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/access-regs.h>
37 #include <asm/asm-offsets.h>
38 #include <asm/lowcore.h>
39 #include <asm/stp.h>
40 #include <asm/gmap.h>
41 #include <asm/nmi.h>
42 #include <asm/isc.h>
43 #include <asm/sclp.h>
44 #include <asm/cpacf.h>
45 #include <asm/timex.h>
46 #include <asm/fpu.h>
47 #include <asm/ap.h>
48 #include <asm/uv.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
62 const 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 STATS_DESC_COUNTER(VM, gmap_shadow_reuse),
71 STATS_DESC_COUNTER(VM, gmap_shadow_create),
72 STATS_DESC_COUNTER(VM, gmap_shadow_r1_entry),
73 STATS_DESC_COUNTER(VM, gmap_shadow_r2_entry),
74 STATS_DESC_COUNTER(VM, gmap_shadow_r3_entry),
75 STATS_DESC_COUNTER(VM, gmap_shadow_sg_entry),
76 STATS_DESC_COUNTER(VM, gmap_shadow_pg_entry),
77 };
78
79 const struct kvm_stats_header kvm_vm_stats_header = {
80 .name_size = KVM_STATS_NAME_SIZE,
81 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
82 .id_offset = sizeof(struct kvm_stats_header),
83 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
84 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
85 sizeof(kvm_vm_stats_desc),
86 };
87
88 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
89 KVM_GENERIC_VCPU_STATS(),
90 STATS_DESC_COUNTER(VCPU, exit_userspace),
91 STATS_DESC_COUNTER(VCPU, exit_null),
92 STATS_DESC_COUNTER(VCPU, exit_external_request),
93 STATS_DESC_COUNTER(VCPU, exit_io_request),
94 STATS_DESC_COUNTER(VCPU, exit_external_interrupt),
95 STATS_DESC_COUNTER(VCPU, exit_stop_request),
96 STATS_DESC_COUNTER(VCPU, exit_validity),
97 STATS_DESC_COUNTER(VCPU, exit_instruction),
98 STATS_DESC_COUNTER(VCPU, exit_pei),
99 STATS_DESC_COUNTER(VCPU, halt_no_poll_steal),
100 STATS_DESC_COUNTER(VCPU, instruction_lctl),
101 STATS_DESC_COUNTER(VCPU, instruction_lctlg),
102 STATS_DESC_COUNTER(VCPU, instruction_stctl),
103 STATS_DESC_COUNTER(VCPU, instruction_stctg),
104 STATS_DESC_COUNTER(VCPU, exit_program_interruption),
105 STATS_DESC_COUNTER(VCPU, exit_instr_and_program),
106 STATS_DESC_COUNTER(VCPU, exit_operation_exception),
107 STATS_DESC_COUNTER(VCPU, deliver_ckc),
108 STATS_DESC_COUNTER(VCPU, deliver_cputm),
109 STATS_DESC_COUNTER(VCPU, deliver_external_call),
110 STATS_DESC_COUNTER(VCPU, deliver_emergency_signal),
111 STATS_DESC_COUNTER(VCPU, deliver_service_signal),
112 STATS_DESC_COUNTER(VCPU, deliver_virtio),
113 STATS_DESC_COUNTER(VCPU, deliver_stop_signal),
114 STATS_DESC_COUNTER(VCPU, deliver_prefix_signal),
115 STATS_DESC_COUNTER(VCPU, deliver_restart_signal),
116 STATS_DESC_COUNTER(VCPU, deliver_program),
117 STATS_DESC_COUNTER(VCPU, deliver_io),
118 STATS_DESC_COUNTER(VCPU, deliver_machine_check),
119 STATS_DESC_COUNTER(VCPU, exit_wait_state),
120 STATS_DESC_COUNTER(VCPU, inject_ckc),
121 STATS_DESC_COUNTER(VCPU, inject_cputm),
122 STATS_DESC_COUNTER(VCPU, inject_external_call),
123 STATS_DESC_COUNTER(VCPU, inject_emergency_signal),
124 STATS_DESC_COUNTER(VCPU, inject_mchk),
125 STATS_DESC_COUNTER(VCPU, inject_pfault_init),
126 STATS_DESC_COUNTER(VCPU, inject_program),
127 STATS_DESC_COUNTER(VCPU, inject_restart),
128 STATS_DESC_COUNTER(VCPU, inject_set_prefix),
129 STATS_DESC_COUNTER(VCPU, inject_stop_signal),
130 STATS_DESC_COUNTER(VCPU, instruction_epsw),
131 STATS_DESC_COUNTER(VCPU, instruction_gs),
132 STATS_DESC_COUNTER(VCPU, instruction_io_other),
133 STATS_DESC_COUNTER(VCPU, instruction_lpsw),
134 STATS_DESC_COUNTER(VCPU, instruction_lpswe),
135 STATS_DESC_COUNTER(VCPU, instruction_lpswey),
136 STATS_DESC_COUNTER(VCPU, instruction_pfmf),
137 STATS_DESC_COUNTER(VCPU, instruction_ptff),
138 STATS_DESC_COUNTER(VCPU, instruction_sck),
139 STATS_DESC_COUNTER(VCPU, instruction_sckpf),
140 STATS_DESC_COUNTER(VCPU, instruction_stidp),
141 STATS_DESC_COUNTER(VCPU, instruction_spx),
142 STATS_DESC_COUNTER(VCPU, instruction_stpx),
143 STATS_DESC_COUNTER(VCPU, instruction_stap),
144 STATS_DESC_COUNTER(VCPU, instruction_iske),
145 STATS_DESC_COUNTER(VCPU, instruction_ri),
146 STATS_DESC_COUNTER(VCPU, instruction_rrbe),
147 STATS_DESC_COUNTER(VCPU, instruction_sske),
148 STATS_DESC_COUNTER(VCPU, instruction_ipte_interlock),
149 STATS_DESC_COUNTER(VCPU, instruction_stsi),
150 STATS_DESC_COUNTER(VCPU, instruction_stfl),
151 STATS_DESC_COUNTER(VCPU, instruction_tb),
152 STATS_DESC_COUNTER(VCPU, instruction_tpi),
153 STATS_DESC_COUNTER(VCPU, instruction_tprot),
154 STATS_DESC_COUNTER(VCPU, instruction_tsch),
155 STATS_DESC_COUNTER(VCPU, instruction_sie),
156 STATS_DESC_COUNTER(VCPU, instruction_essa),
157 STATS_DESC_COUNTER(VCPU, instruction_sthyi),
158 STATS_DESC_COUNTER(VCPU, instruction_sigp_sense),
159 STATS_DESC_COUNTER(VCPU, instruction_sigp_sense_running),
160 STATS_DESC_COUNTER(VCPU, instruction_sigp_external_call),
161 STATS_DESC_COUNTER(VCPU, instruction_sigp_emergency),
162 STATS_DESC_COUNTER(VCPU, instruction_sigp_cond_emergency),
163 STATS_DESC_COUNTER(VCPU, instruction_sigp_start),
164 STATS_DESC_COUNTER(VCPU, instruction_sigp_stop),
165 STATS_DESC_COUNTER(VCPU, instruction_sigp_stop_store_status),
166 STATS_DESC_COUNTER(VCPU, instruction_sigp_store_status),
167 STATS_DESC_COUNTER(VCPU, instruction_sigp_store_adtl_status),
168 STATS_DESC_COUNTER(VCPU, instruction_sigp_arch),
169 STATS_DESC_COUNTER(VCPU, instruction_sigp_prefix),
170 STATS_DESC_COUNTER(VCPU, instruction_sigp_restart),
171 STATS_DESC_COUNTER(VCPU, instruction_sigp_init_cpu_reset),
172 STATS_DESC_COUNTER(VCPU, instruction_sigp_cpu_reset),
173 STATS_DESC_COUNTER(VCPU, instruction_sigp_unknown),
174 STATS_DESC_COUNTER(VCPU, instruction_diagnose_10),
175 STATS_DESC_COUNTER(VCPU, instruction_diagnose_44),
176 STATS_DESC_COUNTER(VCPU, instruction_diagnose_9c),
177 STATS_DESC_COUNTER(VCPU, diag_9c_ignored),
178 STATS_DESC_COUNTER(VCPU, diag_9c_forward),
179 STATS_DESC_COUNTER(VCPU, instruction_diagnose_258),
180 STATS_DESC_COUNTER(VCPU, instruction_diagnose_308),
181 STATS_DESC_COUNTER(VCPU, instruction_diagnose_500),
182 STATS_DESC_COUNTER(VCPU, instruction_diagnose_other),
183 STATS_DESC_COUNTER(VCPU, pfault_sync)
184 };
185
186 const struct kvm_stats_header kvm_vcpu_stats_header = {
187 .name_size = KVM_STATS_NAME_SIZE,
188 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
189 .id_offset = sizeof(struct kvm_stats_header),
190 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
191 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
192 sizeof(kvm_vcpu_stats_desc),
193 };
194
195 /* allow nested virtualization in KVM (if enabled by user space) */
196 static int nested;
197 module_param(nested, int, S_IRUGO);
198 MODULE_PARM_DESC(nested, "Nested virtualization support");
199
200 /* allow 1m huge page guest backing, if !nested */
201 static int hpage;
202 module_param(hpage, int, 0444);
203 MODULE_PARM_DESC(hpage, "1m huge page backing support");
204
205 /* maximum percentage of steal time for polling. >100 is treated like 100 */
206 static u8 halt_poll_max_steal = 10;
207 module_param(halt_poll_max_steal, byte, 0644);
208 MODULE_PARM_DESC(halt_poll_max_steal, "Maximum percentage of steal time to allow polling");
209
210 /* if set to true, the GISA will be initialized and used if available */
211 static bool use_gisa = true;
212 module_param(use_gisa, bool, 0644);
213 MODULE_PARM_DESC(use_gisa, "Use the GISA if the host supports it.");
214
215 /* maximum diag9c forwarding per second */
216 unsigned int diag9c_forwarding_hz;
217 module_param(diag9c_forwarding_hz, uint, 0644);
218 MODULE_PARM_DESC(diag9c_forwarding_hz, "Maximum diag9c forwarding per second, 0 to turn off");
219
220 /*
221 * allow asynchronous deinit for protected guests; enable by default since
222 * the feature is opt-in anyway
223 */
224 static int async_destroy = 1;
225 module_param(async_destroy, int, 0444);
226 MODULE_PARM_DESC(async_destroy, "Asynchronous destroy for protected guests");
227
228 /*
229 * For now we handle at most 16 double words as this is what the s390 base
230 * kernel handles and stores in the prefix page. If we ever need to go beyond
231 * this, this requires changes to code, but the external uapi can stay.
232 */
233 #define SIZE_INTERNAL 16
234
235 /*
236 * Base feature mask that defines default mask for facilities. Consists of the
237 * defines in FACILITIES_KVM and the non-hypervisor managed bits.
238 */
239 static unsigned long kvm_s390_fac_base[SIZE_INTERNAL] = { FACILITIES_KVM };
240 /*
241 * Extended feature mask. Consists of the defines in FACILITIES_KVM_CPUMODEL
242 * and defines the facilities that can be enabled via a cpu model.
243 */
244 static unsigned long kvm_s390_fac_ext[SIZE_INTERNAL] = { FACILITIES_KVM_CPUMODEL };
245
246 static unsigned long kvm_s390_fac_size(void)
247 {
248 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_MASK_SIZE_U64);
249 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_LIST_SIZE_U64);
250 BUILD_BUG_ON(SIZE_INTERNAL * sizeof(unsigned long) >
251 sizeof(stfle_fac_list));
252
253 return SIZE_INTERNAL;
254 }
255
256 /* available cpu features supported by kvm */
257 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
258 /* available subfunctions indicated via query / "test bit" */
259 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
260
261 static struct gmap_notifier gmap_notifier;
262 static struct gmap_notifier vsie_gmap_notifier;
263 debug_info_t *kvm_s390_dbf;
264 debug_info_t *kvm_s390_dbf_uv;
265
266 /* Section: not file related */
267 /* forward declarations */
268 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
269 unsigned long end);
270 static int sca_switch_to_extended(struct kvm *kvm);
271
272 static void kvm_clock_sync_scb(struct kvm_s390_sie_block *scb, u64 delta)
273 {
274 u8 delta_idx = 0;
275
276 /*
277 * The TOD jumps by delta, we have to compensate this by adding
278 * -delta to the epoch.
279 */
280 delta = -delta;
281
282 /* sign-extension - we're adding to signed values below */
283 if ((s64)delta < 0)
284 delta_idx = -1;
285
286 scb->epoch += delta;
287 if (scb->ecd & ECD_MEF) {
288 scb->epdx += delta_idx;
289 if (scb->epoch < delta)
290 scb->epdx += 1;
291 }
292 }
293
294 /*
295 * This callback is executed during stop_machine(). All CPUs are therefore
296 * temporarily stopped. In order not to change guest behavior, we have to
297 * disable preemption whenever we touch the epoch of kvm and the VCPUs,
298 * so a CPU won't be stopped while calculating with the epoch.
299 */
300 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
301 void *v)
302 {
303 struct kvm *kvm;
304 struct kvm_vcpu *vcpu;
305 unsigned long i;
306 unsigned long long *delta = v;
307
308 list_for_each_entry(kvm, &vm_list, vm_list) {
309 kvm_for_each_vcpu(i, vcpu, kvm) {
310 kvm_clock_sync_scb(vcpu->arch.sie_block, *delta);
311 if (i == 0) {
312 kvm->arch.epoch = vcpu->arch.sie_block->epoch;
313 kvm->arch.epdx = vcpu->arch.sie_block->epdx;
314 }
315 if (vcpu->arch.cputm_enabled)
316 vcpu->arch.cputm_start += *delta;
317 if (vcpu->arch.vsie_block)
318 kvm_clock_sync_scb(vcpu->arch.vsie_block,
319 *delta);
320 }
321 }
322 return NOTIFY_OK;
323 }
324
325 static struct notifier_block kvm_clock_notifier = {
326 .notifier_call = kvm_clock_sync,
327 };
328
329 static void allow_cpu_feat(unsigned long nr)
330 {
331 set_bit_inv(nr, kvm_s390_available_cpu_feat);
332 }
333
334 static inline int plo_test_bit(unsigned char nr)
335 {
336 unsigned long function = (unsigned long)nr | 0x100;
337 int cc;
338
339 asm volatile(
340 " lgr 0,%[function]\n"
341 /* Parameter registers are ignored for "test bit" */
342 " plo 0,0,0,0(0)\n"
343 " ipm %0\n"
344 " srl %0,28\n"
345 : "=d" (cc)
346 : [function] "d" (function)
347 : "cc", "");
348 return cc == 0;
349 }
350
351 static __always_inline void __insn32_query(unsigned int opcode, u8 *query)
352 {
353 asm volatile(
354 " lghi 0,0\n"
355 " lgr 1,%[query]\n"
356 /* Parameter registers are ignored */
357 " .insn rrf,%[opc] << 16,2,4,6,0\n"
358 :
359 : [query] "d" ((unsigned long)query), [opc] "i" (opcode)
360 : "cc", "memory", "", "1");
361 }
362
363 #define INSN_SORTL 0xb938
364 #define INSN_DFLTCC 0xb939
365
366 static void __init kvm_s390_cpu_feat_init(void)
367 {
368 int i;
369
370 for (i = 0; i < 256; ++i) {
371 if (plo_test_bit(i))
372 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
373 }
374
375 if (test_facility(28)) /* TOD-clock steering */
376 ptff(kvm_s390_available_subfunc.ptff,
377 sizeof(kvm_s390_available_subfunc.ptff),
378 PTFF_QAF);
379
380 if (test_facility(17)) { /* MSA */
381 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
382 kvm_s390_available_subfunc.kmac);
383 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
384 kvm_s390_available_subfunc.kmc);
385 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
386 kvm_s390_available_subfunc.km);
387 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
388 kvm_s390_available_subfunc.kimd);
389 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
390 kvm_s390_available_subfunc.klmd);
391 }
392 if (test_facility(76)) /* MSA3 */
393 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
394 kvm_s390_available_subfunc.pckmo);
395 if (test_facility(77)) { /* MSA4 */
396 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
397 kvm_s390_available_subfunc.kmctr);
398 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
399 kvm_s390_available_subfunc.kmf);
400 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
401 kvm_s390_available_subfunc.kmo);
402 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
403 kvm_s390_available_subfunc.pcc);
404 }
405 if (test_facility(57)) /* MSA5 */
406 __cpacf_query(CPACF_PRNO, (cpacf_mask_t *)
407 kvm_s390_available_subfunc.ppno);
408
409 if (test_facility(146)) /* MSA8 */
410 __cpacf_query(CPACF_KMA, (cpacf_mask_t *)
411 kvm_s390_available_subfunc.kma);
412
413 if (test_facility(155)) /* MSA9 */
414 __cpacf_query(CPACF_KDSA, (cpacf_mask_t *)
415 kvm_s390_available_subfunc.kdsa);
416
417 if (test_facility(150)) /* SORTL */
418 __insn32_query(INSN_SORTL, kvm_s390_available_subfunc.sortl);
419
420 if (test_facility(151)) /* DFLTCC */
421 __insn32_query(INSN_DFLTCC, kvm_s390_available_subfunc.dfltcc);
422
423 if (MACHINE_HAS_ESOP)
424 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
425 /*
426 * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
427 * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
428 */
429 if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
430 !test_facility(3) || !nested)
431 return;
432 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
433 if (sclp.has_64bscao)
434 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
435 if (sclp.has_siif)
436 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
437 if (sclp.has_gpere)
438 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
439 if (sclp.has_gsls)
440 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
441 if (sclp.has_ib)
442 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
443 if (sclp.has_cei)
444 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
445 if (sclp.has_ibs)
446 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
447 if (sclp.has_kss)
448 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_KSS);
449 /*
450 * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
451 * all skey handling functions read/set the skey from the PGSTE
452 * instead of the real storage key.
453 *
454 * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
455 * pages being detected as preserved although they are resident.
456 *
457 * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
458 * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
459 *
460 * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
461 * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
462 * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
463 *
464 * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
465 * cannot easily shadow the SCA because of the ipte lock.
466 */
467 }
468
469 static int __init __kvm_s390_init(void)
470 {
471 int rc = -ENOMEM;
472
473 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
474 if (!kvm_s390_dbf)
475 return -ENOMEM;
476
477 kvm_s390_dbf_uv = debug_register("kvm-uv", 32, 1, 7 * sizeof(long));
478 if (!kvm_s390_dbf_uv)
479 goto err_kvm_uv;
480
481 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view) ||
482 debug_register_view(kvm_s390_dbf_uv, &debug_sprintf_view))
483 goto err_debug_view;
484
485 kvm_s390_cpu_feat_init();
486
487 /* Register floating interrupt controller interface. */
488 rc = kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
489 if (rc) {
490 pr_err("A FLIC registration call failed with rc=%d\n", rc);
491 goto err_flic;
492 }
493
494 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
495 rc = kvm_s390_pci_init();
496 if (rc) {
497 pr_err("Unable to allocate AIFT for PCI\n");
498 goto err_pci;
499 }
500 }
501
502 rc = kvm_s390_gib_init(GAL_ISC);
503 if (rc)
504 goto err_gib;
505
506 gmap_notifier.notifier_call = kvm_gmap_notifier;
507 gmap_register_pte_notifier(&gmap_notifier);
508 vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
509 gmap_register_pte_notifier(&vsie_gmap_notifier);
510 atomic_notifier_chain_register(&s390_epoch_delta_notifier,
511 &kvm_clock_notifier);
512
513 return 0;
514
515 err_gib:
516 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
517 kvm_s390_pci_exit();
518 err_pci:
519 err_flic:
520 err_debug_view:
521 debug_unregister(kvm_s390_dbf_uv);
522 err_kvm_uv:
523 debug_unregister(kvm_s390_dbf);
524 return rc;
525 }
526
527 static void __kvm_s390_exit(void)
528 {
529 gmap_unregister_pte_notifier(&gmap_notifier);
530 gmap_unregister_pte_notifier(&vsie_gmap_notifier);
531 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
532 &kvm_clock_notifier);
533
534 kvm_s390_gib_destroy();
535 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
536 kvm_s390_pci_exit();
537 debug_unregister(kvm_s390_dbf);
538 debug_unregister(kvm_s390_dbf_uv);
539 }
540
541 /* Section: device related */
542 long kvm_arch_dev_ioctl(struct file *filp,
543 unsigned int ioctl, unsigned long arg)
544 {
545 if (ioctl == KVM_S390_ENABLE_SIE)
546 return s390_enable_sie();
547 return -EINVAL;
548 }
549
550 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
551 {
552 int r;
553
554 switch (ext) {
555 case KVM_CAP_S390_PSW:
556 case KVM_CAP_S390_GMAP:
557 case KVM_CAP_SYNC_MMU:
558 #ifdef CONFIG_KVM_S390_UCONTROL
559 case KVM_CAP_S390_UCONTROL:
560 #endif
561 case KVM_CAP_ASYNC_PF:
562 case KVM_CAP_SYNC_REGS:
563 case KVM_CAP_ONE_REG:
564 case KVM_CAP_ENABLE_CAP:
565 case KVM_CAP_S390_CSS_SUPPORT:
566 case KVM_CAP_IOEVENTFD:
567 case KVM_CAP_S390_IRQCHIP:
568 case KVM_CAP_VM_ATTRIBUTES:
569 case KVM_CAP_MP_STATE:
570 case KVM_CAP_IMMEDIATE_EXIT:
571 case KVM_CAP_S390_INJECT_IRQ:
572 case KVM_CAP_S390_USER_SIGP:
573 case KVM_CAP_S390_USER_STSI:
574 case KVM_CAP_S390_SKEYS:
575 case KVM_CAP_S390_IRQ_STATE:
576 case KVM_CAP_S390_USER_INSTR0:
577 case KVM_CAP_S390_CMMA_MIGRATION:
578 case KVM_CAP_S390_AIS:
579 case KVM_CAP_S390_AIS_MIGRATION:
580 case KVM_CAP_S390_VCPU_RESETS:
581 case KVM_CAP_SET_GUEST_DEBUG:
582 case KVM_CAP_S390_DIAG318:
583 case KVM_CAP_IRQFD_RESAMPLE:
584 r = 1;
585 break;
586 case KVM_CAP_SET_GUEST_DEBUG2:
587 r = KVM_GUESTDBG_VALID_MASK;
588 break;
589 case KVM_CAP_S390_HPAGE_1M:
590 r = 0;
591 if (hpage && !(kvm && kvm_is_ucontrol(kvm)))
592 r = 1;
593 break;
594 case KVM_CAP_S390_MEM_OP:
595 r = MEM_OP_MAX_SIZE;
596 break;
597 case KVM_CAP_S390_MEM_OP_EXTENSION:
598 /*
599 * Flag bits indicating which extensions are supported.
600 * If r > 0, the base extension must also be supported/indicated,
601 * in order to maintain backwards compatibility.
602 */
603 r = KVM_S390_MEMOP_EXTENSION_CAP_BASE |
604 KVM_S390_MEMOP_EXTENSION_CAP_CMPXCHG;
605 break;
606 case KVM_CAP_NR_VCPUS:
607 case KVM_CAP_MAX_VCPUS:
608 case KVM_CAP_MAX_VCPU_ID:
609 r = KVM_S390_BSCA_CPU_SLOTS;
610 if (!kvm_s390_use_sca_entries())
611 r = KVM_MAX_VCPUS;
612 else if (sclp.has_esca && sclp.has_64bscao)
613 r = KVM_S390_ESCA_CPU_SLOTS;
614 if (ext == KVM_CAP_NR_VCPUS)
615 r = min_t(unsigned int, num_online_cpus(), r);
616 break;
617 case KVM_CAP_S390_COW:
618 r = MACHINE_HAS_ESOP;
619 break;
620 case KVM_CAP_S390_VECTOR_REGISTERS:
621 r = test_facility(129);
622 break;
623 case KVM_CAP_S390_RI:
624 r = test_facility(64);
625 break;
626 case KVM_CAP_S390_GS:
627 r = test_facility(133);
628 break;
629 case KVM_CAP_S390_BPB:
630 r = test_facility(82);
631 break;
632 case KVM_CAP_S390_PROTECTED_ASYNC_DISABLE:
633 r = async_destroy && is_prot_virt_host();
634 break;
635 case KVM_CAP_S390_PROTECTED:
636 r = is_prot_virt_host();
637 break;
638 case KVM_CAP_S390_PROTECTED_DUMP: {
639 u64 pv_cmds_dump[] = {
640 BIT_UVC_CMD_DUMP_INIT,
641 BIT_UVC_CMD_DUMP_CONFIG_STOR_STATE,
642 BIT_UVC_CMD_DUMP_CPU,
643 BIT_UVC_CMD_DUMP_COMPLETE,
644 };
645 int i;
646
647 r = is_prot_virt_host();
648
649 for (i = 0; i < ARRAY_SIZE(pv_cmds_dump); i++) {
650 if (!test_bit_inv(pv_cmds_dump[i],
651 (unsigned long *)&uv_info.inst_calls_list)) {
652 r = 0;
653 break;
654 }
655 }
656 break;
657 }
658 case KVM_CAP_S390_ZPCI_OP:
659 r = kvm_s390_pci_interp_allowed();
660 break;
661 case KVM_CAP_S390_CPU_TOPOLOGY:
662 r = test_facility(11);
663 break;
664 default:
665 r = 0;
666 }
667 return r;
668 }
669
670 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
671 {
672 int i;
673 gfn_t cur_gfn, last_gfn;
674 unsigned long gaddr, vmaddr;
675 struct gmap *gmap = kvm->arch.gmap;
676 DECLARE_BITMAP(bitmap, _PAGE_ENTRIES);
677
678 /* Loop over all guest segments */
679 cur_gfn = memslot->base_gfn;
680 last_gfn = memslot->base_gfn + memslot->npages;
681 for (; cur_gfn <= last_gfn; cur_gfn += _PAGE_ENTRIES) {
682 gaddr = gfn_to_gpa(cur_gfn);
683 vmaddr = gfn_to_hva_memslot(memslot, cur_gfn);
684 if (kvm_is_error_hva(vmaddr))
685 continue;
686
687 bitmap_zero(bitmap, _PAGE_ENTRIES);
688 gmap_sync_dirty_log_pmd(gmap, bitmap, gaddr, vmaddr);
689 for (i = 0; i < _PAGE_ENTRIES; i++) {
690 if (test_bit(i, bitmap))
691 mark_page_dirty(kvm, cur_gfn + i);
692 }
693
694 if (fatal_signal_pending(current))
695 return;
696 cond_resched();
697 }
698 }
699
700 /* Section: vm related */
701 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
702
703 /*
704 * Get (and clear) the dirty memory log for a memory slot.
705 */
706 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
707 struct kvm_dirty_log *log)
708 {
709 int r;
710 unsigned long n;
711 struct kvm_memory_slot *memslot;
712 int is_dirty;
713
714 if (kvm_is_ucontrol(kvm))
715 return -EINVAL;
716
717 mutex_lock(&kvm->slots_lock);
718
719 r = -EINVAL;
720 if (log->slot >= KVM_USER_MEM_SLOTS)
721 goto out;
722
723 r = kvm_get_dirty_log(kvm, log, &is_dirty, &memslot);
724 if (r)
725 goto out;
726
727 /* Clear the dirty log */
728 if (is_dirty) {
729 n = kvm_dirty_bitmap_bytes(memslot);
730 memset(memslot->dirty_bitmap, 0, n);
731 }
732 r = 0;
733 out:
734 mutex_unlock(&kvm->slots_lock);
735 return r;
736 }
737
738 static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
739 {
740 unsigned long i;
741 struct kvm_vcpu *vcpu;
742
743 kvm_for_each_vcpu(i, vcpu, kvm) {
744 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
745 }
746 }
747
748 int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
749 {
750 int r;
751
752 if (cap->flags)
753 return -EINVAL;
754
755 switch (cap->cap) {
756 case KVM_CAP_S390_IRQCHIP:
757 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
758 kvm->arch.use_irqchip = 1;
759 r = 0;
760 break;
761 case KVM_CAP_S390_USER_SIGP:
762 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
763 kvm->arch.user_sigp = 1;
764 r = 0;
765 break;
766 case KVM_CAP_S390_VECTOR_REGISTERS:
767 mutex_lock(&kvm->lock);
768 if (kvm->created_vcpus) {
769 r = -EBUSY;
770 } else if (cpu_has_vx()) {
771 set_kvm_facility(kvm->arch.model.fac_mask, 129);
772 set_kvm_facility(kvm->arch.model.fac_list, 129);
773 if (test_facility(134)) {
774 set_kvm_facility(kvm->arch.model.fac_mask, 134);
775 set_kvm_facility(kvm->arch.model.fac_list, 134);
776 }
777 if (test_facility(135)) {
778 set_kvm_facility(kvm->arch.model.fac_mask, 135);
779 set_kvm_facility(kvm->arch.model.fac_list, 135);
780 }
781 if (test_facility(148)) {
782 set_kvm_facility(kvm->arch.model.fac_mask, 148);
783 set_kvm_facility(kvm->arch.model.fac_list, 148);
784 }
785 if (test_facility(152)) {
786 set_kvm_facility(kvm->arch.model.fac_mask, 152);
787 set_kvm_facility(kvm->arch.model.fac_list, 152);
788 }
789 if (test_facility(192)) {
790 set_kvm_facility(kvm->arch.model.fac_mask, 192);
791 set_kvm_facility(kvm->arch.model.fac_list, 192);
792 }
793 r = 0;
794 } else
795 r = -EINVAL;
796 mutex_unlock(&kvm->lock);
797 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
798 r ? "(not available)" : "(success)");
799 break;
800 case KVM_CAP_S390_RI:
801 r = -EINVAL;
802 mutex_lock(&kvm->lock);
803 if (kvm->created_vcpus) {
804 r = -EBUSY;
805 } else if (test_facility(64)) {
806 set_kvm_facility(kvm->arch.model.fac_mask, 64);
807 set_kvm_facility(kvm->arch.model.fac_list, 64);
808 r = 0;
809 }
810 mutex_unlock(&kvm->lock);
811 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
812 r ? "(not available)" : "(success)");
813 break;
814 case KVM_CAP_S390_AIS:
815 mutex_lock(&kvm->lock);
816 if (kvm->created_vcpus) {
817 r = -EBUSY;
818 } else {
819 set_kvm_facility(kvm->arch.model.fac_mask, 72);
820 set_kvm_facility(kvm->arch.model.fac_list, 72);
821 r = 0;
822 }
823 mutex_unlock(&kvm->lock);
824 VM_EVENT(kvm, 3, "ENABLE: AIS %s",
825 r ? "(not available)" : "(success)");
826 break;
827 case KVM_CAP_S390_GS:
828 r = -EINVAL;
829 mutex_lock(&kvm->lock);
830 if (kvm->created_vcpus) {
831 r = -EBUSY;
832 } else if (test_facility(133)) {
833 set_kvm_facility(kvm->arch.model.fac_mask, 133);
834 set_kvm_facility(kvm->arch.model.fac_list, 133);
835 r = 0;
836 }
837 mutex_unlock(&kvm->lock);
838 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_GS %s",
839 r ? "(not available)" : "(success)");
840 break;
841 case KVM_CAP_S390_HPAGE_1M:
842 mutex_lock(&kvm->lock);
843 if (kvm->created_vcpus)
844 r = -EBUSY;
845 else if (!hpage || kvm->arch.use_cmma || kvm_is_ucontrol(kvm))
846 r = -EINVAL;
847 else {
848 r = 0;
849 mmap_write_lock(kvm->mm);
850 kvm->mm->context.allow_gmap_hpage_1m = 1;
851 mmap_write_unlock(kvm->mm);
852 /*
853 * We might have to create fake 4k page
854 * tables. To avoid that the hardware works on
855 * stale PGSTEs, we emulate these instructions.
856 */
857 kvm->arch.use_skf = 0;
858 kvm->arch.use_pfmfi = 0;
859 }
860 mutex_unlock(&kvm->lock);
861 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_HPAGE %s",
862 r ? "(not available)" : "(success)");
863 break;
864 case KVM_CAP_S390_USER_STSI:
865 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
866 kvm->arch.user_stsi = 1;
867 r = 0;
868 break;
869 case KVM_CAP_S390_USER_INSTR0:
870 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
871 kvm->arch.user_instr0 = 1;
872 icpt_operexc_on_all_vcpus(kvm);
873 r = 0;
874 break;
875 case KVM_CAP_S390_CPU_TOPOLOGY:
876 r = -EINVAL;
877 mutex_lock(&kvm->lock);
878 if (kvm->created_vcpus) {
879 r = -EBUSY;
880 } else if (test_facility(11)) {
881 set_kvm_facility(kvm->arch.model.fac_mask, 11);
882 set_kvm_facility(kvm->arch.model.fac_list, 11);
883 r = 0;
884 }
885 mutex_unlock(&kvm->lock);
886 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_CPU_TOPOLOGY %s",
887 r ? "(not available)" : "(success)");
888 break;
889 default:
890 r = -EINVAL;
891 break;
892 }
893 return r;
894 }
895
896 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
897 {
898 int ret;
899
900 switch (attr->attr) {
901 case KVM_S390_VM_MEM_LIMIT_SIZE:
902 ret = 0;
903 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
904 kvm->arch.mem_limit);
905 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
906 ret = -EFAULT;
907 break;
908 default:
909 ret = -ENXIO;
910 break;
911 }
912 return ret;
913 }
914
915 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
916 {
917 int ret;
918 unsigned int idx;
919 switch (attr->attr) {
920 case KVM_S390_VM_MEM_ENABLE_CMMA:
921 ret = -ENXIO;
922 if (!sclp.has_cmma)
923 break;
924
925 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
926 mutex_lock(&kvm->lock);
927 if (kvm->created_vcpus)
928 ret = -EBUSY;
929 else if (kvm->mm->context.allow_gmap_hpage_1m)
930 ret = -EINVAL;
931 else {
932 kvm->arch.use_cmma = 1;
933 /* Not compatible with cmma. */
934 kvm->arch.use_pfmfi = 0;
935 ret = 0;
936 }
937 mutex_unlock(&kvm->lock);
938 break;
939 case KVM_S390_VM_MEM_CLR_CMMA:
940 ret = -ENXIO;
941 if (!sclp.has_cmma)
942 break;
943 ret = -EINVAL;
944 if (!kvm->arch.use_cmma)
945 break;
946
947 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
948 mutex_lock(&kvm->lock);
949 idx = srcu_read_lock(&kvm->srcu);
950 s390_reset_cmma(kvm->arch.gmap->mm);
951 srcu_read_unlock(&kvm->srcu, idx);
952 mutex_unlock(&kvm->lock);
953 ret = 0;
954 break;
955 case KVM_S390_VM_MEM_LIMIT_SIZE: {
956 unsigned long new_limit;
957
958 if (kvm_is_ucontrol(kvm))
959 return -EINVAL;
960
961 if (get_user(new_limit, (u64 __user *)attr->addr))
962 return -EFAULT;
963
964 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
965 new_limit > kvm->arch.mem_limit)
966 return -E2BIG;
967
968 if (!new_limit)
969 return -EINVAL;
970
971 /* gmap_create takes last usable address */
972 if (new_limit != KVM_S390_NO_MEM_LIMIT)
973 new_limit -= 1;
974
975 ret = -EBUSY;
976 mutex_lock(&kvm->lock);
977 if (!kvm->created_vcpus) {
978 /* gmap_create will round the limit up */
979 struct gmap *new = gmap_create(current->mm, new_limit);
980
981 if (!new) {
982 ret = -ENOMEM;
983 } else {
984 gmap_remove(kvm->arch.gmap);
985 new->private = kvm;
986 kvm->arch.gmap = new;
987 ret = 0;
988 }
989 }
990 mutex_unlock(&kvm->lock);
991 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
992 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
993 (void *) kvm->arch.gmap->asce);
994 break;
995 }
996 default:
997 ret = -ENXIO;
998 break;
999 }
1000 return ret;
1001 }
1002
1003 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
1004
1005 void kvm_s390_vcpu_crypto_reset_all(struct kvm *kvm)
1006 {
1007 struct kvm_vcpu *vcpu;
1008 unsigned long i;
1009
1010 kvm_s390_vcpu_block_all(kvm);
1011
1012 kvm_for_each_vcpu(i, vcpu, kvm) {
1013 kvm_s390_vcpu_crypto_setup(vcpu);
1014 /* recreate the shadow crycb by leaving the VSIE handler */
1015 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu);
1016 }
1017
1018 kvm_s390_vcpu_unblock_all(kvm);
1019 }
1020
1021 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
1022 {
1023 mutex_lock(&kvm->lock);
1024 switch (attr->attr) {
1025 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1026 if (!test_kvm_facility(kvm, 76)) {
1027 mutex_unlock(&kvm->lock);
1028 return -EINVAL;
1029 }
1030 get_random_bytes(
1031 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1032 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1033 kvm->arch.crypto.aes_kw = 1;
1034 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
1035 break;
1036 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1037 if (!test_kvm_facility(kvm, 76)) {
1038 mutex_unlock(&kvm->lock);
1039 return -EINVAL;
1040 }
1041 get_random_bytes(
1042 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1043 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1044 kvm->arch.crypto.dea_kw = 1;
1045 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
1046 break;
1047 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1048 if (!test_kvm_facility(kvm, 76)) {
1049 mutex_unlock(&kvm->lock);
1050 return -EINVAL;
1051 }
1052 kvm->arch.crypto.aes_kw = 0;
1053 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
1054 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1055 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
1056 break;
1057 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1058 if (!test_kvm_facility(kvm, 76)) {
1059 mutex_unlock(&kvm->lock);
1060 return -EINVAL;
1061 }
1062 kvm->arch.crypto.dea_kw = 0;
1063 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
1064 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1065 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
1066 break;
1067 case KVM_S390_VM_CRYPTO_ENABLE_APIE:
1068 if (!ap_instructions_available()) {
1069 mutex_unlock(&kvm->lock);
1070 return -EOPNOTSUPP;
1071 }
1072 kvm->arch.crypto.apie = 1;
1073 break;
1074 case KVM_S390_VM_CRYPTO_DISABLE_APIE:
1075 if (!ap_instructions_available()) {
1076 mutex_unlock(&kvm->lock);
1077 return -EOPNOTSUPP;
1078 }
1079 kvm->arch.crypto.apie = 0;
1080 break;
1081 default:
1082 mutex_unlock(&kvm->lock);
1083 return -ENXIO;
1084 }
1085
1086 kvm_s390_vcpu_crypto_reset_all(kvm);
1087 mutex_unlock(&kvm->lock);
1088 return 0;
1089 }
1090
1091 static void kvm_s390_vcpu_pci_setup(struct kvm_vcpu *vcpu)
1092 {
1093 /* Only set the ECB bits after guest requests zPCI interpretation */
1094 if (!vcpu->kvm->arch.use_zpci_interp)
1095 return;
1096
1097 vcpu->arch.sie_block->ecb2 |= ECB2_ZPCI_LSI;
1098 vcpu->arch.sie_block->ecb3 |= ECB3_AISII + ECB3_AISI;
1099 }
1100
1101 void kvm_s390_vcpu_pci_enable_interp(struct kvm *kvm)
1102 {
1103 struct kvm_vcpu *vcpu;
1104 unsigned long i;
1105
1106 lockdep_assert_held(&kvm->lock);
1107
1108 if (!kvm_s390_pci_interp_allowed())
1109 return;
1110
1111 /*
1112 * If host is configured for PCI and the necessary facilities are
1113 * available, turn on interpretation for the life of this guest
1114 */
1115 kvm->arch.use_zpci_interp = 1;
1116
1117 kvm_s390_vcpu_block_all(kvm);
1118
1119 kvm_for_each_vcpu(i, vcpu, kvm) {
1120 kvm_s390_vcpu_pci_setup(vcpu);
1121 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu);
1122 }
1123
1124 kvm_s390_vcpu_unblock_all(kvm);
1125 }
1126
1127 static void kvm_s390_sync_request_broadcast(struct kvm *kvm, int req)
1128 {
1129 unsigned long cx;
1130 struct kvm_vcpu *vcpu;
1131
1132 kvm_for_each_vcpu(cx, vcpu, kvm)
1133 kvm_s390_sync_request(req, vcpu);
1134 }
1135
1136 /*
1137 * Must be called with kvm->srcu held to avoid races on memslots, and with
1138 * kvm->slots_lock to avoid races with ourselves and kvm_s390_vm_stop_migration.
1139 */
1140 static int kvm_s390_vm_start_migration(struct kvm *kvm)
1141 {
1142 struct kvm_memory_slot *ms;
1143 struct kvm_memslots *slots;
1144 unsigned long ram_pages = 0;
1145 int bkt;
1146
1147 /* migration mode already enabled */
1148 if (kvm->arch.migration_mode)
1149 return 0;
1150 slots = kvm_memslots(kvm);
1151 if (!slots || kvm_memslots_empty(slots))
1152 return -EINVAL;
1153
1154 if (!kvm->arch.use_cmma) {
1155 kvm->arch.migration_mode = 1;
1156 return 0;
1157 }
1158 /* mark all the pages in active slots as dirty */
1159 kvm_for_each_memslot(ms, bkt, slots) {
1160 if (!ms->dirty_bitmap)
1161 return -EINVAL;
1162 /*
1163 * The second half of the bitmap is only used on x86,
1164 * and would be wasted otherwise, so we put it to good
1165 * use here to keep track of the state of the storage
1166 * attributes.
1167 */
1168 memset(kvm_second_dirty_bitmap(ms), 0xff, kvm_dirty_bitmap_bytes(ms));
1169 ram_pages += ms->npages;
1170 }
1171 atomic64_set(&kvm->arch.cmma_dirty_pages, ram_pages);
1172 kvm->arch.migration_mode = 1;
1173 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_START_MIGRATION);
1174 return 0;
1175 }
1176
1177 /*
1178 * Must be called with kvm->slots_lock to avoid races with ourselves and
1179 * kvm_s390_vm_start_migration.
1180 */
1181 static int kvm_s390_vm_stop_migration(struct kvm *kvm)
1182 {
1183 /* migration mode already disabled */
1184 if (!kvm->arch.migration_mode)
1185 return 0;
1186 kvm->arch.migration_mode = 0;
1187 if (kvm->arch.use_cmma)
1188 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_STOP_MIGRATION);
1189 return 0;
1190 }
1191
1192 static int kvm_s390_vm_set_migration(struct kvm *kvm,
1193 struct kvm_device_attr *attr)
1194 {
1195 int res = -ENXIO;
1196
1197 mutex_lock(&kvm->slots_lock);
1198 switch (attr->attr) {
1199 case KVM_S390_VM_MIGRATION_START:
1200 res = kvm_s390_vm_start_migration(kvm);
1201 break;
1202 case KVM_S390_VM_MIGRATION_STOP:
1203 res = kvm_s390_vm_stop_migration(kvm);
1204 break;
1205 default:
1206 break;
1207 }
1208 mutex_unlock(&kvm->slots_lock);
1209
1210 return res;
1211 }
1212
1213 static int kvm_s390_vm_get_migration(struct kvm *kvm,
1214 struct kvm_device_attr *attr)
1215 {
1216 u64 mig = kvm->arch.migration_mode;
1217
1218 if (attr->attr != KVM_S390_VM_MIGRATION_STATUS)
1219 return -ENXIO;
1220
1221 if (copy_to_user((void __user *)attr->addr, &mig, sizeof(mig)))
1222 return -EFAULT;
1223 return 0;
1224 }
1225
1226 static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod);
1227
1228 static int kvm_s390_set_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1229 {
1230 struct kvm_s390_vm_tod_clock gtod;
1231
1232 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod)))
1233 return -EFAULT;
1234
1235 if (!test_kvm_facility(kvm, 139) && gtod.epoch_idx)
1236 return -EINVAL;
1237 __kvm_s390_set_tod_clock(kvm, >od);
1238
1239 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x, TOD base: 0x%llx",
1240 gtod.epoch_idx, gtod.tod);
1241
1242 return 0;
1243 }
1244
1245 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1246 {
1247 u8 gtod_high;
1248
1249 if (copy_from_user(>od_high, (void __user *)attr->addr,
1250 sizeof(gtod_high)))
1251 return -EFAULT;
1252
1253 if (gtod_high != 0)
1254 return -EINVAL;
1255 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
1256
1257 return 0;
1258 }
1259
1260 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1261 {
1262 struct kvm_s390_vm_tod_clock gtod = { 0 };
1263
1264 if (copy_from_user(>od.tod, (void __user *)attr->addr,
1265 sizeof(gtod.tod)))
1266 return -EFAULT;
1267
1268 __kvm_s390_set_tod_clock(kvm, >od);
1269 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod.tod);
1270 return 0;
1271 }
1272
1273 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1274 {
1275 int ret;
1276
1277 if (attr->flags)
1278 return -EINVAL;
1279
1280 mutex_lock(&kvm->lock);
1281 /*
1282 * For protected guests, the TOD is managed by the ultravisor, so trying
1283 * to change it will never bring the expected results.
1284 */
1285 if (kvm_s390_pv_is_protected(kvm)) {
1286 ret = -EOPNOTSUPP;
1287 goto out_unlock;
1288 }
1289
1290 switch (attr->attr) {
1291 case KVM_S390_VM_TOD_EXT:
1292 ret = kvm_s390_set_tod_ext(kvm, attr);
1293 break;
1294 case KVM_S390_VM_TOD_HIGH:
1295 ret = kvm_s390_set_tod_high(kvm, attr);
1296 break;
1297 case KVM_S390_VM_TOD_LOW:
1298 ret = kvm_s390_set_tod_low(kvm, attr);
1299 break;
1300 default:
1301 ret = -ENXIO;
1302 break;
1303 }
1304
1305 out_unlock:
1306 mutex_unlock(&kvm->lock);
1307 return ret;
1308 }
1309
1310 static void kvm_s390_get_tod_clock(struct kvm *kvm,
1311 struct kvm_s390_vm_tod_clock *gtod)
1312 {
1313 union tod_clock clk;
1314
1315 preempt_disable();
1316
1317 store_tod_clock_ext(&clk);
1318
1319 gtod->tod = clk.tod + kvm->arch.epoch;
1320 gtod->epoch_idx = 0;
1321 if (test_kvm_facility(kvm, 139)) {
1322 gtod->epoch_idx = clk.ei + kvm->arch.epdx;
1323 if (gtod->tod < clk.tod)
1324 gtod->epoch_idx += 1;
1325 }
1326
1327 preempt_enable();
1328 }
1329
1330 static int kvm_s390_get_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1331 {
1332 struct kvm_s390_vm_tod_clock gtod;
1333
1334 memset(>od, 0, sizeof(gtod));
1335 kvm_s390_get_tod_clock(kvm, >od);
1336 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
1337 return -EFAULT;
1338
1339 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x, TOD base: 0x%llx",
1340 gtod.epoch_idx, gtod.tod);
1341 return 0;
1342 }
1343
1344 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1345 {
1346 u8 gtod_high = 0;
1347
1348 if (copy_to_user((void __user *)attr->addr, >od_high,
1349 sizeof(gtod_high)))
1350 return -EFAULT;
1351 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
1352
1353 return 0;
1354 }
1355
1356 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1357 {
1358 u64 gtod;
1359
1360 gtod = kvm_s390_get_tod_clock_fast(kvm);
1361 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
1362 return -EFAULT;
1363 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
1364
1365 return 0;
1366 }
1367
1368 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1369 {
1370 int ret;
1371
1372 if (attr->flags)
1373 return -EINVAL;
1374
1375 switch (attr->attr) {
1376 case KVM_S390_VM_TOD_EXT:
1377 ret = kvm_s390_get_tod_ext(kvm, attr);
1378 break;
1379 case KVM_S390_VM_TOD_HIGH:
1380 ret = kvm_s390_get_tod_high(kvm, attr);
1381 break;
1382 case KVM_S390_VM_TOD_LOW:
1383 ret = kvm_s390_get_tod_low(kvm, attr);
1384 break;
1385 default:
1386 ret = -ENXIO;
1387 break;
1388 }
1389 return ret;
1390 }
1391
1392 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1393 {
1394 struct kvm_s390_vm_cpu_processor *proc;
1395 u16 lowest_ibc, unblocked_ibc;
1396 int ret = 0;
1397
1398 mutex_lock(&kvm->lock);
1399 if (kvm->created_vcpus) {
1400 ret = -EBUSY;
1401 goto out;
1402 }
1403 proc = kzalloc(sizeof(*proc), GFP_KERNEL_ACCOUNT);
1404 if (!proc) {
1405 ret = -ENOMEM;
1406 goto out;
1407 }
1408 if (!copy_from_user(proc, (void __user *)attr->addr,
1409 sizeof(*proc))) {
1410 kvm->arch.model.cpuid = proc->cpuid;
1411 lowest_ibc = sclp.ibc >> 16 & 0xfff;
1412 unblocked_ibc = sclp.ibc & 0xfff;
1413 if (lowest_ibc && proc->ibc) {
1414 if (proc->ibc > unblocked_ibc)
1415 kvm->arch.model.ibc = unblocked_ibc;
1416 else if (proc->ibc < lowest_ibc)
1417 kvm->arch.model.ibc = lowest_ibc;
1418 else
1419 kvm->arch.model.ibc = proc->ibc;
1420 }
1421 memcpy(kvm->arch.model.fac_list, proc->fac_list,
1422 S390_ARCH_FAC_LIST_SIZE_BYTE);
1423 VM_EVENT(kvm, 3, "SET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1424 kvm->arch.model.ibc,
1425 kvm->arch.model.cpuid);
1426 VM_EVENT(kvm, 3, "SET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1427 kvm->arch.model.fac_list[0],
1428 kvm->arch.model.fac_list[1],
1429 kvm->arch.model.fac_list[2]);
1430 } else
1431 ret = -EFAULT;
1432 kfree(proc);
1433 out:
1434 mutex_unlock(&kvm->lock);
1435 return ret;
1436 }
1437
1438 static int kvm_s390_set_processor_feat(struct kvm *kvm,
1439 struct kvm_device_attr *attr)
1440 {
1441 struct kvm_s390_vm_cpu_feat data;
1442
1443 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
1444 return -EFAULT;
1445 if (!bitmap_subset((unsigned long *) data.feat,
1446 kvm_s390_available_cpu_feat,
1447 KVM_S390_VM_CPU_FEAT_NR_BITS))
1448 return -EINVAL;
1449
1450 mutex_lock(&kvm->lock);
1451 if (kvm->created_vcpus) {
1452 mutex_unlock(&kvm->lock);
1453 return -EBUSY;
1454 }
1455 bitmap_from_arr64(kvm->arch.cpu_feat, data.feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1456 mutex_unlock(&kvm->lock);
1457 VM_EVENT(kvm, 3, "SET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1458 data.feat[0],
1459 data.feat[1],
1460 data.feat[2]);
1461 return 0;
1462 }
1463
1464 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
1465 struct kvm_device_attr *attr)
1466 {
1467 mutex_lock(&kvm->lock);
1468 if (kvm->created_vcpus) {
1469 mutex_unlock(&kvm->lock);
1470 return -EBUSY;
1471 }
1472
1473 if (copy_from_user(&kvm->arch.model.subfuncs, (void __user *)attr->addr,
1474 sizeof(struct kvm_s390_vm_cpu_subfunc))) {
1475 mutex_unlock(&kvm->lock);
1476 return -EFAULT;
1477 }
1478 mutex_unlock(&kvm->lock);
1479
1480 VM_EVENT(kvm, 3, "SET: guest PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1481 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0],
1482 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1],
1483 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2],
1484 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]);
1485 VM_EVENT(kvm, 3, "SET: guest PTFF subfunc 0x%16.16lx.%16.16lx",
1486 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0],
1487 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]);
1488 VM_EVENT(kvm, 3, "SET: guest KMAC subfunc 0x%16.16lx.%16.16lx",
1489 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0],
1490 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]);
1491 VM_EVENT(kvm, 3, "SET: guest KMC subfunc 0x%16.16lx.%16.16lx",
1492 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0],
1493 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]);
1494 VM_EVENT(kvm, 3, "SET: guest KM subfunc 0x%16.16lx.%16.16lx",
1495 ((unsigned long *) &kvm->arch.model.subfuncs.km)[0],
1496 ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]);
1497 VM_EVENT(kvm, 3, "SET: guest KIMD subfunc 0x%16.16lx.%16.16lx",
1498 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0],
1499 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]);
1500 VM_EVENT(kvm, 3, "SET: guest KLMD subfunc 0x%16.16lx.%16.16lx",
1501 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0],
1502 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]);
1503 VM_EVENT(kvm, 3, "SET: guest PCKMO subfunc 0x%16.16lx.%16.16lx",
1504 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0],
1505 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]);
1506 VM_EVENT(kvm, 3, "SET: guest KMCTR subfunc 0x%16.16lx.%16.16lx",
1507 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0],
1508 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]);
1509 VM_EVENT(kvm, 3, "SET: guest KMF subfunc 0x%16.16lx.%16.16lx",
1510 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0],
1511 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]);
1512 VM_EVENT(kvm, 3, "SET: guest KMO subfunc 0x%16.16lx.%16.16lx",
1513 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0],
1514 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]);
1515 VM_EVENT(kvm, 3, "SET: guest PCC subfunc 0x%16.16lx.%16.16lx",
1516 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0],
1517 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]);
1518 VM_EVENT(kvm, 3, "SET: guest PPNO subfunc 0x%16.16lx.%16.16lx",
1519 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0],
1520 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]);
1521 VM_EVENT(kvm, 3, "SET: guest KMA subfunc 0x%16.16lx.%16.16lx",
1522 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0],
1523 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]);
1524 VM_EVENT(kvm, 3, "SET: guest KDSA subfunc 0x%16.16lx.%16.16lx",
1525 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0],
1526 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]);
1527 VM_EVENT(kvm, 3, "SET: guest SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1528 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0],
1529 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1],
1530 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2],
1531 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]);
1532 VM_EVENT(kvm, 3, "SET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1533 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0],
1534 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1],
1535 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2],
1536 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]);
1537
1538 return 0;
1539 }
1540
1541 #define KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK \
1542 ( \
1543 ((struct kvm_s390_vm_cpu_uv_feat){ \
1544 .ap = 1, \
1545 .ap_intr = 1, \
1546 }) \
1547 .feat \
1548 )
1549
1550 static int kvm_s390_set_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1551 {
1552 struct kvm_s390_vm_cpu_uv_feat __user *ptr = (void __user *)attr->addr;
1553 unsigned long data, filter;
1554
1555 filter = uv_info.uv_feature_indications & KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK;
1556 if (get_user(data, &ptr->feat))
1557 return -EFAULT;
1558 if (!bitmap_subset(&data, &filter, KVM_S390_VM_CPU_UV_FEAT_NR_BITS))
1559 return -EINVAL;
1560
1561 mutex_lock(&kvm->lock);
1562 if (kvm->created_vcpus) {
1563 mutex_unlock(&kvm->lock);
1564 return -EBUSY;
1565 }
1566 kvm->arch.model.uv_feat_guest.feat = data;
1567 mutex_unlock(&kvm->lock);
1568
1569 VM_EVENT(kvm, 3, "SET: guest UV-feat: 0x%16.16lx", data);
1570
1571 return 0;
1572 }
1573
1574 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1575 {
1576 int ret = -ENXIO;
1577
1578 switch (attr->attr) {
1579 case KVM_S390_VM_CPU_PROCESSOR:
1580 ret = kvm_s390_set_processor(kvm, attr);
1581 break;
1582 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1583 ret = kvm_s390_set_processor_feat(kvm, attr);
1584 break;
1585 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1586 ret = kvm_s390_set_processor_subfunc(kvm, attr);
1587 break;
1588 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
1589 ret = kvm_s390_set_uv_feat(kvm, attr);
1590 break;
1591 }
1592 return ret;
1593 }
1594
1595 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1596 {
1597 struct kvm_s390_vm_cpu_processor *proc;
1598 int ret = 0;
1599
1600 proc = kzalloc(sizeof(*proc), GFP_KERNEL_ACCOUNT);
1601 if (!proc) {
1602 ret = -ENOMEM;
1603 goto out;
1604 }
1605 proc->cpuid = kvm->arch.model.cpuid;
1606 proc->ibc = kvm->arch.model.ibc;
1607 memcpy(&proc->fac_list, kvm->arch.model.fac_list,
1608 S390_ARCH_FAC_LIST_SIZE_BYTE);
1609 VM_EVENT(kvm, 3, "GET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1610 kvm->arch.model.ibc,
1611 kvm->arch.model.cpuid);
1612 VM_EVENT(kvm, 3, "GET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1613 kvm->arch.model.fac_list[0],
1614 kvm->arch.model.fac_list[1],
1615 kvm->arch.model.fac_list[2]);
1616 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
1617 ret = -EFAULT;
1618 kfree(proc);
1619 out:
1620 return ret;
1621 }
1622
1623 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
1624 {
1625 struct kvm_s390_vm_cpu_machine *mach;
1626 int ret = 0;
1627
1628 mach = kzalloc(sizeof(*mach), GFP_KERNEL_ACCOUNT);
1629 if (!mach) {
1630 ret = -ENOMEM;
1631 goto out;
1632 }
1633 get_cpu_id((struct cpuid *) &mach->cpuid);
1634 mach->ibc = sclp.ibc;
1635 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
1636 S390_ARCH_FAC_LIST_SIZE_BYTE);
1637 memcpy((unsigned long *)&mach->fac_list, stfle_fac_list,
1638 sizeof(stfle_fac_list));
1639 VM_EVENT(kvm, 3, "GET: host ibc: 0x%4.4x, host cpuid: 0x%16.16llx",
1640 kvm->arch.model.ibc,
1641 kvm->arch.model.cpuid);
1642 VM_EVENT(kvm, 3, "GET: host facmask: 0x%16.16llx.%16.16llx.%16.16llx",
1643 mach->fac_mask[0],
1644 mach->fac_mask[1],
1645 mach->fac_mask[2]);
1646 VM_EVENT(kvm, 3, "GET: host faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1647 mach->fac_list[0],
1648 mach->fac_list[1],
1649 mach->fac_list[2]);
1650 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
1651 ret = -EFAULT;
1652 kfree(mach);
1653 out:
1654 return ret;
1655 }
1656
1657 static int kvm_s390_get_processor_feat(struct kvm *kvm,
1658 struct kvm_device_attr *attr)
1659 {
1660 struct kvm_s390_vm_cpu_feat data;
1661
1662 bitmap_to_arr64(data.feat, kvm->arch.cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1663 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1664 return -EFAULT;
1665 VM_EVENT(kvm, 3, "GET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1666 data.feat[0],
1667 data.feat[1],
1668 data.feat[2]);
1669 return 0;
1670 }
1671
1672 static int kvm_s390_get_machine_feat(struct kvm *kvm,
1673 struct kvm_device_attr *attr)
1674 {
1675 struct kvm_s390_vm_cpu_feat data;
1676
1677 bitmap_to_arr64(data.feat, kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1678 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1679 return -EFAULT;
1680 VM_EVENT(kvm, 3, "GET: host feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1681 data.feat[0],
1682 data.feat[1],
1683 data.feat[2]);
1684 return 0;
1685 }
1686
1687 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
1688 struct kvm_device_attr *attr)
1689 {
1690 if (copy_to_user((void __user *)attr->addr, &kvm->arch.model.subfuncs,
1691 sizeof(struct kvm_s390_vm_cpu_subfunc)))
1692 return -EFAULT;
1693
1694 VM_EVENT(kvm, 3, "GET: guest PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1695 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0],
1696 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1],
1697 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2],
1698 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]);
1699 VM_EVENT(kvm, 3, "GET: guest PTFF subfunc 0x%16.16lx.%16.16lx",
1700 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0],
1701 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]);
1702 VM_EVENT(kvm, 3, "GET: guest KMAC subfunc 0x%16.16lx.%16.16lx",
1703 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0],
1704 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]);
1705 VM_EVENT(kvm, 3, "GET: guest KMC subfunc 0x%16.16lx.%16.16lx",
1706 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0],
1707 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]);
1708 VM_EVENT(kvm, 3, "GET: guest KM subfunc 0x%16.16lx.%16.16lx",
1709 ((unsigned long *) &kvm->arch.model.subfuncs.km)[0],
1710 ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]);
1711 VM_EVENT(kvm, 3, "GET: guest KIMD subfunc 0x%16.16lx.%16.16lx",
1712 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0],
1713 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]);
1714 VM_EVENT(kvm, 3, "GET: guest KLMD subfunc 0x%16.16lx.%16.16lx",
1715 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0],
1716 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]);
1717 VM_EVENT(kvm, 3, "GET: guest PCKMO subfunc 0x%16.16lx.%16.16lx",
1718 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0],
1719 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]);
1720 VM_EVENT(kvm, 3, "GET: guest KMCTR subfunc 0x%16.16lx.%16.16lx",
1721 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0],
1722 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]);
1723 VM_EVENT(kvm, 3, "GET: guest KMF subfunc 0x%16.16lx.%16.16lx",
1724 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0],
1725 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]);
1726 VM_EVENT(kvm, 3, "GET: guest KMO subfunc 0x%16.16lx.%16.16lx",
1727 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0],
1728 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]);
1729 VM_EVENT(kvm, 3, "GET: guest PCC subfunc 0x%16.16lx.%16.16lx",
1730 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0],
1731 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]);
1732 VM_EVENT(kvm, 3, "GET: guest PPNO subfunc 0x%16.16lx.%16.16lx",
1733 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0],
1734 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]);
1735 VM_EVENT(kvm, 3, "GET: guest KMA subfunc 0x%16.16lx.%16.16lx",
1736 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0],
1737 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]);
1738 VM_EVENT(kvm, 3, "GET: guest KDSA subfunc 0x%16.16lx.%16.16lx",
1739 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0],
1740 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]);
1741 VM_EVENT(kvm, 3, "GET: guest SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1742 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0],
1743 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1],
1744 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2],
1745 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]);
1746 VM_EVENT(kvm, 3, "GET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1747 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0],
1748 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1],
1749 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2],
1750 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]);
1751
1752 return 0;
1753 }
1754
1755 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
1756 struct kvm_device_attr *attr)
1757 {
1758 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
1759 sizeof(struct kvm_s390_vm_cpu_subfunc)))
1760 return -EFAULT;
1761
1762 VM_EVENT(kvm, 3, "GET: host PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1763 ((unsigned long *) &kvm_s390_available_subfunc.plo)[0],
1764 ((unsigned long *) &kvm_s390_available_subfunc.plo)[1],
1765 ((unsigned long *) &kvm_s390_available_subfunc.plo)[2],
1766 ((unsigned long *) &kvm_s390_available_subfunc.plo)[3]);
1767 VM_EVENT(kvm, 3, "GET: host PTFF subfunc 0x%16.16lx.%16.16lx",
1768 ((unsigned long *) &kvm_s390_available_subfunc.ptff)[0],
1769 ((unsigned long *) &kvm_s390_available_subfunc.ptff)[1]);
1770 VM_EVENT(kvm, 3, "GET: host KMAC subfunc 0x%16.16lx.%16.16lx",
1771 ((unsigned long *) &kvm_s390_available_subfunc.kmac)[0],
1772 ((unsigned long *) &kvm_s390_available_subfunc.kmac)[1]);
1773 VM_EVENT(kvm, 3, "GET: host KMC subfunc 0x%16.16lx.%16.16lx",
1774 ((unsigned long *) &kvm_s390_available_subfunc.kmc)[0],
1775 ((unsigned long *) &kvm_s390_available_subfunc.kmc)[1]);
1776 VM_EVENT(kvm, 3, "GET: host KM subfunc 0x%16.16lx.%16.16lx",
1777 ((unsigned long *) &kvm_s390_available_subfunc.km)[0],
1778 ((unsigned long *) &kvm_s390_available_subfunc.km)[1]);
1779 VM_EVENT(kvm, 3, "GET: host KIMD subfunc 0x%16.16lx.%16.16lx",
1780 ((unsigned long *) &kvm_s390_available_subfunc.kimd)[0],
1781 ((unsigned long *) &kvm_s390_available_subfunc.kimd)[1]);
1782 VM_EVENT(kvm, 3, "GET: host KLMD subfunc 0x%16.16lx.%16.16lx",
1783 ((unsigned long *) &kvm_s390_available_subfunc.klmd)[0],
1784 ((unsigned long *) &kvm_s390_available_subfunc.klmd)[1]);
1785 VM_EVENT(kvm, 3, "GET: host PCKMO subfunc 0x%16.16lx.%16.16lx",
1786 ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[0],
1787 ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[1]);
1788 VM_EVENT(kvm, 3, "GET: host KMCTR subfunc 0x%16.16lx.%16.16lx",
1789 ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[0],
1790 ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[1]);
1791 VM_EVENT(kvm, 3, "GET: host KMF subfunc 0x%16.16lx.%16.16lx",
1792 ((unsigned long *) &kvm_s390_available_subfunc.kmf)[0],
1793 ((unsigned long *) &kvm_s390_available_subfunc.kmf)[1]);
1794 VM_EVENT(kvm, 3, "GET: host KMO subfunc 0x%16.16lx.%16.16lx",
1795 ((unsigned long *) &kvm_s390_available_subfunc.kmo)[0],
1796 ((unsigned long *) &kvm_s390_available_subfunc.kmo)[1]);
1797 VM_EVENT(kvm, 3, "GET: host PCC subfunc 0x%16.16lx.%16.16lx",
1798 ((unsigned long *) &kvm_s390_available_subfunc.pcc)[0],
1799 ((unsigned long *) &kvm_s390_available_subfunc.pcc)[1]);
1800 VM_EVENT(kvm, 3, "GET: host PPNO subfunc 0x%16.16lx.%16.16lx",
1801 ((unsigned long *) &kvm_s390_available_subfunc.ppno)[0],
1802 ((unsigned long *) &kvm_s390_available_subfunc.ppno)[1]);
1803 VM_EVENT(kvm, 3, "GET: host KMA subfunc 0x%16.16lx.%16.16lx",
1804 ((unsigned long *) &kvm_s390_available_subfunc.kma)[0],
1805 ((unsigned long *) &kvm_s390_available_subfunc.kma)[1]);
1806 VM_EVENT(kvm, 3, "GET: host KDSA subfunc 0x%16.16lx.%16.16lx",
1807 ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[0],
1808 ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[1]);
1809 VM_EVENT(kvm, 3, "GET: host SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1810 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[0],
1811 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[1],
1812 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[2],
1813 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[3]);
1814 VM_EVENT(kvm, 3, "GET: host DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1815 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[0],
1816 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[1],
1817 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[2],
1818 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[3]);
1819
1820 return 0;
1821 }
1822
1823 static int kvm_s390_get_processor_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1824 {
1825 struct kvm_s390_vm_cpu_uv_feat __user *dst = (void __user *)attr->addr;
1826 unsigned long feat = kvm->arch.model.uv_feat_guest.feat;
1827
1828 if (put_user(feat, &dst->feat))
1829 return -EFAULT;
1830 VM_EVENT(kvm, 3, "GET: guest UV-feat: 0x%16.16lx", feat);
1831
1832 return 0;
1833 }
1834
1835 static int kvm_s390_get_machine_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1836 {
1837 struct kvm_s390_vm_cpu_uv_feat __user *dst = (void __user *)attr->addr;
1838 unsigned long feat;
1839
1840 BUILD_BUG_ON(sizeof(*dst) != sizeof(uv_info.uv_feature_indications));
1841
1842 feat = uv_info.uv_feature_indications & KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK;
1843 if (put_user(feat, &dst->feat))
1844 return -EFAULT;
1845 VM_EVENT(kvm, 3, "GET: guest UV-feat: 0x%16.16lx", feat);
1846
1847 return 0;
1848 }
1849
1850 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1851 {
1852 int ret = -ENXIO;
1853
1854 switch (attr->attr) {
1855 case KVM_S390_VM_CPU_PROCESSOR:
1856 ret = kvm_s390_get_processor(kvm, attr);
1857 break;
1858 case KVM_S390_VM_CPU_MACHINE:
1859 ret = kvm_s390_get_machine(kvm, attr);
1860 break;
1861 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1862 ret = kvm_s390_get_processor_feat(kvm, attr);
1863 break;
1864 case KVM_S390_VM_CPU_MACHINE_FEAT:
1865 ret = kvm_s390_get_machine_feat(kvm, attr);
1866 break;
1867 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1868 ret = kvm_s390_get_processor_subfunc(kvm, attr);
1869 break;
1870 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1871 ret = kvm_s390_get_machine_subfunc(kvm, attr);
1872 break;
1873 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
1874 ret = kvm_s390_get_processor_uv_feat(kvm, attr);
1875 break;
1876 case KVM_S390_VM_CPU_MACHINE_UV_FEAT_GUEST:
1877 ret = kvm_s390_get_machine_uv_feat(kvm, attr);
1878 break;
1879 }
1880 return ret;
1881 }
1882
1883 /**
1884 * kvm_s390_update_topology_change_report - update CPU topology change report
1885 * @kvm: guest KVM description
1886 * @val: set or clear the MTCR bit
1887 *
1888 * Updates the Multiprocessor Topology-Change-Report bit to signal
1889 * the guest with a topology change.
1890 * This is only relevant if the topology facility is present.
1891 *
1892 * The SCA version, bsca or esca, doesn't matter as offset is the same.
1893 */
1894 static void kvm_s390_update_topology_change_report(struct kvm *kvm, bool val)
1895 {
1896 union sca_utility new, old;
1897 struct bsca_block *sca;
1898
1899 read_lock(&kvm->arch.sca_lock);
1900 sca = kvm->arch.sca;
1901 do {
1902 old = READ_ONCE(sca->utility);
1903 new = old;
1904 new.mtcr = val;
1905 } while (cmpxchg(&sca->utility.val, old.val, new.val) != old.val);
1906 read_unlock(&kvm->arch.sca_lock);
1907 }
1908
1909 static int kvm_s390_set_topo_change_indication(struct kvm *kvm,
1910 struct kvm_device_attr *attr)
1911 {
1912 if (!test_kvm_facility(kvm, 11))
1913 return -ENXIO;
1914
1915 kvm_s390_update_topology_change_report(kvm, !!attr->attr);
1916 return 0;
1917 }
1918
1919 static int kvm_s390_get_topo_change_indication(struct kvm *kvm,
1920 struct kvm_device_attr *attr)
1921 {
1922 u8 topo;
1923
1924 if (!test_kvm_facility(kvm, 11))
1925 return -ENXIO;
1926
1927 read_lock(&kvm->arch.sca_lock);
1928 topo = ((struct bsca_block *)kvm->arch.sca)->utility.mtcr;
1929 read_unlock(&kvm->arch.sca_lock);
1930
1931 return put_user(topo, (u8 __user *)attr->addr);
1932 }
1933
1934 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1935 {
1936 int ret;
1937
1938 switch (attr->group) {
1939 case KVM_S390_VM_MEM_CTRL:
1940 ret = kvm_s390_set_mem_control(kvm, attr);
1941 break;
1942 case KVM_S390_VM_TOD:
1943 ret = kvm_s390_set_tod(kvm, attr);
1944 break;
1945 case KVM_S390_VM_CPU_MODEL:
1946 ret = kvm_s390_set_cpu_model(kvm, attr);
1947 break;
1948 case KVM_S390_VM_CRYPTO:
1949 ret = kvm_s390_vm_set_crypto(kvm, attr);
1950 break;
1951 case KVM_S390_VM_MIGRATION:
1952 ret = kvm_s390_vm_set_migration(kvm, attr);
1953 break;
1954 case KVM_S390_VM_CPU_TOPOLOGY:
1955 ret = kvm_s390_set_topo_change_indication(kvm, attr);
1956 break;
1957 default:
1958 ret = -ENXIO;
1959 break;
1960 }
1961
1962 return ret;
1963 }
1964
1965 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1966 {
1967 int ret;
1968
1969 switch (attr->group) {
1970 case KVM_S390_VM_MEM_CTRL:
1971 ret = kvm_s390_get_mem_control(kvm, attr);
1972 break;
1973 case KVM_S390_VM_TOD:
1974 ret = kvm_s390_get_tod(kvm, attr);
1975 break;
1976 case KVM_S390_VM_CPU_MODEL:
1977 ret = kvm_s390_get_cpu_model(kvm, attr);
1978 break;
1979 case KVM_S390_VM_MIGRATION:
1980 ret = kvm_s390_vm_get_migration(kvm, attr);
1981 break;
1982 case KVM_S390_VM_CPU_TOPOLOGY:
1983 ret = kvm_s390_get_topo_change_indication(kvm, attr);
1984 break;
1985 default:
1986 ret = -ENXIO;
1987 break;
1988 }
1989
1990 return ret;
1991 }
1992
1993 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1994 {
1995 int ret;
1996
1997 switch (attr->group) {
1998 case KVM_S390_VM_MEM_CTRL:
1999 switch (attr->attr) {
2000 case KVM_S390_VM_MEM_ENABLE_CMMA:
2001 case KVM_S390_VM_MEM_CLR_CMMA:
2002 ret = sclp.has_cmma ? 0 : -ENXIO;
2003 break;
2004 case KVM_S390_VM_MEM_LIMIT_SIZE:
2005 ret = 0;
2006 break;
2007 default:
2008 ret = -ENXIO;
2009 break;
2010 }
2011 break;
2012 case KVM_S390_VM_TOD:
2013 switch (attr->attr) {
2014 case KVM_S390_VM_TOD_LOW:
2015 case KVM_S390_VM_TOD_HIGH:
2016 ret = 0;
2017 break;
2018 default:
2019 ret = -ENXIO;
2020 break;
2021 }
2022 break;
2023 case KVM_S390_VM_CPU_MODEL:
2024 switch (attr->attr) {
2025 case KVM_S390_VM_CPU_PROCESSOR:
2026 case KVM_S390_VM_CPU_MACHINE:
2027 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
2028 case KVM_S390_VM_CPU_MACHINE_FEAT:
2029 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
2030 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
2031 case KVM_S390_VM_CPU_MACHINE_UV_FEAT_GUEST:
2032 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
2033 ret = 0;
2034 break;
2035 default:
2036 ret = -ENXIO;
2037 break;
2038 }
2039 break;
2040 case KVM_S390_VM_CRYPTO:
2041 switch (attr->attr) {
2042 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
2043 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
2044 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
2045 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
2046 ret = 0;
2047 break;
2048 case KVM_S390_VM_CRYPTO_ENABLE_APIE:
2049 case KVM_S390_VM_CRYPTO_DISABLE_APIE:
2050 ret = ap_instructions_available() ? 0 : -ENXIO;
2051 break;
2052 default:
2053 ret = -ENXIO;
2054 break;
2055 }
2056 break;
2057 case KVM_S390_VM_MIGRATION:
2058 ret = 0;
2059 break;
2060 case KVM_S390_VM_CPU_TOPOLOGY:
2061 ret = test_kvm_facility(kvm, 11) ? 0 : -ENXIO;
2062 break;
2063 default:
2064 ret = -ENXIO;
2065 break;
2066 }
2067
2068 return ret;
2069 }
2070
2071 static int kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
2072 {
2073 uint8_t *keys;
2074 uint64_t hva;
2075 int srcu_idx, i, r = 0;
2076
2077 if (args->flags != 0)
2078 return -EINVAL;
2079
2080 /* Is this guest using storage keys? */
2081 if (!mm_uses_skeys(current->mm))
2082 return KVM_S390_GET_SKEYS_NONE;
2083
2084 /* Enforce sane limit on memory allocation */
2085 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
2086 return -EINVAL;
2087
2088 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL_ACCOUNT);
2089 if (!keys)
2090 return -ENOMEM;
2091
2092 mmap_read_lock(current->mm);
2093 srcu_idx = srcu_read_lock(&kvm->srcu);
2094 for (i = 0; i < args->count; i++) {
2095 hva = gfn_to_hva(kvm, args->start_gfn + i);
2096 if (kvm_is_error_hva(hva)) {
2097 r = -EFAULT;
2098 break;
2099 }
2100
2101 r = get_guest_storage_key(current->mm, hva, &keys[i]);
2102 if (r)
2103 break;
2104 }
2105 srcu_read_unlock(&kvm->srcu, srcu_idx);
2106 mmap_read_unlock(current->mm);
2107
2108 if (!r) {
2109 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
2110 sizeof(uint8_t) * args->count);
2111 if (r)
2112 r = -EFAULT;
2113 }
2114
2115 kvfree(keys);
2116 return r;
2117 }
2118
2119 static int kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
2120 {
2121 uint8_t *keys;
2122 uint64_t hva;
2123 int srcu_idx, i, r = 0;
2124 bool unlocked;
2125
2126 if (args->flags != 0)
2127 return -EINVAL;
2128
2129 /* Enforce sane limit on memory allocation */
2130 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
2131 return -EINVAL;
2132
2133 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL_ACCOUNT);
2134 if (!keys)
2135 return -ENOMEM;
2136
2137 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
2138 sizeof(uint8_t) * args->count);
2139 if (r) {
2140 r = -EFAULT;
2141 goto out;
2142 }
2143
2144 /* Enable storage key handling for the guest */
2145 r = s390_enable_skey();
2146 if (r)
2147 goto out;
2148
2149 i = 0;
2150 mmap_read_lock(current->mm);
2151 srcu_idx = srcu_read_lock(&kvm->srcu);
2152 while (i < args->count) {
2153 unlocked = false;
2154 hva = gfn_to_hva(kvm, args->start_gfn + i);
2155 if (kvm_is_error_hva(hva)) {
2156 r = -EFAULT;
2157 break;
2158 }
2159
2160 /* Lowest order bit is reserved */
2161 if (keys[i] & 0x01) {
2162 r = -EINVAL;
2163 break;
2164 }
2165
2166 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
2167 if (r) {
2168 r = fixup_user_fault(current->mm, hva,
2169 FAULT_FLAG_WRITE, &unlocked);
2170 if (r)
2171 break;
2172 }
2173 if (!r)
2174 i++;
2175 }
2176 srcu_read_unlock(&kvm->srcu, srcu_idx);
2177 mmap_read_unlock(current->mm);
2178 out:
2179 kvfree(keys);
2180 return r;
2181 }
2182
2183 /*
2184 * Base address and length must be sent at the start of each block, therefore
2185 * it's cheaper to send some clean data, as long as it's less than the size of
2186 * two longs.
2187 */
2188 #define KVM_S390_MAX_BIT_DISTANCE (2 * sizeof(void *))
2189 /* for consistency */
2190 #define KVM_S390_CMMA_SIZE_MAX ((u32)KVM_S390_SKEYS_MAX)
2191
2192 static int kvm_s390_peek_cmma(struct kvm *kvm, struct kvm_s390_cmma_log *args,
2193 u8 *res, unsigned long bufsize)
2194 {
2195 unsigned long pgstev, hva, cur_gfn = args->start_gfn;
2196
2197 args->count = 0;
2198 while (args->count < bufsize) {
2199 hva = gfn_to_hva(kvm, cur_gfn);
2200 /*
2201 * We return an error if the first value was invalid, but we
2202 * return successfully if at least one value was copied.
2203 */
2204 if (kvm_is_error_hva(hva))
2205 return args->count ? 0 : -EFAULT;
2206 if (get_pgste(kvm->mm, hva, &pgstev) < 0)
2207 pgstev = 0;
2208 res[args->count++] = (pgstev >> 24) & 0x43;
2209 cur_gfn++;
2210 }
2211
2212 return 0;
2213 }
2214
2215 static struct kvm_memory_slot *gfn_to_memslot_approx(struct kvm_memslots *slots,
2216 gfn_t gfn)
2217 {
2218 return ____gfn_to_memslot(slots, gfn, true);
2219 }
2220
2221 static unsigned long kvm_s390_next_dirty_cmma(struct kvm_memslots *slots,
2222 unsigned long cur_gfn)
2223 {
2224 struct kvm_memory_slot *ms = gfn_to_memslot_approx(slots, cur_gfn);
2225 unsigned long ofs = cur_gfn - ms->base_gfn;
2226 struct rb_node *mnode = &ms->gfn_node[slots->node_idx];
2227
2228 if (ms->base_gfn + ms->npages <= cur_gfn) {
2229 mnode = rb_next(mnode);
2230 /* If we are above the highest slot, wrap around */
2231 if (!mnode)
2232 mnode = rb_first(&slots->gfn_tree);
2233
2234 ms = container_of(mnode, struct kvm_memory_slot, gfn_node[slots->node_idx]);
2235 ofs = 0;
2236 }
2237
2238 if (cur_gfn < ms->base_gfn)
2239 ofs = 0;
2240
2241 ofs = find_next_bit(kvm_second_dirty_bitmap(ms), ms->npages, ofs);
2242 while (ofs >= ms->npages && (mnode = rb_next(mnode))) {
2243 ms = container_of(mnode, struct kvm_memory_slot, gfn_node[slots->node_idx]);
2244 ofs = find_first_bit(kvm_second_dirty_bitmap(ms), ms->npages);
2245 }
2246 return ms->base_gfn + ofs;
2247 }
2248
2249 static int kvm_s390_get_cmma(struct kvm *kvm, struct kvm_s390_cmma_log *args,
2250 u8 *res, unsigned long bufsize)
2251 {
2252 unsigned long mem_end, cur_gfn, next_gfn, hva, pgstev;
2253 struct kvm_memslots *slots = kvm_memslots(kvm);
2254 struct kvm_memory_slot *ms;
2255
2256 if (unlikely(kvm_memslots_empty(slots)))
2257 return 0;
2258
2259 cur_gfn = kvm_s390_next_dirty_cmma(slots, args->start_gfn);
2260 ms = gfn_to_memslot(kvm, cur_gfn);
2261 args->count = 0;
2262 args->start_gfn = cur_gfn;
2263 if (!ms)
2264 return 0;
2265 next_gfn = kvm_s390_next_dirty_cmma(slots, cur_gfn + 1);
2266 mem_end = kvm_s390_get_gfn_end(slots);
2267
2268 while (args->count < bufsize) {
2269 hva = gfn_to_hva(kvm, cur_gfn);
2270 if (kvm_is_error_hva(hva))
2271 return 0;
2272 /* Decrement only if we actually flipped the bit to 0 */
2273 if (test_and_clear_bit(cur_gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
2274 atomic64_dec(&kvm->arch.cmma_dirty_pages);
2275 if (get_pgste(kvm->mm, hva, &pgstev) < 0)
2276 pgstev = 0;
2277 /* Save the value */
2278 res[args->count++] = (pgstev >> 24) & 0x43;
2279 /* If the next bit is too far away, stop. */
2280 if (next_gfn > cur_gfn + KVM_S390_MAX_BIT_DISTANCE)
2281 return 0;
2282 /* If we reached the previous "next", find the next one */
2283 if (cur_gfn == next_gfn)
2284 next_gfn = kvm_s390_next_dirty_cmma(slots, cur_gfn + 1);
2285 /* Reached the end of memory or of the buffer, stop */
2286 if ((next_gfn >= mem_end) ||
2287 (next_gfn - args->start_gfn >= bufsize))
2288 return 0;
2289 cur_gfn++;
2290 /* Reached the end of the current memslot, take the next one. */
2291 if (cur_gfn - ms->base_gfn >= ms->npages) {
2292 ms = gfn_to_memslot(kvm, cur_gfn);
2293 if (!ms)
2294 return 0;
2295 }
2296 }
2297 return 0;
2298 }
2299
2300 /*
2301 * This function searches for the next page with dirty CMMA attributes, and
2302 * saves the attributes in the buffer up to either the end of the buffer or
2303 * until a block of at least KVM_S390_MAX_BIT_DISTANCE clean bits is found;
2304 * no trailing clean bytes are saved.
2305 * In case no dirty bits were found, or if CMMA was not enabled or used, the
2306 * output buffer will indicate 0 as length.
2307 */
2308 static int kvm_s390_get_cmma_bits(struct kvm *kvm,
2309 struct kvm_s390_cmma_log *args)
2310 {
2311 unsigned long bufsize;
2312 int srcu_idx, peek, ret;
2313 u8 *values;
2314
2315 if (!kvm->arch.use_cmma)
2316 return -ENXIO;
2317 /* Invalid/unsupported flags were specified */
2318 if (args->flags & ~KVM_S390_CMMA_PEEK)
2319 return -EINVAL;
2320 /* Migration mode query, and we are not doing a migration */
2321 peek = !!(args->flags & KVM_S390_CMMA_PEEK);
2322 if (!peek && !kvm->arch.migration_mode)
2323 return -EINVAL;
2324 /* CMMA is disabled or was not used, or the buffer has length zero */
2325 bufsize = min(args->count, KVM_S390_CMMA_SIZE_MAX);
2326 if (!bufsize || !kvm->mm->context.uses_cmm) {
2327 memset(args, 0, sizeof(*args));
2328 return 0;
2329 }
2330 /* We are not peeking, and there are no dirty pages */
2331 if (!peek && !atomic64_read(&kvm->arch.cmma_dirty_pages)) {
2332 memset(args, 0, sizeof(*args));
2333 return 0;
2334 }
2335
2336 values = vmalloc(bufsize);
2337 if (!values)
2338 return -ENOMEM;
2339
2340 mmap_read_lock(kvm->mm);
2341 srcu_idx = srcu_read_lock(&kvm->srcu);
2342 if (peek)
2343 ret = kvm_s390_peek_cmma(kvm, args, values, bufsize);
2344 else
2345 ret = kvm_s390_get_cmma(kvm, args, values, bufsize);
2346 srcu_read_unlock(&kvm->srcu, srcu_idx);
2347 mmap_read_unlock(kvm->mm);
2348
2349 if (kvm->arch.migration_mode)
2350 args->remaining = atomic64_read(&kvm->arch.cmma_dirty_pages);
2351 else
2352 args->remaining = 0;
2353
2354 if (copy_to_user((void __user *)args->values, values, args->count))
2355 ret = -EFAULT;
2356
2357 vfree(values);
2358 return ret;
2359 }
2360
2361 /*
2362 * This function sets the CMMA attributes for the given pages. If the input
2363 * buffer has zero length, no action is taken, otherwise the attributes are
2364 * set and the mm->context.uses_cmm flag is set.
2365 */
2366 static int kvm_s390_set_cmma_bits(struct kvm *kvm,
2367 const struct kvm_s390_cmma_log *args)
2368 {
2369 unsigned long hva, mask, pgstev, i;
2370 uint8_t *bits;
2371 int srcu_idx, r = 0;
2372
2373 mask = args->mask;
2374
2375 if (!kvm->arch.use_cmma)
2376 return -ENXIO;
2377 /* invalid/unsupported flags */
2378 if (args->flags != 0)
2379 return -EINVAL;
2380 /* Enforce sane limit on memory allocation */
2381 if (args->count > KVM_S390_CMMA_SIZE_MAX)
2382 return -EINVAL;
2383 /* Nothing to do */
2384 if (args->count == 0)
2385 return 0;
2386
2387 bits = vmalloc(array_size(sizeof(*bits), args->count));
2388 if (!bits)
2389 return -ENOMEM;
2390
2391 r = copy_from_user(bits, (void __user *)args->values, args->count);
2392 if (r) {
2393 r = -EFAULT;
2394 goto out;
2395 }
2396
2397 mmap_read_lock(kvm->mm);
2398 srcu_idx = srcu_read_lock(&kvm->srcu);
2399 for (i = 0; i < args->count; i++) {
2400 hva = gfn_to_hva(kvm, args->start_gfn + i);
2401 if (kvm_is_error_hva(hva)) {
2402 r = -EFAULT;
2403 break;
2404 }
2405
2406 pgstev = bits[i];
2407 pgstev = pgstev << 24;
2408 mask &= _PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT;
2409 set_pgste_bits(kvm->mm, hva, mask, pgstev);
2410 }
2411 srcu_read_unlock(&kvm->srcu, srcu_idx);
2412 mmap_read_unlock(kvm->mm);
2413
2414 if (!kvm->mm->context.uses_cmm) {
2415 mmap_write_lock(kvm->mm);
2416 kvm->mm->context.uses_cmm = 1;
2417 mmap_write_unlock(kvm->mm);
2418 }
2419 out:
2420 vfree(bits);
2421 return r;
2422 }
2423
2424 /**
2425 * kvm_s390_cpus_from_pv - Convert all protected vCPUs in a protected VM to
2426 * non protected.
2427 * @kvm: the VM whose protected vCPUs are to be converted
2428 * @rc: return value for the RC field of the UVC (in case of error)
2429 * @rrc: return value for the RRC field of the UVC (in case of error)
2430 *
2431 * Does not stop in case of error, tries to convert as many
2432 * CPUs as possible. In case of error, the RC and RRC of the last error are
2433 * returned.
2434 *
2435 * Return: 0 in case of success, otherwise -EIO
2436 */
2437 int kvm_s390_cpus_from_pv(struct kvm *kvm, u16 *rc, u16 *rrc)
2438 {
2439 struct kvm_vcpu *vcpu;
2440 unsigned long i;
2441 u16 _rc, _rrc;
2442 int ret = 0;
2443
2444 /*
2445 * We ignore failures and try to destroy as many CPUs as possible.
2446 * At the same time we must not free the assigned resources when
2447 * this fails, as the ultravisor has still access to that memory.
2448 * So kvm_s390_pv_destroy_cpu can leave a "wanted" memory leak
2449 * behind.
2450 * We want to return the first failure rc and rrc, though.
2451 */
2452 kvm_for_each_vcpu(i, vcpu, kvm) {
2453 mutex_lock(&vcpu->mutex);
2454 if (kvm_s390_pv_destroy_cpu(vcpu, &_rc, &_rrc) && !ret) {
2455 *rc = _rc;
2456 *rrc = _rrc;
2457 ret = -EIO;
2458 }
2459 mutex_unlock(&vcpu->mutex);
2460 }
2461 /* Ensure that we re-enable gisa if the non-PV guest used it but the PV guest did not. */
2462 if (use_gisa)
2463 kvm_s390_gisa_enable(kvm);
2464 return ret;
2465 }
2466
2467 /**
2468 * kvm_s390_cpus_to_pv - Convert all non-protected vCPUs in a protected VM
2469 * to protected.
2470 * @kvm: the VM whose protected vCPUs are to be converted
2471 * @rc: return value for the RC field of the UVC (in case of error)
2472 * @rrc: return value for the RRC field of the UVC (in case of error)
2473 *
2474 * Tries to undo the conversion in case of error.
2475 *
2476 * Return: 0 in case of success, otherwise -EIO
2477 */
2478 static int kvm_s390_cpus_to_pv(struct kvm *kvm, u16 *rc, u16 *rrc)
2479 {
2480 unsigned long i;
2481 int r = 0;
2482 u16 dummy;
2483
2484 struct kvm_vcpu *vcpu;
2485
2486 /* Disable the GISA if the ultravisor does not support AIV. */
2487 if (!uv_has_feature(BIT_UV_FEAT_AIV))
2488 kvm_s390_gisa_disable(kvm);
2489
2490 kvm_for_each_vcpu(i, vcpu, kvm) {
2491 mutex_lock(&vcpu->mutex);
2492 r = kvm_s390_pv_create_cpu(vcpu, rc, rrc);
2493 mutex_unlock(&vcpu->mutex);
2494 if (r)
2495 break;
2496 }
2497 if (r)
2498 kvm_s390_cpus_from_pv(kvm, &dummy, &dummy);
2499 return r;
2500 }
2501
2502 /*
2503 * Here we provide user space with a direct interface to query UV
2504 * related data like UV maxima and available features as well as
2505 * feature specific data.
2506 *
2507 * To facilitate future extension of the data structures we'll try to
2508 * write data up to the maximum requested length.
2509 */
2510 static ssize_t kvm_s390_handle_pv_info(struct kvm_s390_pv_info *info)
2511 {
2512 ssize_t len_min;
2513
2514 switch (info->header.id) {
2515 case KVM_PV_INFO_VM: {
2516 len_min = sizeof(info->header) + sizeof(info->vm);
2517
2518 if (info->header.len_max < len_min)
2519 return -EINVAL;
2520
2521 memcpy(info->vm.inst_calls_list,
2522 uv_info.inst_calls_list,
2523 sizeof(uv_info.inst_calls_list));
2524
2525 /* It's max cpuid not max cpus, so it's off by one */
2526 info->vm.max_cpus = uv_info.max_guest_cpu_id + 1;
2527 info->vm.max_guests = uv_info.max_num_sec_conf;
2528 info->vm.max_guest_addr = uv_info.max_sec_stor_addr;
2529 info->vm.feature_indication = uv_info.uv_feature_indications;
2530
2531 return len_min;
2532 }
2533 case KVM_PV_INFO_DUMP: {
2534 len_min = sizeof(info->header) + sizeof(info->dump);
2535
2536 if (info->header.len_max < len_min)
2537 return -EINVAL;
2538
2539 info->dump.dump_cpu_buffer_len = uv_info.guest_cpu_stor_len;
2540 info->dump.dump_config_mem_buffer_per_1m = uv_info.conf_dump_storage_state_len;
2541 info->dump.dump_config_finalize_len = uv_info.conf_dump_finalize_len;
2542 return len_min;
2543 }
2544 default:
2545 return -EINVAL;
2546 }
2547 }
2548
2549 static int kvm_s390_pv_dmp(struct kvm *kvm, struct kvm_pv_cmd *cmd,
2550 struct kvm_s390_pv_dmp dmp)
2551 {
2552 int r = -EINVAL;
2553 void __user *result_buff = (void __user *)dmp.buff_addr;
2554
2555 switch (dmp.subcmd) {
2556 case KVM_PV_DUMP_INIT: {
2557 if (kvm->arch.pv.dumping)
2558 break;
2559
2560 /*
2561 * Block SIE entry as concurrent dump UVCs could lead
2562 * to validities.
2563 */
2564 kvm_s390_vcpu_block_all(kvm);
2565
2566 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2567 UVC_CMD_DUMP_INIT, &cmd->rc, &cmd->rrc);
2568 KVM_UV_EVENT(kvm, 3, "PROTVIRT DUMP INIT: rc %x rrc %x",
2569 cmd->rc, cmd->rrc);
2570 if (!r) {
2571 kvm->arch.pv.dumping = true;
2572 } else {
2573 kvm_s390_vcpu_unblock_all(kvm);
2574 r = -EINVAL;
2575 }
2576 break;
2577 }
2578 case KVM_PV_DUMP_CONFIG_STOR_STATE: {
2579 if (!kvm->arch.pv.dumping)
2580 break;
2581
2582 /*
2583 * gaddr is an output parameter since we might stop
2584 * early. As dmp will be copied back in our caller, we
2585 * don't need to do it ourselves.
2586 */
2587 r = kvm_s390_pv_dump_stor_state(kvm, result_buff, &dmp.gaddr, dmp.buff_len,
2588 &cmd->rc, &cmd->rrc);
2589 break;
2590 }
2591 case KVM_PV_DUMP_COMPLETE: {
2592 if (!kvm->arch.pv.dumping)
2593 break;
2594
2595 r = -EINVAL;
2596 if (dmp.buff_len < uv_info.conf_dump_finalize_len)
2597 break;
2598
2599 r = kvm_s390_pv_dump_complete(kvm, result_buff,
2600 &cmd->rc, &cmd->rrc);
2601 break;
2602 }
2603 default:
2604 r = -ENOTTY;
2605 break;
2606 }
2607
2608 return r;
2609 }
2610
2611 static int kvm_s390_handle_pv(struct kvm *kvm, struct kvm_pv_cmd *cmd)
2612 {
2613 const bool need_lock = (cmd->cmd != KVM_PV_ASYNC_CLEANUP_PERFORM);
2614 void __user *argp = (void __user *)cmd->data;
2615 int r = 0;
2616 u16 dummy;
2617
2618 if (need_lock)
2619 mutex_lock(&kvm->lock);
2620
2621 switch (cmd->cmd) {
2622 case KVM_PV_ENABLE: {
2623 r = -EINVAL;
2624 if (kvm_s390_pv_is_protected(kvm))
2625 break;
2626
2627 /*
2628 * FMT 4 SIE needs esca. As we never switch back to bsca from
2629 * esca, we need no cleanup in the error cases below
2630 */
2631 r = sca_switch_to_extended(kvm);
2632 if (r)
2633 break;
2634
2635 r = s390_disable_cow_sharing();
2636 if (r)
2637 break;
2638
2639 r = kvm_s390_pv_init_vm(kvm, &cmd->rc, &cmd->rrc);
2640 if (r)
2641 break;
2642
2643 r = kvm_s390_cpus_to_pv(kvm, &cmd->rc, &cmd->rrc);
2644 if (r)
2645 kvm_s390_pv_deinit_vm(kvm, &dummy, &dummy);
2646
2647 /* we need to block service interrupts from now on */
2648 set_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2649 break;
2650 }
2651 case KVM_PV_ASYNC_CLEANUP_PREPARE:
2652 r = -EINVAL;
2653 if (!kvm_s390_pv_is_protected(kvm) || !async_destroy)
2654 break;
2655
2656 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc);
2657 /*
2658 * If a CPU could not be destroyed, destroy VM will also fail.
2659 * There is no point in trying to destroy it. Instead return
2660 * the rc and rrc from the first CPU that failed destroying.
2661 */
2662 if (r)
2663 break;
2664 r = kvm_s390_pv_set_aside(kvm, &cmd->rc, &cmd->rrc);
2665
2666 /* no need to block service interrupts any more */
2667 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2668 break;
2669 case KVM_PV_ASYNC_CLEANUP_PERFORM:
2670 r = -EINVAL;
2671 if (!async_destroy)
2672 break;
2673 /* kvm->lock must not be held; this is asserted inside the function. */
2674 r = kvm_s390_pv_deinit_aside_vm(kvm, &cmd->rc, &cmd->rrc);
2675 break;
2676 case KVM_PV_DISABLE: {
2677 r = -EINVAL;
2678 if (!kvm_s390_pv_is_protected(kvm))
2679 break;
2680
2681 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc);
2682 /*
2683 * If a CPU could not be destroyed, destroy VM will also fail.
2684 * There is no point in trying to destroy it. Instead return
2685 * the rc and rrc from the first CPU that failed destroying.
2686 */
2687 if (r)
2688 break;
2689 r = kvm_s390_pv_deinit_cleanup_all(kvm, &cmd->rc, &cmd->rrc);
2690
2691 /* no need to block service interrupts any more */
2692 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2693 break;
2694 }
2695 case KVM_PV_SET_SEC_PARMS: {
2696 struct kvm_s390_pv_sec_parm parms = {};
2697 void *hdr;
2698
2699 r = -EINVAL;
2700 if (!kvm_s390_pv_is_protected(kvm))
2701 break;
2702
2703 r = -EFAULT;
2704 if (copy_from_user(&parms, argp, sizeof(parms)))
2705 break;
2706
2707 /* Currently restricted to 8KB */
2708 r = -EINVAL;
2709 if (parms.length > PAGE_SIZE * 2)
2710 break;
2711
2712 r = -ENOMEM;
2713 hdr = vmalloc(parms.length);
2714 if (!hdr)
2715 break;
2716
2717 r = -EFAULT;
2718 if (!copy_from_user(hdr, (void __user *)parms.origin,
2719 parms.length))
2720 r = kvm_s390_pv_set_sec_parms(kvm, hdr, parms.length,
2721 &cmd->rc, &cmd->rrc);
2722
2723 vfree(hdr);
2724 break;
2725 }
2726 case KVM_PV_UNPACK: {
2727 struct kvm_s390_pv_unp unp = {};
2728
2729 r = -EINVAL;
2730 if (!kvm_s390_pv_is_protected(kvm) || !mm_is_protected(kvm->mm))
2731 break;
2732
2733 r = -EFAULT;
2734 if (copy_from_user(&unp, argp, sizeof(unp)))
2735 break;
2736
2737 r = kvm_s390_pv_unpack(kvm, unp.addr, unp.size, unp.tweak,
2738 &cmd->rc, &cmd->rrc);
2739 break;
2740 }
2741 case KVM_PV_VERIFY: {
2742 r = -EINVAL;
2743 if (!kvm_s390_pv_is_protected(kvm))
2744 break;
2745
2746 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2747 UVC_CMD_VERIFY_IMG, &cmd->rc, &cmd->rrc);
2748 KVM_UV_EVENT(kvm, 3, "PROTVIRT VERIFY: rc %x rrc %x", cmd->rc,
2749 cmd->rrc);
2750 break;
2751 }
2752 case KVM_PV_PREP_RESET: {
2753 r = -EINVAL;
2754 if (!kvm_s390_pv_is_protected(kvm))
2755 break;
2756
2757 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2758 UVC_CMD_PREPARE_RESET, &cmd->rc, &cmd->rrc);
2759 KVM_UV_EVENT(kvm, 3, "PROTVIRT PREP RESET: rc %x rrc %x",
2760 cmd->rc, cmd->rrc);
2761 break;
2762 }
2763 case KVM_PV_UNSHARE_ALL: {
2764 r = -EINVAL;
2765 if (!kvm_s390_pv_is_protected(kvm))
2766 break;
2767
2768 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2769 UVC_CMD_SET_UNSHARE_ALL, &cmd->rc, &cmd->rrc);
2770 KVM_UV_EVENT(kvm, 3, "PROTVIRT UNSHARE: rc %x rrc %x",
2771 cmd->rc, cmd->rrc);
2772 break;
2773 }
2774 case KVM_PV_INFO: {
2775 struct kvm_s390_pv_info info = {};
2776 ssize_t data_len;
2777
2778 /*
2779 * No need to check the VM protection here.
2780 *
2781 * Maybe user space wants to query some of the data
2782 * when the VM is still unprotected. If we see the
2783 * need to fence a new data command we can still
2784 * return an error in the info handler.
2785 */
2786
2787 r = -EFAULT;
2788 if (copy_from_user(&info, argp, sizeof(info.header)))
2789 break;
2790
2791 r = -EINVAL;
2792 if (info.header.len_max < sizeof(info.header))
2793 break;
2794
2795 data_len = kvm_s390_handle_pv_info(&info);
2796 if (data_len < 0) {
2797 r = data_len;
2798 break;
2799 }
2800 /*
2801 * If a data command struct is extended (multiple
2802 * times) this can be used to determine how much of it
2803 * is valid.
2804 */
2805 info.header.len_written = data_len;
2806
2807 r = -EFAULT;
2808 if (copy_to_user(argp, &info, data_len))
2809 break;
2810
2811 r = 0;
2812 break;
2813 }
2814 case KVM_PV_DUMP: {
2815 struct kvm_s390_pv_dmp dmp;
2816
2817 r = -EINVAL;
2818 if (!kvm_s390_pv_is_protected(kvm))
2819 break;
2820
2821 r = -EFAULT;
2822 if (copy_from_user(&dmp, argp, sizeof(dmp)))
2823 break;
2824
2825 r = kvm_s390_pv_dmp(kvm, cmd, dmp);
2826 if (r)
2827 break;
2828
2829 if (copy_to_user(argp, &dmp, sizeof(dmp))) {
2830 r = -EFAULT;
2831 break;
2832 }
2833
2834 break;
2835 }
2836 default:
2837 r = -ENOTTY;
2838 }
2839 if (need_lock)
2840 mutex_unlock(&kvm->lock);
2841
2842 return r;
2843 }
2844
2845 static int mem_op_validate_common(struct kvm_s390_mem_op *mop, u64 supported_flags)
2846 {
2847 if (mop->flags & ~supported_flags || !mop->size)
2848 return -EINVAL;
2849 if (mop->size > MEM_OP_MAX_SIZE)
2850 return -E2BIG;
2851 if (mop->flags & KVM_S390_MEMOP_F_SKEY_PROTECTION) {
2852 if (mop->key > 0xf)
2853 return -EINVAL;
2854 } else {
2855 mop->key = 0;
2856 }
2857 return 0;
2858 }
2859
2860 static int kvm_s390_vm_mem_op_abs(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2861 {
2862 void __user *uaddr = (void __user *)mop->buf;
2863 enum gacc_mode acc_mode;
2864 void *tmpbuf = NULL;
2865 int r, srcu_idx;
2866
2867 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_SKEY_PROTECTION |
2868 KVM_S390_MEMOP_F_CHECK_ONLY);
2869 if (r)
2870 return r;
2871
2872 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
2873 tmpbuf = vmalloc(mop->size);
2874 if (!tmpbuf)
2875 return -ENOMEM;
2876 }
2877
2878 srcu_idx = srcu_read_lock(&kvm->srcu);
2879
2880 if (!kvm_is_gpa_in_memslot(kvm, mop->gaddr)) {
2881 r = PGM_ADDRESSING;
2882 goto out_unlock;
2883 }
2884
2885 acc_mode = mop->op == KVM_S390_MEMOP_ABSOLUTE_READ ? GACC_FETCH : GACC_STORE;
2886 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2887 r = check_gpa_range(kvm, mop->gaddr, mop->size, acc_mode, mop->key);
2888 goto out_unlock;
2889 }
2890 if (acc_mode == GACC_FETCH) {
2891 r = access_guest_abs_with_key(kvm, mop->gaddr, tmpbuf,
2892 mop->size, GACC_FETCH, mop->key);
2893 if (r)
2894 goto out_unlock;
2895 if (copy_to_user(uaddr, tmpbuf, mop->size))
2896 r = -EFAULT;
2897 } else {
2898 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
2899 r = -EFAULT;
2900 goto out_unlock;
2901 }
2902 r = access_guest_abs_with_key(kvm, mop->gaddr, tmpbuf,
2903 mop->size, GACC_STORE, mop->key);
2904 }
2905
2906 out_unlock:
2907 srcu_read_unlock(&kvm->srcu, srcu_idx);
2908
2909 vfree(tmpbuf);
2910 return r;
2911 }
2912
2913 static int kvm_s390_vm_mem_op_cmpxchg(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2914 {
2915 void __user *uaddr = (void __user *)mop->buf;
2916 void __user *old_addr = (void __user *)mop->old_addr;
2917 union {
2918 __uint128_t quad;
2919 char raw[sizeof(__uint128_t)];
2920 } old = { .quad = 0}, new = { .quad = 0 };
2921 unsigned int off_in_quad = sizeof(new) - mop->size;
2922 int r, srcu_idx;
2923 bool success;
2924
2925 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_SKEY_PROTECTION);
2926 if (r)
2927 return r;
2928 /*
2929 * This validates off_in_quad. Checking that size is a power
2930 * of two is not necessary, as cmpxchg_guest_abs_with_key
2931 * takes care of that
2932 */
2933 if (mop->size > sizeof(new))
2934 return -EINVAL;
2935 if (copy_from_user(&new.raw[off_in_quad], uaddr, mop->size))
2936 return -EFAULT;
2937 if (copy_from_user(&old.raw[off_in_quad], old_addr, mop->size))
2938 return -EFAULT;
2939
2940 srcu_idx = srcu_read_lock(&kvm->srcu);
2941
2942 if (!kvm_is_gpa_in_memslot(kvm, mop->gaddr)) {
2943 r = PGM_ADDRESSING;
2944 goto out_unlock;
2945 }
2946
2947 r = cmpxchg_guest_abs_with_key(kvm, mop->gaddr, mop->size, &old.quad,
2948 new.quad, mop->key, &success);
2949 if (!success && copy_to_user(old_addr, &old.raw[off_in_quad], mop->size))
2950 r = -EFAULT;
2951
2952 out_unlock:
2953 srcu_read_unlock(&kvm->srcu, srcu_idx);
2954 return r;
2955 }
2956
2957 static int kvm_s390_vm_mem_op(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2958 {
2959 /*
2960 * This is technically a heuristic only, if the kvm->lock is not
2961 * taken, it is not guaranteed that the vm is/remains non-protected.
2962 * This is ok from a kernel perspective, wrongdoing is detected
2963 * on the access, -EFAULT is returned and the vm may crash the
2964 * next time it accesses the memory in question.
2965 * There is no sane usecase to do switching and a memop on two
2966 * different CPUs at the same time.
2967 */
2968 if (kvm_s390_pv_get_handle(kvm))
2969 return -EINVAL;
2970
2971 switch (mop->op) {
2972 case KVM_S390_MEMOP_ABSOLUTE_READ:
2973 case KVM_S390_MEMOP_ABSOLUTE_WRITE:
2974 return kvm_s390_vm_mem_op_abs(kvm, mop);
2975 case KVM_S390_MEMOP_ABSOLUTE_CMPXCHG:
2976 return kvm_s390_vm_mem_op_cmpxchg(kvm, mop);
2977 default:
2978 return -EINVAL;
2979 }
2980 }
2981
2982 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
2983 {
2984 struct kvm *kvm = filp->private_data;
2985 void __user *argp = (void __user *)arg;
2986 struct kvm_device_attr attr;
2987 int r;
2988
2989 switch (ioctl) {
2990 case KVM_S390_INTERRUPT: {
2991 struct kvm_s390_interrupt s390int;
2992
2993 r = -EFAULT;
2994 if (copy_from_user(&s390int, argp, sizeof(s390int)))
2995 break;
2996 r = kvm_s390_inject_vm(kvm, &s390int);
2997 break;
2998 }
2999 case KVM_CREATE_IRQCHIP: {
3000 r = -EINVAL;
3001 if (kvm->arch.use_irqchip)
3002 r = 0;
3003 break;
3004 }
3005 case KVM_SET_DEVICE_ATTR: {
3006 r = -EFAULT;
3007 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3008 break;
3009 r = kvm_s390_vm_set_attr(kvm, &attr);
3010 break;
3011 }
3012 case KVM_GET_DEVICE_ATTR: {
3013 r = -EFAULT;
3014 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3015 break;
3016 r = kvm_s390_vm_get_attr(kvm, &attr);
3017 break;
3018 }
3019 case KVM_HAS_DEVICE_ATTR: {
3020 r = -EFAULT;
3021 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3022 break;
3023 r = kvm_s390_vm_has_attr(kvm, &attr);
3024 break;
3025 }
3026 case KVM_S390_GET_SKEYS: {
3027 struct kvm_s390_skeys args;
3028
3029 r = -EFAULT;
3030 if (copy_from_user(&args, argp,
3031 sizeof(struct kvm_s390_skeys)))
3032 break;
3033 r = kvm_s390_get_skeys(kvm, &args);
3034 break;
3035 }
3036 case KVM_S390_SET_SKEYS: {
3037 struct kvm_s390_skeys args;
3038
3039 r = -EFAULT;
3040 if (copy_from_user(&args, argp,
3041 sizeof(struct kvm_s390_skeys)))
3042 break;
3043 r = kvm_s390_set_skeys(kvm, &args);
3044 break;
3045 }
3046 case KVM_S390_GET_CMMA_BITS: {
3047 struct kvm_s390_cmma_log args;
3048
3049 r = -EFAULT;
3050 if (copy_from_user(&args, argp, sizeof(args)))
3051 break;
3052 mutex_lock(&kvm->slots_lock);
3053 r = kvm_s390_get_cmma_bits(kvm, &args);
3054 mutex_unlock(&kvm->slots_lock);
3055 if (!r) {
3056 r = copy_to_user(argp, &args, sizeof(args));
3057 if (r)
3058 r = -EFAULT;
3059 }
3060 break;
3061 }
3062 case KVM_S390_SET_CMMA_BITS: {
3063 struct kvm_s390_cmma_log args;
3064
3065 r = -EFAULT;
3066 if (copy_from_user(&args, argp, sizeof(args)))
3067 break;
3068 mutex_lock(&kvm->slots_lock);
3069 r = kvm_s390_set_cmma_bits(kvm, &args);
3070 mutex_unlock(&kvm->slots_lock);
3071 break;
3072 }
3073 case KVM_S390_PV_COMMAND: {
3074 struct kvm_pv_cmd args;
3075
3076 /* protvirt means user cpu state */
3077 kvm_s390_set_user_cpu_state_ctrl(kvm);
3078 r = 0;
3079 if (!is_prot_virt_host()) {
3080 r = -EINVAL;
3081 break;
3082 }
3083 if (copy_from_user(&args, argp, sizeof(args))) {
3084 r = -EFAULT;
3085 break;
3086 }
3087 if (args.flags) {
3088 r = -EINVAL;
3089 break;
3090 }
3091 /* must be called without kvm->lock */
3092 r = kvm_s390_handle_pv(kvm, &args);
3093 if (copy_to_user(argp, &args, sizeof(args))) {
3094 r = -EFAULT;
3095 break;
3096 }
3097 break;
3098 }
3099 case KVM_S390_MEM_OP: {
3100 struct kvm_s390_mem_op mem_op;
3101
3102 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
3103 r = kvm_s390_vm_mem_op(kvm, &mem_op);
3104 else
3105 r = -EFAULT;
3106 break;
3107 }
3108 case KVM_S390_ZPCI_OP: {
3109 struct kvm_s390_zpci_op args;
3110
3111 r = -EINVAL;
3112 if (!IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
3113 break;
3114 if (copy_from_user(&args, argp, sizeof(args))) {
3115 r = -EFAULT;
3116 break;
3117 }
3118 r = kvm_s390_pci_zpci_op(kvm, &args);
3119 break;
3120 }
3121 default:
3122 r = -ENOTTY;
3123 }
3124
3125 return r;
3126 }
3127
3128 static int kvm_s390_apxa_installed(void)
3129 {
3130 struct ap_config_info info;
3131
3132 if (ap_instructions_available()) {
3133 if (ap_qci(&info) == 0)
3134 return info.apxa;
3135 }
3136
3137 return 0;
3138 }
3139
3140 /*
3141 * The format of the crypto control block (CRYCB) is specified in the 3 low
3142 * order bits of the CRYCB designation (CRYCBD) field as follows:
3143 * Format 0: Neither the message security assist extension 3 (MSAX3) nor the
3144 * AP extended addressing (APXA) facility are installed.
3145 * Format 1: The APXA facility is not installed but the MSAX3 facility is.
3146 * Format 2: Both the APXA and MSAX3 facilities are installed
3147 */
3148 static void kvm_s390_set_crycb_format(struct kvm *kvm)
3149 {
3150 kvm->arch.crypto.crycbd = virt_to_phys(kvm->arch.crypto.crycb);
3151
3152 /* Clear the CRYCB format bits - i.e., set format 0 by default */
3153 kvm->arch.crypto.crycbd &= ~(CRYCB_FORMAT_MASK);
3154
3155 /* Check whether MSAX3 is installed */
3156 if (!test_kvm_facility(kvm, 76))
3157 return;
3158
3159 if (kvm_s390_apxa_installed())
3160 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
3161 else
3162 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
3163 }
3164
3165 /*
3166 * kvm_arch_crypto_set_masks
3167 *
3168 * @kvm: pointer to the target guest's KVM struct containing the crypto masks
3169 * to be set.
3170 * @apm: the mask identifying the accessible AP adapters
3171 * @aqm: the mask identifying the accessible AP domains
3172 * @adm: the mask identifying the accessible AP control domains
3173 *
3174 * Set the masks that identify the adapters, domains and control domains to
3175 * which the KVM guest is granted access.
3176 *
3177 * Note: The kvm->lock mutex must be locked by the caller before invoking this
3178 * function.
3179 */
3180 void kvm_arch_crypto_set_masks(struct kvm *kvm, unsigned long *apm,
3181 unsigned long *aqm, unsigned long *adm)
3182 {
3183 struct kvm_s390_crypto_cb *crycb = kvm->arch.crypto.crycb;
3184
3185 kvm_s390_vcpu_block_all(kvm);
3186
3187 switch (kvm->arch.crypto.crycbd & CRYCB_FORMAT_MASK) {
3188 case CRYCB_FORMAT2: /* APCB1 use 256 bits */
3189 memcpy(crycb->apcb1.apm, apm, 32);
3190 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx %016lx %016lx %016lx",
3191 apm[0], apm[1], apm[2], apm[3]);
3192 memcpy(crycb->apcb1.aqm, aqm, 32);
3193 VM_EVENT(kvm, 3, "SET CRYCB: aqm %016lx %016lx %016lx %016lx",
3194 aqm[0], aqm[1], aqm[2], aqm[3]);
3195 memcpy(crycb->apcb1.adm, adm, 32);
3196 VM_EVENT(kvm, 3, "SET CRYCB: adm %016lx %016lx %016lx %016lx",
3197 adm[0], adm[1], adm[2], adm[3]);
3198 break;
3199 case CRYCB_FORMAT1:
3200 case CRYCB_FORMAT0: /* Fall through both use APCB0 */
3201 memcpy(crycb->apcb0.apm, apm, 8);
3202 memcpy(crycb->apcb0.aqm, aqm, 2);
3203 memcpy(crycb->apcb0.adm, adm, 2);
3204 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx aqm %04x adm %04x",
3205 apm[0], *((unsigned short *)aqm),
3206 *((unsigned short *)adm));
3207 break;
3208 default: /* Can not happen */
3209 break;
3210 }
3211
3212 /* recreate the shadow crycb for each vcpu */
3213 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART);
3214 kvm_s390_vcpu_unblock_all(kvm);
3215 }
3216 EXPORT_SYMBOL_GPL(kvm_arch_crypto_set_masks);
3217
3218 /*
3219 * kvm_arch_crypto_clear_masks
3220 *
3221 * @kvm: pointer to the target guest's KVM struct containing the crypto masks
3222 * to be cleared.
3223 *
3224 * Clear the masks that identify the adapters, domains and control domains to
3225 * which the KVM guest is granted access.
3226 *
3227 * Note: The kvm->lock mutex must be locked by the caller before invoking this
3228 * function.
3229 */
3230 void kvm_arch_crypto_clear_masks(struct kvm *kvm)
3231 {
3232 kvm_s390_vcpu_block_all(kvm);
3233
3234 memset(&kvm->arch.crypto.crycb->apcb0, 0,
3235 sizeof(kvm->arch.crypto.crycb->apcb0));
3236 memset(&kvm->arch.crypto.crycb->apcb1, 0,
3237 sizeof(kvm->arch.crypto.crycb->apcb1));
3238
3239 VM_EVENT(kvm, 3, "%s", "CLR CRYCB:");
3240 /* recreate the shadow crycb for each vcpu */
3241 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART);
3242 kvm_s390_vcpu_unblock_all(kvm);
3243 }
3244 EXPORT_SYMBOL_GPL(kvm_arch_crypto_clear_masks);
3245
3246 static u64 kvm_s390_get_initial_cpuid(void)
3247 {
3248 struct cpuid cpuid;
3249
3250 get_cpu_id(&cpuid);
3251 cpuid.version = 0xff;
3252 return *((u64 *) &cpuid);
3253 }
3254
3255 static void kvm_s390_crypto_init(struct kvm *kvm)
3256 {
3257 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
3258 kvm_s390_set_crycb_format(kvm);
3259 init_rwsem(&kvm->arch.crypto.pqap_hook_rwsem);
3260
3261 if (!test_kvm_facility(kvm, 76))
3262 return;
3263
3264 /* Enable AES/DEA protected key functions by default */
3265 kvm->arch.crypto.aes_kw = 1;
3266 kvm->arch.crypto.dea_kw = 1;
3267 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
3268 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
3269 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
3270 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
3271 }
3272
3273 static void sca_dispose(struct kvm *kvm)
3274 {
3275 if (kvm->arch.use_esca)
3276 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
3277 else
3278 free_page((unsigned long)(kvm->arch.sca));
3279 kvm->arch.sca = NULL;
3280 }
3281
3282 void kvm_arch_free_vm(struct kvm *kvm)
3283 {
3284 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
3285 kvm_s390_pci_clear_list(kvm);
3286
3287 __kvm_arch_free_vm(kvm);
3288 }
3289
3290 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
3291 {
3292 gfp_t alloc_flags = GFP_KERNEL_ACCOUNT;
3293 int i, rc;
3294 char debug_name[16];
3295 static unsigned long sca_offset;
3296
3297 rc = -EINVAL;
3298 #ifdef CONFIG_KVM_S390_UCONTROL
3299 if (type & ~KVM_VM_S390_UCONTROL)
3300 goto out_err;
3301 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
3302 goto out_err;
3303 #else
3304 if (type)
3305 goto out_err;
3306 #endif
3307
3308 rc = s390_enable_sie();
3309 if (rc)
3310 goto out_err;
3311
3312 rc = -ENOMEM;
3313
3314 if (!sclp.has_64bscao)
3315 alloc_flags |= GFP_DMA;
3316 rwlock_init(&kvm->arch.sca_lock);
3317 /* start with basic SCA */
3318 kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
3319 if (!kvm->arch.sca)
3320 goto out_err;
3321 mutex_lock(&kvm_lock);
3322 sca_offset += 16;
3323 if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
3324 sca_offset = 0;
3325 kvm->arch.sca = (struct bsca_block *)
3326 ((char *) kvm->arch.sca + sca_offset);
3327 mutex_unlock(&kvm_lock);
3328
3329 sprintf(debug_name, "kvm-%u", current->pid);
3330
3331 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
3332 if (!kvm->arch.dbf)
3333 goto out_err;
3334
3335 BUILD_BUG_ON(sizeof(struct sie_page2) != 4096);
3336 kvm->arch.sie_page2 =
3337 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3338 if (!kvm->arch.sie_page2)
3339 goto out_err;
3340
3341 kvm->arch.sie_page2->kvm = kvm;
3342 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
3343
3344 for (i = 0; i < kvm_s390_fac_size(); i++) {
3345 kvm->arch.model.fac_mask[i] = stfle_fac_list[i] &
3346 (kvm_s390_fac_base[i] |
3347 kvm_s390_fac_ext[i]);
3348 kvm->arch.model.fac_list[i] = stfle_fac_list[i] &
3349 kvm_s390_fac_base[i];
3350 }
3351 kvm->arch.model.subfuncs = kvm_s390_available_subfunc;
3352
3353 /* we are always in czam mode - even on pre z14 machines */
3354 set_kvm_facility(kvm->arch.model.fac_mask, 138);
3355 set_kvm_facility(kvm->arch.model.fac_list, 138);
3356 /* we emulate STHYI in kvm */
3357 set_kvm_facility(kvm->arch.model.fac_mask, 74);
3358 set_kvm_facility(kvm->arch.model.fac_list, 74);
3359 if (MACHINE_HAS_TLB_GUEST) {
3360 set_kvm_facility(kvm->arch.model.fac_mask, 147);
3361 set_kvm_facility(kvm->arch.model.fac_list, 147);
3362 }
3363
3364 if (css_general_characteristics.aiv && test_facility(65))
3365 set_kvm_facility(kvm->arch.model.fac_mask, 65);
3366
3367 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
3368 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
3369
3370 kvm->arch.model.uv_feat_guest.feat = 0;
3371
3372 kvm_s390_crypto_init(kvm);
3373
3374 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
3375 mutex_lock(&kvm->lock);
3376 kvm_s390_pci_init_list(kvm);
3377 kvm_s390_vcpu_pci_enable_interp(kvm);
3378 mutex_unlock(&kvm->lock);
3379 }
3380
3381 mutex_init(&kvm->arch.float_int.ais_lock);
3382 spin_lock_init(&kvm->arch.float_int.lock);
3383 for (i = 0; i < FIRQ_LIST_COUNT; i++)
3384 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
3385 init_waitqueue_head(&kvm->arch.ipte_wq);
3386 mutex_init(&kvm->arch.ipte_mutex);
3387
3388 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
3389 VM_EVENT(kvm, 3, "vm created with type %lu", type);
3390
3391 if (type & KVM_VM_S390_UCONTROL) {
3392 kvm->arch.gmap = NULL;
3393 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
3394 } else {
3395 if (sclp.hamax == U64_MAX)
3396 kvm->arch.mem_limit = TASK_SIZE_MAX;
3397 else
3398 kvm->arch.mem_limit = min_t(unsigned long, TASK_SIZE_MAX,
3399 sclp.hamax + 1);
3400 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
3401 if (!kvm->arch.gmap)
3402 goto out_err;
3403 kvm->arch.gmap->private = kvm;
3404 kvm->arch.gmap->pfault_enabled = 0;
3405 }
3406
3407 kvm->arch.use_pfmfi = sclp.has_pfmfi;
3408 kvm->arch.use_skf = sclp.has_skey;
3409 spin_lock_init(&kvm->arch.start_stop_lock);
3410 kvm_s390_vsie_init(kvm);
3411 if (use_gisa)
3412 kvm_s390_gisa_init(kvm);
3413 INIT_LIST_HEAD(&kvm->arch.pv.need_cleanup);
3414 kvm->arch.pv.set_aside = NULL;
3415 KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
3416
3417 return 0;
3418 out_err:
3419 free_page((unsigned long)kvm->arch.sie_page2);
3420 debug_unregister(kvm->arch.dbf);
3421 sca_dispose(kvm);
3422 KVM_EVENT(3, "creation of vm failed: %d", rc);
3423 return rc;
3424 }
3425
3426 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3427 {
3428 u16 rc, rrc;
3429
3430 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
3431 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
3432 kvm_s390_clear_local_irqs(vcpu);
3433 kvm_clear_async_pf_completion_queue(vcpu);
3434 if (!kvm_is_ucontrol(vcpu->kvm))
3435 sca_del_vcpu(vcpu);
3436 kvm_s390_update_topology_change_report(vcpu->kvm, 1);
3437
3438 if (kvm_is_ucontrol(vcpu->kvm))
3439 gmap_remove(vcpu->arch.gmap);
3440
3441 if (vcpu->kvm->arch.use_cmma)
3442 kvm_s390_vcpu_unsetup_cmma(vcpu);
3443 /* We can not hold the vcpu mutex here, we are already dying */
3444 if (kvm_s390_pv_cpu_get_handle(vcpu))
3445 kvm_s390_pv_destroy_cpu(vcpu, &rc, &rrc);
3446 free_page((unsigned long)(vcpu->arch.sie_block));
3447 }
3448
3449 void kvm_arch_destroy_vm(struct kvm *kvm)
3450 {
3451 u16 rc, rrc;
3452
3453 kvm_destroy_vcpus(kvm);
3454 sca_dispose(kvm);
3455 kvm_s390_gisa_destroy(kvm);
3456 /*
3457 * We are already at the end of life and kvm->lock is not taken.
3458 * This is ok as the file descriptor is closed by now and nobody
3459 * can mess with the pv state.
3460 */
3461 kvm_s390_pv_deinit_cleanup_all(kvm, &rc, &rrc);
3462 /*
3463 * Remove the mmu notifier only when the whole KVM VM is torn down,
3464 * and only if one was registered to begin with. If the VM is
3465 * currently not protected, but has been previously been protected,
3466 * then it's possible that the notifier is still registered.
3467 */
3468 if (kvm->arch.pv.mmu_notifier.ops)
3469 mmu_notifier_unregister(&kvm->arch.pv.mmu_notifier, kvm->mm);
3470
3471 debug_unregister(kvm->arch.dbf);
3472 free_page((unsigned long)kvm->arch.sie_page2);
3473 if (!kvm_is_ucontrol(kvm))
3474 gmap_remove(kvm->arch.gmap);
3475 kvm_s390_destroy_adapters(kvm);
3476 kvm_s390_clear_float_irqs(kvm);
3477 kvm_s390_vsie_destroy(kvm);
3478 KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
3479 }
3480
3481 /* Section: vcpu related */
3482 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
3483 {
3484 vcpu->arch.gmap = gmap_create(current->mm, -1UL);
3485 if (!vcpu->arch.gmap)
3486 return -ENOMEM;
3487 vcpu->arch.gmap->private = vcpu->kvm;
3488
3489 return 0;
3490 }
3491
3492 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
3493 {
3494 if (!kvm_s390_use_sca_entries())
3495 return;
3496 read_lock(&vcpu->kvm->arch.sca_lock);
3497 if (vcpu->kvm->arch.use_esca) {
3498 struct esca_block *sca = vcpu->kvm->arch.sca;
3499
3500 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
3501 sca->cpu[vcpu->vcpu_id].sda = 0;
3502 } else {
3503 struct bsca_block *sca = vcpu->kvm->arch.sca;
3504
3505 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
3506 sca->cpu[vcpu->vcpu_id].sda = 0;
3507 }
3508 read_unlock(&vcpu->kvm->arch.sca_lock);
3509 }
3510
3511 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
3512 {
3513 if (!kvm_s390_use_sca_entries()) {
3514 phys_addr_t sca_phys = virt_to_phys(vcpu->kvm->arch.sca);
3515
3516 /* we still need the basic sca for the ipte control */
3517 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3518 vcpu->arch.sie_block->scaol = sca_phys;
3519 return;
3520 }
3521 read_lock(&vcpu->kvm->arch.sca_lock);
3522 if (vcpu->kvm->arch.use_esca) {
3523 struct esca_block *sca = vcpu->kvm->arch.sca;
3524 phys_addr_t sca_phys = virt_to_phys(sca);
3525
3526 sca->cpu[vcpu->vcpu_id].sda = virt_to_phys(vcpu->arch.sie_block);
3527 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3528 vcpu->arch.sie_block->scaol = sca_phys & ESCA_SCAOL_MASK;
3529 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
3530 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
3531 } else {
3532 struct bsca_block *sca = vcpu->kvm->arch.sca;
3533 phys_addr_t sca_phys = virt_to_phys(sca);
3534
3535 sca->cpu[vcpu->vcpu_id].sda = virt_to_phys(vcpu->arch.sie_block);
3536 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3537 vcpu->arch.sie_block->scaol = sca_phys;
3538 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
3539 }
3540 read_unlock(&vcpu->kvm->arch.sca_lock);
3541 }
3542
3543 /* Basic SCA to Extended SCA data copy routines */
3544 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
3545 {
3546 d->sda = s->sda;
3547 d->sigp_ctrl.c = s->sigp_ctrl.c;
3548 d->sigp_ctrl.scn = s->sigp_ctrl.scn;
3549 }
3550
3551 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
3552 {
3553 int i;
3554
3555 d->ipte_control = s->ipte_control;
3556 d->mcn[0] = s->mcn;
3557 for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
3558 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
3559 }
3560
3561 static int sca_switch_to_extended(struct kvm *kvm)
3562 {
3563 struct bsca_block *old_sca = kvm->arch.sca;
3564 struct esca_block *new_sca;
3565 struct kvm_vcpu *vcpu;
3566 unsigned long vcpu_idx;
3567 u32 scaol, scaoh;
3568 phys_addr_t new_sca_phys;
3569
3570 if (kvm->arch.use_esca)
3571 return 0;
3572
3573 new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3574 if (!new_sca)
3575 return -ENOMEM;
3576
3577 new_sca_phys = virt_to_phys(new_sca);
3578 scaoh = new_sca_phys >> 32;
3579 scaol = new_sca_phys & ESCA_SCAOL_MASK;
3580
3581 kvm_s390_vcpu_block_all(kvm);
3582 write_lock(&kvm->arch.sca_lock);
3583
3584 sca_copy_b_to_e(new_sca, old_sca);
3585
3586 kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
3587 vcpu->arch.sie_block->scaoh = scaoh;
3588 vcpu->arch.sie_block->scaol = scaol;
3589 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
3590 }
3591 kvm->arch.sca = new_sca;
3592 kvm->arch.use_esca = 1;
3593
3594 write_unlock(&kvm->arch.sca_lock);
3595 kvm_s390_vcpu_unblock_all(kvm);
3596
3597 free_page((unsigned long)old_sca);
3598
3599 VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
3600 old_sca, kvm->arch.sca);
3601 return 0;
3602 }
3603
3604 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
3605 {
3606 int rc;
3607
3608 if (!kvm_s390_use_sca_entries()) {
3609 if (id < KVM_MAX_VCPUS)
3610 return true;
3611 return false;
3612 }
3613 if (id < KVM_S390_BSCA_CPU_SLOTS)
3614 return true;
3615 if (!sclp.has_esca || !sclp.has_64bscao)
3616 return false;
3617
3618 rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
3619
3620 return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
3621 }
3622
3623 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3624 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3625 {
3626 WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
3627 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3628 vcpu->arch.cputm_start = get_tod_clock_fast();
3629 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3630 }
3631
3632 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3633 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3634 {
3635 WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
3636 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3637 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
3638 vcpu->arch.cputm_start = 0;
3639 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3640 }
3641
3642 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3643 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3644 {
3645 WARN_ON_ONCE(vcpu->arch.cputm_enabled);
3646 vcpu->arch.cputm_enabled = true;
3647 __start_cpu_timer_accounting(vcpu);
3648 }
3649
3650 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
3651 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3652 {
3653 WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
3654 __stop_cpu_timer_accounting(vcpu);
3655 vcpu->arch.cputm_enabled = false;
3656 }
3657
3658 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3659 {
3660 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3661 __enable_cpu_timer_accounting(vcpu);
3662 preempt_enable();
3663 }
3664
3665 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3666 {
3667 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3668 __disable_cpu_timer_accounting(vcpu);
3669 preempt_enable();
3670 }
3671
3672 /* set the cpu timer - may only be called from the VCPU thread itself */
3673 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
3674 {
3675 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3676 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3677 if (vcpu->arch.cputm_enabled)
3678 vcpu->arch.cputm_start = get_tod_clock_fast();
3679 vcpu->arch.sie_block->cputm = cputm;
3680 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3681 preempt_enable();
3682 }
3683
3684 /* update and get the cpu timer - can also be called from other VCPU threads */
3685 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
3686 {
3687 unsigned int seq;
3688 __u64 value;
3689
3690 if (unlikely(!vcpu->arch.cputm_enabled))
3691 return vcpu->arch.sie_block->cputm;
3692
3693 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3694 do {
3695 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
3696 /*
3697 * If the writer would ever execute a read in the critical
3698 * section, e.g. in irq context, we have a deadlock.
3699 */
3700 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
3701 value = vcpu->arch.sie_block->cputm;
3702 /* if cputm_start is 0, accounting is being started/stopped */
3703 if (likely(vcpu->arch.cputm_start))
3704 value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
3705 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
3706 preempt_enable();
3707 return value;
3708 }
3709
3710 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3711 {
3712
3713 gmap_enable(vcpu->arch.enabled_gmap);
3714 kvm_s390_set_cpuflags(vcpu, CPUSTAT_RUNNING);
3715 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
3716 __start_cpu_timer_accounting(vcpu);
3717 vcpu->cpu = cpu;
3718 }
3719
3720 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3721 {
3722 vcpu->cpu = -1;
3723 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
3724 __stop_cpu_timer_accounting(vcpu);
3725 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_RUNNING);
3726 vcpu->arch.enabled_gmap = gmap_get_enabled();
3727 gmap_disable(vcpu->arch.enabled_gmap);
3728
3729 }
3730
3731 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
3732 {
3733 mutex_lock(&vcpu->kvm->lock);
3734 preempt_disable();
3735 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
3736 vcpu->arch.sie_block->epdx = vcpu->kvm->arch.epdx;
3737 preempt_enable();
3738 mutex_unlock(&vcpu->kvm->lock);
3739 if (!kvm_is_ucontrol(vcpu->kvm)) {
3740 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
3741 sca_add_vcpu(vcpu);
3742 }
3743 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
3744 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
3745 /* make vcpu_load load the right gmap on the first trigger */
3746 vcpu->arch.enabled_gmap = vcpu->arch.gmap;
3747 }
3748
3749 static bool kvm_has_pckmo_subfunc(struct kvm *kvm, unsigned long nr)
3750 {
3751 if (test_bit_inv(nr, (unsigned long *)&kvm->arch.model.subfuncs.pckmo) &&
3752 test_bit_inv(nr, (unsigned long *)&kvm_s390_available_subfunc.pckmo))
3753 return true;
3754 return false;
3755 }
3756
3757 static bool kvm_has_pckmo_ecc(struct kvm *kvm)
3758 {
3759 /* At least one ECC subfunction must be present */
3760 return kvm_has_pckmo_subfunc(kvm, 32) ||
3761 kvm_has_pckmo_subfunc(kvm, 33) ||
3762 kvm_has_pckmo_subfunc(kvm, 34) ||
3763 kvm_has_pckmo_subfunc(kvm, 40) ||
3764 kvm_has_pckmo_subfunc(kvm, 41);
3765
3766 }
3767
3768 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
3769 {
3770 /*
3771 * If the AP instructions are not being interpreted and the MSAX3
3772 * facility is not configured for the guest, there is nothing to set up.
3773 */
3774 if (!vcpu->kvm->arch.crypto.apie && !test_kvm_facility(vcpu->kvm, 76))
3775 return;
3776
3777 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
3778 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
3779 vcpu->arch.sie_block->eca &= ~ECA_APIE;
3780 vcpu->arch.sie_block->ecd &= ~ECD_ECC;
3781
3782 if (vcpu->kvm->arch.crypto.apie)
3783 vcpu->arch.sie_block->eca |= ECA_APIE;
3784
3785 /* Set up protected key support */
3786 if (vcpu->kvm->arch.crypto.aes_kw) {
3787 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
3788 /* ecc is also wrapped with AES key */
3789 if (kvm_has_pckmo_ecc(vcpu->kvm))
3790 vcpu->arch.sie_block->ecd |= ECD_ECC;
3791 }
3792
3793 if (vcpu->kvm->arch.crypto.dea_kw)
3794 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
3795 }
3796
3797 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
3798 {
3799 free_page((unsigned long)phys_to_virt(vcpu->arch.sie_block->cbrlo));
3800 vcpu->arch.sie_block->cbrlo = 0;
3801 }
3802
3803 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
3804 {
3805 void *cbrlo_page = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
3806
3807 if (!cbrlo_page)
3808 return -ENOMEM;
3809
3810 vcpu->arch.sie_block->cbrlo = virt_to_phys(cbrlo_page);
3811 return 0;
3812 }
3813
3814 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
3815 {
3816 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
3817
3818 vcpu->arch.sie_block->ibc = model->ibc;
3819 if (test_kvm_facility(vcpu->kvm, 7))
3820 vcpu->arch.sie_block->fac = virt_to_phys(model->fac_list);
3821 }
3822
3823 static int kvm_s390_vcpu_setup(struct kvm_vcpu *vcpu)
3824 {
3825 int rc = 0;
3826 u16 uvrc, uvrrc;
3827
3828 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
3829 CPUSTAT_SM |
3830 CPUSTAT_STOPPED);
3831
3832 if (test_kvm_facility(vcpu->kvm, 78))
3833 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED2);
3834 else if (test_kvm_facility(vcpu->kvm, 8))
3835 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED);
3836
3837 kvm_s390_vcpu_setup_model(vcpu);
3838
3839 /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
3840 if (MACHINE_HAS_ESOP)
3841 vcpu->arch.sie_block->ecb |= ECB_HOSTPROTINT;
3842 if (test_kvm_facility(vcpu->kvm, 9))
3843 vcpu->arch.sie_block->ecb |= ECB_SRSI;
3844 if (test_kvm_facility(vcpu->kvm, 11))
3845 vcpu->arch.sie_block->ecb |= ECB_PTF;
3846 if (test_kvm_facility(vcpu->kvm, 73))
3847 vcpu->arch.sie_block->ecb |= ECB_TE;
3848 if (!kvm_is_ucontrol(vcpu->kvm))
3849 vcpu->arch.sie_block->ecb |= ECB_SPECI;
3850
3851 if (test_kvm_facility(vcpu->kvm, 8) && vcpu->kvm->arch.use_pfmfi)
3852 vcpu->arch.sie_block->ecb2 |= ECB2_PFMFI;
3853 if (test_kvm_facility(vcpu->kvm, 130))
3854 vcpu->arch.sie_block->ecb2 |= ECB2_IEP;
3855 vcpu->arch.sie_block->eca = ECA_MVPGI | ECA_PROTEXCI;
3856 if (sclp.has_cei)
3857 vcpu->arch.sie_block->eca |= ECA_CEI;
3858 if (sclp.has_ib)
3859 vcpu->arch.sie_block->eca |= ECA_IB;
3860 if (sclp.has_siif)
3861 vcpu->arch.sie_block->eca |= ECA_SII;
3862 if (sclp.has_sigpif)
3863 vcpu->arch.sie_block->eca |= ECA_SIGPI;
3864 if (test_kvm_facility(vcpu->kvm, 129)) {
3865 vcpu->arch.sie_block->eca |= ECA_VX;
3866 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
3867 }
3868 if (test_kvm_facility(vcpu->kvm, 139))
3869 vcpu->arch.sie_block->ecd |= ECD_MEF;
3870 if (test_kvm_facility(vcpu->kvm, 156))
3871 vcpu->arch.sie_block->ecd |= ECD_ETOKENF;
3872 if (vcpu->arch.sie_block->gd) {
3873 vcpu->arch.sie_block->eca |= ECA_AIV;
3874 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
3875 vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
3876 }
3877 vcpu->arch.sie_block->sdnxo = virt_to_phys(&vcpu->run->s.regs.sdnx) | SDNXC;
3878 vcpu->arch.sie_block->riccbd = virt_to_phys(&vcpu->run->s.regs.riccb);
3879
3880 if (sclp.has_kss)
3881 kvm_s390_set_cpuflags(vcpu, CPUSTAT_KSS);
3882 else
3883 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
3884
3885 if (vcpu->kvm->arch.use_cmma) {
3886 rc = kvm_s390_vcpu_setup_cmma(vcpu);
3887 if (rc)
3888 return rc;
3889 }
3890 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3891 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
3892
3893 vcpu->arch.sie_block->hpid = HPID_KVM;
3894
3895 kvm_s390_vcpu_crypto_setup(vcpu);
3896
3897 kvm_s390_vcpu_pci_setup(vcpu);
3898
3899 mutex_lock(&vcpu->kvm->lock);
3900 if (kvm_s390_pv_is_protected(vcpu->kvm)) {
3901 rc = kvm_s390_pv_create_cpu(vcpu, &uvrc, &uvrrc);
3902 if (rc)
3903 kvm_s390_vcpu_unsetup_cmma(vcpu);
3904 }
3905 mutex_unlock(&vcpu->kvm->lock);
3906
3907 return rc;
3908 }
3909
3910 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
3911 {
3912 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
3913 return -EINVAL;
3914 return 0;
3915 }
3916
3917 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
3918 {
3919 struct sie_page *sie_page;
3920 int rc;
3921
3922 BUILD_BUG_ON(sizeof(struct sie_page) != 4096);
3923 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL_ACCOUNT);
3924 if (!sie_page)
3925 return -ENOMEM;
3926
3927 vcpu->arch.sie_block = &sie_page->sie_block;
3928 vcpu->arch.sie_block->itdba = virt_to_phys(&sie_page->itdb);
3929
3930 /* the real guest size will always be smaller than msl */
3931 vcpu->arch.sie_block->mso = 0;
3932 vcpu->arch.sie_block->msl = sclp.hamax;
3933
3934 vcpu->arch.sie_block->icpua = vcpu->vcpu_id;
3935 spin_lock_init(&vcpu->arch.local_int.lock);
3936 vcpu->arch.sie_block->gd = kvm_s390_get_gisa_desc(vcpu->kvm);
3937 seqcount_init(&vcpu->arch.cputm_seqcount);
3938
3939 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
3940 kvm_clear_async_pf_completion_queue(vcpu);
3941 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
3942 KVM_SYNC_GPRS |
3943 KVM_SYNC_ACRS |
3944 KVM_SYNC_CRS |
3945 KVM_SYNC_ARCH0 |
3946 KVM_SYNC_PFAULT |
3947 KVM_SYNC_DIAG318;
3948 vcpu->arch.acrs_loaded = false;
3949 kvm_s390_set_prefix(vcpu, 0);
3950 if (test_kvm_facility(vcpu->kvm, 64))
3951 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
3952 if (test_kvm_facility(vcpu->kvm, 82))
3953 vcpu->run->kvm_valid_regs |= KVM_SYNC_BPBC;
3954 if (test_kvm_facility(vcpu->kvm, 133))
3955 vcpu->run->kvm_valid_regs |= KVM_SYNC_GSCB;
3956 if (test_kvm_facility(vcpu->kvm, 156))
3957 vcpu->run->kvm_valid_regs |= KVM_SYNC_ETOKEN;
3958 /* fprs can be synchronized via vrs, even if the guest has no vx. With
3959 * cpu_has_vx(), (load|store)_fpu_regs() will work with vrs format.
3960 */
3961 if (cpu_has_vx())
3962 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
3963 else
3964 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
3965
3966 if (kvm_is_ucontrol(vcpu->kvm)) {
3967 rc = __kvm_ucontrol_vcpu_init(vcpu);
3968 if (rc)
3969 goto out_free_sie_block;
3970 }
3971
3972 VM_EVENT(vcpu->kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK",
3973 vcpu->vcpu_id, vcpu, vcpu->arch.sie_block);
3974 trace_kvm_s390_create_vcpu(vcpu->vcpu_id, vcpu, vcpu->arch.sie_block);
3975
3976 rc = kvm_s390_vcpu_setup(vcpu);
3977 if (rc)
3978 goto out_ucontrol_uninit;
3979
3980 kvm_s390_update_topology_change_report(vcpu->kvm, 1);
3981 return 0;
3982
3983 out_ucontrol_uninit:
3984 if (kvm_is_ucontrol(vcpu->kvm))
3985 gmap_remove(vcpu->arch.gmap);
3986 out_free_sie_block:
3987 free_page((unsigned long)(vcpu->arch.sie_block));
3988 return rc;
3989 }
3990
3991 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
3992 {
3993 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask);
3994 return kvm_s390_vcpu_has_irq(vcpu, 0);
3995 }
3996
3997 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
3998 {
3999 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE);
4000 }
4001
4002 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
4003 {
4004 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
4005 exit_sie(vcpu);
4006 }
4007
4008 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
4009 {
4010 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
4011 }
4012
4013 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
4014 {
4015 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
4016 exit_sie(vcpu);
4017 }
4018
4019 bool kvm_s390_vcpu_sie_inhibited(struct kvm_vcpu *vcpu)
4020 {
4021 return atomic_read(&vcpu->arch.sie_block->prog20) &
4022 (PROG_BLOCK_SIE | PROG_REQUEST);
4023 }
4024
4025 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
4026 {
4027 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
4028 }
4029
4030 /*
4031 * Kick a guest cpu out of (v)SIE and wait until (v)SIE is not running.
4032 * If the CPU is not running (e.g. waiting as idle) the function will
4033 * return immediately. */
4034 void exit_sie(struct kvm_vcpu *vcpu)
4035 {
4036 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
4037 kvm_s390_vsie_kick(vcpu);
4038 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
4039 cpu_relax();
4040 }
4041
4042 /* Kick a guest cpu out of SIE to process a request synchronously */
4043 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
4044 {
4045 __kvm_make_request(req, vcpu);
4046 kvm_s390_vcpu_request(vcpu);
4047 }
4048
4049 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
4050 unsigned long end)
4051 {
4052 struct kvm *kvm = gmap->private;
4053 struct kvm_vcpu *vcpu;
4054 unsigned long prefix;
4055 unsigned long i;
4056
4057 trace_kvm_s390_gmap_notifier(start, end, gmap_is_shadow(gmap));
4058
4059 if (gmap_is_shadow(gmap))
4060 return;
4061 if (start >= 1UL << 31)
4062 /* We are only interested in prefix pages */
4063 return;
4064 kvm_for_each_vcpu(i, vcpu, kvm) {
4065 /* match against both prefix pages */
4066 prefix = kvm_s390_get_prefix(vcpu);
4067 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
4068 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
4069 start, end);
4070 kvm_s390_sync_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu);
4071 }
4072 }
4073 }
4074
4075 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
4076 {
4077 /* do not poll with more than halt_poll_max_steal percent of steal time */
4078 if (get_lowcore()->avg_steal_timer * 100 / (TICK_USEC << 12) >=
4079 READ_ONCE(halt_poll_max_steal)) {
4080 vcpu->stat.halt_no_poll_steal++;
4081 return true;
4082 }
4083 return false;
4084 }
4085
4086 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
4087 {
4088 /* kvm common code refers to this, but never calls it */
4089 BUG();
4090 return 0;
4091 }
4092
4093 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
4094 struct kvm_one_reg *reg)
4095 {
4096 int r = -EINVAL;
4097
4098 switch (reg->id) {
4099 case KVM_REG_S390_TODPR:
4100 r = put_user(vcpu->arch.sie_block->todpr,
4101 (u32 __user *)reg->addr);
4102 break;
4103 case KVM_REG_S390_EPOCHDIFF:
4104 r = put_user(vcpu->arch.sie_block->epoch,
4105 (u64 __user *)reg->addr);
4106 break;
4107 case KVM_REG_S390_CPU_TIMER:
4108 r = put_user(kvm_s390_get_cpu_timer(vcpu),
4109 (u64 __user *)reg->addr);
4110 break;
4111 case KVM_REG_S390_CLOCK_COMP:
4112 r = put_user(vcpu->arch.sie_block->ckc,
4113 (u64 __user *)reg->addr);
4114 break;
4115 case KVM_REG_S390_PFTOKEN:
4116 r = put_user(vcpu->arch.pfault_token,
4117 (u64 __user *)reg->addr);
4118 break;
4119 case KVM_REG_S390_PFCOMPARE:
4120 r = put_user(vcpu->arch.pfault_compare,
4121 (u64 __user *)reg->addr);
4122 break;
4123 case KVM_REG_S390_PFSELECT:
4124 r = put_user(vcpu->arch.pfault_select,
4125 (u64 __user *)reg->addr);
4126 break;
4127 case KVM_REG_S390_PP:
4128 r = put_user(vcpu->arch.sie_block->pp,
4129 (u64 __user *)reg->addr);
4130 break;
4131 case KVM_REG_S390_GBEA:
4132 r = put_user(vcpu->arch.sie_block->gbea,
4133 (u64 __user *)reg->addr);
4134 break;
4135 default:
4136 break;
4137 }
4138
4139 return r;
4140 }
4141
4142 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
4143 struct kvm_one_reg *reg)
4144 {
4145 int r = -EINVAL;
4146 __u64 val;
4147
4148 switch (reg->id) {
4149 case KVM_REG_S390_TODPR:
4150 r = get_user(vcpu->arch.sie_block->todpr,
4151 (u32 __user *)reg->addr);
4152 break;
4153 case KVM_REG_S390_EPOCHDIFF:
4154 r = get_user(vcpu->arch.sie_block->epoch,
4155 (u64 __user *)reg->addr);
4156 break;
4157 case KVM_REG_S390_CPU_TIMER:
4158 r = get_user(val, (u64 __user *)reg->addr);
4159 if (!r)
4160 kvm_s390_set_cpu_timer(vcpu, val);
4161 break;
4162 case KVM_REG_S390_CLOCK_COMP:
4163 r = get_user(vcpu->arch.sie_block->ckc,
4164 (u64 __user *)reg->addr);
4165 break;
4166 case KVM_REG_S390_PFTOKEN:
4167 r = get_user(vcpu->arch.pfault_token,
4168 (u64 __user *)reg->addr);
4169 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4170 kvm_clear_async_pf_completion_queue(vcpu);
4171 break;
4172 case KVM_REG_S390_PFCOMPARE:
4173 r = get_user(vcpu->arch.pfault_compare,
4174 (u64 __user *)reg->addr);
4175 break;
4176 case KVM_REG_S390_PFSELECT:
4177 r = get_user(vcpu->arch.pfault_select,
4178 (u64 __user *)reg->addr);
4179 break;
4180 case KVM_REG_S390_PP:
4181 r = get_user(vcpu->arch.sie_block->pp,
4182 (u64 __user *)reg->addr);
4183 break;
4184 case KVM_REG_S390_GBEA:
4185 r = get_user(vcpu->arch.sie_block->gbea,
4186 (u64 __user *)reg->addr);
4187 break;
4188 default:
4189 break;
4190 }
4191
4192 return r;
4193 }
4194
4195 static void kvm_arch_vcpu_ioctl_normal_reset(struct kvm_vcpu *vcpu)
4196 {
4197 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_RI;
4198 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
4199 memset(vcpu->run->s.regs.riccb, 0, sizeof(vcpu->run->s.regs.riccb));
4200
4201 kvm_clear_async_pf_completion_queue(vcpu);
4202 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
4203 kvm_s390_vcpu_stop(vcpu);
4204 kvm_s390_clear_local_irqs(vcpu);
4205 }
4206
4207 static void kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
4208 {
4209 /* Initial reset is a superset of the normal reset */
4210 kvm_arch_vcpu_ioctl_normal_reset(vcpu);
4211
4212 /*
4213 * This equals initial cpu reset in pop, but we don't switch to ESA.
4214 * We do not only reset the internal data, but also ...
4215 */
4216 vcpu->arch.sie_block->gpsw.mask = 0;
4217 vcpu->arch.sie_block->gpsw.addr = 0;
4218 kvm_s390_set_prefix(vcpu, 0);
4219 kvm_s390_set_cpu_timer(vcpu, 0);
4220 vcpu->arch.sie_block->ckc = 0;
4221 memset(vcpu->arch.sie_block->gcr, 0, sizeof(vcpu->arch.sie_block->gcr));
4222 vcpu->arch.sie_block->gcr[0] = CR0_INITIAL_MASK;
4223 vcpu->arch.sie_block->gcr[14] = CR14_INITIAL_MASK;
4224
4225 /* ... the data in sync regs */
4226 memset(vcpu->run->s.regs.crs, 0, sizeof(vcpu->run->s.regs.crs));
4227 vcpu->run->s.regs.ckc = 0;
4228 vcpu->run->s.regs.crs[0] = CR0_INITIAL_MASK;
4229 vcpu->run->s.regs.crs[14] = CR14_INITIAL_MASK;
4230 vcpu->run->psw_addr = 0;
4231 vcpu->run->psw_mask = 0;
4232 vcpu->run->s.regs.todpr = 0;
4233 vcpu->run->s.regs.cputm = 0;
4234 vcpu->run->s.regs.ckc = 0;
4235 vcpu->run->s.regs.pp = 0;
4236 vcpu->run->s.regs.gbea = 1;
4237 vcpu->run->s.regs.fpc = 0;
4238 /*
4239 * Do not reset these registers in the protected case, as some of
4240 * them are overlaid and they are not accessible in this case
4241 * anyway.
4242 */
4243 if (!kvm_s390_pv_cpu_is_protected(vcpu)) {
4244 vcpu->arch.sie_block->gbea = 1;
4245 vcpu->arch.sie_block->pp = 0;
4246 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
4247 vcpu->arch.sie_block->todpr = 0;
4248 }
4249 }
4250
4251 static void kvm_arch_vcpu_ioctl_clear_reset(struct kvm_vcpu *vcpu)
4252 {
4253 struct kvm_sync_regs *regs = &vcpu->run->s.regs;
4254
4255 /* Clear reset is a superset of the initial reset */
4256 kvm_arch_vcpu_ioctl_initial_reset(vcpu);
4257
4258 memset(®s->gprs, 0, sizeof(regs->gprs));
4259 memset(®s->vrs, 0, sizeof(regs->vrs));
4260 memset(®s->acrs, 0, sizeof(regs->acrs));
4261 memset(®s->gscb, 0, sizeof(regs->gscb));
4262
4263 regs->etoken = 0;
4264 regs->etoken_extension = 0;
4265 }
4266
4267 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4268 {
4269 vcpu_load(vcpu);
4270 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs));
4271 vcpu_put(vcpu);
4272 return 0;
4273 }
4274
4275 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4276 {
4277 vcpu_load(vcpu);
4278 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
4279 vcpu_put(vcpu);
4280 return 0;
4281 }
4282
4283 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4284 struct kvm_sregs *sregs)
4285 {
4286 vcpu_load(vcpu);
4287
4288 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
4289 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
4290
4291 vcpu_put(vcpu);
4292 return 0;
4293 }
4294
4295 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4296 struct kvm_sregs *sregs)
4297 {
4298 vcpu_load(vcpu);
4299
4300 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
4301 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
4302
4303 vcpu_put(vcpu);
4304 return 0;
4305 }
4306
4307 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4308 {
4309 int ret = 0;
4310
4311 vcpu_load(vcpu);
4312
4313 vcpu->run->s.regs.fpc = fpu->fpc;
4314 if (cpu_has_vx())
4315 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs,
4316 (freg_t *) fpu->fprs);
4317 else
4318 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs));
4319
4320 vcpu_put(vcpu);
4321 return ret;
4322 }
4323
4324 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4325 {
4326 vcpu_load(vcpu);
4327
4328 if (cpu_has_vx())
4329 convert_vx_to_fp((freg_t *) fpu->fprs,
4330 (__vector128 *) vcpu->run->s.regs.vrs);
4331 else
4332 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs));
4333 fpu->fpc = vcpu->run->s.regs.fpc;
4334
4335 vcpu_put(vcpu);
4336 return 0;
4337 }
4338
4339 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
4340 {
4341 int rc = 0;
4342
4343 if (!is_vcpu_stopped(vcpu))
4344 rc = -EBUSY;
4345 else {
4346 vcpu->run->psw_mask = psw.mask;
4347 vcpu->run->psw_addr = psw.addr;
4348 }
4349 return rc;
4350 }
4351
4352 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4353 struct kvm_translation *tr)
4354 {
4355 return -EINVAL; /* not implemented yet */
4356 }
4357
4358 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
4359 KVM_GUESTDBG_USE_HW_BP | \
4360 KVM_GUESTDBG_ENABLE)
4361
4362 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4363 struct kvm_guest_debug *dbg)
4364 {
4365 int rc = 0;
4366
4367 vcpu_load(vcpu);
4368
4369 vcpu->guest_debug = 0;
4370 kvm_s390_clear_bp_data(vcpu);
4371
4372 if (dbg->control & ~VALID_GUESTDBG_FLAGS) {
4373 rc = -EINVAL;
4374 goto out;
4375 }
4376 if (!sclp.has_gpere) {
4377 rc = -EINVAL;
4378 goto out;
4379 }
4380
4381 if (dbg->control & KVM_GUESTDBG_ENABLE) {
4382 vcpu->guest_debug = dbg->control;
4383 /* enforce guest PER */
4384 kvm_s390_set_cpuflags(vcpu, CPUSTAT_P);
4385
4386 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
4387 rc = kvm_s390_import_bp_data(vcpu, dbg);
4388 } else {
4389 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
4390 vcpu->arch.guestdbg.last_bp = 0;
4391 }
4392
4393 if (rc) {
4394 vcpu->guest_debug = 0;
4395 kvm_s390_clear_bp_data(vcpu);
4396 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
4397 }
4398
4399 out:
4400 vcpu_put(vcpu);
4401 return rc;
4402 }
4403
4404 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4405 struct kvm_mp_state *mp_state)
4406 {
4407 int ret;
4408
4409 vcpu_load(vcpu);
4410
4411 /* CHECK_STOP and LOAD are not supported yet */
4412 ret = is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
4413 KVM_MP_STATE_OPERATING;
4414
4415 vcpu_put(vcpu);
4416 return ret;
4417 }
4418
4419 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4420 struct kvm_mp_state *mp_state)
4421 {
4422 int rc = 0;
4423
4424 vcpu_load(vcpu);
4425
4426 /* user space knows about this interface - let it control the state */
4427 kvm_s390_set_user_cpu_state_ctrl(vcpu->kvm);
4428
4429 switch (mp_state->mp_state) {
4430 case KVM_MP_STATE_STOPPED:
4431 rc = kvm_s390_vcpu_stop(vcpu);
4432 break;
4433 case KVM_MP_STATE_OPERATING:
4434 rc = kvm_s390_vcpu_start(vcpu);
4435 break;
4436 case KVM_MP_STATE_LOAD:
4437 if (!kvm_s390_pv_cpu_is_protected(vcpu)) {
4438 rc = -ENXIO;
4439 break;
4440 }
4441 rc = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR_LOAD);
4442 break;
4443 case KVM_MP_STATE_CHECK_STOP:
4444 fallthrough; /* CHECK_STOP and LOAD are not supported yet */
4445 default:
4446 rc = -ENXIO;
4447 }
4448
4449 vcpu_put(vcpu);
4450 return rc;
4451 }
4452
4453 static bool ibs_enabled(struct kvm_vcpu *vcpu)
4454 {
4455 return kvm_s390_test_cpuflags(vcpu, CPUSTAT_IBS);
4456 }
4457
4458 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
4459 {
4460 retry:
4461 kvm_s390_vcpu_request_handled(vcpu);
4462 if (!kvm_request_pending(vcpu))
4463 return 0;
4464 /*
4465 * If the guest prefix changed, re-arm the ipte notifier for the
4466 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
4467 * This ensures that the ipte instruction for this request has
4468 * already finished. We might race against a second unmapper that
4469 * wants to set the blocking bit. Lets just retry the request loop.
4470 */
4471 if (kvm_check_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu)) {
4472 int rc;
4473 rc = gmap_mprotect_notify(vcpu->arch.gmap,
4474 kvm_s390_get_prefix(vcpu),
4475 PAGE_SIZE * 2, PROT_WRITE);
4476 if (rc) {
4477 kvm_make_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu);
4478 return rc;
4479 }
4480 goto retry;
4481 }
4482
4483 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
4484 vcpu->arch.sie_block->ihcpu = 0xffff;
4485 goto retry;
4486 }
4487
4488 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
4489 if (!ibs_enabled(vcpu)) {
4490 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
4491 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IBS);
4492 }
4493 goto retry;
4494 }
4495
4496 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
4497 if (ibs_enabled(vcpu)) {
4498 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
4499 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IBS);
4500 }
4501 goto retry;
4502 }
4503
4504 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
4505 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
4506 goto retry;
4507 }
4508
4509 if (kvm_check_request(KVM_REQ_START_MIGRATION, vcpu)) {
4510 /*
4511 * Disable CMM virtualization; we will emulate the ESSA
4512 * instruction manually, in order to provide additional
4513 * functionalities needed for live migration.
4514 */
4515 vcpu->arch.sie_block->ecb2 &= ~ECB2_CMMA;
4516 goto retry;
4517 }
4518
4519 if (kvm_check_request(KVM_REQ_STOP_MIGRATION, vcpu)) {
4520 /*
4521 * Re-enable CMM virtualization if CMMA is available and
4522 * CMM has been used.
4523 */
4524 if ((vcpu->kvm->arch.use_cmma) &&
4525 (vcpu->kvm->mm->context.uses_cmm))
4526 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
4527 goto retry;
4528 }
4529
4530 /* we left the vsie handler, nothing to do, just clear the request */
4531 kvm_clear_request(KVM_REQ_VSIE_RESTART, vcpu);
4532
4533 return 0;
4534 }
4535
4536 static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod)
4537 {
4538 struct kvm_vcpu *vcpu;
4539 union tod_clock clk;
4540 unsigned long i;
4541
4542 preempt_disable();
4543
4544 store_tod_clock_ext(&clk);
4545
4546 kvm->arch.epoch = gtod->tod - clk.tod;
4547 kvm->arch.epdx = 0;
4548 if (test_kvm_facility(kvm, 139)) {
4549 kvm->arch.epdx = gtod->epoch_idx - clk.ei;
4550 if (kvm->arch.epoch > gtod->tod)
4551 kvm->arch.epdx -= 1;
4552 }
4553
4554 kvm_s390_vcpu_block_all(kvm);
4555 kvm_for_each_vcpu(i, vcpu, kvm) {
4556 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
4557 vcpu->arch.sie_block->epdx = kvm->arch.epdx;
4558 }
4559
4560 kvm_s390_vcpu_unblock_all(kvm);
4561 preempt_enable();
4562 }
4563
4564 int kvm_s390_try_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod)
4565 {
4566 if (!mutex_trylock(&kvm->lock))
4567 return 0;
4568 __kvm_s390_set_tod_clock(kvm, gtod);
4569 mutex_unlock(&kvm->lock);
4570 return 1;
4571 }
4572
4573 /**
4574 * kvm_arch_fault_in_page - fault-in guest page if necessary
4575 * @vcpu: The corresponding virtual cpu
4576 * @gpa: Guest physical address
4577 * @writable: Whether the page should be writable or not
4578 *
4579 * Make sure that a guest page has been faulted-in on the host.
4580 *
4581 * Return: Zero on success, negative error code otherwise.
4582 */
4583 long kvm_arch_fault_in_page(struct kvm_vcpu *vcpu, gpa_t gpa, int writable)
4584 {
4585 return gmap_fault(vcpu->arch.gmap, gpa,
4586 writable ? FAULT_FLAG_WRITE : 0);
4587 }
4588
4589 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
4590 unsigned long token)
4591 {
4592 struct kvm_s390_interrupt inti;
4593 struct kvm_s390_irq irq;
4594
4595 if (start_token) {
4596 irq.u.ext.ext_params2 = token;
4597 irq.type = KVM_S390_INT_PFAULT_INIT;
4598 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
4599 } else {
4600 inti.type = KVM_S390_INT_PFAULT_DONE;
4601 inti.parm64 = token;
4602 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
4603 }
4604 }
4605
4606 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
4607 struct kvm_async_pf *work)
4608 {
4609 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
4610 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
4611
4612 return true;
4613 }
4614
4615 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
4616 struct kvm_async_pf *work)
4617 {
4618 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
4619 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
4620 }
4621
4622 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
4623 struct kvm_async_pf *work)
4624 {
4625 /* s390 will always inject the page directly */
4626 }
4627
4628 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
4629 {
4630 /*
4631 * s390 will always inject the page directly,
4632 * but we still want check_async_completion to cleanup
4633 */
4634 return true;
4635 }
4636
4637 static bool kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
4638 {
4639 hva_t hva;
4640 struct kvm_arch_async_pf arch;
4641
4642 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4643 return false;
4644 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
4645 vcpu->arch.pfault_compare)
4646 return false;
4647 if (psw_extint_disabled(vcpu))
4648 return false;
4649 if (kvm_s390_vcpu_has_irq(vcpu, 0))
4650 return false;
4651 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK))
4652 return false;
4653 if (!vcpu->arch.gmap->pfault_enabled)
4654 return false;
4655
4656 hva = gfn_to_hva(vcpu->kvm, gpa_to_gfn(current->thread.gmap_addr));
4657 hva += current->thread.gmap_addr & ~PAGE_MASK;
4658 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
4659 return false;
4660
4661 return kvm_setup_async_pf(vcpu, current->thread.gmap_addr, hva, &arch);
4662 }
4663
4664 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
4665 {
4666 int rc, cpuflags;
4667
4668 /*
4669 * On s390 notifications for arriving pages will be delivered directly
4670 * to the guest but the house keeping for completed pfaults is
4671 * handled outside the worker.
4672 */
4673 kvm_check_async_pf_completion(vcpu);
4674
4675 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
4676 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
4677
4678 if (need_resched())
4679 schedule();
4680
4681 if (!kvm_is_ucontrol(vcpu->kvm)) {
4682 rc = kvm_s390_deliver_pending_interrupts(vcpu);
4683 if (rc || guestdbg_exit_pending(vcpu))
4684 return rc;
4685 }
4686
4687 rc = kvm_s390_handle_requests(vcpu);
4688 if (rc)
4689 return rc;
4690
4691 if (guestdbg_enabled(vcpu)) {
4692 kvm_s390_backup_guest_per_regs(vcpu);
4693 kvm_s390_patch_guest_per_regs(vcpu);
4694 }
4695
4696 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask);
4697
4698 vcpu->arch.sie_block->icptcode = 0;
4699 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
4700 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
4701 trace_kvm_s390_sie_enter(vcpu, cpuflags);
4702
4703 return 0;
4704 }
4705
4706 static int vcpu_post_run_fault_in_sie(struct kvm_vcpu *vcpu)
4707 {
4708 struct kvm_s390_pgm_info pgm_info = {
4709 .code = PGM_ADDRESSING,
4710 };
4711 u8 opcode, ilen;
4712 int rc;
4713
4714 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
4715 trace_kvm_s390_sie_fault(vcpu);
4716
4717 /*
4718 * We want to inject an addressing exception, which is defined as a
4719 * suppressing or terminating exception. However, since we came here
4720 * by a DAT access exception, the PSW still points to the faulting
4721 * instruction since DAT exceptions are nullifying. So we've got
4722 * to look up the current opcode to get the length of the instruction
4723 * to be able to forward the PSW.
4724 */
4725 rc = read_guest_instr(vcpu, vcpu->arch.sie_block->gpsw.addr, &opcode, 1);
4726 ilen = insn_length(opcode);
4727 if (rc < 0) {
4728 return rc;
4729 } else if (rc) {
4730 /* Instruction-Fetching Exceptions - we can't detect the ilen.
4731 * Forward by arbitrary ilc, injection will take care of
4732 * nullification if necessary.
4733 */
4734 pgm_info = vcpu->arch.pgm;
4735 ilen = 4;
4736 }
4737 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
4738 kvm_s390_forward_psw(vcpu, ilen);
4739 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
4740 }
4741
4742 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
4743 {
4744 struct mcck_volatile_info *mcck_info;
4745 struct sie_page *sie_page;
4746
4747 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
4748 vcpu->arch.sie_block->icptcode);
4749 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
4750
4751 if (guestdbg_enabled(vcpu))
4752 kvm_s390_restore_guest_per_regs(vcpu);
4753
4754 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
4755 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
4756
4757 if (exit_reason == -EINTR) {
4758 VCPU_EVENT(vcpu, 3, "%s", "machine check");
4759 sie_page = container_of(vcpu->arch.sie_block,
4760 struct sie_page, sie_block);
4761 mcck_info = &sie_page->mcck_info;
4762 kvm_s390_reinject_machine_check(vcpu, mcck_info);
4763 return 0;
4764 }
4765
4766 if (vcpu->arch.sie_block->icptcode > 0) {
4767 int rc = kvm_handle_sie_intercept(vcpu);
4768
4769 if (rc != -EOPNOTSUPP)
4770 return rc;
4771 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
4772 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
4773 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
4774 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
4775 return -EREMOTE;
4776 } else if (exit_reason != -EFAULT) {
4777 vcpu->stat.exit_null++;
4778 return 0;
4779 } else if (kvm_is_ucontrol(vcpu->kvm)) {
4780 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
4781 vcpu->run->s390_ucontrol.trans_exc_code =
4782 current->thread.gmap_addr;
4783 vcpu->run->s390_ucontrol.pgm_code = 0x10;
4784 return -EREMOTE;
4785 } else if (current->thread.gmap_pfault) {
4786 trace_kvm_s390_major_guest_pfault(vcpu);
4787 current->thread.gmap_pfault = 0;
4788 if (kvm_arch_setup_async_pf(vcpu))
4789 return 0;
4790 vcpu->stat.pfault_sync++;
4791 return kvm_arch_fault_in_page(vcpu, current->thread.gmap_addr, 1);
4792 }
4793 return vcpu_post_run_fault_in_sie(vcpu);
4794 }
4795
4796 #define PSW_INT_MASK (PSW_MASK_EXT | PSW_MASK_IO | PSW_MASK_MCHECK)
4797 static int __vcpu_run(struct kvm_vcpu *vcpu)
4798 {
4799 int rc, exit_reason;
4800 struct sie_page *sie_page = (struct sie_page *)vcpu->arch.sie_block;
4801
4802 /*
4803 * We try to hold kvm->srcu during most of vcpu_run (except when run-
4804 * ning the guest), so that memslots (and other stuff) are protected
4805 */
4806 kvm_vcpu_srcu_read_lock(vcpu);
4807
4808 do {
4809 rc = vcpu_pre_run(vcpu);
4810 if (rc || guestdbg_exit_pending(vcpu))
4811 break;
4812
4813 kvm_vcpu_srcu_read_unlock(vcpu);
4814 /*
4815 * As PF_VCPU will be used in fault handler, between
4816 * guest_enter and guest_exit should be no uaccess.
4817 */
4818 local_irq_disable();
4819 guest_enter_irqoff();
4820 __disable_cpu_timer_accounting(vcpu);
4821 local_irq_enable();
4822 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
4823 memcpy(sie_page->pv_grregs,
4824 vcpu->run->s.regs.gprs,
4825 sizeof(sie_page->pv_grregs));
4826 }
4827 exit_reason = sie64a(vcpu->arch.sie_block,
4828 vcpu->run->s.regs.gprs,
4829 gmap_get_enabled()->asce);
4830 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
4831 memcpy(vcpu->run->s.regs.gprs,
4832 sie_page->pv_grregs,
4833 sizeof(sie_page->pv_grregs));
4834 /*
4835 * We're not allowed to inject interrupts on intercepts
4836 * that leave the guest state in an "in-between" state
4837 * where the next SIE entry will do a continuation.
4838 * Fence interrupts in our "internal" PSW.
4839 */
4840 if (vcpu->arch.sie_block->icptcode == ICPT_PV_INSTR ||
4841 vcpu->arch.sie_block->icptcode == ICPT_PV_PREF) {
4842 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK;
4843 }
4844 }
4845 local_irq_disable();
4846 __enable_cpu_timer_accounting(vcpu);
4847 guest_exit_irqoff();
4848 local_irq_enable();
4849 kvm_vcpu_srcu_read_lock(vcpu);
4850
4851 rc = vcpu_post_run(vcpu, exit_reason);
4852 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
4853
4854 kvm_vcpu_srcu_read_unlock(vcpu);
4855 return rc;
4856 }
4857
4858 static void sync_regs_fmt2(struct kvm_vcpu *vcpu)
4859 {
4860 struct kvm_run *kvm_run = vcpu->run;
4861 struct runtime_instr_cb *riccb;
4862 struct gs_cb *gscb;
4863
4864 riccb = (struct runtime_instr_cb *) &kvm_run->s.regs.riccb;
4865 gscb = (struct gs_cb *) &kvm_run->s.regs.gscb;
4866 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
4867 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
4868 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
4869 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
4870 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
4871 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
4872 }
4873 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
4874 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
4875 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
4876 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
4877 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4878 kvm_clear_async_pf_completion_queue(vcpu);
4879 }
4880 if (kvm_run->kvm_dirty_regs & KVM_SYNC_DIAG318) {
4881 vcpu->arch.diag318_info.val = kvm_run->s.regs.diag318;
4882 vcpu->arch.sie_block->cpnc = vcpu->arch.diag318_info.cpnc;
4883 VCPU_EVENT(vcpu, 3, "setting cpnc to %d", vcpu->arch.diag318_info.cpnc);
4884 }
4885 /*
4886 * If userspace sets the riccb (e.g. after migration) to a valid state,
4887 * we should enable RI here instead of doing the lazy enablement.
4888 */
4889 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) &&
4890 test_kvm_facility(vcpu->kvm, 64) &&
4891 riccb->v &&
4892 !(vcpu->arch.sie_block->ecb3 & ECB3_RI)) {
4893 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (sync_regs)");
4894 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
4895 }
4896 /*
4897 * If userspace sets the gscb (e.g. after migration) to non-zero,
4898 * we should enable GS here instead of doing the lazy enablement.
4899 */
4900 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_GSCB) &&
4901 test_kvm_facility(vcpu->kvm, 133) &&
4902 gscb->gssm &&
4903 !vcpu->arch.gs_enabled) {
4904 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (sync_regs)");
4905 vcpu->arch.sie_block->ecb |= ECB_GS;
4906 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
4907 vcpu->arch.gs_enabled = 1;
4908 }
4909 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_BPBC) &&
4910 test_kvm_facility(vcpu->kvm, 82)) {
4911 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
4912 vcpu->arch.sie_block->fpf |= kvm_run->s.regs.bpbc ? FPF_BPBC : 0;
4913 }
4914 if (MACHINE_HAS_GS) {
4915 preempt_disable();
4916 local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT);
4917 if (current->thread.gs_cb) {
4918 vcpu->arch.host_gscb = current->thread.gs_cb;
4919 save_gs_cb(vcpu->arch.host_gscb);
4920 }
4921 if (vcpu->arch.gs_enabled) {
4922 current->thread.gs_cb = (struct gs_cb *)
4923 &vcpu->run->s.regs.gscb;
4924 restore_gs_cb(current->thread.gs_cb);
4925 }
4926 preempt_enable();
4927 }
4928 /* SIE will load etoken directly from SDNX and therefore kvm_run */
4929 }
4930
4931 static void sync_regs(struct kvm_vcpu *vcpu)
4932 {
4933 struct kvm_run *kvm_run = vcpu->run;
4934
4935 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
4936 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
4937 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
4938 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
4939 /* some control register changes require a tlb flush */
4940 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
4941 }
4942 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
4943 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
4944 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
4945 }
4946 save_access_regs(vcpu->arch.host_acrs);
4947 restore_access_regs(vcpu->run->s.regs.acrs);
4948 vcpu->arch.acrs_loaded = true;
4949 kvm_s390_fpu_load(vcpu->run);
4950 /* Sync fmt2 only data */
4951 if (likely(!kvm_s390_pv_cpu_is_protected(vcpu))) {
4952 sync_regs_fmt2(vcpu);
4953 } else {
4954 /*
4955 * In several places we have to modify our internal view to
4956 * not do things that are disallowed by the ultravisor. For
4957 * example we must not inject interrupts after specific exits
4958 * (e.g. 112 prefix page not secure). We do this by turning
4959 * off the machine check, external and I/O interrupt bits
4960 * of our PSW copy. To avoid getting validity intercepts, we
4961 * do only accept the condition code from userspace.
4962 */
4963 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_CC;
4964 vcpu->arch.sie_block->gpsw.mask |= kvm_run->psw_mask &
4965 PSW_MASK_CC;
4966 }
4967
4968 kvm_run->kvm_dirty_regs = 0;
4969 }
4970
4971 static void store_regs_fmt2(struct kvm_vcpu *vcpu)
4972 {
4973 struct kvm_run *kvm_run = vcpu->run;
4974
4975 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
4976 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
4977 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
4978 kvm_run->s.regs.bpbc = (vcpu->arch.sie_block->fpf & FPF_BPBC) == FPF_BPBC;
4979 kvm_run->s.regs.diag318 = vcpu->arch.diag318_info.val;
4980 if (MACHINE_HAS_GS) {
4981 preempt_disable();
4982 local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT);
4983 if (vcpu->arch.gs_enabled)
4984 save_gs_cb(current->thread.gs_cb);
4985 current->thread.gs_cb = vcpu->arch.host_gscb;
4986 restore_gs_cb(vcpu->arch.host_gscb);
4987 if (!vcpu->arch.host_gscb)
4988 local_ctl_clear_bit(2, CR2_GUARDED_STORAGE_BIT);
4989 vcpu->arch.host_gscb = NULL;
4990 preempt_enable();
4991 }
4992 /* SIE will save etoken directly into SDNX and therefore kvm_run */
4993 }
4994
4995 static void store_regs(struct kvm_vcpu *vcpu)
4996 {
4997 struct kvm_run *kvm_run = vcpu->run;
4998
4999 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
5000 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
5001 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
5002 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
5003 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
5004 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
5005 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
5006 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
5007 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
5008 save_access_regs(vcpu->run->s.regs.acrs);
5009 restore_access_regs(vcpu->arch.host_acrs);
5010 vcpu->arch.acrs_loaded = false;
5011 kvm_s390_fpu_store(vcpu->run);
5012 if (likely(!kvm_s390_pv_cpu_is_protected(vcpu)))
5013 store_regs_fmt2(vcpu);
5014 }
5015
5016 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
5017 {
5018 struct kvm_run *kvm_run = vcpu->run;
5019 DECLARE_KERNEL_FPU_ONSTACK32(fpu);
5020 int rc;
5021
5022 /*
5023 * Running a VM while dumping always has the potential to
5024 * produce inconsistent dump data. But for PV vcpus a SIE
5025 * entry while dumping could also lead to a fatal validity
5026 * intercept which we absolutely want to avoid.
5027 */
5028 if (vcpu->kvm->arch.pv.dumping)
5029 return -EINVAL;
5030
5031 if (!vcpu->wants_to_run)
5032 return -EINTR;
5033
5034 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_S390_VALID_FIELDS ||
5035 kvm_run->kvm_dirty_regs & ~KVM_SYNC_S390_VALID_FIELDS)
5036 return -EINVAL;
5037
5038 vcpu_load(vcpu);
5039
5040 if (guestdbg_exit_pending(vcpu)) {
5041 kvm_s390_prepare_debug_exit(vcpu);
5042 rc = 0;
5043 goto out;
5044 }
5045
5046 kvm_sigset_activate(vcpu);
5047
5048 /*
5049 * no need to check the return value of vcpu_start as it can only have
5050 * an error for protvirt, but protvirt means user cpu state
5051 */
5052 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
5053 kvm_s390_vcpu_start(vcpu);
5054 } else if (is_vcpu_stopped(vcpu)) {
5055 pr_err_ratelimited("can't run stopped vcpu %d\n",
5056 vcpu->vcpu_id);
5057 rc = -EINVAL;
5058 goto out;
5059 }
5060
5061 kernel_fpu_begin(&fpu, KERNEL_FPC | KERNEL_VXR);
5062 sync_regs(vcpu);
5063 enable_cpu_timer_accounting(vcpu);
5064
5065 might_fault();
5066 rc = __vcpu_run(vcpu);
5067
5068 if (signal_pending(current) && !rc) {
5069 kvm_run->exit_reason = KVM_EXIT_INTR;
5070 rc = -EINTR;
5071 }
5072
5073 if (guestdbg_exit_pending(vcpu) && !rc) {
5074 kvm_s390_prepare_debug_exit(vcpu);
5075 rc = 0;
5076 }
5077
5078 if (rc == -EREMOTE) {
5079 /* userspace support is needed, kvm_run has been prepared */
5080 rc = 0;
5081 }
5082
5083 disable_cpu_timer_accounting(vcpu);
5084 store_regs(vcpu);
5085 kernel_fpu_end(&fpu, KERNEL_FPC | KERNEL_VXR);
5086
5087 kvm_sigset_deactivate(vcpu);
5088
5089 vcpu->stat.exit_userspace++;
5090 out:
5091 vcpu_put(vcpu);
5092 return rc;
5093 }
5094
5095 /*
5096 * store status at address
5097 * we use have two special cases:
5098 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
5099 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
5100 */
5101 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
5102 {
5103 unsigned char archmode = 1;
5104 freg_t fprs[NUM_FPRS];
5105 unsigned int px;
5106 u64 clkcomp, cputm;
5107 int rc;
5108
5109 px = kvm_s390_get_prefix(vcpu);
5110 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
5111 if (write_guest_abs(vcpu, 163, &archmode, 1))
5112 return -EFAULT;
5113 gpa = 0;
5114 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
5115 if (write_guest_real(vcpu, 163, &archmode, 1))
5116 return -EFAULT;
5117 gpa = px;
5118 } else
5119 gpa -= __LC_FPREGS_SAVE_AREA;
5120
5121 /* manually convert vector registers if necessary */
5122 if (cpu_has_vx()) {
5123 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
5124 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
5125 fprs, 128);
5126 } else {
5127 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
5128 vcpu->run->s.regs.fprs, 128);
5129 }
5130 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
5131 vcpu->run->s.regs.gprs, 128);
5132 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
5133 &vcpu->arch.sie_block->gpsw, 16);
5134 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
5135 &px, 4);
5136 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
5137 &vcpu->run->s.regs.fpc, 4);
5138 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
5139 &vcpu->arch.sie_block->todpr, 4);
5140 cputm = kvm_s390_get_cpu_timer(vcpu);
5141 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
5142 &cputm, 8);
5143 clkcomp = vcpu->arch.sie_block->ckc >> 8;
5144 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
5145 &clkcomp, 8);
5146 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
5147 &vcpu->run->s.regs.acrs, 64);
5148 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
5149 &vcpu->arch.sie_block->gcr, 128);
5150 return rc ? -EFAULT : 0;
5151 }
5152
5153 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
5154 {
5155 /*
5156 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
5157 * switch in the run ioctl. Let's update our copies before we save
5158 * it into the save area
5159 */
5160 kvm_s390_fpu_store(vcpu->run);
5161 save_access_regs(vcpu->run->s.regs.acrs);
5162
5163 return kvm_s390_store_status_unloaded(vcpu, addr);
5164 }
5165
5166 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
5167 {
5168 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
5169 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
5170 }
5171
5172 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
5173 {
5174 unsigned long i;
5175 struct kvm_vcpu *vcpu;
5176
5177 kvm_for_each_vcpu(i, vcpu, kvm) {
5178 __disable_ibs_on_vcpu(vcpu);
5179 }
5180 }
5181
5182 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
5183 {
5184 if (!sclp.has_ibs)
5185 return;
5186 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
5187 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
5188 }
5189
5190 int kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
5191 {
5192 int i, online_vcpus, r = 0, started_vcpus = 0;
5193
5194 if (!is_vcpu_stopped(vcpu))
5195 return 0;
5196
5197 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
5198 /* Only one cpu at a time may enter/leave the STOPPED state. */
5199 spin_lock(&vcpu->kvm->arch.start_stop_lock);
5200 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
5201
5202 /* Let's tell the UV that we want to change into the operating state */
5203 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5204 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR);
5205 if (r) {
5206 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5207 return r;
5208 }
5209 }
5210
5211 for (i = 0; i < online_vcpus; i++) {
5212 if (!is_vcpu_stopped(kvm_get_vcpu(vcpu->kvm, i)))
5213 started_vcpus++;
5214 }
5215
5216 if (started_vcpus == 0) {
5217 /* we're the only active VCPU -> speed it up */
5218 __enable_ibs_on_vcpu(vcpu);
5219 } else if (started_vcpus == 1) {
5220 /*
5221 * As we are starting a second VCPU, we have to disable
5222 * the IBS facility on all VCPUs to remove potentially
5223 * outstanding ENABLE requests.
5224 */
5225 __disable_ibs_on_all_vcpus(vcpu->kvm);
5226 }
5227
5228 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_STOPPED);
5229 /*
5230 * The real PSW might have changed due to a RESTART interpreted by the
5231 * ultravisor. We block all interrupts and let the next sie exit
5232 * refresh our view.
5233 */
5234 if (kvm_s390_pv_cpu_is_protected(vcpu))
5235 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK;
5236 /*
5237 * Another VCPU might have used IBS while we were offline.
5238 * Let's play safe and flush the VCPU at startup.
5239 */
5240 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5241 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5242 return 0;
5243 }
5244
5245 int kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
5246 {
5247 int i, online_vcpus, r = 0, started_vcpus = 0;
5248 struct kvm_vcpu *started_vcpu = NULL;
5249
5250 if (is_vcpu_stopped(vcpu))
5251 return 0;
5252
5253 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
5254 /* Only one cpu at a time may enter/leave the STOPPED state. */
5255 spin_lock(&vcpu->kvm->arch.start_stop_lock);
5256 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
5257
5258 /* Let's tell the UV that we want to change into the stopped state */
5259 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5260 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_STP);
5261 if (r) {
5262 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5263 return r;
5264 }
5265 }
5266
5267 /*
5268 * Set the VCPU to STOPPED and THEN clear the interrupt flag,
5269 * now that the SIGP STOP and SIGP STOP AND STORE STATUS orders
5270 * have been fully processed. This will ensure that the VCPU
5271 * is kept BUSY if another VCPU is inquiring with SIGP SENSE.
5272 */
5273 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOPPED);
5274 kvm_s390_clear_stop_irq(vcpu);
5275
5276 __disable_ibs_on_vcpu(vcpu);
5277
5278 for (i = 0; i < online_vcpus; i++) {
5279 struct kvm_vcpu *tmp = kvm_get_vcpu(vcpu->kvm, i);
5280
5281 if (!is_vcpu_stopped(tmp)) {
5282 started_vcpus++;
5283 started_vcpu = tmp;
5284 }
5285 }
5286
5287 if (started_vcpus == 1) {
5288 /*
5289 * As we only have one VCPU left, we want to enable the
5290 * IBS facility for that VCPU to speed it up.
5291 */
5292 __enable_ibs_on_vcpu(started_vcpu);
5293 }
5294
5295 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5296 return 0;
5297 }
5298
5299 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5300 struct kvm_enable_cap *cap)
5301 {
5302 int r;
5303
5304 if (cap->flags)
5305 return -EINVAL;
5306
5307 switch (cap->cap) {
5308 case KVM_CAP_S390_CSS_SUPPORT:
5309 if (!vcpu->kvm->arch.css_support) {
5310 vcpu->kvm->arch.css_support = 1;
5311 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
5312 trace_kvm_s390_enable_css(vcpu->kvm);
5313 }
5314 r = 0;
5315 break;
5316 default:
5317 r = -EINVAL;
5318 break;
5319 }
5320 return r;
5321 }
5322
5323 static long kvm_s390_vcpu_sida_op(struct kvm_vcpu *vcpu,
5324 struct kvm_s390_mem_op *mop)
5325 {
5326 void __user *uaddr = (void __user *)mop->buf;
5327 void *sida_addr;
5328 int r = 0;
5329
5330 if (mop->flags || !mop->size)
5331 return -EINVAL;
5332 if (mop->size + mop->sida_offset < mop->size)
5333 return -EINVAL;
5334 if (mop->size + mop->sida_offset > sida_size(vcpu->arch.sie_block))
5335 return -E2BIG;
5336 if (!kvm_s390_pv_cpu_is_protected(vcpu))
5337 return -EINVAL;
5338
5339 sida_addr = (char *)sida_addr(vcpu->arch.sie_block) + mop->sida_offset;
5340
5341 switch (mop->op) {
5342 case KVM_S390_MEMOP_SIDA_READ:
5343 if (copy_to_user(uaddr, sida_addr, mop->size))
5344 r = -EFAULT;
5345
5346 break;
5347 case KVM_S390_MEMOP_SIDA_WRITE:
5348 if (copy_from_user(sida_addr, uaddr, mop->size))
5349 r = -EFAULT;
5350 break;
5351 }
5352 return r;
5353 }
5354
5355 static long kvm_s390_vcpu_mem_op(struct kvm_vcpu *vcpu,
5356 struct kvm_s390_mem_op *mop)
5357 {
5358 void __user *uaddr = (void __user *)mop->buf;
5359 enum gacc_mode acc_mode;
5360 void *tmpbuf = NULL;
5361 int r;
5362
5363 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_INJECT_EXCEPTION |
5364 KVM_S390_MEMOP_F_CHECK_ONLY |
5365 KVM_S390_MEMOP_F_SKEY_PROTECTION);
5366 if (r)
5367 return r;
5368 if (mop->ar >= NUM_ACRS)
5369 return -EINVAL;
5370 if (kvm_s390_pv_cpu_is_protected(vcpu))
5371 return -EINVAL;
5372 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
5373 tmpbuf = vmalloc(mop->size);
5374 if (!tmpbuf)
5375 return -ENOMEM;
5376 }
5377
5378 acc_mode = mop->op == KVM_S390_MEMOP_LOGICAL_READ ? GACC_FETCH : GACC_STORE;
5379 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
5380 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size,
5381 acc_mode, mop->key);
5382 goto out_inject;
5383 }
5384 if (acc_mode == GACC_FETCH) {
5385 r = read_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf,
5386 mop->size, mop->key);
5387 if (r)
5388 goto out_inject;
5389 if (copy_to_user(uaddr, tmpbuf, mop->size)) {
5390 r = -EFAULT;
5391 goto out_free;
5392 }
5393 } else {
5394 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
5395 r = -EFAULT;
5396 goto out_free;
5397 }
5398 r = write_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf,
5399 mop->size, mop->key);
5400 }
5401
5402 out_inject:
5403 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
5404 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
5405
5406 out_free:
5407 vfree(tmpbuf);
5408 return r;
5409 }
5410
5411 static long kvm_s390_vcpu_memsida_op(struct kvm_vcpu *vcpu,
5412 struct kvm_s390_mem_op *mop)
5413 {
5414 int r, srcu_idx;
5415
5416 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5417
5418 switch (mop->op) {
5419 case KVM_S390_MEMOP_LOGICAL_READ:
5420 case KVM_S390_MEMOP_LOGICAL_WRITE:
5421 r = kvm_s390_vcpu_mem_op(vcpu, mop);
5422 break;
5423 case KVM_S390_MEMOP_SIDA_READ:
5424 case KVM_S390_MEMOP_SIDA_WRITE:
5425 /* we are locked against sida going away by the vcpu->mutex */
5426 r = kvm_s390_vcpu_sida_op(vcpu, mop);
5427 break;
5428 default:
5429 r = -EINVAL;
5430 }
5431
5432 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
5433 return r;
5434 }
5435
5436 long kvm_arch_vcpu_async_ioctl(struct file *filp,
5437 unsigned int ioctl, unsigned long arg)
5438 {
5439 struct kvm_vcpu *vcpu = filp->private_data;
5440 void __user *argp = (void __user *)arg;
5441 int rc;
5442
5443 switch (ioctl) {
5444 case KVM_S390_IRQ: {
5445 struct kvm_s390_irq s390irq;
5446
5447 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
5448 return -EFAULT;
5449 rc = kvm_s390_inject_vcpu(vcpu, &s390irq);
5450 break;
5451 }
5452 case KVM_S390_INTERRUPT: {
5453 struct kvm_s390_interrupt s390int;
5454 struct kvm_s390_irq s390irq = {};
5455
5456 if (copy_from_user(&s390int, argp, sizeof(s390int)))
5457 return -EFAULT;
5458 if (s390int_to_s390irq(&s390int, &s390irq))
5459 return -EINVAL;
5460 rc = kvm_s390_inject_vcpu(vcpu, &s390irq);
5461 break;
5462 }
5463 default:
5464 rc = -ENOIOCTLCMD;
5465 break;
5466 }
5467
5468 /*
5469 * To simplify single stepping of userspace-emulated instructions,
5470 * KVM_EXIT_S390_SIEIC exit sets KVM_GUESTDBG_EXIT_PENDING (see
5471 * should_handle_per_ifetch()). However, if userspace emulation injects
5472 * an interrupt, it needs to be cleared, so that KVM_EXIT_DEBUG happens
5473 * after (and not before) the interrupt delivery.
5474 */
5475 if (!rc)
5476 vcpu->guest_debug &= ~KVM_GUESTDBG_EXIT_PENDING;
5477
5478 return rc;
5479 }
5480
5481 static int kvm_s390_handle_pv_vcpu_dump(struct kvm_vcpu *vcpu,
5482 struct kvm_pv_cmd *cmd)
5483 {
5484 struct kvm_s390_pv_dmp dmp;
5485 void *data;
5486 int ret;
5487
5488 /* Dump initialization is a prerequisite */
5489 if (!vcpu->kvm->arch.pv.dumping)
5490 return -EINVAL;
5491
5492 if (copy_from_user(&dmp, (__u8 __user *)cmd->data, sizeof(dmp)))
5493 return -EFAULT;
5494
5495 /* We only handle this subcmd right now */
5496 if (dmp.subcmd != KVM_PV_DUMP_CPU)
5497 return -EINVAL;
5498
5499 /* CPU dump length is the same as create cpu storage donation. */
5500 if (dmp.buff_len != uv_info.guest_cpu_stor_len)
5501 return -EINVAL;
5502
5503 data = kvzalloc(uv_info.guest_cpu_stor_len, GFP_KERNEL);
5504 if (!data)
5505 return -ENOMEM;
5506
5507 ret = kvm_s390_pv_dump_cpu(vcpu, data, &cmd->rc, &cmd->rrc);
5508
5509 VCPU_EVENT(vcpu, 3, "PROTVIRT DUMP CPU %d rc %x rrc %x",
5510 vcpu->vcpu_id, cmd->rc, cmd->rrc);
5511
5512 if (ret)
5513 ret = -EINVAL;
5514
5515 /* On success copy over the dump data */
5516 if (!ret && copy_to_user((__u8 __user *)dmp.buff_addr, data, uv_info.guest_cpu_stor_len))
5517 ret = -EFAULT;
5518
5519 kvfree(data);
5520 return ret;
5521 }
5522
5523 long kvm_arch_vcpu_ioctl(struct file *filp,
5524 unsigned int ioctl, unsigned long arg)
5525 {
5526 struct kvm_vcpu *vcpu = filp->private_data;
5527 void __user *argp = (void __user *)arg;
5528 int idx;
5529 long r;
5530 u16 rc, rrc;
5531
5532 vcpu_load(vcpu);
5533
5534 switch (ioctl) {
5535 case KVM_S390_STORE_STATUS:
5536 idx = srcu_read_lock(&vcpu->kvm->srcu);
5537 r = kvm_s390_store_status_unloaded(vcpu, arg);
5538 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5539 break;
5540 case KVM_S390_SET_INITIAL_PSW: {
5541 psw_t psw;
5542
5543 r = -EFAULT;
5544 if (copy_from_user(&psw, argp, sizeof(psw)))
5545 break;
5546 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
5547 break;
5548 }
5549 case KVM_S390_CLEAR_RESET:
5550 r = 0;
5551 kvm_arch_vcpu_ioctl_clear_reset(vcpu);
5552 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5553 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5554 UVC_CMD_CPU_RESET_CLEAR, &rc, &rrc);
5555 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET CLEAR VCPU: rc %x rrc %x",
5556 rc, rrc);
5557 }
5558 break;
5559 case KVM_S390_INITIAL_RESET:
5560 r = 0;
5561 kvm_arch_vcpu_ioctl_initial_reset(vcpu);
5562 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5563 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5564 UVC_CMD_CPU_RESET_INITIAL,
5565 &rc, &rrc);
5566 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET INITIAL VCPU: rc %x rrc %x",
5567 rc, rrc);
5568 }
5569 break;
5570 case KVM_S390_NORMAL_RESET:
5571 r = 0;
5572 kvm_arch_vcpu_ioctl_normal_reset(vcpu);
5573 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5574 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5575 UVC_CMD_CPU_RESET, &rc, &rrc);
5576 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET NORMAL VCPU: rc %x rrc %x",
5577 rc, rrc);
5578 }
5579 break;
5580 case KVM_SET_ONE_REG:
5581 case KVM_GET_ONE_REG: {
5582 struct kvm_one_reg reg;
5583 r = -EINVAL;
5584 if (kvm_s390_pv_cpu_is_protected(vcpu))
5585 break;
5586 r = -EFAULT;
5587 if (copy_from_user(®, argp, sizeof(reg)))
5588 break;
5589 if (ioctl == KVM_SET_ONE_REG)
5590 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®);
5591 else
5592 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®);
5593 break;
5594 }
5595 #ifdef CONFIG_KVM_S390_UCONTROL
5596 case KVM_S390_UCAS_MAP: {
5597 struct kvm_s390_ucas_mapping ucasmap;
5598
5599 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
5600 r = -EFAULT;
5601 break;
5602 }
5603
5604 if (!kvm_is_ucontrol(vcpu->kvm)) {
5605 r = -EINVAL;
5606 break;
5607 }
5608
5609 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
5610 ucasmap.vcpu_addr, ucasmap.length);
5611 break;
5612 }
5613 case KVM_S390_UCAS_UNMAP: {
5614 struct kvm_s390_ucas_mapping ucasmap;
5615
5616 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
5617 r = -EFAULT;
5618 break;
5619 }
5620
5621 if (!kvm_is_ucontrol(vcpu->kvm)) {
5622 r = -EINVAL;
5623 break;
5624 }
5625
5626 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
5627 ucasmap.length);
5628 break;
5629 }
5630 #endif
5631 case KVM_S390_VCPU_FAULT: {
5632 r = gmap_fault(vcpu->arch.gmap, arg, 0);
5633 break;
5634 }
5635 case KVM_ENABLE_CAP:
5636 {
5637 struct kvm_enable_cap cap;
5638 r = -EFAULT;
5639 if (copy_from_user(&cap, argp, sizeof(cap)))
5640 break;
5641 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5642 break;
5643 }
5644 case KVM_S390_MEM_OP: {
5645 struct kvm_s390_mem_op mem_op;
5646
5647 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
5648 r = kvm_s390_vcpu_memsida_op(vcpu, &mem_op);
5649 else
5650 r = -EFAULT;
5651 break;
5652 }
5653 case KVM_S390_SET_IRQ_STATE: {
5654 struct kvm_s390_irq_state irq_state;
5655
5656 r = -EFAULT;
5657 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
5658 break;
5659 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
5660 irq_state.len == 0 ||
5661 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
5662 r = -EINVAL;
5663 break;
5664 }
5665 /* do not use irq_state.flags, it will break old QEMUs */
5666 r = kvm_s390_set_irq_state(vcpu,
5667 (void __user *) irq_state.buf,
5668 irq_state.len);
5669 break;
5670 }
5671 case KVM_S390_GET_IRQ_STATE: {
5672 struct kvm_s390_irq_state irq_state;
5673
5674 r = -EFAULT;
5675 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
5676 break;
5677 if (irq_state.len == 0) {
5678 r = -EINVAL;
5679 break;
5680 }
5681 /* do not use irq_state.flags, it will break old QEMUs */
5682 r = kvm_s390_get_irq_state(vcpu,
5683 (__u8 __user *) irq_state.buf,
5684 irq_state.len);
5685 break;
5686 }
5687 case KVM_S390_PV_CPU_COMMAND: {
5688 struct kvm_pv_cmd cmd;
5689
5690 r = -EINVAL;
5691 if (!is_prot_virt_host())
5692 break;
5693
5694 r = -EFAULT;
5695 if (copy_from_user(&cmd, argp, sizeof(cmd)))
5696 break;
5697
5698 r = -EINVAL;
5699 if (cmd.flags)
5700 break;
5701
5702 /* We only handle this cmd right now */
5703 if (cmd.cmd != KVM_PV_DUMP)
5704 break;
5705
5706 r = kvm_s390_handle_pv_vcpu_dump(vcpu, &cmd);
5707
5708 /* Always copy over UV rc / rrc data */
5709 if (copy_to_user((__u8 __user *)argp, &cmd.rc,
5710 sizeof(cmd.rc) + sizeof(cmd.rrc)))
5711 r = -EFAULT;
5712 break;
5713 }
5714 default:
5715 r = -ENOTTY;
5716 }
5717
5718 vcpu_put(vcpu);
5719 return r;
5720 }
5721
5722 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5723 {
5724 #ifdef CONFIG_KVM_S390_UCONTROL
5725 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
5726 && (kvm_is_ucontrol(vcpu->kvm))) {
5727 vmf->page = virt_to_page(vcpu->arch.sie_block);
5728 get_page(vmf->page);
5729 return 0;
5730 }
5731 #endif
5732 return VM_FAULT_SIGBUS;
5733 }
5734
5735 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
5736 {
5737 return true;
5738 }
5739
5740 /* Section: memory related */
5741 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5742 const struct kvm_memory_slot *old,
5743 struct kvm_memory_slot *new,
5744 enum kvm_mr_change change)
5745 {
5746 gpa_t size;
5747
5748 if (kvm_is_ucontrol(kvm))
5749 return -EINVAL;
5750
5751 /* When we are protected, we should not change the memory slots */
5752 if (kvm_s390_pv_get_handle(kvm))
5753 return -EINVAL;
5754
5755 if (change != KVM_MR_DELETE && change != KVM_MR_FLAGS_ONLY) {
5756 /*
5757 * A few sanity checks. We can have memory slots which have to be
5758 * located/ended at a segment boundary (1MB). The memory in userland is
5759 * ok to be fragmented into various different vmas. It is okay to mmap()
5760 * and munmap() stuff in this slot after doing this call at any time
5761 */
5762
5763 if (new->userspace_addr & 0xffffful)
5764 return -EINVAL;
5765
5766 size = new->npages * PAGE_SIZE;
5767 if (size & 0xffffful)
5768 return -EINVAL;
5769
5770 if ((new->base_gfn * PAGE_SIZE) + size > kvm->arch.mem_limit)
5771 return -EINVAL;
5772 }
5773
5774 if (!kvm->arch.migration_mode)
5775 return 0;
5776
5777 /*
5778 * Turn off migration mode when:
5779 * - userspace creates a new memslot with dirty logging off,
5780 * - userspace modifies an existing memslot (MOVE or FLAGS_ONLY) and
5781 * dirty logging is turned off.
5782 * Migration mode expects dirty page logging being enabled to store
5783 * its dirty bitmap.
5784 */
5785 if (change != KVM_MR_DELETE &&
5786 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
5787 WARN(kvm_s390_vm_stop_migration(kvm),
5788 "Failed to stop migration mode");
5789
5790 return 0;
5791 }
5792
5793 void kvm_arch_commit_memory_region(struct kvm *kvm,
5794 struct kvm_memory_slot *old,
5795 const struct kvm_memory_slot *new,
5796 enum kvm_mr_change change)
5797 {
5798 int rc = 0;
5799
5800 switch (change) {
5801 case KVM_MR_DELETE:
5802 rc = gmap_unmap_segment(kvm->arch.gmap, old->base_gfn * PAGE_SIZE,
5803 old->npages * PAGE_SIZE);
5804 break;
5805 case KVM_MR_MOVE:
5806 rc = gmap_unmap_segment(kvm->arch.gmap, old->base_gfn * PAGE_SIZE,
5807 old->npages * PAGE_SIZE);
5808 if (rc)
5809 break;
5810 fallthrough;
5811 case KVM_MR_CREATE:
5812 rc = gmap_map_segment(kvm->arch.gmap, new->userspace_addr,
5813 new->base_gfn * PAGE_SIZE,
5814 new->npages * PAGE_SIZE);
5815 break;
5816 case KVM_MR_FLAGS_ONLY:
5817 break;
5818 default:
5819 WARN(1, "Unknown KVM MR CHANGE: %d\n", change);
5820 }
5821 if (rc)
5822 pr_warn("failed to commit memory region\n");
5823 return;
5824 }
5825
5826 static inline unsigned long nonhyp_mask(int i)
5827 {
5828 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
5829
5830 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
5831 }
5832
5833 static int __init kvm_s390_init(void)
5834 {
5835 int i, r;
5836
5837 if (!sclp.has_sief2) {
5838 pr_info("SIE is not available\n");
5839 return -ENODEV;
5840 }
5841
5842 if (nested && hpage) {
5843 pr_info("A KVM host that supports nesting cannot back its KVM guests with huge pages\n");
5844 return -EINVAL;
5845 }
5846
5847 for (i = 0; i < 16; i++)
5848 kvm_s390_fac_base[i] |=
5849 stfle_fac_list[i] & nonhyp_mask(i);
5850
5851 r = __kvm_s390_init();
5852 if (r)
5853 return r;
5854
5855 r = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
5856 if (r) {
5857 __kvm_s390_exit();
5858 return r;
5859 }
5860 return 0;
5861 }
5862
5863 static void __exit kvm_s390_exit(void)
5864 {
5865 kvm_exit();
5866
5867 __kvm_s390_exit();
5868 }
5869
5870 module_init(kvm_s390_init);
5871 module_exit(kvm_s390_exit);
5872
5873 /*
5874 * Enable autoloading of the kvm module.
5875 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
5876 * since x86 takes a different approach.
5877 */
5878 #include <linux/miscdevice.h>
5879 MODULE_ALIAS_MISCDEV(KVM_MINOR);
5880 MODULE_ALIAS("devname:kvm");
5881
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.