projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge remote-tracking branch 'mailbox/mailbox-for-next'
[deliverable/linux.git] / arch / s390 / kvm / kvm-s390.c
1 /*
2 * hosting zSeries kernel virtual machines
3 *
4 * Copyright IBM Corp. 2008, 2009
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License (version 2 only)
8 * as published by the Free Software Foundation.
9 *
10 * Author(s): Carsten Otte <cotte@de.ibm.com>
11 * Christian Borntraeger <borntraeger@de.ibm.com>
12 * Heiko Carstens <heiko.carstens@de.ibm.com>
13 * Christian Ehrhardt <ehrhardt@de.ibm.com>
14 * Jason J. Herne <jjherne@us.ibm.com>
15 */
16
17 #include <linux/compiler.h>
18 #include <linux/err.h>
19 #include <linux/fs.h>
20 #include <linux/hrtimer.h>
21 #include <linux/init.h>
22 #include <linux/kvm.h>
23 #include <linux/kvm_host.h>
24 #include <linux/mman.h>
25 #include <linux/module.h>
26 #include <linux/random.h>
27 #include <linux/slab.h>
28 #include <linux/timer.h>
29 #include <linux/vmalloc.h>
30 #include <linux/bitmap.h>
31 #include <asm/asm-offsets.h>
32 #include <asm/lowcore.h>
33 #include <asm/stp.h>
34 #include <asm/pgtable.h>
35 #include <asm/gmap.h>
36 #include <asm/nmi.h>
37 #include <asm/switch_to.h>
38 #include <asm/isc.h>
39 #include <asm/sclp.h>
40 #include <asm/cpacf.h>
41 #include <asm/timex.h>
42 #include "kvm-s390.h"
43 #include "gaccess.h"
44
45 #define KMSG_COMPONENT "kvm-s390"
46 #undef pr_fmt
47 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
48
49 #define CREATE_TRACE_POINTS
50 #include "trace.h"
51 #include "trace-s390.h"
52
53 #define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */
54 #define LOCAL_IRQS 32
55 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
56 (KVM_MAX_VCPUS + LOCAL_IRQS))
57
58 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
59
60 struct kvm_stats_debugfs_item debugfs_entries[] = {
61 { "userspace_handled", VCPU_STAT(exit_userspace) },
62 { "exit_null", VCPU_STAT(exit_null) },
63 { "exit_validity", VCPU_STAT(exit_validity) },
64 { "exit_stop_request", VCPU_STAT(exit_stop_request) },
65 { "exit_external_request", VCPU_STAT(exit_external_request) },
66 { "exit_external_interrupt", VCPU_STAT(exit_external_interrupt) },
67 { "exit_instruction", VCPU_STAT(exit_instruction) },
68 { "exit_pei", VCPU_STAT(exit_pei) },
69 { "exit_program_interruption", VCPU_STAT(exit_program_interruption) },
70 { "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program) },
71 { "exit_operation_exception", VCPU_STAT(exit_operation_exception) },
72 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
73 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
74 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
75 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
76 { "instruction_lctlg", VCPU_STAT(instruction_lctlg) },
77 { "instruction_lctl", VCPU_STAT(instruction_lctl) },
78 { "instruction_stctl", VCPU_STAT(instruction_stctl) },
79 { "instruction_stctg", VCPU_STAT(instruction_stctg) },
80 { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal) },
81 { "deliver_external_call", VCPU_STAT(deliver_external_call) },
82 { "deliver_service_signal", VCPU_STAT(deliver_service_signal) },
83 { "deliver_virtio_interrupt", VCPU_STAT(deliver_virtio_interrupt) },
84 { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal) },
85 { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal) },
86 { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal) },
87 { "deliver_program_interruption", VCPU_STAT(deliver_program_int) },
88 { "exit_wait_state", VCPU_STAT(exit_wait_state) },
89 { "instruction_pfmf", VCPU_STAT(instruction_pfmf) },
90 { "instruction_stidp", VCPU_STAT(instruction_stidp) },
91 { "instruction_spx", VCPU_STAT(instruction_spx) },
92 { "instruction_stpx", VCPU_STAT(instruction_stpx) },
93 { "instruction_stap", VCPU_STAT(instruction_stap) },
94 { "instruction_storage_key", VCPU_STAT(instruction_storage_key) },
95 { "instruction_ipte_interlock", VCPU_STAT(instruction_ipte_interlock) },
96 { "instruction_stsch", VCPU_STAT(instruction_stsch) },
97 { "instruction_chsc", VCPU_STAT(instruction_chsc) },
98 { "instruction_essa", VCPU_STAT(instruction_essa) },
99 { "instruction_stsi", VCPU_STAT(instruction_stsi) },
100 { "instruction_stfl", VCPU_STAT(instruction_stfl) },
101 { "instruction_tprot", VCPU_STAT(instruction_tprot) },
102 { "instruction_sthyi", VCPU_STAT(instruction_sthyi) },
103 { "instruction_sie", VCPU_STAT(instruction_sie) },
104 { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense) },
105 { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running) },
106 { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call) },
107 { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency) },
108 { "instruction_sigp_cond_emergency", VCPU_STAT(instruction_sigp_cond_emergency) },
109 { "instruction_sigp_start", VCPU_STAT(instruction_sigp_start) },
110 { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop) },
111 { "instruction_sigp_stop_store_status", VCPU_STAT(instruction_sigp_stop_store_status) },
112 { "instruction_sigp_store_status", VCPU_STAT(instruction_sigp_store_status) },
113 { "instruction_sigp_store_adtl_status", VCPU_STAT(instruction_sigp_store_adtl_status) },
114 { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch) },
115 { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix) },
116 { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart) },
117 { "instruction_sigp_cpu_reset", VCPU_STAT(instruction_sigp_cpu_reset) },
118 { "instruction_sigp_init_cpu_reset", VCPU_STAT(instruction_sigp_init_cpu_reset) },
119 { "instruction_sigp_unknown", VCPU_STAT(instruction_sigp_unknown) },
120 { "diagnose_10", VCPU_STAT(diagnose_10) },
121 { "diagnose_44", VCPU_STAT(diagnose_44) },
122 { "diagnose_9c", VCPU_STAT(diagnose_9c) },
123 { "diagnose_258", VCPU_STAT(diagnose_258) },
124 { "diagnose_308", VCPU_STAT(diagnose_308) },
125 { "diagnose_500", VCPU_STAT(diagnose_500) },
126 { NULL }
127 };
128
129 /* allow nested virtualization in KVM (if enabled by user space) */
130 static int nested;
131 module_param(nested, int, S_IRUGO);
132 MODULE_PARM_DESC(nested, "Nested virtualization support");
133
134 /* upper facilities limit for kvm */
135 unsigned long kvm_s390_fac_list_mask[16] = { FACILITIES_KVM };
136
137 unsigned long kvm_s390_fac_list_mask_size(void)
138 {
139 BUILD_BUG_ON(ARRAY_SIZE(kvm_s390_fac_list_mask) > S390_ARCH_FAC_MASK_SIZE_U64);
140 return ARRAY_SIZE(kvm_s390_fac_list_mask);
141 }
142
143 /* available cpu features supported by kvm */
144 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
145 /* available subfunctions indicated via query / "test bit" */
146 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
147
148 static struct gmap_notifier gmap_notifier;
149 static struct gmap_notifier vsie_gmap_notifier;
150 debug_info_t *kvm_s390_dbf;
151
152 /* Section: not file related */
153 int kvm_arch_hardware_enable(void)
154 {
155 /* every s390 is virtualization enabled ;-) */
156 return 0;
157 }
158
159 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
160 unsigned long end);
161
162 /*
163 * This callback is executed during stop_machine(). All CPUs are therefore
164 * temporarily stopped. In order not to change guest behavior, we have to
165 * disable preemption whenever we touch the epoch of kvm and the VCPUs,
166 * so a CPU won't be stopped while calculating with the epoch.
167 */
168 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
169 void *v)
170 {
171 struct kvm *kvm;
172 struct kvm_vcpu *vcpu;
173 int i;
174 unsigned long long *delta = v;
175
176 list_for_each_entry(kvm, &vm_list, vm_list) {
177 kvm->arch.epoch -= *delta;
178 kvm_for_each_vcpu(i, vcpu, kvm) {
179 vcpu->arch.sie_block->epoch -= *delta;
180 if (vcpu->arch.cputm_enabled)
181 vcpu->arch.cputm_start += *delta;
182 if (vcpu->arch.vsie_block)
183 vcpu->arch.vsie_block->epoch -= *delta;
184 }
185 }
186 return NOTIFY_OK;
187 }
188
189 static struct notifier_block kvm_clock_notifier = {
190 .notifier_call = kvm_clock_sync,
191 };
192
193 int kvm_arch_hardware_setup(void)
194 {
195 gmap_notifier.notifier_call = kvm_gmap_notifier;
196 gmap_register_pte_notifier(&gmap_notifier);
197 vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
198 gmap_register_pte_notifier(&vsie_gmap_notifier);
199 atomic_notifier_chain_register(&s390_epoch_delta_notifier,
200 &kvm_clock_notifier);
201 return 0;
202 }
203
204 void kvm_arch_hardware_unsetup(void)
205 {
206 gmap_unregister_pte_notifier(&gmap_notifier);
207 gmap_unregister_pte_notifier(&vsie_gmap_notifier);
208 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
209 &kvm_clock_notifier);
210 }
211
212 static void allow_cpu_feat(unsigned long nr)
213 {
214 set_bit_inv(nr, kvm_s390_available_cpu_feat);
215 }
216
217 static inline int plo_test_bit(unsigned char nr)
218 {
219 register unsigned long r0 asm("0") = (unsigned long) nr | 0x100;
220 int cc = 3; /* subfunction not available */
221
222 asm volatile(
223 /* Parameter registers are ignored for "test bit" */
224 " plo 0,0,0,0(0)\n"
225 " ipm %0\n"
226 " srl %0,28\n"
227 : "=d" (cc)
228 : "d" (r0)
229 : "cc");
230 return cc == 0;
231 }
232
233 static void kvm_s390_cpu_feat_init(void)
234 {
235 int i;
236
237 for (i = 0; i < 256; ++i) {
238 if (plo_test_bit(i))
239 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
240 }
241
242 if (test_facility(28)) /* TOD-clock steering */
243 ptff(kvm_s390_available_subfunc.ptff,
244 sizeof(kvm_s390_available_subfunc.ptff),
245 PTFF_QAF);
246
247 if (test_facility(17)) { /* MSA */
248 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
249 kvm_s390_available_subfunc.kmac);
250 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
251 kvm_s390_available_subfunc.kmc);
252 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
253 kvm_s390_available_subfunc.km);
254 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
255 kvm_s390_available_subfunc.kimd);
256 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
257 kvm_s390_available_subfunc.klmd);
258 }
259 if (test_facility(76)) /* MSA3 */
260 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
261 kvm_s390_available_subfunc.pckmo);
262 if (test_facility(77)) { /* MSA4 */
263 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
264 kvm_s390_available_subfunc.kmctr);
265 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
266 kvm_s390_available_subfunc.kmf);
267 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
268 kvm_s390_available_subfunc.kmo);
269 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
270 kvm_s390_available_subfunc.pcc);
271 }
272 if (test_facility(57)) /* MSA5 */
273 __cpacf_query(CPACF_PPNO, (cpacf_mask_t *)
274 kvm_s390_available_subfunc.ppno);
275
276 if (MACHINE_HAS_ESOP)
277 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
278 /*
279 * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
280 * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
281 */
282 if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
283 !test_facility(3) || !nested)
284 return;
285 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
286 if (sclp.has_64bscao)
287 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
288 if (sclp.has_siif)
289 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
290 if (sclp.has_gpere)
291 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
292 if (sclp.has_gsls)
293 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
294 if (sclp.has_ib)
295 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
296 if (sclp.has_cei)
297 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
298 if (sclp.has_ibs)
299 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
300 /*
301 * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
302 * all skey handling functions read/set the skey from the PGSTE
303 * instead of the real storage key.
304 *
305 * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
306 * pages being detected as preserved although they are resident.
307 *
308 * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
309 * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
310 *
311 * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
312 * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
313 * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
314 *
315 * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
316 * cannot easily shadow the SCA because of the ipte lock.
317 */
318 }
319
320 int kvm_arch_init(void *opaque)
321 {
322 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
323 if (!kvm_s390_dbf)
324 return -ENOMEM;
325
326 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view)) {
327 debug_unregister(kvm_s390_dbf);
328 return -ENOMEM;
329 }
330
331 kvm_s390_cpu_feat_init();
332
333 /* Register floating interrupt controller interface. */
334 return kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
335 }
336
337 void kvm_arch_exit(void)
338 {
339 debug_unregister(kvm_s390_dbf);
340 }
341
342 /* Section: device related */
343 long kvm_arch_dev_ioctl(struct file *filp,
344 unsigned int ioctl, unsigned long arg)
345 {
346 if (ioctl == KVM_S390_ENABLE_SIE)
347 return s390_enable_sie();
348 return -EINVAL;
349 }
350
351 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
352 {
353 int r;
354
355 switch (ext) {
356 case KVM_CAP_S390_PSW:
357 case KVM_CAP_S390_GMAP:
358 case KVM_CAP_SYNC_MMU:
359 #ifdef CONFIG_KVM_S390_UCONTROL
360 case KVM_CAP_S390_UCONTROL:
361 #endif
362 case KVM_CAP_ASYNC_PF:
363 case KVM_CAP_SYNC_REGS:
364 case KVM_CAP_ONE_REG:
365 case KVM_CAP_ENABLE_CAP:
366 case KVM_CAP_S390_CSS_SUPPORT:
367 case KVM_CAP_IOEVENTFD:
368 case KVM_CAP_DEVICE_CTRL:
369 case KVM_CAP_ENABLE_CAP_VM:
370 case KVM_CAP_S390_IRQCHIP:
371 case KVM_CAP_VM_ATTRIBUTES:
372 case KVM_CAP_MP_STATE:
373 case KVM_CAP_S390_INJECT_IRQ:
374 case KVM_CAP_S390_USER_SIGP:
375 case KVM_CAP_S390_USER_STSI:
376 case KVM_CAP_S390_SKEYS:
377 case KVM_CAP_S390_IRQ_STATE:
378 case KVM_CAP_S390_USER_INSTR0:
379 r = 1;
380 break;
381 case KVM_CAP_S390_MEM_OP:
382 r = MEM_OP_MAX_SIZE;
383 break;
384 case KVM_CAP_NR_VCPUS:
385 case KVM_CAP_MAX_VCPUS:
386 r = KVM_S390_BSCA_CPU_SLOTS;
387 if (sclp.has_esca && sclp.has_64bscao)
388 r = KVM_S390_ESCA_CPU_SLOTS;
389 break;
390 case KVM_CAP_NR_MEMSLOTS:
391 r = KVM_USER_MEM_SLOTS;
392 break;
393 case KVM_CAP_S390_COW:
394 r = MACHINE_HAS_ESOP;
395 break;
396 case KVM_CAP_S390_VECTOR_REGISTERS:
397 r = MACHINE_HAS_VX;
398 break;
399 case KVM_CAP_S390_RI:
400 r = test_facility(64);
401 break;
402 default:
403 r = 0;
404 }
405 return r;
406 }
407
408 static void kvm_s390_sync_dirty_log(struct kvm *kvm,
409 struct kvm_memory_slot *memslot)
410 {
411 gfn_t cur_gfn, last_gfn;
412 unsigned long address;
413 struct gmap *gmap = kvm->arch.gmap;
414
415 /* Loop over all guest pages */
416 last_gfn = memslot->base_gfn + memslot->npages;
417 for (cur_gfn = memslot->base_gfn; cur_gfn <= last_gfn; cur_gfn++) {
418 address = gfn_to_hva_memslot(memslot, cur_gfn);
419
420 if (test_and_clear_guest_dirty(gmap->mm, address))
421 mark_page_dirty(kvm, cur_gfn);
422 if (fatal_signal_pending(current))
423 return;
424 cond_resched();
425 }
426 }
427
428 /* Section: vm related */
429 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
430
431 /*
432 * Get (and clear) the dirty memory log for a memory slot.
433 */
434 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
435 struct kvm_dirty_log *log)
436 {
437 int r;
438 unsigned long n;
439 struct kvm_memslots *slots;
440 struct kvm_memory_slot *memslot;
441 int is_dirty = 0;
442
443 mutex_lock(&kvm->slots_lock);
444
445 r = -EINVAL;
446 if (log->slot >= KVM_USER_MEM_SLOTS)
447 goto out;
448
449 slots = kvm_memslots(kvm);
450 memslot = id_to_memslot(slots, log->slot);
451 r = -ENOENT;
452 if (!memslot->dirty_bitmap)
453 goto out;
454
455 kvm_s390_sync_dirty_log(kvm, memslot);
456 r = kvm_get_dirty_log(kvm, log, &is_dirty);
457 if (r)
458 goto out;
459
460 /* Clear the dirty log */
461 if (is_dirty) {
462 n = kvm_dirty_bitmap_bytes(memslot);
463 memset(memslot->dirty_bitmap, 0, n);
464 }
465 r = 0;
466 out:
467 mutex_unlock(&kvm->slots_lock);
468 return r;
469 }
470
471 static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
472 {
473 unsigned int i;
474 struct kvm_vcpu *vcpu;
475
476 kvm_for_each_vcpu(i, vcpu, kvm) {
477 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
478 }
479 }
480
481 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
482 {
483 int r;
484
485 if (cap->flags)
486 return -EINVAL;
487
488 switch (cap->cap) {
489 case KVM_CAP_S390_IRQCHIP:
490 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
491 kvm->arch.use_irqchip = 1;
492 r = 0;
493 break;
494 case KVM_CAP_S390_USER_SIGP:
495 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
496 kvm->arch.user_sigp = 1;
497 r = 0;
498 break;
499 case KVM_CAP_S390_VECTOR_REGISTERS:
500 mutex_lock(&kvm->lock);
501 if (kvm->created_vcpus) {
502 r = -EBUSY;
503 } else if (MACHINE_HAS_VX) {
504 set_kvm_facility(kvm->arch.model.fac_mask, 129);
505 set_kvm_facility(kvm->arch.model.fac_list, 129);
506 r = 0;
507 } else
508 r = -EINVAL;
509 mutex_unlock(&kvm->lock);
510 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
511 r ? "(not available)" : "(success)");
512 break;
513 case KVM_CAP_S390_RI:
514 r = -EINVAL;
515 mutex_lock(&kvm->lock);
516 if (kvm->created_vcpus) {
517 r = -EBUSY;
518 } else if (test_facility(64)) {
519 set_kvm_facility(kvm->arch.model.fac_mask, 64);
520 set_kvm_facility(kvm->arch.model.fac_list, 64);
521 r = 0;
522 }
523 mutex_unlock(&kvm->lock);
524 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
525 r ? "(not available)" : "(success)");
526 break;
527 case KVM_CAP_S390_USER_STSI:
528 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
529 kvm->arch.user_stsi = 1;
530 r = 0;
531 break;
532 case KVM_CAP_S390_USER_INSTR0:
533 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
534 kvm->arch.user_instr0 = 1;
535 icpt_operexc_on_all_vcpus(kvm);
536 r = 0;
537 break;
538 default:
539 r = -EINVAL;
540 break;
541 }
542 return r;
543 }
544
545 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
546 {
547 int ret;
548
549 switch (attr->attr) {
550 case KVM_S390_VM_MEM_LIMIT_SIZE:
551 ret = 0;
552 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
553 kvm->arch.mem_limit);
554 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
555 ret = -EFAULT;
556 break;
557 default:
558 ret = -ENXIO;
559 break;
560 }
561 return ret;
562 }
563
564 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
565 {
566 int ret;
567 unsigned int idx;
568 switch (attr->attr) {
569 case KVM_S390_VM_MEM_ENABLE_CMMA:
570 ret = -ENXIO;
571 if (!sclp.has_cmma)
572 break;
573
574 ret = -EBUSY;
575 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
576 mutex_lock(&kvm->lock);
577 if (!kvm->created_vcpus) {
578 kvm->arch.use_cmma = 1;
579 ret = 0;
580 }
581 mutex_unlock(&kvm->lock);
582 break;
583 case KVM_S390_VM_MEM_CLR_CMMA:
584 ret = -ENXIO;
585 if (!sclp.has_cmma)
586 break;
587 ret = -EINVAL;
588 if (!kvm->arch.use_cmma)
589 break;
590
591 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
592 mutex_lock(&kvm->lock);
593 idx = srcu_read_lock(&kvm->srcu);
594 s390_reset_cmma(kvm->arch.gmap->mm);
595 srcu_read_unlock(&kvm->srcu, idx);
596 mutex_unlock(&kvm->lock);
597 ret = 0;
598 break;
599 case KVM_S390_VM_MEM_LIMIT_SIZE: {
600 unsigned long new_limit;
601
602 if (kvm_is_ucontrol(kvm))
603 return -EINVAL;
604
605 if (get_user(new_limit, (u64 __user *)attr->addr))
606 return -EFAULT;
607
608 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
609 new_limit > kvm->arch.mem_limit)
610 return -E2BIG;
611
612 if (!new_limit)
613 return -EINVAL;
614
615 /* gmap_create takes last usable address */
616 if (new_limit != KVM_S390_NO_MEM_LIMIT)
617 new_limit -= 1;
618
619 ret = -EBUSY;
620 mutex_lock(&kvm->lock);
621 if (!kvm->created_vcpus) {
622 /* gmap_create will round the limit up */
623 struct gmap *new = gmap_create(current->mm, new_limit);
624
625 if (!new) {
626 ret = -ENOMEM;
627 } else {
628 gmap_remove(kvm->arch.gmap);
629 new->private = kvm;
630 kvm->arch.gmap = new;
631 ret = 0;
632 }
633 }
634 mutex_unlock(&kvm->lock);
635 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
636 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
637 (void *) kvm->arch.gmap->asce);
638 break;
639 }
640 default:
641 ret = -ENXIO;
642 break;
643 }
644 return ret;
645 }
646
647 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
648
649 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
650 {
651 struct kvm_vcpu *vcpu;
652 int i;
653
654 if (!test_kvm_facility(kvm, 76))
655 return -EINVAL;
656
657 mutex_lock(&kvm->lock);
658 switch (attr->attr) {
659 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
660 get_random_bytes(
661 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
662 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
663 kvm->arch.crypto.aes_kw = 1;
664 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
665 break;
666 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
667 get_random_bytes(
668 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
669 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
670 kvm->arch.crypto.dea_kw = 1;
671 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
672 break;
673 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
674 kvm->arch.crypto.aes_kw = 0;
675 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
676 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
677 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
678 break;
679 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
680 kvm->arch.crypto.dea_kw = 0;
681 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
682 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
683 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
684 break;
685 default:
686 mutex_unlock(&kvm->lock);
687 return -ENXIO;
688 }
689
690 kvm_for_each_vcpu(i, vcpu, kvm) {
691 kvm_s390_vcpu_crypto_setup(vcpu);
692 exit_sie(vcpu);
693 }
694 mutex_unlock(&kvm->lock);
695 return 0;
696 }
697
698 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
699 {
700 u8 gtod_high;
701
702 if (copy_from_user(&gtod_high, (void __user *)attr->addr,
703 sizeof(gtod_high)))
704 return -EFAULT;
705
706 if (gtod_high != 0)
707 return -EINVAL;
708 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
709
710 return 0;
711 }
712
713 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
714 {
715 u64 gtod;
716
717 if (copy_from_user(&gtod, (void __user *)attr->addr, sizeof(gtod)))
718 return -EFAULT;
719
720 kvm_s390_set_tod_clock(kvm, gtod);
721 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod);
722 return 0;
723 }
724
725 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
726 {
727 int ret;
728
729 if (attr->flags)
730 return -EINVAL;
731
732 switch (attr->attr) {
733 case KVM_S390_VM_TOD_HIGH:
734 ret = kvm_s390_set_tod_high(kvm, attr);
735 break;
736 case KVM_S390_VM_TOD_LOW:
737 ret = kvm_s390_set_tod_low(kvm, attr);
738 break;
739 default:
740 ret = -ENXIO;
741 break;
742 }
743 return ret;
744 }
745
746 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
747 {
748 u8 gtod_high = 0;
749
750 if (copy_to_user((void __user *)attr->addr, &gtod_high,
751 sizeof(gtod_high)))
752 return -EFAULT;
753 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
754
755 return 0;
756 }
757
758 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
759 {
760 u64 gtod;
761
762 gtod = kvm_s390_get_tod_clock_fast(kvm);
763 if (copy_to_user((void __user *)attr->addr, &gtod, sizeof(gtod)))
764 return -EFAULT;
765 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
766
767 return 0;
768 }
769
770 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
771 {
772 int ret;
773
774 if (attr->flags)
775 return -EINVAL;
776
777 switch (attr->attr) {
778 case KVM_S390_VM_TOD_HIGH:
779 ret = kvm_s390_get_tod_high(kvm, attr);
780 break;
781 case KVM_S390_VM_TOD_LOW:
782 ret = kvm_s390_get_tod_low(kvm, attr);
783 break;
784 default:
785 ret = -ENXIO;
786 break;
787 }
788 return ret;
789 }
790
791 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
792 {
793 struct kvm_s390_vm_cpu_processor *proc;
794 u16 lowest_ibc, unblocked_ibc;
795 int ret = 0;
796
797 mutex_lock(&kvm->lock);
798 if (kvm->created_vcpus) {
799 ret = -EBUSY;
800 goto out;
801 }
802 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
803 if (!proc) {
804 ret = -ENOMEM;
805 goto out;
806 }
807 if (!copy_from_user(proc, (void __user *)attr->addr,
808 sizeof(*proc))) {
809 kvm->arch.model.cpuid = proc->cpuid;
810 lowest_ibc = sclp.ibc >> 16 & 0xfff;
811 unblocked_ibc = sclp.ibc & 0xfff;
812 if (lowest_ibc && proc->ibc) {
813 if (proc->ibc > unblocked_ibc)
814 kvm->arch.model.ibc = unblocked_ibc;
815 else if (proc->ibc < lowest_ibc)
816 kvm->arch.model.ibc = lowest_ibc;
817 else
818 kvm->arch.model.ibc = proc->ibc;
819 }
820 memcpy(kvm->arch.model.fac_list, proc->fac_list,
821 S390_ARCH_FAC_LIST_SIZE_BYTE);
822 } else
823 ret = -EFAULT;
824 kfree(proc);
825 out:
826 mutex_unlock(&kvm->lock);
827 return ret;
828 }
829
830 static int kvm_s390_set_processor_feat(struct kvm *kvm,
831 struct kvm_device_attr *attr)
832 {
833 struct kvm_s390_vm_cpu_feat data;
834 int ret = -EBUSY;
835
836 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
837 return -EFAULT;
838 if (!bitmap_subset((unsigned long *) data.feat,
839 kvm_s390_available_cpu_feat,
840 KVM_S390_VM_CPU_FEAT_NR_BITS))
841 return -EINVAL;
842
843 mutex_lock(&kvm->lock);
844 if (!atomic_read(&kvm->online_vcpus)) {
845 bitmap_copy(kvm->arch.cpu_feat, (unsigned long *) data.feat,
846 KVM_S390_VM_CPU_FEAT_NR_BITS);
847 ret = 0;
848 }
849 mutex_unlock(&kvm->lock);
850 return ret;
851 }
852
853 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
854 struct kvm_device_attr *attr)
855 {
856 /*
857 * Once supported by kernel + hw, we have to store the subfunctions
858 * in kvm->arch and remember that user space configured them.
859 */
860 return -ENXIO;
861 }
862
863 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
864 {
865 int ret = -ENXIO;
866
867 switch (attr->attr) {
868 case KVM_S390_VM_CPU_PROCESSOR:
869 ret = kvm_s390_set_processor(kvm, attr);
870 break;
871 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
872 ret = kvm_s390_set_processor_feat(kvm, attr);
873 break;
874 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
875 ret = kvm_s390_set_processor_subfunc(kvm, attr);
876 break;
877 }
878 return ret;
879 }
880
881 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
882 {
883 struct kvm_s390_vm_cpu_processor *proc;
884 int ret = 0;
885
886 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
887 if (!proc) {
888 ret = -ENOMEM;
889 goto out;
890 }
891 proc->cpuid = kvm->arch.model.cpuid;
892 proc->ibc = kvm->arch.model.ibc;
893 memcpy(&proc->fac_list, kvm->arch.model.fac_list,
894 S390_ARCH_FAC_LIST_SIZE_BYTE);
895 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
896 ret = -EFAULT;
897 kfree(proc);
898 out:
899 return ret;
900 }
901
902 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
903 {
904 struct kvm_s390_vm_cpu_machine *mach;
905 int ret = 0;
906
907 mach = kzalloc(sizeof(*mach), GFP_KERNEL);
908 if (!mach) {
909 ret = -ENOMEM;
910 goto out;
911 }
912 get_cpu_id((struct cpuid *) &mach->cpuid);
913 mach->ibc = sclp.ibc;
914 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
915 S390_ARCH_FAC_LIST_SIZE_BYTE);
916 memcpy((unsigned long *)&mach->fac_list, S390_lowcore.stfle_fac_list,
917 S390_ARCH_FAC_LIST_SIZE_BYTE);
918 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
919 ret = -EFAULT;
920 kfree(mach);
921 out:
922 return ret;
923 }
924
925 static int kvm_s390_get_processor_feat(struct kvm *kvm,
926 struct kvm_device_attr *attr)
927 {
928 struct kvm_s390_vm_cpu_feat data;
929
930 bitmap_copy((unsigned long *) data.feat, kvm->arch.cpu_feat,
931 KVM_S390_VM_CPU_FEAT_NR_BITS);
932 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
933 return -EFAULT;
934 return 0;
935 }
936
937 static int kvm_s390_get_machine_feat(struct kvm *kvm,
938 struct kvm_device_attr *attr)
939 {
940 struct kvm_s390_vm_cpu_feat data;
941
942 bitmap_copy((unsigned long *) data.feat,
943 kvm_s390_available_cpu_feat,
944 KVM_S390_VM_CPU_FEAT_NR_BITS);
945 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
946 return -EFAULT;
947 return 0;
948 }
949
950 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
951 struct kvm_device_attr *attr)
952 {
953 /*
954 * Once we can actually configure subfunctions (kernel + hw support),
955 * we have to check if they were already set by user space, if so copy
956 * them from kvm->arch.
957 */
958 return -ENXIO;
959 }
960
961 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
962 struct kvm_device_attr *attr)
963 {
964 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
965 sizeof(struct kvm_s390_vm_cpu_subfunc)))
966 return -EFAULT;
967 return 0;
968 }
969 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
970 {
971 int ret = -ENXIO;
972
973 switch (attr->attr) {
974 case KVM_S390_VM_CPU_PROCESSOR:
975 ret = kvm_s390_get_processor(kvm, attr);
976 break;
977 case KVM_S390_VM_CPU_MACHINE:
978 ret = kvm_s390_get_machine(kvm, attr);
979 break;
980 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
981 ret = kvm_s390_get_processor_feat(kvm, attr);
982 break;
983 case KVM_S390_VM_CPU_MACHINE_FEAT:
984 ret = kvm_s390_get_machine_feat(kvm, attr);
985 break;
986 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
987 ret = kvm_s390_get_processor_subfunc(kvm, attr);
988 break;
989 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
990 ret = kvm_s390_get_machine_subfunc(kvm, attr);
991 break;
992 }
993 return ret;
994 }
995
996 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
997 {
998 int ret;
999
1000 switch (attr->group) {
1001 case KVM_S390_VM_MEM_CTRL:
1002 ret = kvm_s390_set_mem_control(kvm, attr);
1003 break;
1004 case KVM_S390_VM_TOD:
1005 ret = kvm_s390_set_tod(kvm, attr);
1006 break;
1007 case KVM_S390_VM_CPU_MODEL:
1008 ret = kvm_s390_set_cpu_model(kvm, attr);
1009 break;
1010 case KVM_S390_VM_CRYPTO:
1011 ret = kvm_s390_vm_set_crypto(kvm, attr);
1012 break;
1013 default:
1014 ret = -ENXIO;
1015 break;
1016 }
1017
1018 return ret;
1019 }
1020
1021 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1022 {
1023 int ret;
1024
1025 switch (attr->group) {
1026 case KVM_S390_VM_MEM_CTRL:
1027 ret = kvm_s390_get_mem_control(kvm, attr);
1028 break;
1029 case KVM_S390_VM_TOD:
1030 ret = kvm_s390_get_tod(kvm, attr);
1031 break;
1032 case KVM_S390_VM_CPU_MODEL:
1033 ret = kvm_s390_get_cpu_model(kvm, attr);
1034 break;
1035 default:
1036 ret = -ENXIO;
1037 break;
1038 }
1039
1040 return ret;
1041 }
1042
1043 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1044 {
1045 int ret;
1046
1047 switch (attr->group) {
1048 case KVM_S390_VM_MEM_CTRL:
1049 switch (attr->attr) {
1050 case KVM_S390_VM_MEM_ENABLE_CMMA:
1051 case KVM_S390_VM_MEM_CLR_CMMA:
1052 ret = sclp.has_cmma ? 0 : -ENXIO;
1053 break;
1054 case KVM_S390_VM_MEM_LIMIT_SIZE:
1055 ret = 0;
1056 break;
1057 default:
1058 ret = -ENXIO;
1059 break;
1060 }
1061 break;
1062 case KVM_S390_VM_TOD:
1063 switch (attr->attr) {
1064 case KVM_S390_VM_TOD_LOW:
1065 case KVM_S390_VM_TOD_HIGH:
1066 ret = 0;
1067 break;
1068 default:
1069 ret = -ENXIO;
1070 break;
1071 }
1072 break;
1073 case KVM_S390_VM_CPU_MODEL:
1074 switch (attr->attr) {
1075 case KVM_S390_VM_CPU_PROCESSOR:
1076 case KVM_S390_VM_CPU_MACHINE:
1077 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1078 case KVM_S390_VM_CPU_MACHINE_FEAT:
1079 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1080 ret = 0;
1081 break;
1082 /* configuring subfunctions is not supported yet */
1083 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1084 default:
1085 ret = -ENXIO;
1086 break;
1087 }
1088 break;
1089 case KVM_S390_VM_CRYPTO:
1090 switch (attr->attr) {
1091 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1092 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1093 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1094 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1095 ret = 0;
1096 break;
1097 default:
1098 ret = -ENXIO;
1099 break;
1100 }
1101 break;
1102 default:
1103 ret = -ENXIO;
1104 break;
1105 }
1106
1107 return ret;
1108 }
1109
1110 static long kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1111 {
1112 uint8_t *keys;
1113 uint64_t hva;
1114 int i, r = 0;
1115
1116 if (args->flags != 0)
1117 return -EINVAL;
1118
1119 /* Is this guest using storage keys? */
1120 if (!mm_use_skey(current->mm))
1121 return KVM_S390_GET_SKEYS_NONE;
1122
1123 /* Enforce sane limit on memory allocation */
1124 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1125 return -EINVAL;
1126
1127 keys = kmalloc_array(args->count, sizeof(uint8_t),
1128 GFP_KERNEL | __GFP_NOWARN);
1129 if (!keys)
1130 keys = vmalloc(sizeof(uint8_t) * args->count);
1131 if (!keys)
1132 return -ENOMEM;
1133
1134 down_read(&current->mm->mmap_sem);
1135 for (i = 0; i < args->count; i++) {
1136 hva = gfn_to_hva(kvm, args->start_gfn + i);
1137 if (kvm_is_error_hva(hva)) {
1138 r = -EFAULT;
1139 break;
1140 }
1141
1142 r = get_guest_storage_key(current->mm, hva, &keys[i]);
1143 if (r)
1144 break;
1145 }
1146 up_read(&current->mm->mmap_sem);
1147
1148 if (!r) {
1149 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
1150 sizeof(uint8_t) * args->count);
1151 if (r)
1152 r = -EFAULT;
1153 }
1154
1155 kvfree(keys);
1156 return r;
1157 }
1158
1159 static long kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1160 {
1161 uint8_t *keys;
1162 uint64_t hva;
1163 int i, r = 0;
1164
1165 if (args->flags != 0)
1166 return -EINVAL;
1167
1168 /* Enforce sane limit on memory allocation */
1169 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1170 return -EINVAL;
1171
1172 keys = kmalloc_array(args->count, sizeof(uint8_t),
1173 GFP_KERNEL | __GFP_NOWARN);
1174 if (!keys)
1175 keys = vmalloc(sizeof(uint8_t) * args->count);
1176 if (!keys)
1177 return -ENOMEM;
1178
1179 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
1180 sizeof(uint8_t) * args->count);
1181 if (r) {
1182 r = -EFAULT;
1183 goto out;
1184 }
1185
1186 /* Enable storage key handling for the guest */
1187 r = s390_enable_skey();
1188 if (r)
1189 goto out;
1190
1191 down_read(&current->mm->mmap_sem);
1192 for (i = 0; i < args->count; i++) {
1193 hva = gfn_to_hva(kvm, args->start_gfn + i);
1194 if (kvm_is_error_hva(hva)) {
1195 r = -EFAULT;
1196 break;
1197 }
1198
1199 /* Lowest order bit is reserved */
1200 if (keys[i] & 0x01) {
1201 r = -EINVAL;
1202 break;
1203 }
1204
1205 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
1206 if (r)
1207 break;
1208 }
1209 up_read(&current->mm->mmap_sem);
1210 out:
1211 kvfree(keys);
1212 return r;
1213 }
1214
1215 long kvm_arch_vm_ioctl(struct file *filp,
1216 unsigned int ioctl, unsigned long arg)
1217 {
1218 struct kvm *kvm = filp->private_data;
1219 void __user *argp = (void __user *)arg;
1220 struct kvm_device_attr attr;
1221 int r;
1222
1223 switch (ioctl) {
1224 case KVM_S390_INTERRUPT: {
1225 struct kvm_s390_interrupt s390int;
1226
1227 r = -EFAULT;
1228 if (copy_from_user(&s390int, argp, sizeof(s390int)))
1229 break;
1230 r = kvm_s390_inject_vm(kvm, &s390int);
1231 break;
1232 }
1233 case KVM_ENABLE_CAP: {
1234 struct kvm_enable_cap cap;
1235 r = -EFAULT;
1236 if (copy_from_user(&cap, argp, sizeof(cap)))
1237 break;
1238 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1239 break;
1240 }
1241 case KVM_CREATE_IRQCHIP: {
1242 struct kvm_irq_routing_entry routing;
1243
1244 r = -EINVAL;
1245 if (kvm->arch.use_irqchip) {
1246 /* Set up dummy routing. */
1247 memset(&routing, 0, sizeof(routing));
1248 r = kvm_set_irq_routing(kvm, &routing, 0, 0);
1249 }
1250 break;
1251 }
1252 case KVM_SET_DEVICE_ATTR: {
1253 r = -EFAULT;
1254 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1255 break;
1256 r = kvm_s390_vm_set_attr(kvm, &attr);
1257 break;
1258 }
1259 case KVM_GET_DEVICE_ATTR: {
1260 r = -EFAULT;
1261 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1262 break;
1263 r = kvm_s390_vm_get_attr(kvm, &attr);
1264 break;
1265 }
1266 case KVM_HAS_DEVICE_ATTR: {
1267 r = -EFAULT;
1268 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1269 break;
1270 r = kvm_s390_vm_has_attr(kvm, &attr);
1271 break;
1272 }
1273 case KVM_S390_GET_SKEYS: {
1274 struct kvm_s390_skeys args;
1275
1276 r = -EFAULT;
1277 if (copy_from_user(&args, argp,
1278 sizeof(struct kvm_s390_skeys)))
1279 break;
1280 r = kvm_s390_get_skeys(kvm, &args);
1281 break;
1282 }
1283 case KVM_S390_SET_SKEYS: {
1284 struct kvm_s390_skeys args;
1285
1286 r = -EFAULT;
1287 if (copy_from_user(&args, argp,
1288 sizeof(struct kvm_s390_skeys)))
1289 break;
1290 r = kvm_s390_set_skeys(kvm, &args);
1291 break;
1292 }
1293 default:
1294 r = -ENOTTY;
1295 }
1296
1297 return r;
1298 }
1299
1300 static int kvm_s390_query_ap_config(u8 *config)
1301 {
1302 u32 fcn_code = 0x04000000UL;
1303 u32 cc = 0;
1304
1305 memset(config, 0, 128);
1306 asm volatile(
1307 "lgr 0,%1\n"
1308 "lgr 2,%2\n"
1309 ".long 0xb2af0000\n" /* PQAP(QCI) */
1310 "0: ipm %0\n"
1311 "srl %0,28\n"
1312 "1:\n"
1313 EX_TABLE(0b, 1b)
1314 : "+r" (cc)
1315 : "r" (fcn_code), "r" (config)
1316 : "cc", "0", "2", "memory"
1317 );
1318
1319 return cc;
1320 }
1321
1322 static int kvm_s390_apxa_installed(void)
1323 {
1324 u8 config[128];
1325 int cc;
1326
1327 if (test_facility(12)) {
1328 cc = kvm_s390_query_ap_config(config);
1329
1330 if (cc)
1331 pr_err("PQAP(QCI) failed with cc=%d", cc);
1332 else
1333 return config[0] & 0x40;
1334 }
1335
1336 return 0;
1337 }
1338
1339 static void kvm_s390_set_crycb_format(struct kvm *kvm)
1340 {
1341 kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
1342
1343 if (kvm_s390_apxa_installed())
1344 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
1345 else
1346 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
1347 }
1348
1349 static u64 kvm_s390_get_initial_cpuid(void)
1350 {
1351 struct cpuid cpuid;
1352
1353 get_cpu_id(&cpuid);
1354 cpuid.version = 0xff;
1355 return *((u64 *) &cpuid);
1356 }
1357
1358 static void kvm_s390_crypto_init(struct kvm *kvm)
1359 {
1360 if (!test_kvm_facility(kvm, 76))
1361 return;
1362
1363 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
1364 kvm_s390_set_crycb_format(kvm);
1365
1366 /* Enable AES/DEA protected key functions by default */
1367 kvm->arch.crypto.aes_kw = 1;
1368 kvm->arch.crypto.dea_kw = 1;
1369 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1370 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1371 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1372 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1373 }
1374
1375 static void sca_dispose(struct kvm *kvm)
1376 {
1377 if (kvm->arch.use_esca)
1378 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
1379 else
1380 free_page((unsigned long)(kvm->arch.sca));
1381 kvm->arch.sca = NULL;
1382 }
1383
1384 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
1385 {
1386 gfp_t alloc_flags = GFP_KERNEL;
1387 int i, rc;
1388 char debug_name[16];
1389 static unsigned long sca_offset;
1390
1391 rc = -EINVAL;
1392 #ifdef CONFIG_KVM_S390_UCONTROL
1393 if (type & ~KVM_VM_S390_UCONTROL)
1394 goto out_err;
1395 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
1396 goto out_err;
1397 #else
1398 if (type)
1399 goto out_err;
1400 #endif
1401
1402 rc = s390_enable_sie();
1403 if (rc)
1404 goto out_err;
1405
1406 rc = -ENOMEM;
1407
1408 ratelimit_state_init(&kvm->arch.sthyi_limit, 5 * HZ, 500);
1409
1410 kvm->arch.use_esca = 0; /* start with basic SCA */
1411 if (!sclp.has_64bscao)
1412 alloc_flags |= GFP_DMA;
1413 rwlock_init(&kvm->arch.sca_lock);
1414 kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
1415 if (!kvm->arch.sca)
1416 goto out_err;
1417 spin_lock(&kvm_lock);
1418 sca_offset += 16;
1419 if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
1420 sca_offset = 0;
1421 kvm->arch.sca = (struct bsca_block *)
1422 ((char *) kvm->arch.sca + sca_offset);
1423 spin_unlock(&kvm_lock);
1424
1425 sprintf(debug_name, "kvm-%u", current->pid);
1426
1427 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
1428 if (!kvm->arch.dbf)
1429 goto out_err;
1430
1431 kvm->arch.sie_page2 =
1432 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1433 if (!kvm->arch.sie_page2)
1434 goto out_err;
1435
1436 /* Populate the facility mask initially. */
1437 memcpy(kvm->arch.model.fac_mask, S390_lowcore.stfle_fac_list,
1438 S390_ARCH_FAC_LIST_SIZE_BYTE);
1439 for (i = 0; i < S390_ARCH_FAC_LIST_SIZE_U64; i++) {
1440 if (i < kvm_s390_fac_list_mask_size())
1441 kvm->arch.model.fac_mask[i] &= kvm_s390_fac_list_mask[i];
1442 else
1443 kvm->arch.model.fac_mask[i] = 0UL;
1444 }
1445
1446 /* Populate the facility list initially. */
1447 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
1448 memcpy(kvm->arch.model.fac_list, kvm->arch.model.fac_mask,
1449 S390_ARCH_FAC_LIST_SIZE_BYTE);
1450
1451 set_kvm_facility(kvm->arch.model.fac_mask, 74);
1452 set_kvm_facility(kvm->arch.model.fac_list, 74);
1453
1454 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
1455 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
1456
1457 kvm_s390_crypto_init(kvm);
1458
1459 spin_lock_init(&kvm->arch.float_int.lock);
1460 for (i = 0; i < FIRQ_LIST_COUNT; i++)
1461 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
1462 init_waitqueue_head(&kvm->arch.ipte_wq);
1463 mutex_init(&kvm->arch.ipte_mutex);
1464
1465 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
1466 VM_EVENT(kvm, 3, "vm created with type %lu", type);
1467
1468 if (type & KVM_VM_S390_UCONTROL) {
1469 kvm->arch.gmap = NULL;
1470 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
1471 } else {
1472 if (sclp.hamax == U64_MAX)
1473 kvm->arch.mem_limit = TASK_MAX_SIZE;
1474 else
1475 kvm->arch.mem_limit = min_t(unsigned long, TASK_MAX_SIZE,
1476 sclp.hamax + 1);
1477 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
1478 if (!kvm->arch.gmap)
1479 goto out_err;
1480 kvm->arch.gmap->private = kvm;
1481 kvm->arch.gmap->pfault_enabled = 0;
1482 }
1483
1484 kvm->arch.css_support = 0;
1485 kvm->arch.use_irqchip = 0;
1486 kvm->arch.epoch = 0;
1487
1488 spin_lock_init(&kvm->arch.start_stop_lock);
1489 kvm_s390_vsie_init(kvm);
1490 KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
1491
1492 return 0;
1493 out_err:
1494 free_page((unsigned long)kvm->arch.sie_page2);
1495 debug_unregister(kvm->arch.dbf);
1496 sca_dispose(kvm);
1497 KVM_EVENT(3, "creation of vm failed: %d", rc);
1498 return rc;
1499 }
1500
1501 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1502 {
1503 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
1504 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
1505 kvm_s390_clear_local_irqs(vcpu);
1506 kvm_clear_async_pf_completion_queue(vcpu);
1507 if (!kvm_is_ucontrol(vcpu->kvm))
1508 sca_del_vcpu(vcpu);
1509
1510 if (kvm_is_ucontrol(vcpu->kvm))
1511 gmap_remove(vcpu->arch.gmap);
1512
1513 if (vcpu->kvm->arch.use_cmma)
1514 kvm_s390_vcpu_unsetup_cmma(vcpu);
1515 free_page((unsigned long)(vcpu->arch.sie_block));
1516
1517 kvm_vcpu_uninit(vcpu);
1518 kmem_cache_free(kvm_vcpu_cache, vcpu);
1519 }
1520
1521 static void kvm_free_vcpus(struct kvm *kvm)
1522 {
1523 unsigned int i;
1524 struct kvm_vcpu *vcpu;
1525
1526 kvm_for_each_vcpu(i, vcpu, kvm)
1527 kvm_arch_vcpu_destroy(vcpu);
1528
1529 mutex_lock(&kvm->lock);
1530 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
1531 kvm->vcpus[i] = NULL;
1532
1533 atomic_set(&kvm->online_vcpus, 0);
1534 mutex_unlock(&kvm->lock);
1535 }
1536
1537 void kvm_arch_destroy_vm(struct kvm *kvm)
1538 {
1539 kvm_free_vcpus(kvm);
1540 sca_dispose(kvm);
1541 debug_unregister(kvm->arch.dbf);
1542 free_page((unsigned long)kvm->arch.sie_page2);
1543 if (!kvm_is_ucontrol(kvm))
1544 gmap_remove(kvm->arch.gmap);
1545 kvm_s390_destroy_adapters(kvm);
1546 kvm_s390_clear_float_irqs(kvm);
1547 kvm_s390_vsie_destroy(kvm);
1548 KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
1549 }
1550
1551 /* Section: vcpu related */
1552 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
1553 {
1554 vcpu->arch.gmap = gmap_create(current->mm, -1UL);
1555 if (!vcpu->arch.gmap)
1556 return -ENOMEM;
1557 vcpu->arch.gmap->private = vcpu->kvm;
1558
1559 return 0;
1560 }
1561
1562 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
1563 {
1564 read_lock(&vcpu->kvm->arch.sca_lock);
1565 if (vcpu->kvm->arch.use_esca) {
1566 struct esca_block *sca = vcpu->kvm->arch.sca;
1567
1568 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
1569 sca->cpu[vcpu->vcpu_id].sda = 0;
1570 } else {
1571 struct bsca_block *sca = vcpu->kvm->arch.sca;
1572
1573 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
1574 sca->cpu[vcpu->vcpu_id].sda = 0;
1575 }
1576 read_unlock(&vcpu->kvm->arch.sca_lock);
1577 }
1578
1579 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
1580 {
1581 read_lock(&vcpu->kvm->arch.sca_lock);
1582 if (vcpu->kvm->arch.use_esca) {
1583 struct esca_block *sca = vcpu->kvm->arch.sca;
1584
1585 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
1586 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1587 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca & ~0x3fU;
1588 vcpu->arch.sie_block->ecb2 |= 0x04U;
1589 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
1590 } else {
1591 struct bsca_block *sca = vcpu->kvm->arch.sca;
1592
1593 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
1594 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1595 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
1596 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
1597 }
1598 read_unlock(&vcpu->kvm->arch.sca_lock);
1599 }
1600
1601 /* Basic SCA to Extended SCA data copy routines */
1602 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
1603 {
1604 d->sda = s->sda;
1605 d->sigp_ctrl.c = s->sigp_ctrl.c;
1606 d->sigp_ctrl.scn = s->sigp_ctrl.scn;
1607 }
1608
1609 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
1610 {
1611 int i;
1612
1613 d->ipte_control = s->ipte_control;
1614 d->mcn[0] = s->mcn;
1615 for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
1616 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
1617 }
1618
1619 static int sca_switch_to_extended(struct kvm *kvm)
1620 {
1621 struct bsca_block *old_sca = kvm->arch.sca;
1622 struct esca_block *new_sca;
1623 struct kvm_vcpu *vcpu;
1624 unsigned int vcpu_idx;
1625 u32 scaol, scaoh;
1626
1627 new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL|__GFP_ZERO);
1628 if (!new_sca)
1629 return -ENOMEM;
1630
1631 scaoh = (u32)((u64)(new_sca) >> 32);
1632 scaol = (u32)(u64)(new_sca) & ~0x3fU;
1633
1634 kvm_s390_vcpu_block_all(kvm);
1635 write_lock(&kvm->arch.sca_lock);
1636
1637 sca_copy_b_to_e(new_sca, old_sca);
1638
1639 kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
1640 vcpu->arch.sie_block->scaoh = scaoh;
1641 vcpu->arch.sie_block->scaol = scaol;
1642 vcpu->arch.sie_block->ecb2 |= 0x04U;
1643 }
1644 kvm->arch.sca = new_sca;
1645 kvm->arch.use_esca = 1;
1646
1647 write_unlock(&kvm->arch.sca_lock);
1648 kvm_s390_vcpu_unblock_all(kvm);
1649
1650 free_page((unsigned long)old_sca);
1651
1652 VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
1653 old_sca, kvm->arch.sca);
1654 return 0;
1655 }
1656
1657 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
1658 {
1659 int rc;
1660
1661 if (id < KVM_S390_BSCA_CPU_SLOTS)
1662 return true;
1663 if (!sclp.has_esca || !sclp.has_64bscao)
1664 return false;
1665
1666 mutex_lock(&kvm->lock);
1667 rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
1668 mutex_unlock(&kvm->lock);
1669
1670 return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
1671 }
1672
1673 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1674 {
1675 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1676 kvm_clear_async_pf_completion_queue(vcpu);
1677 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
1678 KVM_SYNC_GPRS |
1679 KVM_SYNC_ACRS |
1680 KVM_SYNC_CRS |
1681 KVM_SYNC_ARCH0 |
1682 KVM_SYNC_PFAULT;
1683 kvm_s390_set_prefix(vcpu, 0);
1684 if (test_kvm_facility(vcpu->kvm, 64))
1685 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
1686 /* fprs can be synchronized via vrs, even if the guest has no vx. With
1687 * MACHINE_HAS_VX, (load|store)_fpu_regs() will work with vrs format.
1688 */
1689 if (MACHINE_HAS_VX)
1690 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
1691 else
1692 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
1693
1694 if (kvm_is_ucontrol(vcpu->kvm))
1695 return __kvm_ucontrol_vcpu_init(vcpu);
1696
1697 return 0;
1698 }
1699
1700 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1701 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1702 {
1703 WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
1704 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1705 vcpu->arch.cputm_start = get_tod_clock_fast();
1706 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1707 }
1708
1709 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1710 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1711 {
1712 WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
1713 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1714 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
1715 vcpu->arch.cputm_start = 0;
1716 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1717 }
1718
1719 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1720 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1721 {
1722 WARN_ON_ONCE(vcpu->arch.cputm_enabled);
1723 vcpu->arch.cputm_enabled = true;
1724 __start_cpu_timer_accounting(vcpu);
1725 }
1726
1727 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1728 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1729 {
1730 WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
1731 __stop_cpu_timer_accounting(vcpu);
1732 vcpu->arch.cputm_enabled = false;
1733 }
1734
1735 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1736 {
1737 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1738 __enable_cpu_timer_accounting(vcpu);
1739 preempt_enable();
1740 }
1741
1742 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1743 {
1744 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1745 __disable_cpu_timer_accounting(vcpu);
1746 preempt_enable();
1747 }
1748
1749 /* set the cpu timer - may only be called from the VCPU thread itself */
1750 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
1751 {
1752 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1753 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1754 if (vcpu->arch.cputm_enabled)
1755 vcpu->arch.cputm_start = get_tod_clock_fast();
1756 vcpu->arch.sie_block->cputm = cputm;
1757 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1758 preempt_enable();
1759 }
1760
1761 /* update and get the cpu timer - can also be called from other VCPU threads */
1762 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
1763 {
1764 unsigned int seq;
1765 __u64 value;
1766
1767 if (unlikely(!vcpu->arch.cputm_enabled))
1768 return vcpu->arch.sie_block->cputm;
1769
1770 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1771 do {
1772 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
1773 /*
1774 * If the writer would ever execute a read in the critical
1775 * section, e.g. in irq context, we have a deadlock.
1776 */
1777 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
1778 value = vcpu->arch.sie_block->cputm;
1779 /* if cputm_start is 0, accounting is being started/stopped */
1780 if (likely(vcpu->arch.cputm_start))
1781 value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
1782 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
1783 preempt_enable();
1784 return value;
1785 }
1786
1787 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1788 {
1789 /* Save host register state */
1790 save_fpu_regs();
1791 vcpu->arch.host_fpregs.fpc = current->thread.fpu.fpc;
1792 vcpu->arch.host_fpregs.regs = current->thread.fpu.regs;
1793
1794 if (MACHINE_HAS_VX)
1795 current->thread.fpu.regs = vcpu->run->s.regs.vrs;
1796 else
1797 current->thread.fpu.regs = vcpu->run->s.regs.fprs;
1798 current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
1799 if (test_fp_ctl(current->thread.fpu.fpc))
1800 /* User space provided an invalid FPC, let's clear it */
1801 current->thread.fpu.fpc = 0;
1802
1803 save_access_regs(vcpu->arch.host_acrs);
1804 restore_access_regs(vcpu->run->s.regs.acrs);
1805 gmap_enable(vcpu->arch.enabled_gmap);
1806 atomic_or(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1807 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
1808 __start_cpu_timer_accounting(vcpu);
1809 vcpu->cpu = cpu;
1810 }
1811
1812 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1813 {
1814 vcpu->cpu = -1;
1815 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
1816 __stop_cpu_timer_accounting(vcpu);
1817 atomic_andnot(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1818 vcpu->arch.enabled_gmap = gmap_get_enabled();
1819 gmap_disable(vcpu->arch.enabled_gmap);
1820
1821 /* Save guest register state */
1822 save_fpu_regs();
1823 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
1824
1825 /* Restore host register state */
1826 current->thread.fpu.fpc = vcpu->arch.host_fpregs.fpc;
1827 current->thread.fpu.regs = vcpu->arch.host_fpregs.regs;
1828
1829 save_access_regs(vcpu->run->s.regs.acrs);
1830 restore_access_regs(vcpu->arch.host_acrs);
1831 }
1832
1833 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
1834 {
1835 /* this equals initial cpu reset in pop, but we don't switch to ESA */
1836 vcpu->arch.sie_block->gpsw.mask = 0UL;
1837 vcpu->arch.sie_block->gpsw.addr = 0UL;
1838 kvm_s390_set_prefix(vcpu, 0);
1839 kvm_s390_set_cpu_timer(vcpu, 0);
1840 vcpu->arch.sie_block->ckc = 0UL;
1841 vcpu->arch.sie_block->todpr = 0;
1842 memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64));
1843 vcpu->arch.sie_block->gcr[0] = 0xE0UL;
1844 vcpu->arch.sie_block->gcr[14] = 0xC2000000UL;
1845 /* make sure the new fpc will be lazily loaded */
1846 save_fpu_regs();
1847 current->thread.fpu.fpc = 0;
1848 vcpu->arch.sie_block->gbea = 1;
1849 vcpu->arch.sie_block->pp = 0;
1850 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1851 kvm_clear_async_pf_completion_queue(vcpu);
1852 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
1853 kvm_s390_vcpu_stop(vcpu);
1854 kvm_s390_clear_local_irqs(vcpu);
1855 }
1856
1857 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1858 {
1859 mutex_lock(&vcpu->kvm->lock);
1860 preempt_disable();
1861 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
1862 preempt_enable();
1863 mutex_unlock(&vcpu->kvm->lock);
1864 if (!kvm_is_ucontrol(vcpu->kvm)) {
1865 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
1866 sca_add_vcpu(vcpu);
1867 }
1868 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
1869 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
1870 /* make vcpu_load load the right gmap on the first trigger */
1871 vcpu->arch.enabled_gmap = vcpu->arch.gmap;
1872 }
1873
1874 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
1875 {
1876 if (!test_kvm_facility(vcpu->kvm, 76))
1877 return;
1878
1879 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
1880
1881 if (vcpu->kvm->arch.crypto.aes_kw)
1882 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
1883 if (vcpu->kvm->arch.crypto.dea_kw)
1884 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
1885
1886 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
1887 }
1888
1889 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
1890 {
1891 free_page(vcpu->arch.sie_block->cbrlo);
1892 vcpu->arch.sie_block->cbrlo = 0;
1893 }
1894
1895 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
1896 {
1897 vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
1898 if (!vcpu->arch.sie_block->cbrlo)
1899 return -ENOMEM;
1900
1901 vcpu->arch.sie_block->ecb2 |= 0x80;
1902 vcpu->arch.sie_block->ecb2 &= ~0x08;
1903 return 0;
1904 }
1905
1906 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
1907 {
1908 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
1909
1910 vcpu->arch.sie_block->ibc = model->ibc;
1911 if (test_kvm_facility(vcpu->kvm, 7))
1912 vcpu->arch.sie_block->fac = (u32)(u64) model->fac_list;
1913 }
1914
1915 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1916 {
1917 int rc = 0;
1918
1919 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
1920 CPUSTAT_SM |
1921 CPUSTAT_STOPPED);
1922
1923 if (test_kvm_facility(vcpu->kvm, 78))
1924 atomic_or(CPUSTAT_GED2, &vcpu->arch.sie_block->cpuflags);
1925 else if (test_kvm_facility(vcpu->kvm, 8))
1926 atomic_or(CPUSTAT_GED, &vcpu->arch.sie_block->cpuflags);
1927
1928 kvm_s390_vcpu_setup_model(vcpu);
1929
1930 /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
1931 if (MACHINE_HAS_ESOP)
1932 vcpu->arch.sie_block->ecb |= 0x02;
1933 if (test_kvm_facility(vcpu->kvm, 9))
1934 vcpu->arch.sie_block->ecb |= 0x04;
1935 if (test_kvm_facility(vcpu->kvm, 73))
1936 vcpu->arch.sie_block->ecb |= 0x10;
1937
1938 if (test_kvm_facility(vcpu->kvm, 8) && sclp.has_pfmfi)
1939 vcpu->arch.sie_block->ecb2 |= 0x08;
1940 vcpu->arch.sie_block->eca = 0x1002000U;
1941 if (sclp.has_cei)
1942 vcpu->arch.sie_block->eca |= 0x80000000U;
1943 if (sclp.has_ib)
1944 vcpu->arch.sie_block->eca |= 0x40000000U;
1945 if (sclp.has_siif)
1946 vcpu->arch.sie_block->eca |= 1;
1947 if (sclp.has_sigpif)
1948 vcpu->arch.sie_block->eca |= 0x10000000U;
1949 if (test_kvm_facility(vcpu->kvm, 64))
1950 vcpu->arch.sie_block->ecb3 |= 0x01;
1951 if (test_kvm_facility(vcpu->kvm, 129)) {
1952 vcpu->arch.sie_block->eca |= 0x00020000;
1953 vcpu->arch.sie_block->ecd |= 0x20000000;
1954 }
1955 vcpu->arch.sie_block->riccbd = (unsigned long) &vcpu->run->s.regs.riccb;
1956 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
1957
1958 if (vcpu->kvm->arch.use_cmma) {
1959 rc = kvm_s390_vcpu_setup_cmma(vcpu);
1960 if (rc)
1961 return rc;
1962 }
1963 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1964 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
1965
1966 kvm_s390_vcpu_crypto_setup(vcpu);
1967
1968 return rc;
1969 }
1970
1971 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1972 unsigned int id)
1973 {
1974 struct kvm_vcpu *vcpu;
1975 struct sie_page *sie_page;
1976 int rc = -EINVAL;
1977
1978 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
1979 goto out;
1980
1981 rc = -ENOMEM;
1982
1983 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1984 if (!vcpu)
1985 goto out;
1986
1987 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
1988 if (!sie_page)
1989 goto out_free_cpu;
1990
1991 vcpu->arch.sie_block = &sie_page->sie_block;
1992 vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
1993
1994 /* the real guest size will always be smaller than msl */
1995 vcpu->arch.sie_block->mso = 0;
1996 vcpu->arch.sie_block->msl = sclp.hamax;
1997
1998 vcpu->arch.sie_block->icpua = id;
1999 spin_lock_init(&vcpu->arch.local_int.lock);
2000 vcpu->arch.local_int.float_int = &kvm->arch.float_int;
2001 vcpu->arch.local_int.wq = &vcpu->wq;
2002 vcpu->arch.local_int.cpuflags = &vcpu->arch.sie_block->cpuflags;
2003 seqcount_init(&vcpu->arch.cputm_seqcount);
2004
2005 rc = kvm_vcpu_init(vcpu, kvm, id);
2006 if (rc)
2007 goto out_free_sie_block;
2008 VM_EVENT(kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK", id, vcpu,
2009 vcpu->arch.sie_block);
2010 trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
2011
2012 return vcpu;
2013 out_free_sie_block:
2014 free_page((unsigned long)(vcpu->arch.sie_block));
2015 out_free_cpu:
2016 kmem_cache_free(kvm_vcpu_cache, vcpu);
2017 out:
2018 return ERR_PTR(rc);
2019 }
2020
2021 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
2022 {
2023 return kvm_s390_vcpu_has_irq(vcpu, 0);
2024 }
2025
2026 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
2027 {
2028 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2029 exit_sie(vcpu);
2030 }
2031
2032 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
2033 {
2034 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2035 }
2036
2037 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
2038 {
2039 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2040 exit_sie(vcpu);
2041 }
2042
2043 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
2044 {
2045 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2046 }
2047
2048 /*
2049 * Kick a guest cpu out of SIE and wait until SIE is not running.
2050 * If the CPU is not running (e.g. waiting as idle) the function will
2051 * return immediately. */
2052 void exit_sie(struct kvm_vcpu *vcpu)
2053 {
2054 atomic_or(CPUSTAT_STOP_INT, &vcpu->arch.sie_block->cpuflags);
2055 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
2056 cpu_relax();
2057 }
2058
2059 /* Kick a guest cpu out of SIE to process a request synchronously */
2060 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
2061 {
2062 kvm_make_request(req, vcpu);
2063 kvm_s390_vcpu_request(vcpu);
2064 }
2065
2066 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
2067 unsigned long end)
2068 {
2069 struct kvm *kvm = gmap->private;
2070 struct kvm_vcpu *vcpu;
2071 unsigned long prefix;
2072 int i;
2073
2074 if (gmap_is_shadow(gmap))
2075 return;
2076 if (start >= 1UL << 31)
2077 /* We are only interested in prefix pages */
2078 return;
2079 kvm_for_each_vcpu(i, vcpu, kvm) {
2080 /* match against both prefix pages */
2081 prefix = kvm_s390_get_prefix(vcpu);
2082 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
2083 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
2084 start, end);
2085 kvm_s390_sync_request(KVM_REQ_MMU_RELOAD, vcpu);
2086 }
2087 }
2088 }
2089
2090 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
2091 {
2092 /* kvm common code refers to this, but never calls it */
2093 BUG();
2094 return 0;
2095 }
2096
2097 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
2098 struct kvm_one_reg *reg)
2099 {
2100 int r = -EINVAL;
2101
2102 switch (reg->id) {
2103 case KVM_REG_S390_TODPR:
2104 r = put_user(vcpu->arch.sie_block->todpr,
2105 (u32 __user *)reg->addr);
2106 break;
2107 case KVM_REG_S390_EPOCHDIFF:
2108 r = put_user(vcpu->arch.sie_block->epoch,
2109 (u64 __user *)reg->addr);
2110 break;
2111 case KVM_REG_S390_CPU_TIMER:
2112 r = put_user(kvm_s390_get_cpu_timer(vcpu),
2113 (u64 __user *)reg->addr);
2114 break;
2115 case KVM_REG_S390_CLOCK_COMP:
2116 r = put_user(vcpu->arch.sie_block->ckc,
2117 (u64 __user *)reg->addr);
2118 break;
2119 case KVM_REG_S390_PFTOKEN:
2120 r = put_user(vcpu->arch.pfault_token,
2121 (u64 __user *)reg->addr);
2122 break;
2123 case KVM_REG_S390_PFCOMPARE:
2124 r = put_user(vcpu->arch.pfault_compare,
2125 (u64 __user *)reg->addr);
2126 break;
2127 case KVM_REG_S390_PFSELECT:
2128 r = put_user(vcpu->arch.pfault_select,
2129 (u64 __user *)reg->addr);
2130 break;
2131 case KVM_REG_S390_PP:
2132 r = put_user(vcpu->arch.sie_block->pp,
2133 (u64 __user *)reg->addr);
2134 break;
2135 case KVM_REG_S390_GBEA:
2136 r = put_user(vcpu->arch.sie_block->gbea,
2137 (u64 __user *)reg->addr);
2138 break;
2139 default:
2140 break;
2141 }
2142
2143 return r;
2144 }
2145
2146 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
2147 struct kvm_one_reg *reg)
2148 {
2149 int r = -EINVAL;
2150 __u64 val;
2151
2152 switch (reg->id) {
2153 case KVM_REG_S390_TODPR:
2154 r = get_user(vcpu->arch.sie_block->todpr,
2155 (u32 __user *)reg->addr);
2156 break;
2157 case KVM_REG_S390_EPOCHDIFF:
2158 r = get_user(vcpu->arch.sie_block->epoch,
2159 (u64 __user *)reg->addr);
2160 break;
2161 case KVM_REG_S390_CPU_TIMER:
2162 r = get_user(val, (u64 __user *)reg->addr);
2163 if (!r)
2164 kvm_s390_set_cpu_timer(vcpu, val);
2165 break;
2166 case KVM_REG_S390_CLOCK_COMP:
2167 r = get_user(vcpu->arch.sie_block->ckc,
2168 (u64 __user *)reg->addr);
2169 break;
2170 case KVM_REG_S390_PFTOKEN:
2171 r = get_user(vcpu->arch.pfault_token,
2172 (u64 __user *)reg->addr);
2173 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2174 kvm_clear_async_pf_completion_queue(vcpu);
2175 break;
2176 case KVM_REG_S390_PFCOMPARE:
2177 r = get_user(vcpu->arch.pfault_compare,
2178 (u64 __user *)reg->addr);
2179 break;
2180 case KVM_REG_S390_PFSELECT:
2181 r = get_user(vcpu->arch.pfault_select,
2182 (u64 __user *)reg->addr);
2183 break;
2184 case KVM_REG_S390_PP:
2185 r = get_user(vcpu->arch.sie_block->pp,
2186 (u64 __user *)reg->addr);
2187 break;
2188 case KVM_REG_S390_GBEA:
2189 r = get_user(vcpu->arch.sie_block->gbea,
2190 (u64 __user *)reg->addr);
2191 break;
2192 default:
2193 break;
2194 }
2195
2196 return r;
2197 }
2198
2199 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
2200 {
2201 kvm_s390_vcpu_initial_reset(vcpu);
2202 return 0;
2203 }
2204
2205 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2206 {
2207 memcpy(&vcpu->run->s.regs.gprs, &regs->gprs, sizeof(regs->gprs));
2208 return 0;
2209 }
2210
2211 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2212 {
2213 memcpy(&regs->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
2214 return 0;
2215 }
2216
2217 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2218 struct kvm_sregs *sregs)
2219 {
2220 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
2221 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
2222 restore_access_regs(vcpu->run->s.regs.acrs);
2223 return 0;
2224 }
2225
2226 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2227 struct kvm_sregs *sregs)
2228 {
2229 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
2230 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
2231 return 0;
2232 }
2233
2234 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2235 {
2236 /* make sure the new values will be lazily loaded */
2237 save_fpu_regs();
2238 if (test_fp_ctl(fpu->fpc))
2239 return -EINVAL;
2240 current->thread.fpu.fpc = fpu->fpc;
2241 if (MACHINE_HAS_VX)
2242 convert_fp_to_vx(current->thread.fpu.vxrs, (freg_t *)fpu->fprs);
2243 else
2244 memcpy(current->thread.fpu.fprs, &fpu->fprs, sizeof(fpu->fprs));
2245 return 0;
2246 }
2247
2248 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2249 {
2250 /* make sure we have the latest values */
2251 save_fpu_regs();
2252 if (MACHINE_HAS_VX)
2253 convert_vx_to_fp((freg_t *)fpu->fprs, current->thread.fpu.vxrs);
2254 else
2255 memcpy(fpu->fprs, current->thread.fpu.fprs, sizeof(fpu->fprs));
2256 fpu->fpc = current->thread.fpu.fpc;
2257 return 0;
2258 }
2259
2260 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
2261 {
2262 int rc = 0;
2263
2264 if (!is_vcpu_stopped(vcpu))
2265 rc = -EBUSY;
2266 else {
2267 vcpu->run->psw_mask = psw.mask;
2268 vcpu->run->psw_addr = psw.addr;
2269 }
2270 return rc;
2271 }
2272
2273 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2274 struct kvm_translation *tr)
2275 {
2276 return -EINVAL; /* not implemented yet */
2277 }
2278
2279 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
2280 KVM_GUESTDBG_USE_HW_BP | \
2281 KVM_GUESTDBG_ENABLE)
2282
2283 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
2284 struct kvm_guest_debug *dbg)
2285 {
2286 int rc = 0;
2287
2288 vcpu->guest_debug = 0;
2289 kvm_s390_clear_bp_data(vcpu);
2290
2291 if (dbg->control & ~VALID_GUESTDBG_FLAGS)
2292 return -EINVAL;
2293 if (!sclp.has_gpere)
2294 return -EINVAL;
2295
2296 if (dbg->control & KVM_GUESTDBG_ENABLE) {
2297 vcpu->guest_debug = dbg->control;
2298 /* enforce guest PER */
2299 atomic_or(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2300
2301 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
2302 rc = kvm_s390_import_bp_data(vcpu, dbg);
2303 } else {
2304 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2305 vcpu->arch.guestdbg.last_bp = 0;
2306 }
2307
2308 if (rc) {
2309 vcpu->guest_debug = 0;
2310 kvm_s390_clear_bp_data(vcpu);
2311 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2312 }
2313
2314 return rc;
2315 }
2316
2317 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2318 struct kvm_mp_state *mp_state)
2319 {
2320 /* CHECK_STOP and LOAD are not supported yet */
2321 return is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
2322 KVM_MP_STATE_OPERATING;
2323 }
2324
2325 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2326 struct kvm_mp_state *mp_state)
2327 {
2328 int rc = 0;
2329
2330 /* user space knows about this interface - let it control the state */
2331 vcpu->kvm->arch.user_cpu_state_ctrl = 1;
2332
2333 switch (mp_state->mp_state) {
2334 case KVM_MP_STATE_STOPPED:
2335 kvm_s390_vcpu_stop(vcpu);
2336 break;
2337 case KVM_MP_STATE_OPERATING:
2338 kvm_s390_vcpu_start(vcpu);
2339 break;
2340 case KVM_MP_STATE_LOAD:
2341 case KVM_MP_STATE_CHECK_STOP:
2342 /* fall through - CHECK_STOP and LOAD are not supported yet */
2343 default:
2344 rc = -ENXIO;
2345 }
2346
2347 return rc;
2348 }
2349
2350 static bool ibs_enabled(struct kvm_vcpu *vcpu)
2351 {
2352 return atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_IBS;
2353 }
2354
2355 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
2356 {
2357 retry:
2358 kvm_s390_vcpu_request_handled(vcpu);
2359 if (!vcpu->requests)
2360 return 0;
2361 /*
2362 * We use MMU_RELOAD just to re-arm the ipte notifier for the
2363 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
2364 * This ensures that the ipte instruction for this request has
2365 * already finished. We might race against a second unmapper that
2366 * wants to set the blocking bit. Lets just retry the request loop.
2367 */
2368 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
2369 int rc;
2370 rc = gmap_mprotect_notify(vcpu->arch.gmap,
2371 kvm_s390_get_prefix(vcpu),
2372 PAGE_SIZE * 2, PROT_WRITE);
2373 if (rc) {
2374 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
2375 return rc;
2376 }
2377 goto retry;
2378 }
2379
2380 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2381 vcpu->arch.sie_block->ihcpu = 0xffff;
2382 goto retry;
2383 }
2384
2385 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
2386 if (!ibs_enabled(vcpu)) {
2387 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
2388 atomic_or(CPUSTAT_IBS,
2389 &vcpu->arch.sie_block->cpuflags);
2390 }
2391 goto retry;
2392 }
2393
2394 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
2395 if (ibs_enabled(vcpu)) {
2396 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
2397 atomic_andnot(CPUSTAT_IBS,
2398 &vcpu->arch.sie_block->cpuflags);
2399 }
2400 goto retry;
2401 }
2402
2403 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
2404 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
2405 goto retry;
2406 }
2407
2408 /* nothing to do, just clear the request */
2409 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
2410
2411 return 0;
2412 }
2413
2414 void kvm_s390_set_tod_clock(struct kvm *kvm, u64 tod)
2415 {
2416 struct kvm_vcpu *vcpu;
2417 int i;
2418
2419 mutex_lock(&kvm->lock);
2420 preempt_disable();
2421 kvm->arch.epoch = tod - get_tod_clock();
2422 kvm_s390_vcpu_block_all(kvm);
2423 kvm_for_each_vcpu(i, vcpu, kvm)
2424 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
2425 kvm_s390_vcpu_unblock_all(kvm);
2426 preempt_enable();
2427 mutex_unlock(&kvm->lock);
2428 }
2429
2430 /**
2431 * kvm_arch_fault_in_page - fault-in guest page if necessary
2432 * @vcpu: The corresponding virtual cpu
2433 * @gpa: Guest physical address
2434 * @writable: Whether the page should be writable or not
2435 *
2436 * Make sure that a guest page has been faulted-in on the host.
2437 *
2438 * Return: Zero on success, negative error code otherwise.
2439 */
2440 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
2441 {
2442 return gmap_fault(vcpu->arch.gmap, gpa,
2443 writable ? FAULT_FLAG_WRITE : 0);
2444 }
2445
2446 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
2447 unsigned long token)
2448 {
2449 struct kvm_s390_interrupt inti;
2450 struct kvm_s390_irq irq;
2451
2452 if (start_token) {
2453 irq.u.ext.ext_params2 = token;
2454 irq.type = KVM_S390_INT_PFAULT_INIT;
2455 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
2456 } else {
2457 inti.type = KVM_S390_INT_PFAULT_DONE;
2458 inti.parm64 = token;
2459 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
2460 }
2461 }
2462
2463 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
2464 struct kvm_async_pf *work)
2465 {
2466 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
2467 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
2468 }
2469
2470 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
2471 struct kvm_async_pf *work)
2472 {
2473 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
2474 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
2475 }
2476
2477 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
2478 struct kvm_async_pf *work)
2479 {
2480 /* s390 will always inject the page directly */
2481 }
2482
2483 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
2484 {
2485 /*
2486 * s390 will always inject the page directly,
2487 * but we still want check_async_completion to cleanup
2488 */
2489 return true;
2490 }
2491
2492 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
2493 {
2494 hva_t hva;
2495 struct kvm_arch_async_pf arch;
2496 int rc;
2497
2498 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2499 return 0;
2500 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
2501 vcpu->arch.pfault_compare)
2502 return 0;
2503 if (psw_extint_disabled(vcpu))
2504 return 0;
2505 if (kvm_s390_vcpu_has_irq(vcpu, 0))
2506 return 0;
2507 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
2508 return 0;
2509 if (!vcpu->arch.gmap->pfault_enabled)
2510 return 0;
2511
2512 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
2513 hva += current->thread.gmap_addr & ~PAGE_MASK;
2514 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
2515 return 0;
2516
2517 rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
2518 return rc;
2519 }
2520
2521 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
2522 {
2523 int rc, cpuflags;
2524
2525 /*
2526 * On s390 notifications for arriving pages will be delivered directly
2527 * to the guest but the house keeping for completed pfaults is
2528 * handled outside the worker.
2529 */
2530 kvm_check_async_pf_completion(vcpu);
2531
2532 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
2533 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
2534
2535 if (need_resched())
2536 schedule();
2537
2538 if (test_cpu_flag(CIF_MCCK_PENDING))
2539 s390_handle_mcck();
2540
2541 if (!kvm_is_ucontrol(vcpu->kvm)) {
2542 rc = kvm_s390_deliver_pending_interrupts(vcpu);
2543 if (rc)
2544 return rc;
2545 }
2546
2547 rc = kvm_s390_handle_requests(vcpu);
2548 if (rc)
2549 return rc;
2550
2551 if (guestdbg_enabled(vcpu)) {
2552 kvm_s390_backup_guest_per_regs(vcpu);
2553 kvm_s390_patch_guest_per_regs(vcpu);
2554 }
2555
2556 vcpu->arch.sie_block->icptcode = 0;
2557 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
2558 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
2559 trace_kvm_s390_sie_enter(vcpu, cpuflags);
2560
2561 return 0;
2562 }
2563
2564 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
2565 {
2566 struct kvm_s390_pgm_info pgm_info = {
2567 .code = PGM_ADDRESSING,
2568 };
2569 u8 opcode, ilen;
2570 int rc;
2571
2572 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
2573 trace_kvm_s390_sie_fault(vcpu);
2574
2575 /*
2576 * We want to inject an addressing exception, which is defined as a
2577 * suppressing or terminating exception. However, since we came here
2578 * by a DAT access exception, the PSW still points to the faulting
2579 * instruction since DAT exceptions are nullifying. So we've got
2580 * to look up the current opcode to get the length of the instruction
2581 * to be able to forward the PSW.
2582 */
2583 rc = read_guest_instr(vcpu, &opcode, 1);
2584 ilen = insn_length(opcode);
2585 if (rc < 0) {
2586 return rc;
2587 } else if (rc) {
2588 /* Instruction-Fetching Exceptions - we can't detect the ilen.
2589 * Forward by arbitrary ilc, injection will take care of
2590 * nullification if necessary.
2591 */
2592 pgm_info = vcpu->arch.pgm;
2593 ilen = 4;
2594 }
2595 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
2596 kvm_s390_forward_psw(vcpu, ilen);
2597 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
2598 }
2599
2600 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
2601 {
2602 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
2603 vcpu->arch.sie_block->icptcode);
2604 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
2605
2606 if (guestdbg_enabled(vcpu))
2607 kvm_s390_restore_guest_per_regs(vcpu);
2608
2609 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
2610 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
2611
2612 if (vcpu->arch.sie_block->icptcode > 0) {
2613 int rc = kvm_handle_sie_intercept(vcpu);
2614
2615 if (rc != -EOPNOTSUPP)
2616 return rc;
2617 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
2618 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
2619 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
2620 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
2621 return -EREMOTE;
2622 } else if (exit_reason != -EFAULT) {
2623 vcpu->stat.exit_null++;
2624 return 0;
2625 } else if (kvm_is_ucontrol(vcpu->kvm)) {
2626 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
2627 vcpu->run->s390_ucontrol.trans_exc_code =
2628 current->thread.gmap_addr;
2629 vcpu->run->s390_ucontrol.pgm_code = 0x10;
2630 return -EREMOTE;
2631 } else if (current->thread.gmap_pfault) {
2632 trace_kvm_s390_major_guest_pfault(vcpu);
2633 current->thread.gmap_pfault = 0;
2634 if (kvm_arch_setup_async_pf(vcpu))
2635 return 0;
2636 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
2637 }
2638 return vcpu_post_run_fault_in_sie(vcpu);
2639 }
2640
2641 static int __vcpu_run(struct kvm_vcpu *vcpu)
2642 {
2643 int rc, exit_reason;
2644
2645 /*
2646 * We try to hold kvm->srcu during most of vcpu_run (except when run-
2647 * ning the guest), so that memslots (and other stuff) are protected
2648 */
2649 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2650
2651 do {
2652 rc = vcpu_pre_run(vcpu);
2653 if (rc)
2654 break;
2655
2656 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2657 /*
2658 * As PF_VCPU will be used in fault handler, between
2659 * guest_enter and guest_exit should be no uaccess.
2660 */
2661 local_irq_disable();
2662 guest_enter_irqoff();
2663 __disable_cpu_timer_accounting(vcpu);
2664 local_irq_enable();
2665 exit_reason = sie64a(vcpu->arch.sie_block,
2666 vcpu->run->s.regs.gprs);
2667 local_irq_disable();
2668 __enable_cpu_timer_accounting(vcpu);
2669 guest_exit_irqoff();
2670 local_irq_enable();
2671 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2672
2673 rc = vcpu_post_run(vcpu, exit_reason);
2674 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
2675
2676 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2677 return rc;
2678 }
2679
2680 static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2681 {
2682 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
2683 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
2684 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
2685 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
2686 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
2687 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
2688 /* some control register changes require a tlb flush */
2689 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2690 }
2691 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
2692 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
2693 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
2694 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
2695 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
2696 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
2697 }
2698 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
2699 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
2700 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
2701 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
2702 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2703 kvm_clear_async_pf_completion_queue(vcpu);
2704 }
2705 kvm_run->kvm_dirty_regs = 0;
2706 }
2707
2708 static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2709 {
2710 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
2711 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
2712 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
2713 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
2714 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
2715 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
2716 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
2717 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
2718 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
2719 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
2720 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
2721 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
2722 }
2723
2724 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2725 {
2726 int rc;
2727 sigset_t sigsaved;
2728
2729 if (guestdbg_exit_pending(vcpu)) {
2730 kvm_s390_prepare_debug_exit(vcpu);
2731 return 0;
2732 }
2733
2734 if (vcpu->sigset_active)
2735 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2736
2737 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
2738 kvm_s390_vcpu_start(vcpu);
2739 } else if (is_vcpu_stopped(vcpu)) {
2740 pr_err_ratelimited("can't run stopped vcpu %d\n",
2741 vcpu->vcpu_id);
2742 return -EINVAL;
2743 }
2744
2745 sync_regs(vcpu, kvm_run);
2746 enable_cpu_timer_accounting(vcpu);
2747
2748 might_fault();
2749 rc = __vcpu_run(vcpu);
2750
2751 if (signal_pending(current) && !rc) {
2752 kvm_run->exit_reason = KVM_EXIT_INTR;
2753 rc = -EINTR;
2754 }
2755
2756 if (guestdbg_exit_pending(vcpu) && !rc) {
2757 kvm_s390_prepare_debug_exit(vcpu);
2758 rc = 0;
2759 }
2760
2761 if (rc == -EREMOTE) {
2762 /* userspace support is needed, kvm_run has been prepared */
2763 rc = 0;
2764 }
2765
2766 disable_cpu_timer_accounting(vcpu);
2767 store_regs(vcpu, kvm_run);
2768
2769 if (vcpu->sigset_active)
2770 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2771
2772 vcpu->stat.exit_userspace++;
2773 return rc;
2774 }
2775
2776 /*
2777 * store status at address
2778 * we use have two special cases:
2779 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
2780 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
2781 */
2782 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
2783 {
2784 unsigned char archmode = 1;
2785 freg_t fprs[NUM_FPRS];
2786 unsigned int px;
2787 u64 clkcomp, cputm;
2788 int rc;
2789
2790 px = kvm_s390_get_prefix(vcpu);
2791 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
2792 if (write_guest_abs(vcpu, 163, &archmode, 1))
2793 return -EFAULT;
2794 gpa = 0;
2795 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
2796 if (write_guest_real(vcpu, 163, &archmode, 1))
2797 return -EFAULT;
2798 gpa = px;
2799 } else
2800 gpa -= __LC_FPREGS_SAVE_AREA;
2801
2802 /* manually convert vector registers if necessary */
2803 if (MACHINE_HAS_VX) {
2804 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
2805 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2806 fprs, 128);
2807 } else {
2808 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2809 vcpu->run->s.regs.fprs, 128);
2810 }
2811 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
2812 vcpu->run->s.regs.gprs, 128);
2813 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
2814 &vcpu->arch.sie_block->gpsw, 16);
2815 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
2816 &px, 4);
2817 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
2818 &vcpu->run->s.regs.fpc, 4);
2819 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
2820 &vcpu->arch.sie_block->todpr, 4);
2821 cputm = kvm_s390_get_cpu_timer(vcpu);
2822 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
2823 &cputm, 8);
2824 clkcomp = vcpu->arch.sie_block->ckc >> 8;
2825 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
2826 &clkcomp, 8);
2827 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
2828 &vcpu->run->s.regs.acrs, 64);
2829 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
2830 &vcpu->arch.sie_block->gcr, 128);
2831 return rc ? -EFAULT : 0;
2832 }
2833
2834 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
2835 {
2836 /*
2837 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
2838 * copying in vcpu load/put. Lets update our copies before we save
2839 * it into the save area
2840 */
2841 save_fpu_regs();
2842 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
2843 save_access_regs(vcpu->run->s.regs.acrs);
2844
2845 return kvm_s390_store_status_unloaded(vcpu, addr);
2846 }
2847
2848 /*
2849 * store additional status at address
2850 */
2851 int kvm_s390_store_adtl_status_unloaded(struct kvm_vcpu *vcpu,
2852 unsigned long gpa)
2853 {
2854 /* Only bits 0-53 are used for address formation */
2855 if (!(gpa & ~0x3ff))
2856 return 0;
2857
2858 return write_guest_abs(vcpu, gpa & ~0x3ff,
2859 (void *)&vcpu->run->s.regs.vrs, 512);
2860 }
2861
2862 int kvm_s390_vcpu_store_adtl_status(struct kvm_vcpu *vcpu, unsigned long addr)
2863 {
2864 if (!test_kvm_facility(vcpu->kvm, 129))
2865 return 0;
2866
2867 /*
2868 * The guest VXRS are in the host VXRs due to the lazy
2869 * copying in vcpu load/put. We can simply call save_fpu_regs()
2870 * to save the current register state because we are in the
2871 * middle of a load/put cycle.
2872 *
2873 * Let's update our copies before we save it into the save area.
2874 */
2875 save_fpu_regs();
2876
2877 return kvm_s390_store_adtl_status_unloaded(vcpu, addr);
2878 }
2879
2880 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2881 {
2882 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
2883 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
2884 }
2885
2886 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
2887 {
2888 unsigned int i;
2889 struct kvm_vcpu *vcpu;
2890
2891 kvm_for_each_vcpu(i, vcpu, kvm) {
2892 __disable_ibs_on_vcpu(vcpu);
2893 }
2894 }
2895
2896 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2897 {
2898 if (!sclp.has_ibs)
2899 return;
2900 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
2901 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
2902 }
2903
2904 void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
2905 {
2906 int i, online_vcpus, started_vcpus = 0;
2907
2908 if (!is_vcpu_stopped(vcpu))
2909 return;
2910
2911 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
2912 /* Only one cpu at a time may enter/leave the STOPPED state. */
2913 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2914 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2915
2916 for (i = 0; i < online_vcpus; i++) {
2917 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i]))
2918 started_vcpus++;
2919 }
2920
2921 if (started_vcpus == 0) {
2922 /* we're the only active VCPU -> speed it up */
2923 __enable_ibs_on_vcpu(vcpu);
2924 } else if (started_vcpus == 1) {
2925 /*
2926 * As we are starting a second VCPU, we have to disable
2927 * the IBS facility on all VCPUs to remove potentially
2928 * oustanding ENABLE requests.
2929 */
2930 __disable_ibs_on_all_vcpus(vcpu->kvm);
2931 }
2932
2933 atomic_andnot(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2934 /*
2935 * Another VCPU might have used IBS while we were offline.
2936 * Let's play safe and flush the VCPU at startup.
2937 */
2938 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2939 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2940 return;
2941 }
2942
2943 void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
2944 {
2945 int i, online_vcpus, started_vcpus = 0;
2946 struct kvm_vcpu *started_vcpu = NULL;
2947
2948 if (is_vcpu_stopped(vcpu))
2949 return;
2950
2951 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
2952 /* Only one cpu at a time may enter/leave the STOPPED state. */
2953 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2954 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2955
2956 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
2957 kvm_s390_clear_stop_irq(vcpu);
2958
2959 atomic_or(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2960 __disable_ibs_on_vcpu(vcpu);
2961
2962 for (i = 0; i < online_vcpus; i++) {
2963 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) {
2964 started_vcpus++;
2965 started_vcpu = vcpu->kvm->vcpus[i];
2966 }
2967 }
2968
2969 if (started_vcpus == 1) {
2970 /*
2971 * As we only have one VCPU left, we want to enable the
2972 * IBS facility for that VCPU to speed it up.
2973 */
2974 __enable_ibs_on_vcpu(started_vcpu);
2975 }
2976
2977 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2978 return;
2979 }
2980
2981 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
2982 struct kvm_enable_cap *cap)
2983 {
2984 int r;
2985
2986 if (cap->flags)
2987 return -EINVAL;
2988
2989 switch (cap->cap) {
2990 case KVM_CAP_S390_CSS_SUPPORT:
2991 if (!vcpu->kvm->arch.css_support) {
2992 vcpu->kvm->arch.css_support = 1;
2993 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
2994 trace_kvm_s390_enable_css(vcpu->kvm);
2995 }
2996 r = 0;
2997 break;
2998 default:
2999 r = -EINVAL;
3000 break;
3001 }
3002 return r;
3003 }
3004
3005 static long kvm_s390_guest_mem_op(struct kvm_vcpu *vcpu,
3006 struct kvm_s390_mem_op *mop)
3007 {
3008 void __user *uaddr = (void __user *)mop->buf;
3009 void *tmpbuf = NULL;
3010 int r, srcu_idx;
3011 const u64 supported_flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION
3012 | KVM_S390_MEMOP_F_CHECK_ONLY;
3013
3014 if (mop->flags & ~supported_flags)
3015 return -EINVAL;
3016
3017 if (mop->size > MEM_OP_MAX_SIZE)
3018 return -E2BIG;
3019
3020 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
3021 tmpbuf = vmalloc(mop->size);
3022 if (!tmpbuf)
3023 return -ENOMEM;
3024 }
3025
3026 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3027
3028 switch (mop->op) {
3029 case KVM_S390_MEMOP_LOGICAL_READ:
3030 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3031 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3032 mop->size, GACC_FETCH);
3033 break;
3034 }
3035 r = read_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3036 if (r == 0) {
3037 if (copy_to_user(uaddr, tmpbuf, mop->size))
3038 r = -EFAULT;
3039 }
3040 break;
3041 case KVM_S390_MEMOP_LOGICAL_WRITE:
3042 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3043 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3044 mop->size, GACC_STORE);
3045 break;
3046 }
3047 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
3048 r = -EFAULT;
3049 break;
3050 }
3051 r = write_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3052 break;
3053 default:
3054 r = -EINVAL;
3055 }
3056
3057 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
3058
3059 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
3060 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
3061
3062 vfree(tmpbuf);
3063 return r;
3064 }
3065
3066 long kvm_arch_vcpu_ioctl(struct file *filp,
3067 unsigned int ioctl, unsigned long arg)
3068 {
3069 struct kvm_vcpu *vcpu = filp->private_data;
3070 void __user *argp = (void __user *)arg;
3071 int idx;
3072 long r;
3073
3074 switch (ioctl) {
3075 case KVM_S390_IRQ: {
3076 struct kvm_s390_irq s390irq;
3077
3078 r = -EFAULT;
3079 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
3080 break;
3081 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
3082 break;
3083 }
3084 case KVM_S390_INTERRUPT: {
3085 struct kvm_s390_interrupt s390int;
3086 struct kvm_s390_irq s390irq;
3087
3088 r = -EFAULT;
3089 if (copy_from_user(&s390int, argp, sizeof(s390int)))
3090 break;
3091 if (s390int_to_s390irq(&s390int, &s390irq))
3092 return -EINVAL;
3093 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
3094 break;
3095 }
3096 case KVM_S390_STORE_STATUS:
3097 idx = srcu_read_lock(&vcpu->kvm->srcu);
3098 r = kvm_s390_vcpu_store_status(vcpu, arg);
3099 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3100 break;
3101 case KVM_S390_SET_INITIAL_PSW: {
3102 psw_t psw;
3103
3104 r = -EFAULT;
3105 if (copy_from_user(&psw, argp, sizeof(psw)))
3106 break;
3107 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
3108 break;
3109 }
3110 case KVM_S390_INITIAL_RESET:
3111 r = kvm_arch_vcpu_ioctl_initial_reset(vcpu);
3112 break;
3113 case KVM_SET_ONE_REG:
3114 case KVM_GET_ONE_REG: {
3115 struct kvm_one_reg reg;
3116 r = -EFAULT;
3117 if (copy_from_user(&reg, argp, sizeof(reg)))
3118 break;
3119 if (ioctl == KVM_SET_ONE_REG)
3120 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, &reg);
3121 else
3122 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, &reg);
3123 break;
3124 }
3125 #ifdef CONFIG_KVM_S390_UCONTROL
3126 case KVM_S390_UCAS_MAP: {
3127 struct kvm_s390_ucas_mapping ucasmap;
3128
3129 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3130 r = -EFAULT;
3131 break;
3132 }
3133
3134 if (!kvm_is_ucontrol(vcpu->kvm)) {
3135 r = -EINVAL;
3136 break;
3137 }
3138
3139 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
3140 ucasmap.vcpu_addr, ucasmap.length);
3141 break;
3142 }
3143 case KVM_S390_UCAS_UNMAP: {
3144 struct kvm_s390_ucas_mapping ucasmap;
3145
3146 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3147 r = -EFAULT;
3148 break;
3149 }
3150
3151 if (!kvm_is_ucontrol(vcpu->kvm)) {
3152 r = -EINVAL;
3153 break;
3154 }
3155
3156 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
3157 ucasmap.length);
3158 break;
3159 }
3160 #endif
3161 case KVM_S390_VCPU_FAULT: {
3162 r = gmap_fault(vcpu->arch.gmap, arg, 0);
3163 break;
3164 }
3165 case KVM_ENABLE_CAP:
3166 {
3167 struct kvm_enable_cap cap;
3168 r = -EFAULT;
3169 if (copy_from_user(&cap, argp, sizeof(cap)))
3170 break;
3171 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3172 break;
3173 }
3174 case KVM_S390_MEM_OP: {
3175 struct kvm_s390_mem_op mem_op;
3176
3177 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
3178 r = kvm_s390_guest_mem_op(vcpu, &mem_op);
3179 else
3180 r = -EFAULT;
3181 break;
3182 }
3183 case KVM_S390_SET_IRQ_STATE: {
3184 struct kvm_s390_irq_state irq_state;
3185
3186 r = -EFAULT;
3187 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3188 break;
3189 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
3190 irq_state.len == 0 ||
3191 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
3192 r = -EINVAL;
3193 break;
3194 }
3195 r = kvm_s390_set_irq_state(vcpu,
3196 (void __user *) irq_state.buf,
3197 irq_state.len);
3198 break;
3199 }
3200 case KVM_S390_GET_IRQ_STATE: {
3201 struct kvm_s390_irq_state irq_state;
3202
3203 r = -EFAULT;
3204 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3205 break;
3206 if (irq_state.len == 0) {
3207 r = -EINVAL;
3208 break;
3209 }
3210 r = kvm_s390_get_irq_state(vcpu,
3211 (__u8 __user *) irq_state.buf,
3212 irq_state.len);
3213 break;
3214 }
3215 default:
3216 r = -ENOTTY;
3217 }
3218 return r;
3219 }
3220
3221 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3222 {
3223 #ifdef CONFIG_KVM_S390_UCONTROL
3224 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
3225 && (kvm_is_ucontrol(vcpu->kvm))) {
3226 vmf->page = virt_to_page(vcpu->arch.sie_block);
3227 get_page(vmf->page);
3228 return 0;
3229 }
3230 #endif
3231 return VM_FAULT_SIGBUS;
3232 }
3233
3234 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
3235 unsigned long npages)
3236 {
3237 return 0;
3238 }
3239
3240 /* Section: memory related */
3241 int kvm_arch_prepare_memory_region(struct kvm *kvm,
3242 struct kvm_memory_slot *memslot,
3243 const struct kvm_userspace_memory_region *mem,
3244 enum kvm_mr_change change)
3245 {
3246 /* A few sanity checks. We can have memory slots which have to be
3247 located/ended at a segment boundary (1MB). The memory in userland is
3248 ok to be fragmented into various different vmas. It is okay to mmap()
3249 and munmap() stuff in this slot after doing this call at any time */
3250
3251 if (mem->userspace_addr & 0xffffful)
3252 return -EINVAL;
3253
3254 if (mem->memory_size & 0xffffful)
3255 return -EINVAL;
3256
3257 if (mem->guest_phys_addr + mem->memory_size > kvm->arch.mem_limit)
3258 return -EINVAL;
3259
3260 return 0;
3261 }
3262
3263 void kvm_arch_commit_memory_region(struct kvm *kvm,
3264 const struct kvm_userspace_memory_region *mem,
3265 const struct kvm_memory_slot *old,
3266 const struct kvm_memory_slot *new,
3267 enum kvm_mr_change change)
3268 {
3269 int rc;
3270
3271 /* If the basics of the memslot do not change, we do not want
3272 * to update the gmap. Every update causes several unnecessary
3273 * segment translation exceptions. This is usually handled just
3274 * fine by the normal fault handler + gmap, but it will also
3275 * cause faults on the prefix page of running guest CPUs.
3276 */
3277 if (old->userspace_addr == mem->userspace_addr &&
3278 old->base_gfn * PAGE_SIZE == mem->guest_phys_addr &&
3279 old->npages * PAGE_SIZE == mem->memory_size)
3280 return;
3281
3282 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr,
3283 mem->guest_phys_addr, mem->memory_size);
3284 if (rc)
3285 pr_warn("failed to commit memory region\n");
3286 return;
3287 }
3288
3289 static inline unsigned long nonhyp_mask(int i)
3290 {
3291 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
3292
3293 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
3294 }
3295
3296 void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu)
3297 {
3298 vcpu->valid_wakeup = false;
3299 }
3300
3301 static int __init kvm_s390_init(void)
3302 {
3303 int i;
3304
3305 if (!sclp.has_sief2) {
3306 pr_info("SIE not available\n");
3307 return -ENODEV;
3308 }
3309
3310 for (i = 0; i < 16; i++)
3311 kvm_s390_fac_list_mask[i] |=
3312 S390_lowcore.stfle_fac_list[i] & nonhyp_mask(i);
3313
3314 return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
3315 }
3316
3317 static void __exit kvm_s390_exit(void)
3318 {
3319 kvm_exit();
3320 }
3321
3322 module_init(kvm_s390_init);
3323 module_exit(kvm_s390_exit);
3324
3325 /*
3326 * Enable autoloading of the kvm module.
3327 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
3328 * since x86 takes a different approach.
3329 */
3330 #include <linux/miscdevice.h>
3331 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3332 MODULE_ALIAS("devname:kvm");
Linux kernel - Deliverables
This page took 0.137718 seconds and 6 git commands to generate.