Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * hosting IBM Z kernel virtual machines (s390x)
4 *
5 * Copyright IBM Corp. 2008, 2020
6 *
7 * Author(s): Carsten Otte <cotte@de.ibm.com>
8 * Christian Borntraeger <borntraeger@de.ibm.com>
9 * Christian Ehrhardt <ehrhardt@de.ibm.com>
10 * Jason J. Herne <jjherne@us.ibm.com>
11 */
12
13#define KMSG_COMPONENT "kvm-s390"
14#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
15
16#include <linux/compiler.h>
17#include <linux/err.h>
18#include <linux/fs.h>
19#include <linux/hrtimer.h>
20#include <linux/init.h>
21#include <linux/kvm.h>
22#include <linux/kvm_host.h>
23#include <linux/mman.h>
24#include <linux/module.h>
25#include <linux/moduleparam.h>
26#include <linux/random.h>
27#include <linux/slab.h>
28#include <linux/timer.h>
29#include <linux/vmalloc.h>
30#include <linux/bitmap.h>
31#include <linux/sched/signal.h>
32#include <linux/string.h>
33#include <linux/pgtable.h>
34#include <linux/mmu_notifier.h>
35
36#include <asm/asm-offsets.h>
37#include <asm/lowcore.h>
38#include <asm/stp.h>
39#include <asm/gmap.h>
40#include <asm/nmi.h>
41#include <asm/switch_to.h>
42#include <asm/isc.h>
43#include <asm/sclp.h>
44#include <asm/cpacf.h>
45#include <asm/timex.h>
46#include <asm/ap.h>
47#include <asm/uv.h>
48#include <asm/fpu/api.h>
49#include "kvm-s390.h"
50#include "gaccess.h"
51#include "pci.h"
52
53#define CREATE_TRACE_POINTS
54#include "trace.h"
55#include "trace-s390.h"
56
57#define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */
58#define LOCAL_IRQS 32
59#define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
60 (KVM_MAX_VCPUS + LOCAL_IRQS))
61
62const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
63 KVM_GENERIC_VM_STATS(),
64 STATS_DESC_COUNTER(VM, inject_io),
65 STATS_DESC_COUNTER(VM, inject_float_mchk),
66 STATS_DESC_COUNTER(VM, inject_pfault_done),
67 STATS_DESC_COUNTER(VM, inject_service_signal),
68 STATS_DESC_COUNTER(VM, inject_virtio),
69 STATS_DESC_COUNTER(VM, aen_forward)
70};
71
72const struct kvm_stats_header kvm_vm_stats_header = {
73 .name_size = KVM_STATS_NAME_SIZE,
74 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
75 .id_offset = sizeof(struct kvm_stats_header),
76 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
77 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
78 sizeof(kvm_vm_stats_desc),
79};
80
81const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
82 KVM_GENERIC_VCPU_STATS(),
83 STATS_DESC_COUNTER(VCPU, exit_userspace),
84 STATS_DESC_COUNTER(VCPU, exit_null),
85 STATS_DESC_COUNTER(VCPU, exit_external_request),
86 STATS_DESC_COUNTER(VCPU, exit_io_request),
87 STATS_DESC_COUNTER(VCPU, exit_external_interrupt),
88 STATS_DESC_COUNTER(VCPU, exit_stop_request),
89 STATS_DESC_COUNTER(VCPU, exit_validity),
90 STATS_DESC_COUNTER(VCPU, exit_instruction),
91 STATS_DESC_COUNTER(VCPU, exit_pei),
92 STATS_DESC_COUNTER(VCPU, halt_no_poll_steal),
93 STATS_DESC_COUNTER(VCPU, instruction_lctl),
94 STATS_DESC_COUNTER(VCPU, instruction_lctlg),
95 STATS_DESC_COUNTER(VCPU, instruction_stctl),
96 STATS_DESC_COUNTER(VCPU, instruction_stctg),
97 STATS_DESC_COUNTER(VCPU, exit_program_interruption),
98 STATS_DESC_COUNTER(VCPU, exit_instr_and_program),
99 STATS_DESC_COUNTER(VCPU, exit_operation_exception),
100 STATS_DESC_COUNTER(VCPU, deliver_ckc),
101 STATS_DESC_COUNTER(VCPU, deliver_cputm),
102 STATS_DESC_COUNTER(VCPU, deliver_external_call),
103 STATS_DESC_COUNTER(VCPU, deliver_emergency_signal),
104 STATS_DESC_COUNTER(VCPU, deliver_service_signal),
105 STATS_DESC_COUNTER(VCPU, deliver_virtio),
106 STATS_DESC_COUNTER(VCPU, deliver_stop_signal),
107 STATS_DESC_COUNTER(VCPU, deliver_prefix_signal),
108 STATS_DESC_COUNTER(VCPU, deliver_restart_signal),
109 STATS_DESC_COUNTER(VCPU, deliver_program),
110 STATS_DESC_COUNTER(VCPU, deliver_io),
111 STATS_DESC_COUNTER(VCPU, deliver_machine_check),
112 STATS_DESC_COUNTER(VCPU, exit_wait_state),
113 STATS_DESC_COUNTER(VCPU, inject_ckc),
114 STATS_DESC_COUNTER(VCPU, inject_cputm),
115 STATS_DESC_COUNTER(VCPU, inject_external_call),
116 STATS_DESC_COUNTER(VCPU, inject_emergency_signal),
117 STATS_DESC_COUNTER(VCPU, inject_mchk),
118 STATS_DESC_COUNTER(VCPU, inject_pfault_init),
119 STATS_DESC_COUNTER(VCPU, inject_program),
120 STATS_DESC_COUNTER(VCPU, inject_restart),
121 STATS_DESC_COUNTER(VCPU, inject_set_prefix),
122 STATS_DESC_COUNTER(VCPU, inject_stop_signal),
123 STATS_DESC_COUNTER(VCPU, instruction_epsw),
124 STATS_DESC_COUNTER(VCPU, instruction_gs),
125 STATS_DESC_COUNTER(VCPU, instruction_io_other),
126 STATS_DESC_COUNTER(VCPU, instruction_lpsw),
127 STATS_DESC_COUNTER(VCPU, instruction_lpswe),
128 STATS_DESC_COUNTER(VCPU, instruction_pfmf),
129 STATS_DESC_COUNTER(VCPU, instruction_ptff),
130 STATS_DESC_COUNTER(VCPU, instruction_sck),
131 STATS_DESC_COUNTER(VCPU, instruction_sckpf),
132 STATS_DESC_COUNTER(VCPU, instruction_stidp),
133 STATS_DESC_COUNTER(VCPU, instruction_spx),
134 STATS_DESC_COUNTER(VCPU, instruction_stpx),
135 STATS_DESC_COUNTER(VCPU, instruction_stap),
136 STATS_DESC_COUNTER(VCPU, instruction_iske),
137 STATS_DESC_COUNTER(VCPU, instruction_ri),
138 STATS_DESC_COUNTER(VCPU, instruction_rrbe),
139 STATS_DESC_COUNTER(VCPU, instruction_sske),
140 STATS_DESC_COUNTER(VCPU, instruction_ipte_interlock),
141 STATS_DESC_COUNTER(VCPU, instruction_stsi),
142 STATS_DESC_COUNTER(VCPU, instruction_stfl),
143 STATS_DESC_COUNTER(VCPU, instruction_tb),
144 STATS_DESC_COUNTER(VCPU, instruction_tpi),
145 STATS_DESC_COUNTER(VCPU, instruction_tprot),
146 STATS_DESC_COUNTER(VCPU, instruction_tsch),
147 STATS_DESC_COUNTER(VCPU, instruction_sie),
148 STATS_DESC_COUNTER(VCPU, instruction_essa),
149 STATS_DESC_COUNTER(VCPU, instruction_sthyi),
150 STATS_DESC_COUNTER(VCPU, instruction_sigp_sense),
151 STATS_DESC_COUNTER(VCPU, instruction_sigp_sense_running),
152 STATS_DESC_COUNTER(VCPU, instruction_sigp_external_call),
153 STATS_DESC_COUNTER(VCPU, instruction_sigp_emergency),
154 STATS_DESC_COUNTER(VCPU, instruction_sigp_cond_emergency),
155 STATS_DESC_COUNTER(VCPU, instruction_sigp_start),
156 STATS_DESC_COUNTER(VCPU, instruction_sigp_stop),
157 STATS_DESC_COUNTER(VCPU, instruction_sigp_stop_store_status),
158 STATS_DESC_COUNTER(VCPU, instruction_sigp_store_status),
159 STATS_DESC_COUNTER(VCPU, instruction_sigp_store_adtl_status),
160 STATS_DESC_COUNTER(VCPU, instruction_sigp_arch),
161 STATS_DESC_COUNTER(VCPU, instruction_sigp_prefix),
162 STATS_DESC_COUNTER(VCPU, instruction_sigp_restart),
163 STATS_DESC_COUNTER(VCPU, instruction_sigp_init_cpu_reset),
164 STATS_DESC_COUNTER(VCPU, instruction_sigp_cpu_reset),
165 STATS_DESC_COUNTER(VCPU, instruction_sigp_unknown),
166 STATS_DESC_COUNTER(VCPU, instruction_diagnose_10),
167 STATS_DESC_COUNTER(VCPU, instruction_diagnose_44),
168 STATS_DESC_COUNTER(VCPU, instruction_diagnose_9c),
169 STATS_DESC_COUNTER(VCPU, diag_9c_ignored),
170 STATS_DESC_COUNTER(VCPU, diag_9c_forward),
171 STATS_DESC_COUNTER(VCPU, instruction_diagnose_258),
172 STATS_DESC_COUNTER(VCPU, instruction_diagnose_308),
173 STATS_DESC_COUNTER(VCPU, instruction_diagnose_500),
174 STATS_DESC_COUNTER(VCPU, instruction_diagnose_other),
175 STATS_DESC_COUNTER(VCPU, pfault_sync)
176};
177
178const struct kvm_stats_header kvm_vcpu_stats_header = {
179 .name_size = KVM_STATS_NAME_SIZE,
180 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
181 .id_offset = sizeof(struct kvm_stats_header),
182 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
183 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
184 sizeof(kvm_vcpu_stats_desc),
185};
186
187/* allow nested virtualization in KVM (if enabled by user space) */
188static int nested;
189module_param(nested, int, S_IRUGO);
190MODULE_PARM_DESC(nested, "Nested virtualization support");
191
192/* allow 1m huge page guest backing, if !nested */
193static int hpage;
194module_param(hpage, int, 0444);
195MODULE_PARM_DESC(hpage, "1m huge page backing support");
196
197/* maximum percentage of steal time for polling. >100 is treated like 100 */
198static u8 halt_poll_max_steal = 10;
199module_param(halt_poll_max_steal, byte, 0644);
200MODULE_PARM_DESC(halt_poll_max_steal, "Maximum percentage of steal time to allow polling");
201
202/* if set to true, the GISA will be initialized and used if available */
203static bool use_gisa = true;
204module_param(use_gisa, bool, 0644);
205MODULE_PARM_DESC(use_gisa, "Use the GISA if the host supports it.");
206
207/* maximum diag9c forwarding per second */
208unsigned int diag9c_forwarding_hz;
209module_param(diag9c_forwarding_hz, uint, 0644);
210MODULE_PARM_DESC(diag9c_forwarding_hz, "Maximum diag9c forwarding per second, 0 to turn off");
211
212/*
213 * allow asynchronous deinit for protected guests; enable by default since
214 * the feature is opt-in anyway
215 */
216static int async_destroy = 1;
217module_param(async_destroy, int, 0444);
218MODULE_PARM_DESC(async_destroy, "Asynchronous destroy for protected guests");
219
220/*
221 * For now we handle at most 16 double words as this is what the s390 base
222 * kernel handles and stores in the prefix page. If we ever need to go beyond
223 * this, this requires changes to code, but the external uapi can stay.
224 */
225#define SIZE_INTERNAL 16
226
227/*
228 * Base feature mask that defines default mask for facilities. Consists of the
229 * defines in FACILITIES_KVM and the non-hypervisor managed bits.
230 */
231static unsigned long kvm_s390_fac_base[SIZE_INTERNAL] = { FACILITIES_KVM };
232/*
233 * Extended feature mask. Consists of the defines in FACILITIES_KVM_CPUMODEL
234 * and defines the facilities that can be enabled via a cpu model.
235 */
236static unsigned long kvm_s390_fac_ext[SIZE_INTERNAL] = { FACILITIES_KVM_CPUMODEL };
237
238static unsigned long kvm_s390_fac_size(void)
239{
240 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_MASK_SIZE_U64);
241 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_LIST_SIZE_U64);
242 BUILD_BUG_ON(SIZE_INTERNAL * sizeof(unsigned long) >
243 sizeof(stfle_fac_list));
244
245 return SIZE_INTERNAL;
246}
247
248/* available cpu features supported by kvm */
249static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
250/* available subfunctions indicated via query / "test bit" */
251static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
252
253static struct gmap_notifier gmap_notifier;
254static struct gmap_notifier vsie_gmap_notifier;
255debug_info_t *kvm_s390_dbf;
256debug_info_t *kvm_s390_dbf_uv;
257
258/* Section: not file related */
259int kvm_arch_hardware_enable(void)
260{
261 /* every s390 is virtualization enabled ;-) */
262 return 0;
263}
264
265int kvm_arch_check_processor_compat(void *opaque)
266{
267 return 0;
268}
269
270/* forward declarations */
271static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
272 unsigned long end);
273static int sca_switch_to_extended(struct kvm *kvm);
274
275static void kvm_clock_sync_scb(struct kvm_s390_sie_block *scb, u64 delta)
276{
277 u8 delta_idx = 0;
278
279 /*
280 * The TOD jumps by delta, we have to compensate this by adding
281 * -delta to the epoch.
282 */
283 delta = -delta;
284
285 /* sign-extension - we're adding to signed values below */
286 if ((s64)delta < 0)
287 delta_idx = -1;
288
289 scb->epoch += delta;
290 if (scb->ecd & ECD_MEF) {
291 scb->epdx += delta_idx;
292 if (scb->epoch < delta)
293 scb->epdx += 1;
294 }
295}
296
297/*
298 * This callback is executed during stop_machine(). All CPUs are therefore
299 * temporarily stopped. In order not to change guest behavior, we have to
300 * disable preemption whenever we touch the epoch of kvm and the VCPUs,
301 * so a CPU won't be stopped while calculating with the epoch.
302 */
303static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
304 void *v)
305{
306 struct kvm *kvm;
307 struct kvm_vcpu *vcpu;
308 unsigned long i;
309 unsigned long long *delta = v;
310
311 list_for_each_entry(kvm, &vm_list, vm_list) {
312 kvm_for_each_vcpu(i, vcpu, kvm) {
313 kvm_clock_sync_scb(vcpu->arch.sie_block, *delta);
314 if (i == 0) {
315 kvm->arch.epoch = vcpu->arch.sie_block->epoch;
316 kvm->arch.epdx = vcpu->arch.sie_block->epdx;
317 }
318 if (vcpu->arch.cputm_enabled)
319 vcpu->arch.cputm_start += *delta;
320 if (vcpu->arch.vsie_block)
321 kvm_clock_sync_scb(vcpu->arch.vsie_block,
322 *delta);
323 }
324 }
325 return NOTIFY_OK;
326}
327
328static struct notifier_block kvm_clock_notifier = {
329 .notifier_call = kvm_clock_sync,
330};
331
332int kvm_arch_hardware_setup(void *opaque)
333{
334 gmap_notifier.notifier_call = kvm_gmap_notifier;
335 gmap_register_pte_notifier(&gmap_notifier);
336 vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
337 gmap_register_pte_notifier(&vsie_gmap_notifier);
338 atomic_notifier_chain_register(&s390_epoch_delta_notifier,
339 &kvm_clock_notifier);
340 return 0;
341}
342
343void kvm_arch_hardware_unsetup(void)
344{
345 gmap_unregister_pte_notifier(&gmap_notifier);
346 gmap_unregister_pte_notifier(&vsie_gmap_notifier);
347 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
348 &kvm_clock_notifier);
349}
350
351static void allow_cpu_feat(unsigned long nr)
352{
353 set_bit_inv(nr, kvm_s390_available_cpu_feat);
354}
355
356static inline int plo_test_bit(unsigned char nr)
357{
358 unsigned long function = (unsigned long)nr | 0x100;
359 int cc;
360
361 asm volatile(
362 " lgr 0,%[function]\n"
363 /* Parameter registers are ignored for "test bit" */
364 " plo 0,0,0,0(0)\n"
365 " ipm %0\n"
366 " srl %0,28\n"
367 : "=d" (cc)
368 : [function] "d" (function)
369 : "cc", "0");
370 return cc == 0;
371}
372
373static __always_inline void __insn32_query(unsigned int opcode, u8 *query)
374{
375 asm volatile(
376 " lghi 0,0\n"
377 " lgr 1,%[query]\n"
378 /* Parameter registers are ignored */
379 " .insn rrf,%[opc] << 16,2,4,6,0\n"
380 :
381 : [query] "d" ((unsigned long)query), [opc] "i" (opcode)
382 : "cc", "memory", "0", "1");
383}
384
385#define INSN_SORTL 0xb938
386#define INSN_DFLTCC 0xb939
387
388static void kvm_s390_cpu_feat_init(void)
389{
390 int i;
391
392 for (i = 0; i < 256; ++i) {
393 if (plo_test_bit(i))
394 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
395 }
396
397 if (test_facility(28)) /* TOD-clock steering */
398 ptff(kvm_s390_available_subfunc.ptff,
399 sizeof(kvm_s390_available_subfunc.ptff),
400 PTFF_QAF);
401
402 if (test_facility(17)) { /* MSA */
403 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
404 kvm_s390_available_subfunc.kmac);
405 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
406 kvm_s390_available_subfunc.kmc);
407 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
408 kvm_s390_available_subfunc.km);
409 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
410 kvm_s390_available_subfunc.kimd);
411 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
412 kvm_s390_available_subfunc.klmd);
413 }
414 if (test_facility(76)) /* MSA3 */
415 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
416 kvm_s390_available_subfunc.pckmo);
417 if (test_facility(77)) { /* MSA4 */
418 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
419 kvm_s390_available_subfunc.kmctr);
420 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
421 kvm_s390_available_subfunc.kmf);
422 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
423 kvm_s390_available_subfunc.kmo);
424 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
425 kvm_s390_available_subfunc.pcc);
426 }
427 if (test_facility(57)) /* MSA5 */
428 __cpacf_query(CPACF_PRNO, (cpacf_mask_t *)
429 kvm_s390_available_subfunc.ppno);
430
431 if (test_facility(146)) /* MSA8 */
432 __cpacf_query(CPACF_KMA, (cpacf_mask_t *)
433 kvm_s390_available_subfunc.kma);
434
435 if (test_facility(155)) /* MSA9 */
436 __cpacf_query(CPACF_KDSA, (cpacf_mask_t *)
437 kvm_s390_available_subfunc.kdsa);
438
439 if (test_facility(150)) /* SORTL */
440 __insn32_query(INSN_SORTL, kvm_s390_available_subfunc.sortl);
441
442 if (test_facility(151)) /* DFLTCC */
443 __insn32_query(INSN_DFLTCC, kvm_s390_available_subfunc.dfltcc);
444
445 if (MACHINE_HAS_ESOP)
446 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
447 /*
448 * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
449 * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
450 */
451 if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
452 !test_facility(3) || !nested)
453 return;
454 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
455 if (sclp.has_64bscao)
456 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
457 if (sclp.has_siif)
458 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
459 if (sclp.has_gpere)
460 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
461 if (sclp.has_gsls)
462 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
463 if (sclp.has_ib)
464 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
465 if (sclp.has_cei)
466 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
467 if (sclp.has_ibs)
468 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
469 if (sclp.has_kss)
470 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_KSS);
471 /*
472 * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
473 * all skey handling functions read/set the skey from the PGSTE
474 * instead of the real storage key.
475 *
476 * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
477 * pages being detected as preserved although they are resident.
478 *
479 * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
480 * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
481 *
482 * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
483 * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
484 * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
485 *
486 * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
487 * cannot easily shadow the SCA because of the ipte lock.
488 */
489}
490
491int kvm_arch_init(void *opaque)
492{
493 int rc = -ENOMEM;
494
495 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
496 if (!kvm_s390_dbf)
497 return -ENOMEM;
498
499 kvm_s390_dbf_uv = debug_register("kvm-uv", 32, 1, 7 * sizeof(long));
500 if (!kvm_s390_dbf_uv)
501 goto out;
502
503 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view) ||
504 debug_register_view(kvm_s390_dbf_uv, &debug_sprintf_view))
505 goto out;
506
507 kvm_s390_cpu_feat_init();
508
509 /* Register floating interrupt controller interface. */
510 rc = kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
511 if (rc) {
512 pr_err("A FLIC registration call failed with rc=%d\n", rc);
513 goto out;
514 }
515
516 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
517 rc = kvm_s390_pci_init();
518 if (rc) {
519 pr_err("Unable to allocate AIFT for PCI\n");
520 goto out;
521 }
522 }
523
524 rc = kvm_s390_gib_init(GAL_ISC);
525 if (rc)
526 goto out;
527
528 return 0;
529
530out:
531 kvm_arch_exit();
532 return rc;
533}
534
535void kvm_arch_exit(void)
536{
537 kvm_s390_gib_destroy();
538 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
539 kvm_s390_pci_exit();
540 debug_unregister(kvm_s390_dbf);
541 debug_unregister(kvm_s390_dbf_uv);
542}
543
544/* Section: device related */
545long kvm_arch_dev_ioctl(struct file *filp,
546 unsigned int ioctl, unsigned long arg)
547{
548 if (ioctl == KVM_S390_ENABLE_SIE)
549 return s390_enable_sie();
550 return -EINVAL;
551}
552
553int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
554{
555 int r;
556
557 switch (ext) {
558 case KVM_CAP_S390_PSW:
559 case KVM_CAP_S390_GMAP:
560 case KVM_CAP_SYNC_MMU:
561#ifdef CONFIG_KVM_S390_UCONTROL
562 case KVM_CAP_S390_UCONTROL:
563#endif
564 case KVM_CAP_ASYNC_PF:
565 case KVM_CAP_SYNC_REGS:
566 case KVM_CAP_ONE_REG:
567 case KVM_CAP_ENABLE_CAP:
568 case KVM_CAP_S390_CSS_SUPPORT:
569 case KVM_CAP_IOEVENTFD:
570 case KVM_CAP_DEVICE_CTRL:
571 case KVM_CAP_S390_IRQCHIP:
572 case KVM_CAP_VM_ATTRIBUTES:
573 case KVM_CAP_MP_STATE:
574 case KVM_CAP_IMMEDIATE_EXIT:
575 case KVM_CAP_S390_INJECT_IRQ:
576 case KVM_CAP_S390_USER_SIGP:
577 case KVM_CAP_S390_USER_STSI:
578 case KVM_CAP_S390_SKEYS:
579 case KVM_CAP_S390_IRQ_STATE:
580 case KVM_CAP_S390_USER_INSTR0:
581 case KVM_CAP_S390_CMMA_MIGRATION:
582 case KVM_CAP_S390_AIS:
583 case KVM_CAP_S390_AIS_MIGRATION:
584 case KVM_CAP_S390_VCPU_RESETS:
585 case KVM_CAP_SET_GUEST_DEBUG:
586 case KVM_CAP_S390_DIAG318:
587 case KVM_CAP_S390_MEM_OP_EXTENSION:
588 r = 1;
589 break;
590 case KVM_CAP_SET_GUEST_DEBUG2:
591 r = KVM_GUESTDBG_VALID_MASK;
592 break;
593 case KVM_CAP_S390_HPAGE_1M:
594 r = 0;
595 if (hpage && !kvm_is_ucontrol(kvm))
596 r = 1;
597 break;
598 case KVM_CAP_S390_MEM_OP:
599 r = MEM_OP_MAX_SIZE;
600 break;
601 case KVM_CAP_NR_VCPUS:
602 case KVM_CAP_MAX_VCPUS:
603 case KVM_CAP_MAX_VCPU_ID:
604 r = KVM_S390_BSCA_CPU_SLOTS;
605 if (!kvm_s390_use_sca_entries())
606 r = KVM_MAX_VCPUS;
607 else if (sclp.has_esca && sclp.has_64bscao)
608 r = KVM_S390_ESCA_CPU_SLOTS;
609 if (ext == KVM_CAP_NR_VCPUS)
610 r = min_t(unsigned int, num_online_cpus(), r);
611 break;
612 case KVM_CAP_S390_COW:
613 r = MACHINE_HAS_ESOP;
614 break;
615 case KVM_CAP_S390_VECTOR_REGISTERS:
616 r = MACHINE_HAS_VX;
617 break;
618 case KVM_CAP_S390_RI:
619 r = test_facility(64);
620 break;
621 case KVM_CAP_S390_GS:
622 r = test_facility(133);
623 break;
624 case KVM_CAP_S390_BPB:
625 r = test_facility(82);
626 break;
627 case KVM_CAP_S390_PROTECTED_ASYNC_DISABLE:
628 r = async_destroy && is_prot_virt_host();
629 break;
630 case KVM_CAP_S390_PROTECTED:
631 r = is_prot_virt_host();
632 break;
633 case KVM_CAP_S390_PROTECTED_DUMP: {
634 u64 pv_cmds_dump[] = {
635 BIT_UVC_CMD_DUMP_INIT,
636 BIT_UVC_CMD_DUMP_CONFIG_STOR_STATE,
637 BIT_UVC_CMD_DUMP_CPU,
638 BIT_UVC_CMD_DUMP_COMPLETE,
639 };
640 int i;
641
642 r = is_prot_virt_host();
643
644 for (i = 0; i < ARRAY_SIZE(pv_cmds_dump); i++) {
645 if (!test_bit_inv(pv_cmds_dump[i],
646 (unsigned long *)&uv_info.inst_calls_list)) {
647 r = 0;
648 break;
649 }
650 }
651 break;
652 }
653 case KVM_CAP_S390_ZPCI_OP:
654 r = kvm_s390_pci_interp_allowed();
655 break;
656 case KVM_CAP_S390_CPU_TOPOLOGY:
657 r = test_facility(11);
658 break;
659 default:
660 r = 0;
661 }
662 return r;
663}
664
665void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
666{
667 int i;
668 gfn_t cur_gfn, last_gfn;
669 unsigned long gaddr, vmaddr;
670 struct gmap *gmap = kvm->arch.gmap;
671 DECLARE_BITMAP(bitmap, _PAGE_ENTRIES);
672
673 /* Loop over all guest segments */
674 cur_gfn = memslot->base_gfn;
675 last_gfn = memslot->base_gfn + memslot->npages;
676 for (; cur_gfn <= last_gfn; cur_gfn += _PAGE_ENTRIES) {
677 gaddr = gfn_to_gpa(cur_gfn);
678 vmaddr = gfn_to_hva_memslot(memslot, cur_gfn);
679 if (kvm_is_error_hva(vmaddr))
680 continue;
681
682 bitmap_zero(bitmap, _PAGE_ENTRIES);
683 gmap_sync_dirty_log_pmd(gmap, bitmap, gaddr, vmaddr);
684 for (i = 0; i < _PAGE_ENTRIES; i++) {
685 if (test_bit(i, bitmap))
686 mark_page_dirty(kvm, cur_gfn + i);
687 }
688
689 if (fatal_signal_pending(current))
690 return;
691 cond_resched();
692 }
693}
694
695/* Section: vm related */
696static void sca_del_vcpu(struct kvm_vcpu *vcpu);
697
698/*
699 * Get (and clear) the dirty memory log for a memory slot.
700 */
701int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
702 struct kvm_dirty_log *log)
703{
704 int r;
705 unsigned long n;
706 struct kvm_memory_slot *memslot;
707 int is_dirty;
708
709 if (kvm_is_ucontrol(kvm))
710 return -EINVAL;
711
712 mutex_lock(&kvm->slots_lock);
713
714 r = -EINVAL;
715 if (log->slot >= KVM_USER_MEM_SLOTS)
716 goto out;
717
718 r = kvm_get_dirty_log(kvm, log, &is_dirty, &memslot);
719 if (r)
720 goto out;
721
722 /* Clear the dirty log */
723 if (is_dirty) {
724 n = kvm_dirty_bitmap_bytes(memslot);
725 memset(memslot->dirty_bitmap, 0, n);
726 }
727 r = 0;
728out:
729 mutex_unlock(&kvm->slots_lock);
730 return r;
731}
732
733static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
734{
735 unsigned long i;
736 struct kvm_vcpu *vcpu;
737
738 kvm_for_each_vcpu(i, vcpu, kvm) {
739 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
740 }
741}
742
743int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
744{
745 int r;
746
747 if (cap->flags)
748 return -EINVAL;
749
750 switch (cap->cap) {
751 case KVM_CAP_S390_IRQCHIP:
752 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
753 kvm->arch.use_irqchip = 1;
754 r = 0;
755 break;
756 case KVM_CAP_S390_USER_SIGP:
757 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
758 kvm->arch.user_sigp = 1;
759 r = 0;
760 break;
761 case KVM_CAP_S390_VECTOR_REGISTERS:
762 mutex_lock(&kvm->lock);
763 if (kvm->created_vcpus) {
764 r = -EBUSY;
765 } else if (MACHINE_HAS_VX) {
766 set_kvm_facility(kvm->arch.model.fac_mask, 129);
767 set_kvm_facility(kvm->arch.model.fac_list, 129);
768 if (test_facility(134)) {
769 set_kvm_facility(kvm->arch.model.fac_mask, 134);
770 set_kvm_facility(kvm->arch.model.fac_list, 134);
771 }
772 if (test_facility(135)) {
773 set_kvm_facility(kvm->arch.model.fac_mask, 135);
774 set_kvm_facility(kvm->arch.model.fac_list, 135);
775 }
776 if (test_facility(148)) {
777 set_kvm_facility(kvm->arch.model.fac_mask, 148);
778 set_kvm_facility(kvm->arch.model.fac_list, 148);
779 }
780 if (test_facility(152)) {
781 set_kvm_facility(kvm->arch.model.fac_mask, 152);
782 set_kvm_facility(kvm->arch.model.fac_list, 152);
783 }
784 if (test_facility(192)) {
785 set_kvm_facility(kvm->arch.model.fac_mask, 192);
786 set_kvm_facility(kvm->arch.model.fac_list, 192);
787 }
788 r = 0;
789 } else
790 r = -EINVAL;
791 mutex_unlock(&kvm->lock);
792 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
793 r ? "(not available)" : "(success)");
794 break;
795 case KVM_CAP_S390_RI:
796 r = -EINVAL;
797 mutex_lock(&kvm->lock);
798 if (kvm->created_vcpus) {
799 r = -EBUSY;
800 } else if (test_facility(64)) {
801 set_kvm_facility(kvm->arch.model.fac_mask, 64);
802 set_kvm_facility(kvm->arch.model.fac_list, 64);
803 r = 0;
804 }
805 mutex_unlock(&kvm->lock);
806 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
807 r ? "(not available)" : "(success)");
808 break;
809 case KVM_CAP_S390_AIS:
810 mutex_lock(&kvm->lock);
811 if (kvm->created_vcpus) {
812 r = -EBUSY;
813 } else {
814 set_kvm_facility(kvm->arch.model.fac_mask, 72);
815 set_kvm_facility(kvm->arch.model.fac_list, 72);
816 r = 0;
817 }
818 mutex_unlock(&kvm->lock);
819 VM_EVENT(kvm, 3, "ENABLE: AIS %s",
820 r ? "(not available)" : "(success)");
821 break;
822 case KVM_CAP_S390_GS:
823 r = -EINVAL;
824 mutex_lock(&kvm->lock);
825 if (kvm->created_vcpus) {
826 r = -EBUSY;
827 } else if (test_facility(133)) {
828 set_kvm_facility(kvm->arch.model.fac_mask, 133);
829 set_kvm_facility(kvm->arch.model.fac_list, 133);
830 r = 0;
831 }
832 mutex_unlock(&kvm->lock);
833 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_GS %s",
834 r ? "(not available)" : "(success)");
835 break;
836 case KVM_CAP_S390_HPAGE_1M:
837 mutex_lock(&kvm->lock);
838 if (kvm->created_vcpus)
839 r = -EBUSY;
840 else if (!hpage || kvm->arch.use_cmma || kvm_is_ucontrol(kvm))
841 r = -EINVAL;
842 else {
843 r = 0;
844 mmap_write_lock(kvm->mm);
845 kvm->mm->context.allow_gmap_hpage_1m = 1;
846 mmap_write_unlock(kvm->mm);
847 /*
848 * We might have to create fake 4k page
849 * tables. To avoid that the hardware works on
850 * stale PGSTEs, we emulate these instructions.
851 */
852 kvm->arch.use_skf = 0;
853 kvm->arch.use_pfmfi = 0;
854 }
855 mutex_unlock(&kvm->lock);
856 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_HPAGE %s",
857 r ? "(not available)" : "(success)");
858 break;
859 case KVM_CAP_S390_USER_STSI:
860 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
861 kvm->arch.user_stsi = 1;
862 r = 0;
863 break;
864 case KVM_CAP_S390_USER_INSTR0:
865 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
866 kvm->arch.user_instr0 = 1;
867 icpt_operexc_on_all_vcpus(kvm);
868 r = 0;
869 break;
870 case KVM_CAP_S390_CPU_TOPOLOGY:
871 r = -EINVAL;
872 mutex_lock(&kvm->lock);
873 if (kvm->created_vcpus) {
874 r = -EBUSY;
875 } else if (test_facility(11)) {
876 set_kvm_facility(kvm->arch.model.fac_mask, 11);
877 set_kvm_facility(kvm->arch.model.fac_list, 11);
878 r = 0;
879 }
880 mutex_unlock(&kvm->lock);
881 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_CPU_TOPOLOGY %s",
882 r ? "(not available)" : "(success)");
883 break;
884 default:
885 r = -EINVAL;
886 break;
887 }
888 return r;
889}
890
891static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
892{
893 int ret;
894
895 switch (attr->attr) {
896 case KVM_S390_VM_MEM_LIMIT_SIZE:
897 ret = 0;
898 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
899 kvm->arch.mem_limit);
900 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
901 ret = -EFAULT;
902 break;
903 default:
904 ret = -ENXIO;
905 break;
906 }
907 return ret;
908}
909
910static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
911{
912 int ret;
913 unsigned int idx;
914 switch (attr->attr) {
915 case KVM_S390_VM_MEM_ENABLE_CMMA:
916 ret = -ENXIO;
917 if (!sclp.has_cmma)
918 break;
919
920 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
921 mutex_lock(&kvm->lock);
922 if (kvm->created_vcpus)
923 ret = -EBUSY;
924 else if (kvm->mm->context.allow_gmap_hpage_1m)
925 ret = -EINVAL;
926 else {
927 kvm->arch.use_cmma = 1;
928 /* Not compatible with cmma. */
929 kvm->arch.use_pfmfi = 0;
930 ret = 0;
931 }
932 mutex_unlock(&kvm->lock);
933 break;
934 case KVM_S390_VM_MEM_CLR_CMMA:
935 ret = -ENXIO;
936 if (!sclp.has_cmma)
937 break;
938 ret = -EINVAL;
939 if (!kvm->arch.use_cmma)
940 break;
941
942 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
943 mutex_lock(&kvm->lock);
944 idx = srcu_read_lock(&kvm->srcu);
945 s390_reset_cmma(kvm->arch.gmap->mm);
946 srcu_read_unlock(&kvm->srcu, idx);
947 mutex_unlock(&kvm->lock);
948 ret = 0;
949 break;
950 case KVM_S390_VM_MEM_LIMIT_SIZE: {
951 unsigned long new_limit;
952
953 if (kvm_is_ucontrol(kvm))
954 return -EINVAL;
955
956 if (get_user(new_limit, (u64 __user *)attr->addr))
957 return -EFAULT;
958
959 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
960 new_limit > kvm->arch.mem_limit)
961 return -E2BIG;
962
963 if (!new_limit)
964 return -EINVAL;
965
966 /* gmap_create takes last usable address */
967 if (new_limit != KVM_S390_NO_MEM_LIMIT)
968 new_limit -= 1;
969
970 ret = -EBUSY;
971 mutex_lock(&kvm->lock);
972 if (!kvm->created_vcpus) {
973 /* gmap_create will round the limit up */
974 struct gmap *new = gmap_create(current->mm, new_limit);
975
976 if (!new) {
977 ret = -ENOMEM;
978 } else {
979 gmap_remove(kvm->arch.gmap);
980 new->private = kvm;
981 kvm->arch.gmap = new;
982 ret = 0;
983 }
984 }
985 mutex_unlock(&kvm->lock);
986 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
987 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
988 (void *) kvm->arch.gmap->asce);
989 break;
990 }
991 default:
992 ret = -ENXIO;
993 break;
994 }
995 return ret;
996}
997
998static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
999
1000void kvm_s390_vcpu_crypto_reset_all(struct kvm *kvm)
1001{
1002 struct kvm_vcpu *vcpu;
1003 unsigned long i;
1004
1005 kvm_s390_vcpu_block_all(kvm);
1006
1007 kvm_for_each_vcpu(i, vcpu, kvm) {
1008 kvm_s390_vcpu_crypto_setup(vcpu);
1009 /* recreate the shadow crycb by leaving the VSIE handler */
1010 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu);
1011 }
1012
1013 kvm_s390_vcpu_unblock_all(kvm);
1014}
1015
1016static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
1017{
1018 mutex_lock(&kvm->lock);
1019 switch (attr->attr) {
1020 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1021 if (!test_kvm_facility(kvm, 76)) {
1022 mutex_unlock(&kvm->lock);
1023 return -EINVAL;
1024 }
1025 get_random_bytes(
1026 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1027 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1028 kvm->arch.crypto.aes_kw = 1;
1029 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
1030 break;
1031 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1032 if (!test_kvm_facility(kvm, 76)) {
1033 mutex_unlock(&kvm->lock);
1034 return -EINVAL;
1035 }
1036 get_random_bytes(
1037 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1038 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1039 kvm->arch.crypto.dea_kw = 1;
1040 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
1041 break;
1042 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1043 if (!test_kvm_facility(kvm, 76)) {
1044 mutex_unlock(&kvm->lock);
1045 return -EINVAL;
1046 }
1047 kvm->arch.crypto.aes_kw = 0;
1048 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
1049 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1050 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
1051 break;
1052 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1053 if (!test_kvm_facility(kvm, 76)) {
1054 mutex_unlock(&kvm->lock);
1055 return -EINVAL;
1056 }
1057 kvm->arch.crypto.dea_kw = 0;
1058 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
1059 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1060 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
1061 break;
1062 case KVM_S390_VM_CRYPTO_ENABLE_APIE:
1063 if (!ap_instructions_available()) {
1064 mutex_unlock(&kvm->lock);
1065 return -EOPNOTSUPP;
1066 }
1067 kvm->arch.crypto.apie = 1;
1068 break;
1069 case KVM_S390_VM_CRYPTO_DISABLE_APIE:
1070 if (!ap_instructions_available()) {
1071 mutex_unlock(&kvm->lock);
1072 return -EOPNOTSUPP;
1073 }
1074 kvm->arch.crypto.apie = 0;
1075 break;
1076 default:
1077 mutex_unlock(&kvm->lock);
1078 return -ENXIO;
1079 }
1080
1081 kvm_s390_vcpu_crypto_reset_all(kvm);
1082 mutex_unlock(&kvm->lock);
1083 return 0;
1084}
1085
1086static void kvm_s390_vcpu_pci_setup(struct kvm_vcpu *vcpu)
1087{
1088 /* Only set the ECB bits after guest requests zPCI interpretation */
1089 if (!vcpu->kvm->arch.use_zpci_interp)
1090 return;
1091
1092 vcpu->arch.sie_block->ecb2 |= ECB2_ZPCI_LSI;
1093 vcpu->arch.sie_block->ecb3 |= ECB3_AISII + ECB3_AISI;
1094}
1095
1096void kvm_s390_vcpu_pci_enable_interp(struct kvm *kvm)
1097{
1098 struct kvm_vcpu *vcpu;
1099 unsigned long i;
1100
1101 lockdep_assert_held(&kvm->lock);
1102
1103 if (!kvm_s390_pci_interp_allowed())
1104 return;
1105
1106 /*
1107 * If host is configured for PCI and the necessary facilities are
1108 * available, turn on interpretation for the life of this guest
1109 */
1110 kvm->arch.use_zpci_interp = 1;
1111
1112 kvm_s390_vcpu_block_all(kvm);
1113
1114 kvm_for_each_vcpu(i, vcpu, kvm) {
1115 kvm_s390_vcpu_pci_setup(vcpu);
1116 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu);
1117 }
1118
1119 kvm_s390_vcpu_unblock_all(kvm);
1120}
1121
1122static void kvm_s390_sync_request_broadcast(struct kvm *kvm, int req)
1123{
1124 unsigned long cx;
1125 struct kvm_vcpu *vcpu;
1126
1127 kvm_for_each_vcpu(cx, vcpu, kvm)
1128 kvm_s390_sync_request(req, vcpu);
1129}
1130
1131/*
1132 * Must be called with kvm->srcu held to avoid races on memslots, and with
1133 * kvm->slots_lock to avoid races with ourselves and kvm_s390_vm_stop_migration.
1134 */
1135static int kvm_s390_vm_start_migration(struct kvm *kvm)
1136{
1137 struct kvm_memory_slot *ms;
1138 struct kvm_memslots *slots;
1139 unsigned long ram_pages = 0;
1140 int bkt;
1141
1142 /* migration mode already enabled */
1143 if (kvm->arch.migration_mode)
1144 return 0;
1145 slots = kvm_memslots(kvm);
1146 if (!slots || kvm_memslots_empty(slots))
1147 return -EINVAL;
1148
1149 if (!kvm->arch.use_cmma) {
1150 kvm->arch.migration_mode = 1;
1151 return 0;
1152 }
1153 /* mark all the pages in active slots as dirty */
1154 kvm_for_each_memslot(ms, bkt, slots) {
1155 if (!ms->dirty_bitmap)
1156 return -EINVAL;
1157 /*
1158 * The second half of the bitmap is only used on x86,
1159 * and would be wasted otherwise, so we put it to good
1160 * use here to keep track of the state of the storage
1161 * attributes.
1162 */
1163 memset(kvm_second_dirty_bitmap(ms), 0xff, kvm_dirty_bitmap_bytes(ms));
1164 ram_pages += ms->npages;
1165 }
1166 atomic64_set(&kvm->arch.cmma_dirty_pages, ram_pages);
1167 kvm->arch.migration_mode = 1;
1168 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_START_MIGRATION);
1169 return 0;
1170}
1171
1172/*
1173 * Must be called with kvm->slots_lock to avoid races with ourselves and
1174 * kvm_s390_vm_start_migration.
1175 */
1176static int kvm_s390_vm_stop_migration(struct kvm *kvm)
1177{
1178 /* migration mode already disabled */
1179 if (!kvm->arch.migration_mode)
1180 return 0;
1181 kvm->arch.migration_mode = 0;
1182 if (kvm->arch.use_cmma)
1183 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_STOP_MIGRATION);
1184 return 0;
1185}
1186
1187static int kvm_s390_vm_set_migration(struct kvm *kvm,
1188 struct kvm_device_attr *attr)
1189{
1190 int res = -ENXIO;
1191
1192 mutex_lock(&kvm->slots_lock);
1193 switch (attr->attr) {
1194 case KVM_S390_VM_MIGRATION_START:
1195 res = kvm_s390_vm_start_migration(kvm);
1196 break;
1197 case KVM_S390_VM_MIGRATION_STOP:
1198 res = kvm_s390_vm_stop_migration(kvm);
1199 break;
1200 default:
1201 break;
1202 }
1203 mutex_unlock(&kvm->slots_lock);
1204
1205 return res;
1206}
1207
1208static int kvm_s390_vm_get_migration(struct kvm *kvm,
1209 struct kvm_device_attr *attr)
1210{
1211 u64 mig = kvm->arch.migration_mode;
1212
1213 if (attr->attr != KVM_S390_VM_MIGRATION_STATUS)
1214 return -ENXIO;
1215
1216 if (copy_to_user((void __user *)attr->addr, &mig, sizeof(mig)))
1217 return -EFAULT;
1218 return 0;
1219}
1220
1221static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod);
1222
1223static int kvm_s390_set_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1224{
1225 struct kvm_s390_vm_tod_clock gtod;
1226
1227 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod)))
1228 return -EFAULT;
1229
1230 if (!test_kvm_facility(kvm, 139) && gtod.epoch_idx)
1231 return -EINVAL;
1232 __kvm_s390_set_tod_clock(kvm, >od);
1233
1234 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x, TOD base: 0x%llx",
1235 gtod.epoch_idx, gtod.tod);
1236
1237 return 0;
1238}
1239
1240static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1241{
1242 u8 gtod_high;
1243
1244 if (copy_from_user(>od_high, (void __user *)attr->addr,
1245 sizeof(gtod_high)))
1246 return -EFAULT;
1247
1248 if (gtod_high != 0)
1249 return -EINVAL;
1250 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
1251
1252 return 0;
1253}
1254
1255static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1256{
1257 struct kvm_s390_vm_tod_clock gtod = { 0 };
1258
1259 if (copy_from_user(>od.tod, (void __user *)attr->addr,
1260 sizeof(gtod.tod)))
1261 return -EFAULT;
1262
1263 __kvm_s390_set_tod_clock(kvm, >od);
1264 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod.tod);
1265 return 0;
1266}
1267
1268static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1269{
1270 int ret;
1271
1272 if (attr->flags)
1273 return -EINVAL;
1274
1275 mutex_lock(&kvm->lock);
1276 /*
1277 * For protected guests, the TOD is managed by the ultravisor, so trying
1278 * to change it will never bring the expected results.
1279 */
1280 if (kvm_s390_pv_is_protected(kvm)) {
1281 ret = -EOPNOTSUPP;
1282 goto out_unlock;
1283 }
1284
1285 switch (attr->attr) {
1286 case KVM_S390_VM_TOD_EXT:
1287 ret = kvm_s390_set_tod_ext(kvm, attr);
1288 break;
1289 case KVM_S390_VM_TOD_HIGH:
1290 ret = kvm_s390_set_tod_high(kvm, attr);
1291 break;
1292 case KVM_S390_VM_TOD_LOW:
1293 ret = kvm_s390_set_tod_low(kvm, attr);
1294 break;
1295 default:
1296 ret = -ENXIO;
1297 break;
1298 }
1299
1300out_unlock:
1301 mutex_unlock(&kvm->lock);
1302 return ret;
1303}
1304
1305static void kvm_s390_get_tod_clock(struct kvm *kvm,
1306 struct kvm_s390_vm_tod_clock *gtod)
1307{
1308 union tod_clock clk;
1309
1310 preempt_disable();
1311
1312 store_tod_clock_ext(&clk);
1313
1314 gtod->tod = clk.tod + kvm->arch.epoch;
1315 gtod->epoch_idx = 0;
1316 if (test_kvm_facility(kvm, 139)) {
1317 gtod->epoch_idx = clk.ei + kvm->arch.epdx;
1318 if (gtod->tod < clk.tod)
1319 gtod->epoch_idx += 1;
1320 }
1321
1322 preempt_enable();
1323}
1324
1325static int kvm_s390_get_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1326{
1327 struct kvm_s390_vm_tod_clock gtod;
1328
1329 memset(>od, 0, sizeof(gtod));
1330 kvm_s390_get_tod_clock(kvm, >od);
1331 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
1332 return -EFAULT;
1333
1334 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x, TOD base: 0x%llx",
1335 gtod.epoch_idx, gtod.tod);
1336 return 0;
1337}
1338
1339static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1340{
1341 u8 gtod_high = 0;
1342
1343 if (copy_to_user((void __user *)attr->addr, >od_high,
1344 sizeof(gtod_high)))
1345 return -EFAULT;
1346 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
1347
1348 return 0;
1349}
1350
1351static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1352{
1353 u64 gtod;
1354
1355 gtod = kvm_s390_get_tod_clock_fast(kvm);
1356 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
1357 return -EFAULT;
1358 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
1359
1360 return 0;
1361}
1362
1363static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1364{
1365 int ret;
1366
1367 if (attr->flags)
1368 return -EINVAL;
1369
1370 switch (attr->attr) {
1371 case KVM_S390_VM_TOD_EXT:
1372 ret = kvm_s390_get_tod_ext(kvm, attr);
1373 break;
1374 case KVM_S390_VM_TOD_HIGH:
1375 ret = kvm_s390_get_tod_high(kvm, attr);
1376 break;
1377 case KVM_S390_VM_TOD_LOW:
1378 ret = kvm_s390_get_tod_low(kvm, attr);
1379 break;
1380 default:
1381 ret = -ENXIO;
1382 break;
1383 }
1384 return ret;
1385}
1386
1387static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1388{
1389 struct kvm_s390_vm_cpu_processor *proc;
1390 u16 lowest_ibc, unblocked_ibc;
1391 int ret = 0;
1392
1393 mutex_lock(&kvm->lock);
1394 if (kvm->created_vcpus) {
1395 ret = -EBUSY;
1396 goto out;
1397 }
1398 proc = kzalloc(sizeof(*proc), GFP_KERNEL_ACCOUNT);
1399 if (!proc) {
1400 ret = -ENOMEM;
1401 goto out;
1402 }
1403 if (!copy_from_user(proc, (void __user *)attr->addr,
1404 sizeof(*proc))) {
1405 kvm->arch.model.cpuid = proc->cpuid;
1406 lowest_ibc = sclp.ibc >> 16 & 0xfff;
1407 unblocked_ibc = sclp.ibc & 0xfff;
1408 if (lowest_ibc && proc->ibc) {
1409 if (proc->ibc > unblocked_ibc)
1410 kvm->arch.model.ibc = unblocked_ibc;
1411 else if (proc->ibc < lowest_ibc)
1412 kvm->arch.model.ibc = lowest_ibc;
1413 else
1414 kvm->arch.model.ibc = proc->ibc;
1415 }
1416 memcpy(kvm->arch.model.fac_list, proc->fac_list,
1417 S390_ARCH_FAC_LIST_SIZE_BYTE);
1418 VM_EVENT(kvm, 3, "SET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1419 kvm->arch.model.ibc,
1420 kvm->arch.model.cpuid);
1421 VM_EVENT(kvm, 3, "SET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1422 kvm->arch.model.fac_list[0],
1423 kvm->arch.model.fac_list[1],
1424 kvm->arch.model.fac_list[2]);
1425 } else
1426 ret = -EFAULT;
1427 kfree(proc);
1428out:
1429 mutex_unlock(&kvm->lock);
1430 return ret;
1431}
1432
1433static int kvm_s390_set_processor_feat(struct kvm *kvm,
1434 struct kvm_device_attr *attr)
1435{
1436 struct kvm_s390_vm_cpu_feat data;
1437
1438 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
1439 return -EFAULT;
1440 if (!bitmap_subset((unsigned long *) data.feat,
1441 kvm_s390_available_cpu_feat,
1442 KVM_S390_VM_CPU_FEAT_NR_BITS))
1443 return -EINVAL;
1444
1445 mutex_lock(&kvm->lock);
1446 if (kvm->created_vcpus) {
1447 mutex_unlock(&kvm->lock);
1448 return -EBUSY;
1449 }
1450 bitmap_from_arr64(kvm->arch.cpu_feat, data.feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1451 mutex_unlock(&kvm->lock);
1452 VM_EVENT(kvm, 3, "SET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1453 data.feat[0],
1454 data.feat[1],
1455 data.feat[2]);
1456 return 0;
1457}
1458
1459static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
1460 struct kvm_device_attr *attr)
1461{
1462 mutex_lock(&kvm->lock);
1463 if (kvm->created_vcpus) {
1464 mutex_unlock(&kvm->lock);
1465 return -EBUSY;
1466 }
1467
1468 if (copy_from_user(&kvm->arch.model.subfuncs, (void __user *)attr->addr,
1469 sizeof(struct kvm_s390_vm_cpu_subfunc))) {
1470 mutex_unlock(&kvm->lock);
1471 return -EFAULT;
1472 }
1473 mutex_unlock(&kvm->lock);
1474
1475 VM_EVENT(kvm, 3, "SET: guest PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1476 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0],
1477 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1],
1478 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2],
1479 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]);
1480 VM_EVENT(kvm, 3, "SET: guest PTFF subfunc 0x%16.16lx.%16.16lx",
1481 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0],
1482 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]);
1483 VM_EVENT(kvm, 3, "SET: guest KMAC subfunc 0x%16.16lx.%16.16lx",
1484 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0],
1485 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]);
1486 VM_EVENT(kvm, 3, "SET: guest KMC subfunc 0x%16.16lx.%16.16lx",
1487 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0],
1488 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]);
1489 VM_EVENT(kvm, 3, "SET: guest KM subfunc 0x%16.16lx.%16.16lx",
1490 ((unsigned long *) &kvm->arch.model.subfuncs.km)[0],
1491 ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]);
1492 VM_EVENT(kvm, 3, "SET: guest KIMD subfunc 0x%16.16lx.%16.16lx",
1493 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0],
1494 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]);
1495 VM_EVENT(kvm, 3, "SET: guest KLMD subfunc 0x%16.16lx.%16.16lx",
1496 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0],
1497 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]);
1498 VM_EVENT(kvm, 3, "SET: guest PCKMO subfunc 0x%16.16lx.%16.16lx",
1499 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0],
1500 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]);
1501 VM_EVENT(kvm, 3, "SET: guest KMCTR subfunc 0x%16.16lx.%16.16lx",
1502 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0],
1503 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]);
1504 VM_EVENT(kvm, 3, "SET: guest KMF subfunc 0x%16.16lx.%16.16lx",
1505 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0],
1506 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]);
1507 VM_EVENT(kvm, 3, "SET: guest KMO subfunc 0x%16.16lx.%16.16lx",
1508 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0],
1509 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]);
1510 VM_EVENT(kvm, 3, "SET: guest PCC subfunc 0x%16.16lx.%16.16lx",
1511 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0],
1512 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]);
1513 VM_EVENT(kvm, 3, "SET: guest PPNO subfunc 0x%16.16lx.%16.16lx",
1514 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0],
1515 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]);
1516 VM_EVENT(kvm, 3, "SET: guest KMA subfunc 0x%16.16lx.%16.16lx",
1517 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0],
1518 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]);
1519 VM_EVENT(kvm, 3, "SET: guest KDSA subfunc 0x%16.16lx.%16.16lx",
1520 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0],
1521 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]);
1522 VM_EVENT(kvm, 3, "SET: guest SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1523 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0],
1524 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1],
1525 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2],
1526 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]);
1527 VM_EVENT(kvm, 3, "SET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1528 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0],
1529 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1],
1530 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2],
1531 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]);
1532
1533 return 0;
1534}
1535
1536static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1537{
1538 int ret = -ENXIO;
1539
1540 switch (attr->attr) {
1541 case KVM_S390_VM_CPU_PROCESSOR:
1542 ret = kvm_s390_set_processor(kvm, attr);
1543 break;
1544 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1545 ret = kvm_s390_set_processor_feat(kvm, attr);
1546 break;
1547 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1548 ret = kvm_s390_set_processor_subfunc(kvm, attr);
1549 break;
1550 }
1551 return ret;
1552}
1553
1554static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1555{
1556 struct kvm_s390_vm_cpu_processor *proc;
1557 int ret = 0;
1558
1559 proc = kzalloc(sizeof(*proc), GFP_KERNEL_ACCOUNT);
1560 if (!proc) {
1561 ret = -ENOMEM;
1562 goto out;
1563 }
1564 proc->cpuid = kvm->arch.model.cpuid;
1565 proc->ibc = kvm->arch.model.ibc;
1566 memcpy(&proc->fac_list, kvm->arch.model.fac_list,
1567 S390_ARCH_FAC_LIST_SIZE_BYTE);
1568 VM_EVENT(kvm, 3, "GET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1569 kvm->arch.model.ibc,
1570 kvm->arch.model.cpuid);
1571 VM_EVENT(kvm, 3, "GET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1572 kvm->arch.model.fac_list[0],
1573 kvm->arch.model.fac_list[1],
1574 kvm->arch.model.fac_list[2]);
1575 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
1576 ret = -EFAULT;
1577 kfree(proc);
1578out:
1579 return ret;
1580}
1581
1582static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
1583{
1584 struct kvm_s390_vm_cpu_machine *mach;
1585 int ret = 0;
1586
1587 mach = kzalloc(sizeof(*mach), GFP_KERNEL_ACCOUNT);
1588 if (!mach) {
1589 ret = -ENOMEM;
1590 goto out;
1591 }
1592 get_cpu_id((struct cpuid *) &mach->cpuid);
1593 mach->ibc = sclp.ibc;
1594 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
1595 S390_ARCH_FAC_LIST_SIZE_BYTE);
1596 memcpy((unsigned long *)&mach->fac_list, stfle_fac_list,
1597 sizeof(stfle_fac_list));
1598 VM_EVENT(kvm, 3, "GET: host ibc: 0x%4.4x, host cpuid: 0x%16.16llx",
1599 kvm->arch.model.ibc,
1600 kvm->arch.model.cpuid);
1601 VM_EVENT(kvm, 3, "GET: host facmask: 0x%16.16llx.%16.16llx.%16.16llx",
1602 mach->fac_mask[0],
1603 mach->fac_mask[1],
1604 mach->fac_mask[2]);
1605 VM_EVENT(kvm, 3, "GET: host faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1606 mach->fac_list[0],
1607 mach->fac_list[1],
1608 mach->fac_list[2]);
1609 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
1610 ret = -EFAULT;
1611 kfree(mach);
1612out:
1613 return ret;
1614}
1615
1616static int kvm_s390_get_processor_feat(struct kvm *kvm,
1617 struct kvm_device_attr *attr)
1618{
1619 struct kvm_s390_vm_cpu_feat data;
1620
1621 bitmap_to_arr64(data.feat, kvm->arch.cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1622 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1623 return -EFAULT;
1624 VM_EVENT(kvm, 3, "GET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1625 data.feat[0],
1626 data.feat[1],
1627 data.feat[2]);
1628 return 0;
1629}
1630
1631static int kvm_s390_get_machine_feat(struct kvm *kvm,
1632 struct kvm_device_attr *attr)
1633{
1634 struct kvm_s390_vm_cpu_feat data;
1635
1636 bitmap_to_arr64(data.feat, kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1637 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1638 return -EFAULT;
1639 VM_EVENT(kvm, 3, "GET: host feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1640 data.feat[0],
1641 data.feat[1],
1642 data.feat[2]);
1643 return 0;
1644}
1645
1646static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
1647 struct kvm_device_attr *attr)
1648{
1649 if (copy_to_user((void __user *)attr->addr, &kvm->arch.model.subfuncs,
1650 sizeof(struct kvm_s390_vm_cpu_subfunc)))
1651 return -EFAULT;
1652
1653 VM_EVENT(kvm, 3, "GET: guest PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1654 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0],
1655 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1],
1656 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2],
1657 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]);
1658 VM_EVENT(kvm, 3, "GET: guest PTFF subfunc 0x%16.16lx.%16.16lx",
1659 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0],
1660 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]);
1661 VM_EVENT(kvm, 3, "GET: guest KMAC subfunc 0x%16.16lx.%16.16lx",
1662 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0],
1663 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]);
1664 VM_EVENT(kvm, 3, "GET: guest KMC subfunc 0x%16.16lx.%16.16lx",
1665 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0],
1666 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]);
1667 VM_EVENT(kvm, 3, "GET: guest KM subfunc 0x%16.16lx.%16.16lx",
1668 ((unsigned long *) &kvm->arch.model.subfuncs.km)[0],
1669 ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]);
1670 VM_EVENT(kvm, 3, "GET: guest KIMD subfunc 0x%16.16lx.%16.16lx",
1671 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0],
1672 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]);
1673 VM_EVENT(kvm, 3, "GET: guest KLMD subfunc 0x%16.16lx.%16.16lx",
1674 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0],
1675 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]);
1676 VM_EVENT(kvm, 3, "GET: guest PCKMO subfunc 0x%16.16lx.%16.16lx",
1677 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0],
1678 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]);
1679 VM_EVENT(kvm, 3, "GET: guest KMCTR subfunc 0x%16.16lx.%16.16lx",
1680 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0],
1681 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]);
1682 VM_EVENT(kvm, 3, "GET: guest KMF subfunc 0x%16.16lx.%16.16lx",
1683 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0],
1684 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]);
1685 VM_EVENT(kvm, 3, "GET: guest KMO subfunc 0x%16.16lx.%16.16lx",
1686 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0],
1687 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]);
1688 VM_EVENT(kvm, 3, "GET: guest PCC subfunc 0x%16.16lx.%16.16lx",
1689 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0],
1690 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]);
1691 VM_EVENT(kvm, 3, "GET: guest PPNO subfunc 0x%16.16lx.%16.16lx",
1692 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0],
1693 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]);
1694 VM_EVENT(kvm, 3, "GET: guest KMA subfunc 0x%16.16lx.%16.16lx",
1695 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0],
1696 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]);
1697 VM_EVENT(kvm, 3, "GET: guest KDSA subfunc 0x%16.16lx.%16.16lx",
1698 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0],
1699 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]);
1700 VM_EVENT(kvm, 3, "GET: guest SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1701 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0],
1702 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1],
1703 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2],
1704 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]);
1705 VM_EVENT(kvm, 3, "GET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1706 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0],
1707 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1],
1708 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2],
1709 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]);
1710
1711 return 0;
1712}
1713
1714static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
1715 struct kvm_device_attr *attr)
1716{
1717 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
1718 sizeof(struct kvm_s390_vm_cpu_subfunc)))
1719 return -EFAULT;
1720
1721 VM_EVENT(kvm, 3, "GET: host PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1722 ((unsigned long *) &kvm_s390_available_subfunc.plo)[0],
1723 ((unsigned long *) &kvm_s390_available_subfunc.plo)[1],
1724 ((unsigned long *) &kvm_s390_available_subfunc.plo)[2],
1725 ((unsigned long *) &kvm_s390_available_subfunc.plo)[3]);
1726 VM_EVENT(kvm, 3, "GET: host PTFF subfunc 0x%16.16lx.%16.16lx",
1727 ((unsigned long *) &kvm_s390_available_subfunc.ptff)[0],
1728 ((unsigned long *) &kvm_s390_available_subfunc.ptff)[1]);
1729 VM_EVENT(kvm, 3, "GET: host KMAC subfunc 0x%16.16lx.%16.16lx",
1730 ((unsigned long *) &kvm_s390_available_subfunc.kmac)[0],
1731 ((unsigned long *) &kvm_s390_available_subfunc.kmac)[1]);
1732 VM_EVENT(kvm, 3, "GET: host KMC subfunc 0x%16.16lx.%16.16lx",
1733 ((unsigned long *) &kvm_s390_available_subfunc.kmc)[0],
1734 ((unsigned long *) &kvm_s390_available_subfunc.kmc)[1]);
1735 VM_EVENT(kvm, 3, "GET: host KM subfunc 0x%16.16lx.%16.16lx",
1736 ((unsigned long *) &kvm_s390_available_subfunc.km)[0],
1737 ((unsigned long *) &kvm_s390_available_subfunc.km)[1]);
1738 VM_EVENT(kvm, 3, "GET: host KIMD subfunc 0x%16.16lx.%16.16lx",
1739 ((unsigned long *) &kvm_s390_available_subfunc.kimd)[0],
1740 ((unsigned long *) &kvm_s390_available_subfunc.kimd)[1]);
1741 VM_EVENT(kvm, 3, "GET: host KLMD subfunc 0x%16.16lx.%16.16lx",
1742 ((unsigned long *) &kvm_s390_available_subfunc.klmd)[0],
1743 ((unsigned long *) &kvm_s390_available_subfunc.klmd)[1]);
1744 VM_EVENT(kvm, 3, "GET: host PCKMO subfunc 0x%16.16lx.%16.16lx",
1745 ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[0],
1746 ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[1]);
1747 VM_EVENT(kvm, 3, "GET: host KMCTR subfunc 0x%16.16lx.%16.16lx",
1748 ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[0],
1749 ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[1]);
1750 VM_EVENT(kvm, 3, "GET: host KMF subfunc 0x%16.16lx.%16.16lx",
1751 ((unsigned long *) &kvm_s390_available_subfunc.kmf)[0],
1752 ((unsigned long *) &kvm_s390_available_subfunc.kmf)[1]);
1753 VM_EVENT(kvm, 3, "GET: host KMO subfunc 0x%16.16lx.%16.16lx",
1754 ((unsigned long *) &kvm_s390_available_subfunc.kmo)[0],
1755 ((unsigned long *) &kvm_s390_available_subfunc.kmo)[1]);
1756 VM_EVENT(kvm, 3, "GET: host PCC subfunc 0x%16.16lx.%16.16lx",
1757 ((unsigned long *) &kvm_s390_available_subfunc.pcc)[0],
1758 ((unsigned long *) &kvm_s390_available_subfunc.pcc)[1]);
1759 VM_EVENT(kvm, 3, "GET: host PPNO subfunc 0x%16.16lx.%16.16lx",
1760 ((unsigned long *) &kvm_s390_available_subfunc.ppno)[0],
1761 ((unsigned long *) &kvm_s390_available_subfunc.ppno)[1]);
1762 VM_EVENT(kvm, 3, "GET: host KMA subfunc 0x%16.16lx.%16.16lx",
1763 ((unsigned long *) &kvm_s390_available_subfunc.kma)[0],
1764 ((unsigned long *) &kvm_s390_available_subfunc.kma)[1]);
1765 VM_EVENT(kvm, 3, "GET: host KDSA subfunc 0x%16.16lx.%16.16lx",
1766 ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[0],
1767 ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[1]);
1768 VM_EVENT(kvm, 3, "GET: host SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1769 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[0],
1770 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[1],
1771 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[2],
1772 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[3]);
1773 VM_EVENT(kvm, 3, "GET: host DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1774 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[0],
1775 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[1],
1776 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[2],
1777 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[3]);
1778
1779 return 0;
1780}
1781
1782static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1783{
1784 int ret = -ENXIO;
1785
1786 switch (attr->attr) {
1787 case KVM_S390_VM_CPU_PROCESSOR:
1788 ret = kvm_s390_get_processor(kvm, attr);
1789 break;
1790 case KVM_S390_VM_CPU_MACHINE:
1791 ret = kvm_s390_get_machine(kvm, attr);
1792 break;
1793 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1794 ret = kvm_s390_get_processor_feat(kvm, attr);
1795 break;
1796 case KVM_S390_VM_CPU_MACHINE_FEAT:
1797 ret = kvm_s390_get_machine_feat(kvm, attr);
1798 break;
1799 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1800 ret = kvm_s390_get_processor_subfunc(kvm, attr);
1801 break;
1802 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1803 ret = kvm_s390_get_machine_subfunc(kvm, attr);
1804 break;
1805 }
1806 return ret;
1807}
1808
1809/**
1810 * kvm_s390_update_topology_change_report - update CPU topology change report
1811 * @kvm: guest KVM description
1812 * @val: set or clear the MTCR bit
1813 *
1814 * Updates the Multiprocessor Topology-Change-Report bit to signal
1815 * the guest with a topology change.
1816 * This is only relevant if the topology facility is present.
1817 *
1818 * The SCA version, bsca or esca, doesn't matter as offset is the same.
1819 */
1820static void kvm_s390_update_topology_change_report(struct kvm *kvm, bool val)
1821{
1822 union sca_utility new, old;
1823 struct bsca_block *sca;
1824
1825 read_lock(&kvm->arch.sca_lock);
1826 sca = kvm->arch.sca;
1827 do {
1828 old = READ_ONCE(sca->utility);
1829 new = old;
1830 new.mtcr = val;
1831 } while (cmpxchg(&sca->utility.val, old.val, new.val) != old.val);
1832 read_unlock(&kvm->arch.sca_lock);
1833}
1834
1835static int kvm_s390_set_topo_change_indication(struct kvm *kvm,
1836 struct kvm_device_attr *attr)
1837{
1838 if (!test_kvm_facility(kvm, 11))
1839 return -ENXIO;
1840
1841 kvm_s390_update_topology_change_report(kvm, !!attr->attr);
1842 return 0;
1843}
1844
1845static int kvm_s390_get_topo_change_indication(struct kvm *kvm,
1846 struct kvm_device_attr *attr)
1847{
1848 u8 topo;
1849
1850 if (!test_kvm_facility(kvm, 11))
1851 return -ENXIO;
1852
1853 read_lock(&kvm->arch.sca_lock);
1854 topo = ((struct bsca_block *)kvm->arch.sca)->utility.mtcr;
1855 read_unlock(&kvm->arch.sca_lock);
1856
1857 return put_user(topo, (u8 __user *)attr->addr);
1858}
1859
1860static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1861{
1862 int ret;
1863
1864 switch (attr->group) {
1865 case KVM_S390_VM_MEM_CTRL:
1866 ret = kvm_s390_set_mem_control(kvm, attr);
1867 break;
1868 case KVM_S390_VM_TOD:
1869 ret = kvm_s390_set_tod(kvm, attr);
1870 break;
1871 case KVM_S390_VM_CPU_MODEL:
1872 ret = kvm_s390_set_cpu_model(kvm, attr);
1873 break;
1874 case KVM_S390_VM_CRYPTO:
1875 ret = kvm_s390_vm_set_crypto(kvm, attr);
1876 break;
1877 case KVM_S390_VM_MIGRATION:
1878 ret = kvm_s390_vm_set_migration(kvm, attr);
1879 break;
1880 case KVM_S390_VM_CPU_TOPOLOGY:
1881 ret = kvm_s390_set_topo_change_indication(kvm, attr);
1882 break;
1883 default:
1884 ret = -ENXIO;
1885 break;
1886 }
1887
1888 return ret;
1889}
1890
1891static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1892{
1893 int ret;
1894
1895 switch (attr->group) {
1896 case KVM_S390_VM_MEM_CTRL:
1897 ret = kvm_s390_get_mem_control(kvm, attr);
1898 break;
1899 case KVM_S390_VM_TOD:
1900 ret = kvm_s390_get_tod(kvm, attr);
1901 break;
1902 case KVM_S390_VM_CPU_MODEL:
1903 ret = kvm_s390_get_cpu_model(kvm, attr);
1904 break;
1905 case KVM_S390_VM_MIGRATION:
1906 ret = kvm_s390_vm_get_migration(kvm, attr);
1907 break;
1908 case KVM_S390_VM_CPU_TOPOLOGY:
1909 ret = kvm_s390_get_topo_change_indication(kvm, attr);
1910 break;
1911 default:
1912 ret = -ENXIO;
1913 break;
1914 }
1915
1916 return ret;
1917}
1918
1919static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1920{
1921 int ret;
1922
1923 switch (attr->group) {
1924 case KVM_S390_VM_MEM_CTRL:
1925 switch (attr->attr) {
1926 case KVM_S390_VM_MEM_ENABLE_CMMA:
1927 case KVM_S390_VM_MEM_CLR_CMMA:
1928 ret = sclp.has_cmma ? 0 : -ENXIO;
1929 break;
1930 case KVM_S390_VM_MEM_LIMIT_SIZE:
1931 ret = 0;
1932 break;
1933 default:
1934 ret = -ENXIO;
1935 break;
1936 }
1937 break;
1938 case KVM_S390_VM_TOD:
1939 switch (attr->attr) {
1940 case KVM_S390_VM_TOD_LOW:
1941 case KVM_S390_VM_TOD_HIGH:
1942 ret = 0;
1943 break;
1944 default:
1945 ret = -ENXIO;
1946 break;
1947 }
1948 break;
1949 case KVM_S390_VM_CPU_MODEL:
1950 switch (attr->attr) {
1951 case KVM_S390_VM_CPU_PROCESSOR:
1952 case KVM_S390_VM_CPU_MACHINE:
1953 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1954 case KVM_S390_VM_CPU_MACHINE_FEAT:
1955 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1956 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1957 ret = 0;
1958 break;
1959 default:
1960 ret = -ENXIO;
1961 break;
1962 }
1963 break;
1964 case KVM_S390_VM_CRYPTO:
1965 switch (attr->attr) {
1966 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1967 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1968 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1969 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1970 ret = 0;
1971 break;
1972 case KVM_S390_VM_CRYPTO_ENABLE_APIE:
1973 case KVM_S390_VM_CRYPTO_DISABLE_APIE:
1974 ret = ap_instructions_available() ? 0 : -ENXIO;
1975 break;
1976 default:
1977 ret = -ENXIO;
1978 break;
1979 }
1980 break;
1981 case KVM_S390_VM_MIGRATION:
1982 ret = 0;
1983 break;
1984 case KVM_S390_VM_CPU_TOPOLOGY:
1985 ret = test_kvm_facility(kvm, 11) ? 0 : -ENXIO;
1986 break;
1987 default:
1988 ret = -ENXIO;
1989 break;
1990 }
1991
1992 return ret;
1993}
1994
1995static long kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1996{
1997 uint8_t *keys;
1998 uint64_t hva;
1999 int srcu_idx, i, r = 0;
2000
2001 if (args->flags != 0)
2002 return -EINVAL;
2003
2004 /* Is this guest using storage keys? */
2005 if (!mm_uses_skeys(current->mm))
2006 return KVM_S390_GET_SKEYS_NONE;
2007
2008 /* Enforce sane limit on memory allocation */
2009 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
2010 return -EINVAL;
2011
2012 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL_ACCOUNT);
2013 if (!keys)
2014 return -ENOMEM;
2015
2016 mmap_read_lock(current->mm);
2017 srcu_idx = srcu_read_lock(&kvm->srcu);
2018 for (i = 0; i < args->count; i++) {
2019 hva = gfn_to_hva(kvm, args->start_gfn + i);
2020 if (kvm_is_error_hva(hva)) {
2021 r = -EFAULT;
2022 break;
2023 }
2024
2025 r = get_guest_storage_key(current->mm, hva, &keys[i]);
2026 if (r)
2027 break;
2028 }
2029 srcu_read_unlock(&kvm->srcu, srcu_idx);
2030 mmap_read_unlock(current->mm);
2031
2032 if (!r) {
2033 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
2034 sizeof(uint8_t) * args->count);
2035 if (r)
2036 r = -EFAULT;
2037 }
2038
2039 kvfree(keys);
2040 return r;
2041}
2042
2043static long kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
2044{
2045 uint8_t *keys;
2046 uint64_t hva;
2047 int srcu_idx, i, r = 0;
2048 bool unlocked;
2049
2050 if (args->flags != 0)
2051 return -EINVAL;
2052
2053 /* Enforce sane limit on memory allocation */
2054 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
2055 return -EINVAL;
2056
2057 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL_ACCOUNT);
2058 if (!keys)
2059 return -ENOMEM;
2060
2061 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
2062 sizeof(uint8_t) * args->count);
2063 if (r) {
2064 r = -EFAULT;
2065 goto out;
2066 }
2067
2068 /* Enable storage key handling for the guest */
2069 r = s390_enable_skey();
2070 if (r)
2071 goto out;
2072
2073 i = 0;
2074 mmap_read_lock(current->mm);
2075 srcu_idx = srcu_read_lock(&kvm->srcu);
2076 while (i < args->count) {
2077 unlocked = false;
2078 hva = gfn_to_hva(kvm, args->start_gfn + i);
2079 if (kvm_is_error_hva(hva)) {
2080 r = -EFAULT;
2081 break;
2082 }
2083
2084 /* Lowest order bit is reserved */
2085 if (keys[i] & 0x01) {
2086 r = -EINVAL;
2087 break;
2088 }
2089
2090 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
2091 if (r) {
2092 r = fixup_user_fault(current->mm, hva,
2093 FAULT_FLAG_WRITE, &unlocked);
2094 if (r)
2095 break;
2096 }
2097 if (!r)
2098 i++;
2099 }
2100 srcu_read_unlock(&kvm->srcu, srcu_idx);
2101 mmap_read_unlock(current->mm);
2102out:
2103 kvfree(keys);
2104 return r;
2105}
2106
2107/*
2108 * Base address and length must be sent at the start of each block, therefore
2109 * it's cheaper to send some clean data, as long as it's less than the size of
2110 * two longs.
2111 */
2112#define KVM_S390_MAX_BIT_DISTANCE (2 * sizeof(void *))
2113/* for consistency */
2114#define KVM_S390_CMMA_SIZE_MAX ((u32)KVM_S390_SKEYS_MAX)
2115
2116static int kvm_s390_peek_cmma(struct kvm *kvm, struct kvm_s390_cmma_log *args,
2117 u8 *res, unsigned long bufsize)
2118{
2119 unsigned long pgstev, hva, cur_gfn = args->start_gfn;
2120
2121 args->count = 0;
2122 while (args->count < bufsize) {
2123 hva = gfn_to_hva(kvm, cur_gfn);
2124 /*
2125 * We return an error if the first value was invalid, but we
2126 * return successfully if at least one value was copied.
2127 */
2128 if (kvm_is_error_hva(hva))
2129 return args->count ? 0 : -EFAULT;
2130 if (get_pgste(kvm->mm, hva, &pgstev) < 0)
2131 pgstev = 0;
2132 res[args->count++] = (pgstev >> 24) & 0x43;
2133 cur_gfn++;
2134 }
2135
2136 return 0;
2137}
2138
2139static struct kvm_memory_slot *gfn_to_memslot_approx(struct kvm_memslots *slots,
2140 gfn_t gfn)
2141{
2142 return ____gfn_to_memslot(slots, gfn, true);
2143}
2144
2145static unsigned long kvm_s390_next_dirty_cmma(struct kvm_memslots *slots,
2146 unsigned long cur_gfn)
2147{
2148 struct kvm_memory_slot *ms = gfn_to_memslot_approx(slots, cur_gfn);
2149 unsigned long ofs = cur_gfn - ms->base_gfn;
2150 struct rb_node *mnode = &ms->gfn_node[slots->node_idx];
2151
2152 if (ms->base_gfn + ms->npages <= cur_gfn) {
2153 mnode = rb_next(mnode);
2154 /* If we are above the highest slot, wrap around */
2155 if (!mnode)
2156 mnode = rb_first(&slots->gfn_tree);
2157
2158 ms = container_of(mnode, struct kvm_memory_slot, gfn_node[slots->node_idx]);
2159 ofs = 0;
2160 }
2161 ofs = find_next_bit(kvm_second_dirty_bitmap(ms), ms->npages, ofs);
2162 while (ofs >= ms->npages && (mnode = rb_next(mnode))) {
2163 ms = container_of(mnode, struct kvm_memory_slot, gfn_node[slots->node_idx]);
2164 ofs = find_first_bit(kvm_second_dirty_bitmap(ms), ms->npages);
2165 }
2166 return ms->base_gfn + ofs;
2167}
2168
2169static int kvm_s390_get_cmma(struct kvm *kvm, struct kvm_s390_cmma_log *args,
2170 u8 *res, unsigned long bufsize)
2171{
2172 unsigned long mem_end, cur_gfn, next_gfn, hva, pgstev;
2173 struct kvm_memslots *slots = kvm_memslots(kvm);
2174 struct kvm_memory_slot *ms;
2175
2176 if (unlikely(kvm_memslots_empty(slots)))
2177 return 0;
2178
2179 cur_gfn = kvm_s390_next_dirty_cmma(slots, args->start_gfn);
2180 ms = gfn_to_memslot(kvm, cur_gfn);
2181 args->count = 0;
2182 args->start_gfn = cur_gfn;
2183 if (!ms)
2184 return 0;
2185 next_gfn = kvm_s390_next_dirty_cmma(slots, cur_gfn + 1);
2186 mem_end = kvm_s390_get_gfn_end(slots);
2187
2188 while (args->count < bufsize) {
2189 hva = gfn_to_hva(kvm, cur_gfn);
2190 if (kvm_is_error_hva(hva))
2191 return 0;
2192 /* Decrement only if we actually flipped the bit to 0 */
2193 if (test_and_clear_bit(cur_gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
2194 atomic64_dec(&kvm->arch.cmma_dirty_pages);
2195 if (get_pgste(kvm->mm, hva, &pgstev) < 0)
2196 pgstev = 0;
2197 /* Save the value */
2198 res[args->count++] = (pgstev >> 24) & 0x43;
2199 /* If the next bit is too far away, stop. */
2200 if (next_gfn > cur_gfn + KVM_S390_MAX_BIT_DISTANCE)
2201 return 0;
2202 /* If we reached the previous "next", find the next one */
2203 if (cur_gfn == next_gfn)
2204 next_gfn = kvm_s390_next_dirty_cmma(slots, cur_gfn + 1);
2205 /* Reached the end of memory or of the buffer, stop */
2206 if ((next_gfn >= mem_end) ||
2207 (next_gfn - args->start_gfn >= bufsize))
2208 return 0;
2209 cur_gfn++;
2210 /* Reached the end of the current memslot, take the next one. */
2211 if (cur_gfn - ms->base_gfn >= ms->npages) {
2212 ms = gfn_to_memslot(kvm, cur_gfn);
2213 if (!ms)
2214 return 0;
2215 }
2216 }
2217 return 0;
2218}
2219
2220/*
2221 * This function searches for the next page with dirty CMMA attributes, and
2222 * saves the attributes in the buffer up to either the end of the buffer or
2223 * until a block of at least KVM_S390_MAX_BIT_DISTANCE clean bits is found;
2224 * no trailing clean bytes are saved.
2225 * In case no dirty bits were found, or if CMMA was not enabled or used, the
2226 * output buffer will indicate 0 as length.
2227 */
2228static int kvm_s390_get_cmma_bits(struct kvm *kvm,
2229 struct kvm_s390_cmma_log *args)
2230{
2231 unsigned long bufsize;
2232 int srcu_idx, peek, ret;
2233 u8 *values;
2234
2235 if (!kvm->arch.use_cmma)
2236 return -ENXIO;
2237 /* Invalid/unsupported flags were specified */
2238 if (args->flags & ~KVM_S390_CMMA_PEEK)
2239 return -EINVAL;
2240 /* Migration mode query, and we are not doing a migration */
2241 peek = !!(args->flags & KVM_S390_CMMA_PEEK);
2242 if (!peek && !kvm->arch.migration_mode)
2243 return -EINVAL;
2244 /* CMMA is disabled or was not used, or the buffer has length zero */
2245 bufsize = min(args->count, KVM_S390_CMMA_SIZE_MAX);
2246 if (!bufsize || !kvm->mm->context.uses_cmm) {
2247 memset(args, 0, sizeof(*args));
2248 return 0;
2249 }
2250 /* We are not peeking, and there are no dirty pages */
2251 if (!peek && !atomic64_read(&kvm->arch.cmma_dirty_pages)) {
2252 memset(args, 0, sizeof(*args));
2253 return 0;
2254 }
2255
2256 values = vmalloc(bufsize);
2257 if (!values)
2258 return -ENOMEM;
2259
2260 mmap_read_lock(kvm->mm);
2261 srcu_idx = srcu_read_lock(&kvm->srcu);
2262 if (peek)
2263 ret = kvm_s390_peek_cmma(kvm, args, values, bufsize);
2264 else
2265 ret = kvm_s390_get_cmma(kvm, args, values, bufsize);
2266 srcu_read_unlock(&kvm->srcu, srcu_idx);
2267 mmap_read_unlock(kvm->mm);
2268
2269 if (kvm->arch.migration_mode)
2270 args->remaining = atomic64_read(&kvm->arch.cmma_dirty_pages);
2271 else
2272 args->remaining = 0;
2273
2274 if (copy_to_user((void __user *)args->values, values, args->count))
2275 ret = -EFAULT;
2276
2277 vfree(values);
2278 return ret;
2279}
2280
2281/*
2282 * This function sets the CMMA attributes for the given pages. If the input
2283 * buffer has zero length, no action is taken, otherwise the attributes are
2284 * set and the mm->context.uses_cmm flag is set.
2285 */
2286static int kvm_s390_set_cmma_bits(struct kvm *kvm,
2287 const struct kvm_s390_cmma_log *args)
2288{
2289 unsigned long hva, mask, pgstev, i;
2290 uint8_t *bits;
2291 int srcu_idx, r = 0;
2292
2293 mask = args->mask;
2294
2295 if (!kvm->arch.use_cmma)
2296 return -ENXIO;
2297 /* invalid/unsupported flags */
2298 if (args->flags != 0)
2299 return -EINVAL;
2300 /* Enforce sane limit on memory allocation */
2301 if (args->count > KVM_S390_CMMA_SIZE_MAX)
2302 return -EINVAL;
2303 /* Nothing to do */
2304 if (args->count == 0)
2305 return 0;
2306
2307 bits = vmalloc(array_size(sizeof(*bits), args->count));
2308 if (!bits)
2309 return -ENOMEM;
2310
2311 r = copy_from_user(bits, (void __user *)args->values, args->count);
2312 if (r) {
2313 r = -EFAULT;
2314 goto out;
2315 }
2316
2317 mmap_read_lock(kvm->mm);
2318 srcu_idx = srcu_read_lock(&kvm->srcu);
2319 for (i = 0; i < args->count; i++) {
2320 hva = gfn_to_hva(kvm, args->start_gfn + i);
2321 if (kvm_is_error_hva(hva)) {
2322 r = -EFAULT;
2323 break;
2324 }
2325
2326 pgstev = bits[i];
2327 pgstev = pgstev << 24;
2328 mask &= _PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT;
2329 set_pgste_bits(kvm->mm, hva, mask, pgstev);
2330 }
2331 srcu_read_unlock(&kvm->srcu, srcu_idx);
2332 mmap_read_unlock(kvm->mm);
2333
2334 if (!kvm->mm->context.uses_cmm) {
2335 mmap_write_lock(kvm->mm);
2336 kvm->mm->context.uses_cmm = 1;
2337 mmap_write_unlock(kvm->mm);
2338 }
2339out:
2340 vfree(bits);
2341 return r;
2342}
2343
2344/**
2345 * kvm_s390_cpus_from_pv - Convert all protected vCPUs in a protected VM to
2346 * non protected.
2347 * @kvm: the VM whose protected vCPUs are to be converted
2348 * @rc: return value for the RC field of the UVC (in case of error)
2349 * @rrc: return value for the RRC field of the UVC (in case of error)
2350 *
2351 * Does not stop in case of error, tries to convert as many
2352 * CPUs as possible. In case of error, the RC and RRC of the last error are
2353 * returned.
2354 *
2355 * Return: 0 in case of success, otherwise -EIO
2356 */
2357int kvm_s390_cpus_from_pv(struct kvm *kvm, u16 *rc, u16 *rrc)
2358{
2359 struct kvm_vcpu *vcpu;
2360 unsigned long i;
2361 u16 _rc, _rrc;
2362 int ret = 0;
2363
2364 /*
2365 * We ignore failures and try to destroy as many CPUs as possible.
2366 * At the same time we must not free the assigned resources when
2367 * this fails, as the ultravisor has still access to that memory.
2368 * So kvm_s390_pv_destroy_cpu can leave a "wanted" memory leak
2369 * behind.
2370 * We want to return the first failure rc and rrc, though.
2371 */
2372 kvm_for_each_vcpu(i, vcpu, kvm) {
2373 mutex_lock(&vcpu->mutex);
2374 if (kvm_s390_pv_destroy_cpu(vcpu, &_rc, &_rrc) && !ret) {
2375 *rc = _rc;
2376 *rrc = _rrc;
2377 ret = -EIO;
2378 }
2379 mutex_unlock(&vcpu->mutex);
2380 }
2381 /* Ensure that we re-enable gisa if the non-PV guest used it but the PV guest did not. */
2382 if (use_gisa)
2383 kvm_s390_gisa_enable(kvm);
2384 return ret;
2385}
2386
2387/**
2388 * kvm_s390_cpus_to_pv - Convert all non-protected vCPUs in a protected VM
2389 * to protected.
2390 * @kvm: the VM whose protected vCPUs are to be converted
2391 * @rc: return value for the RC field of the UVC (in case of error)
2392 * @rrc: return value for the RRC field of the UVC (in case of error)
2393 *
2394 * Tries to undo the conversion in case of error.
2395 *
2396 * Return: 0 in case of success, otherwise -EIO
2397 */
2398static int kvm_s390_cpus_to_pv(struct kvm *kvm, u16 *rc, u16 *rrc)
2399{
2400 unsigned long i;
2401 int r = 0;
2402 u16 dummy;
2403
2404 struct kvm_vcpu *vcpu;
2405
2406 /* Disable the GISA if the ultravisor does not support AIV. */
2407 if (!test_bit_inv(BIT_UV_FEAT_AIV, &uv_info.uv_feature_indications))
2408 kvm_s390_gisa_disable(kvm);
2409
2410 kvm_for_each_vcpu(i, vcpu, kvm) {
2411 mutex_lock(&vcpu->mutex);
2412 r = kvm_s390_pv_create_cpu(vcpu, rc, rrc);
2413 mutex_unlock(&vcpu->mutex);
2414 if (r)
2415 break;
2416 }
2417 if (r)
2418 kvm_s390_cpus_from_pv(kvm, &dummy, &dummy);
2419 return r;
2420}
2421
2422/*
2423 * Here we provide user space with a direct interface to query UV
2424 * related data like UV maxima and available features as well as
2425 * feature specific data.
2426 *
2427 * To facilitate future extension of the data structures we'll try to
2428 * write data up to the maximum requested length.
2429 */
2430static ssize_t kvm_s390_handle_pv_info(struct kvm_s390_pv_info *info)
2431{
2432 ssize_t len_min;
2433
2434 switch (info->header.id) {
2435 case KVM_PV_INFO_VM: {
2436 len_min = sizeof(info->header) + sizeof(info->vm);
2437
2438 if (info->header.len_max < len_min)
2439 return -EINVAL;
2440
2441 memcpy(info->vm.inst_calls_list,
2442 uv_info.inst_calls_list,
2443 sizeof(uv_info.inst_calls_list));
2444
2445 /* It's max cpuid not max cpus, so it's off by one */
2446 info->vm.max_cpus = uv_info.max_guest_cpu_id + 1;
2447 info->vm.max_guests = uv_info.max_num_sec_conf;
2448 info->vm.max_guest_addr = uv_info.max_sec_stor_addr;
2449 info->vm.feature_indication = uv_info.uv_feature_indications;
2450
2451 return len_min;
2452 }
2453 case KVM_PV_INFO_DUMP: {
2454 len_min = sizeof(info->header) + sizeof(info->dump);
2455
2456 if (info->header.len_max < len_min)
2457 return -EINVAL;
2458
2459 info->dump.dump_cpu_buffer_len = uv_info.guest_cpu_stor_len;
2460 info->dump.dump_config_mem_buffer_per_1m = uv_info.conf_dump_storage_state_len;
2461 info->dump.dump_config_finalize_len = uv_info.conf_dump_finalize_len;
2462 return len_min;
2463 }
2464 default:
2465 return -EINVAL;
2466 }
2467}
2468
2469static int kvm_s390_pv_dmp(struct kvm *kvm, struct kvm_pv_cmd *cmd,
2470 struct kvm_s390_pv_dmp dmp)
2471{
2472 int r = -EINVAL;
2473 void __user *result_buff = (void __user *)dmp.buff_addr;
2474
2475 switch (dmp.subcmd) {
2476 case KVM_PV_DUMP_INIT: {
2477 if (kvm->arch.pv.dumping)
2478 break;
2479
2480 /*
2481 * Block SIE entry as concurrent dump UVCs could lead
2482 * to validities.
2483 */
2484 kvm_s390_vcpu_block_all(kvm);
2485
2486 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2487 UVC_CMD_DUMP_INIT, &cmd->rc, &cmd->rrc);
2488 KVM_UV_EVENT(kvm, 3, "PROTVIRT DUMP INIT: rc %x rrc %x",
2489 cmd->rc, cmd->rrc);
2490 if (!r) {
2491 kvm->arch.pv.dumping = true;
2492 } else {
2493 kvm_s390_vcpu_unblock_all(kvm);
2494 r = -EINVAL;
2495 }
2496 break;
2497 }
2498 case KVM_PV_DUMP_CONFIG_STOR_STATE: {
2499 if (!kvm->arch.pv.dumping)
2500 break;
2501
2502 /*
2503 * gaddr is an output parameter since we might stop
2504 * early. As dmp will be copied back in our caller, we
2505 * don't need to do it ourselves.
2506 */
2507 r = kvm_s390_pv_dump_stor_state(kvm, result_buff, &dmp.gaddr, dmp.buff_len,
2508 &cmd->rc, &cmd->rrc);
2509 break;
2510 }
2511 case KVM_PV_DUMP_COMPLETE: {
2512 if (!kvm->arch.pv.dumping)
2513 break;
2514
2515 r = -EINVAL;
2516 if (dmp.buff_len < uv_info.conf_dump_finalize_len)
2517 break;
2518
2519 r = kvm_s390_pv_dump_complete(kvm, result_buff,
2520 &cmd->rc, &cmd->rrc);
2521 break;
2522 }
2523 default:
2524 r = -ENOTTY;
2525 break;
2526 }
2527
2528 return r;
2529}
2530
2531static int kvm_s390_handle_pv(struct kvm *kvm, struct kvm_pv_cmd *cmd)
2532{
2533 const bool need_lock = (cmd->cmd != KVM_PV_ASYNC_CLEANUP_PERFORM);
2534 void __user *argp = (void __user *)cmd->data;
2535 int r = 0;
2536 u16 dummy;
2537
2538 if (need_lock)
2539 mutex_lock(&kvm->lock);
2540
2541 switch (cmd->cmd) {
2542 case KVM_PV_ENABLE: {
2543 r = -EINVAL;
2544 if (kvm_s390_pv_is_protected(kvm))
2545 break;
2546
2547 /*
2548 * FMT 4 SIE needs esca. As we never switch back to bsca from
2549 * esca, we need no cleanup in the error cases below
2550 */
2551 r = sca_switch_to_extended(kvm);
2552 if (r)
2553 break;
2554
2555 mmap_write_lock(current->mm);
2556 r = gmap_mark_unmergeable();
2557 mmap_write_unlock(current->mm);
2558 if (r)
2559 break;
2560
2561 r = kvm_s390_pv_init_vm(kvm, &cmd->rc, &cmd->rrc);
2562 if (r)
2563 break;
2564
2565 r = kvm_s390_cpus_to_pv(kvm, &cmd->rc, &cmd->rrc);
2566 if (r)
2567 kvm_s390_pv_deinit_vm(kvm, &dummy, &dummy);
2568
2569 /* we need to block service interrupts from now on */
2570 set_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2571 break;
2572 }
2573 case KVM_PV_ASYNC_CLEANUP_PREPARE:
2574 r = -EINVAL;
2575 if (!kvm_s390_pv_is_protected(kvm) || !async_destroy)
2576 break;
2577
2578 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc);
2579 /*
2580 * If a CPU could not be destroyed, destroy VM will also fail.
2581 * There is no point in trying to destroy it. Instead return
2582 * the rc and rrc from the first CPU that failed destroying.
2583 */
2584 if (r)
2585 break;
2586 r = kvm_s390_pv_set_aside(kvm, &cmd->rc, &cmd->rrc);
2587
2588 /* no need to block service interrupts any more */
2589 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2590 break;
2591 case KVM_PV_ASYNC_CLEANUP_PERFORM:
2592 r = -EINVAL;
2593 if (!async_destroy)
2594 break;
2595 /* kvm->lock must not be held; this is asserted inside the function. */
2596 r = kvm_s390_pv_deinit_aside_vm(kvm, &cmd->rc, &cmd->rrc);
2597 break;
2598 case KVM_PV_DISABLE: {
2599 r = -EINVAL;
2600 if (!kvm_s390_pv_is_protected(kvm))
2601 break;
2602
2603 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc);
2604 /*
2605 * If a CPU could not be destroyed, destroy VM will also fail.
2606 * There is no point in trying to destroy it. Instead return
2607 * the rc and rrc from the first CPU that failed destroying.
2608 */
2609 if (r)
2610 break;
2611 r = kvm_s390_pv_deinit_cleanup_all(kvm, &cmd->rc, &cmd->rrc);
2612
2613 /* no need to block service interrupts any more */
2614 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2615 break;
2616 }
2617 case KVM_PV_SET_SEC_PARMS: {
2618 struct kvm_s390_pv_sec_parm parms = {};
2619 void *hdr;
2620
2621 r = -EINVAL;
2622 if (!kvm_s390_pv_is_protected(kvm))
2623 break;
2624
2625 r = -EFAULT;
2626 if (copy_from_user(&parms, argp, sizeof(parms)))
2627 break;
2628
2629 /* Currently restricted to 8KB */
2630 r = -EINVAL;
2631 if (parms.length > PAGE_SIZE * 2)
2632 break;
2633
2634 r = -ENOMEM;
2635 hdr = vmalloc(parms.length);
2636 if (!hdr)
2637 break;
2638
2639 r = -EFAULT;
2640 if (!copy_from_user(hdr, (void __user *)parms.origin,
2641 parms.length))
2642 r = kvm_s390_pv_set_sec_parms(kvm, hdr, parms.length,
2643 &cmd->rc, &cmd->rrc);
2644
2645 vfree(hdr);
2646 break;
2647 }
2648 case KVM_PV_UNPACK: {
2649 struct kvm_s390_pv_unp unp = {};
2650
2651 r = -EINVAL;
2652 if (!kvm_s390_pv_is_protected(kvm) || !mm_is_protected(kvm->mm))
2653 break;
2654
2655 r = -EFAULT;
2656 if (copy_from_user(&unp, argp, sizeof(unp)))
2657 break;
2658
2659 r = kvm_s390_pv_unpack(kvm, unp.addr, unp.size, unp.tweak,
2660 &cmd->rc, &cmd->rrc);
2661 break;
2662 }
2663 case KVM_PV_VERIFY: {
2664 r = -EINVAL;
2665 if (!kvm_s390_pv_is_protected(kvm))
2666 break;
2667
2668 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2669 UVC_CMD_VERIFY_IMG, &cmd->rc, &cmd->rrc);
2670 KVM_UV_EVENT(kvm, 3, "PROTVIRT VERIFY: rc %x rrc %x", cmd->rc,
2671 cmd->rrc);
2672 break;
2673 }
2674 case KVM_PV_PREP_RESET: {
2675 r = -EINVAL;
2676 if (!kvm_s390_pv_is_protected(kvm))
2677 break;
2678
2679 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2680 UVC_CMD_PREPARE_RESET, &cmd->rc, &cmd->rrc);
2681 KVM_UV_EVENT(kvm, 3, "PROTVIRT PREP RESET: rc %x rrc %x",
2682 cmd->rc, cmd->rrc);
2683 break;
2684 }
2685 case KVM_PV_UNSHARE_ALL: {
2686 r = -EINVAL;
2687 if (!kvm_s390_pv_is_protected(kvm))
2688 break;
2689
2690 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2691 UVC_CMD_SET_UNSHARE_ALL, &cmd->rc, &cmd->rrc);
2692 KVM_UV_EVENT(kvm, 3, "PROTVIRT UNSHARE: rc %x rrc %x",
2693 cmd->rc, cmd->rrc);
2694 break;
2695 }
2696 case KVM_PV_INFO: {
2697 struct kvm_s390_pv_info info = {};
2698 ssize_t data_len;
2699
2700 /*
2701 * No need to check the VM protection here.
2702 *
2703 * Maybe user space wants to query some of the data
2704 * when the VM is still unprotected. If we see the
2705 * need to fence a new data command we can still
2706 * return an error in the info handler.
2707 */
2708
2709 r = -EFAULT;
2710 if (copy_from_user(&info, argp, sizeof(info.header)))
2711 break;
2712
2713 r = -EINVAL;
2714 if (info.header.len_max < sizeof(info.header))
2715 break;
2716
2717 data_len = kvm_s390_handle_pv_info(&info);
2718 if (data_len < 0) {
2719 r = data_len;
2720 break;
2721 }
2722 /*
2723 * If a data command struct is extended (multiple
2724 * times) this can be used to determine how much of it
2725 * is valid.
2726 */
2727 info.header.len_written = data_len;
2728
2729 r = -EFAULT;
2730 if (copy_to_user(argp, &info, data_len))
2731 break;
2732
2733 r = 0;
2734 break;
2735 }
2736 case KVM_PV_DUMP: {
2737 struct kvm_s390_pv_dmp dmp;
2738
2739 r = -EINVAL;
2740 if (!kvm_s390_pv_is_protected(kvm))
2741 break;
2742
2743 r = -EFAULT;
2744 if (copy_from_user(&dmp, argp, sizeof(dmp)))
2745 break;
2746
2747 r = kvm_s390_pv_dmp(kvm, cmd, dmp);
2748 if (r)
2749 break;
2750
2751 if (copy_to_user(argp, &dmp, sizeof(dmp))) {
2752 r = -EFAULT;
2753 break;
2754 }
2755
2756 break;
2757 }
2758 default:
2759 r = -ENOTTY;
2760 }
2761 if (need_lock)
2762 mutex_unlock(&kvm->lock);
2763
2764 return r;
2765}
2766
2767static bool access_key_invalid(u8 access_key)
2768{
2769 return access_key > 0xf;
2770}
2771
2772static int kvm_s390_vm_mem_op(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2773{
2774 void __user *uaddr = (void __user *)mop->buf;
2775 u64 supported_flags;
2776 void *tmpbuf = NULL;
2777 int r, srcu_idx;
2778
2779 supported_flags = KVM_S390_MEMOP_F_SKEY_PROTECTION
2780 | KVM_S390_MEMOP_F_CHECK_ONLY;
2781 if (mop->flags & ~supported_flags || !mop->size)
2782 return -EINVAL;
2783 if (mop->size > MEM_OP_MAX_SIZE)
2784 return -E2BIG;
2785 /*
2786 * This is technically a heuristic only, if the kvm->lock is not
2787 * taken, it is not guaranteed that the vm is/remains non-protected.
2788 * This is ok from a kernel perspective, wrongdoing is detected
2789 * on the access, -EFAULT is returned and the vm may crash the
2790 * next time it accesses the memory in question.
2791 * There is no sane usecase to do switching and a memop on two
2792 * different CPUs at the same time.
2793 */
2794 if (kvm_s390_pv_get_handle(kvm))
2795 return -EINVAL;
2796 if (mop->flags & KVM_S390_MEMOP_F_SKEY_PROTECTION) {
2797 if (access_key_invalid(mop->key))
2798 return -EINVAL;
2799 } else {
2800 mop->key = 0;
2801 }
2802 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
2803 tmpbuf = vmalloc(mop->size);
2804 if (!tmpbuf)
2805 return -ENOMEM;
2806 }
2807
2808 srcu_idx = srcu_read_lock(&kvm->srcu);
2809
2810 if (kvm_is_error_gpa(kvm, mop->gaddr)) {
2811 r = PGM_ADDRESSING;
2812 goto out_unlock;
2813 }
2814
2815 switch (mop->op) {
2816 case KVM_S390_MEMOP_ABSOLUTE_READ: {
2817 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2818 r = check_gpa_range(kvm, mop->gaddr, mop->size, GACC_FETCH, mop->key);
2819 } else {
2820 r = access_guest_abs_with_key(kvm, mop->gaddr, tmpbuf,
2821 mop->size, GACC_FETCH, mop->key);
2822 if (r == 0) {
2823 if (copy_to_user(uaddr, tmpbuf, mop->size))
2824 r = -EFAULT;
2825 }
2826 }
2827 break;
2828 }
2829 case KVM_S390_MEMOP_ABSOLUTE_WRITE: {
2830 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2831 r = check_gpa_range(kvm, mop->gaddr, mop->size, GACC_STORE, mop->key);
2832 } else {
2833 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
2834 r = -EFAULT;
2835 break;
2836 }
2837 r = access_guest_abs_with_key(kvm, mop->gaddr, tmpbuf,
2838 mop->size, GACC_STORE, mop->key);
2839 }
2840 break;
2841 }
2842 default:
2843 r = -EINVAL;
2844 }
2845
2846out_unlock:
2847 srcu_read_unlock(&kvm->srcu, srcu_idx);
2848
2849 vfree(tmpbuf);
2850 return r;
2851}
2852
2853long kvm_arch_vm_ioctl(struct file *filp,
2854 unsigned int ioctl, unsigned long arg)
2855{
2856 struct kvm *kvm = filp->private_data;
2857 void __user *argp = (void __user *)arg;
2858 struct kvm_device_attr attr;
2859 int r;
2860
2861 switch (ioctl) {
2862 case KVM_S390_INTERRUPT: {
2863 struct kvm_s390_interrupt s390int;
2864
2865 r = -EFAULT;
2866 if (copy_from_user(&s390int, argp, sizeof(s390int)))
2867 break;
2868 r = kvm_s390_inject_vm(kvm, &s390int);
2869 break;
2870 }
2871 case KVM_CREATE_IRQCHIP: {
2872 struct kvm_irq_routing_entry routing;
2873
2874 r = -EINVAL;
2875 if (kvm->arch.use_irqchip) {
2876 /* Set up dummy routing. */
2877 memset(&routing, 0, sizeof(routing));
2878 r = kvm_set_irq_routing(kvm, &routing, 0, 0);
2879 }
2880 break;
2881 }
2882 case KVM_SET_DEVICE_ATTR: {
2883 r = -EFAULT;
2884 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2885 break;
2886 r = kvm_s390_vm_set_attr(kvm, &attr);
2887 break;
2888 }
2889 case KVM_GET_DEVICE_ATTR: {
2890 r = -EFAULT;
2891 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2892 break;
2893 r = kvm_s390_vm_get_attr(kvm, &attr);
2894 break;
2895 }
2896 case KVM_HAS_DEVICE_ATTR: {
2897 r = -EFAULT;
2898 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2899 break;
2900 r = kvm_s390_vm_has_attr(kvm, &attr);
2901 break;
2902 }
2903 case KVM_S390_GET_SKEYS: {
2904 struct kvm_s390_skeys args;
2905
2906 r = -EFAULT;
2907 if (copy_from_user(&args, argp,
2908 sizeof(struct kvm_s390_skeys)))
2909 break;
2910 r = kvm_s390_get_skeys(kvm, &args);
2911 break;
2912 }
2913 case KVM_S390_SET_SKEYS: {
2914 struct kvm_s390_skeys args;
2915
2916 r = -EFAULT;
2917 if (copy_from_user(&args, argp,
2918 sizeof(struct kvm_s390_skeys)))
2919 break;
2920 r = kvm_s390_set_skeys(kvm, &args);
2921 break;
2922 }
2923 case KVM_S390_GET_CMMA_BITS: {
2924 struct kvm_s390_cmma_log args;
2925
2926 r = -EFAULT;
2927 if (copy_from_user(&args, argp, sizeof(args)))
2928 break;
2929 mutex_lock(&kvm->slots_lock);
2930 r = kvm_s390_get_cmma_bits(kvm, &args);
2931 mutex_unlock(&kvm->slots_lock);
2932 if (!r) {
2933 r = copy_to_user(argp, &args, sizeof(args));
2934 if (r)
2935 r = -EFAULT;
2936 }
2937 break;
2938 }
2939 case KVM_S390_SET_CMMA_BITS: {
2940 struct kvm_s390_cmma_log args;
2941
2942 r = -EFAULT;
2943 if (copy_from_user(&args, argp, sizeof(args)))
2944 break;
2945 mutex_lock(&kvm->slots_lock);
2946 r = kvm_s390_set_cmma_bits(kvm, &args);
2947 mutex_unlock(&kvm->slots_lock);
2948 break;
2949 }
2950 case KVM_S390_PV_COMMAND: {
2951 struct kvm_pv_cmd args;
2952
2953 /* protvirt means user cpu state */
2954 kvm_s390_set_user_cpu_state_ctrl(kvm);
2955 r = 0;
2956 if (!is_prot_virt_host()) {
2957 r = -EINVAL;
2958 break;
2959 }
2960 if (copy_from_user(&args, argp, sizeof(args))) {
2961 r = -EFAULT;
2962 break;
2963 }
2964 if (args.flags) {
2965 r = -EINVAL;
2966 break;
2967 }
2968 /* must be called without kvm->lock */
2969 r = kvm_s390_handle_pv(kvm, &args);
2970 if (copy_to_user(argp, &args, sizeof(args))) {
2971 r = -EFAULT;
2972 break;
2973 }
2974 break;
2975 }
2976 case KVM_S390_MEM_OP: {
2977 struct kvm_s390_mem_op mem_op;
2978
2979 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
2980 r = kvm_s390_vm_mem_op(kvm, &mem_op);
2981 else
2982 r = -EFAULT;
2983 break;
2984 }
2985 case KVM_S390_ZPCI_OP: {
2986 struct kvm_s390_zpci_op args;
2987
2988 r = -EINVAL;
2989 if (!IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
2990 break;
2991 if (copy_from_user(&args, argp, sizeof(args))) {
2992 r = -EFAULT;
2993 break;
2994 }
2995 r = kvm_s390_pci_zpci_op(kvm, &args);
2996 break;
2997 }
2998 default:
2999 r = -ENOTTY;
3000 }
3001
3002 return r;
3003}
3004
3005static int kvm_s390_apxa_installed(void)
3006{
3007 struct ap_config_info info;
3008
3009 if (ap_instructions_available()) {
3010 if (ap_qci(&info) == 0)
3011 return info.apxa;
3012 }
3013
3014 return 0;
3015}
3016
3017/*
3018 * The format of the crypto control block (CRYCB) is specified in the 3 low
3019 * order bits of the CRYCB designation (CRYCBD) field as follows:
3020 * Format 0: Neither the message security assist extension 3 (MSAX3) nor the
3021 * AP extended addressing (APXA) facility are installed.
3022 * Format 1: The APXA facility is not installed but the MSAX3 facility is.
3023 * Format 2: Both the APXA and MSAX3 facilities are installed
3024 */
3025static void kvm_s390_set_crycb_format(struct kvm *kvm)
3026{
3027 kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
3028
3029 /* Clear the CRYCB format bits - i.e., set format 0 by default */
3030 kvm->arch.crypto.crycbd &= ~(CRYCB_FORMAT_MASK);
3031
3032 /* Check whether MSAX3 is installed */
3033 if (!test_kvm_facility(kvm, 76))
3034 return;
3035
3036 if (kvm_s390_apxa_installed())
3037 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
3038 else
3039 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
3040}
3041
3042/*
3043 * kvm_arch_crypto_set_masks
3044 *
3045 * @kvm: pointer to the target guest's KVM struct containing the crypto masks
3046 * to be set.
3047 * @apm: the mask identifying the accessible AP adapters
3048 * @aqm: the mask identifying the accessible AP domains
3049 * @adm: the mask identifying the accessible AP control domains
3050 *
3051 * Set the masks that identify the adapters, domains and control domains to
3052 * which the KVM guest is granted access.
3053 *
3054 * Note: The kvm->lock mutex must be locked by the caller before invoking this
3055 * function.
3056 */
3057void kvm_arch_crypto_set_masks(struct kvm *kvm, unsigned long *apm,
3058 unsigned long *aqm, unsigned long *adm)
3059{
3060 struct kvm_s390_crypto_cb *crycb = kvm->arch.crypto.crycb;
3061
3062 kvm_s390_vcpu_block_all(kvm);
3063
3064 switch (kvm->arch.crypto.crycbd & CRYCB_FORMAT_MASK) {
3065 case CRYCB_FORMAT2: /* APCB1 use 256 bits */
3066 memcpy(crycb->apcb1.apm, apm, 32);
3067 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx %016lx %016lx %016lx",
3068 apm[0], apm[1], apm[2], apm[3]);
3069 memcpy(crycb->apcb1.aqm, aqm, 32);
3070 VM_EVENT(kvm, 3, "SET CRYCB: aqm %016lx %016lx %016lx %016lx",
3071 aqm[0], aqm[1], aqm[2], aqm[3]);
3072 memcpy(crycb->apcb1.adm, adm, 32);
3073 VM_EVENT(kvm, 3, "SET CRYCB: adm %016lx %016lx %016lx %016lx",
3074 adm[0], adm[1], adm[2], adm[3]);
3075 break;
3076 case CRYCB_FORMAT1:
3077 case CRYCB_FORMAT0: /* Fall through both use APCB0 */
3078 memcpy(crycb->apcb0.apm, apm, 8);
3079 memcpy(crycb->apcb0.aqm, aqm, 2);
3080 memcpy(crycb->apcb0.adm, adm, 2);
3081 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx aqm %04x adm %04x",
3082 apm[0], *((unsigned short *)aqm),
3083 *((unsigned short *)adm));
3084 break;
3085 default: /* Can not happen */
3086 break;
3087 }
3088
3089 /* recreate the shadow crycb for each vcpu */
3090 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART);
3091 kvm_s390_vcpu_unblock_all(kvm);
3092}
3093EXPORT_SYMBOL_GPL(kvm_arch_crypto_set_masks);
3094
3095/*
3096 * kvm_arch_crypto_clear_masks
3097 *
3098 * @kvm: pointer to the target guest's KVM struct containing the crypto masks
3099 * to be cleared.
3100 *
3101 * Clear the masks that identify the adapters, domains and control domains to
3102 * which the KVM guest is granted access.
3103 *
3104 * Note: The kvm->lock mutex must be locked by the caller before invoking this
3105 * function.
3106 */
3107void kvm_arch_crypto_clear_masks(struct kvm *kvm)
3108{
3109 kvm_s390_vcpu_block_all(kvm);
3110
3111 memset(&kvm->arch.crypto.crycb->apcb0, 0,
3112 sizeof(kvm->arch.crypto.crycb->apcb0));
3113 memset(&kvm->arch.crypto.crycb->apcb1, 0,
3114 sizeof(kvm->arch.crypto.crycb->apcb1));
3115
3116 VM_EVENT(kvm, 3, "%s", "CLR CRYCB:");
3117 /* recreate the shadow crycb for each vcpu */
3118 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART);
3119 kvm_s390_vcpu_unblock_all(kvm);
3120}
3121EXPORT_SYMBOL_GPL(kvm_arch_crypto_clear_masks);
3122
3123static u64 kvm_s390_get_initial_cpuid(void)
3124{
3125 struct cpuid cpuid;
3126
3127 get_cpu_id(&cpuid);
3128 cpuid.version = 0xff;
3129 return *((u64 *) &cpuid);
3130}
3131
3132static void kvm_s390_crypto_init(struct kvm *kvm)
3133{
3134 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
3135 kvm_s390_set_crycb_format(kvm);
3136 init_rwsem(&kvm->arch.crypto.pqap_hook_rwsem);
3137
3138 if (!test_kvm_facility(kvm, 76))
3139 return;
3140
3141 /* Enable AES/DEA protected key functions by default */
3142 kvm->arch.crypto.aes_kw = 1;
3143 kvm->arch.crypto.dea_kw = 1;
3144 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
3145 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
3146 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
3147 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
3148}
3149
3150static void sca_dispose(struct kvm *kvm)
3151{
3152 if (kvm->arch.use_esca)
3153 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
3154 else
3155 free_page((unsigned long)(kvm->arch.sca));
3156 kvm->arch.sca = NULL;
3157}
3158
3159void kvm_arch_free_vm(struct kvm *kvm)
3160{
3161 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
3162 kvm_s390_pci_clear_list(kvm);
3163
3164 __kvm_arch_free_vm(kvm);
3165}
3166
3167int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
3168{
3169 gfp_t alloc_flags = GFP_KERNEL_ACCOUNT;
3170 int i, rc;
3171 char debug_name[16];
3172 static unsigned long sca_offset;
3173
3174 rc = -EINVAL;
3175#ifdef CONFIG_KVM_S390_UCONTROL
3176 if (type & ~KVM_VM_S390_UCONTROL)
3177 goto out_err;
3178 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
3179 goto out_err;
3180#else
3181 if (type)
3182 goto out_err;
3183#endif
3184
3185 rc = s390_enable_sie();
3186 if (rc)
3187 goto out_err;
3188
3189 rc = -ENOMEM;
3190
3191 if (!sclp.has_64bscao)
3192 alloc_flags |= GFP_DMA;
3193 rwlock_init(&kvm->arch.sca_lock);
3194 /* start with basic SCA */
3195 kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
3196 if (!kvm->arch.sca)
3197 goto out_err;
3198 mutex_lock(&kvm_lock);
3199 sca_offset += 16;
3200 if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
3201 sca_offset = 0;
3202 kvm->arch.sca = (struct bsca_block *)
3203 ((char *) kvm->arch.sca + sca_offset);
3204 mutex_unlock(&kvm_lock);
3205
3206 sprintf(debug_name, "kvm-%u", current->pid);
3207
3208 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
3209 if (!kvm->arch.dbf)
3210 goto out_err;
3211
3212 BUILD_BUG_ON(sizeof(struct sie_page2) != 4096);
3213 kvm->arch.sie_page2 =
3214 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3215 if (!kvm->arch.sie_page2)
3216 goto out_err;
3217
3218 kvm->arch.sie_page2->kvm = kvm;
3219 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
3220
3221 for (i = 0; i < kvm_s390_fac_size(); i++) {
3222 kvm->arch.model.fac_mask[i] = stfle_fac_list[i] &
3223 (kvm_s390_fac_base[i] |
3224 kvm_s390_fac_ext[i]);
3225 kvm->arch.model.fac_list[i] = stfle_fac_list[i] &
3226 kvm_s390_fac_base[i];
3227 }
3228 kvm->arch.model.subfuncs = kvm_s390_available_subfunc;
3229
3230 /* we are always in czam mode - even on pre z14 machines */
3231 set_kvm_facility(kvm->arch.model.fac_mask, 138);
3232 set_kvm_facility(kvm->arch.model.fac_list, 138);
3233 /* we emulate STHYI in kvm */
3234 set_kvm_facility(kvm->arch.model.fac_mask, 74);
3235 set_kvm_facility(kvm->arch.model.fac_list, 74);
3236 if (MACHINE_HAS_TLB_GUEST) {
3237 set_kvm_facility(kvm->arch.model.fac_mask, 147);
3238 set_kvm_facility(kvm->arch.model.fac_list, 147);
3239 }
3240
3241 if (css_general_characteristics.aiv && test_facility(65))
3242 set_kvm_facility(kvm->arch.model.fac_mask, 65);
3243
3244 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
3245 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
3246
3247 kvm_s390_crypto_init(kvm);
3248
3249 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
3250 mutex_lock(&kvm->lock);
3251 kvm_s390_pci_init_list(kvm);
3252 kvm_s390_vcpu_pci_enable_interp(kvm);
3253 mutex_unlock(&kvm->lock);
3254 }
3255
3256 mutex_init(&kvm->arch.float_int.ais_lock);
3257 spin_lock_init(&kvm->arch.float_int.lock);
3258 for (i = 0; i < FIRQ_LIST_COUNT; i++)
3259 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
3260 init_waitqueue_head(&kvm->arch.ipte_wq);
3261 mutex_init(&kvm->arch.ipte_mutex);
3262
3263 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
3264 VM_EVENT(kvm, 3, "vm created with type %lu", type);
3265
3266 if (type & KVM_VM_S390_UCONTROL) {
3267 kvm->arch.gmap = NULL;
3268 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
3269 } else {
3270 if (sclp.hamax == U64_MAX)
3271 kvm->arch.mem_limit = TASK_SIZE_MAX;
3272 else
3273 kvm->arch.mem_limit = min_t(unsigned long, TASK_SIZE_MAX,
3274 sclp.hamax + 1);
3275 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
3276 if (!kvm->arch.gmap)
3277 goto out_err;
3278 kvm->arch.gmap->private = kvm;
3279 kvm->arch.gmap->pfault_enabled = 0;
3280 }
3281
3282 kvm->arch.use_pfmfi = sclp.has_pfmfi;
3283 kvm->arch.use_skf = sclp.has_skey;
3284 spin_lock_init(&kvm->arch.start_stop_lock);
3285 kvm_s390_vsie_init(kvm);
3286 if (use_gisa)
3287 kvm_s390_gisa_init(kvm);
3288 INIT_LIST_HEAD(&kvm->arch.pv.need_cleanup);
3289 kvm->arch.pv.set_aside = NULL;
3290 KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
3291
3292 return 0;
3293out_err:
3294 free_page((unsigned long)kvm->arch.sie_page2);
3295 debug_unregister(kvm->arch.dbf);
3296 sca_dispose(kvm);
3297 KVM_EVENT(3, "creation of vm failed: %d", rc);
3298 return rc;
3299}
3300
3301void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3302{
3303 u16 rc, rrc;
3304
3305 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
3306 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
3307 kvm_s390_clear_local_irqs(vcpu);
3308 kvm_clear_async_pf_completion_queue(vcpu);
3309 if (!kvm_is_ucontrol(vcpu->kvm))
3310 sca_del_vcpu(vcpu);
3311 kvm_s390_update_topology_change_report(vcpu->kvm, 1);
3312
3313 if (kvm_is_ucontrol(vcpu->kvm))
3314 gmap_remove(vcpu->arch.gmap);
3315
3316 if (vcpu->kvm->arch.use_cmma)
3317 kvm_s390_vcpu_unsetup_cmma(vcpu);
3318 /* We can not hold the vcpu mutex here, we are already dying */
3319 if (kvm_s390_pv_cpu_get_handle(vcpu))
3320 kvm_s390_pv_destroy_cpu(vcpu, &rc, &rrc);
3321 free_page((unsigned long)(vcpu->arch.sie_block));
3322}
3323
3324void kvm_arch_destroy_vm(struct kvm *kvm)
3325{
3326 u16 rc, rrc;
3327
3328 kvm_destroy_vcpus(kvm);
3329 sca_dispose(kvm);
3330 kvm_s390_gisa_destroy(kvm);
3331 /*
3332 * We are already at the end of life and kvm->lock is not taken.
3333 * This is ok as the file descriptor is closed by now and nobody
3334 * can mess with the pv state.
3335 */
3336 kvm_s390_pv_deinit_cleanup_all(kvm, &rc, &rrc);
3337 /*
3338 * Remove the mmu notifier only when the whole KVM VM is torn down,
3339 * and only if one was registered to begin with. If the VM is
3340 * currently not protected, but has been previously been protected,
3341 * then it's possible that the notifier is still registered.
3342 */
3343 if (kvm->arch.pv.mmu_notifier.ops)
3344 mmu_notifier_unregister(&kvm->arch.pv.mmu_notifier, kvm->mm);
3345
3346 debug_unregister(kvm->arch.dbf);
3347 free_page((unsigned long)kvm->arch.sie_page2);
3348 if (!kvm_is_ucontrol(kvm))
3349 gmap_remove(kvm->arch.gmap);
3350 kvm_s390_destroy_adapters(kvm);
3351 kvm_s390_clear_float_irqs(kvm);
3352 kvm_s390_vsie_destroy(kvm);
3353 KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
3354}
3355
3356/* Section: vcpu related */
3357static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
3358{
3359 vcpu->arch.gmap = gmap_create(current->mm, -1UL);
3360 if (!vcpu->arch.gmap)
3361 return -ENOMEM;
3362 vcpu->arch.gmap->private = vcpu->kvm;
3363
3364 return 0;
3365}
3366
3367static void sca_del_vcpu(struct kvm_vcpu *vcpu)
3368{
3369 if (!kvm_s390_use_sca_entries())
3370 return;
3371 read_lock(&vcpu->kvm->arch.sca_lock);
3372 if (vcpu->kvm->arch.use_esca) {
3373 struct esca_block *sca = vcpu->kvm->arch.sca;
3374
3375 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
3376 sca->cpu[vcpu->vcpu_id].sda = 0;
3377 } else {
3378 struct bsca_block *sca = vcpu->kvm->arch.sca;
3379
3380 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
3381 sca->cpu[vcpu->vcpu_id].sda = 0;
3382 }
3383 read_unlock(&vcpu->kvm->arch.sca_lock);
3384}
3385
3386static void sca_add_vcpu(struct kvm_vcpu *vcpu)
3387{
3388 if (!kvm_s390_use_sca_entries()) {
3389 phys_addr_t sca_phys = virt_to_phys(vcpu->kvm->arch.sca);
3390
3391 /* we still need the basic sca for the ipte control */
3392 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3393 vcpu->arch.sie_block->scaol = sca_phys;
3394 return;
3395 }
3396 read_lock(&vcpu->kvm->arch.sca_lock);
3397 if (vcpu->kvm->arch.use_esca) {
3398 struct esca_block *sca = vcpu->kvm->arch.sca;
3399 phys_addr_t sca_phys = virt_to_phys(sca);
3400
3401 sca->cpu[vcpu->vcpu_id].sda = virt_to_phys(vcpu->arch.sie_block);
3402 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3403 vcpu->arch.sie_block->scaol = sca_phys & ESCA_SCAOL_MASK;
3404 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
3405 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
3406 } else {
3407 struct bsca_block *sca = vcpu->kvm->arch.sca;
3408 phys_addr_t sca_phys = virt_to_phys(sca);
3409
3410 sca->cpu[vcpu->vcpu_id].sda = virt_to_phys(vcpu->arch.sie_block);
3411 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3412 vcpu->arch.sie_block->scaol = sca_phys;
3413 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
3414 }
3415 read_unlock(&vcpu->kvm->arch.sca_lock);
3416}
3417
3418/* Basic SCA to Extended SCA data copy routines */
3419static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
3420{
3421 d->sda = s->sda;
3422 d->sigp_ctrl.c = s->sigp_ctrl.c;
3423 d->sigp_ctrl.scn = s->sigp_ctrl.scn;
3424}
3425
3426static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
3427{
3428 int i;
3429
3430 d->ipte_control = s->ipte_control;
3431 d->mcn[0] = s->mcn;
3432 for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
3433 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
3434}
3435
3436static int sca_switch_to_extended(struct kvm *kvm)
3437{
3438 struct bsca_block *old_sca = kvm->arch.sca;
3439 struct esca_block *new_sca;
3440 struct kvm_vcpu *vcpu;
3441 unsigned long vcpu_idx;
3442 u32 scaol, scaoh;
3443 phys_addr_t new_sca_phys;
3444
3445 if (kvm->arch.use_esca)
3446 return 0;
3447
3448 new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3449 if (!new_sca)
3450 return -ENOMEM;
3451
3452 new_sca_phys = virt_to_phys(new_sca);
3453 scaoh = new_sca_phys >> 32;
3454 scaol = new_sca_phys & ESCA_SCAOL_MASK;
3455
3456 kvm_s390_vcpu_block_all(kvm);
3457 write_lock(&kvm->arch.sca_lock);
3458
3459 sca_copy_b_to_e(new_sca, old_sca);
3460
3461 kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
3462 vcpu->arch.sie_block->scaoh = scaoh;
3463 vcpu->arch.sie_block->scaol = scaol;
3464 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
3465 }
3466 kvm->arch.sca = new_sca;
3467 kvm->arch.use_esca = 1;
3468
3469 write_unlock(&kvm->arch.sca_lock);
3470 kvm_s390_vcpu_unblock_all(kvm);
3471
3472 free_page((unsigned long)old_sca);
3473
3474 VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
3475 old_sca, kvm->arch.sca);
3476 return 0;
3477}
3478
3479static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
3480{
3481 int rc;
3482
3483 if (!kvm_s390_use_sca_entries()) {
3484 if (id < KVM_MAX_VCPUS)
3485 return true;
3486 return false;
3487 }
3488 if (id < KVM_S390_BSCA_CPU_SLOTS)
3489 return true;
3490 if (!sclp.has_esca || !sclp.has_64bscao)
3491 return false;
3492
3493 rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
3494
3495 return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
3496}
3497
3498/* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3499static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3500{
3501 WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
3502 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3503 vcpu->arch.cputm_start = get_tod_clock_fast();
3504 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3505}
3506
3507/* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3508static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3509{
3510 WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
3511 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3512 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
3513 vcpu->arch.cputm_start = 0;
3514 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3515}
3516
3517/* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3518static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3519{
3520 WARN_ON_ONCE(vcpu->arch.cputm_enabled);
3521 vcpu->arch.cputm_enabled = true;
3522 __start_cpu_timer_accounting(vcpu);
3523}
3524
3525/* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3526static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3527{
3528 WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
3529 __stop_cpu_timer_accounting(vcpu);
3530 vcpu->arch.cputm_enabled = false;
3531}
3532
3533static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3534{
3535 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3536 __enable_cpu_timer_accounting(vcpu);
3537 preempt_enable();
3538}
3539
3540static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3541{
3542 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3543 __disable_cpu_timer_accounting(vcpu);
3544 preempt_enable();
3545}
3546
3547/* set the cpu timer - may only be called from the VCPU thread itself */
3548void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
3549{
3550 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3551 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3552 if (vcpu->arch.cputm_enabled)
3553 vcpu->arch.cputm_start = get_tod_clock_fast();
3554 vcpu->arch.sie_block->cputm = cputm;
3555 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3556 preempt_enable();
3557}
3558
3559/* update and get the cpu timer - can also be called from other VCPU threads */
3560__u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
3561{
3562 unsigned int seq;
3563 __u64 value;
3564
3565 if (unlikely(!vcpu->arch.cputm_enabled))
3566 return vcpu->arch.sie_block->cputm;
3567
3568 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3569 do {
3570 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
3571 /*
3572 * If the writer would ever execute a read in the critical
3573 * section, e.g. in irq context, we have a deadlock.
3574 */
3575 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
3576 value = vcpu->arch.sie_block->cputm;
3577 /* if cputm_start is 0, accounting is being started/stopped */
3578 if (likely(vcpu->arch.cputm_start))
3579 value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
3580 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
3581 preempt_enable();
3582 return value;
3583}
3584
3585void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3586{
3587
3588 gmap_enable(vcpu->arch.enabled_gmap);
3589 kvm_s390_set_cpuflags(vcpu, CPUSTAT_RUNNING);
3590 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
3591 __start_cpu_timer_accounting(vcpu);
3592 vcpu->cpu = cpu;
3593}
3594
3595void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3596{
3597 vcpu->cpu = -1;
3598 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
3599 __stop_cpu_timer_accounting(vcpu);
3600 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_RUNNING);
3601 vcpu->arch.enabled_gmap = gmap_get_enabled();
3602 gmap_disable(vcpu->arch.enabled_gmap);
3603
3604}
3605
3606void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
3607{
3608 mutex_lock(&vcpu->kvm->lock);
3609 preempt_disable();
3610 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
3611 vcpu->arch.sie_block->epdx = vcpu->kvm->arch.epdx;
3612 preempt_enable();
3613 mutex_unlock(&vcpu->kvm->lock);
3614 if (!kvm_is_ucontrol(vcpu->kvm)) {
3615 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
3616 sca_add_vcpu(vcpu);
3617 }
3618 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
3619 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
3620 /* make vcpu_load load the right gmap on the first trigger */
3621 vcpu->arch.enabled_gmap = vcpu->arch.gmap;
3622}
3623
3624static bool kvm_has_pckmo_subfunc(struct kvm *kvm, unsigned long nr)
3625{
3626 if (test_bit_inv(nr, (unsigned long *)&kvm->arch.model.subfuncs.pckmo) &&
3627 test_bit_inv(nr, (unsigned long *)&kvm_s390_available_subfunc.pckmo))
3628 return true;
3629 return false;
3630}
3631
3632static bool kvm_has_pckmo_ecc(struct kvm *kvm)
3633{
3634 /* At least one ECC subfunction must be present */
3635 return kvm_has_pckmo_subfunc(kvm, 32) ||
3636 kvm_has_pckmo_subfunc(kvm, 33) ||
3637 kvm_has_pckmo_subfunc(kvm, 34) ||
3638 kvm_has_pckmo_subfunc(kvm, 40) ||
3639 kvm_has_pckmo_subfunc(kvm, 41);
3640
3641}
3642
3643static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
3644{
3645 /*
3646 * If the AP instructions are not being interpreted and the MSAX3
3647 * facility is not configured for the guest, there is nothing to set up.
3648 */
3649 if (!vcpu->kvm->arch.crypto.apie && !test_kvm_facility(vcpu->kvm, 76))
3650 return;
3651
3652 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
3653 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
3654 vcpu->arch.sie_block->eca &= ~ECA_APIE;
3655 vcpu->arch.sie_block->ecd &= ~ECD_ECC;
3656
3657 if (vcpu->kvm->arch.crypto.apie)
3658 vcpu->arch.sie_block->eca |= ECA_APIE;
3659
3660 /* Set up protected key support */
3661 if (vcpu->kvm->arch.crypto.aes_kw) {
3662 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
3663 /* ecc is also wrapped with AES key */
3664 if (kvm_has_pckmo_ecc(vcpu->kvm))
3665 vcpu->arch.sie_block->ecd |= ECD_ECC;
3666 }
3667
3668 if (vcpu->kvm->arch.crypto.dea_kw)
3669 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
3670}
3671
3672void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
3673{
3674 free_page((unsigned long)phys_to_virt(vcpu->arch.sie_block->cbrlo));
3675 vcpu->arch.sie_block->cbrlo = 0;
3676}
3677
3678int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
3679{
3680 void *cbrlo_page = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
3681
3682 if (!cbrlo_page)
3683 return -ENOMEM;
3684
3685 vcpu->arch.sie_block->cbrlo = virt_to_phys(cbrlo_page);
3686 return 0;
3687}
3688
3689static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
3690{
3691 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
3692
3693 vcpu->arch.sie_block->ibc = model->ibc;
3694 if (test_kvm_facility(vcpu->kvm, 7))
3695 vcpu->arch.sie_block->fac = virt_to_phys(model->fac_list);
3696}
3697
3698static int kvm_s390_vcpu_setup(struct kvm_vcpu *vcpu)
3699{
3700 int rc = 0;
3701 u16 uvrc, uvrrc;
3702
3703 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
3704 CPUSTAT_SM |
3705 CPUSTAT_STOPPED);
3706
3707 if (test_kvm_facility(vcpu->kvm, 78))
3708 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED2);
3709 else if (test_kvm_facility(vcpu->kvm, 8))
3710 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED);
3711
3712 kvm_s390_vcpu_setup_model(vcpu);
3713
3714 /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
3715 if (MACHINE_HAS_ESOP)
3716 vcpu->arch.sie_block->ecb |= ECB_HOSTPROTINT;
3717 if (test_kvm_facility(vcpu->kvm, 9))
3718 vcpu->arch.sie_block->ecb |= ECB_SRSI;
3719 if (test_kvm_facility(vcpu->kvm, 11))
3720 vcpu->arch.sie_block->ecb |= ECB_PTF;
3721 if (test_kvm_facility(vcpu->kvm, 73))
3722 vcpu->arch.sie_block->ecb |= ECB_TE;
3723 if (!kvm_is_ucontrol(vcpu->kvm))
3724 vcpu->arch.sie_block->ecb |= ECB_SPECI;
3725
3726 if (test_kvm_facility(vcpu->kvm, 8) && vcpu->kvm->arch.use_pfmfi)
3727 vcpu->arch.sie_block->ecb2 |= ECB2_PFMFI;
3728 if (test_kvm_facility(vcpu->kvm, 130))
3729 vcpu->arch.sie_block->ecb2 |= ECB2_IEP;
3730 vcpu->arch.sie_block->eca = ECA_MVPGI | ECA_PROTEXCI;
3731 if (sclp.has_cei)
3732 vcpu->arch.sie_block->eca |= ECA_CEI;
3733 if (sclp.has_ib)
3734 vcpu->arch.sie_block->eca |= ECA_IB;
3735 if (sclp.has_siif)
3736 vcpu->arch.sie_block->eca |= ECA_SII;
3737 if (sclp.has_sigpif)
3738 vcpu->arch.sie_block->eca |= ECA_SIGPI;
3739 if (test_kvm_facility(vcpu->kvm, 129)) {
3740 vcpu->arch.sie_block->eca |= ECA_VX;
3741 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
3742 }
3743 if (test_kvm_facility(vcpu->kvm, 139))
3744 vcpu->arch.sie_block->ecd |= ECD_MEF;
3745 if (test_kvm_facility(vcpu->kvm, 156))
3746 vcpu->arch.sie_block->ecd |= ECD_ETOKENF;
3747 if (vcpu->arch.sie_block->gd) {
3748 vcpu->arch.sie_block->eca |= ECA_AIV;
3749 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
3750 vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
3751 }
3752 vcpu->arch.sie_block->sdnxo = virt_to_phys(&vcpu->run->s.regs.sdnx) | SDNXC;
3753 vcpu->arch.sie_block->riccbd = virt_to_phys(&vcpu->run->s.regs.riccb);
3754
3755 if (sclp.has_kss)
3756 kvm_s390_set_cpuflags(vcpu, CPUSTAT_KSS);
3757 else
3758 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
3759
3760 if (vcpu->kvm->arch.use_cmma) {
3761 rc = kvm_s390_vcpu_setup_cmma(vcpu);
3762 if (rc)
3763 return rc;
3764 }
3765 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3766 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
3767
3768 vcpu->arch.sie_block->hpid = HPID_KVM;
3769
3770 kvm_s390_vcpu_crypto_setup(vcpu);
3771
3772 kvm_s390_vcpu_pci_setup(vcpu);
3773
3774 mutex_lock(&vcpu->kvm->lock);
3775 if (kvm_s390_pv_is_protected(vcpu->kvm)) {
3776 rc = kvm_s390_pv_create_cpu(vcpu, &uvrc, &uvrrc);
3777 if (rc)
3778 kvm_s390_vcpu_unsetup_cmma(vcpu);
3779 }
3780 mutex_unlock(&vcpu->kvm->lock);
3781
3782 return rc;
3783}
3784
3785int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
3786{
3787 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
3788 return -EINVAL;
3789 return 0;
3790}
3791
3792int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
3793{
3794 struct sie_page *sie_page;
3795 int rc;
3796
3797 BUILD_BUG_ON(sizeof(struct sie_page) != 4096);
3798 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL_ACCOUNT);
3799 if (!sie_page)
3800 return -ENOMEM;
3801
3802 vcpu->arch.sie_block = &sie_page->sie_block;
3803 vcpu->arch.sie_block->itdba = virt_to_phys(&sie_page->itdb);
3804
3805 /* the real guest size will always be smaller than msl */
3806 vcpu->arch.sie_block->mso = 0;
3807 vcpu->arch.sie_block->msl = sclp.hamax;
3808
3809 vcpu->arch.sie_block->icpua = vcpu->vcpu_id;
3810 spin_lock_init(&vcpu->arch.local_int.lock);
3811 vcpu->arch.sie_block->gd = kvm_s390_get_gisa_desc(vcpu->kvm);
3812 seqcount_init(&vcpu->arch.cputm_seqcount);
3813
3814 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
3815 kvm_clear_async_pf_completion_queue(vcpu);
3816 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
3817 KVM_SYNC_GPRS |
3818 KVM_SYNC_ACRS |
3819 KVM_SYNC_CRS |
3820 KVM_SYNC_ARCH0 |
3821 KVM_SYNC_PFAULT |
3822 KVM_SYNC_DIAG318;
3823 kvm_s390_set_prefix(vcpu, 0);
3824 if (test_kvm_facility(vcpu->kvm, 64))
3825 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
3826 if (test_kvm_facility(vcpu->kvm, 82))
3827 vcpu->run->kvm_valid_regs |= KVM_SYNC_BPBC;
3828 if (test_kvm_facility(vcpu->kvm, 133))
3829 vcpu->run->kvm_valid_regs |= KVM_SYNC_GSCB;
3830 if (test_kvm_facility(vcpu->kvm, 156))
3831 vcpu->run->kvm_valid_regs |= KVM_SYNC_ETOKEN;
3832 /* fprs can be synchronized via vrs, even if the guest has no vx. With
3833 * MACHINE_HAS_VX, (load|store)_fpu_regs() will work with vrs format.
3834 */
3835 if (MACHINE_HAS_VX)
3836 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
3837 else
3838 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
3839
3840 if (kvm_is_ucontrol(vcpu->kvm)) {
3841 rc = __kvm_ucontrol_vcpu_init(vcpu);
3842 if (rc)
3843 goto out_free_sie_block;
3844 }
3845
3846 VM_EVENT(vcpu->kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK",
3847 vcpu->vcpu_id, vcpu, vcpu->arch.sie_block);
3848 trace_kvm_s390_create_vcpu(vcpu->vcpu_id, vcpu, vcpu->arch.sie_block);
3849
3850 rc = kvm_s390_vcpu_setup(vcpu);
3851 if (rc)
3852 goto out_ucontrol_uninit;
3853
3854 kvm_s390_update_topology_change_report(vcpu->kvm, 1);
3855 return 0;
3856
3857out_ucontrol_uninit:
3858 if (kvm_is_ucontrol(vcpu->kvm))
3859 gmap_remove(vcpu->arch.gmap);
3860out_free_sie_block:
3861 free_page((unsigned long)(vcpu->arch.sie_block));
3862 return rc;
3863}
3864
3865int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
3866{
3867 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask);
3868 return kvm_s390_vcpu_has_irq(vcpu, 0);
3869}
3870
3871bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
3872{
3873 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE);
3874}
3875
3876void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
3877{
3878 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
3879 exit_sie(vcpu);
3880}
3881
3882void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
3883{
3884 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
3885}
3886
3887static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
3888{
3889 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
3890 exit_sie(vcpu);
3891}
3892
3893bool kvm_s390_vcpu_sie_inhibited(struct kvm_vcpu *vcpu)
3894{
3895 return atomic_read(&vcpu->arch.sie_block->prog20) &
3896 (PROG_BLOCK_SIE | PROG_REQUEST);
3897}
3898
3899static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
3900{
3901 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
3902}
3903
3904/*
3905 * Kick a guest cpu out of (v)SIE and wait until (v)SIE is not running.
3906 * If the CPU is not running (e.g. waiting as idle) the function will
3907 * return immediately. */
3908void exit_sie(struct kvm_vcpu *vcpu)
3909{
3910 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
3911 kvm_s390_vsie_kick(vcpu);
3912 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
3913 cpu_relax();
3914}
3915
3916/* Kick a guest cpu out of SIE to process a request synchronously */
3917void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
3918{
3919 __kvm_make_request(req, vcpu);
3920 kvm_s390_vcpu_request(vcpu);
3921}
3922
3923static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
3924 unsigned long end)
3925{
3926 struct kvm *kvm = gmap->private;
3927 struct kvm_vcpu *vcpu;
3928 unsigned long prefix;
3929 unsigned long i;
3930
3931 if (gmap_is_shadow(gmap))
3932 return;
3933 if (start >= 1UL << 31)
3934 /* We are only interested in prefix pages */
3935 return;
3936 kvm_for_each_vcpu(i, vcpu, kvm) {
3937 /* match against both prefix pages */
3938 prefix = kvm_s390_get_prefix(vcpu);
3939 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
3940 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
3941 start, end);
3942 kvm_s390_sync_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu);
3943 }
3944 }
3945}
3946
3947bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
3948{
3949 /* do not poll with more than halt_poll_max_steal percent of steal time */
3950 if (S390_lowcore.avg_steal_timer * 100 / (TICK_USEC << 12) >=
3951 READ_ONCE(halt_poll_max_steal)) {
3952 vcpu->stat.halt_no_poll_steal++;
3953 return true;
3954 }
3955 return false;
3956}
3957
3958int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
3959{
3960 /* kvm common code refers to this, but never calls it */
3961 BUG();
3962 return 0;
3963}
3964
3965static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
3966 struct kvm_one_reg *reg)
3967{
3968 int r = -EINVAL;
3969
3970 switch (reg->id) {
3971 case KVM_REG_S390_TODPR:
3972 r = put_user(vcpu->arch.sie_block->todpr,
3973 (u32 __user *)reg->addr);
3974 break;
3975 case KVM_REG_S390_EPOCHDIFF:
3976 r = put_user(vcpu->arch.sie_block->epoch,
3977 (u64 __user *)reg->addr);
3978 break;
3979 case KVM_REG_S390_CPU_TIMER:
3980 r = put_user(kvm_s390_get_cpu_timer(vcpu),
3981 (u64 __user *)reg->addr);
3982 break;
3983 case KVM_REG_S390_CLOCK_COMP:
3984 r = put_user(vcpu->arch.sie_block->ckc,
3985 (u64 __user *)reg->addr);
3986 break;
3987 case KVM_REG_S390_PFTOKEN:
3988 r = put_user(vcpu->arch.pfault_token,
3989 (u64 __user *)reg->addr);
3990 break;
3991 case KVM_REG_S390_PFCOMPARE:
3992 r = put_user(vcpu->arch.pfault_compare,
3993 (u64 __user *)reg->addr);
3994 break;
3995 case KVM_REG_S390_PFSELECT:
3996 r = put_user(vcpu->arch.pfault_select,
3997 (u64 __user *)reg->addr);
3998 break;
3999 case KVM_REG_S390_PP:
4000 r = put_user(vcpu->arch.sie_block->pp,
4001 (u64 __user *)reg->addr);
4002 break;
4003 case KVM_REG_S390_GBEA:
4004 r = put_user(vcpu->arch.sie_block->gbea,
4005 (u64 __user *)reg->addr);
4006 break;
4007 default:
4008 break;
4009 }
4010
4011 return r;
4012}
4013
4014static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
4015 struct kvm_one_reg *reg)
4016{
4017 int r = -EINVAL;
4018 __u64 val;
4019
4020 switch (reg->id) {
4021 case KVM_REG_S390_TODPR:
4022 r = get_user(vcpu->arch.sie_block->todpr,
4023 (u32 __user *)reg->addr);
4024 break;
4025 case KVM_REG_S390_EPOCHDIFF:
4026 r = get_user(vcpu->arch.sie_block->epoch,
4027 (u64 __user *)reg->addr);
4028 break;
4029 case KVM_REG_S390_CPU_TIMER:
4030 r = get_user(val, (u64 __user *)reg->addr);
4031 if (!r)
4032 kvm_s390_set_cpu_timer(vcpu, val);
4033 break;
4034 case KVM_REG_S390_CLOCK_COMP:
4035 r = get_user(vcpu->arch.sie_block->ckc,
4036 (u64 __user *)reg->addr);
4037 break;
4038 case KVM_REG_S390_PFTOKEN:
4039 r = get_user(vcpu->arch.pfault_token,
4040 (u64 __user *)reg->addr);
4041 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4042 kvm_clear_async_pf_completion_queue(vcpu);
4043 break;
4044 case KVM_REG_S390_PFCOMPARE:
4045 r = get_user(vcpu->arch.pfault_compare,
4046 (u64 __user *)reg->addr);
4047 break;
4048 case KVM_REG_S390_PFSELECT:
4049 r = get_user(vcpu->arch.pfault_select,
4050 (u64 __user *)reg->addr);
4051 break;
4052 case KVM_REG_S390_PP:
4053 r = get_user(vcpu->arch.sie_block->pp,
4054 (u64 __user *)reg->addr);
4055 break;
4056 case KVM_REG_S390_GBEA:
4057 r = get_user(vcpu->arch.sie_block->gbea,
4058 (u64 __user *)reg->addr);
4059 break;
4060 default:
4061 break;
4062 }
4063
4064 return r;
4065}
4066
4067static void kvm_arch_vcpu_ioctl_normal_reset(struct kvm_vcpu *vcpu)
4068{
4069 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_RI;
4070 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
4071 memset(vcpu->run->s.regs.riccb, 0, sizeof(vcpu->run->s.regs.riccb));
4072
4073 kvm_clear_async_pf_completion_queue(vcpu);
4074 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
4075 kvm_s390_vcpu_stop(vcpu);
4076 kvm_s390_clear_local_irqs(vcpu);
4077}
4078
4079static void kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
4080{
4081 /* Initial reset is a superset of the normal reset */
4082 kvm_arch_vcpu_ioctl_normal_reset(vcpu);
4083
4084 /*
4085 * This equals initial cpu reset in pop, but we don't switch to ESA.
4086 * We do not only reset the internal data, but also ...
4087 */
4088 vcpu->arch.sie_block->gpsw.mask = 0;
4089 vcpu->arch.sie_block->gpsw.addr = 0;
4090 kvm_s390_set_prefix(vcpu, 0);
4091 kvm_s390_set_cpu_timer(vcpu, 0);
4092 vcpu->arch.sie_block->ckc = 0;
4093 memset(vcpu->arch.sie_block->gcr, 0, sizeof(vcpu->arch.sie_block->gcr));
4094 vcpu->arch.sie_block->gcr[0] = CR0_INITIAL_MASK;
4095 vcpu->arch.sie_block->gcr[14] = CR14_INITIAL_MASK;
4096
4097 /* ... the data in sync regs */
4098 memset(vcpu->run->s.regs.crs, 0, sizeof(vcpu->run->s.regs.crs));
4099 vcpu->run->s.regs.ckc = 0;
4100 vcpu->run->s.regs.crs[0] = CR0_INITIAL_MASK;
4101 vcpu->run->s.regs.crs[14] = CR14_INITIAL_MASK;
4102 vcpu->run->psw_addr = 0;
4103 vcpu->run->psw_mask = 0;
4104 vcpu->run->s.regs.todpr = 0;
4105 vcpu->run->s.regs.cputm = 0;
4106 vcpu->run->s.regs.ckc = 0;
4107 vcpu->run->s.regs.pp = 0;
4108 vcpu->run->s.regs.gbea = 1;
4109 vcpu->run->s.regs.fpc = 0;
4110 /*
4111 * Do not reset these registers in the protected case, as some of
4112 * them are overlayed and they are not accessible in this case
4113 * anyway.
4114 */
4115 if (!kvm_s390_pv_cpu_is_protected(vcpu)) {
4116 vcpu->arch.sie_block->gbea = 1;
4117 vcpu->arch.sie_block->pp = 0;
4118 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
4119 vcpu->arch.sie_block->todpr = 0;
4120 }
4121}
4122
4123static void kvm_arch_vcpu_ioctl_clear_reset(struct kvm_vcpu *vcpu)
4124{
4125 struct kvm_sync_regs *regs = &vcpu->run->s.regs;
4126
4127 /* Clear reset is a superset of the initial reset */
4128 kvm_arch_vcpu_ioctl_initial_reset(vcpu);
4129
4130 memset(®s->gprs, 0, sizeof(regs->gprs));
4131 memset(®s->vrs, 0, sizeof(regs->vrs));
4132 memset(®s->acrs, 0, sizeof(regs->acrs));
4133 memset(®s->gscb, 0, sizeof(regs->gscb));
4134
4135 regs->etoken = 0;
4136 regs->etoken_extension = 0;
4137}
4138
4139int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4140{
4141 vcpu_load(vcpu);
4142 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs));
4143 vcpu_put(vcpu);
4144 return 0;
4145}
4146
4147int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4148{
4149 vcpu_load(vcpu);
4150 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
4151 vcpu_put(vcpu);
4152 return 0;
4153}
4154
4155int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4156 struct kvm_sregs *sregs)
4157{
4158 vcpu_load(vcpu);
4159
4160 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
4161 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
4162
4163 vcpu_put(vcpu);
4164 return 0;
4165}
4166
4167int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4168 struct kvm_sregs *sregs)
4169{
4170 vcpu_load(vcpu);
4171
4172 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
4173 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
4174
4175 vcpu_put(vcpu);
4176 return 0;
4177}
4178
4179int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4180{
4181 int ret = 0;
4182
4183 vcpu_load(vcpu);
4184
4185 if (test_fp_ctl(fpu->fpc)) {
4186 ret = -EINVAL;
4187 goto out;
4188 }
4189 vcpu->run->s.regs.fpc = fpu->fpc;
4190 if (MACHINE_HAS_VX)
4191 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs,
4192 (freg_t *) fpu->fprs);
4193 else
4194 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs));
4195
4196out:
4197 vcpu_put(vcpu);
4198 return ret;
4199}
4200
4201int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4202{
4203 vcpu_load(vcpu);
4204
4205 /* make sure we have the latest values */
4206 save_fpu_regs();
4207 if (MACHINE_HAS_VX)
4208 convert_vx_to_fp((freg_t *) fpu->fprs,
4209 (__vector128 *) vcpu->run->s.regs.vrs);
4210 else
4211 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs));
4212 fpu->fpc = vcpu->run->s.regs.fpc;
4213
4214 vcpu_put(vcpu);
4215 return 0;
4216}
4217
4218static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
4219{
4220 int rc = 0;
4221
4222 if (!is_vcpu_stopped(vcpu))
4223 rc = -EBUSY;
4224 else {
4225 vcpu->run->psw_mask = psw.mask;
4226 vcpu->run->psw_addr = psw.addr;
4227 }
4228 return rc;
4229}
4230
4231int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4232 struct kvm_translation *tr)
4233{
4234 return -EINVAL; /* not implemented yet */
4235}
4236
4237#define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
4238 KVM_GUESTDBG_USE_HW_BP | \
4239 KVM_GUESTDBG_ENABLE)
4240
4241int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4242 struct kvm_guest_debug *dbg)
4243{
4244 int rc = 0;
4245
4246 vcpu_load(vcpu);
4247
4248 vcpu->guest_debug = 0;
4249 kvm_s390_clear_bp_data(vcpu);
4250
4251 if (dbg->control & ~VALID_GUESTDBG_FLAGS) {
4252 rc = -EINVAL;
4253 goto out;
4254 }
4255 if (!sclp.has_gpere) {
4256 rc = -EINVAL;
4257 goto out;
4258 }
4259
4260 if (dbg->control & KVM_GUESTDBG_ENABLE) {
4261 vcpu->guest_debug = dbg->control;
4262 /* enforce guest PER */
4263 kvm_s390_set_cpuflags(vcpu, CPUSTAT_P);
4264
4265 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
4266 rc = kvm_s390_import_bp_data(vcpu, dbg);
4267 } else {
4268 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
4269 vcpu->arch.guestdbg.last_bp = 0;
4270 }
4271
4272 if (rc) {
4273 vcpu->guest_debug = 0;
4274 kvm_s390_clear_bp_data(vcpu);
4275 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
4276 }
4277
4278out:
4279 vcpu_put(vcpu);
4280 return rc;
4281}
4282
4283int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4284 struct kvm_mp_state *mp_state)
4285{
4286 int ret;
4287
4288 vcpu_load(vcpu);
4289
4290 /* CHECK_STOP and LOAD are not supported yet */
4291 ret = is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
4292 KVM_MP_STATE_OPERATING;
4293
4294 vcpu_put(vcpu);
4295 return ret;
4296}
4297
4298int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4299 struct kvm_mp_state *mp_state)
4300{
4301 int rc = 0;
4302
4303 vcpu_load(vcpu);
4304
4305 /* user space knows about this interface - let it control the state */
4306 kvm_s390_set_user_cpu_state_ctrl(vcpu->kvm);
4307
4308 switch (mp_state->mp_state) {
4309 case KVM_MP_STATE_STOPPED:
4310 rc = kvm_s390_vcpu_stop(vcpu);
4311 break;
4312 case KVM_MP_STATE_OPERATING:
4313 rc = kvm_s390_vcpu_start(vcpu);
4314 break;
4315 case KVM_MP_STATE_LOAD:
4316 if (!kvm_s390_pv_cpu_is_protected(vcpu)) {
4317 rc = -ENXIO;
4318 break;
4319 }
4320 rc = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR_LOAD);
4321 break;
4322 case KVM_MP_STATE_CHECK_STOP:
4323 fallthrough; /* CHECK_STOP and LOAD are not supported yet */
4324 default:
4325 rc = -ENXIO;
4326 }
4327
4328 vcpu_put(vcpu);
4329 return rc;
4330}
4331
4332static bool ibs_enabled(struct kvm_vcpu *vcpu)
4333{
4334 return kvm_s390_test_cpuflags(vcpu, CPUSTAT_IBS);
4335}
4336
4337static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
4338{
4339retry:
4340 kvm_s390_vcpu_request_handled(vcpu);
4341 if (!kvm_request_pending(vcpu))
4342 return 0;
4343 /*
4344 * If the guest prefix changed, re-arm the ipte notifier for the
4345 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
4346 * This ensures that the ipte instruction for this request has
4347 * already finished. We might race against a second unmapper that
4348 * wants to set the blocking bit. Lets just retry the request loop.
4349 */
4350 if (kvm_check_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu)) {
4351 int rc;
4352 rc = gmap_mprotect_notify(vcpu->arch.gmap,
4353 kvm_s390_get_prefix(vcpu),
4354 PAGE_SIZE * 2, PROT_WRITE);
4355 if (rc) {
4356 kvm_make_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu);
4357 return rc;
4358 }
4359 goto retry;
4360 }
4361
4362 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
4363 vcpu->arch.sie_block->ihcpu = 0xffff;
4364 goto retry;
4365 }
4366
4367 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
4368 if (!ibs_enabled(vcpu)) {
4369 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
4370 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IBS);
4371 }
4372 goto retry;
4373 }
4374
4375 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
4376 if (ibs_enabled(vcpu)) {
4377 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
4378 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IBS);
4379 }
4380 goto retry;
4381 }
4382
4383 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
4384 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
4385 goto retry;
4386 }
4387
4388 if (kvm_check_request(KVM_REQ_START_MIGRATION, vcpu)) {
4389 /*
4390 * Disable CMM virtualization; we will emulate the ESSA
4391 * instruction manually, in order to provide additional
4392 * functionalities needed for live migration.
4393 */
4394 vcpu->arch.sie_block->ecb2 &= ~ECB2_CMMA;
4395 goto retry;
4396 }
4397
4398 if (kvm_check_request(KVM_REQ_STOP_MIGRATION, vcpu)) {
4399 /*
4400 * Re-enable CMM virtualization if CMMA is available and
4401 * CMM has been used.
4402 */
4403 if ((vcpu->kvm->arch.use_cmma) &&
4404 (vcpu->kvm->mm->context.uses_cmm))
4405 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
4406 goto retry;
4407 }
4408
4409 /* we left the vsie handler, nothing to do, just clear the request */
4410 kvm_clear_request(KVM_REQ_VSIE_RESTART, vcpu);
4411
4412 return 0;
4413}
4414
4415static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod)
4416{
4417 struct kvm_vcpu *vcpu;
4418 union tod_clock clk;
4419 unsigned long i;
4420
4421 preempt_disable();
4422
4423 store_tod_clock_ext(&clk);
4424
4425 kvm->arch.epoch = gtod->tod - clk.tod;
4426 kvm->arch.epdx = 0;
4427 if (test_kvm_facility(kvm, 139)) {
4428 kvm->arch.epdx = gtod->epoch_idx - clk.ei;
4429 if (kvm->arch.epoch > gtod->tod)
4430 kvm->arch.epdx -= 1;
4431 }
4432
4433 kvm_s390_vcpu_block_all(kvm);
4434 kvm_for_each_vcpu(i, vcpu, kvm) {
4435 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
4436 vcpu->arch.sie_block->epdx = kvm->arch.epdx;
4437 }
4438
4439 kvm_s390_vcpu_unblock_all(kvm);
4440 preempt_enable();
4441}
4442
4443int kvm_s390_try_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod)
4444{
4445 if (!mutex_trylock(&kvm->lock))
4446 return 0;
4447 __kvm_s390_set_tod_clock(kvm, gtod);
4448 mutex_unlock(&kvm->lock);
4449 return 1;
4450}
4451
4452/**
4453 * kvm_arch_fault_in_page - fault-in guest page if necessary
4454 * @vcpu: The corresponding virtual cpu
4455 * @gpa: Guest physical address
4456 * @writable: Whether the page should be writable or not
4457 *
4458 * Make sure that a guest page has been faulted-in on the host.
4459 *
4460 * Return: Zero on success, negative error code otherwise.
4461 */
4462long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
4463{
4464 return gmap_fault(vcpu->arch.gmap, gpa,
4465 writable ? FAULT_FLAG_WRITE : 0);
4466}
4467
4468static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
4469 unsigned long token)
4470{
4471 struct kvm_s390_interrupt inti;
4472 struct kvm_s390_irq irq;
4473
4474 if (start_token) {
4475 irq.u.ext.ext_params2 = token;
4476 irq.type = KVM_S390_INT_PFAULT_INIT;
4477 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
4478 } else {
4479 inti.type = KVM_S390_INT_PFAULT_DONE;
4480 inti.parm64 = token;
4481 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
4482 }
4483}
4484
4485bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
4486 struct kvm_async_pf *work)
4487{
4488 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
4489 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
4490
4491 return true;
4492}
4493
4494void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
4495 struct kvm_async_pf *work)
4496{
4497 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
4498 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
4499}
4500
4501void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
4502 struct kvm_async_pf *work)
4503{
4504 /* s390 will always inject the page directly */
4505}
4506
4507bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
4508{
4509 /*
4510 * s390 will always inject the page directly,
4511 * but we still want check_async_completion to cleanup
4512 */
4513 return true;
4514}
4515
4516static bool kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
4517{
4518 hva_t hva;
4519 struct kvm_arch_async_pf arch;
4520
4521 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4522 return false;
4523 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
4524 vcpu->arch.pfault_compare)
4525 return false;
4526 if (psw_extint_disabled(vcpu))
4527 return false;
4528 if (kvm_s390_vcpu_has_irq(vcpu, 0))
4529 return false;
4530 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK))
4531 return false;
4532 if (!vcpu->arch.gmap->pfault_enabled)
4533 return false;
4534
4535 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
4536 hva += current->thread.gmap_addr & ~PAGE_MASK;
4537 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
4538 return false;
4539
4540 return kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
4541}
4542
4543static int vcpu_pre_run(struct kvm_vcpu *vcpu)
4544{
4545 int rc, cpuflags;
4546
4547 /*
4548 * On s390 notifications for arriving pages will be delivered directly
4549 * to the guest but the house keeping for completed pfaults is
4550 * handled outside the worker.
4551 */
4552 kvm_check_async_pf_completion(vcpu);
4553
4554 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
4555 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
4556
4557 if (need_resched())
4558 schedule();
4559
4560 if (!kvm_is_ucontrol(vcpu->kvm)) {
4561 rc = kvm_s390_deliver_pending_interrupts(vcpu);
4562 if (rc)
4563 return rc;
4564 }
4565
4566 rc = kvm_s390_handle_requests(vcpu);
4567 if (rc)
4568 return rc;
4569
4570 if (guestdbg_enabled(vcpu)) {
4571 kvm_s390_backup_guest_per_regs(vcpu);
4572 kvm_s390_patch_guest_per_regs(vcpu);
4573 }
4574
4575 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask);
4576
4577 vcpu->arch.sie_block->icptcode = 0;
4578 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
4579 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
4580 trace_kvm_s390_sie_enter(vcpu, cpuflags);
4581
4582 return 0;
4583}
4584
4585static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
4586{
4587 struct kvm_s390_pgm_info pgm_info = {
4588 .code = PGM_ADDRESSING,
4589 };
4590 u8 opcode, ilen;
4591 int rc;
4592
4593 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
4594 trace_kvm_s390_sie_fault(vcpu);
4595
4596 /*
4597 * We want to inject an addressing exception, which is defined as a
4598 * suppressing or terminating exception. However, since we came here
4599 * by a DAT access exception, the PSW still points to the faulting
4600 * instruction since DAT exceptions are nullifying. So we've got
4601 * to look up the current opcode to get the length of the instruction
4602 * to be able to forward the PSW.
4603 */
4604 rc = read_guest_instr(vcpu, vcpu->arch.sie_block->gpsw.addr, &opcode, 1);
4605 ilen = insn_length(opcode);
4606 if (rc < 0) {
4607 return rc;
4608 } else if (rc) {
4609 /* Instruction-Fetching Exceptions - we can't detect the ilen.
4610 * Forward by arbitrary ilc, injection will take care of
4611 * nullification if necessary.
4612 */
4613 pgm_info = vcpu->arch.pgm;
4614 ilen = 4;
4615 }
4616 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
4617 kvm_s390_forward_psw(vcpu, ilen);
4618 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
4619}
4620
4621static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
4622{
4623 struct mcck_volatile_info *mcck_info;
4624 struct sie_page *sie_page;
4625
4626 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
4627 vcpu->arch.sie_block->icptcode);
4628 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
4629
4630 if (guestdbg_enabled(vcpu))
4631 kvm_s390_restore_guest_per_regs(vcpu);
4632
4633 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
4634 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
4635
4636 if (exit_reason == -EINTR) {
4637 VCPU_EVENT(vcpu, 3, "%s", "machine check");
4638 sie_page = container_of(vcpu->arch.sie_block,
4639 struct sie_page, sie_block);
4640 mcck_info = &sie_page->mcck_info;
4641 kvm_s390_reinject_machine_check(vcpu, mcck_info);
4642 return 0;
4643 }
4644
4645 if (vcpu->arch.sie_block->icptcode > 0) {
4646 int rc = kvm_handle_sie_intercept(vcpu);
4647
4648 if (rc != -EOPNOTSUPP)
4649 return rc;
4650 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
4651 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
4652 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
4653 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
4654 return -EREMOTE;
4655 } else if (exit_reason != -EFAULT) {
4656 vcpu->stat.exit_null++;
4657 return 0;
4658 } else if (kvm_is_ucontrol(vcpu->kvm)) {
4659 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
4660 vcpu->run->s390_ucontrol.trans_exc_code =
4661 current->thread.gmap_addr;
4662 vcpu->run->s390_ucontrol.pgm_code = 0x10;
4663 return -EREMOTE;
4664 } else if (current->thread.gmap_pfault) {
4665 trace_kvm_s390_major_guest_pfault(vcpu);
4666 current->thread.gmap_pfault = 0;
4667 if (kvm_arch_setup_async_pf(vcpu))
4668 return 0;
4669 vcpu->stat.pfault_sync++;
4670 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
4671 }
4672 return vcpu_post_run_fault_in_sie(vcpu);
4673}
4674
4675#define PSW_INT_MASK (PSW_MASK_EXT | PSW_MASK_IO | PSW_MASK_MCHECK)
4676static int __vcpu_run(struct kvm_vcpu *vcpu)
4677{
4678 int rc, exit_reason;
4679 struct sie_page *sie_page = (struct sie_page *)vcpu->arch.sie_block;
4680
4681 /*
4682 * We try to hold kvm->srcu during most of vcpu_run (except when run-
4683 * ning the guest), so that memslots (and other stuff) are protected
4684 */
4685 kvm_vcpu_srcu_read_lock(vcpu);
4686
4687 do {
4688 rc = vcpu_pre_run(vcpu);
4689 if (rc)
4690 break;
4691
4692 kvm_vcpu_srcu_read_unlock(vcpu);
4693 /*
4694 * As PF_VCPU will be used in fault handler, between
4695 * guest_enter and guest_exit should be no uaccess.
4696 */
4697 local_irq_disable();
4698 guest_enter_irqoff();
4699 __disable_cpu_timer_accounting(vcpu);
4700 local_irq_enable();
4701 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
4702 memcpy(sie_page->pv_grregs,
4703 vcpu->run->s.regs.gprs,
4704 sizeof(sie_page->pv_grregs));
4705 }
4706 if (test_cpu_flag(CIF_FPU))
4707 load_fpu_regs();
4708 exit_reason = sie64a(vcpu->arch.sie_block,
4709 vcpu->run->s.regs.gprs);
4710 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
4711 memcpy(vcpu->run->s.regs.gprs,
4712 sie_page->pv_grregs,
4713 sizeof(sie_page->pv_grregs));
4714 /*
4715 * We're not allowed to inject interrupts on intercepts
4716 * that leave the guest state in an "in-between" state
4717 * where the next SIE entry will do a continuation.
4718 * Fence interrupts in our "internal" PSW.
4719 */
4720 if (vcpu->arch.sie_block->icptcode == ICPT_PV_INSTR ||
4721 vcpu->arch.sie_block->icptcode == ICPT_PV_PREF) {
4722 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK;
4723 }
4724 }
4725 local_irq_disable();
4726 __enable_cpu_timer_accounting(vcpu);
4727 guest_exit_irqoff();
4728 local_irq_enable();
4729 kvm_vcpu_srcu_read_lock(vcpu);
4730
4731 rc = vcpu_post_run(vcpu, exit_reason);
4732 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
4733
4734 kvm_vcpu_srcu_read_unlock(vcpu);
4735 return rc;
4736}
4737
4738static void sync_regs_fmt2(struct kvm_vcpu *vcpu)
4739{
4740 struct kvm_run *kvm_run = vcpu->run;
4741 struct runtime_instr_cb *riccb;
4742 struct gs_cb *gscb;
4743
4744 riccb = (struct runtime_instr_cb *) &kvm_run->s.regs.riccb;
4745 gscb = (struct gs_cb *) &kvm_run->s.regs.gscb;
4746 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
4747 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
4748 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
4749 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
4750 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
4751 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
4752 }
4753 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
4754 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
4755 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
4756 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
4757 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4758 kvm_clear_async_pf_completion_queue(vcpu);
4759 }
4760 if (kvm_run->kvm_dirty_regs & KVM_SYNC_DIAG318) {
4761 vcpu->arch.diag318_info.val = kvm_run->s.regs.diag318;
4762 vcpu->arch.sie_block->cpnc = vcpu->arch.diag318_info.cpnc;
4763 VCPU_EVENT(vcpu, 3, "setting cpnc to %d", vcpu->arch.diag318_info.cpnc);
4764 }
4765 /*
4766 * If userspace sets the riccb (e.g. after migration) to a valid state,
4767 * we should enable RI here instead of doing the lazy enablement.
4768 */
4769 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) &&
4770 test_kvm_facility(vcpu->kvm, 64) &&
4771 riccb->v &&
4772 !(vcpu->arch.sie_block->ecb3 & ECB3_RI)) {
4773 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (sync_regs)");
4774 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
4775 }
4776 /*
4777 * If userspace sets the gscb (e.g. after migration) to non-zero,
4778 * we should enable GS here instead of doing the lazy enablement.
4779 */
4780 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_GSCB) &&
4781 test_kvm_facility(vcpu->kvm, 133) &&
4782 gscb->gssm &&
4783 !vcpu->arch.gs_enabled) {
4784 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (sync_regs)");
4785 vcpu->arch.sie_block->ecb |= ECB_GS;
4786 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
4787 vcpu->arch.gs_enabled = 1;
4788 }
4789 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_BPBC) &&
4790 test_kvm_facility(vcpu->kvm, 82)) {
4791 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
4792 vcpu->arch.sie_block->fpf |= kvm_run->s.regs.bpbc ? FPF_BPBC : 0;
4793 }
4794 if (MACHINE_HAS_GS) {
4795 preempt_disable();
4796 __ctl_set_bit(2, 4);
4797 if (current->thread.gs_cb) {
4798 vcpu->arch.host_gscb = current->thread.gs_cb;
4799 save_gs_cb(vcpu->arch.host_gscb);
4800 }
4801 if (vcpu->arch.gs_enabled) {
4802 current->thread.gs_cb = (struct gs_cb *)
4803 &vcpu->run->s.regs.gscb;
4804 restore_gs_cb(current->thread.gs_cb);
4805 }
4806 preempt_enable();
4807 }
4808 /* SIE will load etoken directly from SDNX and therefore kvm_run */
4809}
4810
4811static void sync_regs(struct kvm_vcpu *vcpu)
4812{
4813 struct kvm_run *kvm_run = vcpu->run;
4814
4815 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
4816 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
4817 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
4818 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
4819 /* some control register changes require a tlb flush */
4820 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
4821 }
4822 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
4823 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
4824 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
4825 }
4826 save_access_regs(vcpu->arch.host_acrs);
4827 restore_access_regs(vcpu->run->s.regs.acrs);
4828 /* save host (userspace) fprs/vrs */
4829 save_fpu_regs();
4830 vcpu->arch.host_fpregs.fpc = current->thread.fpu.fpc;
4831 vcpu->arch.host_fpregs.regs = current->thread.fpu.regs;
4832 if (MACHINE_HAS_VX)
4833 current->thread.fpu.regs = vcpu->run->s.regs.vrs;
4834 else
4835 current->thread.fpu.regs = vcpu->run->s.regs.fprs;
4836 current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
4837 if (test_fp_ctl(current->thread.fpu.fpc))
4838 /* User space provided an invalid FPC, let's clear it */
4839 current->thread.fpu.fpc = 0;
4840
4841 /* Sync fmt2 only data */
4842 if (likely(!kvm_s390_pv_cpu_is_protected(vcpu))) {
4843 sync_regs_fmt2(vcpu);
4844 } else {
4845 /*
4846 * In several places we have to modify our internal view to
4847 * not do things that are disallowed by the ultravisor. For
4848 * example we must not inject interrupts after specific exits
4849 * (e.g. 112 prefix page not secure). We do this by turning
4850 * off the machine check, external and I/O interrupt bits
4851 * of our PSW copy. To avoid getting validity intercepts, we
4852 * do only accept the condition code from userspace.
4853 */
4854 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_CC;
4855 vcpu->arch.sie_block->gpsw.mask |= kvm_run->psw_mask &
4856 PSW_MASK_CC;
4857 }
4858
4859 kvm_run->kvm_dirty_regs = 0;
4860}
4861
4862static void store_regs_fmt2(struct kvm_vcpu *vcpu)
4863{
4864 struct kvm_run *kvm_run = vcpu->run;
4865
4866 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
4867 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
4868 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
4869 kvm_run->s.regs.bpbc = (vcpu->arch.sie_block->fpf & FPF_BPBC) == FPF_BPBC;
4870 kvm_run->s.regs.diag318 = vcpu->arch.diag318_info.val;
4871 if (MACHINE_HAS_GS) {
4872 preempt_disable();
4873 __ctl_set_bit(2, 4);
4874 if (vcpu->arch.gs_enabled)
4875 save_gs_cb(current->thread.gs_cb);
4876 current->thread.gs_cb = vcpu->arch.host_gscb;
4877 restore_gs_cb(vcpu->arch.host_gscb);
4878 if (!vcpu->arch.host_gscb)
4879 __ctl_clear_bit(2, 4);
4880 vcpu->arch.host_gscb = NULL;
4881 preempt_enable();
4882 }
4883 /* SIE will save etoken directly into SDNX and therefore kvm_run */
4884}
4885
4886static void store_regs(struct kvm_vcpu *vcpu)
4887{
4888 struct kvm_run *kvm_run = vcpu->run;
4889
4890 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
4891 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
4892 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
4893 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
4894 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
4895 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
4896 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
4897 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
4898 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
4899 save_access_regs(vcpu->run->s.regs.acrs);
4900 restore_access_regs(vcpu->arch.host_acrs);
4901 /* Save guest register state */
4902 save_fpu_regs();
4903 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
4904 /* Restore will be done lazily at return */
4905 current->thread.fpu.fpc = vcpu->arch.host_fpregs.fpc;
4906 current->thread.fpu.regs = vcpu->arch.host_fpregs.regs;
4907 if (likely(!kvm_s390_pv_cpu_is_protected(vcpu)))
4908 store_regs_fmt2(vcpu);
4909}
4910
4911int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
4912{
4913 struct kvm_run *kvm_run = vcpu->run;
4914 int rc;
4915
4916 /*
4917 * Running a VM while dumping always has the potential to
4918 * produce inconsistent dump data. But for PV vcpus a SIE
4919 * entry while dumping could also lead to a fatal validity
4920 * intercept which we absolutely want to avoid.
4921 */
4922 if (vcpu->kvm->arch.pv.dumping)
4923 return -EINVAL;
4924
4925 if (kvm_run->immediate_exit)
4926 return -EINTR;
4927
4928 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_S390_VALID_FIELDS ||
4929 kvm_run->kvm_dirty_regs & ~KVM_SYNC_S390_VALID_FIELDS)
4930 return -EINVAL;
4931
4932 vcpu_load(vcpu);
4933
4934 if (guestdbg_exit_pending(vcpu)) {
4935 kvm_s390_prepare_debug_exit(vcpu);
4936 rc = 0;
4937 goto out;
4938 }
4939
4940 kvm_sigset_activate(vcpu);
4941
4942 /*
4943 * no need to check the return value of vcpu_start as it can only have
4944 * an error for protvirt, but protvirt means user cpu state
4945 */
4946 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
4947 kvm_s390_vcpu_start(vcpu);
4948 } else if (is_vcpu_stopped(vcpu)) {
4949 pr_err_ratelimited("can't run stopped vcpu %d\n",
4950 vcpu->vcpu_id);
4951 rc = -EINVAL;
4952 goto out;
4953 }
4954
4955 sync_regs(vcpu);
4956 enable_cpu_timer_accounting(vcpu);
4957
4958 might_fault();
4959 rc = __vcpu_run(vcpu);
4960
4961 if (signal_pending(current) && !rc) {
4962 kvm_run->exit_reason = KVM_EXIT_INTR;
4963 rc = -EINTR;
4964 }
4965
4966 if (guestdbg_exit_pending(vcpu) && !rc) {
4967 kvm_s390_prepare_debug_exit(vcpu);
4968 rc = 0;
4969 }
4970
4971 if (rc == -EREMOTE) {
4972 /* userspace support is needed, kvm_run has been prepared */
4973 rc = 0;
4974 }
4975
4976 disable_cpu_timer_accounting(vcpu);
4977 store_regs(vcpu);
4978
4979 kvm_sigset_deactivate(vcpu);
4980
4981 vcpu->stat.exit_userspace++;
4982out:
4983 vcpu_put(vcpu);
4984 return rc;
4985}
4986
4987/*
4988 * store status at address
4989 * we use have two special cases:
4990 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
4991 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
4992 */
4993int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
4994{
4995 unsigned char archmode = 1;
4996 freg_t fprs[NUM_FPRS];
4997 unsigned int px;
4998 u64 clkcomp, cputm;
4999 int rc;
5000
5001 px = kvm_s390_get_prefix(vcpu);
5002 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
5003 if (write_guest_abs(vcpu, 163, &archmode, 1))
5004 return -EFAULT;
5005 gpa = 0;
5006 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
5007 if (write_guest_real(vcpu, 163, &archmode, 1))
5008 return -EFAULT;
5009 gpa = px;
5010 } else
5011 gpa -= __LC_FPREGS_SAVE_AREA;
5012
5013 /* manually convert vector registers if necessary */
5014 if (MACHINE_HAS_VX) {
5015 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
5016 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
5017 fprs, 128);
5018 } else {
5019 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
5020 vcpu->run->s.regs.fprs, 128);
5021 }
5022 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
5023 vcpu->run->s.regs.gprs, 128);
5024 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
5025 &vcpu->arch.sie_block->gpsw, 16);
5026 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
5027 &px, 4);
5028 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
5029 &vcpu->run->s.regs.fpc, 4);
5030 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
5031 &vcpu->arch.sie_block->todpr, 4);
5032 cputm = kvm_s390_get_cpu_timer(vcpu);
5033 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
5034 &cputm, 8);
5035 clkcomp = vcpu->arch.sie_block->ckc >> 8;
5036 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
5037 &clkcomp, 8);
5038 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
5039 &vcpu->run->s.regs.acrs, 64);
5040 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
5041 &vcpu->arch.sie_block->gcr, 128);
5042 return rc ? -EFAULT : 0;
5043}
5044
5045int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
5046{
5047 /*
5048 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
5049 * switch in the run ioctl. Let's update our copies before we save
5050 * it into the save area
5051 */
5052 save_fpu_regs();
5053 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
5054 save_access_regs(vcpu->run->s.regs.acrs);
5055
5056 return kvm_s390_store_status_unloaded(vcpu, addr);
5057}
5058
5059static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
5060{
5061 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
5062 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
5063}
5064
5065static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
5066{
5067 unsigned long i;
5068 struct kvm_vcpu *vcpu;
5069
5070 kvm_for_each_vcpu(i, vcpu, kvm) {
5071 __disable_ibs_on_vcpu(vcpu);
5072 }
5073}
5074
5075static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
5076{
5077 if (!sclp.has_ibs)
5078 return;
5079 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
5080 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
5081}
5082
5083int kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
5084{
5085 int i, online_vcpus, r = 0, started_vcpus = 0;
5086
5087 if (!is_vcpu_stopped(vcpu))
5088 return 0;
5089
5090 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
5091 /* Only one cpu at a time may enter/leave the STOPPED state. */
5092 spin_lock(&vcpu->kvm->arch.start_stop_lock);
5093 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
5094
5095 /* Let's tell the UV that we want to change into the operating state */
5096 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5097 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR);
5098 if (r) {
5099 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5100 return r;
5101 }
5102 }
5103
5104 for (i = 0; i < online_vcpus; i++) {
5105 if (!is_vcpu_stopped(kvm_get_vcpu(vcpu->kvm, i)))
5106 started_vcpus++;
5107 }
5108
5109 if (started_vcpus == 0) {
5110 /* we're the only active VCPU -> speed it up */
5111 __enable_ibs_on_vcpu(vcpu);
5112 } else if (started_vcpus == 1) {
5113 /*
5114 * As we are starting a second VCPU, we have to disable
5115 * the IBS facility on all VCPUs to remove potentially
5116 * outstanding ENABLE requests.
5117 */
5118 __disable_ibs_on_all_vcpus(vcpu->kvm);
5119 }
5120
5121 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_STOPPED);
5122 /*
5123 * The real PSW might have changed due to a RESTART interpreted by the
5124 * ultravisor. We block all interrupts and let the next sie exit
5125 * refresh our view.
5126 */
5127 if (kvm_s390_pv_cpu_is_protected(vcpu))
5128 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK;
5129 /*
5130 * Another VCPU might have used IBS while we were offline.
5131 * Let's play safe and flush the VCPU at startup.
5132 */
5133 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5134 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5135 return 0;
5136}
5137
5138int kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
5139{
5140 int i, online_vcpus, r = 0, started_vcpus = 0;
5141 struct kvm_vcpu *started_vcpu = NULL;
5142
5143 if (is_vcpu_stopped(vcpu))
5144 return 0;
5145
5146 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
5147 /* Only one cpu at a time may enter/leave the STOPPED state. */
5148 spin_lock(&vcpu->kvm->arch.start_stop_lock);
5149 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
5150
5151 /* Let's tell the UV that we want to change into the stopped state */
5152 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5153 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_STP);
5154 if (r) {
5155 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5156 return r;
5157 }
5158 }
5159
5160 /*
5161 * Set the VCPU to STOPPED and THEN clear the interrupt flag,
5162 * now that the SIGP STOP and SIGP STOP AND STORE STATUS orders
5163 * have been fully processed. This will ensure that the VCPU
5164 * is kept BUSY if another VCPU is inquiring with SIGP SENSE.
5165 */
5166 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOPPED);
5167 kvm_s390_clear_stop_irq(vcpu);
5168
5169 __disable_ibs_on_vcpu(vcpu);
5170
5171 for (i = 0; i < online_vcpus; i++) {
5172 struct kvm_vcpu *tmp = kvm_get_vcpu(vcpu->kvm, i);
5173
5174 if (!is_vcpu_stopped(tmp)) {
5175 started_vcpus++;
5176 started_vcpu = tmp;
5177 }
5178 }
5179
5180 if (started_vcpus == 1) {
5181 /*
5182 * As we only have one VCPU left, we want to enable the
5183 * IBS facility for that VCPU to speed it up.
5184 */
5185 __enable_ibs_on_vcpu(started_vcpu);
5186 }
5187
5188 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5189 return 0;
5190}
5191
5192static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5193 struct kvm_enable_cap *cap)
5194{
5195 int r;
5196
5197 if (cap->flags)
5198 return -EINVAL;
5199
5200 switch (cap->cap) {
5201 case KVM_CAP_S390_CSS_SUPPORT:
5202 if (!vcpu->kvm->arch.css_support) {
5203 vcpu->kvm->arch.css_support = 1;
5204 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
5205 trace_kvm_s390_enable_css(vcpu->kvm);
5206 }
5207 r = 0;
5208 break;
5209 default:
5210 r = -EINVAL;
5211 break;
5212 }
5213 return r;
5214}
5215
5216static long kvm_s390_vcpu_sida_op(struct kvm_vcpu *vcpu,
5217 struct kvm_s390_mem_op *mop)
5218{
5219 void __user *uaddr = (void __user *)mop->buf;
5220 void *sida_addr;
5221 int r = 0;
5222
5223 if (mop->flags || !mop->size)
5224 return -EINVAL;
5225 if (mop->size + mop->sida_offset < mop->size)
5226 return -EINVAL;
5227 if (mop->size + mop->sida_offset > sida_size(vcpu->arch.sie_block))
5228 return -E2BIG;
5229 if (!kvm_s390_pv_cpu_is_protected(vcpu))
5230 return -EINVAL;
5231
5232 sida_addr = (char *)sida_addr(vcpu->arch.sie_block) + mop->sida_offset;
5233
5234 switch (mop->op) {
5235 case KVM_S390_MEMOP_SIDA_READ:
5236 if (copy_to_user(uaddr, sida_addr, mop->size))
5237 r = -EFAULT;
5238
5239 break;
5240 case KVM_S390_MEMOP_SIDA_WRITE:
5241 if (copy_from_user(sida_addr, uaddr, mop->size))
5242 r = -EFAULT;
5243 break;
5244 }
5245 return r;
5246}
5247
5248static long kvm_s390_vcpu_mem_op(struct kvm_vcpu *vcpu,
5249 struct kvm_s390_mem_op *mop)
5250{
5251 void __user *uaddr = (void __user *)mop->buf;
5252 void *tmpbuf = NULL;
5253 int r = 0;
5254 const u64 supported_flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION
5255 | KVM_S390_MEMOP_F_CHECK_ONLY
5256 | KVM_S390_MEMOP_F_SKEY_PROTECTION;
5257
5258 if (mop->flags & ~supported_flags || mop->ar >= NUM_ACRS || !mop->size)
5259 return -EINVAL;
5260 if (mop->size > MEM_OP_MAX_SIZE)
5261 return -E2BIG;
5262 if (kvm_s390_pv_cpu_is_protected(vcpu))
5263 return -EINVAL;
5264 if (mop->flags & KVM_S390_MEMOP_F_SKEY_PROTECTION) {
5265 if (access_key_invalid(mop->key))
5266 return -EINVAL;
5267 } else {
5268 mop->key = 0;
5269 }
5270 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
5271 tmpbuf = vmalloc(mop->size);
5272 if (!tmpbuf)
5273 return -ENOMEM;
5274 }
5275
5276 switch (mop->op) {
5277 case KVM_S390_MEMOP_LOGICAL_READ:
5278 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
5279 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size,
5280 GACC_FETCH, mop->key);
5281 break;
5282 }
5283 r = read_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf,
5284 mop->size, mop->key);
5285 if (r == 0) {
5286 if (copy_to_user(uaddr, tmpbuf, mop->size))
5287 r = -EFAULT;
5288 }
5289 break;
5290 case KVM_S390_MEMOP_LOGICAL_WRITE:
5291 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
5292 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size,
5293 GACC_STORE, mop->key);
5294 break;
5295 }
5296 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
5297 r = -EFAULT;
5298 break;
5299 }
5300 r = write_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf,
5301 mop->size, mop->key);
5302 break;
5303 }
5304
5305 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
5306 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
5307
5308 vfree(tmpbuf);
5309 return r;
5310}
5311
5312static long kvm_s390_vcpu_memsida_op(struct kvm_vcpu *vcpu,
5313 struct kvm_s390_mem_op *mop)
5314{
5315 int r, srcu_idx;
5316
5317 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5318
5319 switch (mop->op) {
5320 case KVM_S390_MEMOP_LOGICAL_READ:
5321 case KVM_S390_MEMOP_LOGICAL_WRITE:
5322 r = kvm_s390_vcpu_mem_op(vcpu, mop);
5323 break;
5324 case KVM_S390_MEMOP_SIDA_READ:
5325 case KVM_S390_MEMOP_SIDA_WRITE:
5326 /* we are locked against sida going away by the vcpu->mutex */
5327 r = kvm_s390_vcpu_sida_op(vcpu, mop);
5328 break;
5329 default:
5330 r = -EINVAL;
5331 }
5332
5333 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
5334 return r;
5335}
5336
5337long kvm_arch_vcpu_async_ioctl(struct file *filp,
5338 unsigned int ioctl, unsigned long arg)
5339{
5340 struct kvm_vcpu *vcpu = filp->private_data;
5341 void __user *argp = (void __user *)arg;
5342
5343 switch (ioctl) {
5344 case KVM_S390_IRQ: {
5345 struct kvm_s390_irq s390irq;
5346
5347 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
5348 return -EFAULT;
5349 return kvm_s390_inject_vcpu(vcpu, &s390irq);
5350 }
5351 case KVM_S390_INTERRUPT: {
5352 struct kvm_s390_interrupt s390int;
5353 struct kvm_s390_irq s390irq = {};
5354
5355 if (copy_from_user(&s390int, argp, sizeof(s390int)))
5356 return -EFAULT;
5357 if (s390int_to_s390irq(&s390int, &s390irq))
5358 return -EINVAL;
5359 return kvm_s390_inject_vcpu(vcpu, &s390irq);
5360 }
5361 }
5362 return -ENOIOCTLCMD;
5363}
5364
5365static int kvm_s390_handle_pv_vcpu_dump(struct kvm_vcpu *vcpu,
5366 struct kvm_pv_cmd *cmd)
5367{
5368 struct kvm_s390_pv_dmp dmp;
5369 void *data;
5370 int ret;
5371
5372 /* Dump initialization is a prerequisite */
5373 if (!vcpu->kvm->arch.pv.dumping)
5374 return -EINVAL;
5375
5376 if (copy_from_user(&dmp, (__u8 __user *)cmd->data, sizeof(dmp)))
5377 return -EFAULT;
5378
5379 /* We only handle this subcmd right now */
5380 if (dmp.subcmd != KVM_PV_DUMP_CPU)
5381 return -EINVAL;
5382
5383 /* CPU dump length is the same as create cpu storage donation. */
5384 if (dmp.buff_len != uv_info.guest_cpu_stor_len)
5385 return -EINVAL;
5386
5387 data = kvzalloc(uv_info.guest_cpu_stor_len, GFP_KERNEL);
5388 if (!data)
5389 return -ENOMEM;
5390
5391 ret = kvm_s390_pv_dump_cpu(vcpu, data, &cmd->rc, &cmd->rrc);
5392
5393 VCPU_EVENT(vcpu, 3, "PROTVIRT DUMP CPU %d rc %x rrc %x",
5394 vcpu->vcpu_id, cmd->rc, cmd->rrc);
5395
5396 if (ret)
5397 ret = -EINVAL;
5398
5399 /* On success copy over the dump data */
5400 if (!ret && copy_to_user((__u8 __user *)dmp.buff_addr, data, uv_info.guest_cpu_stor_len))
5401 ret = -EFAULT;
5402
5403 kvfree(data);
5404 return ret;
5405}
5406
5407long kvm_arch_vcpu_ioctl(struct file *filp,
5408 unsigned int ioctl, unsigned long arg)
5409{
5410 struct kvm_vcpu *vcpu = filp->private_data;
5411 void __user *argp = (void __user *)arg;
5412 int idx;
5413 long r;
5414 u16 rc, rrc;
5415
5416 vcpu_load(vcpu);
5417
5418 switch (ioctl) {
5419 case KVM_S390_STORE_STATUS:
5420 idx = srcu_read_lock(&vcpu->kvm->srcu);
5421 r = kvm_s390_store_status_unloaded(vcpu, arg);
5422 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5423 break;
5424 case KVM_S390_SET_INITIAL_PSW: {
5425 psw_t psw;
5426
5427 r = -EFAULT;
5428 if (copy_from_user(&psw, argp, sizeof(psw)))
5429 break;
5430 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
5431 break;
5432 }
5433 case KVM_S390_CLEAR_RESET:
5434 r = 0;
5435 kvm_arch_vcpu_ioctl_clear_reset(vcpu);
5436 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5437 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5438 UVC_CMD_CPU_RESET_CLEAR, &rc, &rrc);
5439 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET CLEAR VCPU: rc %x rrc %x",
5440 rc, rrc);
5441 }
5442 break;
5443 case KVM_S390_INITIAL_RESET:
5444 r = 0;
5445 kvm_arch_vcpu_ioctl_initial_reset(vcpu);
5446 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5447 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5448 UVC_CMD_CPU_RESET_INITIAL,
5449 &rc, &rrc);
5450 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET INITIAL VCPU: rc %x rrc %x",
5451 rc, rrc);
5452 }
5453 break;
5454 case KVM_S390_NORMAL_RESET:
5455 r = 0;
5456 kvm_arch_vcpu_ioctl_normal_reset(vcpu);
5457 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5458 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5459 UVC_CMD_CPU_RESET, &rc, &rrc);
5460 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET NORMAL VCPU: rc %x rrc %x",
5461 rc, rrc);
5462 }
5463 break;
5464 case KVM_SET_ONE_REG:
5465 case KVM_GET_ONE_REG: {
5466 struct kvm_one_reg reg;
5467 r = -EINVAL;
5468 if (kvm_s390_pv_cpu_is_protected(vcpu))
5469 break;
5470 r = -EFAULT;
5471 if (copy_from_user(®, argp, sizeof(reg)))
5472 break;
5473 if (ioctl == KVM_SET_ONE_REG)
5474 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®);
5475 else
5476 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®);
5477 break;
5478 }
5479#ifdef CONFIG_KVM_S390_UCONTROL
5480 case KVM_S390_UCAS_MAP: {
5481 struct kvm_s390_ucas_mapping ucasmap;
5482
5483 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
5484 r = -EFAULT;
5485 break;
5486 }
5487
5488 if (!kvm_is_ucontrol(vcpu->kvm)) {
5489 r = -EINVAL;
5490 break;
5491 }
5492
5493 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
5494 ucasmap.vcpu_addr, ucasmap.length);
5495 break;
5496 }
5497 case KVM_S390_UCAS_UNMAP: {
5498 struct kvm_s390_ucas_mapping ucasmap;
5499
5500 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
5501 r = -EFAULT;
5502 break;
5503 }
5504
5505 if (!kvm_is_ucontrol(vcpu->kvm)) {
5506 r = -EINVAL;
5507 break;
5508 }
5509
5510 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
5511 ucasmap.length);
5512 break;
5513 }
5514#endif
5515 case KVM_S390_VCPU_FAULT: {
5516 r = gmap_fault(vcpu->arch.gmap, arg, 0);
5517 break;
5518 }
5519 case KVM_ENABLE_CAP:
5520 {
5521 struct kvm_enable_cap cap;
5522 r = -EFAULT;
5523 if (copy_from_user(&cap, argp, sizeof(cap)))
5524 break;
5525 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5526 break;
5527 }
5528 case KVM_S390_MEM_OP: {
5529 struct kvm_s390_mem_op mem_op;
5530
5531 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
5532 r = kvm_s390_vcpu_memsida_op(vcpu, &mem_op);
5533 else
5534 r = -EFAULT;
5535 break;
5536 }
5537 case KVM_S390_SET_IRQ_STATE: {
5538 struct kvm_s390_irq_state irq_state;
5539
5540 r = -EFAULT;
5541 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
5542 break;
5543 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
5544 irq_state.len == 0 ||
5545 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
5546 r = -EINVAL;
5547 break;
5548 }
5549 /* do not use irq_state.flags, it will break old QEMUs */
5550 r = kvm_s390_set_irq_state(vcpu,
5551 (void __user *) irq_state.buf,
5552 irq_state.len);
5553 break;
5554 }
5555 case KVM_S390_GET_IRQ_STATE: {
5556 struct kvm_s390_irq_state irq_state;
5557
5558 r = -EFAULT;
5559 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
5560 break;
5561 if (irq_state.len == 0) {
5562 r = -EINVAL;
5563 break;
5564 }
5565 /* do not use irq_state.flags, it will break old QEMUs */
5566 r = kvm_s390_get_irq_state(vcpu,
5567 (__u8 __user *) irq_state.buf,
5568 irq_state.len);
5569 break;
5570 }
5571 case KVM_S390_PV_CPU_COMMAND: {
5572 struct kvm_pv_cmd cmd;
5573
5574 r = -EINVAL;
5575 if (!is_prot_virt_host())
5576 break;
5577
5578 r = -EFAULT;
5579 if (copy_from_user(&cmd, argp, sizeof(cmd)))
5580 break;
5581
5582 r = -EINVAL;
5583 if (cmd.flags)
5584 break;
5585
5586 /* We only handle this cmd right now */
5587 if (cmd.cmd != KVM_PV_DUMP)
5588 break;
5589
5590 r = kvm_s390_handle_pv_vcpu_dump(vcpu, &cmd);
5591
5592 /* Always copy over UV rc / rrc data */
5593 if (copy_to_user((__u8 __user *)argp, &cmd.rc,
5594 sizeof(cmd.rc) + sizeof(cmd.rrc)))
5595 r = -EFAULT;
5596 break;
5597 }
5598 default:
5599 r = -ENOTTY;
5600 }
5601
5602 vcpu_put(vcpu);
5603 return r;
5604}
5605
5606vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5607{
5608#ifdef CONFIG_KVM_S390_UCONTROL
5609 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
5610 && (kvm_is_ucontrol(vcpu->kvm))) {
5611 vmf->page = virt_to_page(vcpu->arch.sie_block);
5612 get_page(vmf->page);
5613 return 0;
5614 }
5615#endif
5616 return VM_FAULT_SIGBUS;
5617}
5618
5619bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
5620{
5621 return true;
5622}
5623
5624/* Section: memory related */
5625int kvm_arch_prepare_memory_region(struct kvm *kvm,
5626 const struct kvm_memory_slot *old,
5627 struct kvm_memory_slot *new,
5628 enum kvm_mr_change change)
5629{
5630 gpa_t size;
5631
5632 /* When we are protected, we should not change the memory slots */
5633 if (kvm_s390_pv_get_handle(kvm))
5634 return -EINVAL;
5635
5636 if (change == KVM_MR_DELETE || change == KVM_MR_FLAGS_ONLY)
5637 return 0;
5638
5639 /* A few sanity checks. We can have memory slots which have to be
5640 located/ended at a segment boundary (1MB). The memory in userland is
5641 ok to be fragmented into various different vmas. It is okay to mmap()
5642 and munmap() stuff in this slot after doing this call at any time */
5643
5644 if (new->userspace_addr & 0xffffful)
5645 return -EINVAL;
5646
5647 size = new->npages * PAGE_SIZE;
5648 if (size & 0xffffful)
5649 return -EINVAL;
5650
5651 if ((new->base_gfn * PAGE_SIZE) + size > kvm->arch.mem_limit)
5652 return -EINVAL;
5653
5654 return 0;
5655}
5656
5657void kvm_arch_commit_memory_region(struct kvm *kvm,
5658 struct kvm_memory_slot *old,
5659 const struct kvm_memory_slot *new,
5660 enum kvm_mr_change change)
5661{
5662 int rc = 0;
5663
5664 switch (change) {
5665 case KVM_MR_DELETE:
5666 rc = gmap_unmap_segment(kvm->arch.gmap, old->base_gfn * PAGE_SIZE,
5667 old->npages * PAGE_SIZE);
5668 break;
5669 case KVM_MR_MOVE:
5670 rc = gmap_unmap_segment(kvm->arch.gmap, old->base_gfn * PAGE_SIZE,
5671 old->npages * PAGE_SIZE);
5672 if (rc)
5673 break;
5674 fallthrough;
5675 case KVM_MR_CREATE:
5676 rc = gmap_map_segment(kvm->arch.gmap, new->userspace_addr,
5677 new->base_gfn * PAGE_SIZE,
5678 new->npages * PAGE_SIZE);
5679 break;
5680 case KVM_MR_FLAGS_ONLY:
5681 break;
5682 default:
5683 WARN(1, "Unknown KVM MR CHANGE: %d\n", change);
5684 }
5685 if (rc)
5686 pr_warn("failed to commit memory region\n");
5687 return;
5688}
5689
5690static inline unsigned long nonhyp_mask(int i)
5691{
5692 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
5693
5694 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
5695}
5696
5697static int __init kvm_s390_init(void)
5698{
5699 int i;
5700
5701 if (!sclp.has_sief2) {
5702 pr_info("SIE is not available\n");
5703 return -ENODEV;
5704 }
5705
5706 if (nested && hpage) {
5707 pr_info("A KVM host that supports nesting cannot back its KVM guests with huge pages\n");
5708 return -EINVAL;
5709 }
5710
5711 for (i = 0; i < 16; i++)
5712 kvm_s390_fac_base[i] |=
5713 stfle_fac_list[i] & nonhyp_mask(i);
5714
5715 return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
5716}
5717
5718static void __exit kvm_s390_exit(void)
5719{
5720 kvm_exit();
5721}
5722
5723module_init(kvm_s390_init);
5724module_exit(kvm_s390_exit);
5725
5726/*
5727 * Enable autoloading of the kvm module.
5728 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
5729 * since x86 takes a different approach.
5730 */
5731#include <linux/miscdevice.h>
5732MODULE_ALIAS_MISCDEV(KVM_MINOR);
5733MODULE_ALIAS("devname:kvm");
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * hosting IBM Z kernel virtual machines (s390x)
4 *
5 * Copyright IBM Corp. 2008, 2018
6 *
7 * Author(s): Carsten Otte <cotte@de.ibm.com>
8 * Christian Borntraeger <borntraeger@de.ibm.com>
9 * Heiko Carstens <heiko.carstens@de.ibm.com>
10 * Christian Ehrhardt <ehrhardt@de.ibm.com>
11 * Jason J. Herne <jjherne@us.ibm.com>
12 */
13
14#include <linux/compiler.h>
15#include <linux/err.h>
16#include <linux/fs.h>
17#include <linux/hrtimer.h>
18#include <linux/init.h>
19#include <linux/kvm.h>
20#include <linux/kvm_host.h>
21#include <linux/mman.h>
22#include <linux/module.h>
23#include <linux/moduleparam.h>
24#include <linux/random.h>
25#include <linux/slab.h>
26#include <linux/timer.h>
27#include <linux/vmalloc.h>
28#include <linux/bitmap.h>
29#include <linux/sched/signal.h>
30#include <linux/string.h>
31
32#include <asm/asm-offsets.h>
33#include <asm/lowcore.h>
34#include <asm/stp.h>
35#include <asm/pgtable.h>
36#include <asm/gmap.h>
37#include <asm/nmi.h>
38#include <asm/switch_to.h>
39#include <asm/isc.h>
40#include <asm/sclp.h>
41#include <asm/cpacf.h>
42#include <asm/timex.h>
43#include "kvm-s390.h"
44#include "gaccess.h"
45
46#define KMSG_COMPONENT "kvm-s390"
47#undef pr_fmt
48#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
49
50#define CREATE_TRACE_POINTS
51#include "trace.h"
52#include "trace-s390.h"
53
54#define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */
55#define LOCAL_IRQS 32
56#define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
57 (KVM_MAX_VCPUS + LOCAL_IRQS))
58
59#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
60#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
61
62struct kvm_stats_debugfs_item debugfs_entries[] = {
63 { "userspace_handled", VCPU_STAT(exit_userspace) },
64 { "exit_null", VCPU_STAT(exit_null) },
65 { "exit_validity", VCPU_STAT(exit_validity) },
66 { "exit_stop_request", VCPU_STAT(exit_stop_request) },
67 { "exit_external_request", VCPU_STAT(exit_external_request) },
68 { "exit_io_request", VCPU_STAT(exit_io_request) },
69 { "exit_external_interrupt", VCPU_STAT(exit_external_interrupt) },
70 { "exit_instruction", VCPU_STAT(exit_instruction) },
71 { "exit_pei", VCPU_STAT(exit_pei) },
72 { "exit_program_interruption", VCPU_STAT(exit_program_interruption) },
73 { "exit_instr_and_program_int", VCPU_STAT(exit_instr_and_program) },
74 { "exit_operation_exception", VCPU_STAT(exit_operation_exception) },
75 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
76 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
77 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
78 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
79 { "instruction_lctlg", VCPU_STAT(instruction_lctlg) },
80 { "instruction_lctl", VCPU_STAT(instruction_lctl) },
81 { "instruction_stctl", VCPU_STAT(instruction_stctl) },
82 { "instruction_stctg", VCPU_STAT(instruction_stctg) },
83 { "deliver_ckc", VCPU_STAT(deliver_ckc) },
84 { "deliver_cputm", VCPU_STAT(deliver_cputm) },
85 { "deliver_emergency_signal", VCPU_STAT(deliver_emergency_signal) },
86 { "deliver_external_call", VCPU_STAT(deliver_external_call) },
87 { "deliver_service_signal", VCPU_STAT(deliver_service_signal) },
88 { "deliver_virtio", VCPU_STAT(deliver_virtio) },
89 { "deliver_stop_signal", VCPU_STAT(deliver_stop_signal) },
90 { "deliver_prefix_signal", VCPU_STAT(deliver_prefix_signal) },
91 { "deliver_restart_signal", VCPU_STAT(deliver_restart_signal) },
92 { "deliver_program", VCPU_STAT(deliver_program) },
93 { "deliver_io", VCPU_STAT(deliver_io) },
94 { "deliver_machine_check", VCPU_STAT(deliver_machine_check) },
95 { "exit_wait_state", VCPU_STAT(exit_wait_state) },
96 { "inject_ckc", VCPU_STAT(inject_ckc) },
97 { "inject_cputm", VCPU_STAT(inject_cputm) },
98 { "inject_external_call", VCPU_STAT(inject_external_call) },
99 { "inject_float_mchk", VM_STAT(inject_float_mchk) },
100 { "inject_emergency_signal", VCPU_STAT(inject_emergency_signal) },
101 { "inject_io", VM_STAT(inject_io) },
102 { "inject_mchk", VCPU_STAT(inject_mchk) },
103 { "inject_pfault_done", VM_STAT(inject_pfault_done) },
104 { "inject_program", VCPU_STAT(inject_program) },
105 { "inject_restart", VCPU_STAT(inject_restart) },
106 { "inject_service_signal", VM_STAT(inject_service_signal) },
107 { "inject_set_prefix", VCPU_STAT(inject_set_prefix) },
108 { "inject_stop_signal", VCPU_STAT(inject_stop_signal) },
109 { "inject_pfault_init", VCPU_STAT(inject_pfault_init) },
110 { "inject_virtio", VM_STAT(inject_virtio) },
111 { "instruction_epsw", VCPU_STAT(instruction_epsw) },
112 { "instruction_gs", VCPU_STAT(instruction_gs) },
113 { "instruction_io_other", VCPU_STAT(instruction_io_other) },
114 { "instruction_lpsw", VCPU_STAT(instruction_lpsw) },
115 { "instruction_lpswe", VCPU_STAT(instruction_lpswe) },
116 { "instruction_pfmf", VCPU_STAT(instruction_pfmf) },
117 { "instruction_ptff", VCPU_STAT(instruction_ptff) },
118 { "instruction_stidp", VCPU_STAT(instruction_stidp) },
119 { "instruction_sck", VCPU_STAT(instruction_sck) },
120 { "instruction_sckpf", VCPU_STAT(instruction_sckpf) },
121 { "instruction_spx", VCPU_STAT(instruction_spx) },
122 { "instruction_stpx", VCPU_STAT(instruction_stpx) },
123 { "instruction_stap", VCPU_STAT(instruction_stap) },
124 { "instruction_iske", VCPU_STAT(instruction_iske) },
125 { "instruction_ri", VCPU_STAT(instruction_ri) },
126 { "instruction_rrbe", VCPU_STAT(instruction_rrbe) },
127 { "instruction_sske", VCPU_STAT(instruction_sske) },
128 { "instruction_ipte_interlock", VCPU_STAT(instruction_ipte_interlock) },
129 { "instruction_essa", VCPU_STAT(instruction_essa) },
130 { "instruction_stsi", VCPU_STAT(instruction_stsi) },
131 { "instruction_stfl", VCPU_STAT(instruction_stfl) },
132 { "instruction_tb", VCPU_STAT(instruction_tb) },
133 { "instruction_tpi", VCPU_STAT(instruction_tpi) },
134 { "instruction_tprot", VCPU_STAT(instruction_tprot) },
135 { "instruction_tsch", VCPU_STAT(instruction_tsch) },
136 { "instruction_sthyi", VCPU_STAT(instruction_sthyi) },
137 { "instruction_sie", VCPU_STAT(instruction_sie) },
138 { "instruction_sigp_sense", VCPU_STAT(instruction_sigp_sense) },
139 { "instruction_sigp_sense_running", VCPU_STAT(instruction_sigp_sense_running) },
140 { "instruction_sigp_external_call", VCPU_STAT(instruction_sigp_external_call) },
141 { "instruction_sigp_emergency", VCPU_STAT(instruction_sigp_emergency) },
142 { "instruction_sigp_cond_emergency", VCPU_STAT(instruction_sigp_cond_emergency) },
143 { "instruction_sigp_start", VCPU_STAT(instruction_sigp_start) },
144 { "instruction_sigp_stop", VCPU_STAT(instruction_sigp_stop) },
145 { "instruction_sigp_stop_store_status", VCPU_STAT(instruction_sigp_stop_store_status) },
146 { "instruction_sigp_store_status", VCPU_STAT(instruction_sigp_store_status) },
147 { "instruction_sigp_store_adtl_status", VCPU_STAT(instruction_sigp_store_adtl_status) },
148 { "instruction_sigp_set_arch", VCPU_STAT(instruction_sigp_arch) },
149 { "instruction_sigp_set_prefix", VCPU_STAT(instruction_sigp_prefix) },
150 { "instruction_sigp_restart", VCPU_STAT(instruction_sigp_restart) },
151 { "instruction_sigp_cpu_reset", VCPU_STAT(instruction_sigp_cpu_reset) },
152 { "instruction_sigp_init_cpu_reset", VCPU_STAT(instruction_sigp_init_cpu_reset) },
153 { "instruction_sigp_unknown", VCPU_STAT(instruction_sigp_unknown) },
154 { "instruction_diag_10", VCPU_STAT(diagnose_10) },
155 { "instruction_diag_44", VCPU_STAT(diagnose_44) },
156 { "instruction_diag_9c", VCPU_STAT(diagnose_9c) },
157 { "instruction_diag_258", VCPU_STAT(diagnose_258) },
158 { "instruction_diag_308", VCPU_STAT(diagnose_308) },
159 { "instruction_diag_500", VCPU_STAT(diagnose_500) },
160 { "instruction_diag_other", VCPU_STAT(diagnose_other) },
161 { NULL }
162};
163
164struct kvm_s390_tod_clock_ext {
165 __u8 epoch_idx;
166 __u64 tod;
167 __u8 reserved[7];
168} __packed;
169
170/* allow nested virtualization in KVM (if enabled by user space) */
171static int nested;
172module_param(nested, int, S_IRUGO);
173MODULE_PARM_DESC(nested, "Nested virtualization support");
174
175
176/*
177 * For now we handle at most 16 double words as this is what the s390 base
178 * kernel handles and stores in the prefix page. If we ever need to go beyond
179 * this, this requires changes to code, but the external uapi can stay.
180 */
181#define SIZE_INTERNAL 16
182
183/*
184 * Base feature mask that defines default mask for facilities. Consists of the
185 * defines in FACILITIES_KVM and the non-hypervisor managed bits.
186 */
187static unsigned long kvm_s390_fac_base[SIZE_INTERNAL] = { FACILITIES_KVM };
188/*
189 * Extended feature mask. Consists of the defines in FACILITIES_KVM_CPUMODEL
190 * and defines the facilities that can be enabled via a cpu model.
191 */
192static unsigned long kvm_s390_fac_ext[SIZE_INTERNAL] = { FACILITIES_KVM_CPUMODEL };
193
194static unsigned long kvm_s390_fac_size(void)
195{
196 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_MASK_SIZE_U64);
197 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_LIST_SIZE_U64);
198 BUILD_BUG_ON(SIZE_INTERNAL * sizeof(unsigned long) >
199 sizeof(S390_lowcore.stfle_fac_list));
200
201 return SIZE_INTERNAL;
202}
203
204/* available cpu features supported by kvm */
205static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
206/* available subfunctions indicated via query / "test bit" */
207static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
208
209static struct gmap_notifier gmap_notifier;
210static struct gmap_notifier vsie_gmap_notifier;
211debug_info_t *kvm_s390_dbf;
212
213/* Section: not file related */
214int kvm_arch_hardware_enable(void)
215{
216 /* every s390 is virtualization enabled ;-) */
217 return 0;
218}
219
220static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
221 unsigned long end);
222
223static void kvm_clock_sync_scb(struct kvm_s390_sie_block *scb, u64 delta)
224{
225 u8 delta_idx = 0;
226
227 /*
228 * The TOD jumps by delta, we have to compensate this by adding
229 * -delta to the epoch.
230 */
231 delta = -delta;
232
233 /* sign-extension - we're adding to signed values below */
234 if ((s64)delta < 0)
235 delta_idx = -1;
236
237 scb->epoch += delta;
238 if (scb->ecd & ECD_MEF) {
239 scb->epdx += delta_idx;
240 if (scb->epoch < delta)
241 scb->epdx += 1;
242 }
243}
244
245/*
246 * This callback is executed during stop_machine(). All CPUs are therefore
247 * temporarily stopped. In order not to change guest behavior, we have to
248 * disable preemption whenever we touch the epoch of kvm and the VCPUs,
249 * so a CPU won't be stopped while calculating with the epoch.
250 */
251static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
252 void *v)
253{
254 struct kvm *kvm;
255 struct kvm_vcpu *vcpu;
256 int i;
257 unsigned long long *delta = v;
258
259 list_for_each_entry(kvm, &vm_list, vm_list) {
260 kvm_for_each_vcpu(i, vcpu, kvm) {
261 kvm_clock_sync_scb(vcpu->arch.sie_block, *delta);
262 if (i == 0) {
263 kvm->arch.epoch = vcpu->arch.sie_block->epoch;
264 kvm->arch.epdx = vcpu->arch.sie_block->epdx;
265 }
266 if (vcpu->arch.cputm_enabled)
267 vcpu->arch.cputm_start += *delta;
268 if (vcpu->arch.vsie_block)
269 kvm_clock_sync_scb(vcpu->arch.vsie_block,
270 *delta);
271 }
272 }
273 return NOTIFY_OK;
274}
275
276static struct notifier_block kvm_clock_notifier = {
277 .notifier_call = kvm_clock_sync,
278};
279
280int kvm_arch_hardware_setup(void)
281{
282 gmap_notifier.notifier_call = kvm_gmap_notifier;
283 gmap_register_pte_notifier(&gmap_notifier);
284 vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
285 gmap_register_pte_notifier(&vsie_gmap_notifier);
286 atomic_notifier_chain_register(&s390_epoch_delta_notifier,
287 &kvm_clock_notifier);
288 return 0;
289}
290
291void kvm_arch_hardware_unsetup(void)
292{
293 gmap_unregister_pte_notifier(&gmap_notifier);
294 gmap_unregister_pte_notifier(&vsie_gmap_notifier);
295 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
296 &kvm_clock_notifier);
297}
298
299static void allow_cpu_feat(unsigned long nr)
300{
301 set_bit_inv(nr, kvm_s390_available_cpu_feat);
302}
303
304static inline int plo_test_bit(unsigned char nr)
305{
306 register unsigned long r0 asm("0") = (unsigned long) nr | 0x100;
307 int cc;
308
309 asm volatile(
310 /* Parameter registers are ignored for "test bit" */
311 " plo 0,0,0,0(0)\n"
312 " ipm %0\n"
313 " srl %0,28\n"
314 : "=d" (cc)
315 : "d" (r0)
316 : "cc");
317 return cc == 0;
318}
319
320static void kvm_s390_cpu_feat_init(void)
321{
322 int i;
323
324 for (i = 0; i < 256; ++i) {
325 if (plo_test_bit(i))
326 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
327 }
328
329 if (test_facility(28)) /* TOD-clock steering */
330 ptff(kvm_s390_available_subfunc.ptff,
331 sizeof(kvm_s390_available_subfunc.ptff),
332 PTFF_QAF);
333
334 if (test_facility(17)) { /* MSA */
335 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
336 kvm_s390_available_subfunc.kmac);
337 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
338 kvm_s390_available_subfunc.kmc);
339 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
340 kvm_s390_available_subfunc.km);
341 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
342 kvm_s390_available_subfunc.kimd);
343 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
344 kvm_s390_available_subfunc.klmd);
345 }
346 if (test_facility(76)) /* MSA3 */
347 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
348 kvm_s390_available_subfunc.pckmo);
349 if (test_facility(77)) { /* MSA4 */
350 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
351 kvm_s390_available_subfunc.kmctr);
352 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
353 kvm_s390_available_subfunc.kmf);
354 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
355 kvm_s390_available_subfunc.kmo);
356 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
357 kvm_s390_available_subfunc.pcc);
358 }
359 if (test_facility(57)) /* MSA5 */
360 __cpacf_query(CPACF_PRNO, (cpacf_mask_t *)
361 kvm_s390_available_subfunc.ppno);
362
363 if (test_facility(146)) /* MSA8 */
364 __cpacf_query(CPACF_KMA, (cpacf_mask_t *)
365 kvm_s390_available_subfunc.kma);
366
367 if (MACHINE_HAS_ESOP)
368 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
369 /*
370 * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
371 * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
372 */
373 if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
374 !test_facility(3) || !nested)
375 return;
376 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
377 if (sclp.has_64bscao)
378 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
379 if (sclp.has_siif)
380 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
381 if (sclp.has_gpere)
382 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
383 if (sclp.has_gsls)
384 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
385 if (sclp.has_ib)
386 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
387 if (sclp.has_cei)
388 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
389 if (sclp.has_ibs)
390 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
391 if (sclp.has_kss)
392 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_KSS);
393 /*
394 * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
395 * all skey handling functions read/set the skey from the PGSTE
396 * instead of the real storage key.
397 *
398 * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
399 * pages being detected as preserved although they are resident.
400 *
401 * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
402 * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
403 *
404 * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
405 * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
406 * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
407 *
408 * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
409 * cannot easily shadow the SCA because of the ipte lock.
410 */
411}
412
413int kvm_arch_init(void *opaque)
414{
415 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
416 if (!kvm_s390_dbf)
417 return -ENOMEM;
418
419 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view)) {
420 debug_unregister(kvm_s390_dbf);
421 return -ENOMEM;
422 }
423
424 kvm_s390_cpu_feat_init();
425
426 /* Register floating interrupt controller interface. */
427 return kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
428}
429
430void kvm_arch_exit(void)
431{
432 debug_unregister(kvm_s390_dbf);
433}
434
435/* Section: device related */
436long kvm_arch_dev_ioctl(struct file *filp,
437 unsigned int ioctl, unsigned long arg)
438{
439 if (ioctl == KVM_S390_ENABLE_SIE)
440 return s390_enable_sie();
441 return -EINVAL;
442}
443
444int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
445{
446 int r;
447
448 switch (ext) {
449 case KVM_CAP_S390_PSW:
450 case KVM_CAP_S390_GMAP:
451 case KVM_CAP_SYNC_MMU:
452#ifdef CONFIG_KVM_S390_UCONTROL
453 case KVM_CAP_S390_UCONTROL:
454#endif
455 case KVM_CAP_ASYNC_PF:
456 case KVM_CAP_SYNC_REGS:
457 case KVM_CAP_ONE_REG:
458 case KVM_CAP_ENABLE_CAP:
459 case KVM_CAP_S390_CSS_SUPPORT:
460 case KVM_CAP_IOEVENTFD:
461 case KVM_CAP_DEVICE_CTRL:
462 case KVM_CAP_ENABLE_CAP_VM:
463 case KVM_CAP_S390_IRQCHIP:
464 case KVM_CAP_VM_ATTRIBUTES:
465 case KVM_CAP_MP_STATE:
466 case KVM_CAP_IMMEDIATE_EXIT:
467 case KVM_CAP_S390_INJECT_IRQ:
468 case KVM_CAP_S390_USER_SIGP:
469 case KVM_CAP_S390_USER_STSI:
470 case KVM_CAP_S390_SKEYS:
471 case KVM_CAP_S390_IRQ_STATE:
472 case KVM_CAP_S390_USER_INSTR0:
473 case KVM_CAP_S390_CMMA_MIGRATION:
474 case KVM_CAP_S390_AIS:
475 case KVM_CAP_S390_AIS_MIGRATION:
476 r = 1;
477 break;
478 case KVM_CAP_S390_MEM_OP:
479 r = MEM_OP_MAX_SIZE;
480 break;
481 case KVM_CAP_NR_VCPUS:
482 case KVM_CAP_MAX_VCPUS:
483 r = KVM_S390_BSCA_CPU_SLOTS;
484 if (!kvm_s390_use_sca_entries())
485 r = KVM_MAX_VCPUS;
486 else if (sclp.has_esca && sclp.has_64bscao)
487 r = KVM_S390_ESCA_CPU_SLOTS;
488 break;
489 case KVM_CAP_NR_MEMSLOTS:
490 r = KVM_USER_MEM_SLOTS;
491 break;
492 case KVM_CAP_S390_COW:
493 r = MACHINE_HAS_ESOP;
494 break;
495 case KVM_CAP_S390_VECTOR_REGISTERS:
496 r = MACHINE_HAS_VX;
497 break;
498 case KVM_CAP_S390_RI:
499 r = test_facility(64);
500 break;
501 case KVM_CAP_S390_GS:
502 r = test_facility(133);
503 break;
504 case KVM_CAP_S390_BPB:
505 r = test_facility(82);
506 break;
507 default:
508 r = 0;
509 }
510 return r;
511}
512
513static void kvm_s390_sync_dirty_log(struct kvm *kvm,
514 struct kvm_memory_slot *memslot)
515{
516 gfn_t cur_gfn, last_gfn;
517 unsigned long address;
518 struct gmap *gmap = kvm->arch.gmap;
519
520 /* Loop over all guest pages */
521 last_gfn = memslot->base_gfn + memslot->npages;
522 for (cur_gfn = memslot->base_gfn; cur_gfn <= last_gfn; cur_gfn++) {
523 address = gfn_to_hva_memslot(memslot, cur_gfn);
524
525 if (test_and_clear_guest_dirty(gmap->mm, address))
526 mark_page_dirty(kvm, cur_gfn);
527 if (fatal_signal_pending(current))
528 return;
529 cond_resched();
530 }
531}
532
533/* Section: vm related */
534static void sca_del_vcpu(struct kvm_vcpu *vcpu);
535
536/*
537 * Get (and clear) the dirty memory log for a memory slot.
538 */
539int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
540 struct kvm_dirty_log *log)
541{
542 int r;
543 unsigned long n;
544 struct kvm_memslots *slots;
545 struct kvm_memory_slot *memslot;
546 int is_dirty = 0;
547
548 if (kvm_is_ucontrol(kvm))
549 return -EINVAL;
550
551 mutex_lock(&kvm->slots_lock);
552
553 r = -EINVAL;
554 if (log->slot >= KVM_USER_MEM_SLOTS)
555 goto out;
556
557 slots = kvm_memslots(kvm);
558 memslot = id_to_memslot(slots, log->slot);
559 r = -ENOENT;
560 if (!memslot->dirty_bitmap)
561 goto out;
562
563 kvm_s390_sync_dirty_log(kvm, memslot);
564 r = kvm_get_dirty_log(kvm, log, &is_dirty);
565 if (r)
566 goto out;
567
568 /* Clear the dirty log */
569 if (is_dirty) {
570 n = kvm_dirty_bitmap_bytes(memslot);
571 memset(memslot->dirty_bitmap, 0, n);
572 }
573 r = 0;
574out:
575 mutex_unlock(&kvm->slots_lock);
576 return r;
577}
578
579static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
580{
581 unsigned int i;
582 struct kvm_vcpu *vcpu;
583
584 kvm_for_each_vcpu(i, vcpu, kvm) {
585 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
586 }
587}
588
589static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
590{
591 int r;
592
593 if (cap->flags)
594 return -EINVAL;
595
596 switch (cap->cap) {
597 case KVM_CAP_S390_IRQCHIP:
598 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
599 kvm->arch.use_irqchip = 1;
600 r = 0;
601 break;
602 case KVM_CAP_S390_USER_SIGP:
603 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
604 kvm->arch.user_sigp = 1;
605 r = 0;
606 break;
607 case KVM_CAP_S390_VECTOR_REGISTERS:
608 mutex_lock(&kvm->lock);
609 if (kvm->created_vcpus) {
610 r = -EBUSY;
611 } else if (MACHINE_HAS_VX) {
612 set_kvm_facility(kvm->arch.model.fac_mask, 129);
613 set_kvm_facility(kvm->arch.model.fac_list, 129);
614 if (test_facility(134)) {
615 set_kvm_facility(kvm->arch.model.fac_mask, 134);
616 set_kvm_facility(kvm->arch.model.fac_list, 134);
617 }
618 if (test_facility(135)) {
619 set_kvm_facility(kvm->arch.model.fac_mask, 135);
620 set_kvm_facility(kvm->arch.model.fac_list, 135);
621 }
622 r = 0;
623 } else
624 r = -EINVAL;
625 mutex_unlock(&kvm->lock);
626 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
627 r ? "(not available)" : "(success)");
628 break;
629 case KVM_CAP_S390_RI:
630 r = -EINVAL;
631 mutex_lock(&kvm->lock);
632 if (kvm->created_vcpus) {
633 r = -EBUSY;
634 } else if (test_facility(64)) {
635 set_kvm_facility(kvm->arch.model.fac_mask, 64);
636 set_kvm_facility(kvm->arch.model.fac_list, 64);
637 r = 0;
638 }
639 mutex_unlock(&kvm->lock);
640 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
641 r ? "(not available)" : "(success)");
642 break;
643 case KVM_CAP_S390_AIS:
644 mutex_lock(&kvm->lock);
645 if (kvm->created_vcpus) {
646 r = -EBUSY;
647 } else {
648 set_kvm_facility(kvm->arch.model.fac_mask, 72);
649 set_kvm_facility(kvm->arch.model.fac_list, 72);
650 r = 0;
651 }
652 mutex_unlock(&kvm->lock);
653 VM_EVENT(kvm, 3, "ENABLE: AIS %s",
654 r ? "(not available)" : "(success)");
655 break;
656 case KVM_CAP_S390_GS:
657 r = -EINVAL;
658 mutex_lock(&kvm->lock);
659 if (kvm->created_vcpus) {
660 r = -EBUSY;
661 } else if (test_facility(133)) {
662 set_kvm_facility(kvm->arch.model.fac_mask, 133);
663 set_kvm_facility(kvm->arch.model.fac_list, 133);
664 r = 0;
665 }
666 mutex_unlock(&kvm->lock);
667 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_GS %s",
668 r ? "(not available)" : "(success)");
669 break;
670 case KVM_CAP_S390_USER_STSI:
671 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
672 kvm->arch.user_stsi = 1;
673 r = 0;
674 break;
675 case KVM_CAP_S390_USER_INSTR0:
676 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
677 kvm->arch.user_instr0 = 1;
678 icpt_operexc_on_all_vcpus(kvm);
679 r = 0;
680 break;
681 default:
682 r = -EINVAL;
683 break;
684 }
685 return r;
686}
687
688static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
689{
690 int ret;
691
692 switch (attr->attr) {
693 case KVM_S390_VM_MEM_LIMIT_SIZE:
694 ret = 0;
695 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
696 kvm->arch.mem_limit);
697 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
698 ret = -EFAULT;
699 break;
700 default:
701 ret = -ENXIO;
702 break;
703 }
704 return ret;
705}
706
707static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
708{
709 int ret;
710 unsigned int idx;
711 switch (attr->attr) {
712 case KVM_S390_VM_MEM_ENABLE_CMMA:
713 ret = -ENXIO;
714 if (!sclp.has_cmma)
715 break;
716
717 ret = -EBUSY;
718 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
719 mutex_lock(&kvm->lock);
720 if (!kvm->created_vcpus) {
721 kvm->arch.use_cmma = 1;
722 /* Not compatible with cmma. */
723 kvm->arch.use_pfmfi = 0;
724 ret = 0;
725 }
726 mutex_unlock(&kvm->lock);
727 break;
728 case KVM_S390_VM_MEM_CLR_CMMA:
729 ret = -ENXIO;
730 if (!sclp.has_cmma)
731 break;
732 ret = -EINVAL;
733 if (!kvm->arch.use_cmma)
734 break;
735
736 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
737 mutex_lock(&kvm->lock);
738 idx = srcu_read_lock(&kvm->srcu);
739 s390_reset_cmma(kvm->arch.gmap->mm);
740 srcu_read_unlock(&kvm->srcu, idx);
741 mutex_unlock(&kvm->lock);
742 ret = 0;
743 break;
744 case KVM_S390_VM_MEM_LIMIT_SIZE: {
745 unsigned long new_limit;
746
747 if (kvm_is_ucontrol(kvm))
748 return -EINVAL;
749
750 if (get_user(new_limit, (u64 __user *)attr->addr))
751 return -EFAULT;
752
753 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
754 new_limit > kvm->arch.mem_limit)
755 return -E2BIG;
756
757 if (!new_limit)
758 return -EINVAL;
759
760 /* gmap_create takes last usable address */
761 if (new_limit != KVM_S390_NO_MEM_LIMIT)
762 new_limit -= 1;
763
764 ret = -EBUSY;
765 mutex_lock(&kvm->lock);
766 if (!kvm->created_vcpus) {
767 /* gmap_create will round the limit up */
768 struct gmap *new = gmap_create(current->mm, new_limit);
769
770 if (!new) {
771 ret = -ENOMEM;
772 } else {
773 gmap_remove(kvm->arch.gmap);
774 new->private = kvm;
775 kvm->arch.gmap = new;
776 ret = 0;
777 }
778 }
779 mutex_unlock(&kvm->lock);
780 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
781 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
782 (void *) kvm->arch.gmap->asce);
783 break;
784 }
785 default:
786 ret = -ENXIO;
787 break;
788 }
789 return ret;
790}
791
792static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
793
794static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
795{
796 struct kvm_vcpu *vcpu;
797 int i;
798
799 if (!test_kvm_facility(kvm, 76))
800 return -EINVAL;
801
802 mutex_lock(&kvm->lock);
803 switch (attr->attr) {
804 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
805 get_random_bytes(
806 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
807 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
808 kvm->arch.crypto.aes_kw = 1;
809 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
810 break;
811 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
812 get_random_bytes(
813 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
814 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
815 kvm->arch.crypto.dea_kw = 1;
816 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
817 break;
818 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
819 kvm->arch.crypto.aes_kw = 0;
820 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
821 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
822 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
823 break;
824 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
825 kvm->arch.crypto.dea_kw = 0;
826 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
827 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
828 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
829 break;
830 default:
831 mutex_unlock(&kvm->lock);
832 return -ENXIO;
833 }
834
835 kvm_for_each_vcpu(i, vcpu, kvm) {
836 kvm_s390_vcpu_crypto_setup(vcpu);
837 exit_sie(vcpu);
838 }
839 mutex_unlock(&kvm->lock);
840 return 0;
841}
842
843static void kvm_s390_sync_request_broadcast(struct kvm *kvm, int req)
844{
845 int cx;
846 struct kvm_vcpu *vcpu;
847
848 kvm_for_each_vcpu(cx, vcpu, kvm)
849 kvm_s390_sync_request(req, vcpu);
850}
851
852/*
853 * Must be called with kvm->srcu held to avoid races on memslots, and with
854 * kvm->slots_lock to avoid races with ourselves and kvm_s390_vm_stop_migration.
855 */
856static int kvm_s390_vm_start_migration(struct kvm *kvm)
857{
858 struct kvm_s390_migration_state *mgs;
859 struct kvm_memory_slot *ms;
860 /* should be the only one */
861 struct kvm_memslots *slots;
862 unsigned long ram_pages;
863 int slotnr;
864
865 /* migration mode already enabled */
866 if (kvm->arch.migration_state)
867 return 0;
868
869 slots = kvm_memslots(kvm);
870 if (!slots || !slots->used_slots)
871 return -EINVAL;
872
873 mgs = kzalloc(sizeof(*mgs), GFP_KERNEL);
874 if (!mgs)
875 return -ENOMEM;
876 kvm->arch.migration_state = mgs;
877
878 if (kvm->arch.use_cmma) {
879 /*
880 * Get the first slot. They are reverse sorted by base_gfn, so
881 * the first slot is also the one at the end of the address
882 * space. We have verified above that at least one slot is
883 * present.
884 */
885 ms = slots->memslots;
886 /* round up so we only use full longs */
887 ram_pages = roundup(ms->base_gfn + ms->npages, BITS_PER_LONG);
888 /* allocate enough bytes to store all the bits */
889 mgs->pgste_bitmap = vmalloc(ram_pages / 8);
890 if (!mgs->pgste_bitmap) {
891 kfree(mgs);
892 kvm->arch.migration_state = NULL;
893 return -ENOMEM;
894 }
895
896 mgs->bitmap_size = ram_pages;
897 atomic64_set(&mgs->dirty_pages, ram_pages);
898 /* mark all the pages in active slots as dirty */
899 for (slotnr = 0; slotnr < slots->used_slots; slotnr++) {
900 ms = slots->memslots + slotnr;
901 bitmap_set(mgs->pgste_bitmap, ms->base_gfn, ms->npages);
902 }
903
904 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_START_MIGRATION);
905 }
906 return 0;
907}
908
909/*
910 * Must be called with kvm->slots_lock to avoid races with ourselves and
911 * kvm_s390_vm_start_migration.
912 */
913static int kvm_s390_vm_stop_migration(struct kvm *kvm)
914{
915 struct kvm_s390_migration_state *mgs;
916
917 /* migration mode already disabled */
918 if (!kvm->arch.migration_state)
919 return 0;
920 mgs = kvm->arch.migration_state;
921 kvm->arch.migration_state = NULL;
922
923 if (kvm->arch.use_cmma) {
924 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_STOP_MIGRATION);
925 /* We have to wait for the essa emulation to finish */
926 synchronize_srcu(&kvm->srcu);
927 vfree(mgs->pgste_bitmap);
928 }
929 kfree(mgs);
930 return 0;
931}
932
933static int kvm_s390_vm_set_migration(struct kvm *kvm,
934 struct kvm_device_attr *attr)
935{
936 int res = -ENXIO;
937
938 mutex_lock(&kvm->slots_lock);
939 switch (attr->attr) {
940 case KVM_S390_VM_MIGRATION_START:
941 res = kvm_s390_vm_start_migration(kvm);
942 break;
943 case KVM_S390_VM_MIGRATION_STOP:
944 res = kvm_s390_vm_stop_migration(kvm);
945 break;
946 default:
947 break;
948 }
949 mutex_unlock(&kvm->slots_lock);
950
951 return res;
952}
953
954static int kvm_s390_vm_get_migration(struct kvm *kvm,
955 struct kvm_device_attr *attr)
956{
957 u64 mig = (kvm->arch.migration_state != NULL);
958
959 if (attr->attr != KVM_S390_VM_MIGRATION_STATUS)
960 return -ENXIO;
961
962 if (copy_to_user((void __user *)attr->addr, &mig, sizeof(mig)))
963 return -EFAULT;
964 return 0;
965}
966
967static int kvm_s390_set_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
968{
969 struct kvm_s390_vm_tod_clock gtod;
970
971 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod)))
972 return -EFAULT;
973
974 if (!test_kvm_facility(kvm, 139) && gtod.epoch_idx)
975 return -EINVAL;
976 kvm_s390_set_tod_clock(kvm, >od);
977
978 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x, TOD base: 0x%llx",
979 gtod.epoch_idx, gtod.tod);
980
981 return 0;
982}
983
984static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
985{
986 u8 gtod_high;
987
988 if (copy_from_user(>od_high, (void __user *)attr->addr,
989 sizeof(gtod_high)))
990 return -EFAULT;
991
992 if (gtod_high != 0)
993 return -EINVAL;
994 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
995
996 return 0;
997}
998
999static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1000{
1001 struct kvm_s390_vm_tod_clock gtod = { 0 };
1002
1003 if (copy_from_user(>od.tod, (void __user *)attr->addr,
1004 sizeof(gtod.tod)))
1005 return -EFAULT;
1006
1007 kvm_s390_set_tod_clock(kvm, >od);
1008 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod.tod);
1009 return 0;
1010}
1011
1012static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1013{
1014 int ret;
1015
1016 if (attr->flags)
1017 return -EINVAL;
1018
1019 switch (attr->attr) {
1020 case KVM_S390_VM_TOD_EXT:
1021 ret = kvm_s390_set_tod_ext(kvm, attr);
1022 break;
1023 case KVM_S390_VM_TOD_HIGH:
1024 ret = kvm_s390_set_tod_high(kvm, attr);
1025 break;
1026 case KVM_S390_VM_TOD_LOW:
1027 ret = kvm_s390_set_tod_low(kvm, attr);
1028 break;
1029 default:
1030 ret = -ENXIO;
1031 break;
1032 }
1033 return ret;
1034}
1035
1036static void kvm_s390_get_tod_clock_ext(struct kvm *kvm,
1037 struct kvm_s390_vm_tod_clock *gtod)
1038{
1039 struct kvm_s390_tod_clock_ext htod;
1040
1041 preempt_disable();
1042
1043 get_tod_clock_ext((char *)&htod);
1044
1045 gtod->tod = htod.tod + kvm->arch.epoch;
1046 gtod->epoch_idx = htod.epoch_idx + kvm->arch.epdx;
1047
1048 if (gtod->tod < htod.tod)
1049 gtod->epoch_idx += 1;
1050
1051 preempt_enable();
1052}
1053
1054static int kvm_s390_get_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1055{
1056 struct kvm_s390_vm_tod_clock gtod;
1057
1058 memset(>od, 0, sizeof(gtod));
1059
1060 if (test_kvm_facility(kvm, 139))
1061 kvm_s390_get_tod_clock_ext(kvm, >od);
1062 else
1063 gtod.tod = kvm_s390_get_tod_clock_fast(kvm);
1064
1065 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
1066 return -EFAULT;
1067
1068 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x, TOD base: 0x%llx",
1069 gtod.epoch_idx, gtod.tod);
1070 return 0;
1071}
1072
1073static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1074{
1075 u8 gtod_high = 0;
1076
1077 if (copy_to_user((void __user *)attr->addr, >od_high,
1078 sizeof(gtod_high)))
1079 return -EFAULT;
1080 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
1081
1082 return 0;
1083}
1084
1085static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1086{
1087 u64 gtod;
1088
1089 gtod = kvm_s390_get_tod_clock_fast(kvm);
1090 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
1091 return -EFAULT;
1092 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
1093
1094 return 0;
1095}
1096
1097static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1098{
1099 int ret;
1100
1101 if (attr->flags)
1102 return -EINVAL;
1103
1104 switch (attr->attr) {
1105 case KVM_S390_VM_TOD_EXT:
1106 ret = kvm_s390_get_tod_ext(kvm, attr);
1107 break;
1108 case KVM_S390_VM_TOD_HIGH:
1109 ret = kvm_s390_get_tod_high(kvm, attr);
1110 break;
1111 case KVM_S390_VM_TOD_LOW:
1112 ret = kvm_s390_get_tod_low(kvm, attr);
1113 break;
1114 default:
1115 ret = -ENXIO;
1116 break;
1117 }
1118 return ret;
1119}
1120
1121static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1122{
1123 struct kvm_s390_vm_cpu_processor *proc;
1124 u16 lowest_ibc, unblocked_ibc;
1125 int ret = 0;
1126
1127 mutex_lock(&kvm->lock);
1128 if (kvm->created_vcpus) {
1129 ret = -EBUSY;
1130 goto out;
1131 }
1132 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
1133 if (!proc) {
1134 ret = -ENOMEM;
1135 goto out;
1136 }
1137 if (!copy_from_user(proc, (void __user *)attr->addr,
1138 sizeof(*proc))) {
1139 kvm->arch.model.cpuid = proc->cpuid;
1140 lowest_ibc = sclp.ibc >> 16 & 0xfff;
1141 unblocked_ibc = sclp.ibc & 0xfff;
1142 if (lowest_ibc && proc->ibc) {
1143 if (proc->ibc > unblocked_ibc)
1144 kvm->arch.model.ibc = unblocked_ibc;
1145 else if (proc->ibc < lowest_ibc)
1146 kvm->arch.model.ibc = lowest_ibc;
1147 else
1148 kvm->arch.model.ibc = proc->ibc;
1149 }
1150 memcpy(kvm->arch.model.fac_list, proc->fac_list,
1151 S390_ARCH_FAC_LIST_SIZE_BYTE);
1152 VM_EVENT(kvm, 3, "SET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1153 kvm->arch.model.ibc,
1154 kvm->arch.model.cpuid);
1155 VM_EVENT(kvm, 3, "SET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1156 kvm->arch.model.fac_list[0],
1157 kvm->arch.model.fac_list[1],
1158 kvm->arch.model.fac_list[2]);
1159 } else
1160 ret = -EFAULT;
1161 kfree(proc);
1162out:
1163 mutex_unlock(&kvm->lock);
1164 return ret;
1165}
1166
1167static int kvm_s390_set_processor_feat(struct kvm *kvm,
1168 struct kvm_device_attr *attr)
1169{
1170 struct kvm_s390_vm_cpu_feat data;
1171
1172 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
1173 return -EFAULT;
1174 if (!bitmap_subset((unsigned long *) data.feat,
1175 kvm_s390_available_cpu_feat,
1176 KVM_S390_VM_CPU_FEAT_NR_BITS))
1177 return -EINVAL;
1178
1179 mutex_lock(&kvm->lock);
1180 if (kvm->created_vcpus) {
1181 mutex_unlock(&kvm->lock);
1182 return -EBUSY;
1183 }
1184 bitmap_copy(kvm->arch.cpu_feat, (unsigned long *) data.feat,
1185 KVM_S390_VM_CPU_FEAT_NR_BITS);
1186 mutex_unlock(&kvm->lock);
1187 VM_EVENT(kvm, 3, "SET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1188 data.feat[0],
1189 data.feat[1],
1190 data.feat[2]);
1191 return 0;
1192}
1193
1194static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
1195 struct kvm_device_attr *attr)
1196{
1197 /*
1198 * Once supported by kernel + hw, we have to store the subfunctions
1199 * in kvm->arch and remember that user space configured them.
1200 */
1201 return -ENXIO;
1202}
1203
1204static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1205{
1206 int ret = -ENXIO;
1207
1208 switch (attr->attr) {
1209 case KVM_S390_VM_CPU_PROCESSOR:
1210 ret = kvm_s390_set_processor(kvm, attr);
1211 break;
1212 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1213 ret = kvm_s390_set_processor_feat(kvm, attr);
1214 break;
1215 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1216 ret = kvm_s390_set_processor_subfunc(kvm, attr);
1217 break;
1218 }
1219 return ret;
1220}
1221
1222static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1223{
1224 struct kvm_s390_vm_cpu_processor *proc;
1225 int ret = 0;
1226
1227 proc = kzalloc(sizeof(*proc), GFP_KERNEL);
1228 if (!proc) {
1229 ret = -ENOMEM;
1230 goto out;
1231 }
1232 proc->cpuid = kvm->arch.model.cpuid;
1233 proc->ibc = kvm->arch.model.ibc;
1234 memcpy(&proc->fac_list, kvm->arch.model.fac_list,
1235 S390_ARCH_FAC_LIST_SIZE_BYTE);
1236 VM_EVENT(kvm, 3, "GET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1237 kvm->arch.model.ibc,
1238 kvm->arch.model.cpuid);
1239 VM_EVENT(kvm, 3, "GET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1240 kvm->arch.model.fac_list[0],
1241 kvm->arch.model.fac_list[1],
1242 kvm->arch.model.fac_list[2]);
1243 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
1244 ret = -EFAULT;
1245 kfree(proc);
1246out:
1247 return ret;
1248}
1249
1250static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
1251{
1252 struct kvm_s390_vm_cpu_machine *mach;
1253 int ret = 0;
1254
1255 mach = kzalloc(sizeof(*mach), GFP_KERNEL);
1256 if (!mach) {
1257 ret = -ENOMEM;
1258 goto out;
1259 }
1260 get_cpu_id((struct cpuid *) &mach->cpuid);
1261 mach->ibc = sclp.ibc;
1262 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
1263 S390_ARCH_FAC_LIST_SIZE_BYTE);
1264 memcpy((unsigned long *)&mach->fac_list, S390_lowcore.stfle_fac_list,
1265 sizeof(S390_lowcore.stfle_fac_list));
1266 VM_EVENT(kvm, 3, "GET: host ibc: 0x%4.4x, host cpuid: 0x%16.16llx",
1267 kvm->arch.model.ibc,
1268 kvm->arch.model.cpuid);
1269 VM_EVENT(kvm, 3, "GET: host facmask: 0x%16.16llx.%16.16llx.%16.16llx",
1270 mach->fac_mask[0],
1271 mach->fac_mask[1],
1272 mach->fac_mask[2]);
1273 VM_EVENT(kvm, 3, "GET: host faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1274 mach->fac_list[0],
1275 mach->fac_list[1],
1276 mach->fac_list[2]);
1277 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
1278 ret = -EFAULT;
1279 kfree(mach);
1280out:
1281 return ret;
1282}
1283
1284static int kvm_s390_get_processor_feat(struct kvm *kvm,
1285 struct kvm_device_attr *attr)
1286{
1287 struct kvm_s390_vm_cpu_feat data;
1288
1289 bitmap_copy((unsigned long *) data.feat, kvm->arch.cpu_feat,
1290 KVM_S390_VM_CPU_FEAT_NR_BITS);
1291 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1292 return -EFAULT;
1293 VM_EVENT(kvm, 3, "GET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1294 data.feat[0],
1295 data.feat[1],
1296 data.feat[2]);
1297 return 0;
1298}
1299
1300static int kvm_s390_get_machine_feat(struct kvm *kvm,
1301 struct kvm_device_attr *attr)
1302{
1303 struct kvm_s390_vm_cpu_feat data;
1304
1305 bitmap_copy((unsigned long *) data.feat,
1306 kvm_s390_available_cpu_feat,
1307 KVM_S390_VM_CPU_FEAT_NR_BITS);
1308 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1309 return -EFAULT;
1310 VM_EVENT(kvm, 3, "GET: host feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1311 data.feat[0],
1312 data.feat[1],
1313 data.feat[2]);
1314 return 0;
1315}
1316
1317static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
1318 struct kvm_device_attr *attr)
1319{
1320 /*
1321 * Once we can actually configure subfunctions (kernel + hw support),
1322 * we have to check if they were already set by user space, if so copy
1323 * them from kvm->arch.
1324 */
1325 return -ENXIO;
1326}
1327
1328static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
1329 struct kvm_device_attr *attr)
1330{
1331 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
1332 sizeof(struct kvm_s390_vm_cpu_subfunc)))
1333 return -EFAULT;
1334 return 0;
1335}
1336static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1337{
1338 int ret = -ENXIO;
1339
1340 switch (attr->attr) {
1341 case KVM_S390_VM_CPU_PROCESSOR:
1342 ret = kvm_s390_get_processor(kvm, attr);
1343 break;
1344 case KVM_S390_VM_CPU_MACHINE:
1345 ret = kvm_s390_get_machine(kvm, attr);
1346 break;
1347 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1348 ret = kvm_s390_get_processor_feat(kvm, attr);
1349 break;
1350 case KVM_S390_VM_CPU_MACHINE_FEAT:
1351 ret = kvm_s390_get_machine_feat(kvm, attr);
1352 break;
1353 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1354 ret = kvm_s390_get_processor_subfunc(kvm, attr);
1355 break;
1356 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1357 ret = kvm_s390_get_machine_subfunc(kvm, attr);
1358 break;
1359 }
1360 return ret;
1361}
1362
1363static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1364{
1365 int ret;
1366
1367 switch (attr->group) {
1368 case KVM_S390_VM_MEM_CTRL:
1369 ret = kvm_s390_set_mem_control(kvm, attr);
1370 break;
1371 case KVM_S390_VM_TOD:
1372 ret = kvm_s390_set_tod(kvm, attr);
1373 break;
1374 case KVM_S390_VM_CPU_MODEL:
1375 ret = kvm_s390_set_cpu_model(kvm, attr);
1376 break;
1377 case KVM_S390_VM_CRYPTO:
1378 ret = kvm_s390_vm_set_crypto(kvm, attr);
1379 break;
1380 case KVM_S390_VM_MIGRATION:
1381 ret = kvm_s390_vm_set_migration(kvm, attr);
1382 break;
1383 default:
1384 ret = -ENXIO;
1385 break;
1386 }
1387
1388 return ret;
1389}
1390
1391static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1392{
1393 int ret;
1394
1395 switch (attr->group) {
1396 case KVM_S390_VM_MEM_CTRL:
1397 ret = kvm_s390_get_mem_control(kvm, attr);
1398 break;
1399 case KVM_S390_VM_TOD:
1400 ret = kvm_s390_get_tod(kvm, attr);
1401 break;
1402 case KVM_S390_VM_CPU_MODEL:
1403 ret = kvm_s390_get_cpu_model(kvm, attr);
1404 break;
1405 case KVM_S390_VM_MIGRATION:
1406 ret = kvm_s390_vm_get_migration(kvm, attr);
1407 break;
1408 default:
1409 ret = -ENXIO;
1410 break;
1411 }
1412
1413 return ret;
1414}
1415
1416static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1417{
1418 int ret;
1419
1420 switch (attr->group) {
1421 case KVM_S390_VM_MEM_CTRL:
1422 switch (attr->attr) {
1423 case KVM_S390_VM_MEM_ENABLE_CMMA:
1424 case KVM_S390_VM_MEM_CLR_CMMA:
1425 ret = sclp.has_cmma ? 0 : -ENXIO;
1426 break;
1427 case KVM_S390_VM_MEM_LIMIT_SIZE:
1428 ret = 0;
1429 break;
1430 default:
1431 ret = -ENXIO;
1432 break;
1433 }
1434 break;
1435 case KVM_S390_VM_TOD:
1436 switch (attr->attr) {
1437 case KVM_S390_VM_TOD_LOW:
1438 case KVM_S390_VM_TOD_HIGH:
1439 ret = 0;
1440 break;
1441 default:
1442 ret = -ENXIO;
1443 break;
1444 }
1445 break;
1446 case KVM_S390_VM_CPU_MODEL:
1447 switch (attr->attr) {
1448 case KVM_S390_VM_CPU_PROCESSOR:
1449 case KVM_S390_VM_CPU_MACHINE:
1450 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1451 case KVM_S390_VM_CPU_MACHINE_FEAT:
1452 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1453 ret = 0;
1454 break;
1455 /* configuring subfunctions is not supported yet */
1456 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1457 default:
1458 ret = -ENXIO;
1459 break;
1460 }
1461 break;
1462 case KVM_S390_VM_CRYPTO:
1463 switch (attr->attr) {
1464 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1465 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1466 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1467 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1468 ret = 0;
1469 break;
1470 default:
1471 ret = -ENXIO;
1472 break;
1473 }
1474 break;
1475 case KVM_S390_VM_MIGRATION:
1476 ret = 0;
1477 break;
1478 default:
1479 ret = -ENXIO;
1480 break;
1481 }
1482
1483 return ret;
1484}
1485
1486static long kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1487{
1488 uint8_t *keys;
1489 uint64_t hva;
1490 int srcu_idx, i, r = 0;
1491
1492 if (args->flags != 0)
1493 return -EINVAL;
1494
1495 /* Is this guest using storage keys? */
1496 if (!mm_use_skey(current->mm))
1497 return KVM_S390_GET_SKEYS_NONE;
1498
1499 /* Enforce sane limit on memory allocation */
1500 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1501 return -EINVAL;
1502
1503 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL);
1504 if (!keys)
1505 return -ENOMEM;
1506
1507 down_read(¤t->mm->mmap_sem);
1508 srcu_idx = srcu_read_lock(&kvm->srcu);
1509 for (i = 0; i < args->count; i++) {
1510 hva = gfn_to_hva(kvm, args->start_gfn + i);
1511 if (kvm_is_error_hva(hva)) {
1512 r = -EFAULT;
1513 break;
1514 }
1515
1516 r = get_guest_storage_key(current->mm, hva, &keys[i]);
1517 if (r)
1518 break;
1519 }
1520 srcu_read_unlock(&kvm->srcu, srcu_idx);
1521 up_read(¤t->mm->mmap_sem);
1522
1523 if (!r) {
1524 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
1525 sizeof(uint8_t) * args->count);
1526 if (r)
1527 r = -EFAULT;
1528 }
1529
1530 kvfree(keys);
1531 return r;
1532}
1533
1534static long kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
1535{
1536 uint8_t *keys;
1537 uint64_t hva;
1538 int srcu_idx, i, r = 0;
1539
1540 if (args->flags != 0)
1541 return -EINVAL;
1542
1543 /* Enforce sane limit on memory allocation */
1544 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
1545 return -EINVAL;
1546
1547 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL);
1548 if (!keys)
1549 return -ENOMEM;
1550
1551 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
1552 sizeof(uint8_t) * args->count);
1553 if (r) {
1554 r = -EFAULT;
1555 goto out;
1556 }
1557
1558 /* Enable storage key handling for the guest */
1559 r = s390_enable_skey();
1560 if (r)
1561 goto out;
1562
1563 down_read(¤t->mm->mmap_sem);
1564 srcu_idx = srcu_read_lock(&kvm->srcu);
1565 for (i = 0; i < args->count; i++) {
1566 hva = gfn_to_hva(kvm, args->start_gfn + i);
1567 if (kvm_is_error_hva(hva)) {
1568 r = -EFAULT;
1569 break;
1570 }
1571
1572 /* Lowest order bit is reserved */
1573 if (keys[i] & 0x01) {
1574 r = -EINVAL;
1575 break;
1576 }
1577
1578 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
1579 if (r)
1580 break;
1581 }
1582 srcu_read_unlock(&kvm->srcu, srcu_idx);
1583 up_read(¤t->mm->mmap_sem);
1584out:
1585 kvfree(keys);
1586 return r;
1587}
1588
1589/*
1590 * Base address and length must be sent at the start of each block, therefore
1591 * it's cheaper to send some clean data, as long as it's less than the size of
1592 * two longs.
1593 */
1594#define KVM_S390_MAX_BIT_DISTANCE (2 * sizeof(void *))
1595/* for consistency */
1596#define KVM_S390_CMMA_SIZE_MAX ((u32)KVM_S390_SKEYS_MAX)
1597
1598/*
1599 * This function searches for the next page with dirty CMMA attributes, and
1600 * saves the attributes in the buffer up to either the end of the buffer or
1601 * until a block of at least KVM_S390_MAX_BIT_DISTANCE clean bits is found;
1602 * no trailing clean bytes are saved.
1603 * In case no dirty bits were found, or if CMMA was not enabled or used, the
1604 * output buffer will indicate 0 as length.
1605 */
1606static int kvm_s390_get_cmma_bits(struct kvm *kvm,
1607 struct kvm_s390_cmma_log *args)
1608{
1609 struct kvm_s390_migration_state *s = kvm->arch.migration_state;
1610 unsigned long bufsize, hva, pgstev, i, next, cur;
1611 int srcu_idx, peek, r = 0, rr;
1612 u8 *res;
1613
1614 cur = args->start_gfn;
1615 i = next = pgstev = 0;
1616
1617 if (unlikely(!kvm->arch.use_cmma))
1618 return -ENXIO;
1619 /* Invalid/unsupported flags were specified */
1620 if (args->flags & ~KVM_S390_CMMA_PEEK)
1621 return -EINVAL;
1622 /* Migration mode query, and we are not doing a migration */
1623 peek = !!(args->flags & KVM_S390_CMMA_PEEK);
1624 if (!peek && !s)
1625 return -EINVAL;
1626 /* CMMA is disabled or was not used, or the buffer has length zero */
1627 bufsize = min(args->count, KVM_S390_CMMA_SIZE_MAX);
1628 if (!bufsize || !kvm->mm->context.uses_cmm) {
1629 memset(args, 0, sizeof(*args));
1630 return 0;
1631 }
1632
1633 if (!peek) {
1634 /* We are not peeking, and there are no dirty pages */
1635 if (!atomic64_read(&s->dirty_pages)) {
1636 memset(args, 0, sizeof(*args));
1637 return 0;
1638 }
1639 cur = find_next_bit(s->pgste_bitmap, s->bitmap_size,
1640 args->start_gfn);
1641 if (cur >= s->bitmap_size) /* nothing found, loop back */
1642 cur = find_next_bit(s->pgste_bitmap, s->bitmap_size, 0);
1643 if (cur >= s->bitmap_size) { /* again! (very unlikely) */
1644 memset(args, 0, sizeof(*args));
1645 return 0;
1646 }
1647 next = find_next_bit(s->pgste_bitmap, s->bitmap_size, cur + 1);
1648 }
1649
1650 res = vmalloc(bufsize);
1651 if (!res)
1652 return -ENOMEM;
1653
1654 args->start_gfn = cur;
1655
1656 down_read(&kvm->mm->mmap_sem);
1657 srcu_idx = srcu_read_lock(&kvm->srcu);
1658 while (i < bufsize) {
1659 hva = gfn_to_hva(kvm, cur);
1660 if (kvm_is_error_hva(hva)) {
1661 r = -EFAULT;
1662 break;
1663 }
1664 /* decrement only if we actually flipped the bit to 0 */
1665 if (!peek && test_and_clear_bit(cur, s->pgste_bitmap))
1666 atomic64_dec(&s->dirty_pages);
1667 r = get_pgste(kvm->mm, hva, &pgstev);
1668 if (r < 0)
1669 pgstev = 0;
1670 /* save the value */
1671 res[i++] = (pgstev >> 24) & 0x43;
1672 /*
1673 * if the next bit is too far away, stop.
1674 * if we reached the previous "next", find the next one
1675 */
1676 if (!peek) {
1677 if (next > cur + KVM_S390_MAX_BIT_DISTANCE)
1678 break;
1679 if (cur == next)
1680 next = find_next_bit(s->pgste_bitmap,
1681 s->bitmap_size, cur + 1);
1682 /* reached the end of the bitmap or of the buffer, stop */
1683 if ((next >= s->bitmap_size) ||
1684 (next >= args->start_gfn + bufsize))
1685 break;
1686 }
1687 cur++;
1688 }
1689 srcu_read_unlock(&kvm->srcu, srcu_idx);
1690 up_read(&kvm->mm->mmap_sem);
1691 args->count = i;
1692 args->remaining = s ? atomic64_read(&s->dirty_pages) : 0;
1693
1694 rr = copy_to_user((void __user *)args->values, res, args->count);
1695 if (rr)
1696 r = -EFAULT;
1697
1698 vfree(res);
1699 return r;
1700}
1701
1702/*
1703 * This function sets the CMMA attributes for the given pages. If the input
1704 * buffer has zero length, no action is taken, otherwise the attributes are
1705 * set and the mm->context.uses_cmm flag is set.
1706 */
1707static int kvm_s390_set_cmma_bits(struct kvm *kvm,
1708 const struct kvm_s390_cmma_log *args)
1709{
1710 unsigned long hva, mask, pgstev, i;
1711 uint8_t *bits;
1712 int srcu_idx, r = 0;
1713
1714 mask = args->mask;
1715
1716 if (!kvm->arch.use_cmma)
1717 return -ENXIO;
1718 /* invalid/unsupported flags */
1719 if (args->flags != 0)
1720 return -EINVAL;
1721 /* Enforce sane limit on memory allocation */
1722 if (args->count > KVM_S390_CMMA_SIZE_MAX)
1723 return -EINVAL;
1724 /* Nothing to do */
1725 if (args->count == 0)
1726 return 0;
1727
1728 bits = vmalloc(sizeof(*bits) * args->count);
1729 if (!bits)
1730 return -ENOMEM;
1731
1732 r = copy_from_user(bits, (void __user *)args->values, args->count);
1733 if (r) {
1734 r = -EFAULT;
1735 goto out;
1736 }
1737
1738 down_read(&kvm->mm->mmap_sem);
1739 srcu_idx = srcu_read_lock(&kvm->srcu);
1740 for (i = 0; i < args->count; i++) {
1741 hva = gfn_to_hva(kvm, args->start_gfn + i);
1742 if (kvm_is_error_hva(hva)) {
1743 r = -EFAULT;
1744 break;
1745 }
1746
1747 pgstev = bits[i];
1748 pgstev = pgstev << 24;
1749 mask &= _PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT;
1750 set_pgste_bits(kvm->mm, hva, mask, pgstev);
1751 }
1752 srcu_read_unlock(&kvm->srcu, srcu_idx);
1753 up_read(&kvm->mm->mmap_sem);
1754
1755 if (!kvm->mm->context.uses_cmm) {
1756 down_write(&kvm->mm->mmap_sem);
1757 kvm->mm->context.uses_cmm = 1;
1758 up_write(&kvm->mm->mmap_sem);
1759 }
1760out:
1761 vfree(bits);
1762 return r;
1763}
1764
1765long kvm_arch_vm_ioctl(struct file *filp,
1766 unsigned int ioctl, unsigned long arg)
1767{
1768 struct kvm *kvm = filp->private_data;
1769 void __user *argp = (void __user *)arg;
1770 struct kvm_device_attr attr;
1771 int r;
1772
1773 switch (ioctl) {
1774 case KVM_S390_INTERRUPT: {
1775 struct kvm_s390_interrupt s390int;
1776
1777 r = -EFAULT;
1778 if (copy_from_user(&s390int, argp, sizeof(s390int)))
1779 break;
1780 r = kvm_s390_inject_vm(kvm, &s390int);
1781 break;
1782 }
1783 case KVM_ENABLE_CAP: {
1784 struct kvm_enable_cap cap;
1785 r = -EFAULT;
1786 if (copy_from_user(&cap, argp, sizeof(cap)))
1787 break;
1788 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
1789 break;
1790 }
1791 case KVM_CREATE_IRQCHIP: {
1792 struct kvm_irq_routing_entry routing;
1793
1794 r = -EINVAL;
1795 if (kvm->arch.use_irqchip) {
1796 /* Set up dummy routing. */
1797 memset(&routing, 0, sizeof(routing));
1798 r = kvm_set_irq_routing(kvm, &routing, 0, 0);
1799 }
1800 break;
1801 }
1802 case KVM_SET_DEVICE_ATTR: {
1803 r = -EFAULT;
1804 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1805 break;
1806 r = kvm_s390_vm_set_attr(kvm, &attr);
1807 break;
1808 }
1809 case KVM_GET_DEVICE_ATTR: {
1810 r = -EFAULT;
1811 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1812 break;
1813 r = kvm_s390_vm_get_attr(kvm, &attr);
1814 break;
1815 }
1816 case KVM_HAS_DEVICE_ATTR: {
1817 r = -EFAULT;
1818 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
1819 break;
1820 r = kvm_s390_vm_has_attr(kvm, &attr);
1821 break;
1822 }
1823 case KVM_S390_GET_SKEYS: {
1824 struct kvm_s390_skeys args;
1825
1826 r = -EFAULT;
1827 if (copy_from_user(&args, argp,
1828 sizeof(struct kvm_s390_skeys)))
1829 break;
1830 r = kvm_s390_get_skeys(kvm, &args);
1831 break;
1832 }
1833 case KVM_S390_SET_SKEYS: {
1834 struct kvm_s390_skeys args;
1835
1836 r = -EFAULT;
1837 if (copy_from_user(&args, argp,
1838 sizeof(struct kvm_s390_skeys)))
1839 break;
1840 r = kvm_s390_set_skeys(kvm, &args);
1841 break;
1842 }
1843 case KVM_S390_GET_CMMA_BITS: {
1844 struct kvm_s390_cmma_log args;
1845
1846 r = -EFAULT;
1847 if (copy_from_user(&args, argp, sizeof(args)))
1848 break;
1849 mutex_lock(&kvm->slots_lock);
1850 r = kvm_s390_get_cmma_bits(kvm, &args);
1851 mutex_unlock(&kvm->slots_lock);
1852 if (!r) {
1853 r = copy_to_user(argp, &args, sizeof(args));
1854 if (r)
1855 r = -EFAULT;
1856 }
1857 break;
1858 }
1859 case KVM_S390_SET_CMMA_BITS: {
1860 struct kvm_s390_cmma_log args;
1861
1862 r = -EFAULT;
1863 if (copy_from_user(&args, argp, sizeof(args)))
1864 break;
1865 mutex_lock(&kvm->slots_lock);
1866 r = kvm_s390_set_cmma_bits(kvm, &args);
1867 mutex_unlock(&kvm->slots_lock);
1868 break;
1869 }
1870 default:
1871 r = -ENOTTY;
1872 }
1873
1874 return r;
1875}
1876
1877static int kvm_s390_query_ap_config(u8 *config)
1878{
1879 u32 fcn_code = 0x04000000UL;
1880 u32 cc = 0;
1881
1882 memset(config, 0, 128);
1883 asm volatile(
1884 "lgr 0,%1\n"
1885 "lgr 2,%2\n"
1886 ".long 0xb2af0000\n" /* PQAP(QCI) */
1887 "0: ipm %0\n"
1888 "srl %0,28\n"
1889 "1:\n"
1890 EX_TABLE(0b, 1b)
1891 : "+r" (cc)
1892 : "r" (fcn_code), "r" (config)
1893 : "cc", "0", "2", "memory"
1894 );
1895
1896 return cc;
1897}
1898
1899static int kvm_s390_apxa_installed(void)
1900{
1901 u8 config[128];
1902 int cc;
1903
1904 if (test_facility(12)) {
1905 cc = kvm_s390_query_ap_config(config);
1906
1907 if (cc)
1908 pr_err("PQAP(QCI) failed with cc=%d", cc);
1909 else
1910 return config[0] & 0x40;
1911 }
1912
1913 return 0;
1914}
1915
1916static void kvm_s390_set_crycb_format(struct kvm *kvm)
1917{
1918 kvm->arch.crypto.crycbd = (__u32)(unsigned long) kvm->arch.crypto.crycb;
1919
1920 if (kvm_s390_apxa_installed())
1921 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
1922 else
1923 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
1924}
1925
1926static u64 kvm_s390_get_initial_cpuid(void)
1927{
1928 struct cpuid cpuid;
1929
1930 get_cpu_id(&cpuid);
1931 cpuid.version = 0xff;
1932 return *((u64 *) &cpuid);
1933}
1934
1935static void kvm_s390_crypto_init(struct kvm *kvm)
1936{
1937 if (!test_kvm_facility(kvm, 76))
1938 return;
1939
1940 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
1941 kvm_s390_set_crycb_format(kvm);
1942
1943 /* Enable AES/DEA protected key functions by default */
1944 kvm->arch.crypto.aes_kw = 1;
1945 kvm->arch.crypto.dea_kw = 1;
1946 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1947 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1948 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1949 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1950}
1951
1952static void sca_dispose(struct kvm *kvm)
1953{
1954 if (kvm->arch.use_esca)
1955 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
1956 else
1957 free_page((unsigned long)(kvm->arch.sca));
1958 kvm->arch.sca = NULL;
1959}
1960
1961int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
1962{
1963 gfp_t alloc_flags = GFP_KERNEL;
1964 int i, rc;
1965 char debug_name[16];
1966 static unsigned long sca_offset;
1967
1968 rc = -EINVAL;
1969#ifdef CONFIG_KVM_S390_UCONTROL
1970 if (type & ~KVM_VM_S390_UCONTROL)
1971 goto out_err;
1972 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
1973 goto out_err;
1974#else
1975 if (type)
1976 goto out_err;
1977#endif
1978
1979 rc = s390_enable_sie();
1980 if (rc)
1981 goto out_err;
1982
1983 rc = -ENOMEM;
1984
1985 kvm->arch.use_esca = 0; /* start with basic SCA */
1986 if (!sclp.has_64bscao)
1987 alloc_flags |= GFP_DMA;
1988 rwlock_init(&kvm->arch.sca_lock);
1989 kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
1990 if (!kvm->arch.sca)
1991 goto out_err;
1992 spin_lock(&kvm_lock);
1993 sca_offset += 16;
1994 if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
1995 sca_offset = 0;
1996 kvm->arch.sca = (struct bsca_block *)
1997 ((char *) kvm->arch.sca + sca_offset);
1998 spin_unlock(&kvm_lock);
1999
2000 sprintf(debug_name, "kvm-%u", current->pid);
2001
2002 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
2003 if (!kvm->arch.dbf)
2004 goto out_err;
2005
2006 BUILD_BUG_ON(sizeof(struct sie_page2) != 4096);
2007 kvm->arch.sie_page2 =
2008 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL | GFP_DMA);
2009 if (!kvm->arch.sie_page2)
2010 goto out_err;
2011
2012 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
2013
2014 for (i = 0; i < kvm_s390_fac_size(); i++) {
2015 kvm->arch.model.fac_mask[i] = S390_lowcore.stfle_fac_list[i] &
2016 (kvm_s390_fac_base[i] |
2017 kvm_s390_fac_ext[i]);
2018 kvm->arch.model.fac_list[i] = S390_lowcore.stfle_fac_list[i] &
2019 kvm_s390_fac_base[i];
2020 }
2021
2022 /* we are always in czam mode - even on pre z14 machines */
2023 set_kvm_facility(kvm->arch.model.fac_mask, 138);
2024 set_kvm_facility(kvm->arch.model.fac_list, 138);
2025 /* we emulate STHYI in kvm */
2026 set_kvm_facility(kvm->arch.model.fac_mask, 74);
2027 set_kvm_facility(kvm->arch.model.fac_list, 74);
2028 if (MACHINE_HAS_TLB_GUEST) {
2029 set_kvm_facility(kvm->arch.model.fac_mask, 147);
2030 set_kvm_facility(kvm->arch.model.fac_list, 147);
2031 }
2032
2033 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
2034 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
2035
2036 kvm_s390_crypto_init(kvm);
2037
2038 mutex_init(&kvm->arch.float_int.ais_lock);
2039 kvm->arch.float_int.simm = 0;
2040 kvm->arch.float_int.nimm = 0;
2041 spin_lock_init(&kvm->arch.float_int.lock);
2042 for (i = 0; i < FIRQ_LIST_COUNT; i++)
2043 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
2044 init_waitqueue_head(&kvm->arch.ipte_wq);
2045 mutex_init(&kvm->arch.ipte_mutex);
2046
2047 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
2048 VM_EVENT(kvm, 3, "vm created with type %lu", type);
2049
2050 if (type & KVM_VM_S390_UCONTROL) {
2051 kvm->arch.gmap = NULL;
2052 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
2053 } else {
2054 if (sclp.hamax == U64_MAX)
2055 kvm->arch.mem_limit = TASK_SIZE_MAX;
2056 else
2057 kvm->arch.mem_limit = min_t(unsigned long, TASK_SIZE_MAX,
2058 sclp.hamax + 1);
2059 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
2060 if (!kvm->arch.gmap)
2061 goto out_err;
2062 kvm->arch.gmap->private = kvm;
2063 kvm->arch.gmap->pfault_enabled = 0;
2064 }
2065
2066 kvm->arch.css_support = 0;
2067 kvm->arch.use_irqchip = 0;
2068 kvm->arch.use_pfmfi = sclp.has_pfmfi;
2069 kvm->arch.epoch = 0;
2070
2071 spin_lock_init(&kvm->arch.start_stop_lock);
2072 kvm_s390_vsie_init(kvm);
2073 kvm_s390_gisa_init(kvm);
2074 KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
2075
2076 return 0;
2077out_err:
2078 free_page((unsigned long)kvm->arch.sie_page2);
2079 debug_unregister(kvm->arch.dbf);
2080 sca_dispose(kvm);
2081 KVM_EVENT(3, "creation of vm failed: %d", rc);
2082 return rc;
2083}
2084
2085bool kvm_arch_has_vcpu_debugfs(void)
2086{
2087 return false;
2088}
2089
2090int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2091{
2092 return 0;
2093}
2094
2095void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
2096{
2097 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
2098 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
2099 kvm_s390_clear_local_irqs(vcpu);
2100 kvm_clear_async_pf_completion_queue(vcpu);
2101 if (!kvm_is_ucontrol(vcpu->kvm))
2102 sca_del_vcpu(vcpu);
2103
2104 if (kvm_is_ucontrol(vcpu->kvm))
2105 gmap_remove(vcpu->arch.gmap);
2106
2107 if (vcpu->kvm->arch.use_cmma)
2108 kvm_s390_vcpu_unsetup_cmma(vcpu);
2109 free_page((unsigned long)(vcpu->arch.sie_block));
2110
2111 kvm_vcpu_uninit(vcpu);
2112 kmem_cache_free(kvm_vcpu_cache, vcpu);
2113}
2114
2115static void kvm_free_vcpus(struct kvm *kvm)
2116{
2117 unsigned int i;
2118 struct kvm_vcpu *vcpu;
2119
2120 kvm_for_each_vcpu(i, vcpu, kvm)
2121 kvm_arch_vcpu_destroy(vcpu);
2122
2123 mutex_lock(&kvm->lock);
2124 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
2125 kvm->vcpus[i] = NULL;
2126
2127 atomic_set(&kvm->online_vcpus, 0);
2128 mutex_unlock(&kvm->lock);
2129}
2130
2131void kvm_arch_destroy_vm(struct kvm *kvm)
2132{
2133 kvm_free_vcpus(kvm);
2134 sca_dispose(kvm);
2135 debug_unregister(kvm->arch.dbf);
2136 kvm_s390_gisa_destroy(kvm);
2137 free_page((unsigned long)kvm->arch.sie_page2);
2138 if (!kvm_is_ucontrol(kvm))
2139 gmap_remove(kvm->arch.gmap);
2140 kvm_s390_destroy_adapters(kvm);
2141 kvm_s390_clear_float_irqs(kvm);
2142 kvm_s390_vsie_destroy(kvm);
2143 if (kvm->arch.migration_state) {
2144 vfree(kvm->arch.migration_state->pgste_bitmap);
2145 kfree(kvm->arch.migration_state);
2146 }
2147 KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
2148}
2149
2150/* Section: vcpu related */
2151static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
2152{
2153 vcpu->arch.gmap = gmap_create(current->mm, -1UL);
2154 if (!vcpu->arch.gmap)
2155 return -ENOMEM;
2156 vcpu->arch.gmap->private = vcpu->kvm;
2157
2158 return 0;
2159}
2160
2161static void sca_del_vcpu(struct kvm_vcpu *vcpu)
2162{
2163 if (!kvm_s390_use_sca_entries())
2164 return;
2165 read_lock(&vcpu->kvm->arch.sca_lock);
2166 if (vcpu->kvm->arch.use_esca) {
2167 struct esca_block *sca = vcpu->kvm->arch.sca;
2168
2169 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
2170 sca->cpu[vcpu->vcpu_id].sda = 0;
2171 } else {
2172 struct bsca_block *sca = vcpu->kvm->arch.sca;
2173
2174 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
2175 sca->cpu[vcpu->vcpu_id].sda = 0;
2176 }
2177 read_unlock(&vcpu->kvm->arch.sca_lock);
2178}
2179
2180static void sca_add_vcpu(struct kvm_vcpu *vcpu)
2181{
2182 if (!kvm_s390_use_sca_entries()) {
2183 struct bsca_block *sca = vcpu->kvm->arch.sca;
2184
2185 /* we still need the basic sca for the ipte control */
2186 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
2187 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
2188 return;
2189 }
2190 read_lock(&vcpu->kvm->arch.sca_lock);
2191 if (vcpu->kvm->arch.use_esca) {
2192 struct esca_block *sca = vcpu->kvm->arch.sca;
2193
2194 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
2195 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
2196 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca & ~0x3fU;
2197 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
2198 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
2199 } else {
2200 struct bsca_block *sca = vcpu->kvm->arch.sca;
2201
2202 sca->cpu[vcpu->vcpu_id].sda = (__u64) vcpu->arch.sie_block;
2203 vcpu->arch.sie_block->scaoh = (__u32)(((__u64)sca) >> 32);
2204 vcpu->arch.sie_block->scaol = (__u32)(__u64)sca;
2205 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
2206 }
2207 read_unlock(&vcpu->kvm->arch.sca_lock);
2208}
2209
2210/* Basic SCA to Extended SCA data copy routines */
2211static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
2212{
2213 d->sda = s->sda;
2214 d->sigp_ctrl.c = s->sigp_ctrl.c;
2215 d->sigp_ctrl.scn = s->sigp_ctrl.scn;
2216}
2217
2218static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
2219{
2220 int i;
2221
2222 d->ipte_control = s->ipte_control;
2223 d->mcn[0] = s->mcn;
2224 for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
2225 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
2226}
2227
2228static int sca_switch_to_extended(struct kvm *kvm)
2229{
2230 struct bsca_block *old_sca = kvm->arch.sca;
2231 struct esca_block *new_sca;
2232 struct kvm_vcpu *vcpu;
2233 unsigned int vcpu_idx;
2234 u32 scaol, scaoh;
2235
2236 new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL|__GFP_ZERO);
2237 if (!new_sca)
2238 return -ENOMEM;
2239
2240 scaoh = (u32)((u64)(new_sca) >> 32);
2241 scaol = (u32)(u64)(new_sca) & ~0x3fU;
2242
2243 kvm_s390_vcpu_block_all(kvm);
2244 write_lock(&kvm->arch.sca_lock);
2245
2246 sca_copy_b_to_e(new_sca, old_sca);
2247
2248 kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
2249 vcpu->arch.sie_block->scaoh = scaoh;
2250 vcpu->arch.sie_block->scaol = scaol;
2251 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
2252 }
2253 kvm->arch.sca = new_sca;
2254 kvm->arch.use_esca = 1;
2255
2256 write_unlock(&kvm->arch.sca_lock);
2257 kvm_s390_vcpu_unblock_all(kvm);
2258
2259 free_page((unsigned long)old_sca);
2260
2261 VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
2262 old_sca, kvm->arch.sca);
2263 return 0;
2264}
2265
2266static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
2267{
2268 int rc;
2269
2270 if (!kvm_s390_use_sca_entries()) {
2271 if (id < KVM_MAX_VCPUS)
2272 return true;
2273 return false;
2274 }
2275 if (id < KVM_S390_BSCA_CPU_SLOTS)
2276 return true;
2277 if (!sclp.has_esca || !sclp.has_64bscao)
2278 return false;
2279
2280 mutex_lock(&kvm->lock);
2281 rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
2282 mutex_unlock(&kvm->lock);
2283
2284 return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
2285}
2286
2287int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
2288{
2289 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
2290 kvm_clear_async_pf_completion_queue(vcpu);
2291 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
2292 KVM_SYNC_GPRS |
2293 KVM_SYNC_ACRS |
2294 KVM_SYNC_CRS |
2295 KVM_SYNC_ARCH0 |
2296 KVM_SYNC_PFAULT;
2297 kvm_s390_set_prefix(vcpu, 0);
2298 if (test_kvm_facility(vcpu->kvm, 64))
2299 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
2300 if (test_kvm_facility(vcpu->kvm, 82))
2301 vcpu->run->kvm_valid_regs |= KVM_SYNC_BPBC;
2302 if (test_kvm_facility(vcpu->kvm, 133))
2303 vcpu->run->kvm_valid_regs |= KVM_SYNC_GSCB;
2304 /* fprs can be synchronized via vrs, even if the guest has no vx. With
2305 * MACHINE_HAS_VX, (load|store)_fpu_regs() will work with vrs format.
2306 */
2307 if (MACHINE_HAS_VX)
2308 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
2309 else
2310 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
2311
2312 if (kvm_is_ucontrol(vcpu->kvm))
2313 return __kvm_ucontrol_vcpu_init(vcpu);
2314
2315 return 0;
2316}
2317
2318/* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2319static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2320{
2321 WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
2322 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
2323 vcpu->arch.cputm_start = get_tod_clock_fast();
2324 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
2325}
2326
2327/* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2328static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2329{
2330 WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
2331 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
2332 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
2333 vcpu->arch.cputm_start = 0;
2334 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
2335}
2336
2337/* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2338static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2339{
2340 WARN_ON_ONCE(vcpu->arch.cputm_enabled);
2341 vcpu->arch.cputm_enabled = true;
2342 __start_cpu_timer_accounting(vcpu);
2343}
2344
2345/* needs disabled preemption to protect from TOD sync and vcpu_load/put */
2346static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2347{
2348 WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
2349 __stop_cpu_timer_accounting(vcpu);
2350 vcpu->arch.cputm_enabled = false;
2351}
2352
2353static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2354{
2355 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2356 __enable_cpu_timer_accounting(vcpu);
2357 preempt_enable();
2358}
2359
2360static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
2361{
2362 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2363 __disable_cpu_timer_accounting(vcpu);
2364 preempt_enable();
2365}
2366
2367/* set the cpu timer - may only be called from the VCPU thread itself */
2368void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
2369{
2370 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2371 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
2372 if (vcpu->arch.cputm_enabled)
2373 vcpu->arch.cputm_start = get_tod_clock_fast();
2374 vcpu->arch.sie_block->cputm = cputm;
2375 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
2376 preempt_enable();
2377}
2378
2379/* update and get the cpu timer - can also be called from other VCPU threads */
2380__u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
2381{
2382 unsigned int seq;
2383 __u64 value;
2384
2385 if (unlikely(!vcpu->arch.cputm_enabled))
2386 return vcpu->arch.sie_block->cputm;
2387
2388 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
2389 do {
2390 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
2391 /*
2392 * If the writer would ever execute a read in the critical
2393 * section, e.g. in irq context, we have a deadlock.
2394 */
2395 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
2396 value = vcpu->arch.sie_block->cputm;
2397 /* if cputm_start is 0, accounting is being started/stopped */
2398 if (likely(vcpu->arch.cputm_start))
2399 value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
2400 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
2401 preempt_enable();
2402 return value;
2403}
2404
2405void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2406{
2407
2408 gmap_enable(vcpu->arch.enabled_gmap);
2409 kvm_s390_set_cpuflags(vcpu, CPUSTAT_RUNNING);
2410 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
2411 __start_cpu_timer_accounting(vcpu);
2412 vcpu->cpu = cpu;
2413}
2414
2415void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2416{
2417 vcpu->cpu = -1;
2418 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
2419 __stop_cpu_timer_accounting(vcpu);
2420 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_RUNNING);
2421 vcpu->arch.enabled_gmap = gmap_get_enabled();
2422 gmap_disable(vcpu->arch.enabled_gmap);
2423
2424}
2425
2426static void kvm_s390_vcpu_initial_reset(struct kvm_vcpu *vcpu)
2427{
2428 /* this equals initial cpu reset in pop, but we don't switch to ESA */
2429 vcpu->arch.sie_block->gpsw.mask = 0UL;
2430 vcpu->arch.sie_block->gpsw.addr = 0UL;
2431 kvm_s390_set_prefix(vcpu, 0);
2432 kvm_s390_set_cpu_timer(vcpu, 0);
2433 vcpu->arch.sie_block->ckc = 0UL;
2434 vcpu->arch.sie_block->todpr = 0;
2435 memset(vcpu->arch.sie_block->gcr, 0, 16 * sizeof(__u64));
2436 vcpu->arch.sie_block->gcr[0] = 0xE0UL;
2437 vcpu->arch.sie_block->gcr[14] = 0xC2000000UL;
2438 /* make sure the new fpc will be lazily loaded */
2439 save_fpu_regs();
2440 current->thread.fpu.fpc = 0;
2441 vcpu->arch.sie_block->gbea = 1;
2442 vcpu->arch.sie_block->pp = 0;
2443 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
2444 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
2445 kvm_clear_async_pf_completion_queue(vcpu);
2446 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
2447 kvm_s390_vcpu_stop(vcpu);
2448 kvm_s390_clear_local_irqs(vcpu);
2449}
2450
2451void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
2452{
2453 mutex_lock(&vcpu->kvm->lock);
2454 preempt_disable();
2455 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
2456 vcpu->arch.sie_block->epdx = vcpu->kvm->arch.epdx;
2457 preempt_enable();
2458 mutex_unlock(&vcpu->kvm->lock);
2459 if (!kvm_is_ucontrol(vcpu->kvm)) {
2460 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
2461 sca_add_vcpu(vcpu);
2462 }
2463 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
2464 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
2465 /* make vcpu_load load the right gmap on the first trigger */
2466 vcpu->arch.enabled_gmap = vcpu->arch.gmap;
2467}
2468
2469static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
2470{
2471 if (!test_kvm_facility(vcpu->kvm, 76))
2472 return;
2473
2474 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
2475
2476 if (vcpu->kvm->arch.crypto.aes_kw)
2477 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
2478 if (vcpu->kvm->arch.crypto.dea_kw)
2479 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
2480
2481 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
2482}
2483
2484void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
2485{
2486 free_page(vcpu->arch.sie_block->cbrlo);
2487 vcpu->arch.sie_block->cbrlo = 0;
2488}
2489
2490int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
2491{
2492 vcpu->arch.sie_block->cbrlo = get_zeroed_page(GFP_KERNEL);
2493 if (!vcpu->arch.sie_block->cbrlo)
2494 return -ENOMEM;
2495 return 0;
2496}
2497
2498static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
2499{
2500 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
2501
2502 vcpu->arch.sie_block->ibc = model->ibc;
2503 if (test_kvm_facility(vcpu->kvm, 7))
2504 vcpu->arch.sie_block->fac = (u32)(u64) model->fac_list;
2505}
2506
2507int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
2508{
2509 int rc = 0;
2510
2511 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
2512 CPUSTAT_SM |
2513 CPUSTAT_STOPPED);
2514
2515 if (test_kvm_facility(vcpu->kvm, 78))
2516 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED2);
2517 else if (test_kvm_facility(vcpu->kvm, 8))
2518 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED);
2519
2520 kvm_s390_vcpu_setup_model(vcpu);
2521
2522 /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
2523 if (MACHINE_HAS_ESOP)
2524 vcpu->arch.sie_block->ecb |= ECB_HOSTPROTINT;
2525 if (test_kvm_facility(vcpu->kvm, 9))
2526 vcpu->arch.sie_block->ecb |= ECB_SRSI;
2527 if (test_kvm_facility(vcpu->kvm, 73))
2528 vcpu->arch.sie_block->ecb |= ECB_TE;
2529
2530 if (test_kvm_facility(vcpu->kvm, 8) && vcpu->kvm->arch.use_pfmfi)
2531 vcpu->arch.sie_block->ecb2 |= ECB2_PFMFI;
2532 if (test_kvm_facility(vcpu->kvm, 130))
2533 vcpu->arch.sie_block->ecb2 |= ECB2_IEP;
2534 vcpu->arch.sie_block->eca = ECA_MVPGI | ECA_PROTEXCI;
2535 if (sclp.has_cei)
2536 vcpu->arch.sie_block->eca |= ECA_CEI;
2537 if (sclp.has_ib)
2538 vcpu->arch.sie_block->eca |= ECA_IB;
2539 if (sclp.has_siif)
2540 vcpu->arch.sie_block->eca |= ECA_SII;
2541 if (sclp.has_sigpif)
2542 vcpu->arch.sie_block->eca |= ECA_SIGPI;
2543 if (test_kvm_facility(vcpu->kvm, 129)) {
2544 vcpu->arch.sie_block->eca |= ECA_VX;
2545 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
2546 }
2547 if (test_kvm_facility(vcpu->kvm, 139))
2548 vcpu->arch.sie_block->ecd |= ECD_MEF;
2549
2550 if (vcpu->arch.sie_block->gd) {
2551 vcpu->arch.sie_block->eca |= ECA_AIV;
2552 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
2553 vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
2554 }
2555 vcpu->arch.sie_block->sdnxo = ((unsigned long) &vcpu->run->s.regs.sdnx)
2556 | SDNXC;
2557 vcpu->arch.sie_block->riccbd = (unsigned long) &vcpu->run->s.regs.riccb;
2558
2559 if (sclp.has_kss)
2560 kvm_s390_set_cpuflags(vcpu, CPUSTAT_KSS);
2561 else
2562 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
2563
2564 if (vcpu->kvm->arch.use_cmma) {
2565 rc = kvm_s390_vcpu_setup_cmma(vcpu);
2566 if (rc)
2567 return rc;
2568 }
2569 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
2570 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
2571
2572 kvm_s390_vcpu_crypto_setup(vcpu);
2573
2574 return rc;
2575}
2576
2577struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
2578 unsigned int id)
2579{
2580 struct kvm_vcpu *vcpu;
2581 struct sie_page *sie_page;
2582 int rc = -EINVAL;
2583
2584 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
2585 goto out;
2586
2587 rc = -ENOMEM;
2588
2589 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2590 if (!vcpu)
2591 goto out;
2592
2593 BUILD_BUG_ON(sizeof(struct sie_page) != 4096);
2594 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL);
2595 if (!sie_page)
2596 goto out_free_cpu;
2597
2598 vcpu->arch.sie_block = &sie_page->sie_block;
2599 vcpu->arch.sie_block->itdba = (unsigned long) &sie_page->itdb;
2600
2601 /* the real guest size will always be smaller than msl */
2602 vcpu->arch.sie_block->mso = 0;
2603 vcpu->arch.sie_block->msl = sclp.hamax;
2604
2605 vcpu->arch.sie_block->icpua = id;
2606 spin_lock_init(&vcpu->arch.local_int.lock);
2607 vcpu->arch.sie_block->gd = (u32)(u64)kvm->arch.gisa;
2608 if (vcpu->arch.sie_block->gd && sclp.has_gisaf)
2609 vcpu->arch.sie_block->gd |= GISA_FORMAT1;
2610 seqcount_init(&vcpu->arch.cputm_seqcount);
2611
2612 rc = kvm_vcpu_init(vcpu, kvm, id);
2613 if (rc)
2614 goto out_free_sie_block;
2615 VM_EVENT(kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK", id, vcpu,
2616 vcpu->arch.sie_block);
2617 trace_kvm_s390_create_vcpu(id, vcpu, vcpu->arch.sie_block);
2618
2619 return vcpu;
2620out_free_sie_block:
2621 free_page((unsigned long)(vcpu->arch.sie_block));
2622out_free_cpu:
2623 kmem_cache_free(kvm_vcpu_cache, vcpu);
2624out:
2625 return ERR_PTR(rc);
2626}
2627
2628int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
2629{
2630 return kvm_s390_vcpu_has_irq(vcpu, 0);
2631}
2632
2633bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
2634{
2635 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE);
2636}
2637
2638void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
2639{
2640 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2641 exit_sie(vcpu);
2642}
2643
2644void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
2645{
2646 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
2647}
2648
2649static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
2650{
2651 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2652 exit_sie(vcpu);
2653}
2654
2655static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
2656{
2657 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
2658}
2659
2660/*
2661 * Kick a guest cpu out of SIE and wait until SIE is not running.
2662 * If the CPU is not running (e.g. waiting as idle) the function will
2663 * return immediately. */
2664void exit_sie(struct kvm_vcpu *vcpu)
2665{
2666 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
2667 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
2668 cpu_relax();
2669}
2670
2671/* Kick a guest cpu out of SIE to process a request synchronously */
2672void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
2673{
2674 kvm_make_request(req, vcpu);
2675 kvm_s390_vcpu_request(vcpu);
2676}
2677
2678static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
2679 unsigned long end)
2680{
2681 struct kvm *kvm = gmap->private;
2682 struct kvm_vcpu *vcpu;
2683 unsigned long prefix;
2684 int i;
2685
2686 if (gmap_is_shadow(gmap))
2687 return;
2688 if (start >= 1UL << 31)
2689 /* We are only interested in prefix pages */
2690 return;
2691 kvm_for_each_vcpu(i, vcpu, kvm) {
2692 /* match against both prefix pages */
2693 prefix = kvm_s390_get_prefix(vcpu);
2694 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
2695 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
2696 start, end);
2697 kvm_s390_sync_request(KVM_REQ_MMU_RELOAD, vcpu);
2698 }
2699 }
2700}
2701
2702int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
2703{
2704 /* kvm common code refers to this, but never calls it */
2705 BUG();
2706 return 0;
2707}
2708
2709static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
2710 struct kvm_one_reg *reg)
2711{
2712 int r = -EINVAL;
2713
2714 switch (reg->id) {
2715 case KVM_REG_S390_TODPR:
2716 r = put_user(vcpu->arch.sie_block->todpr,
2717 (u32 __user *)reg->addr);
2718 break;
2719 case KVM_REG_S390_EPOCHDIFF:
2720 r = put_user(vcpu->arch.sie_block->epoch,
2721 (u64 __user *)reg->addr);
2722 break;
2723 case KVM_REG_S390_CPU_TIMER:
2724 r = put_user(kvm_s390_get_cpu_timer(vcpu),
2725 (u64 __user *)reg->addr);
2726 break;
2727 case KVM_REG_S390_CLOCK_COMP:
2728 r = put_user(vcpu->arch.sie_block->ckc,
2729 (u64 __user *)reg->addr);
2730 break;
2731 case KVM_REG_S390_PFTOKEN:
2732 r = put_user(vcpu->arch.pfault_token,
2733 (u64 __user *)reg->addr);
2734 break;
2735 case KVM_REG_S390_PFCOMPARE:
2736 r = put_user(vcpu->arch.pfault_compare,
2737 (u64 __user *)reg->addr);
2738 break;
2739 case KVM_REG_S390_PFSELECT:
2740 r = put_user(vcpu->arch.pfault_select,
2741 (u64 __user *)reg->addr);
2742 break;
2743 case KVM_REG_S390_PP:
2744 r = put_user(vcpu->arch.sie_block->pp,
2745 (u64 __user *)reg->addr);
2746 break;
2747 case KVM_REG_S390_GBEA:
2748 r = put_user(vcpu->arch.sie_block->gbea,
2749 (u64 __user *)reg->addr);
2750 break;
2751 default:
2752 break;
2753 }
2754
2755 return r;
2756}
2757
2758static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
2759 struct kvm_one_reg *reg)
2760{
2761 int r = -EINVAL;
2762 __u64 val;
2763
2764 switch (reg->id) {
2765 case KVM_REG_S390_TODPR:
2766 r = get_user(vcpu->arch.sie_block->todpr,
2767 (u32 __user *)reg->addr);
2768 break;
2769 case KVM_REG_S390_EPOCHDIFF:
2770 r = get_user(vcpu->arch.sie_block->epoch,
2771 (u64 __user *)reg->addr);
2772 break;
2773 case KVM_REG_S390_CPU_TIMER:
2774 r = get_user(val, (u64 __user *)reg->addr);
2775 if (!r)
2776 kvm_s390_set_cpu_timer(vcpu, val);
2777 break;
2778 case KVM_REG_S390_CLOCK_COMP:
2779 r = get_user(vcpu->arch.sie_block->ckc,
2780 (u64 __user *)reg->addr);
2781 break;
2782 case KVM_REG_S390_PFTOKEN:
2783 r = get_user(vcpu->arch.pfault_token,
2784 (u64 __user *)reg->addr);
2785 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
2786 kvm_clear_async_pf_completion_queue(vcpu);
2787 break;
2788 case KVM_REG_S390_PFCOMPARE:
2789 r = get_user(vcpu->arch.pfault_compare,
2790 (u64 __user *)reg->addr);
2791 break;
2792 case KVM_REG_S390_PFSELECT:
2793 r = get_user(vcpu->arch.pfault_select,
2794 (u64 __user *)reg->addr);
2795 break;
2796 case KVM_REG_S390_PP:
2797 r = get_user(vcpu->arch.sie_block->pp,
2798 (u64 __user *)reg->addr);
2799 break;
2800 case KVM_REG_S390_GBEA:
2801 r = get_user(vcpu->arch.sie_block->gbea,
2802 (u64 __user *)reg->addr);
2803 break;
2804 default:
2805 break;
2806 }
2807
2808 return r;
2809}
2810
2811static int kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
2812{
2813 kvm_s390_vcpu_initial_reset(vcpu);
2814 return 0;
2815}
2816
2817int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2818{
2819 vcpu_load(vcpu);
2820 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs));
2821 vcpu_put(vcpu);
2822 return 0;
2823}
2824
2825int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2826{
2827 vcpu_load(vcpu);
2828 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
2829 vcpu_put(vcpu);
2830 return 0;
2831}
2832
2833int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2834 struct kvm_sregs *sregs)
2835{
2836 vcpu_load(vcpu);
2837
2838 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
2839 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
2840
2841 vcpu_put(vcpu);
2842 return 0;
2843}
2844
2845int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2846 struct kvm_sregs *sregs)
2847{
2848 vcpu_load(vcpu);
2849
2850 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
2851 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
2852
2853 vcpu_put(vcpu);
2854 return 0;
2855}
2856
2857int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2858{
2859 int ret = 0;
2860
2861 vcpu_load(vcpu);
2862
2863 if (test_fp_ctl(fpu->fpc)) {
2864 ret = -EINVAL;
2865 goto out;
2866 }
2867 vcpu->run->s.regs.fpc = fpu->fpc;
2868 if (MACHINE_HAS_VX)
2869 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs,
2870 (freg_t *) fpu->fprs);
2871 else
2872 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs));
2873
2874out:
2875 vcpu_put(vcpu);
2876 return ret;
2877}
2878
2879int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2880{
2881 vcpu_load(vcpu);
2882
2883 /* make sure we have the latest values */
2884 save_fpu_regs();
2885 if (MACHINE_HAS_VX)
2886 convert_vx_to_fp((freg_t *) fpu->fprs,
2887 (__vector128 *) vcpu->run->s.regs.vrs);
2888 else
2889 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs));
2890 fpu->fpc = vcpu->run->s.regs.fpc;
2891
2892 vcpu_put(vcpu);
2893 return 0;
2894}
2895
2896static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
2897{
2898 int rc = 0;
2899
2900 if (!is_vcpu_stopped(vcpu))
2901 rc = -EBUSY;
2902 else {
2903 vcpu->run->psw_mask = psw.mask;
2904 vcpu->run->psw_addr = psw.addr;
2905 }
2906 return rc;
2907}
2908
2909int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2910 struct kvm_translation *tr)
2911{
2912 return -EINVAL; /* not implemented yet */
2913}
2914
2915#define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
2916 KVM_GUESTDBG_USE_HW_BP | \
2917 KVM_GUESTDBG_ENABLE)
2918
2919int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
2920 struct kvm_guest_debug *dbg)
2921{
2922 int rc = 0;
2923
2924 vcpu_load(vcpu);
2925
2926 vcpu->guest_debug = 0;
2927 kvm_s390_clear_bp_data(vcpu);
2928
2929 if (dbg->control & ~VALID_GUESTDBG_FLAGS) {
2930 rc = -EINVAL;
2931 goto out;
2932 }
2933 if (!sclp.has_gpere) {
2934 rc = -EINVAL;
2935 goto out;
2936 }
2937
2938 if (dbg->control & KVM_GUESTDBG_ENABLE) {
2939 vcpu->guest_debug = dbg->control;
2940 /* enforce guest PER */
2941 kvm_s390_set_cpuflags(vcpu, CPUSTAT_P);
2942
2943 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
2944 rc = kvm_s390_import_bp_data(vcpu, dbg);
2945 } else {
2946 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
2947 vcpu->arch.guestdbg.last_bp = 0;
2948 }
2949
2950 if (rc) {
2951 vcpu->guest_debug = 0;
2952 kvm_s390_clear_bp_data(vcpu);
2953 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
2954 }
2955
2956out:
2957 vcpu_put(vcpu);
2958 return rc;
2959}
2960
2961int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
2962 struct kvm_mp_state *mp_state)
2963{
2964 int ret;
2965
2966 vcpu_load(vcpu);
2967
2968 /* CHECK_STOP and LOAD are not supported yet */
2969 ret = is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
2970 KVM_MP_STATE_OPERATING;
2971
2972 vcpu_put(vcpu);
2973 return ret;
2974}
2975
2976int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
2977 struct kvm_mp_state *mp_state)
2978{
2979 int rc = 0;
2980
2981 vcpu_load(vcpu);
2982
2983 /* user space knows about this interface - let it control the state */
2984 vcpu->kvm->arch.user_cpu_state_ctrl = 1;
2985
2986 switch (mp_state->mp_state) {
2987 case KVM_MP_STATE_STOPPED:
2988 kvm_s390_vcpu_stop(vcpu);
2989 break;
2990 case KVM_MP_STATE_OPERATING:
2991 kvm_s390_vcpu_start(vcpu);
2992 break;
2993 case KVM_MP_STATE_LOAD:
2994 case KVM_MP_STATE_CHECK_STOP:
2995 /* fall through - CHECK_STOP and LOAD are not supported yet */
2996 default:
2997 rc = -ENXIO;
2998 }
2999
3000 vcpu_put(vcpu);
3001 return rc;
3002}
3003
3004static bool ibs_enabled(struct kvm_vcpu *vcpu)
3005{
3006 return kvm_s390_test_cpuflags(vcpu, CPUSTAT_IBS);
3007}
3008
3009static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
3010{
3011retry:
3012 kvm_s390_vcpu_request_handled(vcpu);
3013 if (!kvm_request_pending(vcpu))
3014 return 0;
3015 /*
3016 * We use MMU_RELOAD just to re-arm the ipte notifier for the
3017 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
3018 * This ensures that the ipte instruction for this request has
3019 * already finished. We might race against a second unmapper that
3020 * wants to set the blocking bit. Lets just retry the request loop.
3021 */
3022 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) {
3023 int rc;
3024 rc = gmap_mprotect_notify(vcpu->arch.gmap,
3025 kvm_s390_get_prefix(vcpu),
3026 PAGE_SIZE * 2, PROT_WRITE);
3027 if (rc) {
3028 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
3029 return rc;
3030 }
3031 goto retry;
3032 }
3033
3034 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
3035 vcpu->arch.sie_block->ihcpu = 0xffff;
3036 goto retry;
3037 }
3038
3039 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
3040 if (!ibs_enabled(vcpu)) {
3041 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
3042 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IBS);
3043 }
3044 goto retry;
3045 }
3046
3047 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
3048 if (ibs_enabled(vcpu)) {
3049 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
3050 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IBS);
3051 }
3052 goto retry;
3053 }
3054
3055 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
3056 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
3057 goto retry;
3058 }
3059
3060 if (kvm_check_request(KVM_REQ_START_MIGRATION, vcpu)) {
3061 /*
3062 * Disable CMM virtualization; we will emulate the ESSA
3063 * instruction manually, in order to provide additional
3064 * functionalities needed for live migration.
3065 */
3066 vcpu->arch.sie_block->ecb2 &= ~ECB2_CMMA;
3067 goto retry;
3068 }
3069
3070 if (kvm_check_request(KVM_REQ_STOP_MIGRATION, vcpu)) {
3071 /*
3072 * Re-enable CMM virtualization if CMMA is available and
3073 * CMM has been used.
3074 */
3075 if ((vcpu->kvm->arch.use_cmma) &&
3076 (vcpu->kvm->mm->context.uses_cmm))
3077 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
3078 goto retry;
3079 }
3080
3081 /* nothing to do, just clear the request */
3082 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
3083
3084 return 0;
3085}
3086
3087void kvm_s390_set_tod_clock(struct kvm *kvm,
3088 const struct kvm_s390_vm_tod_clock *gtod)
3089{
3090 struct kvm_vcpu *vcpu;
3091 struct kvm_s390_tod_clock_ext htod;
3092 int i;
3093
3094 mutex_lock(&kvm->lock);
3095 preempt_disable();
3096
3097 get_tod_clock_ext((char *)&htod);
3098
3099 kvm->arch.epoch = gtod->tod - htod.tod;
3100 kvm->arch.epdx = 0;
3101 if (test_kvm_facility(kvm, 139)) {
3102 kvm->arch.epdx = gtod->epoch_idx - htod.epoch_idx;
3103 if (kvm->arch.epoch > gtod->tod)
3104 kvm->arch.epdx -= 1;
3105 }
3106
3107 kvm_s390_vcpu_block_all(kvm);
3108 kvm_for_each_vcpu(i, vcpu, kvm) {
3109 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
3110 vcpu->arch.sie_block->epdx = kvm->arch.epdx;
3111 }
3112
3113 kvm_s390_vcpu_unblock_all(kvm);
3114 preempt_enable();
3115 mutex_unlock(&kvm->lock);
3116}
3117
3118/**
3119 * kvm_arch_fault_in_page - fault-in guest page if necessary
3120 * @vcpu: The corresponding virtual cpu
3121 * @gpa: Guest physical address
3122 * @writable: Whether the page should be writable or not
3123 *
3124 * Make sure that a guest page has been faulted-in on the host.
3125 *
3126 * Return: Zero on success, negative error code otherwise.
3127 */
3128long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
3129{
3130 return gmap_fault(vcpu->arch.gmap, gpa,
3131 writable ? FAULT_FLAG_WRITE : 0);
3132}
3133
3134static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
3135 unsigned long token)
3136{
3137 struct kvm_s390_interrupt inti;
3138 struct kvm_s390_irq irq;
3139
3140 if (start_token) {
3141 irq.u.ext.ext_params2 = token;
3142 irq.type = KVM_S390_INT_PFAULT_INIT;
3143 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
3144 } else {
3145 inti.type = KVM_S390_INT_PFAULT_DONE;
3146 inti.parm64 = token;
3147 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
3148 }
3149}
3150
3151void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
3152 struct kvm_async_pf *work)
3153{
3154 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
3155 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
3156}
3157
3158void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
3159 struct kvm_async_pf *work)
3160{
3161 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
3162 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
3163}
3164
3165void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
3166 struct kvm_async_pf *work)
3167{
3168 /* s390 will always inject the page directly */
3169}
3170
3171bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
3172{
3173 /*
3174 * s390 will always inject the page directly,
3175 * but we still want check_async_completion to cleanup
3176 */
3177 return true;
3178}
3179
3180static int kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
3181{
3182 hva_t hva;
3183 struct kvm_arch_async_pf arch;
3184 int rc;
3185
3186 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
3187 return 0;
3188 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
3189 vcpu->arch.pfault_compare)
3190 return 0;
3191 if (psw_extint_disabled(vcpu))
3192 return 0;
3193 if (kvm_s390_vcpu_has_irq(vcpu, 0))
3194 return 0;
3195 if (!(vcpu->arch.sie_block->gcr[0] & 0x200ul))
3196 return 0;
3197 if (!vcpu->arch.gmap->pfault_enabled)
3198 return 0;
3199
3200 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
3201 hva += current->thread.gmap_addr & ~PAGE_MASK;
3202 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
3203 return 0;
3204
3205 rc = kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
3206 return rc;
3207}
3208
3209static int vcpu_pre_run(struct kvm_vcpu *vcpu)
3210{
3211 int rc, cpuflags;
3212
3213 /*
3214 * On s390 notifications for arriving pages will be delivered directly
3215 * to the guest but the house keeping for completed pfaults is
3216 * handled outside the worker.
3217 */
3218 kvm_check_async_pf_completion(vcpu);
3219
3220 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
3221 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
3222
3223 if (need_resched())
3224 schedule();
3225
3226 if (test_cpu_flag(CIF_MCCK_PENDING))
3227 s390_handle_mcck();
3228
3229 if (!kvm_is_ucontrol(vcpu->kvm)) {
3230 rc = kvm_s390_deliver_pending_interrupts(vcpu);
3231 if (rc)
3232 return rc;
3233 }
3234
3235 rc = kvm_s390_handle_requests(vcpu);
3236 if (rc)
3237 return rc;
3238
3239 if (guestdbg_enabled(vcpu)) {
3240 kvm_s390_backup_guest_per_regs(vcpu);
3241 kvm_s390_patch_guest_per_regs(vcpu);
3242 }
3243
3244 vcpu->arch.sie_block->icptcode = 0;
3245 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
3246 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
3247 trace_kvm_s390_sie_enter(vcpu, cpuflags);
3248
3249 return 0;
3250}
3251
3252static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
3253{
3254 struct kvm_s390_pgm_info pgm_info = {
3255 .code = PGM_ADDRESSING,
3256 };
3257 u8 opcode, ilen;
3258 int rc;
3259
3260 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
3261 trace_kvm_s390_sie_fault(vcpu);
3262
3263 /*
3264 * We want to inject an addressing exception, which is defined as a
3265 * suppressing or terminating exception. However, since we came here
3266 * by a DAT access exception, the PSW still points to the faulting
3267 * instruction since DAT exceptions are nullifying. So we've got
3268 * to look up the current opcode to get the length of the instruction
3269 * to be able to forward the PSW.
3270 */
3271 rc = read_guest_instr(vcpu, vcpu->arch.sie_block->gpsw.addr, &opcode, 1);
3272 ilen = insn_length(opcode);
3273 if (rc < 0) {
3274 return rc;
3275 } else if (rc) {
3276 /* Instruction-Fetching Exceptions - we can't detect the ilen.
3277 * Forward by arbitrary ilc, injection will take care of
3278 * nullification if necessary.
3279 */
3280 pgm_info = vcpu->arch.pgm;
3281 ilen = 4;
3282 }
3283 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
3284 kvm_s390_forward_psw(vcpu, ilen);
3285 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
3286}
3287
3288static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
3289{
3290 struct mcck_volatile_info *mcck_info;
3291 struct sie_page *sie_page;
3292
3293 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
3294 vcpu->arch.sie_block->icptcode);
3295 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
3296
3297 if (guestdbg_enabled(vcpu))
3298 kvm_s390_restore_guest_per_regs(vcpu);
3299
3300 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
3301 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
3302
3303 if (exit_reason == -EINTR) {
3304 VCPU_EVENT(vcpu, 3, "%s", "machine check");
3305 sie_page = container_of(vcpu->arch.sie_block,
3306 struct sie_page, sie_block);
3307 mcck_info = &sie_page->mcck_info;
3308 kvm_s390_reinject_machine_check(vcpu, mcck_info);
3309 return 0;
3310 }
3311
3312 if (vcpu->arch.sie_block->icptcode > 0) {
3313 int rc = kvm_handle_sie_intercept(vcpu);
3314
3315 if (rc != -EOPNOTSUPP)
3316 return rc;
3317 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
3318 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
3319 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
3320 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
3321 return -EREMOTE;
3322 } else if (exit_reason != -EFAULT) {
3323 vcpu->stat.exit_null++;
3324 return 0;
3325 } else if (kvm_is_ucontrol(vcpu->kvm)) {
3326 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
3327 vcpu->run->s390_ucontrol.trans_exc_code =
3328 current->thread.gmap_addr;
3329 vcpu->run->s390_ucontrol.pgm_code = 0x10;
3330 return -EREMOTE;
3331 } else if (current->thread.gmap_pfault) {
3332 trace_kvm_s390_major_guest_pfault(vcpu);
3333 current->thread.gmap_pfault = 0;
3334 if (kvm_arch_setup_async_pf(vcpu))
3335 return 0;
3336 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
3337 }
3338 return vcpu_post_run_fault_in_sie(vcpu);
3339}
3340
3341static int __vcpu_run(struct kvm_vcpu *vcpu)
3342{
3343 int rc, exit_reason;
3344
3345 /*
3346 * We try to hold kvm->srcu during most of vcpu_run (except when run-
3347 * ning the guest), so that memslots (and other stuff) are protected
3348 */
3349 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3350
3351 do {
3352 rc = vcpu_pre_run(vcpu);
3353 if (rc)
3354 break;
3355
3356 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
3357 /*
3358 * As PF_VCPU will be used in fault handler, between
3359 * guest_enter and guest_exit should be no uaccess.
3360 */
3361 local_irq_disable();
3362 guest_enter_irqoff();
3363 __disable_cpu_timer_accounting(vcpu);
3364 local_irq_enable();
3365 exit_reason = sie64a(vcpu->arch.sie_block,
3366 vcpu->run->s.regs.gprs);
3367 local_irq_disable();
3368 __enable_cpu_timer_accounting(vcpu);
3369 guest_exit_irqoff();
3370 local_irq_enable();
3371 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3372
3373 rc = vcpu_post_run(vcpu, exit_reason);
3374 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
3375
3376 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
3377 return rc;
3378}
3379
3380static void sync_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3381{
3382 struct runtime_instr_cb *riccb;
3383 struct gs_cb *gscb;
3384
3385 riccb = (struct runtime_instr_cb *) &kvm_run->s.regs.riccb;
3386 gscb = (struct gs_cb *) &kvm_run->s.regs.gscb;
3387 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
3388 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
3389 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
3390 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
3391 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
3392 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
3393 /* some control register changes require a tlb flush */
3394 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
3395 }
3396 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
3397 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
3398 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
3399 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
3400 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
3401 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
3402 }
3403 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
3404 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
3405 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
3406 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
3407 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
3408 kvm_clear_async_pf_completion_queue(vcpu);
3409 }
3410 /*
3411 * If userspace sets the riccb (e.g. after migration) to a valid state,
3412 * we should enable RI here instead of doing the lazy enablement.
3413 */
3414 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) &&
3415 test_kvm_facility(vcpu->kvm, 64) &&
3416 riccb->v &&
3417 !(vcpu->arch.sie_block->ecb3 & ECB3_RI)) {
3418 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (sync_regs)");
3419 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
3420 }
3421 /*
3422 * If userspace sets the gscb (e.g. after migration) to non-zero,
3423 * we should enable GS here instead of doing the lazy enablement.
3424 */
3425 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_GSCB) &&
3426 test_kvm_facility(vcpu->kvm, 133) &&
3427 gscb->gssm &&
3428 !vcpu->arch.gs_enabled) {
3429 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (sync_regs)");
3430 vcpu->arch.sie_block->ecb |= ECB_GS;
3431 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
3432 vcpu->arch.gs_enabled = 1;
3433 }
3434 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_BPBC) &&
3435 test_kvm_facility(vcpu->kvm, 82)) {
3436 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
3437 vcpu->arch.sie_block->fpf |= kvm_run->s.regs.bpbc ? FPF_BPBC : 0;
3438 }
3439 save_access_regs(vcpu->arch.host_acrs);
3440 restore_access_regs(vcpu->run->s.regs.acrs);
3441 /* save host (userspace) fprs/vrs */
3442 save_fpu_regs();
3443 vcpu->arch.host_fpregs.fpc = current->thread.fpu.fpc;
3444 vcpu->arch.host_fpregs.regs = current->thread.fpu.regs;
3445 if (MACHINE_HAS_VX)
3446 current->thread.fpu.regs = vcpu->run->s.regs.vrs;
3447 else
3448 current->thread.fpu.regs = vcpu->run->s.regs.fprs;
3449 current->thread.fpu.fpc = vcpu->run->s.regs.fpc;
3450 if (test_fp_ctl(current->thread.fpu.fpc))
3451 /* User space provided an invalid FPC, let's clear it */
3452 current->thread.fpu.fpc = 0;
3453 if (MACHINE_HAS_GS) {
3454 preempt_disable();
3455 __ctl_set_bit(2, 4);
3456 if (current->thread.gs_cb) {
3457 vcpu->arch.host_gscb = current->thread.gs_cb;
3458 save_gs_cb(vcpu->arch.host_gscb);
3459 }
3460 if (vcpu->arch.gs_enabled) {
3461 current->thread.gs_cb = (struct gs_cb *)
3462 &vcpu->run->s.regs.gscb;
3463 restore_gs_cb(current->thread.gs_cb);
3464 }
3465 preempt_enable();
3466 }
3467
3468 kvm_run->kvm_dirty_regs = 0;
3469}
3470
3471static void store_regs(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3472{
3473 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
3474 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
3475 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
3476 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
3477 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
3478 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
3479 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
3480 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
3481 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
3482 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
3483 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
3484 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
3485 kvm_run->s.regs.bpbc = (vcpu->arch.sie_block->fpf & FPF_BPBC) == FPF_BPBC;
3486 save_access_regs(vcpu->run->s.regs.acrs);
3487 restore_access_regs(vcpu->arch.host_acrs);
3488 /* Save guest register state */
3489 save_fpu_regs();
3490 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
3491 /* Restore will be done lazily at return */
3492 current->thread.fpu.fpc = vcpu->arch.host_fpregs.fpc;
3493 current->thread.fpu.regs = vcpu->arch.host_fpregs.regs;
3494 if (MACHINE_HAS_GS) {
3495 __ctl_set_bit(2, 4);
3496 if (vcpu->arch.gs_enabled)
3497 save_gs_cb(current->thread.gs_cb);
3498 preempt_disable();
3499 current->thread.gs_cb = vcpu->arch.host_gscb;
3500 restore_gs_cb(vcpu->arch.host_gscb);
3501 preempt_enable();
3502 if (!vcpu->arch.host_gscb)
3503 __ctl_clear_bit(2, 4);
3504 vcpu->arch.host_gscb = NULL;
3505 }
3506
3507}
3508
3509int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3510{
3511 int rc;
3512
3513 if (kvm_run->immediate_exit)
3514 return -EINTR;
3515
3516 vcpu_load(vcpu);
3517
3518 if (guestdbg_exit_pending(vcpu)) {
3519 kvm_s390_prepare_debug_exit(vcpu);
3520 rc = 0;
3521 goto out;
3522 }
3523
3524 kvm_sigset_activate(vcpu);
3525
3526 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
3527 kvm_s390_vcpu_start(vcpu);
3528 } else if (is_vcpu_stopped(vcpu)) {
3529 pr_err_ratelimited("can't run stopped vcpu %d\n",
3530 vcpu->vcpu_id);
3531 rc = -EINVAL;
3532 goto out;
3533 }
3534
3535 sync_regs(vcpu, kvm_run);
3536 enable_cpu_timer_accounting(vcpu);
3537
3538 might_fault();
3539 rc = __vcpu_run(vcpu);
3540
3541 if (signal_pending(current) && !rc) {
3542 kvm_run->exit_reason = KVM_EXIT_INTR;
3543 rc = -EINTR;
3544 }
3545
3546 if (guestdbg_exit_pending(vcpu) && !rc) {
3547 kvm_s390_prepare_debug_exit(vcpu);
3548 rc = 0;
3549 }
3550
3551 if (rc == -EREMOTE) {
3552 /* userspace support is needed, kvm_run has been prepared */
3553 rc = 0;
3554 }
3555
3556 disable_cpu_timer_accounting(vcpu);
3557 store_regs(vcpu, kvm_run);
3558
3559 kvm_sigset_deactivate(vcpu);
3560
3561 vcpu->stat.exit_userspace++;
3562out:
3563 vcpu_put(vcpu);
3564 return rc;
3565}
3566
3567/*
3568 * store status at address
3569 * we use have two special cases:
3570 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
3571 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
3572 */
3573int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
3574{
3575 unsigned char archmode = 1;
3576 freg_t fprs[NUM_FPRS];
3577 unsigned int px;
3578 u64 clkcomp, cputm;
3579 int rc;
3580
3581 px = kvm_s390_get_prefix(vcpu);
3582 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
3583 if (write_guest_abs(vcpu, 163, &archmode, 1))
3584 return -EFAULT;
3585 gpa = 0;
3586 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
3587 if (write_guest_real(vcpu, 163, &archmode, 1))
3588 return -EFAULT;
3589 gpa = px;
3590 } else
3591 gpa -= __LC_FPREGS_SAVE_AREA;
3592
3593 /* manually convert vector registers if necessary */
3594 if (MACHINE_HAS_VX) {
3595 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
3596 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
3597 fprs, 128);
3598 } else {
3599 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
3600 vcpu->run->s.regs.fprs, 128);
3601 }
3602 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
3603 vcpu->run->s.regs.gprs, 128);
3604 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
3605 &vcpu->arch.sie_block->gpsw, 16);
3606 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
3607 &px, 4);
3608 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
3609 &vcpu->run->s.regs.fpc, 4);
3610 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
3611 &vcpu->arch.sie_block->todpr, 4);
3612 cputm = kvm_s390_get_cpu_timer(vcpu);
3613 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
3614 &cputm, 8);
3615 clkcomp = vcpu->arch.sie_block->ckc >> 8;
3616 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
3617 &clkcomp, 8);
3618 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
3619 &vcpu->run->s.regs.acrs, 64);
3620 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
3621 &vcpu->arch.sie_block->gcr, 128);
3622 return rc ? -EFAULT : 0;
3623}
3624
3625int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
3626{
3627 /*
3628 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
3629 * switch in the run ioctl. Let's update our copies before we save
3630 * it into the save area
3631 */
3632 save_fpu_regs();
3633 vcpu->run->s.regs.fpc = current->thread.fpu.fpc;
3634 save_access_regs(vcpu->run->s.regs.acrs);
3635
3636 return kvm_s390_store_status_unloaded(vcpu, addr);
3637}
3638
3639static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
3640{
3641 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
3642 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
3643}
3644
3645static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
3646{
3647 unsigned int i;
3648 struct kvm_vcpu *vcpu;
3649
3650 kvm_for_each_vcpu(i, vcpu, kvm) {
3651 __disable_ibs_on_vcpu(vcpu);
3652 }
3653}
3654
3655static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
3656{
3657 if (!sclp.has_ibs)
3658 return;
3659 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
3660 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
3661}
3662
3663void kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
3664{
3665 int i, online_vcpus, started_vcpus = 0;
3666
3667 if (!is_vcpu_stopped(vcpu))
3668 return;
3669
3670 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
3671 /* Only one cpu at a time may enter/leave the STOPPED state. */
3672 spin_lock(&vcpu->kvm->arch.start_stop_lock);
3673 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
3674
3675 for (i = 0; i < online_vcpus; i++) {
3676 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i]))
3677 started_vcpus++;
3678 }
3679
3680 if (started_vcpus == 0) {
3681 /* we're the only active VCPU -> speed it up */
3682 __enable_ibs_on_vcpu(vcpu);
3683 } else if (started_vcpus == 1) {
3684 /*
3685 * As we are starting a second VCPU, we have to disable
3686 * the IBS facility on all VCPUs to remove potentially
3687 * oustanding ENABLE requests.
3688 */
3689 __disable_ibs_on_all_vcpus(vcpu->kvm);
3690 }
3691
3692 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_STOPPED);
3693 /*
3694 * Another VCPU might have used IBS while we were offline.
3695 * Let's play safe and flush the VCPU at startup.
3696 */
3697 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
3698 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
3699 return;
3700}
3701
3702void kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
3703{
3704 int i, online_vcpus, started_vcpus = 0;
3705 struct kvm_vcpu *started_vcpu = NULL;
3706
3707 if (is_vcpu_stopped(vcpu))
3708 return;
3709
3710 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
3711 /* Only one cpu at a time may enter/leave the STOPPED state. */
3712 spin_lock(&vcpu->kvm->arch.start_stop_lock);
3713 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
3714
3715 /* SIGP STOP and SIGP STOP AND STORE STATUS has been fully processed */
3716 kvm_s390_clear_stop_irq(vcpu);
3717
3718 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOPPED);
3719 __disable_ibs_on_vcpu(vcpu);
3720
3721 for (i = 0; i < online_vcpus; i++) {
3722 if (!is_vcpu_stopped(vcpu->kvm->vcpus[i])) {
3723 started_vcpus++;
3724 started_vcpu = vcpu->kvm->vcpus[i];
3725 }
3726 }
3727
3728 if (started_vcpus == 1) {
3729 /*
3730 * As we only have one VCPU left, we want to enable the
3731 * IBS facility for that VCPU to speed it up.
3732 */
3733 __enable_ibs_on_vcpu(started_vcpu);
3734 }
3735
3736 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
3737 return;
3738}
3739
3740static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3741 struct kvm_enable_cap *cap)
3742{
3743 int r;
3744
3745 if (cap->flags)
3746 return -EINVAL;
3747
3748 switch (cap->cap) {
3749 case KVM_CAP_S390_CSS_SUPPORT:
3750 if (!vcpu->kvm->arch.css_support) {
3751 vcpu->kvm->arch.css_support = 1;
3752 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
3753 trace_kvm_s390_enable_css(vcpu->kvm);
3754 }
3755 r = 0;
3756 break;
3757 default:
3758 r = -EINVAL;
3759 break;
3760 }
3761 return r;
3762}
3763
3764static long kvm_s390_guest_mem_op(struct kvm_vcpu *vcpu,
3765 struct kvm_s390_mem_op *mop)
3766{
3767 void __user *uaddr = (void __user *)mop->buf;
3768 void *tmpbuf = NULL;
3769 int r, srcu_idx;
3770 const u64 supported_flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION
3771 | KVM_S390_MEMOP_F_CHECK_ONLY;
3772
3773 if (mop->flags & ~supported_flags)
3774 return -EINVAL;
3775
3776 if (mop->size > MEM_OP_MAX_SIZE)
3777 return -E2BIG;
3778
3779 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
3780 tmpbuf = vmalloc(mop->size);
3781 if (!tmpbuf)
3782 return -ENOMEM;
3783 }
3784
3785 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3786
3787 switch (mop->op) {
3788 case KVM_S390_MEMOP_LOGICAL_READ:
3789 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3790 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3791 mop->size, GACC_FETCH);
3792 break;
3793 }
3794 r = read_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3795 if (r == 0) {
3796 if (copy_to_user(uaddr, tmpbuf, mop->size))
3797 r = -EFAULT;
3798 }
3799 break;
3800 case KVM_S390_MEMOP_LOGICAL_WRITE:
3801 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
3802 r = check_gva_range(vcpu, mop->gaddr, mop->ar,
3803 mop->size, GACC_STORE);
3804 break;
3805 }
3806 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
3807 r = -EFAULT;
3808 break;
3809 }
3810 r = write_guest(vcpu, mop->gaddr, mop->ar, tmpbuf, mop->size);
3811 break;
3812 default:
3813 r = -EINVAL;
3814 }
3815
3816 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
3817
3818 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
3819 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
3820
3821 vfree(tmpbuf);
3822 return r;
3823}
3824
3825long kvm_arch_vcpu_async_ioctl(struct file *filp,
3826 unsigned int ioctl, unsigned long arg)
3827{
3828 struct kvm_vcpu *vcpu = filp->private_data;
3829 void __user *argp = (void __user *)arg;
3830
3831 switch (ioctl) {
3832 case KVM_S390_IRQ: {
3833 struct kvm_s390_irq s390irq;
3834
3835 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
3836 return -EFAULT;
3837 return kvm_s390_inject_vcpu(vcpu, &s390irq);
3838 }
3839 case KVM_S390_INTERRUPT: {
3840 struct kvm_s390_interrupt s390int;
3841 struct kvm_s390_irq s390irq;
3842
3843 if (copy_from_user(&s390int, argp, sizeof(s390int)))
3844 return -EFAULT;
3845 if (s390int_to_s390irq(&s390int, &s390irq))
3846 return -EINVAL;
3847 return kvm_s390_inject_vcpu(vcpu, &s390irq);
3848 }
3849 }
3850 return -ENOIOCTLCMD;
3851}
3852
3853long kvm_arch_vcpu_ioctl(struct file *filp,
3854 unsigned int ioctl, unsigned long arg)
3855{
3856 struct kvm_vcpu *vcpu = filp->private_data;
3857 void __user *argp = (void __user *)arg;
3858 int idx;
3859 long r;
3860
3861 vcpu_load(vcpu);
3862
3863 switch (ioctl) {
3864 case KVM_S390_STORE_STATUS:
3865 idx = srcu_read_lock(&vcpu->kvm->srcu);
3866 r = kvm_s390_vcpu_store_status(vcpu, arg);
3867 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3868 break;
3869 case KVM_S390_SET_INITIAL_PSW: {
3870 psw_t psw;
3871
3872 r = -EFAULT;
3873 if (copy_from_user(&psw, argp, sizeof(psw)))
3874 break;
3875 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
3876 break;
3877 }
3878 case KVM_S390_INITIAL_RESET:
3879 r = kvm_arch_vcpu_ioctl_initial_reset(vcpu);
3880 break;
3881 case KVM_SET_ONE_REG:
3882 case KVM_GET_ONE_REG: {
3883 struct kvm_one_reg reg;
3884 r = -EFAULT;
3885 if (copy_from_user(®, argp, sizeof(reg)))
3886 break;
3887 if (ioctl == KVM_SET_ONE_REG)
3888 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®);
3889 else
3890 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®);
3891 break;
3892 }
3893#ifdef CONFIG_KVM_S390_UCONTROL
3894 case KVM_S390_UCAS_MAP: {
3895 struct kvm_s390_ucas_mapping ucasmap;
3896
3897 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3898 r = -EFAULT;
3899 break;
3900 }
3901
3902 if (!kvm_is_ucontrol(vcpu->kvm)) {
3903 r = -EINVAL;
3904 break;
3905 }
3906
3907 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
3908 ucasmap.vcpu_addr, ucasmap.length);
3909 break;
3910 }
3911 case KVM_S390_UCAS_UNMAP: {
3912 struct kvm_s390_ucas_mapping ucasmap;
3913
3914 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
3915 r = -EFAULT;
3916 break;
3917 }
3918
3919 if (!kvm_is_ucontrol(vcpu->kvm)) {
3920 r = -EINVAL;
3921 break;
3922 }
3923
3924 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
3925 ucasmap.length);
3926 break;
3927 }
3928#endif
3929 case KVM_S390_VCPU_FAULT: {
3930 r = gmap_fault(vcpu->arch.gmap, arg, 0);
3931 break;
3932 }
3933 case KVM_ENABLE_CAP:
3934 {
3935 struct kvm_enable_cap cap;
3936 r = -EFAULT;
3937 if (copy_from_user(&cap, argp, sizeof(cap)))
3938 break;
3939 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3940 break;
3941 }
3942 case KVM_S390_MEM_OP: {
3943 struct kvm_s390_mem_op mem_op;
3944
3945 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
3946 r = kvm_s390_guest_mem_op(vcpu, &mem_op);
3947 else
3948 r = -EFAULT;
3949 break;
3950 }
3951 case KVM_S390_SET_IRQ_STATE: {
3952 struct kvm_s390_irq_state irq_state;
3953
3954 r = -EFAULT;
3955 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3956 break;
3957 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
3958 irq_state.len == 0 ||
3959 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
3960 r = -EINVAL;
3961 break;
3962 }
3963 /* do not use irq_state.flags, it will break old QEMUs */
3964 r = kvm_s390_set_irq_state(vcpu,
3965 (void __user *) irq_state.buf,
3966 irq_state.len);
3967 break;
3968 }
3969 case KVM_S390_GET_IRQ_STATE: {
3970 struct kvm_s390_irq_state irq_state;
3971
3972 r = -EFAULT;
3973 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
3974 break;
3975 if (irq_state.len == 0) {
3976 r = -EINVAL;
3977 break;
3978 }
3979 /* do not use irq_state.flags, it will break old QEMUs */
3980 r = kvm_s390_get_irq_state(vcpu,
3981 (__u8 __user *) irq_state.buf,
3982 irq_state.len);
3983 break;
3984 }
3985 default:
3986 r = -ENOTTY;
3987 }
3988
3989 vcpu_put(vcpu);
3990 return r;
3991}
3992
3993int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3994{
3995#ifdef CONFIG_KVM_S390_UCONTROL
3996 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
3997 && (kvm_is_ucontrol(vcpu->kvm))) {
3998 vmf->page = virt_to_page(vcpu->arch.sie_block);
3999 get_page(vmf->page);
4000 return 0;
4001 }
4002#endif
4003 return VM_FAULT_SIGBUS;
4004}
4005
4006int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
4007 unsigned long npages)
4008{
4009 return 0;
4010}
4011
4012/* Section: memory related */
4013int kvm_arch_prepare_memory_region(struct kvm *kvm,
4014 struct kvm_memory_slot *memslot,
4015 const struct kvm_userspace_memory_region *mem,
4016 enum kvm_mr_change change)
4017{
4018 /* A few sanity checks. We can have memory slots which have to be
4019 located/ended at a segment boundary (1MB). The memory in userland is
4020 ok to be fragmented into various different vmas. It is okay to mmap()
4021 and munmap() stuff in this slot after doing this call at any time */
4022
4023 if (mem->userspace_addr & 0xffffful)
4024 return -EINVAL;
4025
4026 if (mem->memory_size & 0xffffful)
4027 return -EINVAL;
4028
4029 if (mem->guest_phys_addr + mem->memory_size > kvm->arch.mem_limit)
4030 return -EINVAL;
4031
4032 return 0;
4033}
4034
4035void kvm_arch_commit_memory_region(struct kvm *kvm,
4036 const struct kvm_userspace_memory_region *mem,
4037 const struct kvm_memory_slot *old,
4038 const struct kvm_memory_slot *new,
4039 enum kvm_mr_change change)
4040{
4041 int rc;
4042
4043 /* If the basics of the memslot do not change, we do not want
4044 * to update the gmap. Every update causes several unnecessary
4045 * segment translation exceptions. This is usually handled just
4046 * fine by the normal fault handler + gmap, but it will also
4047 * cause faults on the prefix page of running guest CPUs.
4048 */
4049 if (old->userspace_addr == mem->userspace_addr &&
4050 old->base_gfn * PAGE_SIZE == mem->guest_phys_addr &&
4051 old->npages * PAGE_SIZE == mem->memory_size)
4052 return;
4053
4054 rc = gmap_map_segment(kvm->arch.gmap, mem->userspace_addr,
4055 mem->guest_phys_addr, mem->memory_size);
4056 if (rc)
4057 pr_warn("failed to commit memory region\n");
4058 return;
4059}
4060
4061static inline unsigned long nonhyp_mask(int i)
4062{
4063 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
4064
4065 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
4066}
4067
4068void kvm_arch_vcpu_block_finish(struct kvm_vcpu *vcpu)
4069{
4070 vcpu->valid_wakeup = false;
4071}
4072
4073static int __init kvm_s390_init(void)
4074{
4075 int i;
4076
4077 if (!sclp.has_sief2) {
4078 pr_info("SIE not available\n");
4079 return -ENODEV;
4080 }
4081
4082 for (i = 0; i < 16; i++)
4083 kvm_s390_fac_base[i] |=
4084 S390_lowcore.stfle_fac_list[i] & nonhyp_mask(i);
4085
4086 return kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
4087}
4088
4089static void __exit kvm_s390_exit(void)
4090{
4091 kvm_exit();
4092}
4093
4094module_init(kvm_s390_init);
4095module_exit(kvm_s390_exit);
4096
4097/*
4098 * Enable autoloading of the kvm module.
4099 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
4100 * since x86 takes a different approach.
4101 */
4102#include <linux/miscdevice.h>
4103MODULE_ALIAS_MISCDEV(KVM_MINOR);
4104MODULE_ALIAS("devname:kvm");