do not use irq_state.flags, it will break old QEMUs */
5693 r = kvm_s390_set_irq_state(vcpu,
5694 (void __user *) irq_state.buf,
5698 case KVM_S390_GET_IRQ_STATE: {
5699 struct kvm_s390_irq_state irq_state;
5702 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
5704 if (irq_state.len == 0) {
5708 /* do not use irq_state.flags, it will break old QEMUs */
5709 r = kvm_s390_get_irq_state(vcpu,
5710 (__u8 __user *) irq_state.buf,
5714 case KVM_S390_PV_CPU_COMMAND: {
5715 struct kvm_pv_cmd cmd;
5718 if (!is_prot_virt_host())
5722 if (copy_from_user(&cmd, argp, sizeof(cmd)))
5729 /* We only handle this cmd right now */
5730 if (cmd.cmd != KVM_PV_DUMP)
5733 r = kvm_s390_handle_pv_vcpu_dump(vcpu, &cmd);
5735 /* Always copy over UV rc / rrc data */
5736 if (copy_to_user((__u8 __user *)argp, &cmd.rc,
5737 sizeof(cmd.rc) + sizeof(cmd.rrc)))
5749 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5751 #ifdef CONFIG_KVM_S390_UCONTROL
5752 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
5753 && (kvm_is_ucontrol(vcpu->kvm))) {
5754 vmf->page = virt_to_page(vcpu->arch.sie_block);
5755 get_page(vmf->page);
5759 return VM_FAULT_SIGBUS;
5762 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
5767 /* Section: memory related */
5768 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5769 const struct kvm_memory_slot *old,
5770 struct kvm_memory_slot *new,
5771 enum kvm_mr_change change)
5775 /* When we are protected, we should not change the memory slots */
5776 if (kvm_s390_pv_get_handle(kvm))
5779 if (change != KVM_MR_DELETE && change != KVM_MR_FLAGS_ONLY) {
5781 * A few sanity checks. We can have memory slots which have to be
5782 * located/ended at a segment boundary (1MB). The memory in userland is
5783 * ok to be fragmented into various different vmas. It is okay to mmap()
5784 * and munmap() stuff in this slot after doing this call at any time
5787 if (new->userspace_addr & 0xffffful)
5790 size = new->npages * PAGE_SIZE;
5791 if (size & 0xffffful)
5794 if ((new->base_gfn * PAGE_SIZE) + size > kvm->arch.mem_limit)
5798 if (!kvm->arch.migration_mode)
5802 * Turn off migration mode when:
5803 * - userspace creates a new memslot with dirty logging off,
5804 * - userspace modifies an existing memslot (MOVE or FLAGS_ONLY) and
5805 * dirty logging is turned off.
5806 * Migration mode expects dirty page logging being enabled to store
5809 if (change != KVM_MR_DELETE &&
5810 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
5811 WARN(kvm_s390_vm_stop_migration(kvm),
5812 "Failed to stop migration mode");
5817 void kvm_arch_commit_memory_region(struct kvm *kvm,
5818 struct kvm_memory_slot *old,
5819 const struct kvm_memory_slot *new,
5820 enum kvm_mr_change change)
5826 rc = gmap_unmap_segment(kvm->arch.gmap, old->base_gfn * PAGE_SIZE,
5827 old->npages * PAGE_SIZE);
5830 rc = gmap_unmap_segment(kvm->arch.gmap, old->base_gfn * PAGE_SIZE,
5831 old->npages * PAGE_SIZE);
5836 rc = gmap_map_segment(kvm->arch.gmap, new->userspace_addr,
5837 new->base_gfn * PAGE_SIZE,
5838 new->npages * PAGE_SIZE);
5840 case KVM_MR_FLAGS_ONLY:
5843 WARN(1, "Unknown KVM MR CHANGE: %d\n", change);
5846 pr_warn("failed to commit memory region\n");
5850 static inline unsigned long nonhyp_mask(int i)
5852 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
5854 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
5857 static int __init kvm_s390_init(void)
5861 if (!sclp.has_sief2) {
5862 pr_info("SIE is not available\n");
5866 if (nested && hpage) {
5867 pr_info("A KVM host that supports nesting cannot back its KVM guests with huge pages\n");
5871 for (i = 0; i < 16; i++)
5872 kvm_s390_fac_base[i] |=
5873 stfle_fac_list[i] & nonhyp_mask(i);
5875 r = __kvm_s390_init();
5879 r = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
5887 static void __exit kvm_s390_exit(void)
5894 module_init(kvm_s390_init);
5895 module_exit(kvm_s390_exit);
5898 * Enable autoloading of the kvm module.
5899 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
5900 * since x86 takes a different approach.
5902 #include <linux/miscdevice.h>
5903 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5904 MODULE_ALIAS("devname:kvm");