?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.11-rc5
[linux/fpc-iii.git] / arch / s390 / kvm / kvm-s390.c
blobfd6cd05bb6a7c761b321a870c06cd79bc2beda47
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/moduleparam.h>
27 #include <linux/random.h>
28 #include <linux/slab.h>
29 #include <linux/timer.h>
30 #include <linux/vmalloc.h>
31 #include <linux/bitmap.h>
32 #include <linux/sched/signal.h>
33
34 #include <asm/asm-offsets.h>
35 #include <asm/lowcore.h>
36 #include <asm/stp.h>
37 #include <asm/pgtable.h>
38 #include <asm/gmap.h>
39 #include <asm/nmi.h>
40 #include <asm/switch_to.h>
41 #include <asm/isc.h>
42 #include <asm/sclp.h>
43 #include <asm/cpacf.h>
44 #include <asm/timex.h>
45 #include "kvm-s390.h"
46 #include "gaccess.h"
47
48 #define KMSG_COMPONENT "kvm-s390"
49 #undef pr_fmt
50 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
51
52 #define CREATE_TRACE_POINTS
53 #include "trace.h"
54 #include "trace-s390.h"
55
56 #define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */
57 #define LOCAL_IRQS 32
58 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
59 (KVM_MAX_VCPUS + LOCAL_IRQS))
60
61 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
62
63 struct kvm_stats_debugfs_item debugfs_entries[] = {
64 { "userspace_handled", VCPU_STAT(exit_userspace) },
65 { "exit_null", VCPU_STAT(exit_null) },
66 { "exit_validity", VCPU_STAT(exit_validity) },
67 { "exit_stop_request", VCPU_STAT(exit_stop_request) },
68 { "exit_external_request", VCPU_STAT(exit_external_request) },
69 { "exit_external_interrupt", VCPU_STAT(exit_external_interrupt) },
70 { "exit_instruction", VCPU_STAT(exit_instruction) },
71 { "exit_pei", VCPU_STAT(exit_pei) },
72 { "exit_program_interruption", VCPU_STAT(exit_program_interruption) },
73 { "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program) },
74 { "exit_operation_exception", VCPU_STAT(exit_operation_exception) },
75 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
76 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
77 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
78 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
79 { "instruction_lctlg", VCPU_STAT(instruction_lctlg) },
80 { "instruction_lctl", VCPU_STAT(instruction_lctl) },
81 { "instruction_stctl", VCPU_STAT(instruction_stctl) },
82 { "instruction_stctg", VCPU_STAT(instruction_stctg) },
83 { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal) },
84 { "deliver_external_call", VCPU_STAT(deliver_external_call) },
85 { "deliver_service_signal", VCPU_STAT(deliver_service_signal) },
86 { "deliver_virtio_interrupt", VCPU_STAT(deliver_virtio_interrupt) },
87 { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal) },
88 { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal) },
89 { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal) },
90 { "deliver_program_interruption", VCPU_STAT(deliver_program_int) },
91 { "exit_wait_state", VCPU_STAT(exit_wait_state) },
92 { "instruction_pfmf", VCPU_STAT(instruction_pfmf) },
93 { "instruction_stidp", VCPU_STAT(instruction_stidp) },
94 { "instruction_spx", VCPU_STAT(instruction_spx) },
95 { "instruction_stpx", VCPU_STAT(instruction_stpx) },
96 { "instruction_stap", VCPU_STAT(instruction_stap) },
97 { "instruction_storage_key", VCPU_STAT(instruction_storage_key) },
98 { "instruction_ipte_interlock", VCPU_STAT(instruction_ipte_interlock) },
99 { "instruction_stsch", VCPU_STAT(instruction_stsch) },
100 { "instruction_chsc", VCPU_STAT(instruction_chsc) },
101 { "instruction_essa", VCPU_STAT(instruction_essa) },
102 { "instruction_stsi", VCPU_STAT(instruction_stsi) },
103 { "instruction_stfl", VCPU_STAT(instruction_stfl) },
104 { "instruction_tprot", VCPU_STAT(instruction_tprot) },
105 { "instruction_sthyi", VCPU_STAT(instruction_sthyi) },
106 { "instruction_sie", VCPU_STAT(instruction_sie) },
107 { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense) },
108 { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running) },
109 { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call) },
110 { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency) },
111 { "instruction_sigp_cond_emergency", VCPU_STAT(instruction_sigp_cond_emergency) },
112 { "instruction_sigp_start", VCPU_STAT(instruction_sigp_start) },
113 { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop) },
114 { "instruction_sigp_stop_store_status", VCPU_STAT(instruction_sigp_stop_store_status) },
115 { "instruction_sigp_store_status", VCPU_STAT(instruction_sigp_store_status) },
116 { "instruction_sigp_store_adtl_status", VCPU_STAT(instruction_sigp_store_adtl_status) },
117 { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch) },
118 { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix) },
119 { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart) },
120 { "instruction_sigp_cpu_reset", VCPU_STAT(instruction_sigp_cpu_reset) },
121 { "instruction_sigp_init_cpu_reset", VCPU_STAT(instruction_sigp_init_cpu_reset) },
122 { "instruction_sigp_unknown", VCPU_STAT(instruction_sigp_unknown) },
123 { "diagnose_10", VCPU_STAT(diagnose_10) },
124 { "diagnose_44", VCPU_STAT(diagnose_44) },
125 { "diagnose_9c", VCPU_STAT(diagnose_9c) },
126 { "diagnose_258", VCPU_STAT(diagnose_258) },
127 { "diagnose_308", VCPU_STAT(diagnose_308) },
128 { "diagnose_500", VCPU_STAT(diagnose_500) },
129 { NULL }
130 };
131
132 /* allow nested virtualization in KVM (if enabled by user space) */
133 static int nested;
134 module_param(nested, int, S_IRUGO);
135 MODULE_PARM_DESC(nested, "Nested virtualization support");
136
137 /* upper facilities limit for kvm */
138 unsigned long kvm_s390_fac_list_mask[16] = { FACILITIES_KVM };
139
140 unsigned long kvm_s390_fac_list_mask_size(void)
141 {
142 BUILD_BUG_ON(ARRAY_SIZE(kvm_s390_fac_list_mask) > S390_ARCH_FAC_MASK_SIZE_U64);
143 return ARRAY_SIZE(kvm_s390_fac_list_mask);
144 }
145
146 /* available cpu features supported by kvm */
147 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
148 /* available subfunctions indicated via query / "test bit" */
149 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
150
151 static struct gmap_notifier gmap_notifier;
152 static struct gmap_notifier vsie_gmap_notifier;
153 debug_info_t *kvm_s390_dbf;
154
155 /* Section: not file related */
156 int kvm_arch_hardware_enable(void)
157 {
158 /* every s390 is virtualization enabled ;-) */
159 return 0;
160 }
161
162 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
163 unsigned long end);
164
165 /*
166 * This callback is executed during stop_machine(). All CPUs are therefore
167 * temporarily stopped. In order not to change guest behavior, we have to
168 * disable preemption whenever we touch the epoch of kvm and the VCPUs,
169 * so a CPU won't be stopped while calculating with the epoch.
170 */
171 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
172 void *v)
173 {
174 struct kvm *kvm;
175 struct kvm_vcpu *vcpu;
176 int i;
177 unsigned long long *delta = v;
178
179 list_for_each_entry(kvm, &vm_list, vm_list) {
180 kvm->arch.epoch -= *delta;
181 kvm_for_each_vcpu(i, vcpu, kvm) {
182 vcpu->arch.sie_block->epoch -= *delta;
183 if (vcpu->arch.cputm_enabled)
184 vcpu->arch.cputm_start += *delta;
185 if (vcpu->arch.vsie_block)
186 vcpu->arch.vsie_block->epoch -= *delta;
187 }
188 }
189 return NOTIFY_OK;
190 }
191
192 static struct notifier_block kvm_clock_notifier = {
193 .notifier_call = kvm_clock_sync,
194 };
195
196 int kvm_arch_hardware_setup(void)
197 {
198 gmap_notifier.notifier_call = kvm_gmap_notifier;
199 gmap_register_pte_notifier(&gmap_notifier);
200 vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
201 gmap_register_pte_notifier(&vsie_gmap_notifier);
202 atomic_notifier_chain_register(&s390_epoch_delta_notifier,
203 &kvm_clock_notifier);
204 return 0;
205 }
206
207 void kvm_arch_hardware_unsetup(void)
208 {
209 gmap_unregister_pte_notifier(&gmap_notifier);
210 gmap_unregister_pte_notifier(&vsie_gmap_notifier);
211 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
212 &kvm_clock_notifier);
213 }
214
215 static void allow_cpu_feat(unsigned long nr)
216 {
217 set_bit_inv(nr, kvm_s390_available_cpu_feat);
218 }
219
220 static inline int plo_test_bit(unsigned char nr)
221 {
222 register unsigned long r0 asm("0") = (unsigned long) nr | 0x100;
223 int cc;
224
225 asm volatile(
226 /* Parameter registers are ignored for "test bit" */
227 " plo 0,0,0,0(0)\n"
228 " ipm %0\n"
229 " srl %0,28\n"
230 : "=d" (cc)
231 : "d" (r0)
232 : "cc");
233 return cc == 0;
234 }
235
236 static void kvm_s390_cpu_feat_init(void)
237 {
238 int i;
239
240 for (i = 0; i < 256; ++i) {
241 if (plo_test_bit(i))
242 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
243 }
244
245 if (test_facility(28)) /* TOD-clock steering */
246 ptff(kvm_s390_available_subfunc.ptff,
247 sizeof(kvm_s390_available_subfunc.ptff),
248 PTFF_QAF);
249
250 if (test_facility(17)) { /* MSA */
251 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
252 kvm_s390_available_subfunc.kmac);
253 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
254 kvm_s390_available_subfunc.kmc);
255 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
256 kvm_s390_available_subfunc.km);
257 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
258 kvm_s390_available_subfunc.kimd);
259 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
260 kvm_s390_available_subfunc.klmd);
261 }
262 if (test_facility(76)) /* MSA3 */
263 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
264 kvm_s390_available_subfunc.pckmo);
265 if (test_facility(77)) { /* MSA4 */
266 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
267 kvm_s390_available_subfunc.kmctr);
268 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
269 kvm_s390_available_subfunc.kmf);
270 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
271 kvm_s390_available_subfunc.kmo);
272 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
273 kvm_s390_available_subfunc.pcc);
274 }
275 if (test_facility(57)) /* MSA5 */
276 __cpacf_query(CPACF_PPNO, (cpacf_mask_t *)
277 kvm_s390_available_subfunc.ppno);
278
279 if (MACHINE_HAS_ESOP)
280 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
281 /*
282 * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
283 * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
284 */
285 if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
286 !test_facility(3) || !nested)
287 return;
288 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
289 if (sclp.has_64bscao)
290 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
291 if (sclp.has_siif)
292 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
293 if (sclp.has_gpere)
294 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
295 if (sclp.has_gsls)
296 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
297 if (sclp.has_ib)
298 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
299 if (sclp.has_cei)
300 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
301 if (sclp.has_ibs)
302 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
303 /*
304 * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
305 * all skey handling functions read/set the skey from the PGSTE
306 * instead of the real storage key.
307 *
308 * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
309 * pages being detected as preserved although they are resident.
310 *
311 * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
312 * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
313 *
314 * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
315 * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
316 * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
317 *
318 * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
319 * cannot easily shadow the SCA because of the ipte lock.
320 */
321 }
322
323 int kvm_arch_init(void *opaque)
324 {
325 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
326 if (!kvm_s390_dbf)
327 return -ENOMEM;
328
329 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view)) {
330 debug_unregister(kvm_s390_dbf);
331 return -ENOMEM;
332 }
333
334 kvm_s390_cpu_feat_init();
335
336 /* Register floating interrupt controller interface. */
337 return kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
338 }
339
340 void kvm_arch_exit(void)
341 {
342 debug_unregister(kvm_s390_dbf);
343 }
344
345 /* Section: device related */
346 long kvm_arch_dev_ioctl(struct file *filp,
347 unsigned int ioctl, unsigned long arg)
348 {
349 if (ioctl == KVM_S390_ENABLE_SIE)
350 return s390_enable_sie();
351 return -EINVAL;
352 }
353
354 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
355 {
356 int r;
357
358 switch (ext) {
359 case KVM_CAP_S390_PSW:
360 case KVM_CAP_S390_GMAP:
361 case KVM_CAP_SYNC_MMU:
362 #ifdef CONFIG_KVM_S390_UCONTROL
363 case KVM_CAP_S390_UCONTROL:
364 #endif
365 case KVM_CAP_ASYNC_PF:
366 case KVM_CAP_SYNC_REGS:
367 case KVM_CAP_ONE_REG:
368 case KVM_CAP_ENABLE_CAP:
369 case KVM_CAP_S390_CSS_SUPPORT:
370 case KVM_CAP_IOEVENTFD:
371 case KVM_CAP_DEVICE_CTRL:
372 case KVM_CAP_ENABLE_CAP_VM:
373 case KVM_CAP_S390_IRQCHIP:
374 case KVM_CAP_VM_ATTRIBUTES:
375 case KVM_CAP_MP_STATE:
376 case KVM_CAP_IMMEDIATE_EXIT:
377 case KVM_CAP_S390_INJECT_IRQ:
378 case KVM_CAP_S390_USER_SIGP:
379 case KVM_CAP_S390_USER_STSI:
380 case KVM_CAP_S390_SKEYS:
381 case KVM_CAP_S390_IRQ_STATE:
382 case KVM_CAP_S390_USER_INSTR0:
383 r = 1;
384 break;
385 case KVM_CAP_S390_MEM_OP:
386 r = MEM_OP_MAX_SIZE;
387 break;
388 case KVM_CAP_NR_VCPUS:
389 case KVM_CAP_MAX_VCPUS:
390 r = KVM_S390_BSCA_CPU_SLOTS;
391 if (!kvm_s390_use_sca_entries())
392 r = KVM_MAX_VCPUS;
393 else if (sclp.has_esca && sclp.has_64bscao)
394 r = KVM_S390_ESCA_CPU_SLOTS;
395 break;
396 case KVM_CAP_NR_MEMSLOTS:
397 r = KVM_USER_MEM_SLOTS;
398 break;
399 case KVM_CAP_S390_COW:
400 r = MACHINE_HAS_ESOP;
401 break;
402 case KVM_CAP_S390_VECTOR_REGISTERS:
403 r = MACHINE_HAS_VX;
404 break;
405 case KVM_CAP_S390_RI:
406 r = test_facility(64);
407 break;
408 default:
409 r = 0;
410 }
411 return r;
412 }
413
414 static void kvm_s390_sync_dirty_log(struct kvm *kvm,
415 struct kvm_memory_slot *memslot)
416 {
417 gfn_t cur_gfn, last_gfn;
418 unsigned long address;
419 struct gmap *gmap = kvm->arch.gmap;
420
421 /* Loop over all guest pages */
422 last_gfn = memslot->base_gfn + memslot->npages;
423 for (cur_gfn = memslot->base_gfn; cur_gfn <= last_gfn; cur_gfn++) {
424 address = gfn_to_hva_memslot(memslot, cur_gfn);
425
426 if (test_and_clear_guest_dirty(gmap->mm, address))
427 mark_page_dirty(kvm, cur_gfn);
428 if (fatal_signal_pending(current))
429 return;
430 cond_resched();
431 }
432 }
433
434 /* Section: vm related */
435 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
436
437 /*
438 * Get (and clear) the dirty memory log for a memory slot.
439 */
440 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
441 struct kvm_dirty_log *log)
442 {
443 int r;
444 unsigned long n;
445 struct kvm_memslots *slots;
446 struct kvm_memory_slot *memslot;
447 int is_dirty = 0;
448
449 if (kvm_is_ucontrol(kvm))
450 return -EINVAL;
451
452 mutex_lock(&kvm->slots_lock);
453
454 r = -EINVAL;
455 if (log->slot >= KVM_USER_MEM_SLOTS)
456 goto out;
457
458 slots = kvm_memslots(kvm);
459 memslot = id_to_memslot(slots, log->slot);
460 r = -ENOENT;
461 if (!memslot->dirty_bitmap)
462 goto out;
463
464 kvm_s390_sync_dirty_log(kvm, memslot);
465 r = kvm_get_dirty_log(kvm, log, &is_dirty);
466 if (r)
467 goto out;
468
469 /* Clear the dirty log */
470 if (is_dirty) {
471 n = kvm_dirty_bitmap_bytes(memslot);
472 memset(memslot->dirty_bitmap, 0, n);
473 }
474 r = 0;
475 out:
476 mutex_unlock(&kvm->slots_lock);
477 return r;
478 }
479
480 static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
481 {
482 unsigned int i;
483 struct kvm_vcpu *vcpu;
484
485 kvm_for_each_vcpu(i, vcpu, kvm) {
486 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
487 }
488 }
489
490 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
491 {
492 int r;
493
494 if (cap->flags)
495 return -EINVAL;
496
497 switch (cap->cap) {
498 case KVM_CAP_S390_IRQCHIP:
499 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
500 kvm->arch.use_irqchip = 1;
501 r = 0;
502 break;
503 case KVM_CAP_S390_USER_SIGP:
504 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
505 kvm->arch.user_sigp = 1;
506 r = 0;
507 break;
508 case KVM_CAP_S390_VECTOR_REGISTERS:
509 mutex_lock(&kvm->lock);
510 if (kvm->created_vcpus) {
511 r = -EBUSY;
512 } else if (MACHINE_HAS_VX) {
513 set_kvm_facility(kvm->arch.model.fac_mask, 129);
514 set_kvm_facility(kvm->arch.model.fac_list, 129);
515 if (test_facility(134)) {
516 set_kvm_facility(kvm->arch.model.fac_mask, 134);
517 set_kvm_facility(kvm->arch.model.fac_list, 134);
518 }
519 if (test_facility(135)) {
520 set_kvm_facility(kvm->arch.model.fac_mask, 135);
521 set_kvm_facility(kvm->arch.model.fac_list, 135);
522 }
523 r = 0;
524 } else
525 r = -EINVAL;
526 mutex_unlock(&kvm->lock);
527 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
528 r ? "(not available)" : "(success)");
529 break;
530 case KVM_CAP_S390_RI:
531 r = -EINVAL;
532 mutex_lock(&kvm->lock);
533 if (kvm->created_vcpus) {
534 r = -EBUSY;
535 } else if (test_facility(64)) {
536 set_kvm_facility(kvm->arch.model.fac_mask, 64);
537 set_kvm_facility(kvm->arch.model.fac_list, 64);
538 r = 0;
539 }
540 mutex_unlock(&kvm->lock);
541 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
542 r ? "(not available)" : "(success)");
543 break;
544 case KVM_CAP_S390_USER_STSI:
545 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
546 kvm->arch.user_stsi = 1;
547 r = 0;
548 break;
549 case KVM_CAP_S390_USER_INSTR0:
550 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
551 kvm->arch.user_instr0 = 1;
552 icpt_operexc_on_all_vcpus(kvm);
553 r = 0;
554 break;
555 default:
556 r = -EINVAL;
557 break;
558 }
559 return r;
560 }
561
562 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
563 {
564 int ret;
565
566 switch (attr->attr) {
567 case KVM_S390_VM_MEM_LIMIT_SIZE:
568 ret = 0;
569 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
570 kvm->arch.mem_limit);
571 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
572 ret = -EFAULT;
573 break;
574 default:
575 ret = -ENXIO;
576 break;
577 }
578 return ret;
579 }
580
581 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
582 {
583 int ret;
584 unsigned int idx;
585 switch (attr->attr) {
586 case KVM_S390_VM_MEM_ENABLE_CMMA:
587 ret = -ENXIO;
588 if (!sclp.has_cmma)
589 break;
590
591 ret = -EBUSY;
592 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
593 mutex_lock(&kvm->lock);
594 if (!kvm->created_vcpus) {
595 kvm->arch.use_cmma = 1;
596 ret = 0;
597 }
598 mutex_unlock(&kvm->lock);
599 break;
600 case KVM_S390_VM_MEM_CLR_CMMA:
601 ret = -ENXIO;
602 if (!sclp.has_cmma)
603 break;
604 ret = -EINVAL;
605 if (!kvm->arch.use_cmma)
606 break;
607
608 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
609 mutex_lock(&kvm->lock);
610 idx = srcu_read_lock(&kvm->srcu);
611 s390_reset_cmma(kvm->arch.gmap->mm);
612 srcu_read_unlock(&kvm->srcu, idx);
613 mutex_unlock(&kvm->lock);
614 ret = 0;
615 break;
616 case KVM_S390_VM_MEM_LIMIT_SIZE: {
617 unsigned long new_limit;
618
619 if (kvm_is_ucontrol(kvm))
620 return -EINVAL;
621
622 if (get_user(new_limit, (u64 __user *)attr->addr))
623 return -EFAULT;
624
625 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
626 new_limit > kvm->arch.mem_limit)
627 return -E2BIG;
628
629 if (!new_limit)
630 return -EINVAL;
631
632 /* gmap_create takes last usable address */
633 if (new_limit != KVM_S390_NO_MEM_LIMIT)
634 new_limit -= 1;
635
636 ret = -EBUSY;
637 mutex_lock(&kvm->lock);
638 if (!kvm->created_vcpus) {
639 /* gmap_create will round the limit up */
640 struct gmap *new = gmap_create(current->mm, new_limit);
641
642 if (!new) {
643 ret = -ENOMEM;
644 } else {
645 gmap_remove(kvm->arch.gmap);
646 new->private = kvm;
647 kvm->arch.gmap = new;
648 ret = 0;
649 }
650 }
651 mutex_unlock(&kvm->lock);
652 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
653 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
654 (void *) kvm->arch.gmap->asce);
655 break;
656 }
657 default:
658 ret = -ENXIO;
659 break;
660 }
661 return ret;
662 }
663
664 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
665
666 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
667 {
668 struct kvm_vcpu *vcpu;
669 int i;
670
671 if (!test_kvm_facility(kvm, 76))
672 return -EINVAL;
673
674 mutex_lock(&kvm->lock);
675 switch (attr->attr) {
676 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
677 get_random_bytes(
678 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
679 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
680 kvm->arch.crypto.aes_kw = 1;
681 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
682 break;
683 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
684 get_random_bytes(
685 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
686 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
687 kvm->arch.crypto.dea_kw = 1;
688 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
689 break;
690 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
691 kvm->arch.crypto.aes_kw = 0;
692 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
693 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
694 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
695 break;
696 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
697 kvm->arch.crypto.dea_kw = 0;
698 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
699 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
700 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
701 break;
702 default:
703 mutex_unlock(&kvm->lock);
704 return -ENXIO;
705 }
706
707 kvm_for_each_vcpu(i, vcpu, kvm) {
708 kvm_s390_vcpu_crypto_setup(vcpu);
709 exit_sie(vcpu);
710 }
711 mutex_unlock(&kvm->lock);
712 return 0;
713 }
714
715 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
716 {
717 u8 gtod_high;
718
719 if (copy_from_user(&gtod_high, (void __user *)attr->addr,
720 sizeof(gtod_high)))
721 return -EFAULT;
722
723 if (gtod_high != 0)
724 return -EINVAL;
725 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
726
727 return 0;
728 }
729
730 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
731 {
732 u64 gtod;
733
734 if (copy_from_user(&gtod, (void __user *)attr->addr, sizeof(gtod)))
735 return -EFAULT;
736
737 kvm_s390_set_tod_clock(kvm, gtod);
738 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod);
739 return 0;
740 }
741
742 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
743 {
744 int ret;
745
746 if (attr->flags)
747 return -EINVAL;
748
749 switch (attr->attr) {
750 case KVM_S390_VM_TOD_HIGH:
751 ret = kvm_s390_set_tod_high(kvm, attr);
752 break;
753 case KVM_S390_VM_TOD_LOW:
754 ret = kvm_s390_set_tod_low(kvm, attr);
755 break;
756 default:
757 ret = -ENXIO;
758 break;
759 }
760 return ret;
761 }
762
763 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
764 {
765 u8 gtod_high = 0;
766
767 if (copy_to_user((void __user *)attr->addr, &gtod_high,
768 sizeof(gtod_high)))
769 return -EFAULT;
770 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
771
772 return 0;
773 }
774
775 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
776 {
777 u64 gtod;
778
779 gtod = kvm_s390_get_tod_clock_fast(kvm);
780 if (copy_to_user((void __user *)attr->addr, &gtod, sizeof(gtod)))
781 return -EFAULT;
782 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
783
784 return 0;
785 }
786
787 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
788 {
789 int ret;
790
791 if (attr->flags)
792 return -EINVAL;
793
794 switch (attr->attr) {
795 case KVM_S390_VM_TOD_HIGH:
796 ret = kvm_s390_get_tod_high(kvm, attr);
797 break;
798 case KVM_S390_VM_TOD_LOW:
799 ret = kvm_s390_get_tod_low(kvm, attr);
800 break;
801 default:
802 ret = -ENXIO;
803 break;
804 }
805 return ret;
806 }
807
808 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
809 {
810 struct kvm_s390_vm_cpu_processor *proc;
811 u16 lowest_ibc, unblocked_ibc;
812 int ret = 0;
813
814 mutex_lock(&kvm->lock);
815 if (kvm->created_vcpus) {
816 ret = -EBUSY;
817 goto out;
818 }
819 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
820 if (!proc) {
821 ret = -ENOMEM;
822 goto out;
823 }
824 if (!copy_from_user(proc, (void __user *)attr->addr,
825 sizeof(*proc))) {
826 kvm->arch.model.cpuid = proc->cpuid;
827 lowest_ibc = sclp.ibc >> 16 & 0xfff;
828 unblocked_ibc = sclp.ibc & 0xfff;
829 if (lowest_ibc && proc->ibc) {
830 if (proc->ibc > unblocked_ibc)
831 kvm->arch.model.ibc = unblocked_ibc;
832 else if (proc->ibc < lowest_ibc)
833 kvm->arch.model.ibc = lowest_ibc;
834 else
835 kvm->arch.model.ibc = proc->ibc;
836 }
837 memcpy(kvm->arch.model.fac_list, proc->fac_list,
838 S390_ARCH_FAC_LIST_SIZE_BYTE);
839 VM_EVENT(kvm, 3, "SET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
840 kvm->arch.model.ibc,
841 kvm->arch.model.cpuid);
842 VM_EVENT(kvm, 3, "SET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
843 kvm->arch.model.fac_list[0],
844 kvm->arch.model.fac_list[1],
845 kvm->arch.model.fac_list[2]);
846 } else
847 ret = -EFAULT;
848 kfree(proc);
849 out:
850 mutex_unlock(&kvm->lock);
851 return ret;
852 }
853
854 static int kvm_s390_set_processor_feat(struct kvm *kvm,
855 struct kvm_device_attr *attr)
856 {
857 struct kvm_s390_vm_cpu_feat data;
858 int ret = -EBUSY;
859
860 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
861 return -EFAULT;
862 if (!bitmap_subset((unsigned long *) data.feat,
863 kvm_s390_available_cpu_feat,
864 KVM_S390_VM_CPU_FEAT_NR_BITS))
865 return -EINVAL;
866
867 mutex_lock(&kvm->lock);
868 if (!atomic_read(&kvm->online_vcpus)) {
869 bitmap_copy(kvm->arch.cpu_feat, (unsigned long *) data.feat,
870 KVM_S390_VM_CPU_FEAT_NR_BITS);
871 ret = 0;
872 }
873 mutex_unlock(&kvm->lock);
874 return ret;
875 }
876
877 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
878 struct kvm_device_attr *attr)
879 {
880 /*
881 * Once supported by kernel + hw, we have to store the subfunctions
882 * in kvm->arch and remember that user space configured them.
883 */
884 return -ENXIO;
885 }
886
887 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
888 {
889 int ret = -ENXIO;
890
891 switch (attr->attr) {
892 case KVM_S390_VM_CPU_PROCESSOR:
893 ret = kvm_s390_set_processor(kvm, attr);
894 break;
895 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
896 ret = kvm_s390_set_processor_feat(kvm, attr);
897 break;
898 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
899 ret = kvm_s390_set_processor_subfunc(kvm, attr);
900 break;
901 }
902 return ret;
903 }
904
905 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
906 {
907 struct kvm_s390_vm_cpu_processor *proc;
908 int ret = 0;
909
910 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
911 if (!proc) {
912 ret = -ENOMEM;
913 goto out;
914 }
915 proc->cpuid = kvm->arch.model.cpuid;
916 proc->ibc = kvm->arch.model.ibc;
917 memcpy(&proc->fac_list, kvm->arch.model.fac_list,
918 S390_ARCH_FAC_LIST_SIZE_BYTE);
919 VM_EVENT(kvm, 3, "GET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
920 kvm->arch.model.ibc,
921 kvm->arch.model.cpuid);
922 VM_EVENT(kvm, 3, "GET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
923 kvm->arch.model.fac_list[0],
924 kvm->arch.model.fac_list[1],
925 kvm->arch.model.fac_list[2]);
926 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
927 ret = -EFAULT;
928 kfree(proc);
929 out:
930 return ret;
931 }
932
933 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
934 {
935 struct kvm_s390_vm_cpu_machine *mach;
936 int ret = 0;
937
938 mach = kzalloc(sizeof(*mach), GFP_KERNEL);
939 if (!mach) {
940 ret = -ENOMEM;
941 goto out;
942 }
943 get_cpu_id((struct cpuid *) &mach->cpuid);
944 mach->ibc = sclp.ibc;
945 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
946 S390_ARCH_FAC_LIST_SIZE_BYTE);
947 memcpy((unsigned long *)&mach->fac_list, S390_lowcore.stfle_fac_list,
948 sizeof(S390_lowcore.stfle_fac_list));
949 VM_EVENT(kvm, 3, "GET: host ibc: 0x%4.4x, host cpuid: 0x%16.16llx",
950 kvm->arch.model.ibc,
951 kvm->arch.model.cpuid);
952 VM_EVENT(kvm, 3, "GET: host facmask: 0x%16.16llx.%16.16llx.%16.16llx",
953 mach->fac_mask[0],
954 mach->fac_mask[1],
955 mach->fac_mask[2]);
956 VM_EVENT(kvm, 3, "GET: host faclist: 0x%16.16llx.%16.16llx.%16.16llx",
957 mach->fac_list[0],
958 mach->fac_list[1],
959 mach->fac_list[2]);
960 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
961 ret = -EFAULT;
962 kfree(mach);
963 out:
964 return ret;
965 }
966
967 static int kvm_s390_get_processor_feat(struct kvm *kvm,
968 struct kvm_device_attr *attr)
969 {
970 struct kvm_s390_vm_cpu_feat data;
971
972 bitmap_copy((unsigned long *) data.feat, kvm->arch.cpu_feat,
973 KVM_S390_VM_CPU_FEAT_NR_BITS);
974 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
975 return -EFAULT;
976 return 0;
977 }
978
979 static int kvm_s390_get_machine_feat(struct kvm *kvm,
980 struct kvm_device_attr *attr)
981 {
982 struct kvm_s390_vm_cpu_feat data;
983
984 bitmap_copy((unsigned long *) data.feat,
985 kvm_s390_available_cpu_feat,
986 KVM_S390_VM_CPU_FEAT_NR_BITS);
987 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
988 return -EFAULT;
989 return 0;
990 }
991
992 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
993 struct kvm_device_attr *attr)
994 {
995 /*
996 * Once we can actually configure subfunctions (kernel + hw support),
997 * we have to check if they were already set by user space, if so copy
998 * them from kvm->arch.
999 */
1000 return -ENXIO;
1001 }
1002
1003 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
1004 struct kvm_device_attr *attr)
1005 {
1006 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
1007 sizeof(struct kvm_s390_vm_cpu_subfunc)))
1008 return -EFAULT;
1009 return 0;
1010 }
1011 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1012 {
1013 int ret = -ENXIO;
1014
1015 switch (attr->attr) {
1016 case KVM_S390_VM_CPU_PROCESSOR:
1017 ret = kvm_s390_get_processor(kvm, attr);
1018 break;
1019 case KVM_S390_VM_CPU_MACHINE:
1020 ret = kvm_s390_get_machine(kvm, attr);
1021 break;
1022 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1023 ret = kvm_s390_get_processor_feat(kvm, attr);
1024 break;
1025 case KVM_S390_VM_CPU_MACHINE_FEAT:
1026 ret = kvm_s390_get_machine_feat(kvm, attr);
1027 break;
1028 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1029 ret = kvm_s390_get_processor_subfunc(kvm, attr);
1030 break;
1031 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1032 ret = kvm_s390_get_machine_subfunc(kvm, attr);
1033 break;
1034 }
1035 return ret;
1036 }
1037
1038 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1039 {
1040 int ret;
1041
1042 switch (attr->group) {
1043 case KVM_S390_VM_MEM_CTRL:
1044 ret = kvm_s390_set_mem_control(kvm, attr);
1045 break;
1046 case KVM_S390_VM_TOD:
1047 ret = kvm_s390_set_tod(kvm, attr);
1048 break;
1049 case KVM_S390_VM_CPU_MODEL:
1050 ret = kvm_s390_set_cpu_model(kvm, attr);
1051 break;
1052 case KVM_S390_VM_CRYPTO:
1053 ret = kvm_s390_vm_set_crypto(kvm, attr);
1054 break;
1055 default:
1056 ret = -ENXIO;
1057 break;
1058 }
1059
1060 return ret;
1061 }
1062
1063 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1064 {
1065 int ret;
1066
1067 switch (attr->group) {
1068 case KVM_S390_VM_MEM_CTRL:
1069 ret = kvm_s390_get_mem_control(kvm, attr);
1070 break;
1071 case KVM_S390_VM_TOD:
1072 ret = kvm_s390_get_tod(kvm, attr);
1073 break;
1074 case KVM_S390_VM_CPU_MODEL:
1075 ret = kvm_s390_get_cpu_model(kvm, attr);
1076 break;
1077 default:
1078 ret = -ENXIO;
1079 break;
1080 }
1081
1082 return ret;
1083 }
1084
1085 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1086 {
1087 int ret;
1088
1089 switch (attr->group) {
1090 case KVM_S390_VM_MEM_CTRL:
1091 switch (attr->attr) {
1092 case KVM_S390_VM_MEM_ENABLE_CMMA:
1093 case KVM_S390_VM_MEM_CLR_CMMA:
1094 ret = sclp.has_cmma ? 0 : -ENXIO;
1095 break;
1096 case KVM_S390_VM_MEM_LIMIT_SIZE:
1097 ret = 0;
1098 break;
1099 default:
1100 ret = -ENXIO;
1101 break;
1102 }
1103 break;
1104 case KVM_S390_VM_TOD:
1105 switch (attr->attr) {
1106 case KVM_S390_VM_TOD_LOW:
1107 case KVM_S390_VM_TOD_HIGH:
1108 ret = 0;
1109 break;
1110 default:
1111 ret = -ENXIO;
1112 break;
1113 }
1114 break;
1115 case KVM_S390_VM_CPU_MODEL:
1116 switch (attr->attr) {
1117 case KVM_S390_VM_CPU_PROCESSOR:
1118 case KVM_S390_VM_CPU_MACHINE:
1119 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1120 case KVM_S390_VM_CPU_MACHINE_FEAT:
1121 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1122 ret = 0;
1123 break;
1124 /* configuring subfunctions is not supported yet */
1125 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1126 default:
1127 ret = -ENXIO;
1128 break;
1129 }
1130 break;
1131 case KVM_S390_VM_CRYPTO:
1132 switch (attr->attr) {
1133 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1134 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1135 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1136 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1137 ret = 0;
1138 break;
1139 default:
1140 ret = -ENXIO;
1141 break;
1142 }
1143 break;
1144 default:
1145 ret = -ENXIO;
1146 break;
1147 }
1148
1149 return ret;
1150 }
1151
1152 static long kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1153 {
1154 uint8_t *keys;
1155 uint64_t hva;
1156 int i, r = 0;
1157
1158 if (args->flags != 0)
1159 return -EINVAL;
1160
1161 /* Is this guest using storage keys? */
1162 if (!mm_use_skey(current->mm))
1163 return KVM_S390_GET_SKEYS_NONE;
1164
1165 /* Enforce sane limit on memory allocation */
1166 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1167 return -EINVAL;
1168
1169 keys = kmalloc_array(args->count, sizeof(uint8_t),
1170 GFP_KERNEL | __GFP_NOWARN);
1171 if (!keys)
1172 keys = vmalloc(sizeof(uint8_t) * args->count);
1173 if (!keys)
1174 return -ENOMEM;
1175
1176 down_read(&current->mm->mmap_sem);
1177 for (i = 0; i < args->count; i++) {
1178 hva = gfn_to_hva(kvm, args->start_gfn + i);
1179 if (kvm_is_error_hva(hva)) {
1180 r = -EFAULT;
1181 break;
1182 }
1183
1184 r = get_guest_storage_key(current->mm, hva, &keys[i]);
1185 if (r)
1186 break;
1187 }
1188 up_read(&current->mm->mmap_sem);
1189
1190 if (!r) {
1191 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
1192 sizeof(uint8_t) * args->count);
1193 if (r)
1194 r = -EFAULT;
1195 }
1196
1197 kvfree(keys);
1198 return r;
1199 }
1200
1201 static long kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1202 {
1203 uint8_t *keys;
1204 uint64_t hva;
1205 int i, r = 0;
1206
1207 if (args->flags != 0)
1208 return -EINVAL;
1209
1210 /* Enforce sane limit on memory allocation */
1211 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1212 return -EINVAL;
1213
1214 keys = kmalloc_array(args->count, sizeof(uint8_t),
1215 GFP_KERNEL | __GFP_NOWARN);
1216 if (!keys)
1217 keys = vmalloc(sizeof(uint8_t) * args->count);
1218 if (!keys)
1219 return -ENOMEM;
1220
1221 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
1222 sizeof(uint8_t) * args->count);
1223 if (r) {
1224 r = -EFAULT;
1225 goto out;
1226 }
1227
1228 /* Enable storage key handling for the guest */
1229 r = s390_enable_skey();
1230 if (r)
1231 goto out;
1232
1233 down_read(&current->mm->mmap_sem);
1234 for (i = 0; i < args->count; i++) {
1235 hva = gfn_to_hva(kvm, args->start_gfn + i);
1236 if (kvm_is_error_hva(hva)) {
1237 r = -EFAULT;
1238 break;
1239 }
1240
1241 /* Lowest order bit is reserved */
1242 if (keys[i] & 0x01) {
1243 r = -EINVAL;
1244 break;
1245 }
1246
1247 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
1248 if (r)
1249 break;
1250 }
1251 up_read(&current->mm->mmap_sem);
1252 out:
1253 kvfree(keys);
1254 return r;
1255 }
1256
1257 long kvm_arch_vm_ioctl(struct file *filp,
1258 unsigned int ioctl, unsigned long arg)
1259 {
1260 struct kvm *kvm = filp->private_data;
1261 void __user *argp = (void __user *)arg;
1262 struct kvm_device_attr attr;
1263 int r;
1264
1265 switch (ioctl) {
1266 case KVM_S390_INTERRUPT: {
1267 struct kvm_s390_interrupt s390int;
1268
1269 r = -EFAULT;
1270 if (copy_from_user(&s390int, argp, sizeof(s390int)))
1271 break;
1272 r = kvm_s390_inject_vm(kvm, &s390int);
1273 break;
1274 }
1275 case KVM_ENABLE_CAP: {
1276 struct kvm_enable_cap cap;
1277 r = -EFAULT;
1278 if (copy_from_user(&cap, argp, sizeof(cap)))
1279 break;
1280 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1281 break;
1282 }
1283 case KVM_CREATE_IRQCHIP: {
1284 struct kvm_irq_routing_entry routing;
1285
1286 r = -EINVAL;
1287 if (kvm->arch.use_irqchip) {
1288 /* Set up dummy routing. */
1289 memset(&routing, 0, sizeof(routing));
1290 r = kvm_set_irq_routing(kvm, &routing, 0, 0);
1291 }
1292 break;
1293 }
1294 case KVM_SET_DEVICE_ATTR: {
1295 r = -EFAULT;
1296 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1297 break;
1298 r = kvm_s390_vm_set_attr(kvm, &attr);
1299 break;
1300 }
1301 case KVM_GET_DEVICE_ATTR: {
1302 r = -EFAULT;
1303 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1304 break;
1305 r = kvm_s390_vm_get_attr(kvm, &attr);
1306 break;
1307 }
1308 case KVM_HAS_DEVICE_ATTR: {
1309 r = -EFAULT;
1310 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1311 break;
1312 r = kvm_s390_vm_has_attr(kvm, &attr);
1313 break;
1314 }
1315 case KVM_S390_GET_SKEYS: {
1316 struct kvm_s390_skeys args;
1317
1318 r = -EFAULT;
1319 if (copy_from_user(&args, argp,
1320 sizeof(struct kvm_s390_skeys)))
1321 break;
1322 r = kvm_s390_get_skeys(kvm, &args);
1323 break;
1324 }
1325 case KVM_S390_SET_SKEYS: {
1326 struct kvm_s390_skeys args;
1327
1328 r = -EFAULT;
1329 if (copy_from_user(&args, argp,
1330 sizeof(struct kvm_s390_skeys)))
1331 break;
1332 r = kvm_s390_set_skeys(kvm, &args);
1333 break;
1334 }
1335 default:
1336 r = -ENOTTY;
1337 }
1338
1339 return r;
1340 }
1341
1342 static int kvm_s390_query_ap_config(u8 *config)
1343 {
1344 u32 fcn_code = 0x04000000UL;
1345 u32 cc = 0;
1346
1347 memset(config, 0, 128);
1348 asm volatile(
1349 "lgr 0,%1\n"
1350 "lgr 2,%2\n"
1351 ".long 0xb2af0000\n" /* PQAP(QCI) */
1352 "0: ipm %0\n"
1353 "srl %0,28\n"
1354 "1:\n"
1355 EX_TABLE(0b, 1b)
1356 : "+r" (cc)
1357 : "r" (fcn_code), "r" (config)
1358 : "cc", "0", "2", "memory"
1359 );
1360
1361 return cc;
1362 }
1363
1364 static int kvm_s390_apxa_installed(void)
1365 {
1366 u8 config[128];
1367 int cc;
1368
1369 if (test_facility(12)) {
1370 cc = kvm_s390_query_ap_config(config);
1371
1372 if (cc)
1373 pr_err("PQAP(QCI) failed with cc=%d", cc);
1374 else
1375 return config[0] & 0x40;
1376 }
1377
1378 return 0;
1379 }
1380
1381 static void kvm_s390_set_crycb_format(struct kvm *kvm)
1382 {
1383 kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
1384
1385 if (kvm_s390_apxa_installed())
1386 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
1387 else
1388 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
1389 }
1390
1391 static u64 kvm_s390_get_initial_cpuid(void)
1392 {
1393 struct cpuid cpuid;
1394
1395 get_cpu_id(&cpuid);
1396 cpuid.version = 0xff;
1397 return *((u64 *) &cpuid);
1398 }
1399
1400 static void kvm_s390_crypto_init(struct kvm *kvm)
1401 {
1402 if (!test_kvm_facility(kvm, 76))
1403 return;
1404
1405 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
1406 kvm_s390_set_crycb_format(kvm);
1407
1408 /* Enable AES/DEA protected key functions by default */
1409 kvm->arch.crypto.aes_kw = 1;
1410 kvm->arch.crypto.dea_kw = 1;
1411 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1412 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1413 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1414 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1415 }
1416
1417 static void sca_dispose(struct kvm *kvm)
1418 {
1419 if (kvm->arch.use_esca)
1420 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
1421 else
1422 free_page((unsigned long)(kvm->arch.sca));
1423 kvm->arch.sca = NULL;
1424 }
1425
1426 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
1427 {
1428 gfp_t alloc_flags = GFP_KERNEL;
1429 int i, rc;
1430 char debug_name[16];
1431 static unsigned long sca_offset;
1432
1433 rc = -EINVAL;
1434 #ifdef CONFIG_KVM_S390_UCONTROL
1435 if (type & ~KVM_VM_S390_UCONTROL)
1436 goto out_err;
1437 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
1438 goto out_err;
1439 #else
1440 if (type)
1441 goto out_err;
1442 #endif
1443
1444 rc = s390_enable_sie();
1445 if (rc)
1446 goto out_err;
1447
1448 rc = -ENOMEM;
1449
1450 ratelimit_state_init(&kvm->arch.sthyi_limit, 5 * HZ, 500);
1451
1452 kvm->arch.use_esca = 0; /* start with basic SCA */
1453 if (!sclp.has_64bscao)
1454 alloc_flags |= GFP_DMA;
1455 rwlock_init(&kvm->arch.sca_lock);
1456 kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
1457 if (!kvm->arch.sca)
1458 goto out_err;
1459 spin_lock(&kvm_lock);
1460 sca_offset += 16;
1461 if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
1462 sca_offset = 0;
1463 kvm->arch.sca = (struct bsca_block *)
1464 ((char *) kvm->arch.sca + sca_offset);
1465 spin_unlock(&kvm_lock);
1466
1467 sprintf(debug_name, "kvm-%u", current->pid);
1468
1469 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
1470 if (!kvm->arch.dbf)
1471 goto out_err;
1472
1473 kvm->arch.sie_page2 =
1474 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1475 if (!kvm->arch.sie_page2)
1476 goto out_err;
1477
1478 /* Populate the facility mask initially. */
1479 memcpy(kvm->arch.model.fac_mask, S390_lowcore.stfle_fac_list,
1480 sizeof(S390_lowcore.stfle_fac_list));
1481 for (i = 0; i < S390_ARCH_FAC_LIST_SIZE_U64; i++) {
1482 if (i < kvm_s390_fac_list_mask_size())
1483 kvm->arch.model.fac_mask[i] &= kvm_s390_fac_list_mask[i];
1484 else
1485 kvm->arch.model.fac_mask[i] = 0UL;
1486 }
1487
1488 /* Populate the facility list initially. */
1489 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
1490 memcpy(kvm->arch.model.fac_list, kvm->arch.model.fac_mask,
1491 S390_ARCH_FAC_LIST_SIZE_BYTE);
1492
1493 set_kvm_facility(kvm->arch.model.fac_mask, 74);
1494 set_kvm_facility(kvm->arch.model.fac_list, 74);
1495
1496 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
1497 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
1498
1499 kvm_s390_crypto_init(kvm);
1500
1501 spin_lock_init(&kvm->arch.float_int.lock);
1502 for (i = 0; i < FIRQ_LIST_COUNT; i++)
1503 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
1504 init_waitqueue_head(&kvm->arch.ipte_wq);
1505 mutex_init(&kvm->arch.ipte_mutex);
1506
1507 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
1508 VM_EVENT(kvm, 3, "vm created with type %lu", type);
1509
1510 if (type & KVM_VM_S390_UCONTROL) {
1511 kvm->arch.gmap = NULL;
1512 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
1513 } else {
1514 if (sclp.hamax == U64_MAX)
1515 kvm->arch.mem_limit = TASK_MAX_SIZE;
1516 else
1517 kvm->arch.mem_limit = min_t(unsigned long, TASK_MAX_SIZE,
1518 sclp.hamax + 1);
1519 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
1520 if (!kvm->arch.gmap)
1521 goto out_err;
1522 kvm->arch.gmap->private = kvm;
1523 kvm->arch.gmap->pfault_enabled = 0;
1524 }
1525
1526 kvm->arch.css_support = 0;
1527 kvm->arch.use_irqchip = 0;
1528 kvm->arch.epoch = 0;
1529
1530 spin_lock_init(&kvm->arch.start_stop_lock);
1531 kvm_s390_vsie_init(kvm);
1532 KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
1533
1534 return 0;
1535 out_err:
1536 free_page((unsigned long)kvm->arch.sie_page2);
1537 debug_unregister(kvm->arch.dbf);
1538 sca_dispose(kvm);
1539 KVM_EVENT(3, "creation of vm failed: %d", rc);
1540 return rc;
1541 }
1542
1543 bool kvm_arch_has_vcpu_debugfs(void)
1544 {
1545 return false;
1546 }
1547
1548 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
1549 {
1550 return 0;
1551 }
1552
1553 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1554 {
1555 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
1556 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
1557 kvm_s390_clear_local_irqs(vcpu);
1558 kvm_clear_async_pf_completion_queue(vcpu);
1559 if (!kvm_is_ucontrol(vcpu->kvm))
1560 sca_del_vcpu(vcpu);
1561
1562 if (kvm_is_ucontrol(vcpu->kvm))
1563 gmap_remove(vcpu->arch.gmap);
1564
1565 if (vcpu->kvm->arch.use_cmma)
1566 kvm_s390_vcpu_unsetup_cmma(vcpu);
1567 free_page((unsigned long)(vcpu->arch.sie_block));
1568
1569 kvm_vcpu_uninit(vcpu);
1570 kmem_cache_free(kvm_vcpu_cache, vcpu);
1571 }
1572
1573 static void kvm_free_vcpus(struct kvm *kvm)
1574 {
1575 unsigned int i;
1576 struct kvm_vcpu *vcpu;
1577
1578 kvm_for_each_vcpu(i, vcpu, kvm)
1579 kvm_arch_vcpu_destroy(vcpu);
1580
1581 mutex_lock(&kvm->lock);
1582 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
1583 kvm->vcpus[i] = NULL;
1584
1585 atomic_set(&kvm->online_vcpus, 0);
1586 mutex_unlock(&kvm->lock);
1587 }
1588
1589 void kvm_arch_destroy_vm(struct kvm *kvm)
1590 {
1591 kvm_free_vcpus(kvm);
1592 sca_dispose(kvm);
1593 debug_unregister(kvm->arch.dbf);
1594 free_page((unsigned long)kvm->arch.sie_page2);
1595 if (!kvm_is_ucontrol(kvm))
1596 gmap_remove(kvm->arch.gmap);
1597 kvm_s390_destroy_adapters(kvm);
1598 kvm_s390_clear_float_irqs(kvm);
1599 kvm_s390_vsie_destroy(kvm);
1600 KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
1601 }
1602
1603 /* Section: vcpu related */
1604 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
1605 {
1606 vcpu->arch.gmap = gmap_create(current->mm, -1UL);
1607 if (!vcpu->arch.gmap)
1608 return -ENOMEM;
1609 vcpu->arch.gmap->private = vcpu->kvm;
1610
1611 return 0;
1612 }
1613
1614 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
1615 {
1616 if (!kvm_s390_use_sca_entries())
1617 return;
1618 read_lock(&vcpu->kvm->arch.sca_lock);
1619 if (vcpu->kvm->arch.use_esca) {
1620 struct esca_block *sca = vcpu->kvm->arch.sca;
1621
1622 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
1623 sca->cpu[vcpu->vcpu_id].sda = 0;
1624 } else {
1625 struct bsca_block *sca = vcpu->kvm->arch.sca;
1626
1627 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
1628 sca->cpu[vcpu->vcpu_id].sda = 0;
1629 }
1630 read_unlock(&vcpu->kvm->arch.sca_lock);
1631 }
1632
1633 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
1634 {
1635 if (!kvm_s390_use_sca_entries()) {
1636 struct bsca_block *sca = vcpu->kvm->arch.sca;
1637
1638 /* we still need the basic sca for the ipte control */
1639 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1640 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
1641 }
1642 read_lock(&vcpu->kvm->arch.sca_lock);
1643 if (vcpu->kvm->arch.use_esca) {
1644 struct esca_block *sca = vcpu->kvm->arch.sca;
1645
1646 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
1647 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1648 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca & ~0x3fU;
1649 vcpu->arch.sie_block->ecb2 |= 0x04U;
1650 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
1651 } else {
1652 struct bsca_block *sca = vcpu->kvm->arch.sca;
1653
1654 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
1655 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1656 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
1657 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
1658 }
1659 read_unlock(&vcpu->kvm->arch.sca_lock);
1660 }
1661
1662 /* Basic SCA to Extended SCA data copy routines */
1663 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
1664 {
1665 d->sda = s->sda;
1666 d->sigp_ctrl.c = s->sigp_ctrl.c;
1667 d->sigp_ctrl.scn = s->sigp_ctrl.scn;
1668 }
1669
1670 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
1671 {
1672 int i;
1673
1674 d->ipte_control = s->ipte_control;
1675 d->mcn[0] = s->mcn;
1676 for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
1677 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
1678 }
1679
1680 static int sca_switch_to_extended(struct kvm *kvm)
1681 {
1682 struct bsca_block *old_sca = kvm->arch.sca;
1683 struct esca_block *new_sca;
1684 struct kvm_vcpu *vcpu;
1685 unsigned int vcpu_idx;
1686 u32 scaol, scaoh;
1687
1688 new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL|__GFP_ZERO);
1689 if (!new_sca)
1690 return -ENOMEM;
1691
1692 scaoh = (u32)((u64)(new_sca) >> 32);
1693 scaol = (u32)(u64)(new_sca) & ~0x3fU;
1694
1695 kvm_s390_vcpu_block_all(kvm);
1696 write_lock(&kvm->arch.sca_lock);
1697
1698 sca_copy_b_to_e(new_sca, old_sca);
1699
1700 kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
1701 vcpu->arch.sie_block->scaoh = scaoh;
1702 vcpu->arch.sie_block->scaol = scaol;
1703 vcpu->arch.sie_block->ecb2 |= 0x04U;
1704 }
1705 kvm->arch.sca = new_sca;
1706 kvm->arch.use_esca = 1;
1707
1708 write_unlock(&kvm->arch.sca_lock);
1709 kvm_s390_vcpu_unblock_all(kvm);
1710
1711 free_page((unsigned long)old_sca);
1712
1713 VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
1714 old_sca, kvm->arch.sca);
1715 return 0;
1716 }
1717
1718 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
1719 {
1720 int rc;
1721
1722 if (!kvm_s390_use_sca_entries()) {
1723 if (id < KVM_MAX_VCPUS)
1724 return true;
1725 return false;
1726 }
1727 if (id < KVM_S390_BSCA_CPU_SLOTS)
1728 return true;
1729 if (!sclp.has_esca || !sclp.has_64bscao)
1730 return false;
1731
1732 mutex_lock(&kvm->lock);
1733 rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
1734 mutex_unlock(&kvm->lock);
1735
1736 return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
1737 }
1738
1739 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1740 {
1741 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1742 kvm_clear_async_pf_completion_queue(vcpu);
1743 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
1744 KVM_SYNC_GPRS |
1745 KVM_SYNC_ACRS |
1746 KVM_SYNC_CRS |
1747 KVM_SYNC_ARCH0 |
1748 KVM_SYNC_PFAULT;
1749 kvm_s390_set_prefix(vcpu, 0);
1750 if (test_kvm_facility(vcpu->kvm, 64))
1751 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
1752 /* fprs can be synchronized via vrs, even if the guest has no vx. With
1753 * MACHINE_HAS_VX, (load|store)_fpu_regs() will work with vrs format.
1754 */
1755 if (MACHINE_HAS_VX)
1756 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
1757 else
1758 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
1759
1760 if (kvm_is_ucontrol(vcpu->kvm))
1761 return __kvm_ucontrol_vcpu_init(vcpu);
1762
1763 return 0;
1764 }
1765
1766 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1767 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1768 {
1769 WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
1770 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1771 vcpu->arch.cputm_start = get_tod_clock_fast();
1772 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1773 }
1774
1775 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1776 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1777 {
1778 WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
1779 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1780 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
1781 vcpu->arch.cputm_start = 0;
1782 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1783 }
1784
1785 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1786 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1787 {
1788 WARN_ON_ONCE(vcpu->arch.cputm_enabled);
1789 vcpu->arch.cputm_enabled = true;
1790 __start_cpu_timer_accounting(vcpu);
1791 }
1792
1793 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1794 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1795 {
1796 WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
1797 __stop_cpu_timer_accounting(vcpu);
1798 vcpu->arch.cputm_enabled = false;
1799 }
1800
1801 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1802 {
1803 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1804 __enable_cpu_timer_accounting(vcpu);
1805 preempt_enable();
1806 }
1807
1808 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1809 {
1810 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1811 __disable_cpu_timer_accounting(vcpu);
1812 preempt_enable();
1813 }
1814
1815 /* set the cpu timer - may only be called from the VCPU thread itself */
1816 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
1817 {
1818 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1819 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1820 if (vcpu->arch.cputm_enabled)
1821 vcpu->arch.cputm_start = get_tod_clock_fast();
1822 vcpu->arch.sie_block->cputm = cputm;
1823 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1824 preempt_enable();
1825 }
1826
1827 /* update and get the cpu timer - can also be called from other VCPU threads */
1828 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
1829 {
1830 unsigned int seq;
1831 __u64 value;
1832
1833 if (unlikely(!vcpu->arch.cputm_enabled))
1834 return vcpu->arch.sie_block->cputm;
1835
1836 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1837 do {
1838 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
1839 /*
1840 * If the writer would ever execute a read in the critical
1841 * section, e.g. in irq context, we have a deadlock.
1842 */
1843 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
1844 value = vcpu->arch.sie_block->cputm;
1845 /* if cputm_start is 0, accounting is being started/stopped */
1846 if (likely(vcpu->arch.cputm_start))
1847 value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
1848 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
1849 preempt_enable();
1850 return value;
1851 }
1852
1853 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1854 {
1855
1856 gmap_enable(vcpu->arch.enabled_gmap);
1857 atomic_or(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1858 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
1859 __start_cpu_timer_accounting(vcpu);
1860 vcpu->cpu = cpu;
1861 }
1862
1863 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1864 {
1865 vcpu->cpu = -1;
1866 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
1867 __stop_cpu_timer_accounting(vcpu);
1868 atomic_andnot(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1869 vcpu->arch.enabled_gmap = gmap_get_enabled();
1870 gmap_disable(vcpu->arch.enabled_gmap);
1871
1872 }
1873
1874 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
1875 {
1876 /* this equals initial cpu reset in pop, but we don't switch to ESA */
1877 vcpu->arch.sie_block->gpsw.mask = 0UL;
1878 vcpu->arch.sie_block->gpsw.addr = 0UL;
1879 kvm_s390_set_prefix(vcpu, 0);
1880 kvm_s390_set_cpu_timer(vcpu, 0);
1881 vcpu->arch.sie_block->ckc = 0UL;
1882 vcpu->arch.sie_block->todpr = 0;
1883 memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64));
1884 vcpu->arch.sie_block->gcr[0] = 0xE0UL;
1885 vcpu->arch.sie_block->gcr[14] = 0xC2000000UL;
1886 /* make sure the new fpc will be lazily loaded */
1887 save_fpu_regs();
1888 current->thread.fpu.fpc = 0;
1889 vcpu->arch.sie_block->gbea = 1;
1890 vcpu->arch.sie_block->pp = 0;
1891 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1892 kvm_clear_async_pf_completion_queue(vcpu);
1893 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
1894 kvm_s390_vcpu_stop(vcpu);
1895 kvm_s390_clear_local_irqs(vcpu);
1896 }
1897
1898 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1899 {
1900 mutex_lock(&vcpu->kvm->lock);
1901 preempt_disable();
1902 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
1903 preempt_enable();
1904 mutex_unlock(&vcpu->kvm->lock);
1905 if (!kvm_is_ucontrol(vcpu->kvm)) {
1906 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
1907 sca_add_vcpu(vcpu);
1908 }
1909 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
1910 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
1911 /* make vcpu_load load the right gmap on the first trigger */
1912 vcpu->arch.enabled_gmap = vcpu->arch.gmap;
1913 }
1914
1915 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
1916 {
1917 if (!test_kvm_facility(vcpu->kvm, 76))
1918 return;
1919
1920 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
1921
1922 if (vcpu->kvm->arch.crypto.aes_kw)
1923 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
1924 if (vcpu->kvm->arch.crypto.dea_kw)
1925 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
1926
1927 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
1928 }
1929
1930 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
1931 {
1932 free_page(vcpu->arch.sie_block->cbrlo);
1933 vcpu->arch.sie_block->cbrlo = 0;
1934 }
1935
1936 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
1937 {
1938 vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
1939 if (!vcpu->arch.sie_block->cbrlo)
1940 return -ENOMEM;
1941
1942 vcpu->arch.sie_block->ecb2 |= 0x80;
1943 vcpu->arch.sie_block->ecb2 &= ~0x08;
1944 return 0;
1945 }
1946
1947 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
1948 {
1949 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
1950
1951 vcpu->arch.sie_block->ibc = model->ibc;
1952 if (test_kvm_facility(vcpu->kvm, 7))
1953 vcpu->arch.sie_block->fac = (u32)(u64) model->fac_list;
1954 }
1955
1956 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1957 {
1958 int rc = 0;
1959
1960 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
1961 CPUSTAT_SM |
1962 CPUSTAT_STOPPED);
1963
1964 if (test_kvm_facility(vcpu->kvm, 78))
1965 atomic_or(CPUSTAT_GED2, &vcpu->arch.sie_block->cpuflags);
1966 else if (test_kvm_facility(vcpu->kvm, 8))
1967 atomic_or(CPUSTAT_GED, &vcpu->arch.sie_block->cpuflags);
1968
1969 kvm_s390_vcpu_setup_model(vcpu);
1970
1971 /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
1972 if (MACHINE_HAS_ESOP)
1973 vcpu->arch.sie_block->ecb |= 0x02;
1974 if (test_kvm_facility(vcpu->kvm, 9))
1975 vcpu->arch.sie_block->ecb |= 0x04;
1976 if (test_kvm_facility(vcpu->kvm, 73))
1977 vcpu->arch.sie_block->ecb |= 0x10;
1978
1979 if (test_kvm_facility(vcpu->kvm, 8) && sclp.has_pfmfi)
1980 vcpu->arch.sie_block->ecb2 |= 0x08;
1981 if (test_kvm_facility(vcpu->kvm, 130))
1982 vcpu->arch.sie_block->ecb2 |= 0x20;
1983 vcpu->arch.sie_block->eca = 0x1002000U;
1984 if (sclp.has_cei)
1985 vcpu->arch.sie_block->eca |= 0x80000000U;
1986 if (sclp.has_ib)
1987 vcpu->arch.sie_block->eca |= 0x40000000U;
1988 if (sclp.has_siif)
1989 vcpu->arch.sie_block->eca |= 1;
1990 if (sclp.has_sigpif)
1991 vcpu->arch.sie_block->eca |= 0x10000000U;
1992 if (test_kvm_facility(vcpu->kvm, 129)) {
1993 vcpu->arch.sie_block->eca |= 0x00020000;
1994 vcpu->arch.sie_block->ecd |= 0x20000000;
1995 }
1996 vcpu->arch.sie_block->riccbd = (unsigned long) &vcpu->run->s.regs.riccb;
1997 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
1998
1999 if (vcpu->kvm->arch.use_cmma) {
2000 rc = kvm_s390_vcpu_setup_cmma(vcpu);
2001 if (rc)
2002 return rc;
2003 }
2004 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
2005 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
2006
2007 kvm_s390_vcpu_crypto_setup(vcpu);
2008
2009 return rc;
2010 }
2011
2012 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
2013 unsigned int id)
2014 {
2015 struct kvm_vcpu *vcpu;
2016 struct sie_page *sie_page;
2017 int rc = -EINVAL;
2018
2019 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
2020 goto out;
2021
2022 rc = -ENOMEM;
2023
2024 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2025 if (!vcpu)
2026 goto out;
2027
2028 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
2029 if (!sie_page)
2030 goto out_free_cpu;
2031
2032 vcpu->arch.sie_block = &sie_page->sie_block;
2033 vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
2034
2035 /* the real guest size will always be smaller than msl */
2036 vcpu->arch.sie_block->mso = 0;
2037 vcpu->arch.sie_block->msl = sclp.hamax;
2038
2039 vcpu->arch.sie_block->icpua = id;
2040 spin_lock_init(&vcpu->arch.local_int.lock);
2041 vcpu->arch.local_int.float_int = &kvm->arch.float_int;
2042 vcpu->arch.local_int.wq = &vcpu->wq;
2043 vcpu->arch.local_int.cpuflags = &vcpu->arch.sie_block->cpuflags;
2044 seqcount_init(&vcpu->arch.cputm_seqcount);
2045
2046 rc = kvm_vcpu_init(vcpu, kvm, id);
2047 if (rc)
2048 goto out_free_sie_block;
2049 VM_EVENT(kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK", id, vcpu,
2050 vcpu->arch.sie_block);
2051 trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
2052
2053 return vcpu;
2054 out_free_sie_block:
2055 free_page((unsigned long)(vcpu->arch.sie_block));
2056 out_free_cpu:
2057 kmem_cache_free(kvm_vcpu_cache, vcpu);
2058 out:
2059 return ERR_PTR(rc);
2060 }
2061
2062 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
2063 {
2064 return kvm_s390_vcpu_has_irq(vcpu, 0);
2065 }
2066
2067 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
2068 {
2069 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2070 exit_sie(vcpu);
2071 }
2072
2073 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
2074 {
2075 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2076 }
2077
2078 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
2079 {
2080 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2081 exit_sie(vcpu);
2082 }
2083
2084 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
2085 {
2086 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2087 }
2088
2089 /*
2090 * Kick a guest cpu out of SIE and wait until SIE is not running.
2091 * If the CPU is not running (e.g. waiting as idle) the function will
2092 * return immediately. */
2093 void exit_sie(struct kvm_vcpu *vcpu)
2094 {
2095 atomic_or(CPUSTAT_STOP_INT, &vcpu->arch.sie_block->cpuflags);
2096 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
2097 cpu_relax();
2098 }
2099
2100 /* Kick a guest cpu out of SIE to process a request synchronously */
2101 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
2102 {
2103 kvm_make_request(req, vcpu);
2104 kvm_s390_vcpu_request(vcpu);
2105 }
2106
2107 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
2108 unsigned long end)
2109 {
2110 struct kvm *kvm = gmap->private;
2111 struct kvm_vcpu *vcpu;
2112 unsigned long prefix;
2113 int i;
2114
2115 if (gmap_is_shadow(gmap))
2116 return;
2117 if (start >= 1UL << 31)
2118 /* We are only interested in prefix pages */
2119 return;
2120 kvm_for_each_vcpu(i, vcpu, kvm) {
2121 /* match against both prefix pages */
2122 prefix = kvm_s390_get_prefix(vcpu);
2123 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
2124 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
2125 start, end);
2126 kvm_s390_sync_request(KVM_REQ_MMU_RELOAD, vcpu);
2127 }
2128 }
2129 }
2130
2131 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
2132 {
2133 /* kvm common code refers to this, but never calls it */
2134 BUG();
2135 return 0;
2136 }
2137
2138 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
2139 struct kvm_one_reg *reg)
2140 {
2141 int r = -EINVAL;
2142
2143 switch (reg->id) {
2144 case KVM_REG_S390_TODPR:
2145 r = put_user(vcpu->arch.sie_block->todpr,
2146 (u32 __user *)reg->addr);
2147 break;
2148 case KVM_REG_S390_EPOCHDIFF:
2149 r = put_user(vcpu->arch.sie_block->epoch,
2150 (u64 __user *)reg->addr);
2151 break;
2152 case KVM_REG_S390_CPU_TIMER:
2153 r = put_user(kvm_s390_get_cpu_timer(vcpu),
2154 (u64 __user *)reg->addr);
2155 break;
2156 case KVM_REG_S390_CLOCK_COMP:
2157 r = put_user(vcpu->arch.sie_block->ckc,
2158 (u64 __user *)reg->addr);
2159 break;
2160 case KVM_REG_S390_PFTOKEN:
2161 r = put_user(vcpu->arch.pfault_token,
2162 (u64 __user *)reg->addr);
2163 break;
2164 case KVM_REG_S390_PFCOMPARE:
2165 r = put_user(vcpu->arch.pfault_compare,
2166 (u64 __user *)reg->addr);
2167 break;
2168 case KVM_REG_S390_PFSELECT:
2169 r = put_user(vcpu->arch.pfault_select,
2170 (u64 __user *)reg->addr);
2171 break;
2172 case KVM_REG_S390_PP:
2173 r = put_user(vcpu->arch.sie_block->pp,
2174 (u64 __user *)reg->addr);
2175 break;
2176 case KVM_REG_S390_GBEA:
2177 r = put_user(vcpu->arch.sie_block->gbea,
2178 (u64 __user *)reg->addr);
2179 break;
2180 default:
2181 break;
2182 }
2183
2184 return r;
2185 }
2186
2187 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
2188 struct kvm_one_reg *reg)
2189 {
2190 int r = -EINVAL;
2191 __u64 val;
2192
2193 switch (reg->id) {
2194 case KVM_REG_S390_TODPR:
2195 r = get_user(vcpu->arch.sie_block->todpr,
2196 (u32 __user *)reg->addr);
2197 break;
2198 case KVM_REG_S390_EPOCHDIFF:
2199 r = get_user(vcpu->arch.sie_block->epoch,
2200 (u64 __user *)reg->addr);
2201 break;
2202 case KVM_REG_S390_CPU_TIMER:
2203 r = get_user(val, (u64 __user *)reg->addr);
2204 if (!r)
2205 kvm_s390_set_cpu_timer(vcpu, val);
2206 break;
2207 case KVM_REG_S390_CLOCK_COMP:
2208 r = get_user(vcpu->arch.sie_block->ckc,
2209 (u64 __user *)reg->addr);
2210 break;
2211 case KVM_REG_S390_PFTOKEN:
2212 r = get_user(vcpu->arch.pfault_token,
2213 (u64 __user *)reg->addr);
2214 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2215 kvm_clear_async_pf_completion_queue(vcpu);
2216 break;
2217 case KVM_REG_S390_PFCOMPARE:
2218 r = get_user(vcpu->arch.pfault_compare,
2219 (u64 __user *)reg->addr);
2220 break;
2221 case KVM_REG_S390_PFSELECT:
2222 r = get_user(vcpu->arch.pfault_select,
2223 (u64 __user *)reg->addr);
2224 break;
2225 case KVM_REG_S390_PP:
2226 r = get_user(vcpu->arch.sie_block->pp,
2227 (u64 __user *)reg->addr);
2228 break;
2229 case KVM_REG_S390_GBEA:
2230 r = get_user(vcpu->arch.sie_block->gbea,
2231 (u64 __user *)reg->addr);
2232 break;
2233 default:
2234 break;
2235 }
2236
2237 return r;
2238 }
2239
2240 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
2241 {
2242 kvm_s390_vcpu_initial_reset(vcpu);
2243 return 0;
2244 }
2245
2246 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2247 {
2248 memcpy(&vcpu->run->s.regs.gprs, &regs->gprs, sizeof(regs->gprs));
2249 return 0;
2250 }
2251
2252 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2253 {
2254 memcpy(&regs->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
2255 return 0;
2256 }
2257
2258 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2259 struct kvm_sregs *sregs)
2260 {
2261 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
2262 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
2263 return 0;
2264 }
2265
2266 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2267 struct kvm_sregs *sregs)
2268 {
2269 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
2270 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
2271 return 0;
2272 }
2273
2274 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2275 {
2276 if (test_fp_ctl(fpu->fpc))
2277 return -EINVAL;
2278 vcpu->run->s.regs.fpc = fpu->fpc;
2279 if (MACHINE_HAS_VX)
2280 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs,
2281 (freg_t *) fpu->fprs);
2282 else
2283 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs));
2284 return 0;
2285 }
2286
2287 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2288 {
2289 /* make sure we have the latest values */
2290 save_fpu_regs();
2291 if (MACHINE_HAS_VX)
2292 convert_vx_to_fp((freg_t *) fpu->fprs,
2293 (__vector128 *) vcpu->run->s.regs.vrs);
2294 else
2295 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs));
2296 fpu->fpc = vcpu->run->s.regs.fpc;
2297 return 0;
2298 }
2299
2300 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
2301 {
2302 int rc = 0;
2303
2304 if (!is_vcpu_stopped(vcpu))
2305 rc = -EBUSY;
2306 else {
2307 vcpu->run->psw_mask = psw.mask;
2308 vcpu->run->psw_addr = psw.addr;
2309 }
2310 return rc;
2311 }
2312
2313 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2314 struct kvm_translation *tr)
2315 {
2316 return -EINVAL; /* not implemented yet */
2317 }
2318
2319 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
2320 KVM_GUESTDBG_USE_HW_BP | \
2321 KVM_GUESTDBG_ENABLE)
2322
2323 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
2324 struct kvm_guest_debug *dbg)
2325 {
2326 int rc = 0;
2327
2328 vcpu->guest_debug = 0;
2329 kvm_s390_clear_bp_data(vcpu);
2330
2331 if (dbg->control & ~VALID_GUESTDBG_FLAGS)
2332 return -EINVAL;
2333 if (!sclp.has_gpere)
2334 return -EINVAL;
2335
2336 if (dbg->control & KVM_GUESTDBG_ENABLE) {
2337 vcpu->guest_debug = dbg->control;
2338 /* enforce guest PER */
2339 atomic_or(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2340
2341 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
2342 rc = kvm_s390_import_bp_data(vcpu, dbg);
2343 } else {
2344 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2345 vcpu->arch.guestdbg.last_bp = 0;
2346 }
2347
2348 if (rc) {
2349 vcpu->guest_debug = 0;
2350 kvm_s390_clear_bp_data(vcpu);
2351 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2352 }
2353
2354 return rc;
2355 }
2356
2357 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2358 struct kvm_mp_state *mp_state)
2359 {
2360 /* CHECK_STOP and LOAD are not supported yet */
2361 return is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
2362 KVM_MP_STATE_OPERATING;
2363 }
2364
2365 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2366 struct kvm_mp_state *mp_state)
2367 {
2368 int rc = 0;
2369
2370 /* user space knows about this interface - let it control the state */
2371 vcpu->kvm->arch.user_cpu_state_ctrl = 1;
2372
2373 switch (mp_state->mp_state) {
2374 case KVM_MP_STATE_STOPPED:
2375 kvm_s390_vcpu_stop(vcpu);
2376 break;
2377 case KVM_MP_STATE_OPERATING:
2378 kvm_s390_vcpu_start(vcpu);
2379 break;
2380 case KVM_MP_STATE_LOAD:
2381 case KVM_MP_STATE_CHECK_STOP:
2382 /* fall through - CHECK_STOP and LOAD are not supported yet */
2383 default:
2384 rc = -ENXIO;
2385 }
2386
2387 return rc;
2388 }
2389
2390 static bool ibs_enabled(struct kvm_vcpu *vcpu)
2391 {
2392 return atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_IBS;
2393 }
2394
2395 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
2396 {
2397 retry:
2398 kvm_s390_vcpu_request_handled(vcpu);
2399 if (!vcpu->requests)
2400 return 0;
2401 /*
2402 * We use MMU_RELOAD just to re-arm the ipte notifier for the
2403 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
2404 * This ensures that the ipte instruction for this request has
2405 * already finished. We might race against a second unmapper that
2406 * wants to set the blocking bit. Lets just retry the request loop.
2407 */
2408 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
2409 int rc;
2410 rc = gmap_mprotect_notify(vcpu->arch.gmap,
2411 kvm_s390_get_prefix(vcpu),
2412 PAGE_SIZE * 2, PROT_WRITE);
2413 if (rc) {
2414 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
2415 return rc;
2416 }
2417 goto retry;
2418 }
2419
2420 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2421 vcpu->arch.sie_block->ihcpu = 0xffff;
2422 goto retry;
2423 }
2424
2425 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
2426 if (!ibs_enabled(vcpu)) {
2427 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
2428 atomic_or(CPUSTAT_IBS,
2429 &vcpu->arch.sie_block->cpuflags);
2430 }
2431 goto retry;
2432 }
2433
2434 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
2435 if (ibs_enabled(vcpu)) {
2436 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
2437 atomic_andnot(CPUSTAT_IBS,
2438 &vcpu->arch.sie_block->cpuflags);
2439 }
2440 goto retry;
2441 }
2442
2443 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
2444 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
2445 goto retry;
2446 }
2447
2448 /* nothing to do, just clear the request */
2449 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
2450
2451 return 0;
2452 }
2453
2454 void kvm_s390_set_tod_clock(struct kvm *kvm, u64 tod)
2455 {
2456 struct kvm_vcpu *vcpu;
2457 int i;
2458
2459 mutex_lock(&kvm->lock);
2460 preempt_disable();
2461 kvm->arch.epoch = tod - get_tod_clock();
2462 kvm_s390_vcpu_block_all(kvm);
2463 kvm_for_each_vcpu(i, vcpu, kvm)
2464 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
2465 kvm_s390_vcpu_unblock_all(kvm);
2466 preempt_enable();
2467 mutex_unlock(&kvm->lock);
2468 }
2469
2470 /**
2471 * kvm_arch_fault_in_page - fault-in guest page if necessary
2472 * @vcpu: The corresponding virtual cpu
2473 * @gpa: Guest physical address
2474 * @writable: Whether the page should be writable or not
2475 *
2476 * Make sure that a guest page has been faulted-in on the host.
2477 *
2478 * Return: Zero on success, negative error code otherwise.
2479 */
2480 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
2481 {
2482 return gmap_fault(vcpu->arch.gmap, gpa,
2483 writable ? FAULT_FLAG_WRITE : 0);
2484 }
2485
2486 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
2487 unsigned long token)
2488 {
2489 struct kvm_s390_interrupt inti;
2490 struct kvm_s390_irq irq;
2491
2492 if (start_token) {
2493 irq.u.ext.ext_params2 = token;
2494 irq.type = KVM_S390_INT_PFAULT_INIT;
2495 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
2496 } else {
2497 inti.type = KVM_S390_INT_PFAULT_DONE;
2498 inti.parm64 = token;
2499 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
2500 }
2501 }
2502
2503 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
2504 struct kvm_async_pf *work)
2505 {
2506 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
2507 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
2508 }
2509
2510 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
2511 struct kvm_async_pf *work)
2512 {
2513 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
2514 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
2515 }
2516
2517 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
2518 struct kvm_async_pf *work)
2519 {
2520 /* s390 will always inject the page directly */
2521 }
2522
2523 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
2524 {
2525 /*
2526 * s390 will always inject the page directly,
2527 * but we still want check_async_completion to cleanup
2528 */
2529 return true;
2530 }
2531
2532 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
2533 {
2534 hva_t hva;
2535 struct kvm_arch_async_pf arch;
2536 int rc;
2537
2538 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2539 return 0;
2540 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
2541 vcpu->arch.pfault_compare)
2542 return 0;
2543 if (psw_extint_disabled(vcpu))
2544 return 0;
2545 if (kvm_s390_vcpu_has_irq(vcpu, 0))
2546 return 0;
2547 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
2548 return 0;
2549 if (!vcpu->arch.gmap->pfault_enabled)
2550 return 0;
2551
2552 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
2553 hva += current->thread.gmap_addr & ~PAGE_MASK;
2554 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
2555 return 0;
2556
2557 rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
2558 return rc;
2559 }
2560
2561 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
2562 {
2563 int rc, cpuflags;
2564
2565 /*
2566 * On s390 notifications for arriving pages will be delivered directly
2567 * to the guest but the house keeping for completed pfaults is
2568 * handled outside the worker.
2569 */
2570 kvm_check_async_pf_completion(vcpu);
2571
2572 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
2573 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
2574
2575 if (need_resched())
2576 schedule();
2577
2578 if (test_cpu_flag(CIF_MCCK_PENDING))
2579 s390_handle_mcck();
2580
2581 if (!kvm_is_ucontrol(vcpu->kvm)) {
2582 rc = kvm_s390_deliver_pending_interrupts(vcpu);
2583 if (rc)
2584 return rc;
2585 }
2586
2587 rc = kvm_s390_handle_requests(vcpu);
2588 if (rc)
2589 return rc;
2590
2591 if (guestdbg_enabled(vcpu)) {
2592 kvm_s390_backup_guest_per_regs(vcpu);
2593 kvm_s390_patch_guest_per_regs(vcpu);
2594 }
2595
2596 vcpu->arch.sie_block->icptcode = 0;
2597 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
2598 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
2599 trace_kvm_s390_sie_enter(vcpu, cpuflags);
2600
2601 return 0;
2602 }
2603
2604 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
2605 {
2606 struct kvm_s390_pgm_info pgm_info = {
2607 .code = PGM_ADDRESSING,
2608 };
2609 u8 opcode, ilen;
2610 int rc;
2611
2612 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
2613 trace_kvm_s390_sie_fault(vcpu);
2614
2615 /*
2616 * We want to inject an addressing exception, which is defined as a
2617 * suppressing or terminating exception. However, since we came here
2618 * by a DAT access exception, the PSW still points to the faulting
2619 * instruction since DAT exceptions are nullifying. So we've got
2620 * to look up the current opcode to get the length of the instruction
2621 * to be able to forward the PSW.
2622 */
2623 rc = read_guest_instr(vcpu, vcpu->arch.sie_block->gpsw.addr, &opcode, 1);
2624 ilen = insn_length(opcode);
2625 if (rc < 0) {
2626 return rc;
2627 } else if (rc) {
2628 /* Instruction-Fetching Exceptions - we can't detect the ilen.
2629 * Forward by arbitrary ilc, injection will take care of
2630 * nullification if necessary.
2631 */
2632 pgm_info = vcpu->arch.pgm;
2633 ilen = 4;
2634 }
2635 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
2636 kvm_s390_forward_psw(vcpu, ilen);
2637 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
2638 }
2639
2640 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
2641 {
2642 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
2643 vcpu->arch.sie_block->icptcode);
2644 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
2645
2646 if (guestdbg_enabled(vcpu))
2647 kvm_s390_restore_guest_per_regs(vcpu);
2648
2649 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
2650 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
2651
2652 if (vcpu->arch.sie_block->icptcode > 0) {
2653 int rc = kvm_handle_sie_intercept(vcpu);
2654
2655 if (rc != -EOPNOTSUPP)
2656 return rc;
2657 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
2658 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
2659 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
2660 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
2661 return -EREMOTE;
2662 } else if (exit_reason != -EFAULT) {
2663 vcpu->stat.exit_null++;
2664 return 0;
2665 } else if (kvm_is_ucontrol(vcpu->kvm)) {
2666 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
2667 vcpu->run->s390_ucontrol.trans_exc_code =
2668 current->thread.gmap_addr;
2669 vcpu->run->s390_ucontrol.pgm_code = 0x10;
2670 return -EREMOTE;
2671 } else if (current->thread.gmap_pfault) {
2672 trace_kvm_s390_major_guest_pfault(vcpu);
2673 current->thread.gmap_pfault = 0;
2674 if (kvm_arch_setup_async_pf(vcpu))
2675 return 0;
2676 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
2677 }
2678 return vcpu_post_run_fault_in_sie(vcpu);
2679 }
2680
2681 static int __vcpu_run(struct kvm_vcpu *vcpu)
2682 {
2683 int rc, exit_reason;
2684
2685 /*
2686 * We try to hold kvm->srcu during most of vcpu_run (except when run-
2687 * ning the guest), so that memslots (and other stuff) are protected
2688 */
2689 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2690
2691 do {
2692 rc = vcpu_pre_run(vcpu);
2693 if (rc)
2694 break;
2695
2696 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2697 /*
2698 * As PF_VCPU will be used in fault handler, between
2699 * guest_enter and guest_exit should be no uaccess.
2700 */
2701 local_irq_disable();
2702 guest_enter_irqoff();
2703 __disable_cpu_timer_accounting(vcpu);
2704 local_irq_enable();
2705 exit_reason = sie64a(vcpu->arch.sie_block,
2706 vcpu->run->s.regs.gprs);
2707 local_irq_disable();
2708 __enable_cpu_timer_accounting(vcpu);
2709 guest_exit_irqoff();
2710 local_irq_enable();
2711 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2712
2713 rc = vcpu_post_run(vcpu, exit_reason);
2714 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
2715
2716 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2717 return rc;
2718 }
2719
2720 static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2721 {
2722 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
2723 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
2724 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
2725 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
2726 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
2727 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
2728 /* some control register changes require a tlb flush */
2729 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2730 }
2731 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
2732 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
2733 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
2734 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
2735 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
2736 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
2737 }
2738 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
2739 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
2740 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
2741 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
2742 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2743 kvm_clear_async_pf_completion_queue(vcpu);
2744 }
2745 /*
2746 * If userspace sets the riccb (e.g. after migration) to a valid state,
2747 * we should enable RI here instead of doing the lazy enablement.
2748 */
2749 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) &&
2750 test_kvm_facility(vcpu->kvm, 64)) {
2751 struct runtime_instr_cb *riccb =
2752 (struct runtime_instr_cb *) &kvm_run->s.regs.riccb;
2753
2754 if (riccb->valid)
2755 vcpu->arch.sie_block->ecb3 |= 0x01;
2756 }
2757 save_access_regs(vcpu->arch.host_acrs);
2758 restore_access_regs(vcpu->run->s.regs.acrs);
2759 /* save host (userspace) fprs/vrs */
2760 save_fpu_regs();
2761 vcpu->arch.host_fpregs.fpc = current->thread.fpu.fpc;
2762 vcpu->arch.host_fpregs.regs = current->thread.fpu.regs;
2763 if (MACHINE_HAS_VX)
2764 current->thread.fpu.regs = vcpu->run->s.regs.vrs;
2765 else
2766 current->thread.fpu.regs = vcpu->run->s.regs.fprs;
2767 current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
2768 if (test_fp_ctl(current->thread.fpu.fpc))
2769 /* User space provided an invalid FPC, let's clear it */
2770 current->thread.fpu.fpc = 0;
2771
2772 kvm_run->kvm_dirty_regs = 0;
2773 }
2774
2775 static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2776 {
2777 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
2778 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
2779 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
2780 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
2781 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
2782 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
2783 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
2784 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
2785 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
2786 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
2787 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
2788 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
2789 save_access_regs(vcpu->run->s.regs.acrs);
2790 restore_access_regs(vcpu->arch.host_acrs);
2791 /* Save guest register state */
2792 save_fpu_regs();
2793 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
2794 /* Restore will be done lazily at return */
2795 current->thread.fpu.fpc = vcpu->arch.host_fpregs.fpc;
2796 current->thread.fpu.regs = vcpu->arch.host_fpregs.regs;
2797
2798 }
2799
2800 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2801 {
2802 int rc;
2803 sigset_t sigsaved;
2804
2805 if (kvm_run->immediate_exit)
2806 return -EINTR;
2807
2808 if (guestdbg_exit_pending(vcpu)) {
2809 kvm_s390_prepare_debug_exit(vcpu);
2810 return 0;
2811 }
2812
2813 if (vcpu->sigset_active)
2814 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2815
2816 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
2817 kvm_s390_vcpu_start(vcpu);
2818 } else if (is_vcpu_stopped(vcpu)) {
2819 pr_err_ratelimited("can't run stopped vcpu %d\n",
2820 vcpu->vcpu_id);
2821 return -EINVAL;
2822 }
2823
2824 sync_regs(vcpu, kvm_run);
2825 enable_cpu_timer_accounting(vcpu);
2826
2827 might_fault();
2828 rc = __vcpu_run(vcpu);
2829
2830 if (signal_pending(current) && !rc) {
2831 kvm_run->exit_reason = KVM_EXIT_INTR;
2832 rc = -EINTR;
2833 }
2834
2835 if (guestdbg_exit_pending(vcpu) && !rc) {
2836 kvm_s390_prepare_debug_exit(vcpu);
2837 rc = 0;
2838 }
2839
2840 if (rc == -EREMOTE) {
2841 /* userspace support is needed, kvm_run has been prepared */
2842 rc = 0;
2843 }
2844
2845 disable_cpu_timer_accounting(vcpu);
2846 store_regs(vcpu, kvm_run);
2847
2848 if (vcpu->sigset_active)
2849 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2850
2851 vcpu->stat.exit_userspace++;
2852 return rc;
2853 }
2854
2855 /*
2856 * store status at address
2857 * we use have two special cases:
2858 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
2859 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
2860 */
2861 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
2862 {
2863 unsigned char archmode = 1;
2864 freg_t fprs[NUM_FPRS];
2865 unsigned int px;
2866 u64 clkcomp, cputm;
2867 int rc;
2868
2869 px = kvm_s390_get_prefix(vcpu);
2870 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
2871 if (write_guest_abs(vcpu, 163, &archmode, 1))
2872 return -EFAULT;
2873 gpa = 0;
2874 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
2875 if (write_guest_real(vcpu, 163, &archmode, 1))
2876 return -EFAULT;
2877 gpa = px;
2878 } else
2879 gpa -= __LC_FPREGS_SAVE_AREA;
2880
2881 /* manually convert vector registers if necessary */
2882 if (MACHINE_HAS_VX) {
2883 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
2884 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2885 fprs, 128);
2886 } else {
2887 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2888 vcpu->run->s.regs.fprs, 128);
2889 }
2890 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
2891 vcpu->run->s.regs.gprs, 128);
2892 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
2893 &vcpu->arch.sie_block->gpsw, 16);
2894 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
2895 &px, 4);
2896 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
2897 &vcpu->run->s.regs.fpc, 4);
2898 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
2899 &vcpu->arch.sie_block->todpr, 4);
2900 cputm = kvm_s390_get_cpu_timer(vcpu);
2901 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
2902 &cputm, 8);
2903 clkcomp = vcpu->arch.sie_block->ckc >> 8;
2904 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
2905 &clkcomp, 8);
2906 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
2907 &vcpu->run->s.regs.acrs, 64);
2908 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
2909 &vcpu->arch.sie_block->gcr, 128);
2910 return rc ? -EFAULT : 0;
2911 }
2912
2913 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
2914 {
2915 /*
2916 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
2917 * switch in the run ioctl. Let's update our copies before we save
2918 * it into the save area
2919 */
2920 save_fpu_regs();
2921 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
2922 save_access_regs(vcpu->run->s.regs.acrs);
2923
2924 return kvm_s390_store_status_unloaded(vcpu, addr);
2925 }
2926
2927 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2928 {
2929 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
2930 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
2931 }
2932
2933 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
2934 {
2935 unsigned int i;
2936 struct kvm_vcpu *vcpu;
2937
2938 kvm_for_each_vcpu(i, vcpu, kvm) {
2939 __disable_ibs_on_vcpu(vcpu);
2940 }
2941 }
2942
2943 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2944 {
2945 if (!sclp.has_ibs)
2946 return;
2947 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
2948 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
2949 }
2950
2951 void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
2952 {
2953 int i, online_vcpus, started_vcpus = 0;
2954
2955 if (!is_vcpu_stopped(vcpu))
2956 return;
2957
2958 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
2959 /* Only one cpu at a time may enter/leave the STOPPED state. */
2960 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2961 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2962
2963 for (i = 0; i < online_vcpus; i++) {
2964 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i]))
2965 started_vcpus++;
2966 }
2967
2968 if (started_vcpus == 0) {
2969 /* we're the only active VCPU -> speed it up */
2970 __enable_ibs_on_vcpu(vcpu);
2971 } else if (started_vcpus == 1) {
2972 /*
2973 * As we are starting a second VCPU, we have to disable
2974 * the IBS facility on all VCPUs to remove potentially
2975 * oustanding ENABLE requests.
2976 */
2977 __disable_ibs_on_all_vcpus(vcpu->kvm);
2978 }
2979
2980 atomic_andnot(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2981 /*
2982 * Another VCPU might have used IBS while we were offline.
2983 * Let's play safe and flush the VCPU at startup.
2984 */
2985 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2986 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2987 return;
2988 }
2989
2990 void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
2991 {
2992 int i, online_vcpus, started_vcpus = 0;
2993 struct kvm_vcpu *started_vcpu = NULL;
2994
2995 if (is_vcpu_stopped(vcpu))
2996 return;
2997
2998 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
2999 /* Only one cpu at a time may enter/leave the STOPPED state. */
3000 spin_lock(&vcpu->kvm->arch.start_stop_lock);
3001 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
3002
3003 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
3004 kvm_s390_clear_stop_irq(vcpu);
3005
3006 atomic_or(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
3007 __disable_ibs_on_vcpu(vcpu);
3008
3009 for (i = 0; i < online_vcpus; i++) {
3010 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) {
3011 started_vcpus++;
3012 started_vcpu = vcpu->kvm->vcpus[i];
3013 }
3014 }
3015
3016 if (started_vcpus == 1) {
3017 /*
3018 * As we only have one VCPU left, we want to enable the
3019 * IBS facility for that VCPU to speed it up.
3020 */
3021 __enable_ibs_on_vcpu(started_vcpu);
3022 }
3023
3024 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
3025 return;
3026 }
3027
3028 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3029 struct kvm_enable_cap *cap)
3030 {
3031 int r;
3032
3033 if (cap->flags)
3034 return -EINVAL;
3035
3036 switch (cap->cap) {
3037 case KVM_CAP_S390_CSS_SUPPORT:
3038 if (!vcpu->kvm->arch.css_support) {
3039 vcpu->kvm->arch.css_support = 1;
3040 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
3041 trace_kvm_s390_enable_css(vcpu->kvm);
3042 }
3043 r = 0;
3044 break;
3045 default:
3046 r = -EINVAL;
3047 break;
3048 }
3049 return r;
3050 }
3051
3052 static long kvm_s390_guest_mem_op(struct kvm_vcpu *vcpu,
3053 struct kvm_s390_mem_op *mop)
3054 {
3055 void __user *uaddr = (void __user *)mop->buf;
3056 void *tmpbuf = NULL;
3057 int r, srcu_idx;
3058 const u64 supported_flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION
3059 | KVM_S390_MEMOP_F_CHECK_ONLY;
3060
3061 if (mop->flags & ~supported_flags)
3062 return -EINVAL;
3063
3064 if (mop->size > MEM_OP_MAX_SIZE)
3065 return -E2BIG;
3066
3067 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
3068 tmpbuf = vmalloc(mop->size);
3069 if (!tmpbuf)
3070 return -ENOMEM;
3071 }
3072
3073 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3074
3075 switch (mop->op) {
3076 case KVM_S390_MEMOP_LOGICAL_READ:
3077 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3078 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3079 mop->size, GACC_FETCH);
3080 break;
3081 }
3082 r = read_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3083 if (r == 0) {
3084 if (copy_to_user(uaddr, tmpbuf, mop->size))
3085 r = -EFAULT;
3086 }
3087 break;
3088 case KVM_S390_MEMOP_LOGICAL_WRITE:
3089 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3090 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3091 mop->size, GACC_STORE);
3092 break;
3093 }
3094 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
3095 r = -EFAULT;
3096 break;
3097 }
3098 r = write_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3099 break;
3100 default:
3101 r = -EINVAL;
3102 }
3103
3104 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
3105
3106 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
3107 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
3108
3109 vfree(tmpbuf);
3110 return r;
3111 }
3112
3113 long kvm_arch_vcpu_ioctl(struct file *filp,
3114 unsigned int ioctl, unsigned long arg)
3115 {
3116 struct kvm_vcpu *vcpu = filp->private_data;
3117 void __user *argp = (void __user *)arg;
3118 int idx;
3119 long r;
3120
3121 switch (ioctl) {
3122 case KVM_S390_IRQ: {
3123 struct kvm_s390_irq s390irq;
3124
3125 r = -EFAULT;
3126 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
3127 break;
3128 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
3129 break;
3130 }
3131 case KVM_S390_INTERRUPT: {
3132 struct kvm_s390_interrupt s390int;
3133 struct kvm_s390_irq s390irq;
3134
3135 r = -EFAULT;
3136 if (copy_from_user(&s390int, argp, sizeof(s390int)))
3137 break;
3138 if (s390int_to_s390irq(&s390int, &s390irq))
3139 return -EINVAL;
3140 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
3141 break;
3142 }
3143 case KVM_S390_STORE_STATUS:
3144 idx = srcu_read_lock(&vcpu->kvm->srcu);
3145 r = kvm_s390_vcpu_store_status(vcpu, arg);
3146 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3147 break;
3148 case KVM_S390_SET_INITIAL_PSW: {
3149 psw_t psw;
3150
3151 r = -EFAULT;
3152 if (copy_from_user(&psw, argp, sizeof(psw)))
3153 break;
3154 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
3155 break;
3156 }
3157 case KVM_S390_INITIAL_RESET:
3158 r = kvm_arch_vcpu_ioctl_initial_reset(vcpu);
3159 break;
3160 case KVM_SET_ONE_REG:
3161 case KVM_GET_ONE_REG: {
3162 struct kvm_one_reg reg;
3163 r = -EFAULT;
3164 if (copy_from_user(&reg, argp, sizeof(reg)))
3165 break;
3166 if (ioctl == KVM_SET_ONE_REG)
3167 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, &reg);
3168 else
3169 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, &reg);
3170 break;
3171 }
3172 #ifdef CONFIG_KVM_S390_UCONTROL
3173 case KVM_S390_UCAS_MAP: {
3174 struct kvm_s390_ucas_mapping ucasmap;
3175
3176 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3177 r = -EFAULT;
3178 break;
3179 }
3180
3181 if (!kvm_is_ucontrol(vcpu->kvm)) {
3182 r = -EINVAL;
3183 break;
3184 }
3185
3186 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
3187 ucasmap.vcpu_addr, ucasmap.length);
3188 break;
3189 }
3190 case KVM_S390_UCAS_UNMAP: {
3191 struct kvm_s390_ucas_mapping ucasmap;
3192
3193 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3194 r = -EFAULT;
3195 break;
3196 }
3197
3198 if (!kvm_is_ucontrol(vcpu->kvm)) {
3199 r = -EINVAL;
3200 break;
3201 }
3202
3203 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
3204 ucasmap.length);
3205 break;
3206 }
3207 #endif
3208 case KVM_S390_VCPU_FAULT: {
3209 r = gmap_fault(vcpu->arch.gmap, arg, 0);
3210 break;
3211 }
3212 case KVM_ENABLE_CAP:
3213 {
3214 struct kvm_enable_cap cap;
3215 r = -EFAULT;
3216 if (copy_from_user(&cap, argp, sizeof(cap)))
3217 break;
3218 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3219 break;
3220 }
3221 case KVM_S390_MEM_OP: {
3222 struct kvm_s390_mem_op mem_op;
3223
3224 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
3225 r = kvm_s390_guest_mem_op(vcpu, &mem_op);
3226 else
3227 r = -EFAULT;
3228 break;
3229 }
3230 case KVM_S390_SET_IRQ_STATE: {
3231 struct kvm_s390_irq_state irq_state;
3232
3233 r = -EFAULT;
3234 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3235 break;
3236 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
3237 irq_state.len == 0 ||
3238 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
3239 r = -EINVAL;
3240 break;
3241 }
3242 r = kvm_s390_set_irq_state(vcpu,
3243 (void __user *) irq_state.buf,
3244 irq_state.len);
3245 break;
3246 }
3247 case KVM_S390_GET_IRQ_STATE: {
3248 struct kvm_s390_irq_state irq_state;
3249
3250 r = -EFAULT;
3251 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3252 break;
3253 if (irq_state.len == 0) {
3254 r = -EINVAL;
3255 break;
3256 }
3257 r = kvm_s390_get_irq_state(vcpu,
3258 (__u8 __user *) irq_state.buf,
3259 irq_state.len);
3260 break;
3261 }
3262 default:
3263 r = -ENOTTY;
3264 }
3265 return r;
3266 }
3267
3268 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3269 {
3270 #ifdef CONFIG_KVM_S390_UCONTROL
3271 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
3272 && (kvm_is_ucontrol(vcpu->kvm))) {
3273 vmf->page = virt_to_page(vcpu->arch.sie_block);
3274 get_page(vmf->page);
3275 return 0;
3276 }
3277 #endif
3278 return VM_FAULT_SIGBUS;
3279 }
3280
3281 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
3282 unsigned long npages)
3283 {
3284 return 0;
3285 }
3286
3287 /* Section: memory related */
3288 int kvm_arch_prepare_memory_region(struct kvm *kvm,
3289 struct kvm_memory_slot *memslot,
3290 const struct kvm_userspace_memory_region *mem,
3291 enum kvm_mr_change change)
3292 {
3293 /* A few sanity checks. We can have memory slots which have to be
3294 located/ended at a segment boundary (1MB). The memory in userland is
3295 ok to be fragmented into various different vmas. It is okay to mmap()
3296 and munmap() stuff in this slot after doing this call at any time */
3297
3298 if (mem->userspace_addr & 0xffffful)
3299 return -EINVAL;
3300
3301 if (mem->memory_size & 0xffffful)
3302 return -EINVAL;
3303
3304 if (mem->guest_phys_addr + mem->memory_size > kvm->arch.mem_limit)
3305 return -EINVAL;
3306
3307 return 0;
3308 }
3309
3310 void kvm_arch_commit_memory_region(struct kvm *kvm,
3311 const struct kvm_userspace_memory_region *mem,
3312 const struct kvm_memory_slot *old,
3313 const struct kvm_memory_slot *new,
3314 enum kvm_mr_change change)
3315 {
3316 int rc;
3317
3318 /* If the basics of the memslot do not change, we do not want
3319 * to update the gmap. Every update causes several unnecessary
3320 * segment translation exceptions. This is usually handled just
3321 * fine by the normal fault handler + gmap, but it will also
3322 * cause faults on the prefix page of running guest CPUs.
3323 */
3324 if (old->userspace_addr == mem->userspace_addr &&
3325 old->base_gfn * PAGE_SIZE == mem->guest_phys_addr &&
3326 old->npages * PAGE_SIZE == mem->memory_size)
3327 return;
3328
3329 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr,
3330 mem->guest_phys_addr, mem->memory_size);
3331 if (rc)
3332 pr_warn("failed to commit memory region\n");
3333 return;
3334 }
3335
3336 static inline unsigned long nonhyp_mask(int i)
3337 {
3338 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
3339
3340 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
3341 }
3342
3343 void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu)
3344 {
3345 vcpu->valid_wakeup = false;
3346 }
3347
3348 static int __init kvm_s390_init(void)
3349 {
3350 int i;
3351
3352 if (!sclp.has_sief2) {
3353 pr_info("SIE not available\n");
3354 return -ENODEV;
3355 }
3356
3357 for (i = 0; i < 16; i++)
3358 kvm_s390_fac_list_mask[i] |=
3359 S390_lowcore.stfle_fac_list[i] & nonhyp_mask(i);
3360
3361 return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
3362 }
3363
3364 static void __exit kvm_s390_exit(void)
3365 {
3366 kvm_exit();
3367 }
3368
3369 module_init(kvm_s390_init);
3370 module_exit(kvm_s390_exit);
3371
3372 /*
3373 * Enable autoloading of the kvm module.
3374 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
3375 * since x86 takes a different approach.
3376 */
3377 #include <linux/miscdevice.h>
3378 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3379 MODULE_ALIAS("devname:kvm");
Linux with added FPC-III support