?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
exec: Don't reset euid and egid when the tracee has CAP_SETUID
[cris-mirror.git] / arch / s390 / kvm / kvm-s390.c
blobbec71e902be3f8030697084540130227ecb55088
1 /*
2 * hosting zSeries kernel virtual machines
3 *
4 * Copyright IBM Corp. 2008, 2009
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License (version 2 only)
8 * as published by the Free Software Foundation.
9 *
10 * Author(s): Carsten Otte <cotte@de.ibm.com>
11 * Christian Borntraeger <borntraeger@de.ibm.com>
12 * Heiko Carstens <heiko.carstens@de.ibm.com>
13 * Christian Ehrhardt <ehrhardt@de.ibm.com>
14 * Jason J. Herne <jjherne@us.ibm.com>
15 */
16
17 #include <linux/compiler.h>
18 #include <linux/err.h>
19 #include <linux/fs.h>
20 #include <linux/hrtimer.h>
21 #include <linux/init.h>
22 #include <linux/kvm.h>
23 #include <linux/kvm_host.h>
24 #include <linux/mman.h>
25 #include <linux/module.h>
26 #include <linux/random.h>
27 #include <linux/slab.h>
28 #include <linux/timer.h>
29 #include <linux/vmalloc.h>
30 #include <linux/bitmap.h>
31 #include <asm/asm-offsets.h>
32 #include <asm/lowcore.h>
33 #include <asm/stp.h>
34 #include <asm/pgtable.h>
35 #include <asm/gmap.h>
36 #include <asm/nmi.h>
37 #include <asm/switch_to.h>
38 #include <asm/isc.h>
39 #include <asm/sclp.h>
40 #include <asm/cpacf.h>
41 #include <asm/timex.h>
42 #include "kvm-s390.h"
43 #include "gaccess.h"
44
45 #define KMSG_COMPONENT "kvm-s390"
46 #undef pr_fmt
47 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
48
49 #define CREATE_TRACE_POINTS
50 #include "trace.h"
51 #include "trace-s390.h"
52
53 #define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */
54 #define LOCAL_IRQS 32
55 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
56 (KVM_MAX_VCPUS + LOCAL_IRQS))
57
58 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
59
60 struct kvm_stats_debugfs_item debugfs_entries[] = {
61 { "userspace_handled", VCPU_STAT(exit_userspace) },
62 { "exit_null", VCPU_STAT(exit_null) },
63 { "exit_validity", VCPU_STAT(exit_validity) },
64 { "exit_stop_request", VCPU_STAT(exit_stop_request) },
65 { "exit_external_request", VCPU_STAT(exit_external_request) },
66 { "exit_external_interrupt", VCPU_STAT(exit_external_interrupt) },
67 { "exit_instruction", VCPU_STAT(exit_instruction) },
68 { "exit_pei", VCPU_STAT(exit_pei) },
69 { "exit_program_interruption", VCPU_STAT(exit_program_interruption) },
70 { "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program) },
71 { "exit_operation_exception", VCPU_STAT(exit_operation_exception) },
72 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
73 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
74 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
75 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
76 { "instruction_lctlg", VCPU_STAT(instruction_lctlg) },
77 { "instruction_lctl", VCPU_STAT(instruction_lctl) },
78 { "instruction_stctl", VCPU_STAT(instruction_stctl) },
79 { "instruction_stctg", VCPU_STAT(instruction_stctg) },
80 { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal) },
81 { "deliver_external_call", VCPU_STAT(deliver_external_call) },
82 { "deliver_service_signal", VCPU_STAT(deliver_service_signal) },
83 { "deliver_virtio_interrupt", VCPU_STAT(deliver_virtio_interrupt) },
84 { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal) },
85 { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal) },
86 { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal) },
87 { "deliver_program_interruption", VCPU_STAT(deliver_program_int) },
88 { "exit_wait_state", VCPU_STAT(exit_wait_state) },
89 { "instruction_pfmf", VCPU_STAT(instruction_pfmf) },
90 { "instruction_stidp", VCPU_STAT(instruction_stidp) },
91 { "instruction_spx", VCPU_STAT(instruction_spx) },
92 { "instruction_stpx", VCPU_STAT(instruction_stpx) },
93 { "instruction_stap", VCPU_STAT(instruction_stap) },
94 { "instruction_storage_key", VCPU_STAT(instruction_storage_key) },
95 { "instruction_ipte_interlock", VCPU_STAT(instruction_ipte_interlock) },
96 { "instruction_stsch", VCPU_STAT(instruction_stsch) },
97 { "instruction_chsc", VCPU_STAT(instruction_chsc) },
98 { "instruction_essa", VCPU_STAT(instruction_essa) },
99 { "instruction_stsi", VCPU_STAT(instruction_stsi) },
100 { "instruction_stfl", VCPU_STAT(instruction_stfl) },
101 { "instruction_tprot", VCPU_STAT(instruction_tprot) },
102 { "instruction_sthyi", VCPU_STAT(instruction_sthyi) },
103 { "instruction_sie", VCPU_STAT(instruction_sie) },
104 { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense) },
105 { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running) },
106 { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call) },
107 { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency) },
108 { "instruction_sigp_cond_emergency", VCPU_STAT(instruction_sigp_cond_emergency) },
109 { "instruction_sigp_start", VCPU_STAT(instruction_sigp_start) },
110 { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop) },
111 { "instruction_sigp_stop_store_status", VCPU_STAT(instruction_sigp_stop_store_status) },
112 { "instruction_sigp_store_status", VCPU_STAT(instruction_sigp_store_status) },
113 { "instruction_sigp_store_adtl_status", VCPU_STAT(instruction_sigp_store_adtl_status) },
114 { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch) },
115 { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix) },
116 { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart) },
117 { "instruction_sigp_cpu_reset", VCPU_STAT(instruction_sigp_cpu_reset) },
118 { "instruction_sigp_init_cpu_reset", VCPU_STAT(instruction_sigp_init_cpu_reset) },
119 { "instruction_sigp_unknown", VCPU_STAT(instruction_sigp_unknown) },
120 { "diagnose_10", VCPU_STAT(diagnose_10) },
121 { "diagnose_44", VCPU_STAT(diagnose_44) },
122 { "diagnose_9c", VCPU_STAT(diagnose_9c) },
123 { "diagnose_258", VCPU_STAT(diagnose_258) },
124 { "diagnose_308", VCPU_STAT(diagnose_308) },
125 { "diagnose_500", VCPU_STAT(diagnose_500) },
126 { NULL }
127 };
128
129 /* allow nested virtualization in KVM (if enabled by user space) */
130 static int nested;
131 module_param(nested, int, S_IRUGO);
132 MODULE_PARM_DESC(nested, "Nested virtualization support");
133
134 /* upper facilities limit for kvm */
135 unsigned long kvm_s390_fac_list_mask[16] = { FACILITIES_KVM };
136
137 unsigned long kvm_s390_fac_list_mask_size(void)
138 {
139 BUILD_BUG_ON(ARRAY_SIZE(kvm_s390_fac_list_mask) > S390_ARCH_FAC_MASK_SIZE_U64);
140 return ARRAY_SIZE(kvm_s390_fac_list_mask);
141 }
142
143 /* available cpu features supported by kvm */
144 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
145 /* available subfunctions indicated via query / "test bit" */
146 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
147
148 static struct gmap_notifier gmap_notifier;
149 static struct gmap_notifier vsie_gmap_notifier;
150 debug_info_t *kvm_s390_dbf;
151
152 /* Section: not file related */
153 int kvm_arch_hardware_enable(void)
154 {
155 /* every s390 is virtualization enabled ;-) */
156 return 0;
157 }
158
159 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
160 unsigned long end);
161
162 /*
163 * This callback is executed during stop_machine(). All CPUs are therefore
164 * temporarily stopped. In order not to change guest behavior, we have to
165 * disable preemption whenever we touch the epoch of kvm and the VCPUs,
166 * so a CPU won't be stopped while calculating with the epoch.
167 */
168 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
169 void *v)
170 {
171 struct kvm *kvm;
172 struct kvm_vcpu *vcpu;
173 int i;
174 unsigned long long *delta = v;
175
176 list_for_each_entry(kvm, &vm_list, vm_list) {
177 kvm->arch.epoch -= *delta;
178 kvm_for_each_vcpu(i, vcpu, kvm) {
179 vcpu->arch.sie_block->epoch -= *delta;
180 if (vcpu->arch.cputm_enabled)
181 vcpu->arch.cputm_start += *delta;
182 if (vcpu->arch.vsie_block)
183 vcpu->arch.vsie_block->epoch -= *delta;
184 }
185 }
186 return NOTIFY_OK;
187 }
188
189 static struct notifier_block kvm_clock_notifier = {
190 .notifier_call = kvm_clock_sync,
191 };
192
193 int kvm_arch_hardware_setup(void)
194 {
195 gmap_notifier.notifier_call = kvm_gmap_notifier;
196 gmap_register_pte_notifier(&gmap_notifier);
197 vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
198 gmap_register_pte_notifier(&vsie_gmap_notifier);
199 atomic_notifier_chain_register(&s390_epoch_delta_notifier,
200 &kvm_clock_notifier);
201 return 0;
202 }
203
204 void kvm_arch_hardware_unsetup(void)
205 {
206 gmap_unregister_pte_notifier(&gmap_notifier);
207 gmap_unregister_pte_notifier(&vsie_gmap_notifier);
208 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
209 &kvm_clock_notifier);
210 }
211
212 static void allow_cpu_feat(unsigned long nr)
213 {
214 set_bit_inv(nr, kvm_s390_available_cpu_feat);
215 }
216
217 static inline int plo_test_bit(unsigned char nr)
218 {
219 register unsigned long r0 asm("0") = (unsigned long) nr | 0x100;
220 int cc = 3; /* subfunction not available */
221
222 asm volatile(
223 /* Parameter registers are ignored for "test bit" */
224 " plo 0,0,0,0(0)\n"
225 " ipm %0\n"
226 " srl %0,28\n"
227 : "=d" (cc)
228 : "d" (r0)
229 : "cc");
230 return cc == 0;
231 }
232
233 static void kvm_s390_cpu_feat_init(void)
234 {
235 int i;
236
237 for (i = 0; i < 256; ++i) {
238 if (plo_test_bit(i))
239 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
240 }
241
242 if (test_facility(28)) /* TOD-clock steering */
243 ptff(kvm_s390_available_subfunc.ptff,
244 sizeof(kvm_s390_available_subfunc.ptff),
245 PTFF_QAF);
246
247 if (test_facility(17)) { /* MSA */
248 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
249 kvm_s390_available_subfunc.kmac);
250 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
251 kvm_s390_available_subfunc.kmc);
252 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
253 kvm_s390_available_subfunc.km);
254 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
255 kvm_s390_available_subfunc.kimd);
256 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
257 kvm_s390_available_subfunc.klmd);
258 }
259 if (test_facility(76)) /* MSA3 */
260 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
261 kvm_s390_available_subfunc.pckmo);
262 if (test_facility(77)) { /* MSA4 */
263 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
264 kvm_s390_available_subfunc.kmctr);
265 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
266 kvm_s390_available_subfunc.kmf);
267 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
268 kvm_s390_available_subfunc.kmo);
269 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
270 kvm_s390_available_subfunc.pcc);
271 }
272 if (test_facility(57)) /* MSA5 */
273 __cpacf_query(CPACF_PPNO, (cpacf_mask_t *)
274 kvm_s390_available_subfunc.ppno);
275
276 if (MACHINE_HAS_ESOP)
277 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
278 /*
279 * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
280 * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
281 */
282 if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
283 !test_facility(3) || !nested)
284 return;
285 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
286 if (sclp.has_64bscao)
287 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
288 if (sclp.has_siif)
289 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
290 if (sclp.has_gpere)
291 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
292 if (sclp.has_gsls)
293 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
294 if (sclp.has_ib)
295 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
296 if (sclp.has_cei)
297 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
298 if (sclp.has_ibs)
299 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
300 /*
301 * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
302 * all skey handling functions read/set the skey from the PGSTE
303 * instead of the real storage key.
304 *
305 * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
306 * pages being detected as preserved although they are resident.
307 *
308 * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
309 * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
310 *
311 * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
312 * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
313 * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
314 *
315 * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
316 * cannot easily shadow the SCA because of the ipte lock.
317 */
318 }
319
320 int kvm_arch_init(void *opaque)
321 {
322 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
323 if (!kvm_s390_dbf)
324 return -ENOMEM;
325
326 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view)) {
327 debug_unregister(kvm_s390_dbf);
328 return -ENOMEM;
329 }
330
331 kvm_s390_cpu_feat_init();
332
333 /* Register floating interrupt controller interface. */
334 return kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
335 }
336
337 void kvm_arch_exit(void)
338 {
339 debug_unregister(kvm_s390_dbf);
340 }
341
342 /* Section: device related */
343 long kvm_arch_dev_ioctl(struct file *filp,
344 unsigned int ioctl, unsigned long arg)
345 {
346 if (ioctl == KVM_S390_ENABLE_SIE)
347 return s390_enable_sie();
348 return -EINVAL;
349 }
350
351 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
352 {
353 int r;
354
355 switch (ext) {
356 case KVM_CAP_S390_PSW:
357 case KVM_CAP_S390_GMAP:
358 case KVM_CAP_SYNC_MMU:
359 #ifdef CONFIG_KVM_S390_UCONTROL
360 case KVM_CAP_S390_UCONTROL:
361 #endif
362 case KVM_CAP_ASYNC_PF:
363 case KVM_CAP_SYNC_REGS:
364 case KVM_CAP_ONE_REG:
365 case KVM_CAP_ENABLE_CAP:
366 case KVM_CAP_S390_CSS_SUPPORT:
367 case KVM_CAP_IOEVENTFD:
368 case KVM_CAP_DEVICE_CTRL:
369 case KVM_CAP_ENABLE_CAP_VM:
370 case KVM_CAP_S390_IRQCHIP:
371 case KVM_CAP_VM_ATTRIBUTES:
372 case KVM_CAP_MP_STATE:
373 case KVM_CAP_S390_INJECT_IRQ:
374 case KVM_CAP_S390_USER_SIGP:
375 case KVM_CAP_S390_USER_STSI:
376 case KVM_CAP_S390_SKEYS:
377 case KVM_CAP_S390_IRQ_STATE:
378 case KVM_CAP_S390_USER_INSTR0:
379 r = 1;
380 break;
381 case KVM_CAP_S390_MEM_OP:
382 r = MEM_OP_MAX_SIZE;
383 break;
384 case KVM_CAP_NR_VCPUS:
385 case KVM_CAP_MAX_VCPUS:
386 r = KVM_S390_BSCA_CPU_SLOTS;
387 if (!kvm_s390_use_sca_entries())
388 r = KVM_MAX_VCPUS;
389 else if (sclp.has_esca && sclp.has_64bscao)
390 r = KVM_S390_ESCA_CPU_SLOTS;
391 break;
392 case KVM_CAP_NR_MEMSLOTS:
393 r = KVM_USER_MEM_SLOTS;
394 break;
395 case KVM_CAP_S390_COW:
396 r = MACHINE_HAS_ESOP;
397 break;
398 case KVM_CAP_S390_VECTOR_REGISTERS:
399 r = MACHINE_HAS_VX;
400 break;
401 case KVM_CAP_S390_RI:
402 r = test_facility(64);
403 break;
404 default:
405 r = 0;
406 }
407 return r;
408 }
409
410 static void kvm_s390_sync_dirty_log(struct kvm *kvm,
411 struct kvm_memory_slot *memslot)
412 {
413 gfn_t cur_gfn, last_gfn;
414 unsigned long address;
415 struct gmap *gmap = kvm->arch.gmap;
416
417 /* Loop over all guest pages */
418 last_gfn = memslot->base_gfn + memslot->npages;
419 for (cur_gfn = memslot->base_gfn; cur_gfn <= last_gfn; cur_gfn++) {
420 address = gfn_to_hva_memslot(memslot, cur_gfn);
421
422 if (test_and_clear_guest_dirty(gmap->mm, address))
423 mark_page_dirty(kvm, cur_gfn);
424 if (fatal_signal_pending(current))
425 return;
426 cond_resched();
427 }
428 }
429
430 /* Section: vm related */
431 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
432
433 /*
434 * Get (and clear) the dirty memory log for a memory slot.
435 */
436 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
437 struct kvm_dirty_log *log)
438 {
439 int r;
440 unsigned long n;
441 struct kvm_memslots *slots;
442 struct kvm_memory_slot *memslot;
443 int is_dirty = 0;
444
445 mutex_lock(&kvm->slots_lock);
446
447 r = -EINVAL;
448 if (log->slot >= KVM_USER_MEM_SLOTS)
449 goto out;
450
451 slots = kvm_memslots(kvm);
452 memslot = id_to_memslot(slots, log->slot);
453 r = -ENOENT;
454 if (!memslot->dirty_bitmap)
455 goto out;
456
457 kvm_s390_sync_dirty_log(kvm, memslot);
458 r = kvm_get_dirty_log(kvm, log, &is_dirty);
459 if (r)
460 goto out;
461
462 /* Clear the dirty log */
463 if (is_dirty) {
464 n = kvm_dirty_bitmap_bytes(memslot);
465 memset(memslot->dirty_bitmap, 0, n);
466 }
467 r = 0;
468 out:
469 mutex_unlock(&kvm->slots_lock);
470 return r;
471 }
472
473 static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
474 {
475 unsigned int i;
476 struct kvm_vcpu *vcpu;
477
478 kvm_for_each_vcpu(i, vcpu, kvm) {
479 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
480 }
481 }
482
483 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
484 {
485 int r;
486
487 if (cap->flags)
488 return -EINVAL;
489
490 switch (cap->cap) {
491 case KVM_CAP_S390_IRQCHIP:
492 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
493 kvm->arch.use_irqchip = 1;
494 r = 0;
495 break;
496 case KVM_CAP_S390_USER_SIGP:
497 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
498 kvm->arch.user_sigp = 1;
499 r = 0;
500 break;
501 case KVM_CAP_S390_VECTOR_REGISTERS:
502 mutex_lock(&kvm->lock);
503 if (kvm->created_vcpus) {
504 r = -EBUSY;
505 } else if (MACHINE_HAS_VX) {
506 set_kvm_facility(kvm->arch.model.fac_mask, 129);
507 set_kvm_facility(kvm->arch.model.fac_list, 129);
508 r = 0;
509 } else
510 r = -EINVAL;
511 mutex_unlock(&kvm->lock);
512 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
513 r ? "(not available)" : "(success)");
514 break;
515 case KVM_CAP_S390_RI:
516 r = -EINVAL;
517 mutex_lock(&kvm->lock);
518 if (kvm->created_vcpus) {
519 r = -EBUSY;
520 } else if (test_facility(64)) {
521 set_kvm_facility(kvm->arch.model.fac_mask, 64);
522 set_kvm_facility(kvm->arch.model.fac_list, 64);
523 r = 0;
524 }
525 mutex_unlock(&kvm->lock);
526 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
527 r ? "(not available)" : "(success)");
528 break;
529 case KVM_CAP_S390_USER_STSI:
530 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
531 kvm->arch.user_stsi = 1;
532 r = 0;
533 break;
534 case KVM_CAP_S390_USER_INSTR0:
535 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
536 kvm->arch.user_instr0 = 1;
537 icpt_operexc_on_all_vcpus(kvm);
538 r = 0;
539 break;
540 default:
541 r = -EINVAL;
542 break;
543 }
544 return r;
545 }
546
547 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
548 {
549 int ret;
550
551 switch (attr->attr) {
552 case KVM_S390_VM_MEM_LIMIT_SIZE:
553 ret = 0;
554 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
555 kvm->arch.mem_limit);
556 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
557 ret = -EFAULT;
558 break;
559 default:
560 ret = -ENXIO;
561 break;
562 }
563 return ret;
564 }
565
566 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
567 {
568 int ret;
569 unsigned int idx;
570 switch (attr->attr) {
571 case KVM_S390_VM_MEM_ENABLE_CMMA:
572 ret = -ENXIO;
573 if (!sclp.has_cmma)
574 break;
575
576 ret = -EBUSY;
577 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
578 mutex_lock(&kvm->lock);
579 if (!kvm->created_vcpus) {
580 kvm->arch.use_cmma = 1;
581 ret = 0;
582 }
583 mutex_unlock(&kvm->lock);
584 break;
585 case KVM_S390_VM_MEM_CLR_CMMA:
586 ret = -ENXIO;
587 if (!sclp.has_cmma)
588 break;
589 ret = -EINVAL;
590 if (!kvm->arch.use_cmma)
591 break;
592
593 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
594 mutex_lock(&kvm->lock);
595 idx = srcu_read_lock(&kvm->srcu);
596 s390_reset_cmma(kvm->arch.gmap->mm);
597 srcu_read_unlock(&kvm->srcu, idx);
598 mutex_unlock(&kvm->lock);
599 ret = 0;
600 break;
601 case KVM_S390_VM_MEM_LIMIT_SIZE: {
602 unsigned long new_limit;
603
604 if (kvm_is_ucontrol(kvm))
605 return -EINVAL;
606
607 if (get_user(new_limit, (u64 __user *)attr->addr))
608 return -EFAULT;
609
610 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
611 new_limit > kvm->arch.mem_limit)
612 return -E2BIG;
613
614 if (!new_limit)
615 return -EINVAL;
616
617 /* gmap_create takes last usable address */
618 if (new_limit != KVM_S390_NO_MEM_LIMIT)
619 new_limit -= 1;
620
621 ret = -EBUSY;
622 mutex_lock(&kvm->lock);
623 if (!kvm->created_vcpus) {
624 /* gmap_create will round the limit up */
625 struct gmap *new = gmap_create(current->mm, new_limit);
626
627 if (!new) {
628 ret = -ENOMEM;
629 } else {
630 gmap_remove(kvm->arch.gmap);
631 new->private = kvm;
632 kvm->arch.gmap = new;
633 ret = 0;
634 }
635 }
636 mutex_unlock(&kvm->lock);
637 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
638 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
639 (void *) kvm->arch.gmap->asce);
640 break;
641 }
642 default:
643 ret = -ENXIO;
644 break;
645 }
646 return ret;
647 }
648
649 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
650
651 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
652 {
653 struct kvm_vcpu *vcpu;
654 int i;
655
656 if (!test_kvm_facility(kvm, 76))
657 return -EINVAL;
658
659 mutex_lock(&kvm->lock);
660 switch (attr->attr) {
661 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
662 get_random_bytes(
663 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
664 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
665 kvm->arch.crypto.aes_kw = 1;
666 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
667 break;
668 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
669 get_random_bytes(
670 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
671 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
672 kvm->arch.crypto.dea_kw = 1;
673 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
674 break;
675 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
676 kvm->arch.crypto.aes_kw = 0;
677 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
678 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
679 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
680 break;
681 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
682 kvm->arch.crypto.dea_kw = 0;
683 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
684 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
685 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
686 break;
687 default:
688 mutex_unlock(&kvm->lock);
689 return -ENXIO;
690 }
691
692 kvm_for_each_vcpu(i, vcpu, kvm) {
693 kvm_s390_vcpu_crypto_setup(vcpu);
694 exit_sie(vcpu);
695 }
696 mutex_unlock(&kvm->lock);
697 return 0;
698 }
699
700 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
701 {
702 u8 gtod_high;
703
704 if (copy_from_user(&gtod_high, (void __user *)attr->addr,
705 sizeof(gtod_high)))
706 return -EFAULT;
707
708 if (gtod_high != 0)
709 return -EINVAL;
710 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
711
712 return 0;
713 }
714
715 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
716 {
717 u64 gtod;
718
719 if (copy_from_user(&gtod, (void __user *)attr->addr, sizeof(gtod)))
720 return -EFAULT;
721
722 kvm_s390_set_tod_clock(kvm, gtod);
723 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod);
724 return 0;
725 }
726
727 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
728 {
729 int ret;
730
731 if (attr->flags)
732 return -EINVAL;
733
734 switch (attr->attr) {
735 case KVM_S390_VM_TOD_HIGH:
736 ret = kvm_s390_set_tod_high(kvm, attr);
737 break;
738 case KVM_S390_VM_TOD_LOW:
739 ret = kvm_s390_set_tod_low(kvm, attr);
740 break;
741 default:
742 ret = -ENXIO;
743 break;
744 }
745 return ret;
746 }
747
748 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
749 {
750 u8 gtod_high = 0;
751
752 if (copy_to_user((void __user *)attr->addr, &gtod_high,
753 sizeof(gtod_high)))
754 return -EFAULT;
755 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
756
757 return 0;
758 }
759
760 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
761 {
762 u64 gtod;
763
764 gtod = kvm_s390_get_tod_clock_fast(kvm);
765 if (copy_to_user((void __user *)attr->addr, &gtod, sizeof(gtod)))
766 return -EFAULT;
767 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
768
769 return 0;
770 }
771
772 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
773 {
774 int ret;
775
776 if (attr->flags)
777 return -EINVAL;
778
779 switch (attr->attr) {
780 case KVM_S390_VM_TOD_HIGH:
781 ret = kvm_s390_get_tod_high(kvm, attr);
782 break;
783 case KVM_S390_VM_TOD_LOW:
784 ret = kvm_s390_get_tod_low(kvm, attr);
785 break;
786 default:
787 ret = -ENXIO;
788 break;
789 }
790 return ret;
791 }
792
793 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
794 {
795 struct kvm_s390_vm_cpu_processor *proc;
796 u16 lowest_ibc, unblocked_ibc;
797 int ret = 0;
798
799 mutex_lock(&kvm->lock);
800 if (kvm->created_vcpus) {
801 ret = -EBUSY;
802 goto out;
803 }
804 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
805 if (!proc) {
806 ret = -ENOMEM;
807 goto out;
808 }
809 if (!copy_from_user(proc, (void __user *)attr->addr,
810 sizeof(*proc))) {
811 kvm->arch.model.cpuid = proc->cpuid;
812 lowest_ibc = sclp.ibc >> 16 & 0xfff;
813 unblocked_ibc = sclp.ibc & 0xfff;
814 if (lowest_ibc && proc->ibc) {
815 if (proc->ibc > unblocked_ibc)
816 kvm->arch.model.ibc = unblocked_ibc;
817 else if (proc->ibc < lowest_ibc)
818 kvm->arch.model.ibc = lowest_ibc;
819 else
820 kvm->arch.model.ibc = proc->ibc;
821 }
822 memcpy(kvm->arch.model.fac_list, proc->fac_list,
823 S390_ARCH_FAC_LIST_SIZE_BYTE);
824 } else
825 ret = -EFAULT;
826 kfree(proc);
827 out:
828 mutex_unlock(&kvm->lock);
829 return ret;
830 }
831
832 static int kvm_s390_set_processor_feat(struct kvm *kvm,
833 struct kvm_device_attr *attr)
834 {
835 struct kvm_s390_vm_cpu_feat data;
836 int ret = -EBUSY;
837
838 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
839 return -EFAULT;
840 if (!bitmap_subset((unsigned long *) data.feat,
841 kvm_s390_available_cpu_feat,
842 KVM_S390_VM_CPU_FEAT_NR_BITS))
843 return -EINVAL;
844
845 mutex_lock(&kvm->lock);
846 if (!atomic_read(&kvm->online_vcpus)) {
847 bitmap_copy(kvm->arch.cpu_feat, (unsigned long *) data.feat,
848 KVM_S390_VM_CPU_FEAT_NR_BITS);
849 ret = 0;
850 }
851 mutex_unlock(&kvm->lock);
852 return ret;
853 }
854
855 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
856 struct kvm_device_attr *attr)
857 {
858 /*
859 * Once supported by kernel + hw, we have to store the subfunctions
860 * in kvm->arch and remember that user space configured them.
861 */
862 return -ENXIO;
863 }
864
865 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
866 {
867 int ret = -ENXIO;
868
869 switch (attr->attr) {
870 case KVM_S390_VM_CPU_PROCESSOR:
871 ret = kvm_s390_set_processor(kvm, attr);
872 break;
873 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
874 ret = kvm_s390_set_processor_feat(kvm, attr);
875 break;
876 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
877 ret = kvm_s390_set_processor_subfunc(kvm, attr);
878 break;
879 }
880 return ret;
881 }
882
883 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
884 {
885 struct kvm_s390_vm_cpu_processor *proc;
886 int ret = 0;
887
888 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
889 if (!proc) {
890 ret = -ENOMEM;
891 goto out;
892 }
893 proc->cpuid = kvm->arch.model.cpuid;
894 proc->ibc = kvm->arch.model.ibc;
895 memcpy(&proc->fac_list, kvm->arch.model.fac_list,
896 S390_ARCH_FAC_LIST_SIZE_BYTE);
897 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
898 ret = -EFAULT;
899 kfree(proc);
900 out:
901 return ret;
902 }
903
904 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
905 {
906 struct kvm_s390_vm_cpu_machine *mach;
907 int ret = 0;
908
909 mach = kzalloc(sizeof(*mach), GFP_KERNEL);
910 if (!mach) {
911 ret = -ENOMEM;
912 goto out;
913 }
914 get_cpu_id((struct cpuid *) &mach->cpuid);
915 mach->ibc = sclp.ibc;
916 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
917 S390_ARCH_FAC_LIST_SIZE_BYTE);
918 memcpy((unsigned long *)&mach->fac_list, S390_lowcore.stfle_fac_list,
919 S390_ARCH_FAC_LIST_SIZE_BYTE);
920 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
921 ret = -EFAULT;
922 kfree(mach);
923 out:
924 return ret;
925 }
926
927 static int kvm_s390_get_processor_feat(struct kvm *kvm,
928 struct kvm_device_attr *attr)
929 {
930 struct kvm_s390_vm_cpu_feat data;
931
932 bitmap_copy((unsigned long *) data.feat, kvm->arch.cpu_feat,
933 KVM_S390_VM_CPU_FEAT_NR_BITS);
934 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
935 return -EFAULT;
936 return 0;
937 }
938
939 static int kvm_s390_get_machine_feat(struct kvm *kvm,
940 struct kvm_device_attr *attr)
941 {
942 struct kvm_s390_vm_cpu_feat data;
943
944 bitmap_copy((unsigned long *) data.feat,
945 kvm_s390_available_cpu_feat,
946 KVM_S390_VM_CPU_FEAT_NR_BITS);
947 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
948 return -EFAULT;
949 return 0;
950 }
951
952 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
953 struct kvm_device_attr *attr)
954 {
955 /*
956 * Once we can actually configure subfunctions (kernel + hw support),
957 * we have to check if they were already set by user space, if so copy
958 * them from kvm->arch.
959 */
960 return -ENXIO;
961 }
962
963 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
964 struct kvm_device_attr *attr)
965 {
966 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
967 sizeof(struct kvm_s390_vm_cpu_subfunc)))
968 return -EFAULT;
969 return 0;
970 }
971 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
972 {
973 int ret = -ENXIO;
974
975 switch (attr->attr) {
976 case KVM_S390_VM_CPU_PROCESSOR:
977 ret = kvm_s390_get_processor(kvm, attr);
978 break;
979 case KVM_S390_VM_CPU_MACHINE:
980 ret = kvm_s390_get_machine(kvm, attr);
981 break;
982 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
983 ret = kvm_s390_get_processor_feat(kvm, attr);
984 break;
985 case KVM_S390_VM_CPU_MACHINE_FEAT:
986 ret = kvm_s390_get_machine_feat(kvm, attr);
987 break;
988 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
989 ret = kvm_s390_get_processor_subfunc(kvm, attr);
990 break;
991 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
992 ret = kvm_s390_get_machine_subfunc(kvm, attr);
993 break;
994 }
995 return ret;
996 }
997
998 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
999 {
1000 int ret;
1001
1002 switch (attr->group) {
1003 case KVM_S390_VM_MEM_CTRL:
1004 ret = kvm_s390_set_mem_control(kvm, attr);
1005 break;
1006 case KVM_S390_VM_TOD:
1007 ret = kvm_s390_set_tod(kvm, attr);
1008 break;
1009 case KVM_S390_VM_CPU_MODEL:
1010 ret = kvm_s390_set_cpu_model(kvm, attr);
1011 break;
1012 case KVM_S390_VM_CRYPTO:
1013 ret = kvm_s390_vm_set_crypto(kvm, attr);
1014 break;
1015 default:
1016 ret = -ENXIO;
1017 break;
1018 }
1019
1020 return ret;
1021 }
1022
1023 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1024 {
1025 int ret;
1026
1027 switch (attr->group) {
1028 case KVM_S390_VM_MEM_CTRL:
1029 ret = kvm_s390_get_mem_control(kvm, attr);
1030 break;
1031 case KVM_S390_VM_TOD:
1032 ret = kvm_s390_get_tod(kvm, attr);
1033 break;
1034 case KVM_S390_VM_CPU_MODEL:
1035 ret = kvm_s390_get_cpu_model(kvm, attr);
1036 break;
1037 default:
1038 ret = -ENXIO;
1039 break;
1040 }
1041
1042 return ret;
1043 }
1044
1045 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1046 {
1047 int ret;
1048
1049 switch (attr->group) {
1050 case KVM_S390_VM_MEM_CTRL:
1051 switch (attr->attr) {
1052 case KVM_S390_VM_MEM_ENABLE_CMMA:
1053 case KVM_S390_VM_MEM_CLR_CMMA:
1054 ret = sclp.has_cmma ? 0 : -ENXIO;
1055 break;
1056 case KVM_S390_VM_MEM_LIMIT_SIZE:
1057 ret = 0;
1058 break;
1059 default:
1060 ret = -ENXIO;
1061 break;
1062 }
1063 break;
1064 case KVM_S390_VM_TOD:
1065 switch (attr->attr) {
1066 case KVM_S390_VM_TOD_LOW:
1067 case KVM_S390_VM_TOD_HIGH:
1068 ret = 0;
1069 break;
1070 default:
1071 ret = -ENXIO;
1072 break;
1073 }
1074 break;
1075 case KVM_S390_VM_CPU_MODEL:
1076 switch (attr->attr) {
1077 case KVM_S390_VM_CPU_PROCESSOR:
1078 case KVM_S390_VM_CPU_MACHINE:
1079 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1080 case KVM_S390_VM_CPU_MACHINE_FEAT:
1081 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1082 ret = 0;
1083 break;
1084 /* configuring subfunctions is not supported yet */
1085 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1086 default:
1087 ret = -ENXIO;
1088 break;
1089 }
1090 break;
1091 case KVM_S390_VM_CRYPTO:
1092 switch (attr->attr) {
1093 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1094 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1095 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1096 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1097 ret = 0;
1098 break;
1099 default:
1100 ret = -ENXIO;
1101 break;
1102 }
1103 break;
1104 default:
1105 ret = -ENXIO;
1106 break;
1107 }
1108
1109 return ret;
1110 }
1111
1112 static long kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1113 {
1114 uint8_t *keys;
1115 uint64_t hva;
1116 int i, r = 0;
1117
1118 if (args->flags != 0)
1119 return -EINVAL;
1120
1121 /* Is this guest using storage keys? */
1122 if (!mm_use_skey(current->mm))
1123 return KVM_S390_GET_SKEYS_NONE;
1124
1125 /* Enforce sane limit on memory allocation */
1126 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1127 return -EINVAL;
1128
1129 keys = kmalloc_array(args->count, sizeof(uint8_t),
1130 GFP_KERNEL | __GFP_NOWARN);
1131 if (!keys)
1132 keys = vmalloc(sizeof(uint8_t) * args->count);
1133 if (!keys)
1134 return -ENOMEM;
1135
1136 down_read(&current->mm->mmap_sem);
1137 for (i = 0; i < args->count; i++) {
1138 hva = gfn_to_hva(kvm, args->start_gfn + i);
1139 if (kvm_is_error_hva(hva)) {
1140 r = -EFAULT;
1141 break;
1142 }
1143
1144 r = get_guest_storage_key(current->mm, hva, &keys[i]);
1145 if (r)
1146 break;
1147 }
1148 up_read(&current->mm->mmap_sem);
1149
1150 if (!r) {
1151 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
1152 sizeof(uint8_t) * args->count);
1153 if (r)
1154 r = -EFAULT;
1155 }
1156
1157 kvfree(keys);
1158 return r;
1159 }
1160
1161 static long kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1162 {
1163 uint8_t *keys;
1164 uint64_t hva;
1165 int i, r = 0;
1166
1167 if (args->flags != 0)
1168 return -EINVAL;
1169
1170 /* Enforce sane limit on memory allocation */
1171 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1172 return -EINVAL;
1173
1174 keys = kmalloc_array(args->count, sizeof(uint8_t),
1175 GFP_KERNEL | __GFP_NOWARN);
1176 if (!keys)
1177 keys = vmalloc(sizeof(uint8_t) * args->count);
1178 if (!keys)
1179 return -ENOMEM;
1180
1181 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
1182 sizeof(uint8_t) * args->count);
1183 if (r) {
1184 r = -EFAULT;
1185 goto out;
1186 }
1187
1188 /* Enable storage key handling for the guest */
1189 r = s390_enable_skey();
1190 if (r)
1191 goto out;
1192
1193 down_read(&current->mm->mmap_sem);
1194 for (i = 0; i < args->count; i++) {
1195 hva = gfn_to_hva(kvm, args->start_gfn + i);
1196 if (kvm_is_error_hva(hva)) {
1197 r = -EFAULT;
1198 break;
1199 }
1200
1201 /* Lowest order bit is reserved */
1202 if (keys[i] & 0x01) {
1203 r = -EINVAL;
1204 break;
1205 }
1206
1207 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
1208 if (r)
1209 break;
1210 }
1211 up_read(&current->mm->mmap_sem);
1212 out:
1213 kvfree(keys);
1214 return r;
1215 }
1216
1217 long kvm_arch_vm_ioctl(struct file *filp,
1218 unsigned int ioctl, unsigned long arg)
1219 {
1220 struct kvm *kvm = filp->private_data;
1221 void __user *argp = (void __user *)arg;
1222 struct kvm_device_attr attr;
1223 int r;
1224
1225 switch (ioctl) {
1226 case KVM_S390_INTERRUPT: {
1227 struct kvm_s390_interrupt s390int;
1228
1229 r = -EFAULT;
1230 if (copy_from_user(&s390int, argp, sizeof(s390int)))
1231 break;
1232 r = kvm_s390_inject_vm(kvm, &s390int);
1233 break;
1234 }
1235 case KVM_ENABLE_CAP: {
1236 struct kvm_enable_cap cap;
1237 r = -EFAULT;
1238 if (copy_from_user(&cap, argp, sizeof(cap)))
1239 break;
1240 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1241 break;
1242 }
1243 case KVM_CREATE_IRQCHIP: {
1244 struct kvm_irq_routing_entry routing;
1245
1246 r = -EINVAL;
1247 if (kvm->arch.use_irqchip) {
1248 /* Set up dummy routing. */
1249 memset(&routing, 0, sizeof(routing));
1250 r = kvm_set_irq_routing(kvm, &routing, 0, 0);
1251 }
1252 break;
1253 }
1254 case KVM_SET_DEVICE_ATTR: {
1255 r = -EFAULT;
1256 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1257 break;
1258 r = kvm_s390_vm_set_attr(kvm, &attr);
1259 break;
1260 }
1261 case KVM_GET_DEVICE_ATTR: {
1262 r = -EFAULT;
1263 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1264 break;
1265 r = kvm_s390_vm_get_attr(kvm, &attr);
1266 break;
1267 }
1268 case KVM_HAS_DEVICE_ATTR: {
1269 r = -EFAULT;
1270 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1271 break;
1272 r = kvm_s390_vm_has_attr(kvm, &attr);
1273 break;
1274 }
1275 case KVM_S390_GET_SKEYS: {
1276 struct kvm_s390_skeys args;
1277
1278 r = -EFAULT;
1279 if (copy_from_user(&args, argp,
1280 sizeof(struct kvm_s390_skeys)))
1281 break;
1282 r = kvm_s390_get_skeys(kvm, &args);
1283 break;
1284 }
1285 case KVM_S390_SET_SKEYS: {
1286 struct kvm_s390_skeys args;
1287
1288 r = -EFAULT;
1289 if (copy_from_user(&args, argp,
1290 sizeof(struct kvm_s390_skeys)))
1291 break;
1292 r = kvm_s390_set_skeys(kvm, &args);
1293 break;
1294 }
1295 default:
1296 r = -ENOTTY;
1297 }
1298
1299 return r;
1300 }
1301
1302 static int kvm_s390_query_ap_config(u8 *config)
1303 {
1304 u32 fcn_code = 0x04000000UL;
1305 u32 cc = 0;
1306
1307 memset(config, 0, 128);
1308 asm volatile(
1309 "lgr 0,%1\n"
1310 "lgr 2,%2\n"
1311 ".long 0xb2af0000\n" /* PQAP(QCI) */
1312 "0: ipm %0\n"
1313 "srl %0,28\n"
1314 "1:\n"
1315 EX_TABLE(0b, 1b)
1316 : "+r" (cc)
1317 : "r" (fcn_code), "r" (config)
1318 : "cc", "0", "2", "memory"
1319 );
1320
1321 return cc;
1322 }
1323
1324 static int kvm_s390_apxa_installed(void)
1325 {
1326 u8 config[128];
1327 int cc;
1328
1329 if (test_facility(12)) {
1330 cc = kvm_s390_query_ap_config(config);
1331
1332 if (cc)
1333 pr_err("PQAP(QCI) failed with cc=%d", cc);
1334 else
1335 return config[0] & 0x40;
1336 }
1337
1338 return 0;
1339 }
1340
1341 static void kvm_s390_set_crycb_format(struct kvm *kvm)
1342 {
1343 kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
1344
1345 if (kvm_s390_apxa_installed())
1346 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
1347 else
1348 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
1349 }
1350
1351 static u64 kvm_s390_get_initial_cpuid(void)
1352 {
1353 struct cpuid cpuid;
1354
1355 get_cpu_id(&cpuid);
1356 cpuid.version = 0xff;
1357 return *((u64 *) &cpuid);
1358 }
1359
1360 static void kvm_s390_crypto_init(struct kvm *kvm)
1361 {
1362 if (!test_kvm_facility(kvm, 76))
1363 return;
1364
1365 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
1366 kvm_s390_set_crycb_format(kvm);
1367
1368 /* Enable AES/DEA protected key functions by default */
1369 kvm->arch.crypto.aes_kw = 1;
1370 kvm->arch.crypto.dea_kw = 1;
1371 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1372 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1373 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1374 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1375 }
1376
1377 static void sca_dispose(struct kvm *kvm)
1378 {
1379 if (kvm->arch.use_esca)
1380 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
1381 else
1382 free_page((unsigned long)(kvm->arch.sca));
1383 kvm->arch.sca = NULL;
1384 }
1385
1386 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
1387 {
1388 gfp_t alloc_flags = GFP_KERNEL;
1389 int i, rc;
1390 char debug_name[16];
1391 static unsigned long sca_offset;
1392
1393 rc = -EINVAL;
1394 #ifdef CONFIG_KVM_S390_UCONTROL
1395 if (type & ~KVM_VM_S390_UCONTROL)
1396 goto out_err;
1397 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
1398 goto out_err;
1399 #else
1400 if (type)
1401 goto out_err;
1402 #endif
1403
1404 rc = s390_enable_sie();
1405 if (rc)
1406 goto out_err;
1407
1408 rc = -ENOMEM;
1409
1410 ratelimit_state_init(&kvm->arch.sthyi_limit, 5 * HZ, 500);
1411
1412 kvm->arch.use_esca = 0; /* start with basic SCA */
1413 if (!sclp.has_64bscao)
1414 alloc_flags |= GFP_DMA;
1415 rwlock_init(&kvm->arch.sca_lock);
1416 kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
1417 if (!kvm->arch.sca)
1418 goto out_err;
1419 spin_lock(&kvm_lock);
1420 sca_offset += 16;
1421 if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
1422 sca_offset = 0;
1423 kvm->arch.sca = (struct bsca_block *)
1424 ((char *) kvm->arch.sca + sca_offset);
1425 spin_unlock(&kvm_lock);
1426
1427 sprintf(debug_name, "kvm-%u", current->pid);
1428
1429 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
1430 if (!kvm->arch.dbf)
1431 goto out_err;
1432
1433 kvm->arch.sie_page2 =
1434 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
1435 if (!kvm->arch.sie_page2)
1436 goto out_err;
1437
1438 /* Populate the facility mask initially. */
1439 memcpy(kvm->arch.model.fac_mask, S390_lowcore.stfle_fac_list,
1440 S390_ARCH_FAC_LIST_SIZE_BYTE);
1441 for (i = 0; i < S390_ARCH_FAC_LIST_SIZE_U64; i++) {
1442 if (i < kvm_s390_fac_list_mask_size())
1443 kvm->arch.model.fac_mask[i] &= kvm_s390_fac_list_mask[i];
1444 else
1445 kvm->arch.model.fac_mask[i] = 0UL;
1446 }
1447
1448 /* Populate the facility list initially. */
1449 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
1450 memcpy(kvm->arch.model.fac_list, kvm->arch.model.fac_mask,
1451 S390_ARCH_FAC_LIST_SIZE_BYTE);
1452
1453 set_kvm_facility(kvm->arch.model.fac_mask, 74);
1454 set_kvm_facility(kvm->arch.model.fac_list, 74);
1455
1456 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
1457 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
1458
1459 kvm_s390_crypto_init(kvm);
1460
1461 spin_lock_init(&kvm->arch.float_int.lock);
1462 for (i = 0; i < FIRQ_LIST_COUNT; i++)
1463 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
1464 init_waitqueue_head(&kvm->arch.ipte_wq);
1465 mutex_init(&kvm->arch.ipte_mutex);
1466
1467 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
1468 VM_EVENT(kvm, 3, "vm created with type %lu", type);
1469
1470 if (type & KVM_VM_S390_UCONTROL) {
1471 kvm->arch.gmap = NULL;
1472 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
1473 } else {
1474 if (sclp.hamax == U64_MAX)
1475 kvm->arch.mem_limit = TASK_MAX_SIZE;
1476 else
1477 kvm->arch.mem_limit = min_t(unsigned long, TASK_MAX_SIZE,
1478 sclp.hamax + 1);
1479 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
1480 if (!kvm->arch.gmap)
1481 goto out_err;
1482 kvm->arch.gmap->private = kvm;
1483 kvm->arch.gmap->pfault_enabled = 0;
1484 }
1485
1486 kvm->arch.css_support = 0;
1487 kvm->arch.use_irqchip = 0;
1488 kvm->arch.epoch = 0;
1489
1490 spin_lock_init(&kvm->arch.start_stop_lock);
1491 kvm_s390_vsie_init(kvm);
1492 KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
1493
1494 return 0;
1495 out_err:
1496 free_page((unsigned long)kvm->arch.sie_page2);
1497 debug_unregister(kvm->arch.dbf);
1498 sca_dispose(kvm);
1499 KVM_EVENT(3, "creation of vm failed: %d", rc);
1500 return rc;
1501 }
1502
1503 bool kvm_arch_has_vcpu_debugfs(void)
1504 {
1505 return false;
1506 }
1507
1508 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
1509 {
1510 return 0;
1511 }
1512
1513 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
1514 {
1515 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
1516 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
1517 kvm_s390_clear_local_irqs(vcpu);
1518 kvm_clear_async_pf_completion_queue(vcpu);
1519 if (!kvm_is_ucontrol(vcpu->kvm))
1520 sca_del_vcpu(vcpu);
1521
1522 if (kvm_is_ucontrol(vcpu->kvm))
1523 gmap_remove(vcpu->arch.gmap);
1524
1525 if (vcpu->kvm->arch.use_cmma)
1526 kvm_s390_vcpu_unsetup_cmma(vcpu);
1527 free_page((unsigned long)(vcpu->arch.sie_block));
1528
1529 kvm_vcpu_uninit(vcpu);
1530 kmem_cache_free(kvm_vcpu_cache, vcpu);
1531 }
1532
1533 static void kvm_free_vcpus(struct kvm *kvm)
1534 {
1535 unsigned int i;
1536 struct kvm_vcpu *vcpu;
1537
1538 kvm_for_each_vcpu(i, vcpu, kvm)
1539 kvm_arch_vcpu_destroy(vcpu);
1540
1541 mutex_lock(&kvm->lock);
1542 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
1543 kvm->vcpus[i] = NULL;
1544
1545 atomic_set(&kvm->online_vcpus, 0);
1546 mutex_unlock(&kvm->lock);
1547 }
1548
1549 void kvm_arch_destroy_vm(struct kvm *kvm)
1550 {
1551 kvm_free_vcpus(kvm);
1552 sca_dispose(kvm);
1553 debug_unregister(kvm->arch.dbf);
1554 free_page((unsigned long)kvm->arch.sie_page2);
1555 if (!kvm_is_ucontrol(kvm))
1556 gmap_remove(kvm->arch.gmap);
1557 kvm_s390_destroy_adapters(kvm);
1558 kvm_s390_clear_float_irqs(kvm);
1559 kvm_s390_vsie_destroy(kvm);
1560 KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
1561 }
1562
1563 /* Section: vcpu related */
1564 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
1565 {
1566 vcpu->arch.gmap = gmap_create(current->mm, -1UL);
1567 if (!vcpu->arch.gmap)
1568 return -ENOMEM;
1569 vcpu->arch.gmap->private = vcpu->kvm;
1570
1571 return 0;
1572 }
1573
1574 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
1575 {
1576 if (!kvm_s390_use_sca_entries())
1577 return;
1578 read_lock(&vcpu->kvm->arch.sca_lock);
1579 if (vcpu->kvm->arch.use_esca) {
1580 struct esca_block *sca = vcpu->kvm->arch.sca;
1581
1582 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
1583 sca->cpu[vcpu->vcpu_id].sda = 0;
1584 } else {
1585 struct bsca_block *sca = vcpu->kvm->arch.sca;
1586
1587 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
1588 sca->cpu[vcpu->vcpu_id].sda = 0;
1589 }
1590 read_unlock(&vcpu->kvm->arch.sca_lock);
1591 }
1592
1593 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
1594 {
1595 if (!kvm_s390_use_sca_entries()) {
1596 struct bsca_block *sca = vcpu->kvm->arch.sca;
1597
1598 /* we still need the basic sca for the ipte control */
1599 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1600 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
1601 }
1602 read_lock(&vcpu->kvm->arch.sca_lock);
1603 if (vcpu->kvm->arch.use_esca) {
1604 struct esca_block *sca = vcpu->kvm->arch.sca;
1605
1606 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
1607 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1608 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca & ~0x3fU;
1609 vcpu->arch.sie_block->ecb2 |= 0x04U;
1610 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
1611 } else {
1612 struct bsca_block *sca = vcpu->kvm->arch.sca;
1613
1614 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
1615 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
1616 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
1617 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
1618 }
1619 read_unlock(&vcpu->kvm->arch.sca_lock);
1620 }
1621
1622 /* Basic SCA to Extended SCA data copy routines */
1623 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
1624 {
1625 d->sda = s->sda;
1626 d->sigp_ctrl.c = s->sigp_ctrl.c;
1627 d->sigp_ctrl.scn = s->sigp_ctrl.scn;
1628 }
1629
1630 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
1631 {
1632 int i;
1633
1634 d->ipte_control = s->ipte_control;
1635 d->mcn[0] = s->mcn;
1636 for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
1637 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
1638 }
1639
1640 static int sca_switch_to_extended(struct kvm *kvm)
1641 {
1642 struct bsca_block *old_sca = kvm->arch.sca;
1643 struct esca_block *new_sca;
1644 struct kvm_vcpu *vcpu;
1645 unsigned int vcpu_idx;
1646 u32 scaol, scaoh;
1647
1648 new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL|__GFP_ZERO);
1649 if (!new_sca)
1650 return -ENOMEM;
1651
1652 scaoh = (u32)((u64)(new_sca) >> 32);
1653 scaol = (u32)(u64)(new_sca) & ~0x3fU;
1654
1655 kvm_s390_vcpu_block_all(kvm);
1656 write_lock(&kvm->arch.sca_lock);
1657
1658 sca_copy_b_to_e(new_sca, old_sca);
1659
1660 kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
1661 vcpu->arch.sie_block->scaoh = scaoh;
1662 vcpu->arch.sie_block->scaol = scaol;
1663 vcpu->arch.sie_block->ecb2 |= 0x04U;
1664 }
1665 kvm->arch.sca = new_sca;
1666 kvm->arch.use_esca = 1;
1667
1668 write_unlock(&kvm->arch.sca_lock);
1669 kvm_s390_vcpu_unblock_all(kvm);
1670
1671 free_page((unsigned long)old_sca);
1672
1673 VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
1674 old_sca, kvm->arch.sca);
1675 return 0;
1676 }
1677
1678 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
1679 {
1680 int rc;
1681
1682 if (!kvm_s390_use_sca_entries()) {
1683 if (id < KVM_MAX_VCPUS)
1684 return true;
1685 return false;
1686 }
1687 if (id < KVM_S390_BSCA_CPU_SLOTS)
1688 return true;
1689 if (!sclp.has_esca || !sclp.has_64bscao)
1690 return false;
1691
1692 mutex_lock(&kvm->lock);
1693 rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
1694 mutex_unlock(&kvm->lock);
1695
1696 return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
1697 }
1698
1699 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
1700 {
1701 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1702 kvm_clear_async_pf_completion_queue(vcpu);
1703 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
1704 KVM_SYNC_GPRS |
1705 KVM_SYNC_ACRS |
1706 KVM_SYNC_CRS |
1707 KVM_SYNC_ARCH0 |
1708 KVM_SYNC_PFAULT;
1709 kvm_s390_set_prefix(vcpu, 0);
1710 if (test_kvm_facility(vcpu->kvm, 64))
1711 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
1712 /* fprs can be synchronized via vrs, even if the guest has no vx. With
1713 * MACHINE_HAS_VX, (load|store)_fpu_regs() will work with vrs format.
1714 */
1715 if (MACHINE_HAS_VX)
1716 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
1717 else
1718 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
1719
1720 if (kvm_is_ucontrol(vcpu->kvm))
1721 return __kvm_ucontrol_vcpu_init(vcpu);
1722
1723 return 0;
1724 }
1725
1726 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1727 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1728 {
1729 WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
1730 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1731 vcpu->arch.cputm_start = get_tod_clock_fast();
1732 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1733 }
1734
1735 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1736 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1737 {
1738 WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
1739 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1740 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
1741 vcpu->arch.cputm_start = 0;
1742 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1743 }
1744
1745 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1746 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1747 {
1748 WARN_ON_ONCE(vcpu->arch.cputm_enabled);
1749 vcpu->arch.cputm_enabled = true;
1750 __start_cpu_timer_accounting(vcpu);
1751 }
1752
1753 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
1754 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1755 {
1756 WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
1757 __stop_cpu_timer_accounting(vcpu);
1758 vcpu->arch.cputm_enabled = false;
1759 }
1760
1761 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1762 {
1763 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1764 __enable_cpu_timer_accounting(vcpu);
1765 preempt_enable();
1766 }
1767
1768 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
1769 {
1770 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1771 __disable_cpu_timer_accounting(vcpu);
1772 preempt_enable();
1773 }
1774
1775 /* set the cpu timer - may only be called from the VCPU thread itself */
1776 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
1777 {
1778 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1779 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
1780 if (vcpu->arch.cputm_enabled)
1781 vcpu->arch.cputm_start = get_tod_clock_fast();
1782 vcpu->arch.sie_block->cputm = cputm;
1783 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
1784 preempt_enable();
1785 }
1786
1787 /* update and get the cpu timer - can also be called from other VCPU threads */
1788 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
1789 {
1790 unsigned int seq;
1791 __u64 value;
1792
1793 if (unlikely(!vcpu->arch.cputm_enabled))
1794 return vcpu->arch.sie_block->cputm;
1795
1796 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
1797 do {
1798 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
1799 /*
1800 * If the writer would ever execute a read in the critical
1801 * section, e.g. in irq context, we have a deadlock.
1802 */
1803 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
1804 value = vcpu->arch.sie_block->cputm;
1805 /* if cputm_start is 0, accounting is being started/stopped */
1806 if (likely(vcpu->arch.cputm_start))
1807 value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
1808 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
1809 preempt_enable();
1810 return value;
1811 }
1812
1813 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1814 {
1815
1816 gmap_enable(vcpu->arch.enabled_gmap);
1817 atomic_or(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1818 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
1819 __start_cpu_timer_accounting(vcpu);
1820 vcpu->cpu = cpu;
1821 }
1822
1823 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1824 {
1825 vcpu->cpu = -1;
1826 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
1827 __stop_cpu_timer_accounting(vcpu);
1828 atomic_andnot(CPUSTAT_RUNNING, &vcpu->arch.sie_block->cpuflags);
1829 vcpu->arch.enabled_gmap = gmap_get_enabled();
1830 gmap_disable(vcpu->arch.enabled_gmap);
1831
1832 }
1833
1834 static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
1835 {
1836 /* this equals initial cpu reset in pop, but we don't switch to ESA */
1837 vcpu->arch.sie_block->gpsw.mask = 0UL;
1838 vcpu->arch.sie_block->gpsw.addr = 0UL;
1839 kvm_s390_set_prefix(vcpu, 0);
1840 kvm_s390_set_cpu_timer(vcpu, 0);
1841 vcpu->arch.sie_block->ckc = 0UL;
1842 vcpu->arch.sie_block->todpr = 0;
1843 memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64));
1844 vcpu->arch.sie_block->gcr[0] = 0xE0UL;
1845 vcpu->arch.sie_block->gcr[14] = 0xC2000000UL;
1846 /* make sure the new fpc will be lazily loaded */
1847 save_fpu_regs();
1848 current->thread.fpu.fpc = 0;
1849 vcpu->arch.sie_block->gbea = 1;
1850 vcpu->arch.sie_block->pp = 0;
1851 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
1852 kvm_clear_async_pf_completion_queue(vcpu);
1853 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
1854 kvm_s390_vcpu_stop(vcpu);
1855 kvm_s390_clear_local_irqs(vcpu);
1856 }
1857
1858 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
1859 {
1860 mutex_lock(&vcpu->kvm->lock);
1861 preempt_disable();
1862 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
1863 preempt_enable();
1864 mutex_unlock(&vcpu->kvm->lock);
1865 if (!kvm_is_ucontrol(vcpu->kvm)) {
1866 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
1867 sca_add_vcpu(vcpu);
1868 }
1869 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
1870 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
1871 /* make vcpu_load load the right gmap on the first trigger */
1872 vcpu->arch.enabled_gmap = vcpu->arch.gmap;
1873 }
1874
1875 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
1876 {
1877 if (!test_kvm_facility(vcpu->kvm, 76))
1878 return;
1879
1880 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
1881
1882 if (vcpu->kvm->arch.crypto.aes_kw)
1883 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
1884 if (vcpu->kvm->arch.crypto.dea_kw)
1885 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
1886
1887 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
1888 }
1889
1890 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
1891 {
1892 free_page(vcpu->arch.sie_block->cbrlo);
1893 vcpu->arch.sie_block->cbrlo = 0;
1894 }
1895
1896 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
1897 {
1898 vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
1899 if (!vcpu->arch.sie_block->cbrlo)
1900 return -ENOMEM;
1901
1902 vcpu->arch.sie_block->ecb2 |= 0x80;
1903 vcpu->arch.sie_block->ecb2 &= ~0x08;
1904 return 0;
1905 }
1906
1907 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
1908 {
1909 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
1910
1911 vcpu->arch.sie_block->ibc = model->ibc;
1912 if (test_kvm_facility(vcpu->kvm, 7))
1913 vcpu->arch.sie_block->fac = (u32)(u64) model->fac_list;
1914 }
1915
1916 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
1917 {
1918 int rc = 0;
1919
1920 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
1921 CPUSTAT_SM |
1922 CPUSTAT_STOPPED);
1923
1924 if (test_kvm_facility(vcpu->kvm, 78))
1925 atomic_or(CPUSTAT_GED2, &vcpu->arch.sie_block->cpuflags);
1926 else if (test_kvm_facility(vcpu->kvm, 8))
1927 atomic_or(CPUSTAT_GED, &vcpu->arch.sie_block->cpuflags);
1928
1929 kvm_s390_vcpu_setup_model(vcpu);
1930
1931 /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
1932 if (MACHINE_HAS_ESOP)
1933 vcpu->arch.sie_block->ecb |= 0x02;
1934 if (test_kvm_facility(vcpu->kvm, 9))
1935 vcpu->arch.sie_block->ecb |= 0x04;
1936 if (test_kvm_facility(vcpu->kvm, 73))
1937 vcpu->arch.sie_block->ecb |= 0x10;
1938
1939 if (test_kvm_facility(vcpu->kvm, 8) && sclp.has_pfmfi)
1940 vcpu->arch.sie_block->ecb2 |= 0x08;
1941 vcpu->arch.sie_block->eca = 0x1002000U;
1942 if (sclp.has_cei)
1943 vcpu->arch.sie_block->eca |= 0x80000000U;
1944 if (sclp.has_ib)
1945 vcpu->arch.sie_block->eca |= 0x40000000U;
1946 if (sclp.has_siif)
1947 vcpu->arch.sie_block->eca |= 1;
1948 if (sclp.has_sigpif)
1949 vcpu->arch.sie_block->eca |= 0x10000000U;
1950 if (test_kvm_facility(vcpu->kvm, 129)) {
1951 vcpu->arch.sie_block->eca |= 0x00020000;
1952 vcpu->arch.sie_block->ecd |= 0x20000000;
1953 }
1954 vcpu->arch.sie_block->riccbd = (unsigned long) &vcpu->run->s.regs.riccb;
1955 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
1956
1957 if (vcpu->kvm->arch.use_cmma) {
1958 rc = kvm_s390_vcpu_setup_cmma(vcpu);
1959 if (rc)
1960 return rc;
1961 }
1962 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1963 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
1964
1965 kvm_s390_vcpu_crypto_setup(vcpu);
1966
1967 return rc;
1968 }
1969
1970 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
1971 unsigned int id)
1972 {
1973 struct kvm_vcpu *vcpu;
1974 struct sie_page *sie_page;
1975 int rc = -EINVAL;
1976
1977 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
1978 goto out;
1979
1980 rc = -ENOMEM;
1981
1982 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1983 if (!vcpu)
1984 goto out;
1985
1986 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
1987 if (!sie_page)
1988 goto out_free_cpu;
1989
1990 vcpu->arch.sie_block = &sie_page->sie_block;
1991 vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
1992
1993 /* the real guest size will always be smaller than msl */
1994 vcpu->arch.sie_block->mso = 0;
1995 vcpu->arch.sie_block->msl = sclp.hamax;
1996
1997 vcpu->arch.sie_block->icpua = id;
1998 spin_lock_init(&vcpu->arch.local_int.lock);
1999 vcpu->arch.local_int.float_int = &kvm->arch.float_int;
2000 vcpu->arch.local_int.wq = &vcpu->wq;
2001 vcpu->arch.local_int.cpuflags = &vcpu->arch.sie_block->cpuflags;
2002 seqcount_init(&vcpu->arch.cputm_seqcount);
2003
2004 rc = kvm_vcpu_init(vcpu, kvm, id);
2005 if (rc)
2006 goto out_free_sie_block;
2007 VM_EVENT(kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK", id, vcpu,
2008 vcpu->arch.sie_block);
2009 trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
2010
2011 return vcpu;
2012 out_free_sie_block:
2013 free_page((unsigned long)(vcpu->arch.sie_block));
2014 out_free_cpu:
2015 kmem_cache_free(kvm_vcpu_cache, vcpu);
2016 out:
2017 return ERR_PTR(rc);
2018 }
2019
2020 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
2021 {
2022 return kvm_s390_vcpu_has_irq(vcpu, 0);
2023 }
2024
2025 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
2026 {
2027 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2028 exit_sie(vcpu);
2029 }
2030
2031 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
2032 {
2033 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2034 }
2035
2036 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
2037 {
2038 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2039 exit_sie(vcpu);
2040 }
2041
2042 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
2043 {
2044 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2045 }
2046
2047 /*
2048 * Kick a guest cpu out of SIE and wait until SIE is not running.
2049 * If the CPU is not running (e.g. waiting as idle) the function will
2050 * return immediately. */
2051 void exit_sie(struct kvm_vcpu *vcpu)
2052 {
2053 atomic_or(CPUSTAT_STOP_INT, &vcpu->arch.sie_block->cpuflags);
2054 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
2055 cpu_relax();
2056 }
2057
2058 /* Kick a guest cpu out of SIE to process a request synchronously */
2059 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
2060 {
2061 kvm_make_request(req, vcpu);
2062 kvm_s390_vcpu_request(vcpu);
2063 }
2064
2065 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
2066 unsigned long end)
2067 {
2068 struct kvm *kvm = gmap->private;
2069 struct kvm_vcpu *vcpu;
2070 unsigned long prefix;
2071 int i;
2072
2073 if (gmap_is_shadow(gmap))
2074 return;
2075 if (start >= 1UL << 31)
2076 /* We are only interested in prefix pages */
2077 return;
2078 kvm_for_each_vcpu(i, vcpu, kvm) {
2079 /* match against both prefix pages */
2080 prefix = kvm_s390_get_prefix(vcpu);
2081 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
2082 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
2083 start, end);
2084 kvm_s390_sync_request(KVM_REQ_MMU_RELOAD, vcpu);
2085 }
2086 }
2087 }
2088
2089 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
2090 {
2091 /* kvm common code refers to this, but never calls it */
2092 BUG();
2093 return 0;
2094 }
2095
2096 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
2097 struct kvm_one_reg *reg)
2098 {
2099 int r = -EINVAL;
2100
2101 switch (reg->id) {
2102 case KVM_REG_S390_TODPR:
2103 r = put_user(vcpu->arch.sie_block->todpr,
2104 (u32 __user *)reg->addr);
2105 break;
2106 case KVM_REG_S390_EPOCHDIFF:
2107 r = put_user(vcpu->arch.sie_block->epoch,
2108 (u64 __user *)reg->addr);
2109 break;
2110 case KVM_REG_S390_CPU_TIMER:
2111 r = put_user(kvm_s390_get_cpu_timer(vcpu),
2112 (u64 __user *)reg->addr);
2113 break;
2114 case KVM_REG_S390_CLOCK_COMP:
2115 r = put_user(vcpu->arch.sie_block->ckc,
2116 (u64 __user *)reg->addr);
2117 break;
2118 case KVM_REG_S390_PFTOKEN:
2119 r = put_user(vcpu->arch.pfault_token,
2120 (u64 __user *)reg->addr);
2121 break;
2122 case KVM_REG_S390_PFCOMPARE:
2123 r = put_user(vcpu->arch.pfault_compare,
2124 (u64 __user *)reg->addr);
2125 break;
2126 case KVM_REG_S390_PFSELECT:
2127 r = put_user(vcpu->arch.pfault_select,
2128 (u64 __user *)reg->addr);
2129 break;
2130 case KVM_REG_S390_PP:
2131 r = put_user(vcpu->arch.sie_block->pp,
2132 (u64 __user *)reg->addr);
2133 break;
2134 case KVM_REG_S390_GBEA:
2135 r = put_user(vcpu->arch.sie_block->gbea,
2136 (u64 __user *)reg->addr);
2137 break;
2138 default:
2139 break;
2140 }
2141
2142 return r;
2143 }
2144
2145 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
2146 struct kvm_one_reg *reg)
2147 {
2148 int r = -EINVAL;
2149 __u64 val;
2150
2151 switch (reg->id) {
2152 case KVM_REG_S390_TODPR:
2153 r = get_user(vcpu->arch.sie_block->todpr,
2154 (u32 __user *)reg->addr);
2155 break;
2156 case KVM_REG_S390_EPOCHDIFF:
2157 r = get_user(vcpu->arch.sie_block->epoch,
2158 (u64 __user *)reg->addr);
2159 break;
2160 case KVM_REG_S390_CPU_TIMER:
2161 r = get_user(val, (u64 __user *)reg->addr);
2162 if (!r)
2163 kvm_s390_set_cpu_timer(vcpu, val);
2164 break;
2165 case KVM_REG_S390_CLOCK_COMP:
2166 r = get_user(vcpu->arch.sie_block->ckc,
2167 (u64 __user *)reg->addr);
2168 break;
2169 case KVM_REG_S390_PFTOKEN:
2170 r = get_user(vcpu->arch.pfault_token,
2171 (u64 __user *)reg->addr);
2172 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2173 kvm_clear_async_pf_completion_queue(vcpu);
2174 break;
2175 case KVM_REG_S390_PFCOMPARE:
2176 r = get_user(vcpu->arch.pfault_compare,
2177 (u64 __user *)reg->addr);
2178 break;
2179 case KVM_REG_S390_PFSELECT:
2180 r = get_user(vcpu->arch.pfault_select,
2181 (u64 __user *)reg->addr);
2182 break;
2183 case KVM_REG_S390_PP:
2184 r = get_user(vcpu->arch.sie_block->pp,
2185 (u64 __user *)reg->addr);
2186 break;
2187 case KVM_REG_S390_GBEA:
2188 r = get_user(vcpu->arch.sie_block->gbea,
2189 (u64 __user *)reg->addr);
2190 break;
2191 default:
2192 break;
2193 }
2194
2195 return r;
2196 }
2197
2198 static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
2199 {
2200 kvm_s390_vcpu_initial_reset(vcpu);
2201 return 0;
2202 }
2203
2204 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2205 {
2206 memcpy(&vcpu->run->s.regs.gprs, &regs->gprs, sizeof(regs->gprs));
2207 return 0;
2208 }
2209
2210 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2211 {
2212 memcpy(&regs->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
2213 return 0;
2214 }
2215
2216 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2217 struct kvm_sregs *sregs)
2218 {
2219 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
2220 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
2221 return 0;
2222 }
2223
2224 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2225 struct kvm_sregs *sregs)
2226 {
2227 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
2228 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
2229 return 0;
2230 }
2231
2232 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2233 {
2234 if (test_fp_ctl(fpu->fpc))
2235 return -EINVAL;
2236 vcpu->run->s.regs.fpc = fpu->fpc;
2237 if (MACHINE_HAS_VX)
2238 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs,
2239 (freg_t *) fpu->fprs);
2240 else
2241 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs));
2242 return 0;
2243 }
2244
2245 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2246 {
2247 /* make sure we have the latest values */
2248 save_fpu_regs();
2249 if (MACHINE_HAS_VX)
2250 convert_vx_to_fp((freg_t *) fpu->fprs,
2251 (__vector128 *) vcpu->run->s.regs.vrs);
2252 else
2253 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs));
2254 fpu->fpc = vcpu->run->s.regs.fpc;
2255 return 0;
2256 }
2257
2258 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
2259 {
2260 int rc = 0;
2261
2262 if (!is_vcpu_stopped(vcpu))
2263 rc = -EBUSY;
2264 else {
2265 vcpu->run->psw_mask = psw.mask;
2266 vcpu->run->psw_addr = psw.addr;
2267 }
2268 return rc;
2269 }
2270
2271 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2272 struct kvm_translation *tr)
2273 {
2274 return -EINVAL; /* not implemented yet */
2275 }
2276
2277 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
2278 KVM_GUESTDBG_USE_HW_BP | \
2279 KVM_GUESTDBG_ENABLE)
2280
2281 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
2282 struct kvm_guest_debug *dbg)
2283 {
2284 int rc = 0;
2285
2286 vcpu->guest_debug = 0;
2287 kvm_s390_clear_bp_data(vcpu);
2288
2289 if (dbg->control & ~VALID_GUESTDBG_FLAGS)
2290 return -EINVAL;
2291 if (!sclp.has_gpere)
2292 return -EINVAL;
2293
2294 if (dbg->control & KVM_GUESTDBG_ENABLE) {
2295 vcpu->guest_debug = dbg->control;
2296 /* enforce guest PER */
2297 atomic_or(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2298
2299 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
2300 rc = kvm_s390_import_bp_data(vcpu, dbg);
2301 } else {
2302 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2303 vcpu->arch.guestdbg.last_bp = 0;
2304 }
2305
2306 if (rc) {
2307 vcpu->guest_debug = 0;
2308 kvm_s390_clear_bp_data(vcpu);
2309 atomic_andnot(CPUSTAT_P, &vcpu->arch.sie_block->cpuflags);
2310 }
2311
2312 return rc;
2313 }
2314
2315 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2316 struct kvm_mp_state *mp_state)
2317 {
2318 /* CHECK_STOP and LOAD are not supported yet */
2319 return is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
2320 KVM_MP_STATE_OPERATING;
2321 }
2322
2323 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2324 struct kvm_mp_state *mp_state)
2325 {
2326 int rc = 0;
2327
2328 /* user space knows about this interface - let it control the state */
2329 vcpu->kvm->arch.user_cpu_state_ctrl = 1;
2330
2331 switch (mp_state->mp_state) {
2332 case KVM_MP_STATE_STOPPED:
2333 kvm_s390_vcpu_stop(vcpu);
2334 break;
2335 case KVM_MP_STATE_OPERATING:
2336 kvm_s390_vcpu_start(vcpu);
2337 break;
2338 case KVM_MP_STATE_LOAD:
2339 case KVM_MP_STATE_CHECK_STOP:
2340 /* fall through - CHECK_STOP and LOAD are not supported yet */
2341 default:
2342 rc = -ENXIO;
2343 }
2344
2345 return rc;
2346 }
2347
2348 static bool ibs_enabled(struct kvm_vcpu *vcpu)
2349 {
2350 return atomic_read(&vcpu->arch.sie_block->cpuflags) & CPUSTAT_IBS;
2351 }
2352
2353 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
2354 {
2355 retry:
2356 kvm_s390_vcpu_request_handled(vcpu);
2357 if (!vcpu->requests)
2358 return 0;
2359 /*
2360 * We use MMU_RELOAD just to re-arm the ipte notifier for the
2361 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
2362 * This ensures that the ipte instruction for this request has
2363 * already finished. We might race against a second unmapper that
2364 * wants to set the blocking bit. Lets just retry the request loop.
2365 */
2366 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
2367 int rc;
2368 rc = gmap_mprotect_notify(vcpu->arch.gmap,
2369 kvm_s390_get_prefix(vcpu),
2370 PAGE_SIZE * 2, PROT_WRITE);
2371 if (rc) {
2372 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
2373 return rc;
2374 }
2375 goto retry;
2376 }
2377
2378 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
2379 vcpu->arch.sie_block->ihcpu = 0xffff;
2380 goto retry;
2381 }
2382
2383 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
2384 if (!ibs_enabled(vcpu)) {
2385 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
2386 atomic_or(CPUSTAT_IBS,
2387 &vcpu->arch.sie_block->cpuflags);
2388 }
2389 goto retry;
2390 }
2391
2392 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
2393 if (ibs_enabled(vcpu)) {
2394 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
2395 atomic_andnot(CPUSTAT_IBS,
2396 &vcpu->arch.sie_block->cpuflags);
2397 }
2398 goto retry;
2399 }
2400
2401 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
2402 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
2403 goto retry;
2404 }
2405
2406 /* nothing to do, just clear the request */
2407 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
2408
2409 return 0;
2410 }
2411
2412 void kvm_s390_set_tod_clock(struct kvm *kvm, u64 tod)
2413 {
2414 struct kvm_vcpu *vcpu;
2415 int i;
2416
2417 mutex_lock(&kvm->lock);
2418 preempt_disable();
2419 kvm->arch.epoch = tod - get_tod_clock();
2420 kvm_s390_vcpu_block_all(kvm);
2421 kvm_for_each_vcpu(i, vcpu, kvm)
2422 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
2423 kvm_s390_vcpu_unblock_all(kvm);
2424 preempt_enable();
2425 mutex_unlock(&kvm->lock);
2426 }
2427
2428 /**
2429 * kvm_arch_fault_in_page - fault-in guest page if necessary
2430 * @vcpu: The corresponding virtual cpu
2431 * @gpa: Guest physical address
2432 * @writable: Whether the page should be writable or not
2433 *
2434 * Make sure that a guest page has been faulted-in on the host.
2435 *
2436 * Return: Zero on success, negative error code otherwise.
2437 */
2438 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
2439 {
2440 return gmap_fault(vcpu->arch.gmap, gpa,
2441 writable ? FAULT_FLAG_WRITE : 0);
2442 }
2443
2444 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
2445 unsigned long token)
2446 {
2447 struct kvm_s390_interrupt inti;
2448 struct kvm_s390_irq irq;
2449
2450 if (start_token) {
2451 irq.u.ext.ext_params2 = token;
2452 irq.type = KVM_S390_INT_PFAULT_INIT;
2453 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
2454 } else {
2455 inti.type = KVM_S390_INT_PFAULT_DONE;
2456 inti.parm64 = token;
2457 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
2458 }
2459 }
2460
2461 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
2462 struct kvm_async_pf *work)
2463 {
2464 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
2465 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
2466 }
2467
2468 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
2469 struct kvm_async_pf *work)
2470 {
2471 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
2472 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
2473 }
2474
2475 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
2476 struct kvm_async_pf *work)
2477 {
2478 /* s390 will always inject the page directly */
2479 }
2480
2481 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
2482 {
2483 /*
2484 * s390 will always inject the page directly,
2485 * but we still want check_async_completion to cleanup
2486 */
2487 return true;
2488 }
2489
2490 static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
2491 {
2492 hva_t hva;
2493 struct kvm_arch_async_pf arch;
2494 int rc;
2495
2496 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2497 return 0;
2498 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
2499 vcpu->arch.pfault_compare)
2500 return 0;
2501 if (psw_extint_disabled(vcpu))
2502 return 0;
2503 if (kvm_s390_vcpu_has_irq(vcpu, 0))
2504 return 0;
2505 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
2506 return 0;
2507 if (!vcpu->arch.gmap->pfault_enabled)
2508 return 0;
2509
2510 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
2511 hva += current->thread.gmap_addr & ~PAGE_MASK;
2512 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
2513 return 0;
2514
2515 rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
2516 return rc;
2517 }
2518
2519 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
2520 {
2521 int rc, cpuflags;
2522
2523 /*
2524 * On s390 notifications for arriving pages will be delivered directly
2525 * to the guest but the house keeping for completed pfaults is
2526 * handled outside the worker.
2527 */
2528 kvm_check_async_pf_completion(vcpu);
2529
2530 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
2531 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
2532
2533 if (need_resched())
2534 schedule();
2535
2536 if (test_cpu_flag(CIF_MCCK_PENDING))
2537 s390_handle_mcck();
2538
2539 if (!kvm_is_ucontrol(vcpu->kvm)) {
2540 rc = kvm_s390_deliver_pending_interrupts(vcpu);
2541 if (rc)
2542 return rc;
2543 }
2544
2545 rc = kvm_s390_handle_requests(vcpu);
2546 if (rc)
2547 return rc;
2548
2549 if (guestdbg_enabled(vcpu)) {
2550 kvm_s390_backup_guest_per_regs(vcpu);
2551 kvm_s390_patch_guest_per_regs(vcpu);
2552 }
2553
2554 vcpu->arch.sie_block->icptcode = 0;
2555 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
2556 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
2557 trace_kvm_s390_sie_enter(vcpu, cpuflags);
2558
2559 return 0;
2560 }
2561
2562 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
2563 {
2564 struct kvm_s390_pgm_info pgm_info = {
2565 .code = PGM_ADDRESSING,
2566 };
2567 u8 opcode, ilen;
2568 int rc;
2569
2570 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
2571 trace_kvm_s390_sie_fault(vcpu);
2572
2573 /*
2574 * We want to inject an addressing exception, which is defined as a
2575 * suppressing or terminating exception. However, since we came here
2576 * by a DAT access exception, the PSW still points to the faulting
2577 * instruction since DAT exceptions are nullifying. So we've got
2578 * to look up the current opcode to get the length of the instruction
2579 * to be able to forward the PSW.
2580 */
2581 rc = read_guest_instr(vcpu, &opcode, 1);
2582 ilen = insn_length(opcode);
2583 if (rc < 0) {
2584 return rc;
2585 } else if (rc) {
2586 /* Instruction-Fetching Exceptions - we can't detect the ilen.
2587 * Forward by arbitrary ilc, injection will take care of
2588 * nullification if necessary.
2589 */
2590 pgm_info = vcpu->arch.pgm;
2591 ilen = 4;
2592 }
2593 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
2594 kvm_s390_forward_psw(vcpu, ilen);
2595 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
2596 }
2597
2598 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
2599 {
2600 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
2601 vcpu->arch.sie_block->icptcode);
2602 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
2603
2604 if (guestdbg_enabled(vcpu))
2605 kvm_s390_restore_guest_per_regs(vcpu);
2606
2607 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
2608 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
2609
2610 if (vcpu->arch.sie_block->icptcode > 0) {
2611 int rc = kvm_handle_sie_intercept(vcpu);
2612
2613 if (rc != -EOPNOTSUPP)
2614 return rc;
2615 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
2616 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
2617 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
2618 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
2619 return -EREMOTE;
2620 } else if (exit_reason != -EFAULT) {
2621 vcpu->stat.exit_null++;
2622 return 0;
2623 } else if (kvm_is_ucontrol(vcpu->kvm)) {
2624 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
2625 vcpu->run->s390_ucontrol.trans_exc_code =
2626 current->thread.gmap_addr;
2627 vcpu->run->s390_ucontrol.pgm_code = 0x10;
2628 return -EREMOTE;
2629 } else if (current->thread.gmap_pfault) {
2630 trace_kvm_s390_major_guest_pfault(vcpu);
2631 current->thread.gmap_pfault = 0;
2632 if (kvm_arch_setup_async_pf(vcpu))
2633 return 0;
2634 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
2635 }
2636 return vcpu_post_run_fault_in_sie(vcpu);
2637 }
2638
2639 static int __vcpu_run(struct kvm_vcpu *vcpu)
2640 {
2641 int rc, exit_reason;
2642
2643 /*
2644 * We try to hold kvm->srcu during most of vcpu_run (except when run-
2645 * ning the guest), so that memslots (and other stuff) are protected
2646 */
2647 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2648
2649 do {
2650 rc = vcpu_pre_run(vcpu);
2651 if (rc)
2652 break;
2653
2654 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2655 /*
2656 * As PF_VCPU will be used in fault handler, between
2657 * guest_enter and guest_exit should be no uaccess.
2658 */
2659 local_irq_disable();
2660 guest_enter_irqoff();
2661 __disable_cpu_timer_accounting(vcpu);
2662 local_irq_enable();
2663 exit_reason = sie64a(vcpu->arch.sie_block,
2664 vcpu->run->s.regs.gprs);
2665 local_irq_disable();
2666 __enable_cpu_timer_accounting(vcpu);
2667 guest_exit_irqoff();
2668 local_irq_enable();
2669 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2670
2671 rc = vcpu_post_run(vcpu, exit_reason);
2672 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
2673
2674 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2675 return rc;
2676 }
2677
2678 static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2679 {
2680 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
2681 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
2682 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
2683 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
2684 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
2685 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
2686 /* some control register changes require a tlb flush */
2687 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2688 }
2689 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
2690 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
2691 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
2692 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
2693 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
2694 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
2695 }
2696 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
2697 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
2698 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
2699 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
2700 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2701 kvm_clear_async_pf_completion_queue(vcpu);
2702 }
2703 /*
2704 * If userspace sets the riccb (e.g. after migration) to a valid state,
2705 * we should enable RI here instead of doing the lazy enablement.
2706 */
2707 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) &&
2708 test_kvm_facility(vcpu->kvm, 64)) {
2709 struct runtime_instr_cb *riccb =
2710 (struct runtime_instr_cb *) &kvm_run->s.regs.riccb;
2711
2712 if (riccb->valid)
2713 vcpu->arch.sie_block->ecb3 |= 0x01;
2714 }
2715 save_access_regs(vcpu->arch.host_acrs);
2716 restore_access_regs(vcpu->run->s.regs.acrs);
2717 /* save host (userspace) fprs/vrs */
2718 save_fpu_regs();
2719 vcpu->arch.host_fpregs.fpc = current->thread.fpu.fpc;
2720 vcpu->arch.host_fpregs.regs = current->thread.fpu.regs;
2721 if (MACHINE_HAS_VX)
2722 current->thread.fpu.regs = vcpu->run->s.regs.vrs;
2723 else
2724 current->thread.fpu.regs = vcpu->run->s.regs.fprs;
2725 current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
2726 if (test_fp_ctl(current->thread.fpu.fpc))
2727 /* User space provided an invalid FPC, let's clear it */
2728 current->thread.fpu.fpc = 0;
2729
2730 kvm_run->kvm_dirty_regs = 0;
2731 }
2732
2733 static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2734 {
2735 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
2736 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
2737 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
2738 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
2739 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
2740 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
2741 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
2742 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
2743 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
2744 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
2745 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
2746 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
2747 save_access_regs(vcpu->run->s.regs.acrs);
2748 restore_access_regs(vcpu->arch.host_acrs);
2749 /* Save guest register state */
2750 save_fpu_regs();
2751 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
2752 /* Restore will be done lazily at return */
2753 current->thread.fpu.fpc = vcpu->arch.host_fpregs.fpc;
2754 current->thread.fpu.regs = vcpu->arch.host_fpregs.regs;
2755
2756 }
2757
2758 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2759 {
2760 int rc;
2761 sigset_t sigsaved;
2762
2763 if (guestdbg_exit_pending(vcpu)) {
2764 kvm_s390_prepare_debug_exit(vcpu);
2765 return 0;
2766 }
2767
2768 if (vcpu->sigset_active)
2769 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2770
2771 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
2772 kvm_s390_vcpu_start(vcpu);
2773 } else if (is_vcpu_stopped(vcpu)) {
2774 pr_err_ratelimited("can't run stopped vcpu %d\n",
2775 vcpu->vcpu_id);
2776 return -EINVAL;
2777 }
2778
2779 sync_regs(vcpu, kvm_run);
2780 enable_cpu_timer_accounting(vcpu);
2781
2782 might_fault();
2783 rc = __vcpu_run(vcpu);
2784
2785 if (signal_pending(current) && !rc) {
2786 kvm_run->exit_reason = KVM_EXIT_INTR;
2787 rc = -EINTR;
2788 }
2789
2790 if (guestdbg_exit_pending(vcpu) && !rc) {
2791 kvm_s390_prepare_debug_exit(vcpu);
2792 rc = 0;
2793 }
2794
2795 if (rc == -EREMOTE) {
2796 /* userspace support is needed, kvm_run has been prepared */
2797 rc = 0;
2798 }
2799
2800 disable_cpu_timer_accounting(vcpu);
2801 store_regs(vcpu, kvm_run);
2802
2803 if (vcpu->sigset_active)
2804 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2805
2806 vcpu->stat.exit_userspace++;
2807 return rc;
2808 }
2809
2810 /*
2811 * store status at address
2812 * we use have two special cases:
2813 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
2814 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
2815 */
2816 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
2817 {
2818 unsigned char archmode = 1;
2819 freg_t fprs[NUM_FPRS];
2820 unsigned int px;
2821 u64 clkcomp, cputm;
2822 int rc;
2823
2824 px = kvm_s390_get_prefix(vcpu);
2825 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
2826 if (write_guest_abs(vcpu, 163, &archmode, 1))
2827 return -EFAULT;
2828 gpa = 0;
2829 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
2830 if (write_guest_real(vcpu, 163, &archmode, 1))
2831 return -EFAULT;
2832 gpa = px;
2833 } else
2834 gpa -= __LC_FPREGS_SAVE_AREA;
2835
2836 /* manually convert vector registers if necessary */
2837 if (MACHINE_HAS_VX) {
2838 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
2839 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2840 fprs, 128);
2841 } else {
2842 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
2843 vcpu->run->s.regs.fprs, 128);
2844 }
2845 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
2846 vcpu->run->s.regs.gprs, 128);
2847 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
2848 &vcpu->arch.sie_block->gpsw, 16);
2849 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
2850 &px, 4);
2851 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
2852 &vcpu->run->s.regs.fpc, 4);
2853 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
2854 &vcpu->arch.sie_block->todpr, 4);
2855 cputm = kvm_s390_get_cpu_timer(vcpu);
2856 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
2857 &cputm, 8);
2858 clkcomp = vcpu->arch.sie_block->ckc >> 8;
2859 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
2860 &clkcomp, 8);
2861 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
2862 &vcpu->run->s.regs.acrs, 64);
2863 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
2864 &vcpu->arch.sie_block->gcr, 128);
2865 return rc ? -EFAULT : 0;
2866 }
2867
2868 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
2869 {
2870 /*
2871 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
2872 * switch in the run ioctl. Let's update our copies before we save
2873 * it into the save area
2874 */
2875 save_fpu_regs();
2876 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
2877 save_access_regs(vcpu->run->s.regs.acrs);
2878
2879 return kvm_s390_store_status_unloaded(vcpu, addr);
2880 }
2881
2882 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2883 {
2884 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
2885 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
2886 }
2887
2888 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
2889 {
2890 unsigned int i;
2891 struct kvm_vcpu *vcpu;
2892
2893 kvm_for_each_vcpu(i, vcpu, kvm) {
2894 __disable_ibs_on_vcpu(vcpu);
2895 }
2896 }
2897
2898 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
2899 {
2900 if (!sclp.has_ibs)
2901 return;
2902 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
2903 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
2904 }
2905
2906 void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
2907 {
2908 int i, online_vcpus, started_vcpus = 0;
2909
2910 if (!is_vcpu_stopped(vcpu))
2911 return;
2912
2913 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
2914 /* Only one cpu at a time may enter/leave the STOPPED state. */
2915 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2916 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2917
2918 for (i = 0; i < online_vcpus; i++) {
2919 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i]))
2920 started_vcpus++;
2921 }
2922
2923 if (started_vcpus == 0) {
2924 /* we're the only active VCPU -> speed it up */
2925 __enable_ibs_on_vcpu(vcpu);
2926 } else if (started_vcpus == 1) {
2927 /*
2928 * As we are starting a second VCPU, we have to disable
2929 * the IBS facility on all VCPUs to remove potentially
2930 * oustanding ENABLE requests.
2931 */
2932 __disable_ibs_on_all_vcpus(vcpu->kvm);
2933 }
2934
2935 atomic_andnot(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2936 /*
2937 * Another VCPU might have used IBS while we were offline.
2938 * Let's play safe and flush the VCPU at startup.
2939 */
2940 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2941 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2942 return;
2943 }
2944
2945 void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
2946 {
2947 int i, online_vcpus, started_vcpus = 0;
2948 struct kvm_vcpu *started_vcpu = NULL;
2949
2950 if (is_vcpu_stopped(vcpu))
2951 return;
2952
2953 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
2954 /* Only one cpu at a time may enter/leave the STOPPED state. */
2955 spin_lock(&vcpu->kvm->arch.start_stop_lock);
2956 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
2957
2958 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
2959 kvm_s390_clear_stop_irq(vcpu);
2960
2961 atomic_or(CPUSTAT_STOPPED, &vcpu->arch.sie_block->cpuflags);
2962 __disable_ibs_on_vcpu(vcpu);
2963
2964 for (i = 0; i < online_vcpus; i++) {
2965 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) {
2966 started_vcpus++;
2967 started_vcpu = vcpu->kvm->vcpus[i];
2968 }
2969 }
2970
2971 if (started_vcpus == 1) {
2972 /*
2973 * As we only have one VCPU left, we want to enable the
2974 * IBS facility for that VCPU to speed it up.
2975 */
2976 __enable_ibs_on_vcpu(started_vcpu);
2977 }
2978
2979 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
2980 return;
2981 }
2982
2983 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
2984 struct kvm_enable_cap *cap)
2985 {
2986 int r;
2987
2988 if (cap->flags)
2989 return -EINVAL;
2990
2991 switch (cap->cap) {
2992 case KVM_CAP_S390_CSS_SUPPORT:
2993 if (!vcpu->kvm->arch.css_support) {
2994 vcpu->kvm->arch.css_support = 1;
2995 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
2996 trace_kvm_s390_enable_css(vcpu->kvm);
2997 }
2998 r = 0;
2999 break;
3000 default:
3001 r = -EINVAL;
3002 break;
3003 }
3004 return r;
3005 }
3006
3007 static long kvm_s390_guest_mem_op(struct kvm_vcpu *vcpu,
3008 struct kvm_s390_mem_op *mop)
3009 {
3010 void __user *uaddr = (void __user *)mop->buf;
3011 void *tmpbuf = NULL;
3012 int r, srcu_idx;
3013 const u64 supported_flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION
3014 | KVM_S390_MEMOP_F_CHECK_ONLY;
3015
3016 if (mop->flags & ~supported_flags)
3017 return -EINVAL;
3018
3019 if (mop->size > MEM_OP_MAX_SIZE)
3020 return -E2BIG;
3021
3022 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
3023 tmpbuf = vmalloc(mop->size);
3024 if (!tmpbuf)
3025 return -ENOMEM;
3026 }
3027
3028 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3029
3030 switch (mop->op) {
3031 case KVM_S390_MEMOP_LOGICAL_READ:
3032 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3033 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3034 mop->size, GACC_FETCH);
3035 break;
3036 }
3037 r = read_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3038 if (r == 0) {
3039 if (copy_to_user(uaddr, tmpbuf, mop->size))
3040 r = -EFAULT;
3041 }
3042 break;
3043 case KVM_S390_MEMOP_LOGICAL_WRITE:
3044 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3045 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3046 mop->size, GACC_STORE);
3047 break;
3048 }
3049 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
3050 r = -EFAULT;
3051 break;
3052 }
3053 r = write_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3054 break;
3055 default:
3056 r = -EINVAL;
3057 }
3058
3059 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
3060
3061 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
3062 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
3063
3064 vfree(tmpbuf);
3065 return r;
3066 }
3067
3068 long kvm_arch_vcpu_ioctl(struct file *filp,
3069 unsigned int ioctl, unsigned long arg)
3070 {
3071 struct kvm_vcpu *vcpu = filp->private_data;
3072 void __user *argp = (void __user *)arg;
3073 int idx;
3074 long r;
3075
3076 switch (ioctl) {
3077 case KVM_S390_IRQ: {
3078 struct kvm_s390_irq s390irq;
3079
3080 r = -EFAULT;
3081 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
3082 break;
3083 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
3084 break;
3085 }
3086 case KVM_S390_INTERRUPT: {
3087 struct kvm_s390_interrupt s390int;
3088 struct kvm_s390_irq s390irq;
3089
3090 r = -EFAULT;
3091 if (copy_from_user(&s390int, argp, sizeof(s390int)))
3092 break;
3093 if (s390int_to_s390irq(&s390int, &s390irq))
3094 return -EINVAL;
3095 r = kvm_s390_inject_vcpu(vcpu, &s390irq);
3096 break;
3097 }
3098 case KVM_S390_STORE_STATUS:
3099 idx = srcu_read_lock(&vcpu->kvm->srcu);
3100 r = kvm_s390_vcpu_store_status(vcpu, arg);
3101 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3102 break;
3103 case KVM_S390_SET_INITIAL_PSW: {
3104 psw_t psw;
3105
3106 r = -EFAULT;
3107 if (copy_from_user(&psw, argp, sizeof(psw)))
3108 break;
3109 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
3110 break;
3111 }
3112 case KVM_S390_INITIAL_RESET:
3113 r = kvm_arch_vcpu_ioctl_initial_reset(vcpu);
3114 break;
3115 case KVM_SET_ONE_REG:
3116 case KVM_GET_ONE_REG: {
3117 struct kvm_one_reg reg;
3118 r = -EFAULT;
3119 if (copy_from_user(&reg, argp, sizeof(reg)))
3120 break;
3121 if (ioctl == KVM_SET_ONE_REG)
3122 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, &reg);
3123 else
3124 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, &reg);
3125 break;
3126 }
3127 #ifdef CONFIG_KVM_S390_UCONTROL
3128 case KVM_S390_UCAS_MAP: {
3129 struct kvm_s390_ucas_mapping ucasmap;
3130
3131 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3132 r = -EFAULT;
3133 break;
3134 }
3135
3136 if (!kvm_is_ucontrol(vcpu->kvm)) {
3137 r = -EINVAL;
3138 break;
3139 }
3140
3141 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
3142 ucasmap.vcpu_addr, ucasmap.length);
3143 break;
3144 }
3145 case KVM_S390_UCAS_UNMAP: {
3146 struct kvm_s390_ucas_mapping ucasmap;
3147
3148 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3149 r = -EFAULT;
3150 break;
3151 }
3152
3153 if (!kvm_is_ucontrol(vcpu->kvm)) {
3154 r = -EINVAL;
3155 break;
3156 }
3157
3158 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
3159 ucasmap.length);
3160 break;
3161 }
3162 #endif
3163 case KVM_S390_VCPU_FAULT: {
3164 r = gmap_fault(vcpu->arch.gmap, arg, 0);
3165 break;
3166 }
3167 case KVM_ENABLE_CAP:
3168 {
3169 struct kvm_enable_cap cap;
3170 r = -EFAULT;
3171 if (copy_from_user(&cap, argp, sizeof(cap)))
3172 break;
3173 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3174 break;
3175 }
3176 case KVM_S390_MEM_OP: {
3177 struct kvm_s390_mem_op mem_op;
3178
3179 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
3180 r = kvm_s390_guest_mem_op(vcpu, &mem_op);
3181 else
3182 r = -EFAULT;
3183 break;
3184 }
3185 case KVM_S390_SET_IRQ_STATE: {
3186 struct kvm_s390_irq_state irq_state;
3187
3188 r = -EFAULT;
3189 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3190 break;
3191 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
3192 irq_state.len == 0 ||
3193 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
3194 r = -EINVAL;
3195 break;
3196 }
3197 r = kvm_s390_set_irq_state(vcpu,
3198 (void __user *) irq_state.buf,
3199 irq_state.len);
3200 break;
3201 }
3202 case KVM_S390_GET_IRQ_STATE: {
3203 struct kvm_s390_irq_state irq_state;
3204
3205 r = -EFAULT;
3206 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3207 break;
3208 if (irq_state.len == 0) {
3209 r = -EINVAL;
3210 break;
3211 }
3212 r = kvm_s390_get_irq_state(vcpu,
3213 (__u8 __user *) irq_state.buf,
3214 irq_state.len);
3215 break;
3216 }
3217 default:
3218 r = -ENOTTY;
3219 }
3220 return r;
3221 }
3222
3223 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3224 {
3225 #ifdef CONFIG_KVM_S390_UCONTROL
3226 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
3227 && (kvm_is_ucontrol(vcpu->kvm))) {
3228 vmf->page = virt_to_page(vcpu->arch.sie_block);
3229 get_page(vmf->page);
3230 return 0;
3231 }
3232 #endif
3233 return VM_FAULT_SIGBUS;
3234 }
3235
3236 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
3237 unsigned long npages)
3238 {
3239 return 0;
3240 }
3241
3242 /* Section: memory related */
3243 int kvm_arch_prepare_memory_region(struct kvm *kvm,
3244 struct kvm_memory_slot *memslot,
3245 const struct kvm_userspace_memory_region *mem,
3246 enum kvm_mr_change change)
3247 {
3248 /* A few sanity checks. We can have memory slots which have to be
3249 located/ended at a segment boundary (1MB). The memory in userland is
3250 ok to be fragmented into various different vmas. It is okay to mmap()
3251 and munmap() stuff in this slot after doing this call at any time */
3252
3253 if (mem->userspace_addr & 0xffffful)
3254 return -EINVAL;
3255
3256 if (mem->memory_size & 0xffffful)
3257 return -EINVAL;
3258
3259 if (mem->guest_phys_addr + mem->memory_size > kvm->arch.mem_limit)
3260 return -EINVAL;
3261
3262 return 0;
3263 }
3264
3265 void kvm_arch_commit_memory_region(struct kvm *kvm,
3266 const struct kvm_userspace_memory_region *mem,
3267 const struct kvm_memory_slot *old,
3268 const struct kvm_memory_slot *new,
3269 enum kvm_mr_change change)
3270 {
3271 int rc;
3272
3273 /* If the basics of the memslot do not change, we do not want
3274 * to update the gmap. Every update causes several unnecessary
3275 * segment translation exceptions. This is usually handled just
3276 * fine by the normal fault handler + gmap, but it will also
3277 * cause faults on the prefix page of running guest CPUs.
3278 */
3279 if (old->userspace_addr == mem->userspace_addr &&
3280 old->base_gfn * PAGE_SIZE == mem->guest_phys_addr &&
3281 old->npages * PAGE_SIZE == mem->memory_size)
3282 return;
3283
3284 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr,
3285 mem->guest_phys_addr, mem->memory_size);
3286 if (rc)
3287 pr_warn("failed to commit memory region\n");
3288 return;
3289 }
3290
3291 static inline unsigned long nonhyp_mask(int i)
3292 {
3293 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
3294
3295 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
3296 }
3297
3298 void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu)
3299 {
3300 vcpu->valid_wakeup = false;
3301 }
3302
3303 static int __init kvm_s390_init(void)
3304 {
3305 int i;
3306
3307 if (!sclp.has_sief2) {
3308 pr_info("SIE not available\n");
3309 return -ENODEV;
3310 }
3311
3312 for (i = 0; i < 16; i++)
3313 kvm_s390_fac_list_mask[i] |=
3314 S390_lowcore.stfle_fac_list[i] & nonhyp_mask(i);
3315
3316 return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
3317 }
3318
3319 static void __exit kvm_s390_exit(void)
3320 {
3321 kvm_exit();
3322 }
3323
3324 module_init(kvm_s390_init);
3325 module_exit(kvm_s390_exit);
3326
3327 /*
3328 * Enable autoloading of the kvm module.
3329 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
3330 * since x86 takes a different approach.
3331 */
3332 #include <linux/miscdevice.h>
3333 MODULE_ALIAS_MISCDEV(KVM_MINOR);
3334 MODULE_ALIAS("devname:kvm");
CRIS linux kernel tree