projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge remote-tracking branch 'kvm/linux-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 (!kvm_s390_use_sca_entries())
388 r = KVM_MAX_VCPUS;
389 else if (sclp.has_esca && sclp.has_64bscao)
390 r = KVM_S390_ESCA_CPU_SLOTS;
391 break;
392 case KVM_CAP_NR_MEMSLOTS:
393 r = KVM_USER_MEM_SLOTS;
394 break;
395 case KVM_CAP_S390_COW:
396 r = MACHINE_HAS_ESOP;
397 break;
398 case KVM_CAP_S390_VECTOR_REGISTERS:
399 r = MACHINE_HAS_VX;
400 break;
401 case KVM_CAP_S390_RI:
402 r = test_facility(64);
403 break;
404 default:
405 r = 0;
406 }
407 return r;
408 }
409
410 static void kvm_s390_sync_dirty_log(struct kvm *kvm,
411 struct kvm_memory_slot *memslot)
412 {
413 gfn_t cur_gfn, last_gfn;
414 unsigned long address;
415 struct gmap *gmap = kvm->arch.gmap;
416
417 /* Loop over all guest pages */
418 last_gfn = memslot->base_gfn + memslot->npages;
419 for (cur_gfn = memslot->base_gfn; cur_gfn <= last_gfn; cur_gfn++) {
420 address = gfn_to_hva_memslot(memslot, cur_gfn);
421
422 if (test_and_clear_guest_dirty(gmap->mm, address))
423 mark_page_dirty(kvm, cur_gfn);
424 if (fatal_signal_pending(current))
425 return;
426 cond_resched();
427 }
428 }
429
430 /* Section: vm related */
431 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
432
433 /*
434 * Get (and clear) the dirty memory log for a memory slot.
435 */
436 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
437 struct kvm_dirty_log *log)
438 {
439 int r;
440 unsigned long n;
441 struct kvm_memslots *slots;
442 struct kvm_memory_slot *memslot;
443 int is_dirty = 0;
444
445 mutex_lock(&kvm->slots_lock);
446
447 r = -EINVAL;
448 if (log->slot >= KVM_USER_MEM_SLOTS)
449 goto out;
450
451 slots = kvm_memslots(kvm);
452 memslot = id_to_memslot(slots, log->slot);
453 r = -ENOENT;
454 if (!memslot->dirty_bitmap)
455 goto out;
456
457 kvm_s390_sync_dirty_log(kvm, memslot);
458 r = kvm_get_dirty_log(kvm, log, &is_dirty);
459 if (r)
460 goto out;
461
462 /* Clear the dirty log */
463 if (is_dirty) {
464 n = kvm_dirty_bitmap_bytes(memslot);
465 memset(memslot->dirty_bitmap, 0, n);
466 }
467 r = 0;
468 out:
469 mutex_unlock(&kvm->slots_lock);
470 return r;
471 }
472
473 static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
474 {
475 unsigned int i;
476 struct kvm_vcpu *vcpu;
477
478 kvm_for_each_vcpu(i, vcpu, kvm) {
479 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
480 }
481 }
482
483 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
484 {
485 int r;
486
487 if (cap->flags)
488 return -EINVAL;
489
490 switch (cap->cap) {
491 case KVM_CAP_S390_IRQCHIP:
492 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
493 kvm->arch.use_irqchip = 1;
494 r = 0;
495 break;
496 case KVM_CAP_S390_USER_SIGP:
497 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
498 kvm->arch.user_sigp = 1;
499 r = 0;
500 break;
501 case KVM_CAP_S390_VECTOR_REGISTERS:
502 mutex_lock(&kvm->lock);
503 if (kvm->created_vcpus) {
504 r = -EBUSY;
505 } else if (MACHINE_HAS_VX) {
506 set_kvm_facility(kvm->arch.model.fac_mask, 129);
507 set_kvm_facility(kvm->arch.model.fac_list, 129);
508 r = 0;
509 } else
510 r = -EINVAL;
511 mutex_unlock(&kvm->lock);
512 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
513 r ? "(not available)" : "(success)");
514 break;
515 case KVM_CAP_S390_RI:
516 r = -EINVAL;
517 mutex_lock(&kvm->lock);
518 if (kvm->created_vcpus) {
519 r = -EBUSY;
520 } else if (test_facility(64)) {
521 set_kvm_facility(kvm->arch.model.fac_mask, 64);
522 set_kvm_facility(kvm->arch.model.fac_list, 64);
523 r = 0;
524 }
525 mutex_unlock(&kvm->lock);
526 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
527 r ? "(not available)" : "(success)");
528 break;
529 case KVM_CAP_S390_USER_STSI:
530 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
531 kvm->arch.user_stsi = 1;
532 r = 0;
533 break;
534 case KVM_CAP_S390_USER_INSTR0:
535 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
536 kvm->arch.user_instr0 = 1;
537 icpt_operexc_on_all_vcpus(kvm);
538 r = 0;
539 break;
540 default:
541 r = -EINVAL;
542 break;
543 }
544 return r;
545 }
546
547 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
548 {
549 int ret;
550
551 switch (attr->attr) {
552 case KVM_S390_VM_MEM_LIMIT_SIZE:
553 ret = 0;
554 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
555 kvm->arch.mem_limit);
556 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
557 ret = -EFAULT;
558 break;
559 default:
560 ret = -ENXIO;
561 break;
562 }
563 return ret;
564 }
565
566 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
567 {
568 int ret;
569 unsigned int idx;
570 switch (attr->attr) {
571 case KVM_S390_VM_MEM_ENABLE_CMMA:
572 ret = -ENXIO;
573 if (!sclp.has_cmma)
574 break;
575
576 ret = -EBUSY;
577 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
578 mutex_lock(&kvm->lock);
579 if (!kvm->created_vcpus) {
580 kvm->arch.use_cmma = 1;
581 ret = 0;
582 }
583 mutex_unlock(&kvm->lock);
584 break;
585 case KVM_S390_VM_MEM_CLR_CMMA:
586 ret = -ENXIO;
587 if (!sclp.has_cmma)
588 break;
589 ret = -EINVAL;
590 if (!kvm->arch.use_cmma)
591 break;
592
593 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
594 mutex_lock(&kvm->lock);
595 idx = srcu_read_lock(&kvm->srcu);
596 s390_reset_cmma(kvm->arch.gmap->mm);
597 srcu_read_unlock(&kvm->srcu, idx);
598 mutex_unlock(&kvm->lock);
599 ret = 0;
600 break;
601 case KVM_S390_VM_MEM_LIMIT_SIZE: {
602 unsigned long new_limit;
603
604 if (kvm_is_ucontrol(kvm))
605 return -EINVAL;
606
607 if (get_user(new_limit, (u64 __user *)attr->addr))
608 return -EFAULT;
609
610 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
611 new_limit > kvm->arch.mem_limit)
612 return -E2BIG;
613
614 if (!new_limit)
615 return -EINVAL;
616
617 /* gmap_create takes last usable address */
618 if (new_limit != KVM_S390_NO_MEM_LIMIT)
619 new_limit -= 1;
620
621 ret = -EBUSY;
622 mutex_lock(&kvm->lock);
623 if (!kvm->created_vcpus) {
624 /* gmap_create will round the limit up */
625 struct gmap *new = gmap_create(current->mm, new_limit);
626
627 if (!new) {
628 ret = -ENOMEM;
629 } else {
630 gmap_remove(kvm->arch.gmap);
631 new->private = kvm;
632 kvm->arch.gmap = new;
633 ret = 0;
634 }
635 }
636 mutex_unlock(&kvm->lock);
637 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
638 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
639 (void *) kvm->arch.gmap->asce);
640 break;
641 }
642 default:
643 ret = -ENXIO;
644 break;
645 }
646 return ret;
647 }
648
649 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
650
651 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
652 {
653 struct kvm_vcpu *vcpu;
654 int i;
655
656 if (!test_kvm_facility(kvm, 76))
657 return -EINVAL;
658
659 mutex_lock(&kvm->lock);
660 switch (attr->attr) {
661 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
662 get_random_bytes(
663 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
664 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
665 kvm->arch.crypto.aes_kw = 1;
666 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
667 break;
668 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
669 get_random_bytes(
670 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
671 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
672 kvm->arch.crypto.dea_kw = 1;
673 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
674 break;
675 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
676 kvm->arch.crypto.aes_kw = 0;
677 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
678 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
679 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
680 break;
681 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
682 kvm->arch.crypto.dea_kw = 0;
683 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
684 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
685 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
686 break;
687 default:
688 mutex_unlock(&kvm->lock);
689 return -ENXIO;
690 }
691
692 kvm_for_each_vcpu(i, vcpu, kvm) {
693 kvm_s390_vcpu_crypto_setup(vcpu);
694 exit_sie(vcpu);
695 }
696 mutex_unlock(&kvm->lock);
697 return 0;
698 }
699
700 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
701 {
702 u8 gtod_high;
703
704 if (copy_from_user(&gtod_high, (void __user *)attr->addr,
705 sizeof(gtod_high)))
706 return -EFAULT;
707
708 if (gtod_high != 0)
709 return -EINVAL;
710 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
711
712 return 0;
713 }
714
715 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
716 {
717 u64 gtod;
718
719 if (copy_from_user(&gtod, (void __user *)attr->addr, sizeof(gtod)))
720 return -EFAULT;
721
722 kvm_s390_set_tod_clock(kvm, gtod);
723 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod);
724 return 0;
725 }
726
727 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
728 {
729 int ret;
730
731 if (attr->flags)
732 return -EINVAL;
733
734 switch (attr->attr) {
735 case KVM_S390_VM_TOD_HIGH:
736 ret = kvm_s390_set_tod_high(kvm, attr);
737 break;
738 case KVM_S390_VM_TOD_LOW:
739 ret = kvm_s390_set_tod_low(kvm, attr);
740 break;
741 default:
742 ret = -ENXIO;
743 break;
744 }
745 return ret;
746 }
747
748 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
749 {
750 u8 gtod_high = 0;
751
752 if (copy_to_user((void __user *)attr->addr, &gtod_high,
753 sizeof(gtod_high)))
754 return -EFAULT;
755 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
756
757 return 0;
758 }
759
760 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
761 {
762 u64 gtod;
763
764 gtod = kvm_s390_get_tod_clock_fast(kvm);
765 if (copy_to_user((void __user *)attr->addr, &gtod, sizeof(gtod)))
766 return -EFAULT;
767 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
768
769 return 0;
770 }
771
772 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
773 {
774 int ret;
775
776 if (attr->flags)
777 return -EINVAL;
778
779 switch (attr->attr) {
780 case KVM_S390_VM_TOD_HIGH:
781 ret = kvm_s390_get_tod_high(kvm, attr);
782 break;
783 case KVM_S390_VM_TOD_LOW:
784 ret = kvm_s390_get_tod_low(kvm, attr);
785 break;
786 default:
787 ret = -ENXIO;
788 break;
789 }
790 return ret;
791 }
792
793 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
794 {
795 struct kvm_s390_vm_cpu_processor *proc;
796 u16 lowest_ibc, unblocked_ibc;
797 int ret = 0;
798
799 mutex_lock(&kvm->lock);
800 if (kvm->created_vcpus) {
801 ret = -EBUSY;
802 goto out;
803 }
804 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
805 if (!proc) {
806 ret = -ENOMEM;
807 goto out;
808 }
809 if (!copy_from_user(proc, (void __user *)attr->addr,
810 sizeof(*proc))) {
811 kvm->arch.model.cpuid = proc->cpuid;
812 lowest_ibc = sclp.ibc >> 16 & 0xfff;
813 unblocked_ibc = sclp.ibc & 0xfff;
814 if (lowest_ibc && proc->ibc) {
815 if (proc->ibc > unblocked_ibc)
816 kvm->arch.model.ibc = unblocked_ibc;
817 else if (proc->ibc < lowest_ibc)
818 kvm->arch.model.ibc = lowest_ibc;
819 else
820 kvm->arch.model.ibc = proc->ibc;
821 }
822 memcpy(kvm->arch.model.fac_list, proc->fac_list,
823 S390_ARCH_FAC_LIST_SIZE_BYTE);
824 } else
825 ret = -EFAULT;
826 kfree(proc);
827 out:
828 mutex_unlock(&kvm->lock);
829 return ret;
830 }
831
832 static int kvm_s390_set_processor_feat(struct kvm *kvm,
833 struct kvm_device_attr *attr)
834 {
835 struct kvm_s390_vm_cpu_feat data;
836 int ret = -EBUSY;
837
838 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
839 return -EFAULT;
840 if (!bitmap_subset((unsigned long *) data.feat,
841 kvm_s390_available_cpu_feat,
842 KVM_S390_VM_CPU_FEAT_NR_BITS))
843 return -EINVAL;
844
845 mutex_lock(&kvm->lock);
846 if (!atomic_read(&kvm->online_vcpus)) {
847 bitmap_copy(kvm->arch.cpu_feat, (unsigned long *) data.feat,
848 KVM_S390_VM_CPU_FEAT_NR_BITS);
849 ret = 0;
850 }
851 mutex_unlock(&kvm->lock);
852 return ret;
853 }
854
855 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
856 struct kvm_device_attr *attr)
857 {
858 /*
859 * Once supported by kernel + hw, we have to store the subfunctions
860 * in kvm->arch and remember that user space configured them.
861 */
862 return -ENXIO;
863 }
864
865 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
866 {
867 int ret = -ENXIO;
868
869 switch (attr->attr) {
870 case KVM_S390_VM_CPU_PROCESSOR:
871 ret = kvm_s390_set_processor(kvm, attr);
872 break;
873 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
874 ret = kvm_s390_set_processor_feat(kvm, attr);
875 break;
876 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
877 ret = kvm_s390_set_processor_subfunc(kvm, attr);
878 break;
879 }
880 return ret;
881 }
882
883 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
884 {
885 struct kvm_s390_vm_cpu_processor *proc;
886 int ret = 0;
887
888 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
889 if (!proc) {
890 ret = -ENOMEM;
891 goto out;
892 }
893 proc->cpuid = kvm->arch.model.cpuid;
894 proc->ibc = kvm->arch.model.ibc;
895 memcpy(&proc->fac_list, kvm->arch.model.fac_list,
896 S390_ARCH_FAC_LIST_SIZE_BYTE);
897 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
898 ret = -EFAULT;
899 kfree(proc);
900 out:
901 return ret;
902 }
903
904 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
905 {
906 struct kvm_s390_vm_cpu_machine *mach;
907 int ret = 0;
908
909 mach = kzalloc(sizeof(*mach), GFP_KERNEL);
910 if (!mach) {
911 ret = -ENOMEM;
912 goto out;
913 }
914 get_cpu_id((struct cpuid *) &mach->cpuid);
915 mach->ibc = sclp.ibc;
916 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
917 S390_ARCH_FAC_LIST_SIZE_BYTE);
918 memcpy((unsigned long *)&mach->fac_list, S390_lowcore.stfle_fac_list,
919 S390_ARCH_FAC_LIST_SIZE_BYTE);
920 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
921 ret = -EFAULT;
922 kfree(mach);
923 out:
924 return ret;
925 }
926
927 static int kvm_s390_get_processor_feat(struct kvm *kvm,
928 struct kvm_device_attr *attr)
929 {
930 struct kvm_s390_vm_cpu_feat data;
931
932 bitmap_copy((unsigned long *) data.feat, kvm->arch.cpu_feat,
933 KVM_S390_VM_CPU_FEAT_NR_BITS);
934 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
935 return -EFAULT;
936 return 0;
937 }
938
939 static int kvm_s390_get_machine_feat(struct kvm *kvm,
940 struct kvm_device_attr *attr)
941 {
942 struct kvm_s390_vm_cpu_feat data;
943
944 bitmap_copy((unsigned long *) data.feat,
945 kvm_s390_available_cpu_feat,
946 KVM_S390_VM_CPU_FEAT_NR_BITS);
947 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
948 return -EFAULT;
949 return 0;
950 }
951
952 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
953 struct kvm_device_attr *attr)
954 {
955 /*
956 * Once we can actually configure subfunctions (kernel + hw support),
957 * we have to check if they were already set by user space, if so copy
958 * them from kvm->arch.
959 */
960 return -ENXIO;
961 }
962
963 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
964 struct kvm_device_attr *attr)
965 {
966 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
967 sizeof(struct kvm_s390_vm_cpu_subfunc)))
968 return -EFAULT;
969 return 0;
970 }
971 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
972 {
973 int ret = -ENXIO;
974
975 switch (attr->attr) {
976 case KVM_S390_VM_CPU_PROCESSOR:
977 ret = kvm_s390_get_processor(kvm, attr);
978 break;
979 case KVM_S390_VM_CPU_MACHINE:
980 ret = kvm_s390_get_machine(kvm, attr);
981 break;
982 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
983 ret = kvm_s390_get_processor_feat(kvm, attr);
984 break;
985 case KVM_S390_VM_CPU_MACHINE_FEAT:
986 ret = kvm_s390_get_machine_feat(kvm, attr);
987 break;
988 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
989 ret = kvm_s390_get_processor_subfunc(kvm, attr);
990 break;
991 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
992 ret = kvm_s390_get_machine_subfunc(kvm, attr);
993 break;
994 }
995 return ret;
996 }
997
998 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
999 {
1000 int ret;
1001
1002 switch (attr->group) {
1003 case KVM_S390_VM_MEM_CTRL:
1004 ret = kvm_s390_set_mem_control(kvm, attr);
1005 break;
1006 case KVM_S390_VM_TOD:
1007 ret = kvm_s390_set_tod(kvm, attr);
1008 break;
1009 case KVM_S390_VM_CPU_MODEL:
1010 ret = kvm_s390_set_cpu_model(kvm, attr);
1011 break;
1012 case KVM_S390_VM_CRYPTO:
1013 ret = kvm_s390_vm_set_crypto(kvm, attr);
1014 break;
1015 default:
1016 ret = -ENXIO;
1017 break;
1018 }
1019
1020 return ret;
1021 }
1022
1023 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1024 {
1025 int ret;
1026
1027 switch (attr->group) {
1028 case KVM_S390_VM_MEM_CTRL:
1029 ret = kvm_s390_get_mem_control(kvm, attr);
1030 break;
1031 case KVM_S390_VM_TOD:
1032 ret = kvm_s390_get_tod(kvm, attr);
1033 break;
1034 case KVM_S390_VM_CPU_MODEL:
1035 ret = kvm_s390_get_cpu_model(kvm, attr);
1036 break;
1037 default:
1038 ret = -ENXIO;
1039 break;
1040 }
1041
1042 return ret;
1043 }
1044
1045 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1046 {
1047 int ret;
1048
1049 switch (attr->group) {
1050 case KVM_S390_VM_MEM_CTRL:
1051 switch (attr->attr) {
1052 case KVM_S390_VM_MEM_ENABLE_CMMA:
1053 case KVM_S390_VM_MEM_CLR_CMMA:
1054 ret = sclp.has_cmma ? 0 : -ENXIO;
1055 break;
1056 case KVM_S390_VM_MEM_LIMIT_SIZE:
1057 ret = 0;
1058 break;
1059 default:
1060 ret = -ENXIO;
1061 break;
1062 }
1063 break;
1064 case KVM_S390_VM_TOD:
1065 switch (attr->attr) {
1066 case KVM_S390_VM_TOD_LOW:
1067 case KVM_S390_VM_TOD_HIGH:
1068 ret = 0;
1069 break;
1070 default:
1071 ret = -ENXIO;
1072 break;
1073 }
1074 break;
1075 case KVM_S390_VM_CPU_MODEL:
1076 switch (attr->attr) {
1077 case KVM_S390_VM_CPU_PROCESSOR:
1078 case KVM_S390_VM_CPU_MACHINE:
1079 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1080 case KVM_S390_VM_CPU_MACHINE_FEAT:
1081 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1082 ret = 0;
1083 break;
1084 /* configuring subfunctions is not supported yet */
1085 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1086 default:
1087 ret = -ENXIO;
1088 break;
1089 }
1090 break;
1091 case KVM_S390_VM_CRYPTO:
1092 switch (attr->attr) {
1093 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1094 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1095 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1096 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1097 ret = 0;
1098 break;
1099 default:
1100 ret = -ENXIO;
1101 break;
1102 }
1103 break;
1104 default:
1105 ret = -ENXIO;
1106 break;
1107 }
1108
1109 return ret;
1110 }
1111
1112 static long kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1113 {
1114 uint8_t *keys;
1115 uint64_t hva;
1116 int i, r = 0;
1117
1118 if (args->flags != 0)
1119 return -EINVAL;
1120
1121 /* Is this guest using storage keys? */
1122 if (!mm_use_skey(current->mm))
1123 return KVM_S390_GET_SKEYS_NONE;
1124
1125 /* Enforce sane limit on memory allocation */
1126 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1127 return -EINVAL;
1128
1129 keys = kmalloc_array(args->count, sizeof(uint8_t),
1130 GFP_KERNEL | __GFP_NOWARN);
1131 if (!keys)
1132 keys = vmalloc(sizeof(uint8_t) * args->count);
1133 if (!keys)
1134 return -ENOMEM;
1135
1136 down_read(&current->mm->mmap_sem);
1137 for (i = 0; i < args->count; i++) {
1138 hva = gfn_to_hva(kvm, args->start_gfn + i);
1139 if (kvm_is_error_hva(hva)) {
1140 r = -EFAULT;
1141 break;
1142 }
1143
1144 r = get_guest_storage_key(current->mm, hva, &keys[i]);
1145 if (r)
1146 break;
1147 }
1148 up_read(&current->mm->mmap_sem);
1149
1150 if (!r) {
1151 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
1152 sizeof(uint8_t) * args->count);
1153 if (r)
1154 r = -EFAULT;
1155 }
1156
1157 kvfree(keys);
1158 return r;
1159 }
1160
1161 static long kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1162 {
1163 uint8_t *keys;
1164 uint64_t hva;
1165 int i, r = 0;
1166
1167 if (args->flags != 0)
1168 return -EINVAL;
1169
1170 /* Enforce sane limit on memory allocation */
1171 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1172 return -EINVAL;
1173
1174 keys = kmalloc_array(args->count, sizeof(uint8_t),
1175 GFP_KERNEL | __GFP_NOWARN);
1176 if (!keys)
1177 keys = vmalloc(sizeof(uint8_t) * args->count);
1178 if (!keys)
1179 return -ENOMEM;
1180
1181 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
1182 sizeof(uint8_t) * args->count);
1183 if (r) {
1184 r = -EFAULT;
1185 goto out;
1186 }
1187
1188 /* Enable storage key handling for the guest */
1189 r = s390_enable_skey();
1190 if (r)
1191 goto out;
1192
1193 down_read(&current->mm->mmap_sem);
1194 for (i = 0; i < args->count; i++) {
1195 hva = gfn_to_hva(kvm, args->start_gfn + i);
1196 if (kvm_is_error_hva(hva)) {
1197 r = -EFAULT;
1198 break;
1199 }
1200
1201 /* Lowest order bit is reserved */
1202 if (keys[i] & 0x01) {
1203 r = -EINVAL;
1204 break;
1205 }
1206
1207 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
1208 if (r)
1209 break;
1210 }
1211 up_read(&current->mm->mmap_sem);
1212 out:
1213 kvfree(keys);
1214 return r;
1215 }
1216
1217 long kvm_arch_vm_ioctl(struct file *filp,
1218 unsigned int ioctl, unsigned long arg)
1219 {
1220 struct kvm *kvm = filp->private_data;
1221 void __user *argp = (void __user *)arg;
1222 struct kvm_device_attr attr;
1223 int r;
1224
1225 switch (ioctl) {
1226 case KVM_S390_INTERRUPT: {
1227 struct kvm_s390_interrupt s390int;
1228
1229 r = -EFAULT;
1230 if (copy_from_user(&s390int, argp, sizeof(s390int)))
1231 break;
1232 r = kvm_s390_inject_vm(kvm, &s390int);
1233 break;
1234 }
1235 case KVM_ENABLE_CAP: {
1236 struct kvm_enable_cap cap;
1237 r = -EFAULT;
1238 if (copy_from_user(&cap, argp, sizeof(cap)))
1239 break;
1240 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1241 break;
1242 }
1243 case KVM_CREATE_IRQCHIP: {
1244 struct kvm_irq_routing_entry routing;
1245
1246 r = -EINVAL;
1247 if (kvm->arch.use_irqchip) {
1248 /* Set up dummy routing. */
1249 memset(&routing, 0, sizeof(routing));
1250 r = kvm_set_irq_routing(kvm, &routing, 0, 0);
1251 }
1252 break;
1253 }
1254 case KVM_SET_DEVICE_ATTR: {
1255 r = -EFAULT;
1256 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1257 break;
1258 r = kvm_s390_vm_set_attr(kvm, &attr);
1259 break;
1260 }
1261 case KVM_GET_DEVICE_ATTR: {
1262 r = -EFAULT;
1263 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1264 break;
1265 r = kvm_s390_vm_get_attr(kvm, &attr);
1266 break;
1267 }
1268 case KVM_HAS_DEVICE_ATTR: {
1269 r = -EFAULT;
1270 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1271 break;
1272 r = kvm_s390_vm_has_attr(kvm, &attr);
1273 break;
1274 }
1275 case KVM_S390_GET_SKEYS: {
1276 struct kvm_s390_skeys args;
1277
1278 r = -EFAULT;
1279 if (copy_from_user(&args, argp,
1280 sizeof(struct kvm_s390_skeys)))
1281 break;
1282 r = kvm_s390_get_skeys(kvm, &args);
1283 break;
1284 }
1285 case KVM_S390_SET_SKEYS: {
1286 struct kvm_s390_skeys args;
1287
1288 r = -EFAULT;
1289 if (copy_from_user(&args, argp,
1290 sizeof(struct kvm_s390_skeys)))
1291 break;
1292 r = kvm_s390_set_skeys(kvm, &args);
1293 break;
1294 }
1295 default:
1296 r = -ENOTTY;
1297 }
1298
1299 return r;
1300 }
1301
1302 static int kvm_s390_query_ap_config(u8 *config)
1303 {
1304 u32 fcn_code = 0x04000000UL;
1305 u32 cc = 0;
1306
1307 memset(config, 0, 128);
1308 asm volatile(
1309 "lgr 0,%1\n"
1310 "lgr 2,%2\n"
1311 ".long 0xb2af0000\n" /* PQAP(QCI) */
1312 "0: ipm %0\n"
1313 "srl %0,28\n"
1314 "1:\n"
1315 EX_TABLE(0b, 1b)
1316 : "+r" (cc)
1317 : "r" (fcn_code), "r" (config)
1318 : "cc", "0", "2", "memory"
1319 );
1320
1321 return cc;
1322 }
1323
1324 static int kvm_s390_apxa_installed(void)
1325 {
1326 u8 config[128];
1327 int cc;
1328
1329 if (test_facility(12)) {
1330 cc = kvm_s390_query_ap_config(config);
1331
1332 if (cc)
1333 pr_err("PQAP(QCI) failed with cc=%d", cc);
1334 else
1335 return config[0] & 0x40;
1336 }
1337
1338 return 0;
1339 }
1340
1341 static void kvm_s390_set_crycb_format(struct kvm *kvm)
1342 {
1343 kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
1344
1345 if (kvm_s390_apxa_installed())
1346 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
1347 else
1348 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
1349 }
1350
1351 static u64 kvm_s390_get_initial_cpuid(void)
1352 {
1353 struct cpuid cpuid;
1354
1355 get_cpu_id(&cpuid);
1356 cpuid.version = 0xff;
1357 return *((u64 *) &cpuid);
1358 }
1359
1360 static void kvm_s390_crypto_init(struct kvm *kvm)
1361 {
1362 if (!test_kvm_facility(kvm, 76))
1363 return;
1364
1365 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
1366 kvm_s390_set_crycb_format(kvm);
1367
1368 /* Enable AES/DEA protected key functions by default */
1369 kvm->arch.crypto.aes_kw = 1;
1370 kvm->arch.crypto.dea_kw = 1;
1371 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1372 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1373 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1374 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1375 }
1376
1377 static void sca_dispose(struct kvm *kvm)
1378 {
1379 if (kvm->arch.use_esca)
1380 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
1381 else
1382 free_page((unsigned long)(kvm->arch.sca));
1383 kvm->arch.sca = NULL;
1384 }
1385
1386 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
1387 {
1388 gfp_t alloc_flags = GFP_KERNEL;
1389 int i, rc;
1390 char debug_name[16];
1391 static unsigned long sca_offset;
1392
1393 rc = -EINVAL;
1394 #ifdef CONFIG_KVM_S390_UCONTROL
1395 if (type & ~KVM_VM_S390_UCONTROL)
1396 goto out_err;
1397 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
1398 goto out_err;
1399 #else
1400 if (type)
1401 goto out_err;
1402 #endif
1403
1404 rc = s390_enable_sie();
1405 if (rc)
1406 goto out_err;
1407
1408 rc = -ENOMEM;
1409
1410 ratelimit_state_init(&kvm->arch.sthyi_limit, 5 * HZ, 500);
1411
1412 kvm->arch.use_esca = 0; /* start with basic SCA */
1413 if (!sclp.has_64bscao)
1414 alloc_flags |= GFP_DMA;
1415 rwlock_init(&kvm->arch.sca_lock);
1416 kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
1417 if (!kvm->arch.sca)
1418 goto out_err;
1419 spin_lock(&kvm_lock);
1420 sca_offset += 16;
1421 if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
1422 sca_offset = 0;
1423 kvm->arch.sca = (struct bsca_block *)
1424 ((char *) kvm->arch.sca + sca_offset);
1425 spin_unlock(&kvm_lock);
1426
1427 sprintf(debug_name, "kvm-%u", current->pid);
1428
1429 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
1430 if (!kvm->arch.dbf)
1431 goto out_err;
1432
1433 kvm->arch.sie_page2 =
1434 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1435 if (!kvm->arch.sie_page2)
1436 goto out_err;
1437
1438 /* Populate the facility mask initially. */
1439 memcpy(kvm->arch.model.fac_mask, S390_lowcore.stfle_fac_list,
1440 S390_ARCH_FAC_LIST_SIZE_BYTE);
1441 for (i = 0; i < S390_ARCH_FAC_LIST_SIZE_U64; i++) {
1442 if (i < kvm_s390_fac_list_mask_size())
1443 kvm->arch.model.fac_mask[i] &= kvm_s390_fac_list_mask[i];
1444 else
1445 kvm->arch.model.fac_mask[i] = 0UL;
1446 }
1447
1448 /* Populate the facility list initially. */
1449 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
1450 memcpy(kvm->arch.model.fac_list, kvm->arch.model.fac_mask,
1451 S390_ARCH_FAC_LIST_SIZE_BYTE);
1452
1453 set_kvm_facility(kvm->arch.model.fac_mask, 74);
1454 set_kvm_facility(kvm->arch.model.fac_list, 74);
1455
1456 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
1457 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
1458
1459 kvm_s390_crypto_init(kvm);
1460
1461 spin_lock_init(&kvm->arch.float_int.lock);
1462 for (i = 0; i < FIRQ_LIST_COUNT; i++)
1463 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
1464 init_waitqueue_head(&kvm->arch.ipte_wq);
1465 mutex_init(&kvm->arch.ipte_mutex);
1466
1467 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
1468 VM_EVENT(kvm, 3, "vm created with type %lu", type);
1469
1470 if (type & KVM_VM_S390_UCONTROL) {
1471 kvm->arch.gmap = NULL;
1472 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
1473 } else {
1474 if (sclp.hamax == U64_MAX)
1475 kvm->arch.mem_limit = TASK_MAX_SIZE;
1476 else
1477 kvm->arch.mem_limit = min_t(unsigned long, TASK_MAX_SIZE,
1478 sclp.hamax + 1);
1479 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
1480 if (!kvm->arch.gmap)
1481 goto out_err;
1482 kvm->arch.gmap->private = kvm;
1483 kvm->arch.gmap->pfault_enabled = 0;
1484 }
1485
1486 kvm->arch.css_support = 0;
1487 kvm->arch.use_irqchip = 0;
1488 kvm->arch.epoch = 0;
1489
1490 spin_lock_init(&kvm->arch.start_stop_lock);
1491 kvm_s390_vsie_init(kvm);
1492 KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
1493
1494 return 0;
1495 out_err:
1496 free_page((unsigned long)kvm->arch.sie_page2);
1497 debug_unregister(kvm->arch.dbf);
1498 sca_dispose(kvm);
1499 KVM_EVENT(3, "creation of vm failed: %d", rc);
1500 return rc;
1501 }
1502
1503 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1504 {
1505 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
1506 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
1507 kvm_s390_clear_local_irqs(vcpu);
1508 kvm_clear_async_pf_completion_queue(vcpu);
1509 if (!kvm_is_ucontrol(vcpu->kvm))
1510 sca_del_vcpu(vcpu);
1511
1512 if (kvm_is_ucontrol(vcpu->kvm))
1513 gmap_remove(vcpu->arch.gmap);
1514
1515 if (vcpu->kvm->arch.use_cmma)
1516 kvm_s390_vcpu_unsetup_cmma(vcpu);
1517 free_page((unsigned long)(vcpu->arch.sie_block));
1518
1519 kvm_vcpu_uninit(vcpu);
1520 kmem_cache_free(kvm_vcpu_cache, vcpu);
1521 }
1522
1523 static void kvm_free_vcpus(struct kvm *kvm)
1524 {
1525 unsigned int i;
1526 struct kvm_vcpu *vcpu;
1527
1528 kvm_for_each_vcpu(i, vcpu, kvm)
1529 kvm_arch_vcpu_destroy(vcpu);
1530
1531 mutex_lock(&kvm->lock);
1532 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
1533 kvm->vcpus[i] = NULL;
1534
1535 atomic_set(&kvm->online_vcpus, 0);
1536 mutex_unlock(&kvm->lock);
1537 }
1538
1539 void kvm_arch_destroy_vm(struct kvm *kvm)
1540 {
1541 kvm_free_vcpus(kvm);
1542 sca_dispose(kvm);
1543 debug_unregister(kvm->arch.dbf);
1544 free_page((unsigned long)kvm->arch.sie_page2);
1545 if (!kvm_is_ucontrol(kvm))
1546 gmap_remove(kvm->arch.gmap);
1547 kvm_s390_destroy_adapters(kvm);
1548 kvm_s390_clear_float_irqs(kvm);
1549 kvm_s390_vsie_destroy(kvm);
1550 KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
1551 }
1552
1553 /* Section: vcpu related */
1554 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
1555 {
1556 vcpu->arch.gmap = gmap_create(current->mm, -1UL);
1557 if (!vcpu->arch.gmap)
1558 return -ENOMEM;
1559 vcpu->arch.gmap->private = vcpu->kvm;
1560
1561 return 0;
1562 }
1563
1564 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
1565 {
1566 if (!kvm_s390_use_sca_entries())
1567 return;
1568 read_lock(&vcpu->kvm->arch.sca_lock);
1569 if (vcpu->kvm->arch.use_esca) {
1570 struct esca_block *sca = vcpu->kvm->arch.sca;
1571
1572 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
1573 sca->cpu[vcpu->vcpu_id].sda = 0;
1574 } else {
1575 struct bsca_block *sca = vcpu->kvm->arch.sca;
1576
1577 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
1578 sca->cpu[vcpu->vcpu_id].sda = 0;
1579 }
1580 read_unlock(&vcpu->kvm->arch.sca_lock);
1581 }
1582
1583 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
1584 {
1585 if (!kvm_s390_use_sca_entries()) {
1586 struct bsca_block *sca = vcpu->kvm->arch.sca;
1587
1588 /* we still need the basic sca for the ipte control */
1589 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1590 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
1591 }
1592 read_lock(&vcpu->kvm->arch.sca_lock);
1593 if (vcpu->kvm->arch.use_esca) {
1594 struct esca_block *sca = vcpu->kvm->arch.sca;
1595
1596 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
1597 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1598 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca & ~0x3fU;
1599 vcpu->arch.sie_block->ecb2 |= 0x04U;
1600 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
1601 } else {
1602 struct bsca_block *sca = vcpu->kvm->arch.sca;
1603
1604 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
1605 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1606 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
1607 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
1608 }
1609 read_unlock(&vcpu->kvm->arch.sca_lock);
1610 }
1611
1612 /* Basic SCA to Extended SCA data copy routines */
1613 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
1614 {
1615 d->sda = s->sda;
1616 d->sigp_ctrl.c = s->sigp_ctrl.c;
1617 d->sigp_ctrl.scn = s->sigp_ctrl.scn;
1618 }
1619
1620 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
1621 {
1622 int i;
1623
1624 d->ipte_control = s->ipte_control;
1625 d->mcn[0] = s->mcn;
1626 for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
1627 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
1628 }
1629
1630 static int sca_switch_to_extended(struct kvm *kvm)
1631 {
1632 struct bsca_block *old_sca = kvm->arch.sca;
1633 struct esca_block *new_sca;
1634 struct kvm_vcpu *vcpu;
1635 unsigned int vcpu_idx;
1636 u32 scaol, scaoh;
1637
1638 new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL|__GFP_ZERO);
1639 if (!new_sca)
1640 return -ENOMEM;
1641
1642 scaoh = (u32)((u64)(new_sca) >> 32);
1643 scaol = (u32)(u64)(new_sca) & ~0x3fU;
1644
1645 kvm_s390_vcpu_block_all(kvm);
1646 write_lock(&kvm->arch.sca_lock);
1647
1648 sca_copy_b_to_e(new_sca, old_sca);
1649
1650 kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
1651 vcpu->arch.sie_block->scaoh = scaoh;
1652 vcpu->arch.sie_block->scaol = scaol;
1653 vcpu->arch.sie_block->ecb2 |= 0x04U;
1654 }
1655 kvm->arch.sca = new_sca;
1656 kvm->arch.use_esca = 1;
1657
1658 write_unlock(&kvm->arch.sca_lock);
1659 kvm_s390_vcpu_unblock_all(kvm);
1660
1661 free_page((unsigned long)old_sca);
1662
1663 VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
1664 old_sca, kvm->arch.sca);
1665 return 0;
1666 }
1667
1668 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
1669 {
1670 int rc;
1671
1672 if (!kvm_s390_use_sca_entries()) {
1673 if (id < KVM_MAX_VCPUS)
1674 return true;
1675 return false;
1676 }
1677 if (id < KVM_S390_BSCA_CPU_SLOTS)
1678 return true;
1679 if (!sclp.has_esca || !sclp.has_64bscao)
1680 return false;
1681
1682 mutex_lock(&kvm->lock);
1683 rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
1684 mutex_unlock(&kvm->lock);
1685
1686 return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
1687 }
1688
1689 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1690 {
1691 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1692 kvm_clear_async_pf_completion_queue(vcpu);
1693 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
1694 KVM_SYNC_GPRS |
1695 KVM_SYNC_ACRS |
1696 KVM_SYNC_CRS |
1697 KVM_SYNC_ARCH0 |
1698 KVM_SYNC_PFAULT;
1699 kvm_s390_set_prefix(vcpu, 0);
1700 if (test_kvm_facility(vcpu->kvm, 64))
1701 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
1702 /* fprs can be synchronized via vrs, even if the guest has no vx. With
1703 * MACHINE_HAS_VX, (load|store)_fpu_regs() will work with vrs format.
1704 */
1705 if (MACHINE_HAS_VX)
1706 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
1707 else
1708 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
1709
1710 if (kvm_is_ucontrol(vcpu->kvm))
1711 return __kvm_ucontrol_vcpu_init(vcpu);
1712
1713 return 0;
1714 }
1715
1716 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1717 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1718 {
1719 WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
1720 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1721 vcpu->arch.cputm_start = get_tod_clock_fast();
1722 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1723 }
1724
1725 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1726 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1727 {
1728 WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
1729 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1730 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
1731 vcpu->arch.cputm_start = 0;
1732 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1733 }
1734
1735 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1736 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1737 {
1738 WARN_ON_ONCE(vcpu->arch.cputm_enabled);
1739 vcpu->arch.cputm_enabled = true;
1740 __start_cpu_timer_accounting(vcpu);
1741 }
1742
1743 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1744 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1745 {
1746 WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
1747 __stop_cpu_timer_accounting(vcpu);
1748 vcpu->arch.cputm_enabled = false;
1749 }
1750
1751 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1752 {
1753 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1754 __enable_cpu_timer_accounting(vcpu);
1755 preempt_enable();
1756 }
1757
1758 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1759 {
1760 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1761 __disable_cpu_timer_accounting(vcpu);
1762 preempt_enable();
1763 }
1764
1765 /* set the cpu timer - may only be called from the VCPU thread itself */
1766 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
1767 {
1768 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1769 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1770 if (vcpu->arch.cputm_enabled)
1771 vcpu->arch.cputm_start = get_tod_clock_fast();
1772 vcpu->arch.sie_block->cputm = cputm;
1773 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1774 preempt_enable();
1775 }
1776
1777 /* update and get the cpu timer - can also be called from other VCPU threads */
1778 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
1779 {
1780 unsigned int seq;
1781 __u64 value;
1782
1783 if (unlikely(!vcpu->arch.cputm_enabled))
1784 return vcpu->arch.sie_block->cputm;
1785
1786 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1787 do {
1788 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
1789 /*
1790 * If the writer would ever execute a read in the critical
1791 * section, e.g. in irq context, we have a deadlock.
1792 */
1793 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
1794 value = vcpu->arch.sie_block->cputm;
1795 /* if cputm_start is 0, accounting is being started/stopped */
1796 if (likely(vcpu->arch.cputm_start))
1797 value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
1798 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
1799 preempt_enable();
1800 return value;
1801 }
1802
1803 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1804 {
1805 /* Save host register state */
1806 save_fpu_regs();
1807 vcpu->arch.host_fpregs.fpc = current->thread.fpu.fpc;
1808 vcpu->arch.host_fpregs.regs = current->thread.fpu.regs;
1809
1810 if (MACHINE_HAS_VX)
1811 current->thread.fpu.regs = vcpu->run->s.regs.vrs;
1812 else
1813 current->thread.fpu.regs = vcpu->run->s.regs.fprs;
1814 current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
1815 if (test_fp_ctl(current->thread.fpu.fpc))
1816 /* User space provided an invalid FPC, let's clear it */
1817 current->thread.fpu.fpc = 0;
1818
1819 save_access_regs(vcpu->arch.host_acrs);
1820 restore_access_regs(vcpu->run->s.regs.acrs);
1821 gmap_enable(vcpu->arch.enabled_gmap);
1822 atomic_or(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1823 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
1824 __start_cpu_timer_accounting(vcpu);
1825 vcpu->cpu = cpu;
1826 }
1827
1828 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1829 {
1830 vcpu->cpu = -1;
1831 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
1832 __stop_cpu_timer_accounting(vcpu);
1833 atomic_andnot(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1834 vcpu->arch.enabled_gmap = gmap_get_enabled();
1835 gmap_disable(vcpu->arch.enabled_gmap);
1836
1837 /* Save guest register state */
1838 save_fpu_regs();
1839 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
1840
1841 /* Restore host register state */
1842 current->thread.fpu.fpc = vcpu->arch.host_fpregs.fpc;
1843 current->thread.fpu.regs = vcpu->arch.host_fpregs.regs;
1844
1845 save_access_regs(vcpu->run->s.regs.acrs);
1846 restore_access_regs(vcpu->arch.host_acrs);
1847 }
1848
1849 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
1850 {
1851 /* this equals initial cpu reset in pop, but we don't switch to ESA */
1852 vcpu->arch.sie_block->gpsw.mask = 0UL;
1853 vcpu->arch.sie_block->gpsw.addr = 0UL;
1854 kvm_s390_set_prefix(vcpu, 0);
1855 kvm_s390_set_cpu_timer(vcpu, 0);
1856 vcpu->arch.sie_block->ckc = 0UL;
1857 vcpu->arch.sie_block->todpr = 0;
1858 memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64));
1859 vcpu->arch.sie_block->gcr[0] = 0xE0UL;
1860 vcpu->arch.sie_block->gcr[14] = 0xC2000000UL;
1861 /* make sure the new fpc will be lazily loaded */
1862 save_fpu_regs();
1863 current->thread.fpu.fpc = 0;
1864 vcpu->arch.sie_block->gbea = 1;
1865 vcpu->arch.sie_block->pp = 0;
1866 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1867 kvm_clear_async_pf_completion_queue(vcpu);
1868 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
1869 kvm_s390_vcpu_stop(vcpu);
1870 kvm_s390_clear_local_irqs(vcpu);
1871 }
1872
1873 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1874 {
1875 mutex_lock(&vcpu->kvm->lock);
1876 preempt_disable();
1877 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
1878 preempt_enable();
1879 mutex_unlock(&vcpu->kvm->lock);
1880 if (!kvm_is_ucontrol(vcpu->kvm)) {
1881 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
1882 sca_add_vcpu(vcpu);
1883 }
1884 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
1885 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
1886 /* make vcpu_load load the right gmap on the first trigger */
1887 vcpu->arch.enabled_gmap = vcpu->arch.gmap;
1888 }
1889
1890 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
1891 {
1892 if (!test_kvm_facility(vcpu->kvm, 76))
1893 return;
1894
1895 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
1896
1897 if (vcpu->kvm->arch.crypto.aes_kw)
1898 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
1899 if (vcpu->kvm->arch.crypto.dea_kw)
1900 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
1901
1902 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
1903 }
1904
1905 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
1906 {
1907 free_page(vcpu->arch.sie_block->cbrlo);
1908 vcpu->arch.sie_block->cbrlo = 0;
1909 }
1910
1911 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
1912 {
1913 vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
1914 if (!vcpu->arch.sie_block->cbrlo)
1915 return -ENOMEM;
1916
1917 vcpu->arch.sie_block->ecb2 |= 0x80;
1918 vcpu->arch.sie_block->ecb2 &= ~0x08;
1919 return 0;
1920 }
1921
1922 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
1923 {
1924 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
1925
1926 vcpu->arch.sie_block->ibc = model->ibc;
1927 if (test_kvm_facility(vcpu->kvm, 7))
1928 vcpu->arch.sie_block->fac = (u32)(u64) model->fac_list;
1929 }
1930
1931 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1932 {
1933 int rc = 0;
1934
1935 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
1936 CPUSTAT_SM |
1937 CPUSTAT_STOPPED);
1938
1939 if (test_kvm_facility(vcpu->kvm, 78))
1940 atomic_or(CPUSTAT_GED2, &vcpu->arch.sie_block->cpuflags);
1941 else if (test_kvm_facility(vcpu->kvm, 8))
1942 atomic_or(CPUSTAT_GED, &vcpu->arch.sie_block->cpuflags);
1943
1944 kvm_s390_vcpu_setup_model(vcpu);
1945
1946 /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
1947 if (MACHINE_HAS_ESOP)
1948 vcpu->arch.sie_block->ecb |= 0x02;
1949 if (test_kvm_facility(vcpu->kvm, 9))
1950 vcpu->arch.sie_block->ecb |= 0x04;
1951 if (test_kvm_facility(vcpu->kvm, 73))
1952 vcpu->arch.sie_block->ecb |= 0x10;
1953
1954 if (test_kvm_facility(vcpu->kvm, 8) && sclp.has_pfmfi)
1955 vcpu->arch.sie_block->ecb2 |= 0x08;
1956 vcpu->arch.sie_block->eca = 0x1002000U;
1957 if (sclp.has_cei)
1958 vcpu->arch.sie_block->eca |= 0x80000000U;
1959 if (sclp.has_ib)
1960 vcpu->arch.sie_block->eca |= 0x40000000U;
1961 if (sclp.has_siif)
1962 vcpu->arch.sie_block->eca |= 1;
1963 if (sclp.has_sigpif)
1964 vcpu->arch.sie_block->eca |= 0x10000000U;
1965 if (test_kvm_facility(vcpu->kvm, 129)) {
1966 vcpu->arch.sie_block->eca |= 0x00020000;
1967 vcpu->arch.sie_block->ecd |= 0x20000000;
1968 }
1969 vcpu->arch.sie_block->riccbd = (unsigned long) &vcpu->run->s.regs.riccb;
1970 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
1971
1972 if (vcpu->kvm->arch.use_cmma) {
1973 rc = kvm_s390_vcpu_setup_cmma(vcpu);
1974 if (rc)
1975 return rc;
1976 }
1977 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1978 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
1979
1980 kvm_s390_vcpu_crypto_setup(vcpu);
1981
1982 return rc;
1983 }
1984
1985 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1986 unsigned int id)
1987 {
1988 struct kvm_vcpu *vcpu;
1989 struct sie_page *sie_page;
1990 int rc = -EINVAL;
1991
1992 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
1993 goto out;
1994
1995 rc = -ENOMEM;
1996
1997 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1998 if (!vcpu)
1999 goto out;
2000
2001 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
2002 if (!sie_page)
2003 goto out_free_cpu;
2004
2005 vcpu->arch.sie_block = &sie_page->sie_block;
2006 vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
2007
2008 /* the real guest size will always be smaller than msl */
2009 vcpu->arch.sie_block->mso = 0;
2010 vcpu->arch.sie_block->msl = sclp.hamax;
2011
2012 vcpu->arch.sie_block->icpua = id;
2013 spin_lock_init(&vcpu->arch.local_int.lock);
2014 vcpu->arch.local_int.float_int = &kvm->arch.float_int;
2015 vcpu->arch.local_int.wq = &vcpu->wq;
2016 vcpu->arch.local_int.cpuflags = &vcpu->arch.sie_block->cpuflags;
2017 seqcount_init(&vcpu->arch.cputm_seqcount);
2018
2019 rc = kvm_vcpu_init(vcpu, kvm, id);
2020 if (rc)
2021 goto out_free_sie_block;
2022 VM_EVENT(kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK", id, vcpu,
2023 vcpu->arch.sie_block);
2024 trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
2025
2026 return vcpu;
2027 out_free_sie_block:
2028 free_page((unsigned long)(vcpu->arch.sie_block));
2029 out_free_cpu:
2030 kmem_cache_free(kvm_vcpu_cache, vcpu);
2031 out:
2032 return ERR_PTR(rc);
2033 }
2034
2035 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
2036 {
2037 return kvm_s390_vcpu_has_irq(vcpu, 0);
2038 }
2039
2040 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
2041 {
2042 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2043 exit_sie(vcpu);
2044 }
2045
2046 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
2047 {
2048 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2049 }
2050
2051 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
2052 {
2053 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2054 exit_sie(vcpu);
2055 }
2056
2057 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
2058 {
2059 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2060 }
2061
2062 /*
2063 * Kick a guest cpu out of SIE and wait until SIE is not running.
2064 * If the CPU is not running (e.g. waiting as idle) the function will
2065 * return immediately. */
2066 void exit_sie(struct kvm_vcpu *vcpu)
2067 {
2068 atomic_or(CPUSTAT_STOP_INT, &vcpu->arch.sie_block->cpuflags);
2069 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
2070 cpu_relax();
2071 }
2072
2073 /* Kick a guest cpu out of SIE to process a request synchronously */
2074 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
2075 {
2076 kvm_make_request(req, vcpu);
2077 kvm_s390_vcpu_request(vcpu);
2078 }
2079
2080 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
2081 unsigned long end)
2082 {
2083 struct kvm *kvm = gmap->private;
2084 struct kvm_vcpu *vcpu;
2085 unsigned long prefix;
2086 int i;
2087
2088 if (gmap_is_shadow(gmap))
2089 return;
2090 if (start >= 1UL << 31)
2091 /* We are only interested in prefix pages */
2092 return;
2093 kvm_for_each_vcpu(i, vcpu, kvm) {
2094 /* match against both prefix pages */
2095 prefix = kvm_s390_get_prefix(vcpu);
2096 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
2097 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
2098 start, end);
2099 kvm_s390_sync_request(KVM_REQ_MMU_RELOAD, vcpu);
2100 }
2101 }
2102 }
2103
2104 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
2105 {
2106 /* kvm common code refers to this, but never calls it */
2107 BUG();
2108 return 0;
2109 }
2110
2111 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
2112 struct kvm_one_reg *reg)
2113 {
2114 int r = -EINVAL;
2115
2116 switch (reg->id) {
2117 case KVM_REG_S390_TODPR:
2118 r = put_user(vcpu->arch.sie_block->todpr,
2119 (u32 __user *)reg->addr);
2120 break;
2121 case KVM_REG_S390_EPOCHDIFF:
2122 r = put_user(vcpu->arch.sie_block->epoch,
2123 (u64 __user *)reg->addr);
2124 break;
2125 case KVM_REG_S390_CPU_TIMER:
2126 r = put_user(kvm_s390_get_cpu_timer(vcpu),
2127 (u64 __user *)reg->addr);
2128 break;
2129 case KVM_REG_S390_CLOCK_COMP:
2130 r = put_user(vcpu->arch.sie_block->ckc,
2131 (u64 __user *)reg->addr);
2132 break;
2133 case KVM_REG_S390_PFTOKEN:
2134 r = put_user(vcpu->arch.pfault_token,
2135 (u64 __user *)reg->addr);
2136 break;
2137 case KVM_REG_S390_PFCOMPARE:
2138 r = put_user(vcpu->arch.pfault_compare,
2139 (u64 __user *)reg->addr);
2140 break;
2141 case KVM_REG_S390_PFSELECT:
2142 r = put_user(vcpu->arch.pfault_select,
2143 (u64 __user *)reg->addr);
2144 break;
2145 case KVM_REG_S390_PP:
2146 r = put_user(vcpu->arch.sie_block->pp,
2147 (u64 __user *)reg->addr);
2148 break;
2149 case KVM_REG_S390_GBEA:
2150 r = put_user(vcpu->arch.sie_block->gbea,
2151 (u64 __user *)reg->addr);
2152 break;
2153 default:
2154 break;
2155 }
2156
2157 return r;
2158 }
2159
2160 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
2161 struct kvm_one_reg *reg)
2162 {
2163 int r = -EINVAL;
2164 __u64 val;
2165
2166 switch (reg->id) {
2167 case KVM_REG_S390_TODPR:
2168 r = get_user(vcpu->arch.sie_block->todpr,
2169 (u32 __user *)reg->addr);
2170 break;
2171 case KVM_REG_S390_EPOCHDIFF:
2172 r = get_user(vcpu->arch.sie_block->epoch,
2173 (u64 __user *)reg->addr);
2174 break;
2175 case KVM_REG_S390_CPU_TIMER:
2176 r = get_user(val, (u64 __user *)reg->addr);
2177 if (!r)
2178 kvm_s390_set_cpu_timer(vcpu, val);
2179 break;
2180 case KVM_REG_S390_CLOCK_COMP:
2181 r = get_user(vcpu->arch.sie_block->ckc,
2182 (u64 __user *)reg->addr);
2183 break;
2184 case KVM_REG_S390_PFTOKEN:
2185 r = get_user(vcpu->arch.pfault_token,
2186 (u64 __user *)reg->addr);
2187 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2188 kvm_clear_async_pf_completion_queue(vcpu);
2189 break;
2190 case KVM_REG_S390_PFCOMPARE:
2191 r = get_user(vcpu->arch.pfault_compare,
2192 (u64 __user *)reg->addr);
2193 break;
2194 case KVM_REG_S390_PFSELECT:
2195 r = get_user(vcpu->arch.pfault_select,
2196 (u64 __user *)reg->addr);
2197 break;
2198 case KVM_REG_S390_PP:
2199 r = get_user(vcpu->arch.sie_block->pp,
2200 (u64 __user *)reg->addr);
2201 break;
2202 case KVM_REG_S390_GBEA:
2203 r = get_user(vcpu->arch.sie_block->gbea,
2204 (u64 __user *)reg->addr);
2205 break;
2206 default:
2207 break;
2208 }
2209
2210 return r;
2211 }
2212
2213 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
2214 {
2215 kvm_s390_vcpu_initial_reset(vcpu);
2216 return 0;
2217 }
2218
2219 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2220 {
2221 memcpy(&vcpu->run->s.regs.gprs, &regs->gprs, sizeof(regs->gprs));
2222 return 0;
2223 }
2224
2225 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2226 {
2227 memcpy(&regs->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
2228 return 0;
2229 }
2230
2231 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2232 struct kvm_sregs *sregs)
2233 {
2234 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
2235 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
2236 restore_access_regs(vcpu->run->s.regs.acrs);
2237 return 0;
2238 }
2239
2240 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2241 struct kvm_sregs *sregs)
2242 {
2243 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
2244 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
2245 return 0;
2246 }
2247
2248 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2249 {
2250 /* make sure the new values will be lazily loaded */
2251 save_fpu_regs();
2252 if (test_fp_ctl(fpu->fpc))
2253 return -EINVAL;
2254 current->thread.fpu.fpc = fpu->fpc;
2255 if (MACHINE_HAS_VX)
2256 convert_fp_to_vx(current->thread.fpu.vxrs, (freg_t *)fpu->fprs);
2257 else
2258 memcpy(current->thread.fpu.fprs, &fpu->fprs, sizeof(fpu->fprs));
2259 return 0;
2260 }
2261
2262 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2263 {
2264 /* make sure we have the latest values */
2265 save_fpu_regs();
2266 if (MACHINE_HAS_VX)
2267 convert_vx_to_fp((freg_t *)fpu->fprs, current->thread.fpu.vxrs);
2268 else
2269 memcpy(fpu->fprs, current->thread.fpu.fprs, sizeof(fpu->fprs));
2270 fpu->fpc = current->thread.fpu.fpc;
2271 return 0;
2272 }
2273
2274 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
2275 {
2276 int rc = 0;
2277
2278 if (!is_vcpu_stopped(vcpu))
2279 rc = -EBUSY;
2280 else {
2281 vcpu->run->psw_mask = psw.mask;
2282 vcpu->run->psw_addr = psw.addr;
2283 }
2284 return rc;
2285 }
2286
2287 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2288 struct kvm_translation *tr)
2289 {
2290 return -EINVAL; /* not implemented yet */
2291 }
2292
2293 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
2294 KVM_GUESTDBG_USE_HW_BP | \
2295 KVM_GUESTDBG_ENABLE)
2296
2297 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
2298 struct kvm_guest_debug *dbg)
2299 {
2300 int rc = 0;
2301
2302 vcpu->guest_debug = 0;
2303 kvm_s390_clear_bp_data(vcpu);
2304
2305 if (dbg->control & ~VALID_GUESTDBG_FLAGS)
2306 return -EINVAL;
2307 if (!sclp.has_gpere)
2308 return -EINVAL;
2309
2310 if (dbg->control & KVM_GUESTDBG_ENABLE) {
2311 vcpu->guest_debug = dbg->control;
2312 /* enforce guest PER */
2313 atomic_or(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2314
2315 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
2316 rc = kvm_s390_import_bp_data(vcpu, dbg);
2317 } else {
2318 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2319 vcpu->arch.guestdbg.last_bp = 0;
2320 }
2321
2322 if (rc) {
2323 vcpu->guest_debug = 0;
2324 kvm_s390_clear_bp_data(vcpu);
2325 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2326 }
2327
2328 return rc;
2329 }
2330
2331 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2332 struct kvm_mp_state *mp_state)
2333 {
2334 /* CHECK_STOP and LOAD are not supported yet */
2335 return is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
2336 KVM_MP_STATE_OPERATING;
2337 }
2338
2339 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2340 struct kvm_mp_state *mp_state)
2341 {
2342 int rc = 0;
2343
2344 /* user space knows about this interface - let it control the state */
2345 vcpu->kvm->arch.user_cpu_state_ctrl = 1;
2346
2347 switch (mp_state->mp_state) {
2348 case KVM_MP_STATE_STOPPED:
2349 kvm_s390_vcpu_stop(vcpu);
2350 break;
2351 case KVM_MP_STATE_OPERATING:
2352 kvm_s390_vcpu_start(vcpu);
2353 break;
2354 case KVM_MP_STATE_LOAD:
2355 case KVM_MP_STATE_CHECK_STOP:
2356 /* fall through - CHECK_STOP and LOAD are not supported yet */
2357 default:
2358 rc = -ENXIO;
2359 }
2360
2361 return rc;
2362 }
2363
2364 static bool ibs_enabled(struct kvm_vcpu *vcpu)
2365 {
2366 return atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_IBS;
2367 }
2368
2369 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
2370 {
2371 retry:
2372 kvm_s390_vcpu_request_handled(vcpu);
2373 if (!vcpu->requests)
2374 return 0;
2375 /*
2376 * We use MMU_RELOAD just to re-arm the ipte notifier for the
2377 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
2378 * This ensures that the ipte instruction for this request has
2379 * already finished. We might race against a second unmapper that
2380 * wants to set the blocking bit. Lets just retry the request loop.
2381 */
2382 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
2383 int rc;
2384 rc = gmap_mprotect_notify(vcpu->arch.gmap,
2385 kvm_s390_get_prefix(vcpu),
2386 PAGE_SIZE * 2, PROT_WRITE);
2387 if (rc) {
2388 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
2389 return rc;
2390 }
2391 goto retry;
2392 }
2393
2394 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2395 vcpu->arch.sie_block->ihcpu = 0xffff;
2396 goto retry;
2397 }
2398
2399 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
2400 if (!ibs_enabled(vcpu)) {
2401 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
2402 atomic_or(CPUSTAT_IBS,
2403 &vcpu->arch.sie_block->cpuflags);
2404 }
2405 goto retry;
2406 }
2407
2408 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
2409 if (ibs_enabled(vcpu)) {
2410 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
2411 atomic_andnot(CPUSTAT_IBS,
2412 &vcpu->arch.sie_block->cpuflags);
2413 }
2414 goto retry;
2415 }
2416
2417 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
2418 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
2419 goto retry;
2420 }
2421
2422 /* nothing to do, just clear the request */
2423 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
2424
2425 return 0;
2426 }
2427
2428 void kvm_s390_set_tod_clock(struct kvm *kvm, u64 tod)
2429 {
2430 struct kvm_vcpu *vcpu;
2431 int i;
2432
2433 mutex_lock(&kvm->lock);
2434 preempt_disable();
2435 kvm->arch.epoch = tod - get_tod_clock();
2436 kvm_s390_vcpu_block_all(kvm);
2437 kvm_for_each_vcpu(i, vcpu, kvm)
2438 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
2439 kvm_s390_vcpu_unblock_all(kvm);
2440 preempt_enable();
2441 mutex_unlock(&kvm->lock);
2442 }
2443
2444 /**
2445 * kvm_arch_fault_in_page - fault-in guest page if necessary
2446 * @vcpu: The corresponding virtual cpu
2447 * @gpa: Guest physical address
2448 * @writable: Whether the page should be writable or not
2449 *
2450 * Make sure that a guest page has been faulted-in on the host.
2451 *
2452 * Return: Zero on success, negative error code otherwise.
2453 */
2454 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
2455 {
2456 return gmap_fault(vcpu->arch.gmap, gpa,
2457 writable ? FAULT_FLAG_WRITE : 0);
2458 }
2459
2460 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
2461 unsigned long token)
2462 {
2463 struct kvm_s390_interrupt inti;
2464 struct kvm_s390_irq irq;
2465
2466 if (start_token) {
2467 irq.u.ext.ext_params2 = token;
2468 irq.type = KVM_S390_INT_PFAULT_INIT;
2469 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
2470 } else {
2471 inti.type = KVM_S390_INT_PFAULT_DONE;
2472 inti.parm64 = token;
2473 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
2474 }
2475 }
2476
2477 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
2478 struct kvm_async_pf *work)
2479 {
2480 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
2481 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
2482 }
2483
2484 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
2485 struct kvm_async_pf *work)
2486 {
2487 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
2488 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
2489 }
2490
2491 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
2492 struct kvm_async_pf *work)
2493 {
2494 /* s390 will always inject the page directly */
2495 }
2496
2497 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
2498 {
2499 /*
2500 * s390 will always inject the page directly,
2501 * but we still want check_async_completion to cleanup
2502 */
2503 return true;
2504 }
2505
2506 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
2507 {
2508 hva_t hva;
2509 struct kvm_arch_async_pf arch;
2510 int rc;
2511
2512 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2513 return 0;
2514 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
2515 vcpu->arch.pfault_compare)
2516 return 0;
2517 if (psw_extint_disabled(vcpu))
2518 return 0;
2519 if (kvm_s390_vcpu_has_irq(vcpu, 0))
2520 return 0;
2521 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
2522 return 0;
2523 if (!vcpu->arch.gmap->pfault_enabled)
2524 return 0;
2525
2526 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
2527 hva += current->thread.gmap_addr & ~PAGE_MASK;
2528 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
2529 return 0;
2530
2531 rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
2532 return rc;
2533 }
2534
2535 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
2536 {
2537 int rc, cpuflags;
2538
2539 /*
2540 * On s390 notifications for arriving pages will be delivered directly
2541 * to the guest but the house keeping for completed pfaults is
2542 * handled outside the worker.
2543 */
2544 kvm_check_async_pf_completion(vcpu);
2545
2546 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
2547 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
2548
2549 if (need_resched())
2550 schedule();
2551
2552 if (test_cpu_flag(CIF_MCCK_PENDING))
2553 s390_handle_mcck();
2554
2555 if (!kvm_is_ucontrol(vcpu->kvm)) {
2556 rc = kvm_s390_deliver_pending_interrupts(vcpu);
2557 if (rc)
2558 return rc;
2559 }
2560
2561 rc = kvm_s390_handle_requests(vcpu);
2562 if (rc)
2563 return rc;
2564
2565 if (guestdbg_enabled(vcpu)) {
2566 kvm_s390_backup_guest_per_regs(vcpu);
2567 kvm_s390_patch_guest_per_regs(vcpu);
2568 }
2569
2570 vcpu->arch.sie_block->icptcode = 0;
2571 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
2572 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
2573 trace_kvm_s390_sie_enter(vcpu, cpuflags);
2574
2575 return 0;
2576 }
2577
2578 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
2579 {
2580 struct kvm_s390_pgm_info pgm_info = {
2581 .code = PGM_ADDRESSING,
2582 };
2583 u8 opcode, ilen;
2584 int rc;
2585
2586 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
2587 trace_kvm_s390_sie_fault(vcpu);
2588
2589 /*
2590 * We want to inject an addressing exception, which is defined as a
2591 * suppressing or terminating exception. However, since we came here
2592 * by a DAT access exception, the PSW still points to the faulting
2593 * instruction since DAT exceptions are nullifying. So we've got
2594 * to look up the current opcode to get the length of the instruction
2595 * to be able to forward the PSW.
2596 */
2597 rc = read_guest_instr(vcpu, &opcode, 1);
2598 ilen = insn_length(opcode);
2599 if (rc < 0) {
2600 return rc;
2601 } else if (rc) {
2602 /* Instruction-Fetching Exceptions - we can't detect the ilen.
2603 * Forward by arbitrary ilc, injection will take care of
2604 * nullification if necessary.
2605 */
2606 pgm_info = vcpu->arch.pgm;
2607 ilen = 4;
2608 }
2609 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
2610 kvm_s390_forward_psw(vcpu, ilen);
2611 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
2612 }
2613
2614 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
2615 {
2616 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
2617 vcpu->arch.sie_block->icptcode);
2618 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
2619
2620 if (guestdbg_enabled(vcpu))
2621 kvm_s390_restore_guest_per_regs(vcpu);
2622
2623 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
2624 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
2625
2626 if (vcpu->arch.sie_block->icptcode > 0) {
2627 int rc = kvm_handle_sie_intercept(vcpu);
2628
2629 if (rc != -EOPNOTSUPP)
2630 return rc;
2631 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
2632 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
2633 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
2634 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
2635 return -EREMOTE;
2636 } else if (exit_reason != -EFAULT) {
2637 vcpu->stat.exit_null++;
2638 return 0;
2639 } else if (kvm_is_ucontrol(vcpu->kvm)) {
2640 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
2641 vcpu->run->s390_ucontrol.trans_exc_code =
2642 current->thread.gmap_addr;
2643 vcpu->run->s390_ucontrol.pgm_code = 0x10;
2644 return -EREMOTE;
2645 } else if (current->thread.gmap_pfault) {
2646 trace_kvm_s390_major_guest_pfault(vcpu);
2647 current->thread.gmap_pfault = 0;
2648 if (kvm_arch_setup_async_pf(vcpu))
2649 return 0;
2650 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
2651 }
2652 return vcpu_post_run_fault_in_sie(vcpu);
2653 }
2654
2655 static int __vcpu_run(struct kvm_vcpu *vcpu)
2656 {
2657 int rc, exit_reason;
2658
2659 /*
2660 * We try to hold kvm->srcu during most of vcpu_run (except when run-
2661 * ning the guest), so that memslots (and other stuff) are protected
2662 */
2663 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2664
2665 do {
2666 rc = vcpu_pre_run(vcpu);
2667 if (rc)
2668 break;
2669
2670 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2671 /*
2672 * As PF_VCPU will be used in fault handler, between
2673 * guest_enter and guest_exit should be no uaccess.
2674 */
2675 local_irq_disable();
2676 guest_enter_irqoff();
2677 __disable_cpu_timer_accounting(vcpu);
2678 local_irq_enable();
2679 exit_reason = sie64a(vcpu->arch.sie_block,
2680 vcpu->run->s.regs.gprs);
2681 local_irq_disable();
2682 __enable_cpu_timer_accounting(vcpu);
2683 guest_exit_irqoff();
2684 local_irq_enable();
2685 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2686
2687 rc = vcpu_post_run(vcpu, exit_reason);
2688 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
2689
2690 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2691 return rc;
2692 }
2693
2694 static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2695 {
2696 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
2697 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
2698 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
2699 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
2700 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
2701 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
2702 /* some control register changes require a tlb flush */
2703 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2704 }
2705 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
2706 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
2707 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
2708 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
2709 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
2710 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
2711 }
2712 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
2713 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
2714 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
2715 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
2716 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2717 kvm_clear_async_pf_completion_queue(vcpu);
2718 }
2719 /*
2720 * If userspace sets the riccb (e.g. after migration) to a valid state,
2721 * we should enable RI here instead of doing the lazy enablement.
2722 */
2723 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) &&
2724 test_kvm_facility(vcpu->kvm, 64)) {
2725 struct runtime_instr_cb *riccb =
2726 (struct runtime_instr_cb *) &kvm_run->s.regs.riccb;
2727
2728 if (riccb->valid)
2729 vcpu->arch.sie_block->ecb3 |= 0x01;
2730 }
2731
2732 kvm_run->kvm_dirty_regs = 0;
2733 }
2734
2735 static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2736 {
2737 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
2738 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
2739 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
2740 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
2741 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
2742 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
2743 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
2744 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
2745 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
2746 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
2747 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
2748 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
2749 }
2750
2751 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2752 {
2753 int rc;
2754 sigset_t sigsaved;
2755
2756 if (guestdbg_exit_pending(vcpu)) {
2757 kvm_s390_prepare_debug_exit(vcpu);
2758 return 0;
2759 }
2760
2761 if (vcpu->sigset_active)
2762 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2763
2764 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
2765 kvm_s390_vcpu_start(vcpu);
2766 } else if (is_vcpu_stopped(vcpu)) {
2767 pr_err_ratelimited("can't run stopped vcpu %d\n",
2768 vcpu->vcpu_id);
2769 return -EINVAL;
2770 }
2771
2772 sync_regs(vcpu, kvm_run);
2773 enable_cpu_timer_accounting(vcpu);
2774
2775 might_fault();
2776 rc = __vcpu_run(vcpu);
2777
2778 if (signal_pending(current) && !rc) {
2779 kvm_run->exit_reason = KVM_EXIT_INTR;
2780 rc = -EINTR;
2781 }
2782
2783 if (guestdbg_exit_pending(vcpu) && !rc) {
2784 kvm_s390_prepare_debug_exit(vcpu);
2785 rc = 0;
2786 }
2787
2788 if (rc == -EREMOTE) {
2789 /* userspace support is needed, kvm_run has been prepared */
2790 rc = 0;
2791 }
2792
2793 disable_cpu_timer_accounting(vcpu);
2794 store_regs(vcpu, kvm_run);
2795
2796 if (vcpu->sigset_active)
2797 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2798
2799 vcpu->stat.exit_userspace++;
2800 return rc;
2801 }
2802
2803 /*
2804 * store status at address
2805 * we use have two special cases:
2806 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
2807 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
2808 */
2809 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
2810 {
2811 unsigned char archmode = 1;
2812 freg_t fprs[NUM_FPRS];
2813 unsigned int px;
2814 u64 clkcomp, cputm;
2815 int rc;
2816
2817 px = kvm_s390_get_prefix(vcpu);
2818 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
2819 if (write_guest_abs(vcpu, 163, &archmode, 1))
2820 return -EFAULT;
2821 gpa = 0;
2822 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
2823 if (write_guest_real(vcpu, 163, &archmode, 1))
2824 return -EFAULT;
2825 gpa = px;
2826 } else
2827 gpa -= __LC_FPREGS_SAVE_AREA;
2828
2829 /* manually convert vector registers if necessary */
2830 if (MACHINE_HAS_VX) {
2831 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
2832 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2833 fprs, 128);
2834 } else {
2835 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2836 vcpu->run->s.regs.fprs, 128);
2837 }
2838 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
2839 vcpu->run->s.regs.gprs, 128);
2840 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
2841 &vcpu->arch.sie_block->gpsw, 16);
2842 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
2843 &px, 4);
2844 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
2845 &vcpu->run->s.regs.fpc, 4);
2846 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
2847 &vcpu->arch.sie_block->todpr, 4);
2848 cputm = kvm_s390_get_cpu_timer(vcpu);
2849 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
2850 &cputm, 8);
2851 clkcomp = vcpu->arch.sie_block->ckc >> 8;
2852 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
2853 &clkcomp, 8);
2854 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
2855 &vcpu->run->s.regs.acrs, 64);
2856 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
2857 &vcpu->arch.sie_block->gcr, 128);
2858 return rc ? -EFAULT : 0;
2859 }
2860
2861 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
2862 {
2863 /*
2864 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
2865 * copying in vcpu load/put. Lets update our copies before we save
2866 * it into the save area
2867 */
2868 save_fpu_regs();
2869 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
2870 save_access_regs(vcpu->run->s.regs.acrs);
2871
2872 return kvm_s390_store_status_unloaded(vcpu, addr);
2873 }
2874
2875 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2876 {
2877 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
2878 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
2879 }
2880
2881 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
2882 {
2883 unsigned int i;
2884 struct kvm_vcpu *vcpu;
2885
2886 kvm_for_each_vcpu(i, vcpu, kvm) {
2887 __disable_ibs_on_vcpu(vcpu);
2888 }
2889 }
2890
2891 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2892 {
2893 if (!sclp.has_ibs)
2894 return;
2895 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
2896 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
2897 }
2898
2899 void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
2900 {
2901 int i, online_vcpus, started_vcpus = 0;
2902
2903 if (!is_vcpu_stopped(vcpu))
2904 return;
2905
2906 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
2907 /* Only one cpu at a time may enter/leave the STOPPED state. */
2908 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2909 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2910
2911 for (i = 0; i < online_vcpus; i++) {
2912 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i]))
2913 started_vcpus++;
2914 }
2915
2916 if (started_vcpus == 0) {
2917 /* we're the only active VCPU -> speed it up */
2918 __enable_ibs_on_vcpu(vcpu);
2919 } else if (started_vcpus == 1) {
2920 /*
2921 * As we are starting a second VCPU, we have to disable
2922 * the IBS facility on all VCPUs to remove potentially
2923 * oustanding ENABLE requests.
2924 */
2925 __disable_ibs_on_all_vcpus(vcpu->kvm);
2926 }
2927
2928 atomic_andnot(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2929 /*
2930 * Another VCPU might have used IBS while we were offline.
2931 * Let's play safe and flush the VCPU at startup.
2932 */
2933 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2934 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2935 return;
2936 }
2937
2938 void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
2939 {
2940 int i, online_vcpus, started_vcpus = 0;
2941 struct kvm_vcpu *started_vcpu = NULL;
2942
2943 if (is_vcpu_stopped(vcpu))
2944 return;
2945
2946 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
2947 /* Only one cpu at a time may enter/leave the STOPPED state. */
2948 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2949 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2950
2951 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
2952 kvm_s390_clear_stop_irq(vcpu);
2953
2954 atomic_or(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2955 __disable_ibs_on_vcpu(vcpu);
2956
2957 for (i = 0; i < online_vcpus; i++) {
2958 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) {
2959 started_vcpus++;
2960 started_vcpu = vcpu->kvm->vcpus[i];
2961 }
2962 }
2963
2964 if (started_vcpus == 1) {
2965 /*
2966 * As we only have one VCPU left, we want to enable the
2967 * IBS facility for that VCPU to speed it up.
2968 */
2969 __enable_ibs_on_vcpu(started_vcpu);
2970 }
2971
2972 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2973 return;
2974 }
2975
2976 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
2977 struct kvm_enable_cap *cap)
2978 {
2979 int r;
2980
2981 if (cap->flags)
2982 return -EINVAL;
2983
2984 switch (cap->cap) {
2985 case KVM_CAP_S390_CSS_SUPPORT:
2986 if (!vcpu->kvm->arch.css_support) {
2987 vcpu->kvm->arch.css_support = 1;
2988 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
2989 trace_kvm_s390_enable_css(vcpu->kvm);
2990 }
2991 r = 0;
2992 break;
2993 default:
2994 r = -EINVAL;
2995 break;
2996 }
2997 return r;
2998 }
2999
3000 static long kvm_s390_guest_mem_op(struct kvm_vcpu *vcpu,
3001 struct kvm_s390_mem_op *mop)
3002 {
3003 void __user *uaddr = (void __user *)mop->buf;
3004 void *tmpbuf = NULL;
3005 int r, srcu_idx;
3006 const u64 supported_flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION
3007 | KVM_S390_MEMOP_F_CHECK_ONLY;
3008
3009 if (mop->flags & ~supported_flags)
3010 return -EINVAL;
3011
3012 if (mop->size > MEM_OP_MAX_SIZE)
3013 return -E2BIG;
3014
3015 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
3016 tmpbuf = vmalloc(mop->size);
3017 if (!tmpbuf)
3018 return -ENOMEM;
3019 }
3020
3021 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3022
3023 switch (mop->op) {
3024 case KVM_S390_MEMOP_LOGICAL_READ:
3025 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3026 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3027 mop->size, GACC_FETCH);
3028 break;
3029 }
3030 r = read_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3031 if (r == 0) {
3032 if (copy_to_user(uaddr, tmpbuf, mop->size))
3033 r = -EFAULT;
3034 }
3035 break;
3036 case KVM_S390_MEMOP_LOGICAL_WRITE:
3037 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3038 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3039 mop->size, GACC_STORE);
3040 break;
3041 }
3042 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
3043 r = -EFAULT;
3044 break;
3045 }
3046 r = write_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3047 break;
3048 default:
3049 r = -EINVAL;
3050 }
3051
3052 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
3053
3054 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
3055 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
3056
3057 vfree(tmpbuf);
3058 return r;
3059 }
3060
3061 long kvm_arch_vcpu_ioctl(struct file *filp,
3062 unsigned int ioctl, unsigned long arg)
3063 {
3064 struct kvm_vcpu *vcpu = filp->private_data;
3065 void __user *argp = (void __user *)arg;
3066 int idx;
3067 long r;
3068
3069 switch (ioctl) {
3070 case KVM_S390_IRQ: {
3071 struct kvm_s390_irq s390irq;
3072
3073 r = -EFAULT;
3074 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
3075 break;
3076 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
3077 break;
3078 }
3079 case KVM_S390_INTERRUPT: {
3080 struct kvm_s390_interrupt s390int;
3081 struct kvm_s390_irq s390irq;
3082
3083 r = -EFAULT;
3084 if (copy_from_user(&s390int, argp, sizeof(s390int)))
3085 break;
3086 if (s390int_to_s390irq(&s390int, &s390irq))
3087 return -EINVAL;
3088 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
3089 break;
3090 }
3091 case KVM_S390_STORE_STATUS:
3092 idx = srcu_read_lock(&vcpu->kvm->srcu);
3093 r = kvm_s390_vcpu_store_status(vcpu, arg);
3094 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3095 break;
3096 case KVM_S390_SET_INITIAL_PSW: {
3097 psw_t psw;
3098
3099 r = -EFAULT;
3100 if (copy_from_user(&psw, argp, sizeof(psw)))
3101 break;
3102 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
3103 break;
3104 }
3105 case KVM_S390_INITIAL_RESET:
3106 r = kvm_arch_vcpu_ioctl_initial_reset(vcpu);
3107 break;
3108 case KVM_SET_ONE_REG:
3109 case KVM_GET_ONE_REG: {
3110 struct kvm_one_reg reg;
3111 r = -EFAULT;
3112 if (copy_from_user(&reg, argp, sizeof(reg)))
3113 break;
3114 if (ioctl == KVM_SET_ONE_REG)
3115 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, &reg);
3116 else
3117 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, &reg);
3118 break;
3119 }
3120 #ifdef CONFIG_KVM_S390_UCONTROL
3121 case KVM_S390_UCAS_MAP: {
3122 struct kvm_s390_ucas_mapping ucasmap;
3123
3124 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3125 r = -EFAULT;
3126 break;
3127 }
3128
3129 if (!kvm_is_ucontrol(vcpu->kvm)) {
3130 r = -EINVAL;
3131 break;
3132 }
3133
3134 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
3135 ucasmap.vcpu_addr, ucasmap.length);
3136 break;
3137 }
3138 case KVM_S390_UCAS_UNMAP: {
3139 struct kvm_s390_ucas_mapping ucasmap;
3140
3141 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3142 r = -EFAULT;
3143 break;
3144 }
3145
3146 if (!kvm_is_ucontrol(vcpu->kvm)) {
3147 r = -EINVAL;
3148 break;
3149 }
3150
3151 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
3152 ucasmap.length);
3153 break;
3154 }
3155 #endif
3156 case KVM_S390_VCPU_FAULT: {
3157 r = gmap_fault(vcpu->arch.gmap, arg, 0);
3158 break;
3159 }
3160 case KVM_ENABLE_CAP:
3161 {
3162 struct kvm_enable_cap cap;
3163 r = -EFAULT;
3164 if (copy_from_user(&cap, argp, sizeof(cap)))
3165 break;
3166 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3167 break;
3168 }
3169 case KVM_S390_MEM_OP: {
3170 struct kvm_s390_mem_op mem_op;
3171
3172 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
3173 r = kvm_s390_guest_mem_op(vcpu, &mem_op);
3174 else
3175 r = -EFAULT;
3176 break;
3177 }
3178 case KVM_S390_SET_IRQ_STATE: {
3179 struct kvm_s390_irq_state irq_state;
3180
3181 r = -EFAULT;
3182 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3183 break;
3184 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
3185 irq_state.len == 0 ||
3186 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
3187 r = -EINVAL;
3188 break;
3189 }
3190 r = kvm_s390_set_irq_state(vcpu,
3191 (void __user *) irq_state.buf,
3192 irq_state.len);
3193 break;
3194 }
3195 case KVM_S390_GET_IRQ_STATE: {
3196 struct kvm_s390_irq_state irq_state;
3197
3198 r = -EFAULT;
3199 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3200 break;
3201 if (irq_state.len == 0) {
3202 r = -EINVAL;
3203 break;
3204 }
3205 r = kvm_s390_get_irq_state(vcpu,
3206 (__u8 __user *) irq_state.buf,
3207 irq_state.len);
3208 break;
3209 }
3210 default:
3211 r = -ENOTTY;
3212 }
3213 return r;
3214 }
3215
3216 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3217 {
3218 #ifdef CONFIG_KVM_S390_UCONTROL
3219 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
3220 && (kvm_is_ucontrol(vcpu->kvm))) {
3221 vmf->page = virt_to_page(vcpu->arch.sie_block);
3222 get_page(vmf->page);
3223 return 0;
3224 }
3225 #endif
3226 return VM_FAULT_SIGBUS;
3227 }
3228
3229 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
3230 unsigned long npages)
3231 {
3232 return 0;
3233 }
3234
3235 /* Section: memory related */
3236 int kvm_arch_prepare_memory_region(struct kvm *kvm,
3237 struct kvm_memory_slot *memslot,
3238 const struct kvm_userspace_memory_region *mem,
3239 enum kvm_mr_change change)
3240 {
3241 /* A few sanity checks. We can have memory slots which have to be
3242 located/ended at a segment boundary (1MB). The memory in userland is
3243 ok to be fragmented into various different vmas. It is okay to mmap()
3244 and munmap() stuff in this slot after doing this call at any time */
3245
3246 if (mem->userspace_addr & 0xffffful)
3247 return -EINVAL;
3248
3249 if (mem->memory_size & 0xffffful)
3250 return -EINVAL;
3251
3252 if (mem->guest_phys_addr + mem->memory_size > kvm->arch.mem_limit)
3253 return -EINVAL;
3254
3255 return 0;
3256 }
3257
3258 void kvm_arch_commit_memory_region(struct kvm *kvm,
3259 const struct kvm_userspace_memory_region *mem,
3260 const struct kvm_memory_slot *old,
3261 const struct kvm_memory_slot *new,
3262 enum kvm_mr_change change)
3263 {
3264 int rc;
3265
3266 /* If the basics of the memslot do not change, we do not want
3267 * to update the gmap. Every update causes several unnecessary
3268 * segment translation exceptions. This is usually handled just
3269 * fine by the normal fault handler + gmap, but it will also
3270 * cause faults on the prefix page of running guest CPUs.
3271 */
3272 if (old->userspace_addr == mem->userspace_addr &&
3273 old->base_gfn * PAGE_SIZE == mem->guest_phys_addr &&
3274 old->npages * PAGE_SIZE == mem->memory_size)
3275 return;
3276
3277 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr,
3278 mem->guest_phys_addr, mem->memory_size);
3279 if (rc)
3280 pr_warn("failed to commit memory region\n");
3281 return;
3282 }
3283
3284 static inline unsigned long nonhyp_mask(int i)
3285 {
3286 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
3287
3288 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
3289 }
3290
3291 void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu)
3292 {
3293 vcpu->valid_wakeup = false;
3294 }
3295
3296 static int __init kvm_s390_init(void)
3297 {
3298 int i;
3299
3300 if (!sclp.has_sief2) {
3301 pr_info("SIE not available\n");
3302 return -ENODEV;
3303 }
3304
3305 for (i = 0; i < 16; i++)
3306 kvm_s390_fac_list_mask[i] |=
3307 S390_lowcore.stfle_fac_list[i] & nonhyp_mask(i);
3308
3309 return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
3310 }
3311
3312 static void __exit kvm_s390_exit(void)
3313 {
3314 kvm_exit();
3315 }
3316
3317 module_init(kvm_s390_init);
3318 module_exit(kvm_s390_exit);
3319
3320 /*
3321 * Enable autoloading of the kvm module.
3322 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
3323 * since x86 takes a different approach.
3324 */
3325 #include <linux/miscdevice.h>
3326 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3327 MODULE_ALIAS("devname:kvm");
Linux kernel - Deliverables
This page took 0.104103 seconds and 5 git commands to generate.