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/asm.h>
47 #include <asm/fpu.h>
48 #include <asm/ap.h>
49 #include <asm/uv.h>
50 #include "kvm-s390.h"
51 #include "gaccess.h"
52 #include "pci.h"
53 #include "gmap.h"
54
55 #define CREATE_TRACE_POINTS
56 #include "trace.h"
57 #include "trace-s390.h"
58
59 #define MEM_OP_MAX_SIZE 65536 /* Maximum transfer size for KVM_S390_MEM_OP */
60 #define LOCAL_IRQS 32
61 #define VCPU_IRQS_MAX_BUF (sizeof(struct kvm_s390_irq) * \
62 (KVM_MAX_VCPUS + LOCAL_IRQS))
63
64 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
65 KVM_GENERIC_VM_STATS(),
66 STATS_DESC_COUNTER(VM, inject_io),
67 STATS_DESC_COUNTER(VM, inject_float_mchk),
68 STATS_DESC_COUNTER(VM, inject_pfault_done),
69 STATS_DESC_COUNTER(VM, inject_service_signal),
70 STATS_DESC_COUNTER(VM, inject_virtio),
71 STATS_DESC_COUNTER(VM, aen_forward),
72 STATS_DESC_COUNTER(VM, gmap_shadow_reuse),
73 STATS_DESC_COUNTER(VM, gmap_shadow_create),
74 STATS_DESC_COUNTER(VM, gmap_shadow_r1_entry),
75 STATS_DESC_COUNTER(VM, gmap_shadow_r2_entry),
76 STATS_DESC_COUNTER(VM, gmap_shadow_r3_entry),
77 STATS_DESC_COUNTER(VM, gmap_shadow_sg_entry),
78 STATS_DESC_COUNTER(VM, gmap_shadow_pg_entry),
79 };
80
81 const struct kvm_stats_header kvm_vm_stats_header = {
82 .name_size = KVM_STATS_NAME_SIZE,
83 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
84 .id_offset = sizeof(struct kvm_stats_header),
85 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
86 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
87 sizeof(kvm_vm_stats_desc),
88 };
89
90 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
91 KVM_GENERIC_VCPU_STATS(),
92 STATS_DESC_COUNTER(VCPU, exit_userspace),
93 STATS_DESC_COUNTER(VCPU, exit_null),
94 STATS_DESC_COUNTER(VCPU, exit_external_request),
95 STATS_DESC_COUNTER(VCPU, exit_io_request),
96 STATS_DESC_COUNTER(VCPU, exit_external_interrupt),
97 STATS_DESC_COUNTER(VCPU, exit_stop_request),
98 STATS_DESC_COUNTER(VCPU, exit_validity),
99 STATS_DESC_COUNTER(VCPU, exit_instruction),
100 STATS_DESC_COUNTER(VCPU, exit_pei),
101 STATS_DESC_COUNTER(VCPU, halt_no_poll_steal),
102 STATS_DESC_COUNTER(VCPU, instruction_lctl),
103 STATS_DESC_COUNTER(VCPU, instruction_lctlg),
104 STATS_DESC_COUNTER(VCPU, instruction_stctl),
105 STATS_DESC_COUNTER(VCPU, instruction_stctg),
106 STATS_DESC_COUNTER(VCPU, exit_program_interruption),
107 STATS_DESC_COUNTER(VCPU, exit_instr_and_program),
108 STATS_DESC_COUNTER(VCPU, exit_operation_exception),
109 STATS_DESC_COUNTER(VCPU, deliver_ckc),
110 STATS_DESC_COUNTER(VCPU, deliver_cputm),
111 STATS_DESC_COUNTER(VCPU, deliver_external_call),
112 STATS_DESC_COUNTER(VCPU, deliver_emergency_signal),
113 STATS_DESC_COUNTER(VCPU, deliver_service_signal),
114 STATS_DESC_COUNTER(VCPU, deliver_virtio),
115 STATS_DESC_COUNTER(VCPU, deliver_stop_signal),
116 STATS_DESC_COUNTER(VCPU, deliver_prefix_signal),
117 STATS_DESC_COUNTER(VCPU, deliver_restart_signal),
118 STATS_DESC_COUNTER(VCPU, deliver_program),
119 STATS_DESC_COUNTER(VCPU, deliver_io),
120 STATS_DESC_COUNTER(VCPU, deliver_machine_check),
121 STATS_DESC_COUNTER(VCPU, exit_wait_state),
122 STATS_DESC_COUNTER(VCPU, inject_ckc),
123 STATS_DESC_COUNTER(VCPU, inject_cputm),
124 STATS_DESC_COUNTER(VCPU, inject_external_call),
125 STATS_DESC_COUNTER(VCPU, inject_emergency_signal),
126 STATS_DESC_COUNTER(VCPU, inject_mchk),
127 STATS_DESC_COUNTER(VCPU, inject_pfault_init),
128 STATS_DESC_COUNTER(VCPU, inject_program),
129 STATS_DESC_COUNTER(VCPU, inject_restart),
130 STATS_DESC_COUNTER(VCPU, inject_set_prefix),
131 STATS_DESC_COUNTER(VCPU, inject_stop_signal),
132 STATS_DESC_COUNTER(VCPU, instruction_epsw),
133 STATS_DESC_COUNTER(VCPU, instruction_gs),
134 STATS_DESC_COUNTER(VCPU, instruction_io_other),
135 STATS_DESC_COUNTER(VCPU, instruction_lpsw),
136 STATS_DESC_COUNTER(VCPU, instruction_lpswe),
137 STATS_DESC_COUNTER(VCPU, instruction_lpswey),
138 STATS_DESC_COUNTER(VCPU, instruction_pfmf),
139 STATS_DESC_COUNTER(VCPU, instruction_ptff),
140 STATS_DESC_COUNTER(VCPU, instruction_sck),
141 STATS_DESC_COUNTER(VCPU, instruction_sckpf),
142 STATS_DESC_COUNTER(VCPU, instruction_stidp),
143 STATS_DESC_COUNTER(VCPU, instruction_spx),
144 STATS_DESC_COUNTER(VCPU, instruction_stpx),
145 STATS_DESC_COUNTER(VCPU, instruction_stap),
146 STATS_DESC_COUNTER(VCPU, instruction_iske),
147 STATS_DESC_COUNTER(VCPU, instruction_ri),
148 STATS_DESC_COUNTER(VCPU, instruction_rrbe),
149 STATS_DESC_COUNTER(VCPU, instruction_sske),
150 STATS_DESC_COUNTER(VCPU, instruction_ipte_interlock),
151 STATS_DESC_COUNTER(VCPU, instruction_stsi),
152 STATS_DESC_COUNTER(VCPU, instruction_stfl),
153 STATS_DESC_COUNTER(VCPU, instruction_tb),
154 STATS_DESC_COUNTER(VCPU, instruction_tpi),
155 STATS_DESC_COUNTER(VCPU, instruction_tprot),
156 STATS_DESC_COUNTER(VCPU, instruction_tsch),
157 STATS_DESC_COUNTER(VCPU, instruction_sie),
158 STATS_DESC_COUNTER(VCPU, instruction_essa),
159 STATS_DESC_COUNTER(VCPU, instruction_sthyi),
160 STATS_DESC_COUNTER(VCPU, instruction_sigp_sense),
161 STATS_DESC_COUNTER(VCPU, instruction_sigp_sense_running),
162 STATS_DESC_COUNTER(VCPU, instruction_sigp_external_call),
163 STATS_DESC_COUNTER(VCPU, instruction_sigp_emergency),
164 STATS_DESC_COUNTER(VCPU, instruction_sigp_cond_emergency),
165 STATS_DESC_COUNTER(VCPU, instruction_sigp_start),
166 STATS_DESC_COUNTER(VCPU, instruction_sigp_stop),
167 STATS_DESC_COUNTER(VCPU, instruction_sigp_stop_store_status),
168 STATS_DESC_COUNTER(VCPU, instruction_sigp_store_status),
169 STATS_DESC_COUNTER(VCPU, instruction_sigp_store_adtl_status),
170 STATS_DESC_COUNTER(VCPU, instruction_sigp_arch),
171 STATS_DESC_COUNTER(VCPU, instruction_sigp_prefix),
172 STATS_DESC_COUNTER(VCPU, instruction_sigp_restart),
173 STATS_DESC_COUNTER(VCPU, instruction_sigp_init_cpu_reset),
174 STATS_DESC_COUNTER(VCPU, instruction_sigp_cpu_reset),
175 STATS_DESC_COUNTER(VCPU, instruction_sigp_unknown),
176 STATS_DESC_COUNTER(VCPU, instruction_diagnose_10),
177 STATS_DESC_COUNTER(VCPU, instruction_diagnose_44),
178 STATS_DESC_COUNTER(VCPU, instruction_diagnose_9c),
179 STATS_DESC_COUNTER(VCPU, diag_9c_ignored),
180 STATS_DESC_COUNTER(VCPU, diag_9c_forward),
181 STATS_DESC_COUNTER(VCPU, instruction_diagnose_258),
182 STATS_DESC_COUNTER(VCPU, instruction_diagnose_308),
183 STATS_DESC_COUNTER(VCPU, instruction_diagnose_500),
184 STATS_DESC_COUNTER(VCPU, instruction_diagnose_other),
185 STATS_DESC_COUNTER(VCPU, pfault_sync)
186 };
187
188 const struct kvm_stats_header kvm_vcpu_stats_header = {
189 .name_size = KVM_STATS_NAME_SIZE,
190 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
191 .id_offset = sizeof(struct kvm_stats_header),
192 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
193 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
194 sizeof(kvm_vcpu_stats_desc),
195 };
196
197 /* allow nested virtualization in KVM (if enabled by user space) */
198 static int nested;
199 module_param(nested, int, S_IRUGO);
200 MODULE_PARM_DESC(nested, "Nested virtualization support");
201
202 /* allow 1m huge page guest backing, if !nested */
203 static int hpage;
204 module_param(hpage, int, 0444);
205 MODULE_PARM_DESC(hpage, "1m huge page backing support");
206
207 /* maximum percentage of steal time for polling. >100 is treated like 100 */
208 static u8 halt_poll_max_steal = 10;
209 module_param(halt_poll_max_steal, byte, 0644);
210 MODULE_PARM_DESC(halt_poll_max_steal, "Maximum percentage of steal time to allow polling");
211
212 /* if set to true, the GISA will be initialized and used if available */
213 static bool use_gisa = true;
214 module_param(use_gisa, bool, 0644);
215 MODULE_PARM_DESC(use_gisa, "Use the GISA if the host supports it.");
216
217 /* maximum diag9c forwarding per second */
218 unsigned int diag9c_forwarding_hz;
219 module_param(diag9c_forwarding_hz, uint, 0644);
220 MODULE_PARM_DESC(diag9c_forwarding_hz, "Maximum diag9c forwarding per second, 0 to turn off");
221
222 /*
223 * allow asynchronous deinit for protected guests; enable by default since
224 * the feature is opt-in anyway
225 */
226 static int async_destroy = 1;
227 module_param(async_destroy, int, 0444);
228 MODULE_PARM_DESC(async_destroy, "Asynchronous destroy for protected guests");
229
230 /*
231 * For now we handle at most 16 double words as this is what the s390 base
232 * kernel handles and stores in the prefix page. If we ever need to go beyond
233 * this, this requires changes to code, but the external uapi can stay.
234 */
235 #define SIZE_INTERNAL 16
236
237 /*
238 * Base feature mask that defines default mask for facilities. Consists of the
239 * defines in FACILITIES_KVM and the non-hypervisor managed bits.
240 */
241 static unsigned long kvm_s390_fac_base[SIZE_INTERNAL] = { FACILITIES_KVM };
242 /*
243 * Extended feature mask. Consists of the defines in FACILITIES_KVM_CPUMODEL
244 * and defines the facilities that can be enabled via a cpu model.
245 */
246 static unsigned long kvm_s390_fac_ext[SIZE_INTERNAL] = { FACILITIES_KVM_CPUMODEL };
247
kvm_s390_fac_size(void)248 static unsigned long kvm_s390_fac_size(void)
249 {
250 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_MASK_SIZE_U64);
251 BUILD_BUG_ON(SIZE_INTERNAL > S390_ARCH_FAC_LIST_SIZE_U64);
252 BUILD_BUG_ON(SIZE_INTERNAL * sizeof(unsigned long) >
253 sizeof(stfle_fac_list));
254
255 return SIZE_INTERNAL;
256 }
257
258 /* available cpu features supported by kvm */
259 static DECLARE_BITMAP(kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
260 /* available subfunctions indicated via query / "test bit" */
261 static struct kvm_s390_vm_cpu_subfunc kvm_s390_available_subfunc;
262
263 static struct gmap_notifier gmap_notifier;
264 static struct gmap_notifier vsie_gmap_notifier;
265 debug_info_t *kvm_s390_dbf;
266 debug_info_t *kvm_s390_dbf_uv;
267
268 /* Section: not file related */
269 /* forward declarations */
270 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
271 unsigned long end);
272 static int sca_switch_to_extended(struct kvm *kvm);
273
kvm_clock_sync_scb(struct kvm_s390_sie_block * scb,u64 delta)274 static void kvm_clock_sync_scb(struct kvm_s390_sie_block *scb, u64 delta)
275 {
276 u8 delta_idx = 0;
277
278 /*
279 * The TOD jumps by delta, we have to compensate this by adding
280 * -delta to the epoch.
281 */
282 delta = -delta;
283
284 /* sign-extension - we're adding to signed values below */
285 if ((s64)delta < 0)
286 delta_idx = -1;
287
288 scb->epoch += delta;
289 if (scb->ecd & ECD_MEF) {
290 scb->epdx += delta_idx;
291 if (scb->epoch < delta)
292 scb->epdx += 1;
293 }
294 }
295
296 /*
297 * This callback is executed during stop_machine(). All CPUs are therefore
298 * temporarily stopped. In order not to change guest behavior, we have to
299 * disable preemption whenever we touch the epoch of kvm and the VCPUs,
300 * so a CPU won't be stopped while calculating with the epoch.
301 */
kvm_clock_sync(struct notifier_block * notifier,unsigned long val,void * v)302 static int kvm_clock_sync(struct notifier_block *notifier, unsigned long val,
303 void *v)
304 {
305 struct kvm *kvm;
306 struct kvm_vcpu *vcpu;
307 unsigned long i;
308 unsigned long long *delta = v;
309
310 list_for_each_entry(kvm, &vm_list, vm_list) {
311 kvm_for_each_vcpu(i, vcpu, kvm) {
312 kvm_clock_sync_scb(vcpu->arch.sie_block, *delta);
313 if (i == 0) {
314 kvm->arch.epoch = vcpu->arch.sie_block->epoch;
315 kvm->arch.epdx = vcpu->arch.sie_block->epdx;
316 }
317 if (vcpu->arch.cputm_enabled)
318 vcpu->arch.cputm_start += *delta;
319 if (vcpu->arch.vsie_block)
320 kvm_clock_sync_scb(vcpu->arch.vsie_block,
321 *delta);
322 }
323 }
324 return NOTIFY_OK;
325 }
326
327 static struct notifier_block kvm_clock_notifier = {
328 .notifier_call = kvm_clock_sync,
329 };
330
allow_cpu_feat(unsigned long nr)331 static void allow_cpu_feat(unsigned long nr)
332 {
333 set_bit_inv(nr, kvm_s390_available_cpu_feat);
334 }
335
plo_test_bit(unsigned char nr)336 static inline int plo_test_bit(unsigned char nr)
337 {
338 unsigned long function = (unsigned long)nr | 0x100;
339 int cc;
340
341 asm volatile(
342 " lgr 0,%[function]\n"
343 /* Parameter registers are ignored for "test bit" */
344 " plo 0,0,0,0(0)\n"
345 CC_IPM(cc)
346 : CC_OUT(cc, cc)
347 : [function] "d" (function)
348 : CC_CLOBBER_LIST("0"));
349 return CC_TRANSFORM(cc) == 0;
350 }
351
pfcr_query(u8 (* query)[16])352 static __always_inline void pfcr_query(u8 (*query)[16])
353 {
354 asm volatile(
355 " lghi 0,0\n"
356 " .insn rsy,0xeb0000000016,0,0,%[query]\n"
357 : [query] "=QS" (*query)
358 :
359 : "cc", "0");
360 }
361
__sortl_query(u8 (* query)[32])362 static __always_inline void __sortl_query(u8 (*query)[32])
363 {
364 asm volatile(
365 " lghi 0,0\n"
366 " la 1,%[query]\n"
367 /* Parameter registers are ignored */
368 " .insn rre,0xb9380000,2,4\n"
369 : [query] "=R" (*query)
370 :
371 : "cc", "0", "1");
372 }
373
__dfltcc_query(u8 (* query)[32])374 static __always_inline void __dfltcc_query(u8 (*query)[32])
375 {
376 asm volatile(
377 " lghi 0,0\n"
378 " la 1,%[query]\n"
379 /* Parameter registers are ignored */
380 " .insn rrf,0xb9390000,2,4,6,0\n"
381 : [query] "=R" (*query)
382 :
383 : "cc", "0", "1");
384 }
385
kvm_s390_cpu_feat_init(void)386 static void __init kvm_s390_cpu_feat_init(void)
387 {
388 int i;
389
390 for (i = 0; i < 256; ++i) {
391 if (plo_test_bit(i))
392 kvm_s390_available_subfunc.plo[i >> 3] |= 0x80 >> (i & 7);
393 }
394
395 if (test_facility(28)) /* TOD-clock steering */
396 ptff(kvm_s390_available_subfunc.ptff,
397 sizeof(kvm_s390_available_subfunc.ptff),
398 PTFF_QAF);
399
400 if (test_facility(17)) { /* MSA */
401 __cpacf_query(CPACF_KMAC, (cpacf_mask_t *)
402 kvm_s390_available_subfunc.kmac);
403 __cpacf_query(CPACF_KMC, (cpacf_mask_t *)
404 kvm_s390_available_subfunc.kmc);
405 __cpacf_query(CPACF_KM, (cpacf_mask_t *)
406 kvm_s390_available_subfunc.km);
407 __cpacf_query(CPACF_KIMD, (cpacf_mask_t *)
408 kvm_s390_available_subfunc.kimd);
409 __cpacf_query(CPACF_KLMD, (cpacf_mask_t *)
410 kvm_s390_available_subfunc.klmd);
411 }
412 if (test_facility(76)) /* MSA3 */
413 __cpacf_query(CPACF_PCKMO, (cpacf_mask_t *)
414 kvm_s390_available_subfunc.pckmo);
415 if (test_facility(77)) { /* MSA4 */
416 __cpacf_query(CPACF_KMCTR, (cpacf_mask_t *)
417 kvm_s390_available_subfunc.kmctr);
418 __cpacf_query(CPACF_KMF, (cpacf_mask_t *)
419 kvm_s390_available_subfunc.kmf);
420 __cpacf_query(CPACF_KMO, (cpacf_mask_t *)
421 kvm_s390_available_subfunc.kmo);
422 __cpacf_query(CPACF_PCC, (cpacf_mask_t *)
423 kvm_s390_available_subfunc.pcc);
424 }
425 if (test_facility(57)) /* MSA5 */
426 __cpacf_query(CPACF_PRNO, (cpacf_mask_t *)
427 kvm_s390_available_subfunc.ppno);
428
429 if (test_facility(146)) /* MSA8 */
430 __cpacf_query(CPACF_KMA, (cpacf_mask_t *)
431 kvm_s390_available_subfunc.kma);
432
433 if (test_facility(155)) /* MSA9 */
434 __cpacf_query(CPACF_KDSA, (cpacf_mask_t *)
435 kvm_s390_available_subfunc.kdsa);
436
437 if (test_facility(150)) /* SORTL */
438 __sortl_query(&kvm_s390_available_subfunc.sortl);
439
440 if (test_facility(151)) /* DFLTCC */
441 __dfltcc_query(&kvm_s390_available_subfunc.dfltcc);
442
443 if (test_facility(201)) /* PFCR */
444 pfcr_query(&kvm_s390_available_subfunc.pfcr);
445
446 if (MACHINE_HAS_ESOP)
447 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_ESOP);
448 /*
449 * We need SIE support, ESOP (PROT_READ protection for gmap_shadow),
450 * 64bit SCAO (SCA passthrough) and IDTE (for gmap_shadow unshadowing).
451 */
452 if (!sclp.has_sief2 || !MACHINE_HAS_ESOP || !sclp.has_64bscao ||
453 !test_facility(3) || !nested)
454 return;
455 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIEF2);
456 if (sclp.has_64bscao)
457 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_64BSCAO);
458 if (sclp.has_siif)
459 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_SIIF);
460 if (sclp.has_gpere)
461 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GPERE);
462 if (sclp.has_gsls)
463 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_GSLS);
464 if (sclp.has_ib)
465 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IB);
466 if (sclp.has_cei)
467 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_CEI);
468 if (sclp.has_ibs)
469 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_IBS);
470 if (sclp.has_kss)
471 allow_cpu_feat(KVM_S390_VM_CPU_FEAT_KSS);
472 /*
473 * KVM_S390_VM_CPU_FEAT_SKEY: Wrong shadow of PTE.I bits will make
474 * all skey handling functions read/set the skey from the PGSTE
475 * instead of the real storage key.
476 *
477 * KVM_S390_VM_CPU_FEAT_CMMA: Wrong shadow of PTE.I bits will make
478 * pages being detected as preserved although they are resident.
479 *
480 * KVM_S390_VM_CPU_FEAT_PFMFI: Wrong shadow of PTE.I bits will
481 * have the same effect as for KVM_S390_VM_CPU_FEAT_SKEY.
482 *
483 * For KVM_S390_VM_CPU_FEAT_SKEY, KVM_S390_VM_CPU_FEAT_CMMA and
484 * KVM_S390_VM_CPU_FEAT_PFMFI, all PTE.I and PGSTE bits have to be
485 * correctly shadowed. We can do that for the PGSTE but not for PTE.I.
486 *
487 * KVM_S390_VM_CPU_FEAT_SIGPIF: Wrong SCB addresses in the SCA. We
488 * cannot easily shadow the SCA because of the ipte lock.
489 */
490 }
491
__kvm_s390_init(void)492 static int __init __kvm_s390_init(void)
493 {
494 int rc = -ENOMEM;
495
496 kvm_s390_dbf = debug_register("kvm-trace", 32, 1, 7 * sizeof(long));
497 if (!kvm_s390_dbf)
498 return -ENOMEM;
499
500 kvm_s390_dbf_uv = debug_register("kvm-uv", 32, 1, 7 * sizeof(long));
501 if (!kvm_s390_dbf_uv)
502 goto err_kvm_uv;
503
504 if (debug_register_view(kvm_s390_dbf, &debug_sprintf_view) ||
505 debug_register_view(kvm_s390_dbf_uv, &debug_sprintf_view))
506 goto err_debug_view;
507
508 kvm_s390_cpu_feat_init();
509
510 /* Register floating interrupt controller interface. */
511 rc = kvm_register_device_ops(&kvm_flic_ops, KVM_DEV_TYPE_FLIC);
512 if (rc) {
513 pr_err("A FLIC registration call failed with rc=%d\n", rc);
514 goto err_flic;
515 }
516
517 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
518 rc = kvm_s390_pci_init();
519 if (rc) {
520 pr_err("Unable to allocate AIFT for PCI\n");
521 goto err_pci;
522 }
523 }
524
525 rc = kvm_s390_gib_init(GAL_ISC);
526 if (rc)
527 goto err_gib;
528
529 gmap_notifier.notifier_call = kvm_gmap_notifier;
530 gmap_register_pte_notifier(&gmap_notifier);
531 vsie_gmap_notifier.notifier_call = kvm_s390_vsie_gmap_notifier;
532 gmap_register_pte_notifier(&vsie_gmap_notifier);
533 atomic_notifier_chain_register(&s390_epoch_delta_notifier,
534 &kvm_clock_notifier);
535
536 return 0;
537
538 err_gib:
539 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
540 kvm_s390_pci_exit();
541 err_pci:
542 err_flic:
543 err_debug_view:
544 debug_unregister(kvm_s390_dbf_uv);
545 err_kvm_uv:
546 debug_unregister(kvm_s390_dbf);
547 return rc;
548 }
549
__kvm_s390_exit(void)550 static void __kvm_s390_exit(void)
551 {
552 gmap_unregister_pte_notifier(&gmap_notifier);
553 gmap_unregister_pte_notifier(&vsie_gmap_notifier);
554 atomic_notifier_chain_unregister(&s390_epoch_delta_notifier,
555 &kvm_clock_notifier);
556
557 kvm_s390_gib_destroy();
558 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
559 kvm_s390_pci_exit();
560 debug_unregister(kvm_s390_dbf);
561 debug_unregister(kvm_s390_dbf_uv);
562 }
563
564 /* Section: device related */
kvm_arch_dev_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)565 long kvm_arch_dev_ioctl(struct file *filp,
566 unsigned int ioctl, unsigned long arg)
567 {
568 if (ioctl == KVM_S390_ENABLE_SIE)
569 return s390_enable_sie();
570 return -EINVAL;
571 }
572
kvm_vm_ioctl_check_extension(struct kvm * kvm,long ext)573 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
574 {
575 int r;
576
577 switch (ext) {
578 case KVM_CAP_S390_PSW:
579 case KVM_CAP_S390_GMAP:
580 case KVM_CAP_SYNC_MMU:
581 #ifdef CONFIG_KVM_S390_UCONTROL
582 case KVM_CAP_S390_UCONTROL:
583 #endif
584 case KVM_CAP_ASYNC_PF:
585 case KVM_CAP_SYNC_REGS:
586 case KVM_CAP_ONE_REG:
587 case KVM_CAP_ENABLE_CAP:
588 case KVM_CAP_S390_CSS_SUPPORT:
589 case KVM_CAP_IOEVENTFD:
590 case KVM_CAP_S390_IRQCHIP:
591 case KVM_CAP_VM_ATTRIBUTES:
592 case KVM_CAP_MP_STATE:
593 case KVM_CAP_IMMEDIATE_EXIT:
594 case KVM_CAP_S390_INJECT_IRQ:
595 case KVM_CAP_S390_USER_SIGP:
596 case KVM_CAP_S390_USER_STSI:
597 case KVM_CAP_S390_SKEYS:
598 case KVM_CAP_S390_IRQ_STATE:
599 case KVM_CAP_S390_USER_INSTR0:
600 case KVM_CAP_S390_CMMA_MIGRATION:
601 case KVM_CAP_S390_AIS:
602 case KVM_CAP_S390_AIS_MIGRATION:
603 case KVM_CAP_S390_VCPU_RESETS:
604 case KVM_CAP_SET_GUEST_DEBUG:
605 case KVM_CAP_S390_DIAG318:
606 case KVM_CAP_IRQFD_RESAMPLE:
607 r = 1;
608 break;
609 case KVM_CAP_SET_GUEST_DEBUG2:
610 r = KVM_GUESTDBG_VALID_MASK;
611 break;
612 case KVM_CAP_S390_HPAGE_1M:
613 r = 0;
614 if (hpage && !(kvm && kvm_is_ucontrol(kvm)))
615 r = 1;
616 break;
617 case KVM_CAP_S390_MEM_OP:
618 r = MEM_OP_MAX_SIZE;
619 break;
620 case KVM_CAP_S390_MEM_OP_EXTENSION:
621 /*
622 * Flag bits indicating which extensions are supported.
623 * If r > 0, the base extension must also be supported/indicated,
624 * in order to maintain backwards compatibility.
625 */
626 r = KVM_S390_MEMOP_EXTENSION_CAP_BASE |
627 KVM_S390_MEMOP_EXTENSION_CAP_CMPXCHG;
628 break;
629 case KVM_CAP_NR_VCPUS:
630 case KVM_CAP_MAX_VCPUS:
631 case KVM_CAP_MAX_VCPU_ID:
632 r = KVM_S390_BSCA_CPU_SLOTS;
633 if (!kvm_s390_use_sca_entries())
634 r = KVM_MAX_VCPUS;
635 else if (sclp.has_esca && sclp.has_64bscao)
636 r = KVM_S390_ESCA_CPU_SLOTS;
637 if (ext == KVM_CAP_NR_VCPUS)
638 r = min_t(unsigned int, num_online_cpus(), r);
639 break;
640 case KVM_CAP_S390_COW:
641 r = MACHINE_HAS_ESOP;
642 break;
643 case KVM_CAP_S390_VECTOR_REGISTERS:
644 r = test_facility(129);
645 break;
646 case KVM_CAP_S390_RI:
647 r = test_facility(64);
648 break;
649 case KVM_CAP_S390_GS:
650 r = test_facility(133);
651 break;
652 case KVM_CAP_S390_BPB:
653 r = test_facility(82);
654 break;
655 case KVM_CAP_S390_PROTECTED_ASYNC_DISABLE:
656 r = async_destroy && is_prot_virt_host();
657 break;
658 case KVM_CAP_S390_PROTECTED:
659 r = is_prot_virt_host();
660 break;
661 case KVM_CAP_S390_PROTECTED_DUMP: {
662 u64 pv_cmds_dump[] = {
663 BIT_UVC_CMD_DUMP_INIT,
664 BIT_UVC_CMD_DUMP_CONFIG_STOR_STATE,
665 BIT_UVC_CMD_DUMP_CPU,
666 BIT_UVC_CMD_DUMP_COMPLETE,
667 };
668 int i;
669
670 r = is_prot_virt_host();
671
672 for (i = 0; i < ARRAY_SIZE(pv_cmds_dump); i++) {
673 if (!test_bit_inv(pv_cmds_dump[i],
674 (unsigned long *)&uv_info.inst_calls_list)) {
675 r = 0;
676 break;
677 }
678 }
679 break;
680 }
681 case KVM_CAP_S390_ZPCI_OP:
682 r = kvm_s390_pci_interp_allowed();
683 break;
684 case KVM_CAP_S390_CPU_TOPOLOGY:
685 r = test_facility(11);
686 break;
687 default:
688 r = 0;
689 }
690 return r;
691 }
692
kvm_arch_sync_dirty_log(struct kvm * kvm,struct kvm_memory_slot * memslot)693 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
694 {
695 int i;
696 gfn_t cur_gfn, last_gfn;
697 unsigned long gaddr, vmaddr;
698 struct gmap *gmap = kvm->arch.gmap;
699 DECLARE_BITMAP(bitmap, _PAGE_ENTRIES);
700
701 /* Loop over all guest segments */
702 cur_gfn = memslot->base_gfn;
703 last_gfn = memslot->base_gfn + memslot->npages;
704 for (; cur_gfn <= last_gfn; cur_gfn += _PAGE_ENTRIES) {
705 gaddr = gfn_to_gpa(cur_gfn);
706 vmaddr = gfn_to_hva_memslot(memslot, cur_gfn);
707 if (kvm_is_error_hva(vmaddr))
708 continue;
709
710 bitmap_zero(bitmap, _PAGE_ENTRIES);
711 gmap_sync_dirty_log_pmd(gmap, bitmap, gaddr, vmaddr);
712 for (i = 0; i < _PAGE_ENTRIES; i++) {
713 if (test_bit(i, bitmap))
714 mark_page_dirty(kvm, cur_gfn + i);
715 }
716
717 if (fatal_signal_pending(current))
718 return;
719 cond_resched();
720 }
721 }
722
723 /* Section: vm related */
724 static void sca_del_vcpu(struct kvm_vcpu *vcpu);
725
726 /*
727 * Get (and clear) the dirty memory log for a memory slot.
728 */
kvm_vm_ioctl_get_dirty_log(struct kvm * kvm,struct kvm_dirty_log * log)729 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
730 struct kvm_dirty_log *log)
731 {
732 int r;
733 unsigned long n;
734 struct kvm_memory_slot *memslot;
735 int is_dirty;
736
737 if (kvm_is_ucontrol(kvm))
738 return -EINVAL;
739
740 mutex_lock(&kvm->slots_lock);
741
742 r = -EINVAL;
743 if (log->slot >= KVM_USER_MEM_SLOTS)
744 goto out;
745
746 r = kvm_get_dirty_log(kvm, log, &is_dirty, &memslot);
747 if (r)
748 goto out;
749
750 /* Clear the dirty log */
751 if (is_dirty) {
752 n = kvm_dirty_bitmap_bytes(memslot);
753 memset(memslot->dirty_bitmap, 0, n);
754 }
755 r = 0;
756 out:
757 mutex_unlock(&kvm->slots_lock);
758 return r;
759 }
760
icpt_operexc_on_all_vcpus(struct kvm * kvm)761 static void icpt_operexc_on_all_vcpus(struct kvm *kvm)
762 {
763 unsigned long i;
764 struct kvm_vcpu *vcpu;
765
766 kvm_for_each_vcpu(i, vcpu, kvm) {
767 kvm_s390_sync_request(KVM_REQ_ICPT_OPEREXC, vcpu);
768 }
769 }
770
kvm_vm_ioctl_enable_cap(struct kvm * kvm,struct kvm_enable_cap * cap)771 int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
772 {
773 int r;
774
775 if (cap->flags)
776 return -EINVAL;
777
778 switch (cap->cap) {
779 case KVM_CAP_S390_IRQCHIP:
780 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_IRQCHIP");
781 kvm->arch.use_irqchip = 1;
782 r = 0;
783 break;
784 case KVM_CAP_S390_USER_SIGP:
785 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_SIGP");
786 kvm->arch.user_sigp = 1;
787 r = 0;
788 break;
789 case KVM_CAP_S390_VECTOR_REGISTERS:
790 mutex_lock(&kvm->lock);
791 if (kvm->created_vcpus) {
792 r = -EBUSY;
793 } else if (cpu_has_vx()) {
794 set_kvm_facility(kvm->arch.model.fac_mask, 129);
795 set_kvm_facility(kvm->arch.model.fac_list, 129);
796 if (test_facility(134)) {
797 set_kvm_facility(kvm->arch.model.fac_mask, 134);
798 set_kvm_facility(kvm->arch.model.fac_list, 134);
799 }
800 if (test_facility(135)) {
801 set_kvm_facility(kvm->arch.model.fac_mask, 135);
802 set_kvm_facility(kvm->arch.model.fac_list, 135);
803 }
804 if (test_facility(148)) {
805 set_kvm_facility(kvm->arch.model.fac_mask, 148);
806 set_kvm_facility(kvm->arch.model.fac_list, 148);
807 }
808 if (test_facility(152)) {
809 set_kvm_facility(kvm->arch.model.fac_mask, 152);
810 set_kvm_facility(kvm->arch.model.fac_list, 152);
811 }
812 if (test_facility(192)) {
813 set_kvm_facility(kvm->arch.model.fac_mask, 192);
814 set_kvm_facility(kvm->arch.model.fac_list, 192);
815 }
816 if (test_facility(198)) {
817 set_kvm_facility(kvm->arch.model.fac_mask, 198);
818 set_kvm_facility(kvm->arch.model.fac_list, 198);
819 }
820 if (test_facility(199)) {
821 set_kvm_facility(kvm->arch.model.fac_mask, 199);
822 set_kvm_facility(kvm->arch.model.fac_list, 199);
823 }
824 r = 0;
825 } else
826 r = -EINVAL;
827 mutex_unlock(&kvm->lock);
828 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_VECTOR_REGISTERS %s",
829 r ? "(not available)" : "(success)");
830 break;
831 case KVM_CAP_S390_RI:
832 r = -EINVAL;
833 mutex_lock(&kvm->lock);
834 if (kvm->created_vcpus) {
835 r = -EBUSY;
836 } else if (test_facility(64)) {
837 set_kvm_facility(kvm->arch.model.fac_mask, 64);
838 set_kvm_facility(kvm->arch.model.fac_list, 64);
839 r = 0;
840 }
841 mutex_unlock(&kvm->lock);
842 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_RI %s",
843 r ? "(not available)" : "(success)");
844 break;
845 case KVM_CAP_S390_AIS:
846 mutex_lock(&kvm->lock);
847 if (kvm->created_vcpus) {
848 r = -EBUSY;
849 } else {
850 set_kvm_facility(kvm->arch.model.fac_mask, 72);
851 set_kvm_facility(kvm->arch.model.fac_list, 72);
852 r = 0;
853 }
854 mutex_unlock(&kvm->lock);
855 VM_EVENT(kvm, 3, "ENABLE: AIS %s",
856 r ? "(not available)" : "(success)");
857 break;
858 case KVM_CAP_S390_GS:
859 r = -EINVAL;
860 mutex_lock(&kvm->lock);
861 if (kvm->created_vcpus) {
862 r = -EBUSY;
863 } else if (test_facility(133)) {
864 set_kvm_facility(kvm->arch.model.fac_mask, 133);
865 set_kvm_facility(kvm->arch.model.fac_list, 133);
866 r = 0;
867 }
868 mutex_unlock(&kvm->lock);
869 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_GS %s",
870 r ? "(not available)" : "(success)");
871 break;
872 case KVM_CAP_S390_HPAGE_1M:
873 mutex_lock(&kvm->lock);
874 if (kvm->created_vcpus)
875 r = -EBUSY;
876 else if (!hpage || kvm->arch.use_cmma || kvm_is_ucontrol(kvm))
877 r = -EINVAL;
878 else {
879 r = 0;
880 mmap_write_lock(kvm->mm);
881 kvm->mm->context.allow_gmap_hpage_1m = 1;
882 mmap_write_unlock(kvm->mm);
883 /*
884 * We might have to create fake 4k page
885 * tables. To avoid that the hardware works on
886 * stale PGSTEs, we emulate these instructions.
887 */
888 kvm->arch.use_skf = 0;
889 kvm->arch.use_pfmfi = 0;
890 }
891 mutex_unlock(&kvm->lock);
892 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_HPAGE %s",
893 r ? "(not available)" : "(success)");
894 break;
895 case KVM_CAP_S390_USER_STSI:
896 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_STSI");
897 kvm->arch.user_stsi = 1;
898 r = 0;
899 break;
900 case KVM_CAP_S390_USER_INSTR0:
901 VM_EVENT(kvm, 3, "%s", "ENABLE: CAP_S390_USER_INSTR0");
902 kvm->arch.user_instr0 = 1;
903 icpt_operexc_on_all_vcpus(kvm);
904 r = 0;
905 break;
906 case KVM_CAP_S390_CPU_TOPOLOGY:
907 r = -EINVAL;
908 mutex_lock(&kvm->lock);
909 if (kvm->created_vcpus) {
910 r = -EBUSY;
911 } else if (test_facility(11)) {
912 set_kvm_facility(kvm->arch.model.fac_mask, 11);
913 set_kvm_facility(kvm->arch.model.fac_list, 11);
914 r = 0;
915 }
916 mutex_unlock(&kvm->lock);
917 VM_EVENT(kvm, 3, "ENABLE: CAP_S390_CPU_TOPOLOGY %s",
918 r ? "(not available)" : "(success)");
919 break;
920 default:
921 r = -EINVAL;
922 break;
923 }
924 return r;
925 }
926
kvm_s390_get_mem_control(struct kvm * kvm,struct kvm_device_attr * attr)927 static int kvm_s390_get_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
928 {
929 int ret;
930
931 switch (attr->attr) {
932 case KVM_S390_VM_MEM_LIMIT_SIZE:
933 ret = 0;
934 VM_EVENT(kvm, 3, "QUERY: max guest memory: %lu bytes",
935 kvm->arch.mem_limit);
936 if (put_user(kvm->arch.mem_limit, (u64 __user *)attr->addr))
937 ret = -EFAULT;
938 break;
939 default:
940 ret = -ENXIO;
941 break;
942 }
943 return ret;
944 }
945
kvm_s390_set_mem_control(struct kvm * kvm,struct kvm_device_attr * attr)946 static int kvm_s390_set_mem_control(struct kvm *kvm, struct kvm_device_attr *attr)
947 {
948 int ret;
949 unsigned int idx;
950 switch (attr->attr) {
951 case KVM_S390_VM_MEM_ENABLE_CMMA:
952 ret = -ENXIO;
953 if (!sclp.has_cmma)
954 break;
955
956 VM_EVENT(kvm, 3, "%s", "ENABLE: CMMA support");
957 mutex_lock(&kvm->lock);
958 if (kvm->created_vcpus)
959 ret = -EBUSY;
960 else if (kvm->mm->context.allow_gmap_hpage_1m)
961 ret = -EINVAL;
962 else {
963 kvm->arch.use_cmma = 1;
964 /* Not compatible with cmma. */
965 kvm->arch.use_pfmfi = 0;
966 ret = 0;
967 }
968 mutex_unlock(&kvm->lock);
969 break;
970 case KVM_S390_VM_MEM_CLR_CMMA:
971 ret = -ENXIO;
972 if (!sclp.has_cmma)
973 break;
974 ret = -EINVAL;
975 if (!kvm->arch.use_cmma)
976 break;
977
978 VM_EVENT(kvm, 3, "%s", "RESET: CMMA states");
979 mutex_lock(&kvm->lock);
980 idx = srcu_read_lock(&kvm->srcu);
981 s390_reset_cmma(kvm->arch.gmap->mm);
982 srcu_read_unlock(&kvm->srcu, idx);
983 mutex_unlock(&kvm->lock);
984 ret = 0;
985 break;
986 case KVM_S390_VM_MEM_LIMIT_SIZE: {
987 unsigned long new_limit;
988
989 if (kvm_is_ucontrol(kvm))
990 return -EINVAL;
991
992 if (get_user(new_limit, (u64 __user *)attr->addr))
993 return -EFAULT;
994
995 if (kvm->arch.mem_limit != KVM_S390_NO_MEM_LIMIT &&
996 new_limit > kvm->arch.mem_limit)
997 return -E2BIG;
998
999 if (!new_limit)
1000 return -EINVAL;
1001
1002 /* gmap_create takes last usable address */
1003 if (new_limit != KVM_S390_NO_MEM_LIMIT)
1004 new_limit -= 1;
1005
1006 ret = -EBUSY;
1007 mutex_lock(&kvm->lock);
1008 if (!kvm->created_vcpus) {
1009 /* gmap_create will round the limit up */
1010 struct gmap *new = gmap_create(current->mm, new_limit);
1011
1012 if (!new) {
1013 ret = -ENOMEM;
1014 } else {
1015 gmap_remove(kvm->arch.gmap);
1016 new->private = kvm;
1017 kvm->arch.gmap = new;
1018 ret = 0;
1019 }
1020 }
1021 mutex_unlock(&kvm->lock);
1022 VM_EVENT(kvm, 3, "SET: max guest address: %lu", new_limit);
1023 VM_EVENT(kvm, 3, "New guest asce: 0x%pK",
1024 (void *) kvm->arch.gmap->asce);
1025 break;
1026 }
1027 default:
1028 ret = -ENXIO;
1029 break;
1030 }
1031 return ret;
1032 }
1033
1034 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu);
1035
kvm_s390_vcpu_crypto_reset_all(struct kvm * kvm)1036 void kvm_s390_vcpu_crypto_reset_all(struct kvm *kvm)
1037 {
1038 struct kvm_vcpu *vcpu;
1039 unsigned long i;
1040
1041 kvm_s390_vcpu_block_all(kvm);
1042
1043 kvm_for_each_vcpu(i, vcpu, kvm) {
1044 kvm_s390_vcpu_crypto_setup(vcpu);
1045 /* recreate the shadow crycb by leaving the VSIE handler */
1046 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu);
1047 }
1048
1049 kvm_s390_vcpu_unblock_all(kvm);
1050 }
1051
kvm_s390_vm_set_crypto(struct kvm * kvm,struct kvm_device_attr * attr)1052 static int kvm_s390_vm_set_crypto(struct kvm *kvm, struct kvm_device_attr *attr)
1053 {
1054 mutex_lock(&kvm->lock);
1055 switch (attr->attr) {
1056 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
1057 if (!test_kvm_facility(kvm, 76)) {
1058 mutex_unlock(&kvm->lock);
1059 return -EINVAL;
1060 }
1061 get_random_bytes(
1062 kvm->arch.crypto.crycb->aes_wrapping_key_mask,
1063 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1064 kvm->arch.crypto.aes_kw = 1;
1065 VM_EVENT(kvm, 3, "%s", "ENABLE: AES keywrapping support");
1066 break;
1067 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
1068 if (!test_kvm_facility(kvm, 76)) {
1069 mutex_unlock(&kvm->lock);
1070 return -EINVAL;
1071 }
1072 get_random_bytes(
1073 kvm->arch.crypto.crycb->dea_wrapping_key_mask,
1074 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1075 kvm->arch.crypto.dea_kw = 1;
1076 VM_EVENT(kvm, 3, "%s", "ENABLE: DEA keywrapping support");
1077 break;
1078 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
1079 if (!test_kvm_facility(kvm, 76)) {
1080 mutex_unlock(&kvm->lock);
1081 return -EINVAL;
1082 }
1083 kvm->arch.crypto.aes_kw = 0;
1084 memset(kvm->arch.crypto.crycb->aes_wrapping_key_mask, 0,
1085 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
1086 VM_EVENT(kvm, 3, "%s", "DISABLE: AES keywrapping support");
1087 break;
1088 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
1089 if (!test_kvm_facility(kvm, 76)) {
1090 mutex_unlock(&kvm->lock);
1091 return -EINVAL;
1092 }
1093 kvm->arch.crypto.dea_kw = 0;
1094 memset(kvm->arch.crypto.crycb->dea_wrapping_key_mask, 0,
1095 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
1096 VM_EVENT(kvm, 3, "%s", "DISABLE: DEA keywrapping support");
1097 break;
1098 case KVM_S390_VM_CRYPTO_ENABLE_APIE:
1099 if (!ap_instructions_available()) {
1100 mutex_unlock(&kvm->lock);
1101 return -EOPNOTSUPP;
1102 }
1103 kvm->arch.crypto.apie = 1;
1104 break;
1105 case KVM_S390_VM_CRYPTO_DISABLE_APIE:
1106 if (!ap_instructions_available()) {
1107 mutex_unlock(&kvm->lock);
1108 return -EOPNOTSUPP;
1109 }
1110 kvm->arch.crypto.apie = 0;
1111 break;
1112 default:
1113 mutex_unlock(&kvm->lock);
1114 return -ENXIO;
1115 }
1116
1117 kvm_s390_vcpu_crypto_reset_all(kvm);
1118 mutex_unlock(&kvm->lock);
1119 return 0;
1120 }
1121
kvm_s390_vcpu_pci_setup(struct kvm_vcpu * vcpu)1122 static void kvm_s390_vcpu_pci_setup(struct kvm_vcpu *vcpu)
1123 {
1124 /* Only set the ECB bits after guest requests zPCI interpretation */
1125 if (!vcpu->kvm->arch.use_zpci_interp)
1126 return;
1127
1128 vcpu->arch.sie_block->ecb2 |= ECB2_ZPCI_LSI;
1129 vcpu->arch.sie_block->ecb3 |= ECB3_AISII + ECB3_AISI;
1130 }
1131
kvm_s390_vcpu_pci_enable_interp(struct kvm * kvm)1132 void kvm_s390_vcpu_pci_enable_interp(struct kvm *kvm)
1133 {
1134 struct kvm_vcpu *vcpu;
1135 unsigned long i;
1136
1137 lockdep_assert_held(&kvm->lock);
1138
1139 if (!kvm_s390_pci_interp_allowed())
1140 return;
1141
1142 /*
1143 * If host is configured for PCI and the necessary facilities are
1144 * available, turn on interpretation for the life of this guest
1145 */
1146 kvm->arch.use_zpci_interp = 1;
1147
1148 kvm_s390_vcpu_block_all(kvm);
1149
1150 kvm_for_each_vcpu(i, vcpu, kvm) {
1151 kvm_s390_vcpu_pci_setup(vcpu);
1152 kvm_s390_sync_request(KVM_REQ_VSIE_RESTART, vcpu);
1153 }
1154
1155 kvm_s390_vcpu_unblock_all(kvm);
1156 }
1157
kvm_s390_sync_request_broadcast(struct kvm * kvm,int req)1158 static void kvm_s390_sync_request_broadcast(struct kvm *kvm, int req)
1159 {
1160 unsigned long cx;
1161 struct kvm_vcpu *vcpu;
1162
1163 kvm_for_each_vcpu(cx, vcpu, kvm)
1164 kvm_s390_sync_request(req, vcpu);
1165 }
1166
1167 /*
1168 * Must be called with kvm->srcu held to avoid races on memslots, and with
1169 * kvm->slots_lock to avoid races with ourselves and kvm_s390_vm_stop_migration.
1170 */
kvm_s390_vm_start_migration(struct kvm * kvm)1171 static int kvm_s390_vm_start_migration(struct kvm *kvm)
1172 {
1173 struct kvm_memory_slot *ms;
1174 struct kvm_memslots *slots;
1175 unsigned long ram_pages = 0;
1176 int bkt;
1177
1178 /* migration mode already enabled */
1179 if (kvm->arch.migration_mode)
1180 return 0;
1181 slots = kvm_memslots(kvm);
1182 if (!slots || kvm_memslots_empty(slots))
1183 return -EINVAL;
1184
1185 if (!kvm->arch.use_cmma) {
1186 kvm->arch.migration_mode = 1;
1187 return 0;
1188 }
1189 /* mark all the pages in active slots as dirty */
1190 kvm_for_each_memslot(ms, bkt, slots) {
1191 if (!ms->dirty_bitmap)
1192 return -EINVAL;
1193 /*
1194 * The second half of the bitmap is only used on x86,
1195 * and would be wasted otherwise, so we put it to good
1196 * use here to keep track of the state of the storage
1197 * attributes.
1198 */
1199 memset(kvm_second_dirty_bitmap(ms), 0xff, kvm_dirty_bitmap_bytes(ms));
1200 ram_pages += ms->npages;
1201 }
1202 atomic64_set(&kvm->arch.cmma_dirty_pages, ram_pages);
1203 kvm->arch.migration_mode = 1;
1204 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_START_MIGRATION);
1205 return 0;
1206 }
1207
1208 /*
1209 * Must be called with kvm->slots_lock to avoid races with ourselves and
1210 * kvm_s390_vm_start_migration.
1211 */
kvm_s390_vm_stop_migration(struct kvm * kvm)1212 static int kvm_s390_vm_stop_migration(struct kvm *kvm)
1213 {
1214 /* migration mode already disabled */
1215 if (!kvm->arch.migration_mode)
1216 return 0;
1217 kvm->arch.migration_mode = 0;
1218 if (kvm->arch.use_cmma)
1219 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_STOP_MIGRATION);
1220 return 0;
1221 }
1222
kvm_s390_vm_set_migration(struct kvm * kvm,struct kvm_device_attr * attr)1223 static int kvm_s390_vm_set_migration(struct kvm *kvm,
1224 struct kvm_device_attr *attr)
1225 {
1226 int res = -ENXIO;
1227
1228 mutex_lock(&kvm->slots_lock);
1229 switch (attr->attr) {
1230 case KVM_S390_VM_MIGRATION_START:
1231 res = kvm_s390_vm_start_migration(kvm);
1232 break;
1233 case KVM_S390_VM_MIGRATION_STOP:
1234 res = kvm_s390_vm_stop_migration(kvm);
1235 break;
1236 default:
1237 break;
1238 }
1239 mutex_unlock(&kvm->slots_lock);
1240
1241 return res;
1242 }
1243
kvm_s390_vm_get_migration(struct kvm * kvm,struct kvm_device_attr * attr)1244 static int kvm_s390_vm_get_migration(struct kvm *kvm,
1245 struct kvm_device_attr *attr)
1246 {
1247 u64 mig = kvm->arch.migration_mode;
1248
1249 if (attr->attr != KVM_S390_VM_MIGRATION_STATUS)
1250 return -ENXIO;
1251
1252 if (copy_to_user((void __user *)attr->addr, &mig, sizeof(mig)))
1253 return -EFAULT;
1254 return 0;
1255 }
1256
1257 static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod);
1258
kvm_s390_set_tod_ext(struct kvm * kvm,struct kvm_device_attr * attr)1259 static int kvm_s390_set_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1260 {
1261 struct kvm_s390_vm_tod_clock gtod;
1262
1263 if (copy_from_user(>od, (void __user *)attr->addr, sizeof(gtod)))
1264 return -EFAULT;
1265
1266 if (!test_kvm_facility(kvm, 139) && gtod.epoch_idx)
1267 return -EINVAL;
1268 __kvm_s390_set_tod_clock(kvm, >od);
1269
1270 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x, TOD base: 0x%llx",
1271 gtod.epoch_idx, gtod.tod);
1272
1273 return 0;
1274 }
1275
kvm_s390_set_tod_high(struct kvm * kvm,struct kvm_device_attr * attr)1276 static int kvm_s390_set_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1277 {
1278 u8 gtod_high;
1279
1280 if (copy_from_user(>od_high, (void __user *)attr->addr,
1281 sizeof(gtod_high)))
1282 return -EFAULT;
1283
1284 if (gtod_high != 0)
1285 return -EINVAL;
1286 VM_EVENT(kvm, 3, "SET: TOD extension: 0x%x", gtod_high);
1287
1288 return 0;
1289 }
1290
kvm_s390_set_tod_low(struct kvm * kvm,struct kvm_device_attr * attr)1291 static int kvm_s390_set_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1292 {
1293 struct kvm_s390_vm_tod_clock gtod = { 0 };
1294
1295 if (copy_from_user(>od.tod, (void __user *)attr->addr,
1296 sizeof(gtod.tod)))
1297 return -EFAULT;
1298
1299 __kvm_s390_set_tod_clock(kvm, >od);
1300 VM_EVENT(kvm, 3, "SET: TOD base: 0x%llx", gtod.tod);
1301 return 0;
1302 }
1303
kvm_s390_set_tod(struct kvm * kvm,struct kvm_device_attr * attr)1304 static int kvm_s390_set_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1305 {
1306 int ret;
1307
1308 if (attr->flags)
1309 return -EINVAL;
1310
1311 mutex_lock(&kvm->lock);
1312 /*
1313 * For protected guests, the TOD is managed by the ultravisor, so trying
1314 * to change it will never bring the expected results.
1315 */
1316 if (kvm_s390_pv_is_protected(kvm)) {
1317 ret = -EOPNOTSUPP;
1318 goto out_unlock;
1319 }
1320
1321 switch (attr->attr) {
1322 case KVM_S390_VM_TOD_EXT:
1323 ret = kvm_s390_set_tod_ext(kvm, attr);
1324 break;
1325 case KVM_S390_VM_TOD_HIGH:
1326 ret = kvm_s390_set_tod_high(kvm, attr);
1327 break;
1328 case KVM_S390_VM_TOD_LOW:
1329 ret = kvm_s390_set_tod_low(kvm, attr);
1330 break;
1331 default:
1332 ret = -ENXIO;
1333 break;
1334 }
1335
1336 out_unlock:
1337 mutex_unlock(&kvm->lock);
1338 return ret;
1339 }
1340
kvm_s390_get_tod_clock(struct kvm * kvm,struct kvm_s390_vm_tod_clock * gtod)1341 static void kvm_s390_get_tod_clock(struct kvm *kvm,
1342 struct kvm_s390_vm_tod_clock *gtod)
1343 {
1344 union tod_clock clk;
1345
1346 preempt_disable();
1347
1348 store_tod_clock_ext(&clk);
1349
1350 gtod->tod = clk.tod + kvm->arch.epoch;
1351 gtod->epoch_idx = 0;
1352 if (test_kvm_facility(kvm, 139)) {
1353 gtod->epoch_idx = clk.ei + kvm->arch.epdx;
1354 if (gtod->tod < clk.tod)
1355 gtod->epoch_idx += 1;
1356 }
1357
1358 preempt_enable();
1359 }
1360
kvm_s390_get_tod_ext(struct kvm * kvm,struct kvm_device_attr * attr)1361 static int kvm_s390_get_tod_ext(struct kvm *kvm, struct kvm_device_attr *attr)
1362 {
1363 struct kvm_s390_vm_tod_clock gtod;
1364
1365 memset(>od, 0, sizeof(gtod));
1366 kvm_s390_get_tod_clock(kvm, >od);
1367 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
1368 return -EFAULT;
1369
1370 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x, TOD base: 0x%llx",
1371 gtod.epoch_idx, gtod.tod);
1372 return 0;
1373 }
1374
kvm_s390_get_tod_high(struct kvm * kvm,struct kvm_device_attr * attr)1375 static int kvm_s390_get_tod_high(struct kvm *kvm, struct kvm_device_attr *attr)
1376 {
1377 u8 gtod_high = 0;
1378
1379 if (copy_to_user((void __user *)attr->addr, >od_high,
1380 sizeof(gtod_high)))
1381 return -EFAULT;
1382 VM_EVENT(kvm, 3, "QUERY: TOD extension: 0x%x", gtod_high);
1383
1384 return 0;
1385 }
1386
kvm_s390_get_tod_low(struct kvm * kvm,struct kvm_device_attr * attr)1387 static int kvm_s390_get_tod_low(struct kvm *kvm, struct kvm_device_attr *attr)
1388 {
1389 u64 gtod;
1390
1391 gtod = kvm_s390_get_tod_clock_fast(kvm);
1392 if (copy_to_user((void __user *)attr->addr, >od, sizeof(gtod)))
1393 return -EFAULT;
1394 VM_EVENT(kvm, 3, "QUERY: TOD base: 0x%llx", gtod);
1395
1396 return 0;
1397 }
1398
kvm_s390_get_tod(struct kvm * kvm,struct kvm_device_attr * attr)1399 static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
1400 {
1401 int ret;
1402
1403 if (attr->flags)
1404 return -EINVAL;
1405
1406 switch (attr->attr) {
1407 case KVM_S390_VM_TOD_EXT:
1408 ret = kvm_s390_get_tod_ext(kvm, attr);
1409 break;
1410 case KVM_S390_VM_TOD_HIGH:
1411 ret = kvm_s390_get_tod_high(kvm, attr);
1412 break;
1413 case KVM_S390_VM_TOD_LOW:
1414 ret = kvm_s390_get_tod_low(kvm, attr);
1415 break;
1416 default:
1417 ret = -ENXIO;
1418 break;
1419 }
1420 return ret;
1421 }
1422
kvm_s390_set_processor(struct kvm * kvm,struct kvm_device_attr * attr)1423 static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1424 {
1425 struct kvm_s390_vm_cpu_processor *proc;
1426 u16 lowest_ibc, unblocked_ibc;
1427 int ret = 0;
1428
1429 mutex_lock(&kvm->lock);
1430 if (kvm->created_vcpus) {
1431 ret = -EBUSY;
1432 goto out;
1433 }
1434 proc = kzalloc(sizeof(*proc), GFP_KERNEL_ACCOUNT);
1435 if (!proc) {
1436 ret = -ENOMEM;
1437 goto out;
1438 }
1439 if (!copy_from_user(proc, (void __user *)attr->addr,
1440 sizeof(*proc))) {
1441 kvm->arch.model.cpuid = proc->cpuid;
1442 lowest_ibc = sclp.ibc >> 16 & 0xfff;
1443 unblocked_ibc = sclp.ibc & 0xfff;
1444 if (lowest_ibc && proc->ibc) {
1445 if (proc->ibc > unblocked_ibc)
1446 kvm->arch.model.ibc = unblocked_ibc;
1447 else if (proc->ibc < lowest_ibc)
1448 kvm->arch.model.ibc = lowest_ibc;
1449 else
1450 kvm->arch.model.ibc = proc->ibc;
1451 }
1452 memcpy(kvm->arch.model.fac_list, proc->fac_list,
1453 S390_ARCH_FAC_LIST_SIZE_BYTE);
1454 VM_EVENT(kvm, 3, "SET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1455 kvm->arch.model.ibc,
1456 kvm->arch.model.cpuid);
1457 VM_EVENT(kvm, 3, "SET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1458 kvm->arch.model.fac_list[0],
1459 kvm->arch.model.fac_list[1],
1460 kvm->arch.model.fac_list[2]);
1461 } else
1462 ret = -EFAULT;
1463 kfree(proc);
1464 out:
1465 mutex_unlock(&kvm->lock);
1466 return ret;
1467 }
1468
kvm_s390_set_processor_feat(struct kvm * kvm,struct kvm_device_attr * attr)1469 static int kvm_s390_set_processor_feat(struct kvm *kvm,
1470 struct kvm_device_attr *attr)
1471 {
1472 struct kvm_s390_vm_cpu_feat data;
1473
1474 if (copy_from_user(&data, (void __user *)attr->addr, sizeof(data)))
1475 return -EFAULT;
1476 if (!bitmap_subset((unsigned long *) data.feat,
1477 kvm_s390_available_cpu_feat,
1478 KVM_S390_VM_CPU_FEAT_NR_BITS))
1479 return -EINVAL;
1480
1481 mutex_lock(&kvm->lock);
1482 if (kvm->created_vcpus) {
1483 mutex_unlock(&kvm->lock);
1484 return -EBUSY;
1485 }
1486 bitmap_from_arr64(kvm->arch.cpu_feat, data.feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1487 mutex_unlock(&kvm->lock);
1488 VM_EVENT(kvm, 3, "SET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1489 data.feat[0],
1490 data.feat[1],
1491 data.feat[2]);
1492 return 0;
1493 }
1494
kvm_s390_set_processor_subfunc(struct kvm * kvm,struct kvm_device_attr * attr)1495 static int kvm_s390_set_processor_subfunc(struct kvm *kvm,
1496 struct kvm_device_attr *attr)
1497 {
1498 mutex_lock(&kvm->lock);
1499 if (kvm->created_vcpus) {
1500 mutex_unlock(&kvm->lock);
1501 return -EBUSY;
1502 }
1503
1504 if (copy_from_user(&kvm->arch.model.subfuncs, (void __user *)attr->addr,
1505 sizeof(struct kvm_s390_vm_cpu_subfunc))) {
1506 mutex_unlock(&kvm->lock);
1507 return -EFAULT;
1508 }
1509 mutex_unlock(&kvm->lock);
1510
1511 VM_EVENT(kvm, 3, "SET: guest PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1512 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0],
1513 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1],
1514 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2],
1515 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]);
1516 VM_EVENT(kvm, 3, "SET: guest PTFF subfunc 0x%16.16lx.%16.16lx",
1517 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0],
1518 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]);
1519 VM_EVENT(kvm, 3, "SET: guest KMAC subfunc 0x%16.16lx.%16.16lx",
1520 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0],
1521 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]);
1522 VM_EVENT(kvm, 3, "SET: guest KMC subfunc 0x%16.16lx.%16.16lx",
1523 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0],
1524 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]);
1525 VM_EVENT(kvm, 3, "SET: guest KM subfunc 0x%16.16lx.%16.16lx",
1526 ((unsigned long *) &kvm->arch.model.subfuncs.km)[0],
1527 ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]);
1528 VM_EVENT(kvm, 3, "SET: guest KIMD subfunc 0x%16.16lx.%16.16lx",
1529 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0],
1530 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]);
1531 VM_EVENT(kvm, 3, "SET: guest KLMD subfunc 0x%16.16lx.%16.16lx",
1532 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0],
1533 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]);
1534 VM_EVENT(kvm, 3, "SET: guest PCKMO subfunc 0x%16.16lx.%16.16lx",
1535 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0],
1536 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]);
1537 VM_EVENT(kvm, 3, "SET: guest KMCTR subfunc 0x%16.16lx.%16.16lx",
1538 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0],
1539 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]);
1540 VM_EVENT(kvm, 3, "SET: guest KMF subfunc 0x%16.16lx.%16.16lx",
1541 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0],
1542 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]);
1543 VM_EVENT(kvm, 3, "SET: guest KMO subfunc 0x%16.16lx.%16.16lx",
1544 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0],
1545 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]);
1546 VM_EVENT(kvm, 3, "SET: guest PCC subfunc 0x%16.16lx.%16.16lx",
1547 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0],
1548 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]);
1549 VM_EVENT(kvm, 3, "SET: guest PPNO subfunc 0x%16.16lx.%16.16lx",
1550 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0],
1551 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]);
1552 VM_EVENT(kvm, 3, "SET: guest KMA subfunc 0x%16.16lx.%16.16lx",
1553 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0],
1554 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]);
1555 VM_EVENT(kvm, 3, "SET: guest KDSA subfunc 0x%16.16lx.%16.16lx",
1556 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0],
1557 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]);
1558 VM_EVENT(kvm, 3, "SET: guest SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1559 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0],
1560 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1],
1561 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2],
1562 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]);
1563 VM_EVENT(kvm, 3, "SET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1564 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0],
1565 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1],
1566 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2],
1567 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]);
1568 VM_EVENT(kvm, 3, "GET: guest PFCR subfunc 0x%16.16lx.%16.16lx",
1569 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[0],
1570 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[1]);
1571
1572 return 0;
1573 }
1574
1575 #define KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK \
1576 ( \
1577 ((struct kvm_s390_vm_cpu_uv_feat){ \
1578 .ap = 1, \
1579 .ap_intr = 1, \
1580 }) \
1581 .feat \
1582 )
1583
kvm_s390_set_uv_feat(struct kvm * kvm,struct kvm_device_attr * attr)1584 static int kvm_s390_set_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1585 {
1586 struct kvm_s390_vm_cpu_uv_feat __user *ptr = (void __user *)attr->addr;
1587 unsigned long data, filter;
1588
1589 filter = uv_info.uv_feature_indications & KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK;
1590 if (get_user(data, &ptr->feat))
1591 return -EFAULT;
1592 if (!bitmap_subset(&data, &filter, KVM_S390_VM_CPU_UV_FEAT_NR_BITS))
1593 return -EINVAL;
1594
1595 mutex_lock(&kvm->lock);
1596 if (kvm->created_vcpus) {
1597 mutex_unlock(&kvm->lock);
1598 return -EBUSY;
1599 }
1600 kvm->arch.model.uv_feat_guest.feat = data;
1601 mutex_unlock(&kvm->lock);
1602
1603 VM_EVENT(kvm, 3, "SET: guest UV-feat: 0x%16.16lx", data);
1604
1605 return 0;
1606 }
1607
kvm_s390_set_cpu_model(struct kvm * kvm,struct kvm_device_attr * attr)1608 static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1609 {
1610 int ret = -ENXIO;
1611
1612 switch (attr->attr) {
1613 case KVM_S390_VM_CPU_PROCESSOR:
1614 ret = kvm_s390_set_processor(kvm, attr);
1615 break;
1616 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1617 ret = kvm_s390_set_processor_feat(kvm, attr);
1618 break;
1619 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1620 ret = kvm_s390_set_processor_subfunc(kvm, attr);
1621 break;
1622 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
1623 ret = kvm_s390_set_uv_feat(kvm, attr);
1624 break;
1625 }
1626 return ret;
1627 }
1628
kvm_s390_get_processor(struct kvm * kvm,struct kvm_device_attr * attr)1629 static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
1630 {
1631 struct kvm_s390_vm_cpu_processor *proc;
1632 int ret = 0;
1633
1634 proc = kzalloc(sizeof(*proc), GFP_KERNEL_ACCOUNT);
1635 if (!proc) {
1636 ret = -ENOMEM;
1637 goto out;
1638 }
1639 proc->cpuid = kvm->arch.model.cpuid;
1640 proc->ibc = kvm->arch.model.ibc;
1641 memcpy(&proc->fac_list, kvm->arch.model.fac_list,
1642 S390_ARCH_FAC_LIST_SIZE_BYTE);
1643 VM_EVENT(kvm, 3, "GET: guest ibc: 0x%4.4x, guest cpuid: 0x%16.16llx",
1644 kvm->arch.model.ibc,
1645 kvm->arch.model.cpuid);
1646 VM_EVENT(kvm, 3, "GET: guest faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1647 kvm->arch.model.fac_list[0],
1648 kvm->arch.model.fac_list[1],
1649 kvm->arch.model.fac_list[2]);
1650 if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
1651 ret = -EFAULT;
1652 kfree(proc);
1653 out:
1654 return ret;
1655 }
1656
kvm_s390_get_machine(struct kvm * kvm,struct kvm_device_attr * attr)1657 static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
1658 {
1659 struct kvm_s390_vm_cpu_machine *mach;
1660 int ret = 0;
1661
1662 mach = kzalloc(sizeof(*mach), GFP_KERNEL_ACCOUNT);
1663 if (!mach) {
1664 ret = -ENOMEM;
1665 goto out;
1666 }
1667 get_cpu_id((struct cpuid *) &mach->cpuid);
1668 mach->ibc = sclp.ibc;
1669 memcpy(&mach->fac_mask, kvm->arch.model.fac_mask,
1670 S390_ARCH_FAC_LIST_SIZE_BYTE);
1671 memcpy((unsigned long *)&mach->fac_list, stfle_fac_list,
1672 sizeof(stfle_fac_list));
1673 VM_EVENT(kvm, 3, "GET: host ibc: 0x%4.4x, host cpuid: 0x%16.16llx",
1674 kvm->arch.model.ibc,
1675 kvm->arch.model.cpuid);
1676 VM_EVENT(kvm, 3, "GET: host facmask: 0x%16.16llx.%16.16llx.%16.16llx",
1677 mach->fac_mask[0],
1678 mach->fac_mask[1],
1679 mach->fac_mask[2]);
1680 VM_EVENT(kvm, 3, "GET: host faclist: 0x%16.16llx.%16.16llx.%16.16llx",
1681 mach->fac_list[0],
1682 mach->fac_list[1],
1683 mach->fac_list[2]);
1684 if (copy_to_user((void __user *)attr->addr, mach, sizeof(*mach)))
1685 ret = -EFAULT;
1686 kfree(mach);
1687 out:
1688 return ret;
1689 }
1690
kvm_s390_get_processor_feat(struct kvm * kvm,struct kvm_device_attr * attr)1691 static int kvm_s390_get_processor_feat(struct kvm *kvm,
1692 struct kvm_device_attr *attr)
1693 {
1694 struct kvm_s390_vm_cpu_feat data;
1695
1696 bitmap_to_arr64(data.feat, kvm->arch.cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1697 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1698 return -EFAULT;
1699 VM_EVENT(kvm, 3, "GET: guest feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1700 data.feat[0],
1701 data.feat[1],
1702 data.feat[2]);
1703 return 0;
1704 }
1705
kvm_s390_get_machine_feat(struct kvm * kvm,struct kvm_device_attr * attr)1706 static int kvm_s390_get_machine_feat(struct kvm *kvm,
1707 struct kvm_device_attr *attr)
1708 {
1709 struct kvm_s390_vm_cpu_feat data;
1710
1711 bitmap_to_arr64(data.feat, kvm_s390_available_cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
1712 if (copy_to_user((void __user *)attr->addr, &data, sizeof(data)))
1713 return -EFAULT;
1714 VM_EVENT(kvm, 3, "GET: host feat: 0x%16.16llx.0x%16.16llx.0x%16.16llx",
1715 data.feat[0],
1716 data.feat[1],
1717 data.feat[2]);
1718 return 0;
1719 }
1720
kvm_s390_get_processor_subfunc(struct kvm * kvm,struct kvm_device_attr * attr)1721 static int kvm_s390_get_processor_subfunc(struct kvm *kvm,
1722 struct kvm_device_attr *attr)
1723 {
1724 if (copy_to_user((void __user *)attr->addr, &kvm->arch.model.subfuncs,
1725 sizeof(struct kvm_s390_vm_cpu_subfunc)))
1726 return -EFAULT;
1727
1728 VM_EVENT(kvm, 3, "GET: guest PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1729 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[0],
1730 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[1],
1731 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[2],
1732 ((unsigned long *) &kvm->arch.model.subfuncs.plo)[3]);
1733 VM_EVENT(kvm, 3, "GET: guest PTFF subfunc 0x%16.16lx.%16.16lx",
1734 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[0],
1735 ((unsigned long *) &kvm->arch.model.subfuncs.ptff)[1]);
1736 VM_EVENT(kvm, 3, "GET: guest KMAC subfunc 0x%16.16lx.%16.16lx",
1737 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[0],
1738 ((unsigned long *) &kvm->arch.model.subfuncs.kmac)[1]);
1739 VM_EVENT(kvm, 3, "GET: guest KMC subfunc 0x%16.16lx.%16.16lx",
1740 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[0],
1741 ((unsigned long *) &kvm->arch.model.subfuncs.kmc)[1]);
1742 VM_EVENT(kvm, 3, "GET: guest KM subfunc 0x%16.16lx.%16.16lx",
1743 ((unsigned long *) &kvm->arch.model.subfuncs.km)[0],
1744 ((unsigned long *) &kvm->arch.model.subfuncs.km)[1]);
1745 VM_EVENT(kvm, 3, "GET: guest KIMD subfunc 0x%16.16lx.%16.16lx",
1746 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[0],
1747 ((unsigned long *) &kvm->arch.model.subfuncs.kimd)[1]);
1748 VM_EVENT(kvm, 3, "GET: guest KLMD subfunc 0x%16.16lx.%16.16lx",
1749 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[0],
1750 ((unsigned long *) &kvm->arch.model.subfuncs.klmd)[1]);
1751 VM_EVENT(kvm, 3, "GET: guest PCKMO subfunc 0x%16.16lx.%16.16lx",
1752 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[0],
1753 ((unsigned long *) &kvm->arch.model.subfuncs.pckmo)[1]);
1754 VM_EVENT(kvm, 3, "GET: guest KMCTR subfunc 0x%16.16lx.%16.16lx",
1755 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[0],
1756 ((unsigned long *) &kvm->arch.model.subfuncs.kmctr)[1]);
1757 VM_EVENT(kvm, 3, "GET: guest KMF subfunc 0x%16.16lx.%16.16lx",
1758 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[0],
1759 ((unsigned long *) &kvm->arch.model.subfuncs.kmf)[1]);
1760 VM_EVENT(kvm, 3, "GET: guest KMO subfunc 0x%16.16lx.%16.16lx",
1761 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[0],
1762 ((unsigned long *) &kvm->arch.model.subfuncs.kmo)[1]);
1763 VM_EVENT(kvm, 3, "GET: guest PCC subfunc 0x%16.16lx.%16.16lx",
1764 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[0],
1765 ((unsigned long *) &kvm->arch.model.subfuncs.pcc)[1]);
1766 VM_EVENT(kvm, 3, "GET: guest PPNO subfunc 0x%16.16lx.%16.16lx",
1767 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[0],
1768 ((unsigned long *) &kvm->arch.model.subfuncs.ppno)[1]);
1769 VM_EVENT(kvm, 3, "GET: guest KMA subfunc 0x%16.16lx.%16.16lx",
1770 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[0],
1771 ((unsigned long *) &kvm->arch.model.subfuncs.kma)[1]);
1772 VM_EVENT(kvm, 3, "GET: guest KDSA subfunc 0x%16.16lx.%16.16lx",
1773 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[0],
1774 ((unsigned long *) &kvm->arch.model.subfuncs.kdsa)[1]);
1775 VM_EVENT(kvm, 3, "GET: guest SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1776 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[0],
1777 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[1],
1778 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[2],
1779 ((unsigned long *) &kvm->arch.model.subfuncs.sortl)[3]);
1780 VM_EVENT(kvm, 3, "GET: guest DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1781 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[0],
1782 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[1],
1783 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[2],
1784 ((unsigned long *) &kvm->arch.model.subfuncs.dfltcc)[3]);
1785 VM_EVENT(kvm, 3, "GET: guest PFCR subfunc 0x%16.16lx.%16.16lx",
1786 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[0],
1787 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[1]);
1788
1789 return 0;
1790 }
1791
kvm_s390_get_machine_subfunc(struct kvm * kvm,struct kvm_device_attr * attr)1792 static int kvm_s390_get_machine_subfunc(struct kvm *kvm,
1793 struct kvm_device_attr *attr)
1794 {
1795 if (copy_to_user((void __user *)attr->addr, &kvm_s390_available_subfunc,
1796 sizeof(struct kvm_s390_vm_cpu_subfunc)))
1797 return -EFAULT;
1798
1799 VM_EVENT(kvm, 3, "GET: host PLO subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1800 ((unsigned long *) &kvm_s390_available_subfunc.plo)[0],
1801 ((unsigned long *) &kvm_s390_available_subfunc.plo)[1],
1802 ((unsigned long *) &kvm_s390_available_subfunc.plo)[2],
1803 ((unsigned long *) &kvm_s390_available_subfunc.plo)[3]);
1804 VM_EVENT(kvm, 3, "GET: host PTFF subfunc 0x%16.16lx.%16.16lx",
1805 ((unsigned long *) &kvm_s390_available_subfunc.ptff)[0],
1806 ((unsigned long *) &kvm_s390_available_subfunc.ptff)[1]);
1807 VM_EVENT(kvm, 3, "GET: host KMAC subfunc 0x%16.16lx.%16.16lx",
1808 ((unsigned long *) &kvm_s390_available_subfunc.kmac)[0],
1809 ((unsigned long *) &kvm_s390_available_subfunc.kmac)[1]);
1810 VM_EVENT(kvm, 3, "GET: host KMC subfunc 0x%16.16lx.%16.16lx",
1811 ((unsigned long *) &kvm_s390_available_subfunc.kmc)[0],
1812 ((unsigned long *) &kvm_s390_available_subfunc.kmc)[1]);
1813 VM_EVENT(kvm, 3, "GET: host KM subfunc 0x%16.16lx.%16.16lx",
1814 ((unsigned long *) &kvm_s390_available_subfunc.km)[0],
1815 ((unsigned long *) &kvm_s390_available_subfunc.km)[1]);
1816 VM_EVENT(kvm, 3, "GET: host KIMD subfunc 0x%16.16lx.%16.16lx",
1817 ((unsigned long *) &kvm_s390_available_subfunc.kimd)[0],
1818 ((unsigned long *) &kvm_s390_available_subfunc.kimd)[1]);
1819 VM_EVENT(kvm, 3, "GET: host KLMD subfunc 0x%16.16lx.%16.16lx",
1820 ((unsigned long *) &kvm_s390_available_subfunc.klmd)[0],
1821 ((unsigned long *) &kvm_s390_available_subfunc.klmd)[1]);
1822 VM_EVENT(kvm, 3, "GET: host PCKMO subfunc 0x%16.16lx.%16.16lx",
1823 ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[0],
1824 ((unsigned long *) &kvm_s390_available_subfunc.pckmo)[1]);
1825 VM_EVENT(kvm, 3, "GET: host KMCTR subfunc 0x%16.16lx.%16.16lx",
1826 ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[0],
1827 ((unsigned long *) &kvm_s390_available_subfunc.kmctr)[1]);
1828 VM_EVENT(kvm, 3, "GET: host KMF subfunc 0x%16.16lx.%16.16lx",
1829 ((unsigned long *) &kvm_s390_available_subfunc.kmf)[0],
1830 ((unsigned long *) &kvm_s390_available_subfunc.kmf)[1]);
1831 VM_EVENT(kvm, 3, "GET: host KMO subfunc 0x%16.16lx.%16.16lx",
1832 ((unsigned long *) &kvm_s390_available_subfunc.kmo)[0],
1833 ((unsigned long *) &kvm_s390_available_subfunc.kmo)[1]);
1834 VM_EVENT(kvm, 3, "GET: host PCC subfunc 0x%16.16lx.%16.16lx",
1835 ((unsigned long *) &kvm_s390_available_subfunc.pcc)[0],
1836 ((unsigned long *) &kvm_s390_available_subfunc.pcc)[1]);
1837 VM_EVENT(kvm, 3, "GET: host PPNO subfunc 0x%16.16lx.%16.16lx",
1838 ((unsigned long *) &kvm_s390_available_subfunc.ppno)[0],
1839 ((unsigned long *) &kvm_s390_available_subfunc.ppno)[1]);
1840 VM_EVENT(kvm, 3, "GET: host KMA subfunc 0x%16.16lx.%16.16lx",
1841 ((unsigned long *) &kvm_s390_available_subfunc.kma)[0],
1842 ((unsigned long *) &kvm_s390_available_subfunc.kma)[1]);
1843 VM_EVENT(kvm, 3, "GET: host KDSA subfunc 0x%16.16lx.%16.16lx",
1844 ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[0],
1845 ((unsigned long *) &kvm_s390_available_subfunc.kdsa)[1]);
1846 VM_EVENT(kvm, 3, "GET: host SORTL subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1847 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[0],
1848 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[1],
1849 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[2],
1850 ((unsigned long *) &kvm_s390_available_subfunc.sortl)[3]);
1851 VM_EVENT(kvm, 3, "GET: host DFLTCC subfunc 0x%16.16lx.%16.16lx.%16.16lx.%16.16lx",
1852 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[0],
1853 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[1],
1854 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[2],
1855 ((unsigned long *) &kvm_s390_available_subfunc.dfltcc)[3]);
1856 VM_EVENT(kvm, 3, "GET: host PFCR subfunc 0x%16.16lx.%16.16lx",
1857 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[0],
1858 ((unsigned long *) &kvm_s390_available_subfunc.pfcr)[1]);
1859
1860 return 0;
1861 }
1862
kvm_s390_get_processor_uv_feat(struct kvm * kvm,struct kvm_device_attr * attr)1863 static int kvm_s390_get_processor_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1864 {
1865 struct kvm_s390_vm_cpu_uv_feat __user *dst = (void __user *)attr->addr;
1866 unsigned long feat = kvm->arch.model.uv_feat_guest.feat;
1867
1868 if (put_user(feat, &dst->feat))
1869 return -EFAULT;
1870 VM_EVENT(kvm, 3, "GET: guest UV-feat: 0x%16.16lx", feat);
1871
1872 return 0;
1873 }
1874
kvm_s390_get_machine_uv_feat(struct kvm * kvm,struct kvm_device_attr * attr)1875 static int kvm_s390_get_machine_uv_feat(struct kvm *kvm, struct kvm_device_attr *attr)
1876 {
1877 struct kvm_s390_vm_cpu_uv_feat __user *dst = (void __user *)attr->addr;
1878 unsigned long feat;
1879
1880 BUILD_BUG_ON(sizeof(*dst) != sizeof(uv_info.uv_feature_indications));
1881
1882 feat = uv_info.uv_feature_indications & KVM_S390_VM_CPU_UV_FEAT_GUEST_MASK;
1883 if (put_user(feat, &dst->feat))
1884 return -EFAULT;
1885 VM_EVENT(kvm, 3, "GET: guest UV-feat: 0x%16.16lx", feat);
1886
1887 return 0;
1888 }
1889
kvm_s390_get_cpu_model(struct kvm * kvm,struct kvm_device_attr * attr)1890 static int kvm_s390_get_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
1891 {
1892 int ret = -ENXIO;
1893
1894 switch (attr->attr) {
1895 case KVM_S390_VM_CPU_PROCESSOR:
1896 ret = kvm_s390_get_processor(kvm, attr);
1897 break;
1898 case KVM_S390_VM_CPU_MACHINE:
1899 ret = kvm_s390_get_machine(kvm, attr);
1900 break;
1901 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
1902 ret = kvm_s390_get_processor_feat(kvm, attr);
1903 break;
1904 case KVM_S390_VM_CPU_MACHINE_FEAT:
1905 ret = kvm_s390_get_machine_feat(kvm, attr);
1906 break;
1907 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
1908 ret = kvm_s390_get_processor_subfunc(kvm, attr);
1909 break;
1910 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
1911 ret = kvm_s390_get_machine_subfunc(kvm, attr);
1912 break;
1913 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
1914 ret = kvm_s390_get_processor_uv_feat(kvm, attr);
1915 break;
1916 case KVM_S390_VM_CPU_MACHINE_UV_FEAT_GUEST:
1917 ret = kvm_s390_get_machine_uv_feat(kvm, attr);
1918 break;
1919 }
1920 return ret;
1921 }
1922
1923 /**
1924 * kvm_s390_update_topology_change_report - update CPU topology change report
1925 * @kvm: guest KVM description
1926 * @val: set or clear the MTCR bit
1927 *
1928 * Updates the Multiprocessor Topology-Change-Report bit to signal
1929 * the guest with a topology change.
1930 * This is only relevant if the topology facility is present.
1931 *
1932 * The SCA version, bsca or esca, doesn't matter as offset is the same.
1933 */
kvm_s390_update_topology_change_report(struct kvm * kvm,bool val)1934 static void kvm_s390_update_topology_change_report(struct kvm *kvm, bool val)
1935 {
1936 union sca_utility new, old;
1937 struct bsca_block *sca;
1938
1939 read_lock(&kvm->arch.sca_lock);
1940 sca = kvm->arch.sca;
1941 old = READ_ONCE(sca->utility);
1942 do {
1943 new = old;
1944 new.mtcr = val;
1945 } while (!try_cmpxchg(&sca->utility.val, &old.val, new.val));
1946 read_unlock(&kvm->arch.sca_lock);
1947 }
1948
kvm_s390_set_topo_change_indication(struct kvm * kvm,struct kvm_device_attr * attr)1949 static int kvm_s390_set_topo_change_indication(struct kvm *kvm,
1950 struct kvm_device_attr *attr)
1951 {
1952 if (!test_kvm_facility(kvm, 11))
1953 return -ENXIO;
1954
1955 kvm_s390_update_topology_change_report(kvm, !!attr->attr);
1956 return 0;
1957 }
1958
kvm_s390_get_topo_change_indication(struct kvm * kvm,struct kvm_device_attr * attr)1959 static int kvm_s390_get_topo_change_indication(struct kvm *kvm,
1960 struct kvm_device_attr *attr)
1961 {
1962 u8 topo;
1963
1964 if (!test_kvm_facility(kvm, 11))
1965 return -ENXIO;
1966
1967 read_lock(&kvm->arch.sca_lock);
1968 topo = ((struct bsca_block *)kvm->arch.sca)->utility.mtcr;
1969 read_unlock(&kvm->arch.sca_lock);
1970
1971 return put_user(topo, (u8 __user *)attr->addr);
1972 }
1973
kvm_s390_vm_set_attr(struct kvm * kvm,struct kvm_device_attr * attr)1974 static int kvm_s390_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1975 {
1976 int ret;
1977
1978 switch (attr->group) {
1979 case KVM_S390_VM_MEM_CTRL:
1980 ret = kvm_s390_set_mem_control(kvm, attr);
1981 break;
1982 case KVM_S390_VM_TOD:
1983 ret = kvm_s390_set_tod(kvm, attr);
1984 break;
1985 case KVM_S390_VM_CPU_MODEL:
1986 ret = kvm_s390_set_cpu_model(kvm, attr);
1987 break;
1988 case KVM_S390_VM_CRYPTO:
1989 ret = kvm_s390_vm_set_crypto(kvm, attr);
1990 break;
1991 case KVM_S390_VM_MIGRATION:
1992 ret = kvm_s390_vm_set_migration(kvm, attr);
1993 break;
1994 case KVM_S390_VM_CPU_TOPOLOGY:
1995 ret = kvm_s390_set_topo_change_indication(kvm, attr);
1996 break;
1997 default:
1998 ret = -ENXIO;
1999 break;
2000 }
2001
2002 return ret;
2003 }
2004
kvm_s390_vm_get_attr(struct kvm * kvm,struct kvm_device_attr * attr)2005 static int kvm_s390_vm_get_attr(struct kvm *kvm, struct kvm_device_attr *attr)
2006 {
2007 int ret;
2008
2009 switch (attr->group) {
2010 case KVM_S390_VM_MEM_CTRL:
2011 ret = kvm_s390_get_mem_control(kvm, attr);
2012 break;
2013 case KVM_S390_VM_TOD:
2014 ret = kvm_s390_get_tod(kvm, attr);
2015 break;
2016 case KVM_S390_VM_CPU_MODEL:
2017 ret = kvm_s390_get_cpu_model(kvm, attr);
2018 break;
2019 case KVM_S390_VM_MIGRATION:
2020 ret = kvm_s390_vm_get_migration(kvm, attr);
2021 break;
2022 case KVM_S390_VM_CPU_TOPOLOGY:
2023 ret = kvm_s390_get_topo_change_indication(kvm, attr);
2024 break;
2025 default:
2026 ret = -ENXIO;
2027 break;
2028 }
2029
2030 return ret;
2031 }
2032
kvm_s390_vm_has_attr(struct kvm * kvm,struct kvm_device_attr * attr)2033 static int kvm_s390_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
2034 {
2035 int ret;
2036
2037 switch (attr->group) {
2038 case KVM_S390_VM_MEM_CTRL:
2039 switch (attr->attr) {
2040 case KVM_S390_VM_MEM_ENABLE_CMMA:
2041 case KVM_S390_VM_MEM_CLR_CMMA:
2042 ret = sclp.has_cmma ? 0 : -ENXIO;
2043 break;
2044 case KVM_S390_VM_MEM_LIMIT_SIZE:
2045 ret = 0;
2046 break;
2047 default:
2048 ret = -ENXIO;
2049 break;
2050 }
2051 break;
2052 case KVM_S390_VM_TOD:
2053 switch (attr->attr) {
2054 case KVM_S390_VM_TOD_LOW:
2055 case KVM_S390_VM_TOD_HIGH:
2056 ret = 0;
2057 break;
2058 default:
2059 ret = -ENXIO;
2060 break;
2061 }
2062 break;
2063 case KVM_S390_VM_CPU_MODEL:
2064 switch (attr->attr) {
2065 case KVM_S390_VM_CPU_PROCESSOR:
2066 case KVM_S390_VM_CPU_MACHINE:
2067 case KVM_S390_VM_CPU_PROCESSOR_FEAT:
2068 case KVM_S390_VM_CPU_MACHINE_FEAT:
2069 case KVM_S390_VM_CPU_MACHINE_SUBFUNC:
2070 case KVM_S390_VM_CPU_PROCESSOR_SUBFUNC:
2071 case KVM_S390_VM_CPU_MACHINE_UV_FEAT_GUEST:
2072 case KVM_S390_VM_CPU_PROCESSOR_UV_FEAT_GUEST:
2073 ret = 0;
2074 break;
2075 default:
2076 ret = -ENXIO;
2077 break;
2078 }
2079 break;
2080 case KVM_S390_VM_CRYPTO:
2081 switch (attr->attr) {
2082 case KVM_S390_VM_CRYPTO_ENABLE_AES_KW:
2083 case KVM_S390_VM_CRYPTO_ENABLE_DEA_KW:
2084 case KVM_S390_VM_CRYPTO_DISABLE_AES_KW:
2085 case KVM_S390_VM_CRYPTO_DISABLE_DEA_KW:
2086 ret = 0;
2087 break;
2088 case KVM_S390_VM_CRYPTO_ENABLE_APIE:
2089 case KVM_S390_VM_CRYPTO_DISABLE_APIE:
2090 ret = ap_instructions_available() ? 0 : -ENXIO;
2091 break;
2092 default:
2093 ret = -ENXIO;
2094 break;
2095 }
2096 break;
2097 case KVM_S390_VM_MIGRATION:
2098 ret = 0;
2099 break;
2100 case KVM_S390_VM_CPU_TOPOLOGY:
2101 ret = test_kvm_facility(kvm, 11) ? 0 : -ENXIO;
2102 break;
2103 default:
2104 ret = -ENXIO;
2105 break;
2106 }
2107
2108 return ret;
2109 }
2110
kvm_s390_get_skeys(struct kvm * kvm,struct kvm_s390_skeys * args)2111 static int kvm_s390_get_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
2112 {
2113 uint8_t *keys;
2114 uint64_t hva;
2115 int srcu_idx, i, r = 0;
2116
2117 if (args->flags != 0)
2118 return -EINVAL;
2119
2120 /* Is this guest using storage keys? */
2121 if (!mm_uses_skeys(current->mm))
2122 return KVM_S390_GET_SKEYS_NONE;
2123
2124 /* Enforce sane limit on memory allocation */
2125 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
2126 return -EINVAL;
2127
2128 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL_ACCOUNT);
2129 if (!keys)
2130 return -ENOMEM;
2131
2132 mmap_read_lock(current->mm);
2133 srcu_idx = srcu_read_lock(&kvm->srcu);
2134 for (i = 0; i < args->count; i++) {
2135 hva = gfn_to_hva(kvm, args->start_gfn + i);
2136 if (kvm_is_error_hva(hva)) {
2137 r = -EFAULT;
2138 break;
2139 }
2140
2141 r = get_guest_storage_key(current->mm, hva, &keys[i]);
2142 if (r)
2143 break;
2144 }
2145 srcu_read_unlock(&kvm->srcu, srcu_idx);
2146 mmap_read_unlock(current->mm);
2147
2148 if (!r) {
2149 r = copy_to_user((uint8_t __user *)args->skeydata_addr, keys,
2150 sizeof(uint8_t) * args->count);
2151 if (r)
2152 r = -EFAULT;
2153 }
2154
2155 kvfree(keys);
2156 return r;
2157 }
2158
kvm_s390_set_skeys(struct kvm * kvm,struct kvm_s390_skeys * args)2159 static int kvm_s390_set_skeys(struct kvm *kvm, struct kvm_s390_skeys *args)
2160 {
2161 uint8_t *keys;
2162 uint64_t hva;
2163 int srcu_idx, i, r = 0;
2164 bool unlocked;
2165
2166 if (args->flags != 0)
2167 return -EINVAL;
2168
2169 /* Enforce sane limit on memory allocation */
2170 if (args->count < 1 || args->count > KVM_S390_SKEYS_MAX)
2171 return -EINVAL;
2172
2173 keys = kvmalloc_array(args->count, sizeof(uint8_t), GFP_KERNEL_ACCOUNT);
2174 if (!keys)
2175 return -ENOMEM;
2176
2177 r = copy_from_user(keys, (uint8_t __user *)args->skeydata_addr,
2178 sizeof(uint8_t) * args->count);
2179 if (r) {
2180 r = -EFAULT;
2181 goto out;
2182 }
2183
2184 /* Enable storage key handling for the guest */
2185 r = s390_enable_skey();
2186 if (r)
2187 goto out;
2188
2189 i = 0;
2190 mmap_read_lock(current->mm);
2191 srcu_idx = srcu_read_lock(&kvm->srcu);
2192 while (i < args->count) {
2193 unlocked = false;
2194 hva = gfn_to_hva(kvm, args->start_gfn + i);
2195 if (kvm_is_error_hva(hva)) {
2196 r = -EFAULT;
2197 break;
2198 }
2199
2200 /* Lowest order bit is reserved */
2201 if (keys[i] & 0x01) {
2202 r = -EINVAL;
2203 break;
2204 }
2205
2206 r = set_guest_storage_key(current->mm, hva, keys[i], 0);
2207 if (r) {
2208 r = fixup_user_fault(current->mm, hva,
2209 FAULT_FLAG_WRITE, &unlocked);
2210 if (r)
2211 break;
2212 }
2213 if (!r)
2214 i++;
2215 }
2216 srcu_read_unlock(&kvm->srcu, srcu_idx);
2217 mmap_read_unlock(current->mm);
2218 out:
2219 kvfree(keys);
2220 return r;
2221 }
2222
2223 /*
2224 * Base address and length must be sent at the start of each block, therefore
2225 * it's cheaper to send some clean data, as long as it's less than the size of
2226 * two longs.
2227 */
2228 #define KVM_S390_MAX_BIT_DISTANCE (2 * sizeof(void *))
2229 /* for consistency */
2230 #define KVM_S390_CMMA_SIZE_MAX ((u32)KVM_S390_SKEYS_MAX)
2231
kvm_s390_peek_cmma(struct kvm * kvm,struct kvm_s390_cmma_log * args,u8 * res,unsigned long bufsize)2232 static int kvm_s390_peek_cmma(struct kvm *kvm, struct kvm_s390_cmma_log *args,
2233 u8 *res, unsigned long bufsize)
2234 {
2235 unsigned long pgstev, hva, cur_gfn = args->start_gfn;
2236
2237 args->count = 0;
2238 while (args->count < bufsize) {
2239 hva = gfn_to_hva(kvm, cur_gfn);
2240 /*
2241 * We return an error if the first value was invalid, but we
2242 * return successfully if at least one value was copied.
2243 */
2244 if (kvm_is_error_hva(hva))
2245 return args->count ? 0 : -EFAULT;
2246 if (get_pgste(kvm->mm, hva, &pgstev) < 0)
2247 pgstev = 0;
2248 res[args->count++] = (pgstev >> 24) & 0x43;
2249 cur_gfn++;
2250 }
2251
2252 return 0;
2253 }
2254
gfn_to_memslot_approx(struct kvm_memslots * slots,gfn_t gfn)2255 static struct kvm_memory_slot *gfn_to_memslot_approx(struct kvm_memslots *slots,
2256 gfn_t gfn)
2257 {
2258 return ____gfn_to_memslot(slots, gfn, true);
2259 }
2260
kvm_s390_next_dirty_cmma(struct kvm_memslots * slots,unsigned long cur_gfn)2261 static unsigned long kvm_s390_next_dirty_cmma(struct kvm_memslots *slots,
2262 unsigned long cur_gfn)
2263 {
2264 struct kvm_memory_slot *ms = gfn_to_memslot_approx(slots, cur_gfn);
2265 unsigned long ofs = cur_gfn - ms->base_gfn;
2266 struct rb_node *mnode = &ms->gfn_node[slots->node_idx];
2267
2268 if (ms->base_gfn + ms->npages <= cur_gfn) {
2269 mnode = rb_next(mnode);
2270 /* If we are above the highest slot, wrap around */
2271 if (!mnode)
2272 mnode = rb_first(&slots->gfn_tree);
2273
2274 ms = container_of(mnode, struct kvm_memory_slot, gfn_node[slots->node_idx]);
2275 ofs = 0;
2276 }
2277
2278 if (cur_gfn < ms->base_gfn)
2279 ofs = 0;
2280
2281 ofs = find_next_bit(kvm_second_dirty_bitmap(ms), ms->npages, ofs);
2282 while (ofs >= ms->npages && (mnode = rb_next(mnode))) {
2283 ms = container_of(mnode, struct kvm_memory_slot, gfn_node[slots->node_idx]);
2284 ofs = find_first_bit(kvm_second_dirty_bitmap(ms), ms->npages);
2285 }
2286 return ms->base_gfn + ofs;
2287 }
2288
kvm_s390_get_cmma(struct kvm * kvm,struct kvm_s390_cmma_log * args,u8 * res,unsigned long bufsize)2289 static int kvm_s390_get_cmma(struct kvm *kvm, struct kvm_s390_cmma_log *args,
2290 u8 *res, unsigned long bufsize)
2291 {
2292 unsigned long mem_end, cur_gfn, next_gfn, hva, pgstev;
2293 struct kvm_memslots *slots = kvm_memslots(kvm);
2294 struct kvm_memory_slot *ms;
2295
2296 if (unlikely(kvm_memslots_empty(slots)))
2297 return 0;
2298
2299 cur_gfn = kvm_s390_next_dirty_cmma(slots, args->start_gfn);
2300 ms = gfn_to_memslot(kvm, cur_gfn);
2301 args->count = 0;
2302 args->start_gfn = cur_gfn;
2303 if (!ms)
2304 return 0;
2305 next_gfn = kvm_s390_next_dirty_cmma(slots, cur_gfn + 1);
2306 mem_end = kvm_s390_get_gfn_end(slots);
2307
2308 while (args->count < bufsize) {
2309 hva = gfn_to_hva(kvm, cur_gfn);
2310 if (kvm_is_error_hva(hva))
2311 return 0;
2312 /* Decrement only if we actually flipped the bit to 0 */
2313 if (test_and_clear_bit(cur_gfn - ms->base_gfn, kvm_second_dirty_bitmap(ms)))
2314 atomic64_dec(&kvm->arch.cmma_dirty_pages);
2315 if (get_pgste(kvm->mm, hva, &pgstev) < 0)
2316 pgstev = 0;
2317 /* Save the value */
2318 res[args->count++] = (pgstev >> 24) & 0x43;
2319 /* If the next bit is too far away, stop. */
2320 if (next_gfn > cur_gfn + KVM_S390_MAX_BIT_DISTANCE)
2321 return 0;
2322 /* If we reached the previous "next", find the next one */
2323 if (cur_gfn == next_gfn)
2324 next_gfn = kvm_s390_next_dirty_cmma(slots, cur_gfn + 1);
2325 /* Reached the end of memory or of the buffer, stop */
2326 if ((next_gfn >= mem_end) ||
2327 (next_gfn - args->start_gfn >= bufsize))
2328 return 0;
2329 cur_gfn++;
2330 /* Reached the end of the current memslot, take the next one. */
2331 if (cur_gfn - ms->base_gfn >= ms->npages) {
2332 ms = gfn_to_memslot(kvm, cur_gfn);
2333 if (!ms)
2334 return 0;
2335 }
2336 }
2337 return 0;
2338 }
2339
2340 /*
2341 * This function searches for the next page with dirty CMMA attributes, and
2342 * saves the attributes in the buffer up to either the end of the buffer or
2343 * until a block of at least KVM_S390_MAX_BIT_DISTANCE clean bits is found;
2344 * no trailing clean bytes are saved.
2345 * In case no dirty bits were found, or if CMMA was not enabled or used, the
2346 * output buffer will indicate 0 as length.
2347 */
kvm_s390_get_cmma_bits(struct kvm * kvm,struct kvm_s390_cmma_log * args)2348 static int kvm_s390_get_cmma_bits(struct kvm *kvm,
2349 struct kvm_s390_cmma_log *args)
2350 {
2351 unsigned long bufsize;
2352 int srcu_idx, peek, ret;
2353 u8 *values;
2354
2355 if (!kvm->arch.use_cmma)
2356 return -ENXIO;
2357 /* Invalid/unsupported flags were specified */
2358 if (args->flags & ~KVM_S390_CMMA_PEEK)
2359 return -EINVAL;
2360 /* Migration mode query, and we are not doing a migration */
2361 peek = !!(args->flags & KVM_S390_CMMA_PEEK);
2362 if (!peek && !kvm->arch.migration_mode)
2363 return -EINVAL;
2364 /* CMMA is disabled or was not used, or the buffer has length zero */
2365 bufsize = min(args->count, KVM_S390_CMMA_SIZE_MAX);
2366 if (!bufsize || !kvm->mm->context.uses_cmm) {
2367 memset(args, 0, sizeof(*args));
2368 return 0;
2369 }
2370 /* We are not peeking, and there are no dirty pages */
2371 if (!peek && !atomic64_read(&kvm->arch.cmma_dirty_pages)) {
2372 memset(args, 0, sizeof(*args));
2373 return 0;
2374 }
2375
2376 values = vmalloc(bufsize);
2377 if (!values)
2378 return -ENOMEM;
2379
2380 mmap_read_lock(kvm->mm);
2381 srcu_idx = srcu_read_lock(&kvm->srcu);
2382 if (peek)
2383 ret = kvm_s390_peek_cmma(kvm, args, values, bufsize);
2384 else
2385 ret = kvm_s390_get_cmma(kvm, args, values, bufsize);
2386 srcu_read_unlock(&kvm->srcu, srcu_idx);
2387 mmap_read_unlock(kvm->mm);
2388
2389 if (kvm->arch.migration_mode)
2390 args->remaining = atomic64_read(&kvm->arch.cmma_dirty_pages);
2391 else
2392 args->remaining = 0;
2393
2394 if (copy_to_user((void __user *)args->values, values, args->count))
2395 ret = -EFAULT;
2396
2397 vfree(values);
2398 return ret;
2399 }
2400
2401 /*
2402 * This function sets the CMMA attributes for the given pages. If the input
2403 * buffer has zero length, no action is taken, otherwise the attributes are
2404 * set and the mm->context.uses_cmm flag is set.
2405 */
kvm_s390_set_cmma_bits(struct kvm * kvm,const struct kvm_s390_cmma_log * args)2406 static int kvm_s390_set_cmma_bits(struct kvm *kvm,
2407 const struct kvm_s390_cmma_log *args)
2408 {
2409 unsigned long hva, mask, pgstev, i;
2410 uint8_t *bits;
2411 int srcu_idx, r = 0;
2412
2413 mask = args->mask;
2414
2415 if (!kvm->arch.use_cmma)
2416 return -ENXIO;
2417 /* invalid/unsupported flags */
2418 if (args->flags != 0)
2419 return -EINVAL;
2420 /* Enforce sane limit on memory allocation */
2421 if (args->count > KVM_S390_CMMA_SIZE_MAX)
2422 return -EINVAL;
2423 /* Nothing to do */
2424 if (args->count == 0)
2425 return 0;
2426
2427 bits = vmalloc(array_size(sizeof(*bits), args->count));
2428 if (!bits)
2429 return -ENOMEM;
2430
2431 r = copy_from_user(bits, (void __user *)args->values, args->count);
2432 if (r) {
2433 r = -EFAULT;
2434 goto out;
2435 }
2436
2437 mmap_read_lock(kvm->mm);
2438 srcu_idx = srcu_read_lock(&kvm->srcu);
2439 for (i = 0; i < args->count; i++) {
2440 hva = gfn_to_hva(kvm, args->start_gfn + i);
2441 if (kvm_is_error_hva(hva)) {
2442 r = -EFAULT;
2443 break;
2444 }
2445
2446 pgstev = bits[i];
2447 pgstev = pgstev << 24;
2448 mask &= _PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT;
2449 set_pgste_bits(kvm->mm, hva, mask, pgstev);
2450 }
2451 srcu_read_unlock(&kvm->srcu, srcu_idx);
2452 mmap_read_unlock(kvm->mm);
2453
2454 if (!kvm->mm->context.uses_cmm) {
2455 mmap_write_lock(kvm->mm);
2456 kvm->mm->context.uses_cmm = 1;
2457 mmap_write_unlock(kvm->mm);
2458 }
2459 out:
2460 vfree(bits);
2461 return r;
2462 }
2463
2464 /**
2465 * kvm_s390_cpus_from_pv - Convert all protected vCPUs in a protected VM to
2466 * non protected.
2467 * @kvm: the VM whose protected vCPUs are to be converted
2468 * @rc: return value for the RC field of the UVC (in case of error)
2469 * @rrc: return value for the RRC field of the UVC (in case of error)
2470 *
2471 * Does not stop in case of error, tries to convert as many
2472 * CPUs as possible. In case of error, the RC and RRC of the last error are
2473 * returned.
2474 *
2475 * Return: 0 in case of success, otherwise -EIO
2476 */
kvm_s390_cpus_from_pv(struct kvm * kvm,u16 * rc,u16 * rrc)2477 int kvm_s390_cpus_from_pv(struct kvm *kvm, u16 *rc, u16 *rrc)
2478 {
2479 struct kvm_vcpu *vcpu;
2480 unsigned long i;
2481 u16 _rc, _rrc;
2482 int ret = 0;
2483
2484 /*
2485 * We ignore failures and try to destroy as many CPUs as possible.
2486 * At the same time we must not free the assigned resources when
2487 * this fails, as the ultravisor has still access to that memory.
2488 * So kvm_s390_pv_destroy_cpu can leave a "wanted" memory leak
2489 * behind.
2490 * We want to return the first failure rc and rrc, though.
2491 */
2492 kvm_for_each_vcpu(i, vcpu, kvm) {
2493 mutex_lock(&vcpu->mutex);
2494 if (kvm_s390_pv_destroy_cpu(vcpu, &_rc, &_rrc) && !ret) {
2495 *rc = _rc;
2496 *rrc = _rrc;
2497 ret = -EIO;
2498 }
2499 mutex_unlock(&vcpu->mutex);
2500 }
2501 /* Ensure that we re-enable gisa if the non-PV guest used it but the PV guest did not. */
2502 if (use_gisa)
2503 kvm_s390_gisa_enable(kvm);
2504 return ret;
2505 }
2506
2507 /**
2508 * kvm_s390_cpus_to_pv - Convert all non-protected vCPUs in a protected VM
2509 * to protected.
2510 * @kvm: the VM whose protected vCPUs are to be converted
2511 * @rc: return value for the RC field of the UVC (in case of error)
2512 * @rrc: return value for the RRC field of the UVC (in case of error)
2513 *
2514 * Tries to undo the conversion in case of error.
2515 *
2516 * Return: 0 in case of success, otherwise -EIO
2517 */
kvm_s390_cpus_to_pv(struct kvm * kvm,u16 * rc,u16 * rrc)2518 static int kvm_s390_cpus_to_pv(struct kvm *kvm, u16 *rc, u16 *rrc)
2519 {
2520 unsigned long i;
2521 int r = 0;
2522 u16 dummy;
2523
2524 struct kvm_vcpu *vcpu;
2525
2526 /* Disable the GISA if the ultravisor does not support AIV. */
2527 if (!uv_has_feature(BIT_UV_FEAT_AIV))
2528 kvm_s390_gisa_disable(kvm);
2529
2530 kvm_for_each_vcpu(i, vcpu, kvm) {
2531 mutex_lock(&vcpu->mutex);
2532 r = kvm_s390_pv_create_cpu(vcpu, rc, rrc);
2533 mutex_unlock(&vcpu->mutex);
2534 if (r)
2535 break;
2536 }
2537 if (r)
2538 kvm_s390_cpus_from_pv(kvm, &dummy, &dummy);
2539 return r;
2540 }
2541
2542 /*
2543 * Here we provide user space with a direct interface to query UV
2544 * related data like UV maxima and available features as well as
2545 * feature specific data.
2546 *
2547 * To facilitate future extension of the data structures we'll try to
2548 * write data up to the maximum requested length.
2549 */
kvm_s390_handle_pv_info(struct kvm_s390_pv_info * info)2550 static ssize_t kvm_s390_handle_pv_info(struct kvm_s390_pv_info *info)
2551 {
2552 ssize_t len_min;
2553
2554 switch (info->header.id) {
2555 case KVM_PV_INFO_VM: {
2556 len_min = sizeof(info->header) + sizeof(info->vm);
2557
2558 if (info->header.len_max < len_min)
2559 return -EINVAL;
2560
2561 memcpy(info->vm.inst_calls_list,
2562 uv_info.inst_calls_list,
2563 sizeof(uv_info.inst_calls_list));
2564
2565 /* It's max cpuid not max cpus, so it's off by one */
2566 info->vm.max_cpus = uv_info.max_guest_cpu_id + 1;
2567 info->vm.max_guests = uv_info.max_num_sec_conf;
2568 info->vm.max_guest_addr = uv_info.max_sec_stor_addr;
2569 info->vm.feature_indication = uv_info.uv_feature_indications;
2570
2571 return len_min;
2572 }
2573 case KVM_PV_INFO_DUMP: {
2574 len_min = sizeof(info->header) + sizeof(info->dump);
2575
2576 if (info->header.len_max < len_min)
2577 return -EINVAL;
2578
2579 info->dump.dump_cpu_buffer_len = uv_info.guest_cpu_stor_len;
2580 info->dump.dump_config_mem_buffer_per_1m = uv_info.conf_dump_storage_state_len;
2581 info->dump.dump_config_finalize_len = uv_info.conf_dump_finalize_len;
2582 return len_min;
2583 }
2584 default:
2585 return -EINVAL;
2586 }
2587 }
2588
kvm_s390_pv_dmp(struct kvm * kvm,struct kvm_pv_cmd * cmd,struct kvm_s390_pv_dmp dmp)2589 static int kvm_s390_pv_dmp(struct kvm *kvm, struct kvm_pv_cmd *cmd,
2590 struct kvm_s390_pv_dmp dmp)
2591 {
2592 int r = -EINVAL;
2593 void __user *result_buff = (void __user *)dmp.buff_addr;
2594
2595 switch (dmp.subcmd) {
2596 case KVM_PV_DUMP_INIT: {
2597 if (kvm->arch.pv.dumping)
2598 break;
2599
2600 /*
2601 * Block SIE entry as concurrent dump UVCs could lead
2602 * to validities.
2603 */
2604 kvm_s390_vcpu_block_all(kvm);
2605
2606 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2607 UVC_CMD_DUMP_INIT, &cmd->rc, &cmd->rrc);
2608 KVM_UV_EVENT(kvm, 3, "PROTVIRT DUMP INIT: rc %x rrc %x",
2609 cmd->rc, cmd->rrc);
2610 if (!r) {
2611 kvm->arch.pv.dumping = true;
2612 } else {
2613 kvm_s390_vcpu_unblock_all(kvm);
2614 r = -EINVAL;
2615 }
2616 break;
2617 }
2618 case KVM_PV_DUMP_CONFIG_STOR_STATE: {
2619 if (!kvm->arch.pv.dumping)
2620 break;
2621
2622 /*
2623 * gaddr is an output parameter since we might stop
2624 * early. As dmp will be copied back in our caller, we
2625 * don't need to do it ourselves.
2626 */
2627 r = kvm_s390_pv_dump_stor_state(kvm, result_buff, &dmp.gaddr, dmp.buff_len,
2628 &cmd->rc, &cmd->rrc);
2629 break;
2630 }
2631 case KVM_PV_DUMP_COMPLETE: {
2632 if (!kvm->arch.pv.dumping)
2633 break;
2634
2635 r = -EINVAL;
2636 if (dmp.buff_len < uv_info.conf_dump_finalize_len)
2637 break;
2638
2639 r = kvm_s390_pv_dump_complete(kvm, result_buff,
2640 &cmd->rc, &cmd->rrc);
2641 break;
2642 }
2643 default:
2644 r = -ENOTTY;
2645 break;
2646 }
2647
2648 return r;
2649 }
2650
kvm_s390_handle_pv(struct kvm * kvm,struct kvm_pv_cmd * cmd)2651 static int kvm_s390_handle_pv(struct kvm *kvm, struct kvm_pv_cmd *cmd)
2652 {
2653 const bool need_lock = (cmd->cmd != KVM_PV_ASYNC_CLEANUP_PERFORM);
2654 void __user *argp = (void __user *)cmd->data;
2655 int r = 0;
2656 u16 dummy;
2657
2658 if (need_lock)
2659 mutex_lock(&kvm->lock);
2660
2661 switch (cmd->cmd) {
2662 case KVM_PV_ENABLE: {
2663 r = -EINVAL;
2664 if (kvm_s390_pv_is_protected(kvm))
2665 break;
2666
2667 /*
2668 * FMT 4 SIE needs esca. As we never switch back to bsca from
2669 * esca, we need no cleanup in the error cases below
2670 */
2671 r = sca_switch_to_extended(kvm);
2672 if (r)
2673 break;
2674
2675 r = s390_disable_cow_sharing();
2676 if (r)
2677 break;
2678
2679 r = kvm_s390_pv_init_vm(kvm, &cmd->rc, &cmd->rrc);
2680 if (r)
2681 break;
2682
2683 r = kvm_s390_cpus_to_pv(kvm, &cmd->rc, &cmd->rrc);
2684 if (r)
2685 kvm_s390_pv_deinit_vm(kvm, &dummy, &dummy);
2686
2687 /* we need to block service interrupts from now on */
2688 set_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2689 break;
2690 }
2691 case KVM_PV_ASYNC_CLEANUP_PREPARE:
2692 r = -EINVAL;
2693 if (!kvm_s390_pv_is_protected(kvm) || !async_destroy)
2694 break;
2695
2696 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc);
2697 /*
2698 * If a CPU could not be destroyed, destroy VM will also fail.
2699 * There is no point in trying to destroy it. Instead return
2700 * the rc and rrc from the first CPU that failed destroying.
2701 */
2702 if (r)
2703 break;
2704 r = kvm_s390_pv_set_aside(kvm, &cmd->rc, &cmd->rrc);
2705
2706 /* no need to block service interrupts any more */
2707 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2708 break;
2709 case KVM_PV_ASYNC_CLEANUP_PERFORM:
2710 r = -EINVAL;
2711 if (!async_destroy)
2712 break;
2713 /* kvm->lock must not be held; this is asserted inside the function. */
2714 r = kvm_s390_pv_deinit_aside_vm(kvm, &cmd->rc, &cmd->rrc);
2715 break;
2716 case KVM_PV_DISABLE: {
2717 r = -EINVAL;
2718 if (!kvm_s390_pv_is_protected(kvm))
2719 break;
2720
2721 r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc);
2722 /*
2723 * If a CPU could not be destroyed, destroy VM will also fail.
2724 * There is no point in trying to destroy it. Instead return
2725 * the rc and rrc from the first CPU that failed destroying.
2726 */
2727 if (r)
2728 break;
2729 r = kvm_s390_pv_deinit_cleanup_all(kvm, &cmd->rc, &cmd->rrc);
2730
2731 /* no need to block service interrupts any more */
2732 clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
2733 break;
2734 }
2735 case KVM_PV_SET_SEC_PARMS: {
2736 struct kvm_s390_pv_sec_parm parms = {};
2737 void *hdr;
2738
2739 r = -EINVAL;
2740 if (!kvm_s390_pv_is_protected(kvm))
2741 break;
2742
2743 r = -EFAULT;
2744 if (copy_from_user(&parms, argp, sizeof(parms)))
2745 break;
2746
2747 /* Currently restricted to 8KB */
2748 r = -EINVAL;
2749 if (parms.length > PAGE_SIZE * 2)
2750 break;
2751
2752 r = -ENOMEM;
2753 hdr = vmalloc(parms.length);
2754 if (!hdr)
2755 break;
2756
2757 r = -EFAULT;
2758 if (!copy_from_user(hdr, (void __user *)parms.origin,
2759 parms.length))
2760 r = kvm_s390_pv_set_sec_parms(kvm, hdr, parms.length,
2761 &cmd->rc, &cmd->rrc);
2762
2763 vfree(hdr);
2764 break;
2765 }
2766 case KVM_PV_UNPACK: {
2767 struct kvm_s390_pv_unp unp = {};
2768
2769 r = -EINVAL;
2770 if (!kvm_s390_pv_is_protected(kvm) || !mm_is_protected(kvm->mm))
2771 break;
2772
2773 r = -EFAULT;
2774 if (copy_from_user(&unp, argp, sizeof(unp)))
2775 break;
2776
2777 r = kvm_s390_pv_unpack(kvm, unp.addr, unp.size, unp.tweak,
2778 &cmd->rc, &cmd->rrc);
2779 break;
2780 }
2781 case KVM_PV_VERIFY: {
2782 r = -EINVAL;
2783 if (!kvm_s390_pv_is_protected(kvm))
2784 break;
2785
2786 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2787 UVC_CMD_VERIFY_IMG, &cmd->rc, &cmd->rrc);
2788 KVM_UV_EVENT(kvm, 3, "PROTVIRT VERIFY: rc %x rrc %x", cmd->rc,
2789 cmd->rrc);
2790 break;
2791 }
2792 case KVM_PV_PREP_RESET: {
2793 r = -EINVAL;
2794 if (!kvm_s390_pv_is_protected(kvm))
2795 break;
2796
2797 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2798 UVC_CMD_PREPARE_RESET, &cmd->rc, &cmd->rrc);
2799 KVM_UV_EVENT(kvm, 3, "PROTVIRT PREP RESET: rc %x rrc %x",
2800 cmd->rc, cmd->rrc);
2801 break;
2802 }
2803 case KVM_PV_UNSHARE_ALL: {
2804 r = -EINVAL;
2805 if (!kvm_s390_pv_is_protected(kvm))
2806 break;
2807
2808 r = uv_cmd_nodata(kvm_s390_pv_get_handle(kvm),
2809 UVC_CMD_SET_UNSHARE_ALL, &cmd->rc, &cmd->rrc);
2810 KVM_UV_EVENT(kvm, 3, "PROTVIRT UNSHARE: rc %x rrc %x",
2811 cmd->rc, cmd->rrc);
2812 break;
2813 }
2814 case KVM_PV_INFO: {
2815 struct kvm_s390_pv_info info = {};
2816 ssize_t data_len;
2817
2818 /*
2819 * No need to check the VM protection here.
2820 *
2821 * Maybe user space wants to query some of the data
2822 * when the VM is still unprotected. If we see the
2823 * need to fence a new data command we can still
2824 * return an error in the info handler.
2825 */
2826
2827 r = -EFAULT;
2828 if (copy_from_user(&info, argp, sizeof(info.header)))
2829 break;
2830
2831 r = -EINVAL;
2832 if (info.header.len_max < sizeof(info.header))
2833 break;
2834
2835 data_len = kvm_s390_handle_pv_info(&info);
2836 if (data_len < 0) {
2837 r = data_len;
2838 break;
2839 }
2840 /*
2841 * If a data command struct is extended (multiple
2842 * times) this can be used to determine how much of it
2843 * is valid.
2844 */
2845 info.header.len_written = data_len;
2846
2847 r = -EFAULT;
2848 if (copy_to_user(argp, &info, data_len))
2849 break;
2850
2851 r = 0;
2852 break;
2853 }
2854 case KVM_PV_DUMP: {
2855 struct kvm_s390_pv_dmp dmp;
2856
2857 r = -EINVAL;
2858 if (!kvm_s390_pv_is_protected(kvm))
2859 break;
2860
2861 r = -EFAULT;
2862 if (copy_from_user(&dmp, argp, sizeof(dmp)))
2863 break;
2864
2865 r = kvm_s390_pv_dmp(kvm, cmd, dmp);
2866 if (r)
2867 break;
2868
2869 if (copy_to_user(argp, &dmp, sizeof(dmp))) {
2870 r = -EFAULT;
2871 break;
2872 }
2873
2874 break;
2875 }
2876 default:
2877 r = -ENOTTY;
2878 }
2879 if (need_lock)
2880 mutex_unlock(&kvm->lock);
2881
2882 return r;
2883 }
2884
mem_op_validate_common(struct kvm_s390_mem_op * mop,u64 supported_flags)2885 static int mem_op_validate_common(struct kvm_s390_mem_op *mop, u64 supported_flags)
2886 {
2887 if (mop->flags & ~supported_flags || !mop->size)
2888 return -EINVAL;
2889 if (mop->size > MEM_OP_MAX_SIZE)
2890 return -E2BIG;
2891 if (mop->flags & KVM_S390_MEMOP_F_SKEY_PROTECTION) {
2892 if (mop->key > 0xf)
2893 return -EINVAL;
2894 } else {
2895 mop->key = 0;
2896 }
2897 return 0;
2898 }
2899
kvm_s390_vm_mem_op_abs(struct kvm * kvm,struct kvm_s390_mem_op * mop)2900 static int kvm_s390_vm_mem_op_abs(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2901 {
2902 void __user *uaddr = (void __user *)mop->buf;
2903 enum gacc_mode acc_mode;
2904 void *tmpbuf = NULL;
2905 int r, srcu_idx;
2906
2907 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_SKEY_PROTECTION |
2908 KVM_S390_MEMOP_F_CHECK_ONLY);
2909 if (r)
2910 return r;
2911
2912 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
2913 tmpbuf = vmalloc(mop->size);
2914 if (!tmpbuf)
2915 return -ENOMEM;
2916 }
2917
2918 srcu_idx = srcu_read_lock(&kvm->srcu);
2919
2920 if (!kvm_is_gpa_in_memslot(kvm, mop->gaddr)) {
2921 r = PGM_ADDRESSING;
2922 goto out_unlock;
2923 }
2924
2925 acc_mode = mop->op == KVM_S390_MEMOP_ABSOLUTE_READ ? GACC_FETCH : GACC_STORE;
2926 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
2927 r = check_gpa_range(kvm, mop->gaddr, mop->size, acc_mode, mop->key);
2928 goto out_unlock;
2929 }
2930 if (acc_mode == GACC_FETCH) {
2931 r = access_guest_abs_with_key(kvm, mop->gaddr, tmpbuf,
2932 mop->size, GACC_FETCH, mop->key);
2933 if (r)
2934 goto out_unlock;
2935 if (copy_to_user(uaddr, tmpbuf, mop->size))
2936 r = -EFAULT;
2937 } else {
2938 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
2939 r = -EFAULT;
2940 goto out_unlock;
2941 }
2942 r = access_guest_abs_with_key(kvm, mop->gaddr, tmpbuf,
2943 mop->size, GACC_STORE, mop->key);
2944 }
2945
2946 out_unlock:
2947 srcu_read_unlock(&kvm->srcu, srcu_idx);
2948
2949 vfree(tmpbuf);
2950 return r;
2951 }
2952
kvm_s390_vm_mem_op_cmpxchg(struct kvm * kvm,struct kvm_s390_mem_op * mop)2953 static int kvm_s390_vm_mem_op_cmpxchg(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2954 {
2955 void __user *uaddr = (void __user *)mop->buf;
2956 void __user *old_addr = (void __user *)mop->old_addr;
2957 union {
2958 __uint128_t quad;
2959 char raw[sizeof(__uint128_t)];
2960 } old = { .quad = 0}, new = { .quad = 0 };
2961 unsigned int off_in_quad = sizeof(new) - mop->size;
2962 int r, srcu_idx;
2963 bool success;
2964
2965 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_SKEY_PROTECTION);
2966 if (r)
2967 return r;
2968 /*
2969 * This validates off_in_quad. Checking that size is a power
2970 * of two is not necessary, as cmpxchg_guest_abs_with_key
2971 * takes care of that
2972 */
2973 if (mop->size > sizeof(new))
2974 return -EINVAL;
2975 if (copy_from_user(&new.raw[off_in_quad], uaddr, mop->size))
2976 return -EFAULT;
2977 if (copy_from_user(&old.raw[off_in_quad], old_addr, mop->size))
2978 return -EFAULT;
2979
2980 srcu_idx = srcu_read_lock(&kvm->srcu);
2981
2982 if (!kvm_is_gpa_in_memslot(kvm, mop->gaddr)) {
2983 r = PGM_ADDRESSING;
2984 goto out_unlock;
2985 }
2986
2987 r = cmpxchg_guest_abs_with_key(kvm, mop->gaddr, mop->size, &old.quad,
2988 new.quad, mop->key, &success);
2989 if (!success && copy_to_user(old_addr, &old.raw[off_in_quad], mop->size))
2990 r = -EFAULT;
2991
2992 out_unlock:
2993 srcu_read_unlock(&kvm->srcu, srcu_idx);
2994 return r;
2995 }
2996
kvm_s390_vm_mem_op(struct kvm * kvm,struct kvm_s390_mem_op * mop)2997 static int kvm_s390_vm_mem_op(struct kvm *kvm, struct kvm_s390_mem_op *mop)
2998 {
2999 /*
3000 * This is technically a heuristic only, if the kvm->lock is not
3001 * taken, it is not guaranteed that the vm is/remains non-protected.
3002 * This is ok from a kernel perspective, wrongdoing is detected
3003 * on the access, -EFAULT is returned and the vm may crash the
3004 * next time it accesses the memory in question.
3005 * There is no sane usecase to do switching and a memop on two
3006 * different CPUs at the same time.
3007 */
3008 if (kvm_s390_pv_get_handle(kvm))
3009 return -EINVAL;
3010
3011 switch (mop->op) {
3012 case KVM_S390_MEMOP_ABSOLUTE_READ:
3013 case KVM_S390_MEMOP_ABSOLUTE_WRITE:
3014 return kvm_s390_vm_mem_op_abs(kvm, mop);
3015 case KVM_S390_MEMOP_ABSOLUTE_CMPXCHG:
3016 return kvm_s390_vm_mem_op_cmpxchg(kvm, mop);
3017 default:
3018 return -EINVAL;
3019 }
3020 }
3021
kvm_arch_vm_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)3022 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
3023 {
3024 struct kvm *kvm = filp->private_data;
3025 void __user *argp = (void __user *)arg;
3026 struct kvm_device_attr attr;
3027 int r;
3028
3029 switch (ioctl) {
3030 case KVM_S390_INTERRUPT: {
3031 struct kvm_s390_interrupt s390int;
3032
3033 r = -EFAULT;
3034 if (copy_from_user(&s390int, argp, sizeof(s390int)))
3035 break;
3036 r = kvm_s390_inject_vm(kvm, &s390int);
3037 break;
3038 }
3039 case KVM_CREATE_IRQCHIP: {
3040 r = -EINVAL;
3041 if (kvm->arch.use_irqchip)
3042 r = 0;
3043 break;
3044 }
3045 case KVM_SET_DEVICE_ATTR: {
3046 r = -EFAULT;
3047 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3048 break;
3049 r = kvm_s390_vm_set_attr(kvm, &attr);
3050 break;
3051 }
3052 case KVM_GET_DEVICE_ATTR: {
3053 r = -EFAULT;
3054 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3055 break;
3056 r = kvm_s390_vm_get_attr(kvm, &attr);
3057 break;
3058 }
3059 case KVM_HAS_DEVICE_ATTR: {
3060 r = -EFAULT;
3061 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3062 break;
3063 r = kvm_s390_vm_has_attr(kvm, &attr);
3064 break;
3065 }
3066 case KVM_S390_GET_SKEYS: {
3067 struct kvm_s390_skeys args;
3068
3069 r = -EFAULT;
3070 if (copy_from_user(&args, argp,
3071 sizeof(struct kvm_s390_skeys)))
3072 break;
3073 r = kvm_s390_get_skeys(kvm, &args);
3074 break;
3075 }
3076 case KVM_S390_SET_SKEYS: {
3077 struct kvm_s390_skeys args;
3078
3079 r = -EFAULT;
3080 if (copy_from_user(&args, argp,
3081 sizeof(struct kvm_s390_skeys)))
3082 break;
3083 r = kvm_s390_set_skeys(kvm, &args);
3084 break;
3085 }
3086 case KVM_S390_GET_CMMA_BITS: {
3087 struct kvm_s390_cmma_log args;
3088
3089 r = -EFAULT;
3090 if (copy_from_user(&args, argp, sizeof(args)))
3091 break;
3092 mutex_lock(&kvm->slots_lock);
3093 r = kvm_s390_get_cmma_bits(kvm, &args);
3094 mutex_unlock(&kvm->slots_lock);
3095 if (!r) {
3096 r = copy_to_user(argp, &args, sizeof(args));
3097 if (r)
3098 r = -EFAULT;
3099 }
3100 break;
3101 }
3102 case KVM_S390_SET_CMMA_BITS: {
3103 struct kvm_s390_cmma_log args;
3104
3105 r = -EFAULT;
3106 if (copy_from_user(&args, argp, sizeof(args)))
3107 break;
3108 mutex_lock(&kvm->slots_lock);
3109 r = kvm_s390_set_cmma_bits(kvm, &args);
3110 mutex_unlock(&kvm->slots_lock);
3111 break;
3112 }
3113 case KVM_S390_PV_COMMAND: {
3114 struct kvm_pv_cmd args;
3115
3116 /* protvirt means user cpu state */
3117 kvm_s390_set_user_cpu_state_ctrl(kvm);
3118 r = 0;
3119 if (!is_prot_virt_host()) {
3120 r = -EINVAL;
3121 break;
3122 }
3123 if (copy_from_user(&args, argp, sizeof(args))) {
3124 r = -EFAULT;
3125 break;
3126 }
3127 if (args.flags) {
3128 r = -EINVAL;
3129 break;
3130 }
3131 /* must be called without kvm->lock */
3132 r = kvm_s390_handle_pv(kvm, &args);
3133 if (copy_to_user(argp, &args, sizeof(args))) {
3134 r = -EFAULT;
3135 break;
3136 }
3137 break;
3138 }
3139 case KVM_S390_MEM_OP: {
3140 struct kvm_s390_mem_op mem_op;
3141
3142 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
3143 r = kvm_s390_vm_mem_op(kvm, &mem_op);
3144 else
3145 r = -EFAULT;
3146 break;
3147 }
3148 case KVM_S390_ZPCI_OP: {
3149 struct kvm_s390_zpci_op args;
3150
3151 r = -EINVAL;
3152 if (!IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
3153 break;
3154 if (copy_from_user(&args, argp, sizeof(args))) {
3155 r = -EFAULT;
3156 break;
3157 }
3158 r = kvm_s390_pci_zpci_op(kvm, &args);
3159 break;
3160 }
3161 default:
3162 r = -ENOTTY;
3163 }
3164
3165 return r;
3166 }
3167
kvm_s390_apxa_installed(void)3168 static int kvm_s390_apxa_installed(void)
3169 {
3170 struct ap_config_info info;
3171
3172 if (ap_instructions_available()) {
3173 if (ap_qci(&info) == 0)
3174 return info.apxa;
3175 }
3176
3177 return 0;
3178 }
3179
3180 /*
3181 * The format of the crypto control block (CRYCB) is specified in the 3 low
3182 * order bits of the CRYCB designation (CRYCBD) field as follows:
3183 * Format 0: Neither the message security assist extension 3 (MSAX3) nor the
3184 * AP extended addressing (APXA) facility are installed.
3185 * Format 1: The APXA facility is not installed but the MSAX3 facility is.
3186 * Format 2: Both the APXA and MSAX3 facilities are installed
3187 */
kvm_s390_set_crycb_format(struct kvm * kvm)3188 static void kvm_s390_set_crycb_format(struct kvm *kvm)
3189 {
3190 kvm->arch.crypto.crycbd = virt_to_phys(kvm->arch.crypto.crycb);
3191
3192 /* Clear the CRYCB format bits - i.e., set format 0 by default */
3193 kvm->arch.crypto.crycbd &= ~(CRYCB_FORMAT_MASK);
3194
3195 /* Check whether MSAX3 is installed */
3196 if (!test_kvm_facility(kvm, 76))
3197 return;
3198
3199 if (kvm_s390_apxa_installed())
3200 kvm->arch.crypto.crycbd |= CRYCB_FORMAT2;
3201 else
3202 kvm->arch.crypto.crycbd |= CRYCB_FORMAT1;
3203 }
3204
3205 /*
3206 * kvm_arch_crypto_set_masks
3207 *
3208 * @kvm: pointer to the target guest's KVM struct containing the crypto masks
3209 * to be set.
3210 * @apm: the mask identifying the accessible AP adapters
3211 * @aqm: the mask identifying the accessible AP domains
3212 * @adm: the mask identifying the accessible AP control domains
3213 *
3214 * Set the masks that identify the adapters, domains and control domains to
3215 * which the KVM guest is granted access.
3216 *
3217 * Note: The kvm->lock mutex must be locked by the caller before invoking this
3218 * function.
3219 */
kvm_arch_crypto_set_masks(struct kvm * kvm,unsigned long * apm,unsigned long * aqm,unsigned long * adm)3220 void kvm_arch_crypto_set_masks(struct kvm *kvm, unsigned long *apm,
3221 unsigned long *aqm, unsigned long *adm)
3222 {
3223 struct kvm_s390_crypto_cb *crycb = kvm->arch.crypto.crycb;
3224
3225 kvm_s390_vcpu_block_all(kvm);
3226
3227 switch (kvm->arch.crypto.crycbd & CRYCB_FORMAT_MASK) {
3228 case CRYCB_FORMAT2: /* APCB1 use 256 bits */
3229 memcpy(crycb->apcb1.apm, apm, 32);
3230 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx %016lx %016lx %016lx",
3231 apm[0], apm[1], apm[2], apm[3]);
3232 memcpy(crycb->apcb1.aqm, aqm, 32);
3233 VM_EVENT(kvm, 3, "SET CRYCB: aqm %016lx %016lx %016lx %016lx",
3234 aqm[0], aqm[1], aqm[2], aqm[3]);
3235 memcpy(crycb->apcb1.adm, adm, 32);
3236 VM_EVENT(kvm, 3, "SET CRYCB: adm %016lx %016lx %016lx %016lx",
3237 adm[0], adm[1], adm[2], adm[3]);
3238 break;
3239 case CRYCB_FORMAT1:
3240 case CRYCB_FORMAT0: /* Fall through both use APCB0 */
3241 memcpy(crycb->apcb0.apm, apm, 8);
3242 memcpy(crycb->apcb0.aqm, aqm, 2);
3243 memcpy(crycb->apcb0.adm, adm, 2);
3244 VM_EVENT(kvm, 3, "SET CRYCB: apm %016lx aqm %04x adm %04x",
3245 apm[0], *((unsigned short *)aqm),
3246 *((unsigned short *)adm));
3247 break;
3248 default: /* Can not happen */
3249 break;
3250 }
3251
3252 /* recreate the shadow crycb for each vcpu */
3253 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART);
3254 kvm_s390_vcpu_unblock_all(kvm);
3255 }
3256 EXPORT_SYMBOL_GPL(kvm_arch_crypto_set_masks);
3257
3258 /*
3259 * kvm_arch_crypto_clear_masks
3260 *
3261 * @kvm: pointer to the target guest's KVM struct containing the crypto masks
3262 * to be cleared.
3263 *
3264 * Clear the masks that identify the adapters, domains and control domains to
3265 * which the KVM guest is granted access.
3266 *
3267 * Note: The kvm->lock mutex must be locked by the caller before invoking this
3268 * function.
3269 */
kvm_arch_crypto_clear_masks(struct kvm * kvm)3270 void kvm_arch_crypto_clear_masks(struct kvm *kvm)
3271 {
3272 kvm_s390_vcpu_block_all(kvm);
3273
3274 memset(&kvm->arch.crypto.crycb->apcb0, 0,
3275 sizeof(kvm->arch.crypto.crycb->apcb0));
3276 memset(&kvm->arch.crypto.crycb->apcb1, 0,
3277 sizeof(kvm->arch.crypto.crycb->apcb1));
3278
3279 VM_EVENT(kvm, 3, "%s", "CLR CRYCB:");
3280 /* recreate the shadow crycb for each vcpu */
3281 kvm_s390_sync_request_broadcast(kvm, KVM_REQ_VSIE_RESTART);
3282 kvm_s390_vcpu_unblock_all(kvm);
3283 }
3284 EXPORT_SYMBOL_GPL(kvm_arch_crypto_clear_masks);
3285
kvm_s390_get_initial_cpuid(void)3286 static u64 kvm_s390_get_initial_cpuid(void)
3287 {
3288 struct cpuid cpuid;
3289
3290 get_cpu_id(&cpuid);
3291 cpuid.version = 0xff;
3292 return *((u64 *) &cpuid);
3293 }
3294
kvm_s390_crypto_init(struct kvm * kvm)3295 static void kvm_s390_crypto_init(struct kvm *kvm)
3296 {
3297 kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
3298 kvm_s390_set_crycb_format(kvm);
3299 init_rwsem(&kvm->arch.crypto.pqap_hook_rwsem);
3300
3301 if (!test_kvm_facility(kvm, 76))
3302 return;
3303
3304 /* Enable AES/DEA protected key functions by default */
3305 kvm->arch.crypto.aes_kw = 1;
3306 kvm->arch.crypto.dea_kw = 1;
3307 get_random_bytes(kvm->arch.crypto.crycb->aes_wrapping_key_mask,
3308 sizeof(kvm->arch.crypto.crycb->aes_wrapping_key_mask));
3309 get_random_bytes(kvm->arch.crypto.crycb->dea_wrapping_key_mask,
3310 sizeof(kvm->arch.crypto.crycb->dea_wrapping_key_mask));
3311 }
3312
sca_dispose(struct kvm * kvm)3313 static void sca_dispose(struct kvm *kvm)
3314 {
3315 if (kvm->arch.use_esca)
3316 free_pages_exact(kvm->arch.sca, sizeof(struct esca_block));
3317 else
3318 free_page((unsigned long)(kvm->arch.sca));
3319 kvm->arch.sca = NULL;
3320 }
3321
kvm_arch_free_vm(struct kvm * kvm)3322 void kvm_arch_free_vm(struct kvm *kvm)
3323 {
3324 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM))
3325 kvm_s390_pci_clear_list(kvm);
3326
3327 __kvm_arch_free_vm(kvm);
3328 }
3329
kvm_arch_init_vm(struct kvm * kvm,unsigned long type)3330 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
3331 {
3332 gfp_t alloc_flags = GFP_KERNEL_ACCOUNT;
3333 int i, rc;
3334 char debug_name[16];
3335 static unsigned long sca_offset;
3336
3337 rc = -EINVAL;
3338 #ifdef CONFIG_KVM_S390_UCONTROL
3339 if (type & ~KVM_VM_S390_UCONTROL)
3340 goto out_err;
3341 if ((type & KVM_VM_S390_UCONTROL) && (!capable(CAP_SYS_ADMIN)))
3342 goto out_err;
3343 #else
3344 if (type)
3345 goto out_err;
3346 #endif
3347
3348 rc = s390_enable_sie();
3349 if (rc)
3350 goto out_err;
3351
3352 rc = -ENOMEM;
3353
3354 if (!sclp.has_64bscao)
3355 alloc_flags |= GFP_DMA;
3356 rwlock_init(&kvm->arch.sca_lock);
3357 /* start with basic SCA */
3358 kvm->arch.sca = (struct bsca_block *) get_zeroed_page(alloc_flags);
3359 if (!kvm->arch.sca)
3360 goto out_err;
3361 mutex_lock(&kvm_lock);
3362 sca_offset += 16;
3363 if (sca_offset + sizeof(struct bsca_block) > PAGE_SIZE)
3364 sca_offset = 0;
3365 kvm->arch.sca = (struct bsca_block *)
3366 ((char *) kvm->arch.sca + sca_offset);
3367 mutex_unlock(&kvm_lock);
3368
3369 sprintf(debug_name, "kvm-%u", current->pid);
3370
3371 kvm->arch.dbf = debug_register(debug_name, 32, 1, 7 * sizeof(long));
3372 if (!kvm->arch.dbf)
3373 goto out_err;
3374
3375 BUILD_BUG_ON(sizeof(struct sie_page2) != 4096);
3376 kvm->arch.sie_page2 =
3377 (struct sie_page2 *) get_zeroed_page(GFP_KERNEL_ACCOUNT | GFP_DMA);
3378 if (!kvm->arch.sie_page2)
3379 goto out_err;
3380
3381 kvm->arch.sie_page2->kvm = kvm;
3382 kvm->arch.model.fac_list = kvm->arch.sie_page2->fac_list;
3383
3384 for (i = 0; i < kvm_s390_fac_size(); i++) {
3385 kvm->arch.model.fac_mask[i] = stfle_fac_list[i] &
3386 (kvm_s390_fac_base[i] |
3387 kvm_s390_fac_ext[i]);
3388 kvm->arch.model.fac_list[i] = stfle_fac_list[i] &
3389 kvm_s390_fac_base[i];
3390 }
3391 kvm->arch.model.subfuncs = kvm_s390_available_subfunc;
3392
3393 /* we are always in czam mode - even on pre z14 machines */
3394 set_kvm_facility(kvm->arch.model.fac_mask, 138);
3395 set_kvm_facility(kvm->arch.model.fac_list, 138);
3396 /* we emulate STHYI in kvm */
3397 set_kvm_facility(kvm->arch.model.fac_mask, 74);
3398 set_kvm_facility(kvm->arch.model.fac_list, 74);
3399 if (MACHINE_HAS_TLB_GUEST) {
3400 set_kvm_facility(kvm->arch.model.fac_mask, 147);
3401 set_kvm_facility(kvm->arch.model.fac_list, 147);
3402 }
3403
3404 if (css_general_characteristics.aiv && test_facility(65))
3405 set_kvm_facility(kvm->arch.model.fac_mask, 65);
3406
3407 kvm->arch.model.cpuid = kvm_s390_get_initial_cpuid();
3408 kvm->arch.model.ibc = sclp.ibc & 0x0fff;
3409
3410 kvm->arch.model.uv_feat_guest.feat = 0;
3411
3412 kvm_s390_crypto_init(kvm);
3413
3414 if (IS_ENABLED(CONFIG_VFIO_PCI_ZDEV_KVM)) {
3415 mutex_lock(&kvm->lock);
3416 kvm_s390_pci_init_list(kvm);
3417 kvm_s390_vcpu_pci_enable_interp(kvm);
3418 mutex_unlock(&kvm->lock);
3419 }
3420
3421 mutex_init(&kvm->arch.float_int.ais_lock);
3422 spin_lock_init(&kvm->arch.float_int.lock);
3423 for (i = 0; i < FIRQ_LIST_COUNT; i++)
3424 INIT_LIST_HEAD(&kvm->arch.float_int.lists[i]);
3425 init_waitqueue_head(&kvm->arch.ipte_wq);
3426 mutex_init(&kvm->arch.ipte_mutex);
3427
3428 debug_register_view(kvm->arch.dbf, &debug_sprintf_view);
3429 VM_EVENT(kvm, 3, "vm created with type %lu", type);
3430
3431 if (type & KVM_VM_S390_UCONTROL) {
3432 struct kvm_userspace_memory_region2 fake_memslot = {
3433 .slot = KVM_S390_UCONTROL_MEMSLOT,
3434 .guest_phys_addr = 0,
3435 .userspace_addr = 0,
3436 .memory_size = ALIGN_DOWN(TASK_SIZE, _SEGMENT_SIZE),
3437 .flags = 0,
3438 };
3439
3440 kvm->arch.gmap = NULL;
3441 kvm->arch.mem_limit = KVM_S390_NO_MEM_LIMIT;
3442 /* one flat fake memslot covering the whole address-space */
3443 mutex_lock(&kvm->slots_lock);
3444 KVM_BUG_ON(kvm_set_internal_memslot(kvm, &fake_memslot), kvm);
3445 mutex_unlock(&kvm->slots_lock);
3446 } else {
3447 if (sclp.hamax == U64_MAX)
3448 kvm->arch.mem_limit = TASK_SIZE_MAX;
3449 else
3450 kvm->arch.mem_limit = min_t(unsigned long, TASK_SIZE_MAX,
3451 sclp.hamax + 1);
3452 kvm->arch.gmap = gmap_create(current->mm, kvm->arch.mem_limit - 1);
3453 if (!kvm->arch.gmap)
3454 goto out_err;
3455 kvm->arch.gmap->private = kvm;
3456 kvm->arch.gmap->pfault_enabled = 0;
3457 }
3458
3459 kvm->arch.use_pfmfi = sclp.has_pfmfi;
3460 kvm->arch.use_skf = sclp.has_skey;
3461 spin_lock_init(&kvm->arch.start_stop_lock);
3462 kvm_s390_vsie_init(kvm);
3463 if (use_gisa)
3464 kvm_s390_gisa_init(kvm);
3465 INIT_LIST_HEAD(&kvm->arch.pv.need_cleanup);
3466 kvm->arch.pv.set_aside = NULL;
3467 KVM_EVENT(3, "vm 0x%pK created by pid %u", kvm, current->pid);
3468
3469 return 0;
3470 out_err:
3471 free_page((unsigned long)kvm->arch.sie_page2);
3472 debug_unregister(kvm->arch.dbf);
3473 sca_dispose(kvm);
3474 KVM_EVENT(3, "creation of vm failed: %d", rc);
3475 return rc;
3476 }
3477
kvm_arch_vcpu_destroy(struct kvm_vcpu * vcpu)3478 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
3479 {
3480 u16 rc, rrc;
3481
3482 VCPU_EVENT(vcpu, 3, "%s", "free cpu");
3483 trace_kvm_s390_destroy_vcpu(vcpu->vcpu_id);
3484 kvm_s390_clear_local_irqs(vcpu);
3485 kvm_clear_async_pf_completion_queue(vcpu);
3486 if (!kvm_is_ucontrol(vcpu->kvm))
3487 sca_del_vcpu(vcpu);
3488 kvm_s390_update_topology_change_report(vcpu->kvm, 1);
3489
3490 if (kvm_is_ucontrol(vcpu->kvm))
3491 gmap_remove(vcpu->arch.gmap);
3492
3493 if (vcpu->kvm->arch.use_cmma)
3494 kvm_s390_vcpu_unsetup_cmma(vcpu);
3495 /* We can not hold the vcpu mutex here, we are already dying */
3496 if (kvm_s390_pv_cpu_get_handle(vcpu))
3497 kvm_s390_pv_destroy_cpu(vcpu, &rc, &rrc);
3498 free_page((unsigned long)(vcpu->arch.sie_block));
3499 }
3500
kvm_arch_destroy_vm(struct kvm * kvm)3501 void kvm_arch_destroy_vm(struct kvm *kvm)
3502 {
3503 u16 rc, rrc;
3504
3505 kvm_destroy_vcpus(kvm);
3506 sca_dispose(kvm);
3507 kvm_s390_gisa_destroy(kvm);
3508 /*
3509 * We are already at the end of life and kvm->lock is not taken.
3510 * This is ok as the file descriptor is closed by now and nobody
3511 * can mess with the pv state.
3512 */
3513 kvm_s390_pv_deinit_cleanup_all(kvm, &rc, &rrc);
3514 /*
3515 * Remove the mmu notifier only when the whole KVM VM is torn down,
3516 * and only if one was registered to begin with. If the VM is
3517 * currently not protected, but has been previously been protected,
3518 * then it's possible that the notifier is still registered.
3519 */
3520 if (kvm->arch.pv.mmu_notifier.ops)
3521 mmu_notifier_unregister(&kvm->arch.pv.mmu_notifier, kvm->mm);
3522
3523 debug_unregister(kvm->arch.dbf);
3524 free_page((unsigned long)kvm->arch.sie_page2);
3525 if (!kvm_is_ucontrol(kvm))
3526 gmap_remove(kvm->arch.gmap);
3527 kvm_s390_destroy_adapters(kvm);
3528 kvm_s390_clear_float_irqs(kvm);
3529 kvm_s390_vsie_destroy(kvm);
3530 KVM_EVENT(3, "vm 0x%pK destroyed", kvm);
3531 }
3532
3533 /* Section: vcpu related */
__kvm_ucontrol_vcpu_init(struct kvm_vcpu * vcpu)3534 static int __kvm_ucontrol_vcpu_init(struct kvm_vcpu *vcpu)
3535 {
3536 vcpu->arch.gmap = gmap_create(current->mm, -1UL);
3537 if (!vcpu->arch.gmap)
3538 return -ENOMEM;
3539 vcpu->arch.gmap->private = vcpu->kvm;
3540
3541 return 0;
3542 }
3543
sca_del_vcpu(struct kvm_vcpu * vcpu)3544 static void sca_del_vcpu(struct kvm_vcpu *vcpu)
3545 {
3546 if (!kvm_s390_use_sca_entries())
3547 return;
3548 read_lock(&vcpu->kvm->arch.sca_lock);
3549 if (vcpu->kvm->arch.use_esca) {
3550 struct esca_block *sca = vcpu->kvm->arch.sca;
3551
3552 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
3553 sca->cpu[vcpu->vcpu_id].sda = 0;
3554 } else {
3555 struct bsca_block *sca = vcpu->kvm->arch.sca;
3556
3557 clear_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
3558 sca->cpu[vcpu->vcpu_id].sda = 0;
3559 }
3560 read_unlock(&vcpu->kvm->arch.sca_lock);
3561 }
3562
sca_add_vcpu(struct kvm_vcpu * vcpu)3563 static void sca_add_vcpu(struct kvm_vcpu *vcpu)
3564 {
3565 if (!kvm_s390_use_sca_entries()) {
3566 phys_addr_t sca_phys = virt_to_phys(vcpu->kvm->arch.sca);
3567
3568 /* we still need the basic sca for the ipte control */
3569 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3570 vcpu->arch.sie_block->scaol = sca_phys;
3571 return;
3572 }
3573 read_lock(&vcpu->kvm->arch.sca_lock);
3574 if (vcpu->kvm->arch.use_esca) {
3575 struct esca_block *sca = vcpu->kvm->arch.sca;
3576 phys_addr_t sca_phys = virt_to_phys(sca);
3577
3578 sca->cpu[vcpu->vcpu_id].sda = virt_to_phys(vcpu->arch.sie_block);
3579 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3580 vcpu->arch.sie_block->scaol = sca_phys & ESCA_SCAOL_MASK;
3581 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
3582 set_bit_inv(vcpu->vcpu_id, (unsigned long *) sca->mcn);
3583 } else {
3584 struct bsca_block *sca = vcpu->kvm->arch.sca;
3585 phys_addr_t sca_phys = virt_to_phys(sca);
3586
3587 sca->cpu[vcpu->vcpu_id].sda = virt_to_phys(vcpu->arch.sie_block);
3588 vcpu->arch.sie_block->scaoh = sca_phys >> 32;
3589 vcpu->arch.sie_block->scaol = sca_phys;
3590 set_bit_inv(vcpu->vcpu_id, (unsigned long *) &sca->mcn);
3591 }
3592 read_unlock(&vcpu->kvm->arch.sca_lock);
3593 }
3594
3595 /* Basic SCA to Extended SCA data copy routines */
sca_copy_entry(struct esca_entry * d,struct bsca_entry * s)3596 static inline void sca_copy_entry(struct esca_entry *d, struct bsca_entry *s)
3597 {
3598 d->sda = s->sda;
3599 d->sigp_ctrl.c = s->sigp_ctrl.c;
3600 d->sigp_ctrl.scn = s->sigp_ctrl.scn;
3601 }
3602
sca_copy_b_to_e(struct esca_block * d,struct bsca_block * s)3603 static void sca_copy_b_to_e(struct esca_block *d, struct bsca_block *s)
3604 {
3605 int i;
3606
3607 d->ipte_control = s->ipte_control;
3608 d->mcn[0] = s->mcn;
3609 for (i = 0; i < KVM_S390_BSCA_CPU_SLOTS; i++)
3610 sca_copy_entry(&d->cpu[i], &s->cpu[i]);
3611 }
3612
sca_switch_to_extended(struct kvm * kvm)3613 static int sca_switch_to_extended(struct kvm *kvm)
3614 {
3615 struct bsca_block *old_sca = kvm->arch.sca;
3616 struct esca_block *new_sca;
3617 struct kvm_vcpu *vcpu;
3618 unsigned long vcpu_idx;
3619 u32 scaol, scaoh;
3620 phys_addr_t new_sca_phys;
3621
3622 if (kvm->arch.use_esca)
3623 return 0;
3624
3625 new_sca = alloc_pages_exact(sizeof(*new_sca), GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3626 if (!new_sca)
3627 return -ENOMEM;
3628
3629 new_sca_phys = virt_to_phys(new_sca);
3630 scaoh = new_sca_phys >> 32;
3631 scaol = new_sca_phys & ESCA_SCAOL_MASK;
3632
3633 kvm_s390_vcpu_block_all(kvm);
3634 write_lock(&kvm->arch.sca_lock);
3635
3636 sca_copy_b_to_e(new_sca, old_sca);
3637
3638 kvm_for_each_vcpu(vcpu_idx, vcpu, kvm) {
3639 vcpu->arch.sie_block->scaoh = scaoh;
3640 vcpu->arch.sie_block->scaol = scaol;
3641 vcpu->arch.sie_block->ecb2 |= ECB2_ESCA;
3642 }
3643 kvm->arch.sca = new_sca;
3644 kvm->arch.use_esca = 1;
3645
3646 write_unlock(&kvm->arch.sca_lock);
3647 kvm_s390_vcpu_unblock_all(kvm);
3648
3649 free_page((unsigned long)old_sca);
3650
3651 VM_EVENT(kvm, 2, "Switched to ESCA (0x%pK -> 0x%pK)",
3652 old_sca, kvm->arch.sca);
3653 return 0;
3654 }
3655
sca_can_add_vcpu(struct kvm * kvm,unsigned int id)3656 static int sca_can_add_vcpu(struct kvm *kvm, unsigned int id)
3657 {
3658 int rc;
3659
3660 if (!kvm_s390_use_sca_entries()) {
3661 if (id < KVM_MAX_VCPUS)
3662 return true;
3663 return false;
3664 }
3665 if (id < KVM_S390_BSCA_CPU_SLOTS)
3666 return true;
3667 if (!sclp.has_esca || !sclp.has_64bscao)
3668 return false;
3669
3670 rc = kvm->arch.use_esca ? 0 : sca_switch_to_extended(kvm);
3671
3672 return rc == 0 && id < KVM_S390_ESCA_CPU_SLOTS;
3673 }
3674
3675 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
__start_cpu_timer_accounting(struct kvm_vcpu * vcpu)3676 static void __start_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3677 {
3678 WARN_ON_ONCE(vcpu->arch.cputm_start != 0);
3679 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3680 vcpu->arch.cputm_start = get_tod_clock_fast();
3681 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3682 }
3683
3684 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
__stop_cpu_timer_accounting(struct kvm_vcpu * vcpu)3685 static void __stop_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3686 {
3687 WARN_ON_ONCE(vcpu->arch.cputm_start == 0);
3688 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3689 vcpu->arch.sie_block->cputm -= get_tod_clock_fast() - vcpu->arch.cputm_start;
3690 vcpu->arch.cputm_start = 0;
3691 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3692 }
3693
3694 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
__enable_cpu_timer_accounting(struct kvm_vcpu * vcpu)3695 static void __enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3696 {
3697 WARN_ON_ONCE(vcpu->arch.cputm_enabled);
3698 vcpu->arch.cputm_enabled = true;
3699 __start_cpu_timer_accounting(vcpu);
3700 }
3701
3702 /* needs disabled preemption to protect from TOD sync and vcpu_load/put */
__disable_cpu_timer_accounting(struct kvm_vcpu * vcpu)3703 static void __disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3704 {
3705 WARN_ON_ONCE(!vcpu->arch.cputm_enabled);
3706 __stop_cpu_timer_accounting(vcpu);
3707 vcpu->arch.cputm_enabled = false;
3708 }
3709
enable_cpu_timer_accounting(struct kvm_vcpu * vcpu)3710 static void enable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3711 {
3712 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3713 __enable_cpu_timer_accounting(vcpu);
3714 preempt_enable();
3715 }
3716
disable_cpu_timer_accounting(struct kvm_vcpu * vcpu)3717 static void disable_cpu_timer_accounting(struct kvm_vcpu *vcpu)
3718 {
3719 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3720 __disable_cpu_timer_accounting(vcpu);
3721 preempt_enable();
3722 }
3723
3724 /* set the cpu timer - may only be called from the VCPU thread itself */
kvm_s390_set_cpu_timer(struct kvm_vcpu * vcpu,__u64 cputm)3725 void kvm_s390_set_cpu_timer(struct kvm_vcpu *vcpu, __u64 cputm)
3726 {
3727 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3728 raw_write_seqcount_begin(&vcpu->arch.cputm_seqcount);
3729 if (vcpu->arch.cputm_enabled)
3730 vcpu->arch.cputm_start = get_tod_clock_fast();
3731 vcpu->arch.sie_block->cputm = cputm;
3732 raw_write_seqcount_end(&vcpu->arch.cputm_seqcount);
3733 preempt_enable();
3734 }
3735
3736 /* update and get the cpu timer - can also be called from other VCPU threads */
kvm_s390_get_cpu_timer(struct kvm_vcpu * vcpu)3737 __u64 kvm_s390_get_cpu_timer(struct kvm_vcpu *vcpu)
3738 {
3739 unsigned int seq;
3740 __u64 value;
3741
3742 if (unlikely(!vcpu->arch.cputm_enabled))
3743 return vcpu->arch.sie_block->cputm;
3744
3745 preempt_disable(); /* protect from TOD sync and vcpu_load/put */
3746 do {
3747 seq = raw_read_seqcount(&vcpu->arch.cputm_seqcount);
3748 /*
3749 * If the writer would ever execute a read in the critical
3750 * section, e.g. in irq context, we have a deadlock.
3751 */
3752 WARN_ON_ONCE((seq & 1) && smp_processor_id() == vcpu->cpu);
3753 value = vcpu->arch.sie_block->cputm;
3754 /* if cputm_start is 0, accounting is being started/stopped */
3755 if (likely(vcpu->arch.cputm_start))
3756 value -= get_tod_clock_fast() - vcpu->arch.cputm_start;
3757 } while (read_seqcount_retry(&vcpu->arch.cputm_seqcount, seq & ~1));
3758 preempt_enable();
3759 return value;
3760 }
3761
kvm_arch_vcpu_load(struct kvm_vcpu * vcpu,int cpu)3762 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3763 {
3764
3765 kvm_s390_set_cpuflags(vcpu, CPUSTAT_RUNNING);
3766 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
3767 __start_cpu_timer_accounting(vcpu);
3768 vcpu->cpu = cpu;
3769 }
3770
kvm_arch_vcpu_put(struct kvm_vcpu * vcpu)3771 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3772 {
3773 vcpu->cpu = -1;
3774 if (vcpu->arch.cputm_enabled && !is_vcpu_idle(vcpu))
3775 __stop_cpu_timer_accounting(vcpu);
3776 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_RUNNING);
3777
3778 }
3779
kvm_arch_vcpu_postcreate(struct kvm_vcpu * vcpu)3780 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
3781 {
3782 mutex_lock(&vcpu->kvm->lock);
3783 preempt_disable();
3784 vcpu->arch.sie_block->epoch = vcpu->kvm->arch.epoch;
3785 vcpu->arch.sie_block->epdx = vcpu->kvm->arch.epdx;
3786 preempt_enable();
3787 mutex_unlock(&vcpu->kvm->lock);
3788 if (!kvm_is_ucontrol(vcpu->kvm)) {
3789 vcpu->arch.gmap = vcpu->kvm->arch.gmap;
3790 sca_add_vcpu(vcpu);
3791 }
3792 if (test_kvm_facility(vcpu->kvm, 74) || vcpu->kvm->arch.user_instr0)
3793 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
3794 }
3795
kvm_has_pckmo_subfunc(struct kvm * kvm,unsigned long nr)3796 static bool kvm_has_pckmo_subfunc(struct kvm *kvm, unsigned long nr)
3797 {
3798 if (test_bit_inv(nr, (unsigned long *)&kvm->arch.model.subfuncs.pckmo) &&
3799 test_bit_inv(nr, (unsigned long *)&kvm_s390_available_subfunc.pckmo))
3800 return true;
3801 return false;
3802 }
3803
kvm_has_pckmo_ecc(struct kvm * kvm)3804 static bool kvm_has_pckmo_ecc(struct kvm *kvm)
3805 {
3806 /* At least one ECC subfunction must be present */
3807 return kvm_has_pckmo_subfunc(kvm, 32) ||
3808 kvm_has_pckmo_subfunc(kvm, 33) ||
3809 kvm_has_pckmo_subfunc(kvm, 34) ||
3810 kvm_has_pckmo_subfunc(kvm, 40) ||
3811 kvm_has_pckmo_subfunc(kvm, 41);
3812
3813 }
3814
kvm_has_pckmo_hmac(struct kvm * kvm)3815 static bool kvm_has_pckmo_hmac(struct kvm *kvm)
3816 {
3817 /* At least one HMAC subfunction must be present */
3818 return kvm_has_pckmo_subfunc(kvm, 118) ||
3819 kvm_has_pckmo_subfunc(kvm, 122);
3820 }
3821
kvm_s390_vcpu_crypto_setup(struct kvm_vcpu * vcpu)3822 static void kvm_s390_vcpu_crypto_setup(struct kvm_vcpu *vcpu)
3823 {
3824 /*
3825 * If the AP instructions are not being interpreted and the MSAX3
3826 * facility is not configured for the guest, there is nothing to set up.
3827 */
3828 if (!vcpu->kvm->arch.crypto.apie && !test_kvm_facility(vcpu->kvm, 76))
3829 return;
3830
3831 vcpu->arch.sie_block->crycbd = vcpu->kvm->arch.crypto.crycbd;
3832 vcpu->arch.sie_block->ecb3 &= ~(ECB3_AES | ECB3_DEA);
3833 vcpu->arch.sie_block->eca &= ~ECA_APIE;
3834 vcpu->arch.sie_block->ecd &= ~(ECD_ECC | ECD_HMAC);
3835
3836 if (vcpu->kvm->arch.crypto.apie)
3837 vcpu->arch.sie_block->eca |= ECA_APIE;
3838
3839 /* Set up protected key support */
3840 if (vcpu->kvm->arch.crypto.aes_kw) {
3841 vcpu->arch.sie_block->ecb3 |= ECB3_AES;
3842 /* ecc/hmac is also wrapped with AES key */
3843 if (kvm_has_pckmo_ecc(vcpu->kvm))
3844 vcpu->arch.sie_block->ecd |= ECD_ECC;
3845 if (kvm_has_pckmo_hmac(vcpu->kvm))
3846 vcpu->arch.sie_block->ecd |= ECD_HMAC;
3847 }
3848
3849 if (vcpu->kvm->arch.crypto.dea_kw)
3850 vcpu->arch.sie_block->ecb3 |= ECB3_DEA;
3851 }
3852
kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu * vcpu)3853 void kvm_s390_vcpu_unsetup_cmma(struct kvm_vcpu *vcpu)
3854 {
3855 free_page((unsigned long)phys_to_virt(vcpu->arch.sie_block->cbrlo));
3856 vcpu->arch.sie_block->cbrlo = 0;
3857 }
3858
kvm_s390_vcpu_setup_cmma(struct kvm_vcpu * vcpu)3859 int kvm_s390_vcpu_setup_cmma(struct kvm_vcpu *vcpu)
3860 {
3861 void *cbrlo_page = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
3862
3863 if (!cbrlo_page)
3864 return -ENOMEM;
3865
3866 vcpu->arch.sie_block->cbrlo = virt_to_phys(cbrlo_page);
3867 return 0;
3868 }
3869
kvm_s390_vcpu_setup_model(struct kvm_vcpu * vcpu)3870 static void kvm_s390_vcpu_setup_model(struct kvm_vcpu *vcpu)
3871 {
3872 struct kvm_s390_cpu_model *model = &vcpu->kvm->arch.model;
3873
3874 vcpu->arch.sie_block->ibc = model->ibc;
3875 if (test_kvm_facility(vcpu->kvm, 7))
3876 vcpu->arch.sie_block->fac = virt_to_phys(model->fac_list);
3877 }
3878
kvm_s390_vcpu_setup(struct kvm_vcpu * vcpu)3879 static int kvm_s390_vcpu_setup(struct kvm_vcpu *vcpu)
3880 {
3881 int rc = 0;
3882 u16 uvrc, uvrrc;
3883
3884 atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
3885 CPUSTAT_SM |
3886 CPUSTAT_STOPPED);
3887
3888 if (test_kvm_facility(vcpu->kvm, 78))
3889 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED2);
3890 else if (test_kvm_facility(vcpu->kvm, 8))
3891 kvm_s390_set_cpuflags(vcpu, CPUSTAT_GED);
3892
3893 kvm_s390_vcpu_setup_model(vcpu);
3894
3895 /* pgste_set_pte has special handling for !MACHINE_HAS_ESOP */
3896 if (MACHINE_HAS_ESOP)
3897 vcpu->arch.sie_block->ecb |= ECB_HOSTPROTINT;
3898 if (test_kvm_facility(vcpu->kvm, 9))
3899 vcpu->arch.sie_block->ecb |= ECB_SRSI;
3900 if (test_kvm_facility(vcpu->kvm, 11))
3901 vcpu->arch.sie_block->ecb |= ECB_PTF;
3902 if (test_kvm_facility(vcpu->kvm, 73))
3903 vcpu->arch.sie_block->ecb |= ECB_TE;
3904 if (!kvm_is_ucontrol(vcpu->kvm))
3905 vcpu->arch.sie_block->ecb |= ECB_SPECI;
3906
3907 if (test_kvm_facility(vcpu->kvm, 8) && vcpu->kvm->arch.use_pfmfi)
3908 vcpu->arch.sie_block->ecb2 |= ECB2_PFMFI;
3909 if (test_kvm_facility(vcpu->kvm, 130))
3910 vcpu->arch.sie_block->ecb2 |= ECB2_IEP;
3911 vcpu->arch.sie_block->eca = ECA_MVPGI | ECA_PROTEXCI;
3912 if (sclp.has_cei)
3913 vcpu->arch.sie_block->eca |= ECA_CEI;
3914 if (sclp.has_ib)
3915 vcpu->arch.sie_block->eca |= ECA_IB;
3916 if (sclp.has_siif)
3917 vcpu->arch.sie_block->eca |= ECA_SII;
3918 if (sclp.has_sigpif)
3919 vcpu->arch.sie_block->eca |= ECA_SIGPI;
3920 if (test_kvm_facility(vcpu->kvm, 129)) {
3921 vcpu->arch.sie_block->eca |= ECA_VX;
3922 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
3923 }
3924 if (test_kvm_facility(vcpu->kvm, 139))
3925 vcpu->arch.sie_block->ecd |= ECD_MEF;
3926 if (test_kvm_facility(vcpu->kvm, 156))
3927 vcpu->arch.sie_block->ecd |= ECD_ETOKENF;
3928 if (vcpu->arch.sie_block->gd) {
3929 vcpu->arch.sie_block->eca |= ECA_AIV;
3930 VCPU_EVENT(vcpu, 3, "AIV gisa format-%u enabled for cpu %03u",
3931 vcpu->arch.sie_block->gd & 0x3, vcpu->vcpu_id);
3932 }
3933 vcpu->arch.sie_block->sdnxo = virt_to_phys(&vcpu->run->s.regs.sdnx) | SDNXC;
3934 vcpu->arch.sie_block->riccbd = virt_to_phys(&vcpu->run->s.regs.riccb);
3935
3936 if (sclp.has_kss)
3937 kvm_s390_set_cpuflags(vcpu, CPUSTAT_KSS);
3938 else
3939 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
3940
3941 if (vcpu->kvm->arch.use_cmma) {
3942 rc = kvm_s390_vcpu_setup_cmma(vcpu);
3943 if (rc)
3944 return rc;
3945 }
3946 hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
3947 vcpu->arch.ckc_timer.function = kvm_s390_idle_wakeup;
3948
3949 vcpu->arch.sie_block->hpid = HPID_KVM;
3950
3951 kvm_s390_vcpu_crypto_setup(vcpu);
3952
3953 kvm_s390_vcpu_pci_setup(vcpu);
3954
3955 mutex_lock(&vcpu->kvm->lock);
3956 if (kvm_s390_pv_is_protected(vcpu->kvm)) {
3957 rc = kvm_s390_pv_create_cpu(vcpu, &uvrc, &uvrrc);
3958 if (rc)
3959 kvm_s390_vcpu_unsetup_cmma(vcpu);
3960 }
3961 mutex_unlock(&vcpu->kvm->lock);
3962
3963 return rc;
3964 }
3965
kvm_arch_vcpu_precreate(struct kvm * kvm,unsigned int id)3966 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
3967 {
3968 if (!kvm_is_ucontrol(kvm) && !sca_can_add_vcpu(kvm, id))
3969 return -EINVAL;
3970 return 0;
3971 }
3972
kvm_arch_vcpu_create(struct kvm_vcpu * vcpu)3973 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
3974 {
3975 struct sie_page *sie_page;
3976 int rc;
3977
3978 BUILD_BUG_ON(sizeof(struct sie_page) != 4096);
3979 sie_page = (struct sie_page *) get_zeroed_page(GFP_KERNEL_ACCOUNT);
3980 if (!sie_page)
3981 return -ENOMEM;
3982
3983 vcpu->arch.sie_block = &sie_page->sie_block;
3984 vcpu->arch.sie_block->itdba = virt_to_phys(&sie_page->itdb);
3985
3986 /* the real guest size will always be smaller than msl */
3987 vcpu->arch.sie_block->mso = 0;
3988 vcpu->arch.sie_block->msl = sclp.hamax;
3989
3990 vcpu->arch.sie_block->icpua = vcpu->vcpu_id;
3991 spin_lock_init(&vcpu->arch.local_int.lock);
3992 vcpu->arch.sie_block->gd = kvm_s390_get_gisa_desc(vcpu->kvm);
3993 seqcount_init(&vcpu->arch.cputm_seqcount);
3994
3995 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
3996 kvm_clear_async_pf_completion_queue(vcpu);
3997 vcpu->run->kvm_valid_regs = KVM_SYNC_PREFIX |
3998 KVM_SYNC_GPRS |
3999 KVM_SYNC_ACRS |
4000 KVM_SYNC_CRS |
4001 KVM_SYNC_ARCH0 |
4002 KVM_SYNC_PFAULT |
4003 KVM_SYNC_DIAG318;
4004 vcpu->arch.acrs_loaded = false;
4005 kvm_s390_set_prefix(vcpu, 0);
4006 if (test_kvm_facility(vcpu->kvm, 64))
4007 vcpu->run->kvm_valid_regs |= KVM_SYNC_RICCB;
4008 if (test_kvm_facility(vcpu->kvm, 82))
4009 vcpu->run->kvm_valid_regs |= KVM_SYNC_BPBC;
4010 if (test_kvm_facility(vcpu->kvm, 133))
4011 vcpu->run->kvm_valid_regs |= KVM_SYNC_GSCB;
4012 if (test_kvm_facility(vcpu->kvm, 156))
4013 vcpu->run->kvm_valid_regs |= KVM_SYNC_ETOKEN;
4014 /* fprs can be synchronized via vrs, even if the guest has no vx. With
4015 * cpu_has_vx(), (load|store)_fpu_regs() will work with vrs format.
4016 */
4017 if (cpu_has_vx())
4018 vcpu->run->kvm_valid_regs |= KVM_SYNC_VRS;
4019 else
4020 vcpu->run->kvm_valid_regs |= KVM_SYNC_FPRS;
4021
4022 if (kvm_is_ucontrol(vcpu->kvm)) {
4023 rc = __kvm_ucontrol_vcpu_init(vcpu);
4024 if (rc)
4025 goto out_free_sie_block;
4026 }
4027
4028 VM_EVENT(vcpu->kvm, 3, "create cpu %d at 0x%pK, sie block at 0x%pK",
4029 vcpu->vcpu_id, vcpu, vcpu->arch.sie_block);
4030 trace_kvm_s390_create_vcpu(vcpu->vcpu_id, vcpu, vcpu->arch.sie_block);
4031
4032 rc = kvm_s390_vcpu_setup(vcpu);
4033 if (rc)
4034 goto out_ucontrol_uninit;
4035
4036 kvm_s390_update_topology_change_report(vcpu->kvm, 1);
4037 return 0;
4038
4039 out_ucontrol_uninit:
4040 if (kvm_is_ucontrol(vcpu->kvm))
4041 gmap_remove(vcpu->arch.gmap);
4042 out_free_sie_block:
4043 free_page((unsigned long)(vcpu->arch.sie_block));
4044 return rc;
4045 }
4046
kvm_arch_vcpu_runnable(struct kvm_vcpu * vcpu)4047 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4048 {
4049 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask);
4050 return kvm_s390_vcpu_has_irq(vcpu, 0);
4051 }
4052
kvm_arch_vcpu_in_kernel(struct kvm_vcpu * vcpu)4053 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
4054 {
4055 return !(vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE);
4056 }
4057
kvm_s390_vcpu_block(struct kvm_vcpu * vcpu)4058 void kvm_s390_vcpu_block(struct kvm_vcpu *vcpu)
4059 {
4060 atomic_or(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
4061 exit_sie(vcpu);
4062 }
4063
kvm_s390_vcpu_unblock(struct kvm_vcpu * vcpu)4064 void kvm_s390_vcpu_unblock(struct kvm_vcpu *vcpu)
4065 {
4066 atomic_andnot(PROG_BLOCK_SIE, &vcpu->arch.sie_block->prog20);
4067 }
4068
kvm_s390_vcpu_request(struct kvm_vcpu * vcpu)4069 static void kvm_s390_vcpu_request(struct kvm_vcpu *vcpu)
4070 {
4071 atomic_or(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
4072 exit_sie(vcpu);
4073 }
4074
kvm_s390_vcpu_sie_inhibited(struct kvm_vcpu * vcpu)4075 bool kvm_s390_vcpu_sie_inhibited(struct kvm_vcpu *vcpu)
4076 {
4077 return atomic_read(&vcpu->arch.sie_block->prog20) &
4078 (PROG_BLOCK_SIE | PROG_REQUEST);
4079 }
4080
kvm_s390_vcpu_request_handled(struct kvm_vcpu * vcpu)4081 static void kvm_s390_vcpu_request_handled(struct kvm_vcpu *vcpu)
4082 {
4083 atomic_andnot(PROG_REQUEST, &vcpu->arch.sie_block->prog20);
4084 }
4085
4086 /*
4087 * Kick a guest cpu out of (v)SIE and wait until (v)SIE is not running.
4088 * If the CPU is not running (e.g. waiting as idle) the function will
4089 * return immediately. */
exit_sie(struct kvm_vcpu * vcpu)4090 void exit_sie(struct kvm_vcpu *vcpu)
4091 {
4092 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOP_INT);
4093 kvm_s390_vsie_kick(vcpu);
4094 while (vcpu->arch.sie_block->prog0c & PROG_IN_SIE)
4095 cpu_relax();
4096 }
4097
4098 /* Kick a guest cpu out of SIE to process a request synchronously */
kvm_s390_sync_request(int req,struct kvm_vcpu * vcpu)4099 void kvm_s390_sync_request(int req, struct kvm_vcpu *vcpu)
4100 {
4101 __kvm_make_request(req, vcpu);
4102 kvm_s390_vcpu_request(vcpu);
4103 }
4104
kvm_gmap_notifier(struct gmap * gmap,unsigned long start,unsigned long end)4105 static void kvm_gmap_notifier(struct gmap *gmap, unsigned long start,
4106 unsigned long end)
4107 {
4108 struct kvm *kvm = gmap->private;
4109 struct kvm_vcpu *vcpu;
4110 unsigned long prefix;
4111 unsigned long i;
4112
4113 trace_kvm_s390_gmap_notifier(start, end, gmap_is_shadow(gmap));
4114
4115 if (gmap_is_shadow(gmap))
4116 return;
4117 if (start >= 1UL << 31)
4118 /* We are only interested in prefix pages */
4119 return;
4120 kvm_for_each_vcpu(i, vcpu, kvm) {
4121 /* match against both prefix pages */
4122 prefix = kvm_s390_get_prefix(vcpu);
4123 if (prefix <= end && start <= prefix + 2*PAGE_SIZE - 1) {
4124 VCPU_EVENT(vcpu, 2, "gmap notifier for %lx-%lx",
4125 start, end);
4126 kvm_s390_sync_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu);
4127 }
4128 }
4129 }
4130
kvm_arch_no_poll(struct kvm_vcpu * vcpu)4131 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
4132 {
4133 /* do not poll with more than halt_poll_max_steal percent of steal time */
4134 if (get_lowcore()->avg_steal_timer * 100 / (TICK_USEC << 12) >=
4135 READ_ONCE(halt_poll_max_steal)) {
4136 vcpu->stat.halt_no_poll_steal++;
4137 return true;
4138 }
4139 return false;
4140 }
4141
kvm_arch_vcpu_should_kick(struct kvm_vcpu * vcpu)4142 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
4143 {
4144 /* kvm common code refers to this, but never calls it */
4145 BUG();
4146 return 0;
4147 }
4148
kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu * vcpu,struct kvm_one_reg * reg)4149 static int kvm_arch_vcpu_ioctl_get_one_reg(struct kvm_vcpu *vcpu,
4150 struct kvm_one_reg *reg)
4151 {
4152 int r = -EINVAL;
4153
4154 switch (reg->id) {
4155 case KVM_REG_S390_TODPR:
4156 r = put_user(vcpu->arch.sie_block->todpr,
4157 (u32 __user *)reg->addr);
4158 break;
4159 case KVM_REG_S390_EPOCHDIFF:
4160 r = put_user(vcpu->arch.sie_block->epoch,
4161 (u64 __user *)reg->addr);
4162 break;
4163 case KVM_REG_S390_CPU_TIMER:
4164 r = put_user(kvm_s390_get_cpu_timer(vcpu),
4165 (u64 __user *)reg->addr);
4166 break;
4167 case KVM_REG_S390_CLOCK_COMP:
4168 r = put_user(vcpu->arch.sie_block->ckc,
4169 (u64 __user *)reg->addr);
4170 break;
4171 case KVM_REG_S390_PFTOKEN:
4172 r = put_user(vcpu->arch.pfault_token,
4173 (u64 __user *)reg->addr);
4174 break;
4175 case KVM_REG_S390_PFCOMPARE:
4176 r = put_user(vcpu->arch.pfault_compare,
4177 (u64 __user *)reg->addr);
4178 break;
4179 case KVM_REG_S390_PFSELECT:
4180 r = put_user(vcpu->arch.pfault_select,
4181 (u64 __user *)reg->addr);
4182 break;
4183 case KVM_REG_S390_PP:
4184 r = put_user(vcpu->arch.sie_block->pp,
4185 (u64 __user *)reg->addr);
4186 break;
4187 case KVM_REG_S390_GBEA:
4188 r = put_user(vcpu->arch.sie_block->gbea,
4189 (u64 __user *)reg->addr);
4190 break;
4191 default:
4192 break;
4193 }
4194
4195 return r;
4196 }
4197
kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu * vcpu,struct kvm_one_reg * reg)4198 static int kvm_arch_vcpu_ioctl_set_one_reg(struct kvm_vcpu *vcpu,
4199 struct kvm_one_reg *reg)
4200 {
4201 int r = -EINVAL;
4202 __u64 val;
4203
4204 switch (reg->id) {
4205 case KVM_REG_S390_TODPR:
4206 r = get_user(vcpu->arch.sie_block->todpr,
4207 (u32 __user *)reg->addr);
4208 break;
4209 case KVM_REG_S390_EPOCHDIFF:
4210 r = get_user(vcpu->arch.sie_block->epoch,
4211 (u64 __user *)reg->addr);
4212 break;
4213 case KVM_REG_S390_CPU_TIMER:
4214 r = get_user(val, (u64 __user *)reg->addr);
4215 if (!r)
4216 kvm_s390_set_cpu_timer(vcpu, val);
4217 break;
4218 case KVM_REG_S390_CLOCK_COMP:
4219 r = get_user(vcpu->arch.sie_block->ckc,
4220 (u64 __user *)reg->addr);
4221 break;
4222 case KVM_REG_S390_PFTOKEN:
4223 r = get_user(vcpu->arch.pfault_token,
4224 (u64 __user *)reg->addr);
4225 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4226 kvm_clear_async_pf_completion_queue(vcpu);
4227 break;
4228 case KVM_REG_S390_PFCOMPARE:
4229 r = get_user(vcpu->arch.pfault_compare,
4230 (u64 __user *)reg->addr);
4231 break;
4232 case KVM_REG_S390_PFSELECT:
4233 r = get_user(vcpu->arch.pfault_select,
4234 (u64 __user *)reg->addr);
4235 break;
4236 case KVM_REG_S390_PP:
4237 r = get_user(vcpu->arch.sie_block->pp,
4238 (u64 __user *)reg->addr);
4239 break;
4240 case KVM_REG_S390_GBEA:
4241 r = get_user(vcpu->arch.sie_block->gbea,
4242 (u64 __user *)reg->addr);
4243 break;
4244 default:
4245 break;
4246 }
4247
4248 return r;
4249 }
4250
kvm_arch_vcpu_ioctl_normal_reset(struct kvm_vcpu * vcpu)4251 static void kvm_arch_vcpu_ioctl_normal_reset(struct kvm_vcpu *vcpu)
4252 {
4253 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_RI;
4254 vcpu->arch.pfault_token = KVM_S390_PFAULT_TOKEN_INVALID;
4255 memset(vcpu->run->s.regs.riccb, 0, sizeof(vcpu->run->s.regs.riccb));
4256
4257 kvm_clear_async_pf_completion_queue(vcpu);
4258 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm))
4259 kvm_s390_vcpu_stop(vcpu);
4260 kvm_s390_clear_local_irqs(vcpu);
4261 }
4262
kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu * vcpu)4263 static void kvm_arch_vcpu_ioctl_initial_reset(struct kvm_vcpu *vcpu)
4264 {
4265 /* Initial reset is a superset of the normal reset */
4266 kvm_arch_vcpu_ioctl_normal_reset(vcpu);
4267
4268 /*
4269 * This equals initial cpu reset in pop, but we don't switch to ESA.
4270 * We do not only reset the internal data, but also ...
4271 */
4272 vcpu->arch.sie_block->gpsw.mask = 0;
4273 vcpu->arch.sie_block->gpsw.addr = 0;
4274 kvm_s390_set_prefix(vcpu, 0);
4275 kvm_s390_set_cpu_timer(vcpu, 0);
4276 vcpu->arch.sie_block->ckc = 0;
4277 memset(vcpu->arch.sie_block->gcr, 0, sizeof(vcpu->arch.sie_block->gcr));
4278 vcpu->arch.sie_block->gcr[0] = CR0_INITIAL_MASK;
4279 vcpu->arch.sie_block->gcr[14] = CR14_INITIAL_MASK;
4280
4281 /* ... the data in sync regs */
4282 memset(vcpu->run->s.regs.crs, 0, sizeof(vcpu->run->s.regs.crs));
4283 vcpu->run->s.regs.ckc = 0;
4284 vcpu->run->s.regs.crs[0] = CR0_INITIAL_MASK;
4285 vcpu->run->s.regs.crs[14] = CR14_INITIAL_MASK;
4286 vcpu->run->psw_addr = 0;
4287 vcpu->run->psw_mask = 0;
4288 vcpu->run->s.regs.todpr = 0;
4289 vcpu->run->s.regs.cputm = 0;
4290 vcpu->run->s.regs.ckc = 0;
4291 vcpu->run->s.regs.pp = 0;
4292 vcpu->run->s.regs.gbea = 1;
4293 vcpu->run->s.regs.fpc = 0;
4294 /*
4295 * Do not reset these registers in the protected case, as some of
4296 * them are overlaid and they are not accessible in this case
4297 * anyway.
4298 */
4299 if (!kvm_s390_pv_cpu_is_protected(vcpu)) {
4300 vcpu->arch.sie_block->gbea = 1;
4301 vcpu->arch.sie_block->pp = 0;
4302 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
4303 vcpu->arch.sie_block->todpr = 0;
4304 }
4305 }
4306
kvm_arch_vcpu_ioctl_clear_reset(struct kvm_vcpu * vcpu)4307 static void kvm_arch_vcpu_ioctl_clear_reset(struct kvm_vcpu *vcpu)
4308 {
4309 struct kvm_sync_regs *regs = &vcpu->run->s.regs;
4310
4311 /* Clear reset is a superset of the initial reset */
4312 kvm_arch_vcpu_ioctl_initial_reset(vcpu);
4313
4314 memset(®s->gprs, 0, sizeof(regs->gprs));
4315 memset(®s->vrs, 0, sizeof(regs->vrs));
4316 memset(®s->acrs, 0, sizeof(regs->acrs));
4317 memset(®s->gscb, 0, sizeof(regs->gscb));
4318
4319 regs->etoken = 0;
4320 regs->etoken_extension = 0;
4321 }
4322
kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu * vcpu,struct kvm_regs * regs)4323 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4324 {
4325 vcpu_load(vcpu);
4326 memcpy(&vcpu->run->s.regs.gprs, ®s->gprs, sizeof(regs->gprs));
4327 vcpu_put(vcpu);
4328 return 0;
4329 }
4330
kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu * vcpu,struct kvm_regs * regs)4331 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4332 {
4333 vcpu_load(vcpu);
4334 memcpy(®s->gprs, &vcpu->run->s.regs.gprs, sizeof(regs->gprs));
4335 vcpu_put(vcpu);
4336 return 0;
4337 }
4338
kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu * vcpu,struct kvm_sregs * sregs)4339 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4340 struct kvm_sregs *sregs)
4341 {
4342 vcpu_load(vcpu);
4343
4344 memcpy(&vcpu->run->s.regs.acrs, &sregs->acrs, sizeof(sregs->acrs));
4345 memcpy(&vcpu->arch.sie_block->gcr, &sregs->crs, sizeof(sregs->crs));
4346
4347 vcpu_put(vcpu);
4348 return 0;
4349 }
4350
kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu * vcpu,struct kvm_sregs * sregs)4351 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4352 struct kvm_sregs *sregs)
4353 {
4354 vcpu_load(vcpu);
4355
4356 memcpy(&sregs->acrs, &vcpu->run->s.regs.acrs, sizeof(sregs->acrs));
4357 memcpy(&sregs->crs, &vcpu->arch.sie_block->gcr, sizeof(sregs->crs));
4358
4359 vcpu_put(vcpu);
4360 return 0;
4361 }
4362
kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu * vcpu,struct kvm_fpu * fpu)4363 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4364 {
4365 int ret = 0;
4366
4367 vcpu_load(vcpu);
4368
4369 vcpu->run->s.regs.fpc = fpu->fpc;
4370 if (cpu_has_vx())
4371 convert_fp_to_vx((__vector128 *) vcpu->run->s.regs.vrs,
4372 (freg_t *) fpu->fprs);
4373 else
4374 memcpy(vcpu->run->s.regs.fprs, &fpu->fprs, sizeof(fpu->fprs));
4375
4376 vcpu_put(vcpu);
4377 return ret;
4378 }
4379
kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu * vcpu,struct kvm_fpu * fpu)4380 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4381 {
4382 vcpu_load(vcpu);
4383
4384 if (cpu_has_vx())
4385 convert_vx_to_fp((freg_t *) fpu->fprs,
4386 (__vector128 *) vcpu->run->s.regs.vrs);
4387 else
4388 memcpy(fpu->fprs, vcpu->run->s.regs.fprs, sizeof(fpu->fprs));
4389 fpu->fpc = vcpu->run->s.regs.fpc;
4390
4391 vcpu_put(vcpu);
4392 return 0;
4393 }
4394
kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu * vcpu,psw_t psw)4395 static int kvm_arch_vcpu_ioctl_set_initial_psw(struct kvm_vcpu *vcpu, psw_t psw)
4396 {
4397 int rc = 0;
4398
4399 if (!is_vcpu_stopped(vcpu))
4400 rc = -EBUSY;
4401 else {
4402 vcpu->run->psw_mask = psw.mask;
4403 vcpu->run->psw_addr = psw.addr;
4404 }
4405 return rc;
4406 }
4407
kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu * vcpu,struct kvm_translation * tr)4408 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4409 struct kvm_translation *tr)
4410 {
4411 return -EINVAL; /* not implemented yet */
4412 }
4413
4414 #define VALID_GUESTDBG_FLAGS (KVM_GUESTDBG_SINGLESTEP | \
4415 KVM_GUESTDBG_USE_HW_BP | \
4416 KVM_GUESTDBG_ENABLE)
4417
kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu * vcpu,struct kvm_guest_debug * dbg)4418 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4419 struct kvm_guest_debug *dbg)
4420 {
4421 int rc = 0;
4422
4423 vcpu_load(vcpu);
4424
4425 vcpu->guest_debug = 0;
4426 kvm_s390_clear_bp_data(vcpu);
4427
4428 if (dbg->control & ~VALID_GUESTDBG_FLAGS) {
4429 rc = -EINVAL;
4430 goto out;
4431 }
4432 if (!sclp.has_gpere) {
4433 rc = -EINVAL;
4434 goto out;
4435 }
4436
4437 if (dbg->control & KVM_GUESTDBG_ENABLE) {
4438 vcpu->guest_debug = dbg->control;
4439 /* enforce guest PER */
4440 kvm_s390_set_cpuflags(vcpu, CPUSTAT_P);
4441
4442 if (dbg->control & KVM_GUESTDBG_USE_HW_BP)
4443 rc = kvm_s390_import_bp_data(vcpu, dbg);
4444 } else {
4445 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
4446 vcpu->arch.guestdbg.last_bp = 0;
4447 }
4448
4449 if (rc) {
4450 vcpu->guest_debug = 0;
4451 kvm_s390_clear_bp_data(vcpu);
4452 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_P);
4453 }
4454
4455 out:
4456 vcpu_put(vcpu);
4457 return rc;
4458 }
4459
kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu * vcpu,struct kvm_mp_state * mp_state)4460 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4461 struct kvm_mp_state *mp_state)
4462 {
4463 int ret;
4464
4465 vcpu_load(vcpu);
4466
4467 /* CHECK_STOP and LOAD are not supported yet */
4468 ret = is_vcpu_stopped(vcpu) ? KVM_MP_STATE_STOPPED :
4469 KVM_MP_STATE_OPERATING;
4470
4471 vcpu_put(vcpu);
4472 return ret;
4473 }
4474
kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu * vcpu,struct kvm_mp_state * mp_state)4475 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4476 struct kvm_mp_state *mp_state)
4477 {
4478 int rc = 0;
4479
4480 vcpu_load(vcpu);
4481
4482 /* user space knows about this interface - let it control the state */
4483 kvm_s390_set_user_cpu_state_ctrl(vcpu->kvm);
4484
4485 switch (mp_state->mp_state) {
4486 case KVM_MP_STATE_STOPPED:
4487 rc = kvm_s390_vcpu_stop(vcpu);
4488 break;
4489 case KVM_MP_STATE_OPERATING:
4490 rc = kvm_s390_vcpu_start(vcpu);
4491 break;
4492 case KVM_MP_STATE_LOAD:
4493 if (!kvm_s390_pv_cpu_is_protected(vcpu)) {
4494 rc = -ENXIO;
4495 break;
4496 }
4497 rc = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR_LOAD);
4498 break;
4499 case KVM_MP_STATE_CHECK_STOP:
4500 fallthrough; /* CHECK_STOP and LOAD are not supported yet */
4501 default:
4502 rc = -ENXIO;
4503 }
4504
4505 vcpu_put(vcpu);
4506 return rc;
4507 }
4508
ibs_enabled(struct kvm_vcpu * vcpu)4509 static bool ibs_enabled(struct kvm_vcpu *vcpu)
4510 {
4511 return kvm_s390_test_cpuflags(vcpu, CPUSTAT_IBS);
4512 }
4513
__kvm_s390_fixup_fault_sync(struct gmap * gmap,gpa_t gaddr,unsigned int flags)4514 static int __kvm_s390_fixup_fault_sync(struct gmap *gmap, gpa_t gaddr, unsigned int flags)
4515 {
4516 struct kvm *kvm = gmap->private;
4517 gfn_t gfn = gpa_to_gfn(gaddr);
4518 bool unlocked;
4519 hva_t vmaddr;
4520 gpa_t tmp;
4521 int rc;
4522
4523 if (kvm_is_ucontrol(kvm)) {
4524 tmp = __gmap_translate(gmap, gaddr);
4525 gfn = gpa_to_gfn(tmp);
4526 }
4527
4528 vmaddr = gfn_to_hva(kvm, gfn);
4529 rc = fixup_user_fault(gmap->mm, vmaddr, FAULT_FLAG_WRITE, &unlocked);
4530 if (!rc)
4531 rc = __gmap_link(gmap, gaddr, vmaddr);
4532 return rc;
4533 }
4534
4535 /**
4536 * __kvm_s390_mprotect_many() - Apply specified protection to guest pages
4537 * @gmap: the gmap of the guest
4538 * @gpa: the starting guest address
4539 * @npages: how many pages to protect
4540 * @prot: indicates access rights: PROT_NONE, PROT_READ or PROT_WRITE
4541 * @bits: pgste notification bits to set
4542 *
4543 * Returns: 0 in case of success, < 0 in case of error - see gmap_protect_one()
4544 *
4545 * Context: kvm->srcu and gmap->mm need to be held in read mode
4546 */
__kvm_s390_mprotect_many(struct gmap * gmap,gpa_t gpa,u8 npages,unsigned int prot,unsigned long bits)4547 int __kvm_s390_mprotect_many(struct gmap *gmap, gpa_t gpa, u8 npages, unsigned int prot,
4548 unsigned long bits)
4549 {
4550 unsigned int fault_flag = (prot & PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
4551 gpa_t end = gpa + npages * PAGE_SIZE;
4552 int rc;
4553
4554 for (; gpa < end; gpa = ALIGN(gpa + 1, rc)) {
4555 rc = gmap_protect_one(gmap, gpa, prot, bits);
4556 if (rc == -EAGAIN) {
4557 __kvm_s390_fixup_fault_sync(gmap, gpa, fault_flag);
4558 rc = gmap_protect_one(gmap, gpa, prot, bits);
4559 }
4560 if (rc < 0)
4561 return rc;
4562 }
4563
4564 return 0;
4565 }
4566
kvm_s390_mprotect_notify_prefix(struct kvm_vcpu * vcpu)4567 static int kvm_s390_mprotect_notify_prefix(struct kvm_vcpu *vcpu)
4568 {
4569 gpa_t gaddr = kvm_s390_get_prefix(vcpu);
4570 int idx, rc;
4571
4572 idx = srcu_read_lock(&vcpu->kvm->srcu);
4573 mmap_read_lock(vcpu->arch.gmap->mm);
4574
4575 rc = __kvm_s390_mprotect_many(vcpu->arch.gmap, gaddr, 2, PROT_WRITE, GMAP_NOTIFY_MPROT);
4576
4577 mmap_read_unlock(vcpu->arch.gmap->mm);
4578 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4579
4580 return rc;
4581 }
4582
kvm_s390_handle_requests(struct kvm_vcpu * vcpu)4583 static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
4584 {
4585 retry:
4586 kvm_s390_vcpu_request_handled(vcpu);
4587 if (!kvm_request_pending(vcpu))
4588 return 0;
4589 /*
4590 * If the guest prefix changed, re-arm the ipte notifier for the
4591 * guest prefix page. gmap_mprotect_notify will wait on the ptl lock.
4592 * This ensures that the ipte instruction for this request has
4593 * already finished. We might race against a second unmapper that
4594 * wants to set the blocking bit. Lets just retry the request loop.
4595 */
4596 if (kvm_check_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu)) {
4597 int rc;
4598
4599 rc = kvm_s390_mprotect_notify_prefix(vcpu);
4600 if (rc) {
4601 kvm_make_request(KVM_REQ_REFRESH_GUEST_PREFIX, vcpu);
4602 return rc;
4603 }
4604 goto retry;
4605 }
4606
4607 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
4608 vcpu->arch.sie_block->ihcpu = 0xffff;
4609 goto retry;
4610 }
4611
4612 if (kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu)) {
4613 if (!ibs_enabled(vcpu)) {
4614 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 1);
4615 kvm_s390_set_cpuflags(vcpu, CPUSTAT_IBS);
4616 }
4617 goto retry;
4618 }
4619
4620 if (kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu)) {
4621 if (ibs_enabled(vcpu)) {
4622 trace_kvm_s390_enable_disable_ibs(vcpu->vcpu_id, 0);
4623 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_IBS);
4624 }
4625 goto retry;
4626 }
4627
4628 if (kvm_check_request(KVM_REQ_ICPT_OPEREXC, vcpu)) {
4629 vcpu->arch.sie_block->ictl |= ICTL_OPEREXC;
4630 goto retry;
4631 }
4632
4633 if (kvm_check_request(KVM_REQ_START_MIGRATION, vcpu)) {
4634 /*
4635 * Disable CMM virtualization; we will emulate the ESSA
4636 * instruction manually, in order to provide additional
4637 * functionalities needed for live migration.
4638 */
4639 vcpu->arch.sie_block->ecb2 &= ~ECB2_CMMA;
4640 goto retry;
4641 }
4642
4643 if (kvm_check_request(KVM_REQ_STOP_MIGRATION, vcpu)) {
4644 /*
4645 * Re-enable CMM virtualization if CMMA is available and
4646 * CMM has been used.
4647 */
4648 if ((vcpu->kvm->arch.use_cmma) &&
4649 (vcpu->kvm->mm->context.uses_cmm))
4650 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
4651 goto retry;
4652 }
4653
4654 /* we left the vsie handler, nothing to do, just clear the request */
4655 kvm_clear_request(KVM_REQ_VSIE_RESTART, vcpu);
4656
4657 return 0;
4658 }
4659
__kvm_s390_set_tod_clock(struct kvm * kvm,const struct kvm_s390_vm_tod_clock * gtod)4660 static void __kvm_s390_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod)
4661 {
4662 struct kvm_vcpu *vcpu;
4663 union tod_clock clk;
4664 unsigned long i;
4665
4666 preempt_disable();
4667
4668 store_tod_clock_ext(&clk);
4669
4670 kvm->arch.epoch = gtod->tod - clk.tod;
4671 kvm->arch.epdx = 0;
4672 if (test_kvm_facility(kvm, 139)) {
4673 kvm->arch.epdx = gtod->epoch_idx - clk.ei;
4674 if (kvm->arch.epoch > gtod->tod)
4675 kvm->arch.epdx -= 1;
4676 }
4677
4678 kvm_s390_vcpu_block_all(kvm);
4679 kvm_for_each_vcpu(i, vcpu, kvm) {
4680 vcpu->arch.sie_block->epoch = kvm->arch.epoch;
4681 vcpu->arch.sie_block->epdx = kvm->arch.epdx;
4682 }
4683
4684 kvm_s390_vcpu_unblock_all(kvm);
4685 preempt_enable();
4686 }
4687
kvm_s390_try_set_tod_clock(struct kvm * kvm,const struct kvm_s390_vm_tod_clock * gtod)4688 int kvm_s390_try_set_tod_clock(struct kvm *kvm, const struct kvm_s390_vm_tod_clock *gtod)
4689 {
4690 if (!mutex_trylock(&kvm->lock))
4691 return 0;
4692 __kvm_s390_set_tod_clock(kvm, gtod);
4693 mutex_unlock(&kvm->lock);
4694 return 1;
4695 }
4696
__kvm_inject_pfault_token(struct kvm_vcpu * vcpu,bool start_token,unsigned long token)4697 static void __kvm_inject_pfault_token(struct kvm_vcpu *vcpu, bool start_token,
4698 unsigned long token)
4699 {
4700 struct kvm_s390_interrupt inti;
4701 struct kvm_s390_irq irq;
4702
4703 if (start_token) {
4704 irq.u.ext.ext_params2 = token;
4705 irq.type = KVM_S390_INT_PFAULT_INIT;
4706 WARN_ON_ONCE(kvm_s390_inject_vcpu(vcpu, &irq));
4707 } else {
4708 inti.type = KVM_S390_INT_PFAULT_DONE;
4709 inti.parm64 = token;
4710 WARN_ON_ONCE(kvm_s390_inject_vm(vcpu->kvm, &inti));
4711 }
4712 }
4713
kvm_arch_async_page_not_present(struct kvm_vcpu * vcpu,struct kvm_async_pf * work)4714 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
4715 struct kvm_async_pf *work)
4716 {
4717 trace_kvm_s390_pfault_init(vcpu, work->arch.pfault_token);
4718 __kvm_inject_pfault_token(vcpu, true, work->arch.pfault_token);
4719
4720 return true;
4721 }
4722
kvm_arch_async_page_present(struct kvm_vcpu * vcpu,struct kvm_async_pf * work)4723 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
4724 struct kvm_async_pf *work)
4725 {
4726 trace_kvm_s390_pfault_done(vcpu, work->arch.pfault_token);
4727 __kvm_inject_pfault_token(vcpu, false, work->arch.pfault_token);
4728 }
4729
kvm_arch_async_page_ready(struct kvm_vcpu * vcpu,struct kvm_async_pf * work)4730 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu,
4731 struct kvm_async_pf *work)
4732 {
4733 /* s390 will always inject the page directly */
4734 }
4735
kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu * vcpu)4736 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
4737 {
4738 /*
4739 * s390 will always inject the page directly,
4740 * but we still want check_async_completion to cleanup
4741 */
4742 return true;
4743 }
4744
kvm_arch_setup_async_pf(struct kvm_vcpu * vcpu)4745 static bool kvm_arch_setup_async_pf(struct kvm_vcpu *vcpu)
4746 {
4747 hva_t hva;
4748 struct kvm_arch_async_pf arch;
4749
4750 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
4751 return false;
4752 if ((vcpu->arch.sie_block->gpsw.mask & vcpu->arch.pfault_select) !=
4753 vcpu->arch.pfault_compare)
4754 return false;
4755 if (psw_extint_disabled(vcpu))
4756 return false;
4757 if (kvm_s390_vcpu_has_irq(vcpu, 0))
4758 return false;
4759 if (!(vcpu->arch.sie_block->gcr[0] & CR0_SERVICE_SIGNAL_SUBMASK))
4760 return false;
4761 if (!vcpu->arch.gmap->pfault_enabled)
4762 return false;
4763
4764 hva = gfn_to_hva(vcpu->kvm, current->thread.gmap_teid.addr);
4765 if (read_guest_real(vcpu, vcpu->arch.pfault_token, &arch.pfault_token, 8))
4766 return false;
4767
4768 return kvm_setup_async_pf(vcpu, current->thread.gmap_teid.addr * PAGE_SIZE, hva, &arch);
4769 }
4770
vcpu_pre_run(struct kvm_vcpu * vcpu)4771 static int vcpu_pre_run(struct kvm_vcpu *vcpu)
4772 {
4773 int rc, cpuflags;
4774
4775 /*
4776 * On s390 notifications for arriving pages will be delivered directly
4777 * to the guest but the house keeping for completed pfaults is
4778 * handled outside the worker.
4779 */
4780 kvm_check_async_pf_completion(vcpu);
4781
4782 vcpu->arch.sie_block->gg14 = vcpu->run->s.regs.gprs[14];
4783 vcpu->arch.sie_block->gg15 = vcpu->run->s.regs.gprs[15];
4784
4785 if (need_resched())
4786 schedule();
4787
4788 if (!kvm_is_ucontrol(vcpu->kvm)) {
4789 rc = kvm_s390_deliver_pending_interrupts(vcpu);
4790 if (rc || guestdbg_exit_pending(vcpu))
4791 return rc;
4792 }
4793
4794 rc = kvm_s390_handle_requests(vcpu);
4795 if (rc)
4796 return rc;
4797
4798 if (guestdbg_enabled(vcpu)) {
4799 kvm_s390_backup_guest_per_regs(vcpu);
4800 kvm_s390_patch_guest_per_regs(vcpu);
4801 }
4802
4803 clear_bit(vcpu->vcpu_idx, vcpu->kvm->arch.gisa_int.kicked_mask);
4804
4805 vcpu->arch.sie_block->icptcode = 0;
4806 current->thread.gmap_int_code = 0;
4807 cpuflags = atomic_read(&vcpu->arch.sie_block->cpuflags);
4808 VCPU_EVENT(vcpu, 6, "entering sie flags %x", cpuflags);
4809 trace_kvm_s390_sie_enter(vcpu, cpuflags);
4810
4811 return 0;
4812 }
4813
vcpu_post_run_addressing_exception(struct kvm_vcpu * vcpu)4814 static int vcpu_post_run_addressing_exception(struct kvm_vcpu *vcpu)
4815 {
4816 struct kvm_s390_pgm_info pgm_info = {
4817 .code = PGM_ADDRESSING,
4818 };
4819 u8 opcode, ilen;
4820 int rc;
4821
4822 VCPU_EVENT(vcpu, 3, "%s", "fault in sie instruction");
4823 trace_kvm_s390_sie_fault(vcpu);
4824
4825 /*
4826 * We want to inject an addressing exception, which is defined as a
4827 * suppressing or terminating exception. However, since we came here
4828 * by a DAT access exception, the PSW still points to the faulting
4829 * instruction since DAT exceptions are nullifying. So we've got
4830 * to look up the current opcode to get the length of the instruction
4831 * to be able to forward the PSW.
4832 */
4833 rc = read_guest_instr(vcpu, vcpu->arch.sie_block->gpsw.addr, &opcode, 1);
4834 ilen = insn_length(opcode);
4835 if (rc < 0) {
4836 return rc;
4837 } else if (rc) {
4838 /* Instruction-Fetching Exceptions - we can't detect the ilen.
4839 * Forward by arbitrary ilc, injection will take care of
4840 * nullification if necessary.
4841 */
4842 pgm_info = vcpu->arch.pgm;
4843 ilen = 4;
4844 }
4845 pgm_info.flags = ilen | KVM_S390_PGM_FLAGS_ILC_VALID;
4846 kvm_s390_forward_psw(vcpu, ilen);
4847 return kvm_s390_inject_prog_irq(vcpu, &pgm_info);
4848 }
4849
kvm_s390_assert_primary_as(struct kvm_vcpu * vcpu)4850 static void kvm_s390_assert_primary_as(struct kvm_vcpu *vcpu)
4851 {
4852 KVM_BUG(current->thread.gmap_teid.as != PSW_BITS_AS_PRIMARY, vcpu->kvm,
4853 "Unexpected program interrupt 0x%x, TEID 0x%016lx",
4854 current->thread.gmap_int_code, current->thread.gmap_teid.val);
4855 }
4856
4857 /*
4858 * __kvm_s390_handle_dat_fault() - handle a dat fault for the gmap of a vcpu
4859 * @vcpu: the vCPU whose gmap is to be fixed up
4860 * @gfn: the guest frame number used for memslots (including fake memslots)
4861 * @gaddr: the gmap address, does not have to match @gfn for ucontrol gmaps
4862 * @flags: FOLL_* flags
4863 *
4864 * Return: 0 on success, < 0 in case of error.
4865 * Context: The mm lock must not be held before calling. May sleep.
4866 */
__kvm_s390_handle_dat_fault(struct kvm_vcpu * vcpu,gfn_t gfn,gpa_t gaddr,unsigned int flags)4867 int __kvm_s390_handle_dat_fault(struct kvm_vcpu *vcpu, gfn_t gfn, gpa_t gaddr, unsigned int flags)
4868 {
4869 struct kvm_memory_slot *slot;
4870 unsigned int fault_flags;
4871 bool writable, unlocked;
4872 unsigned long vmaddr;
4873 struct page *page;
4874 kvm_pfn_t pfn;
4875 int rc;
4876
4877 slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
4878 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
4879 return vcpu_post_run_addressing_exception(vcpu);
4880
4881 fault_flags = flags & FOLL_WRITE ? FAULT_FLAG_WRITE : 0;
4882 if (vcpu->arch.gmap->pfault_enabled)
4883 flags |= FOLL_NOWAIT;
4884 vmaddr = __gfn_to_hva_memslot(slot, gfn);
4885
4886 try_again:
4887 pfn = __kvm_faultin_pfn(slot, gfn, flags, &writable, &page);
4888
4889 /* Access outside memory, inject addressing exception */
4890 if (is_noslot_pfn(pfn))
4891 return vcpu_post_run_addressing_exception(vcpu);
4892 /* Signal pending: try again */
4893 if (pfn == KVM_PFN_ERR_SIGPENDING)
4894 return -EAGAIN;
4895
4896 /* Needs I/O, try to setup async pfault (only possible with FOLL_NOWAIT) */
4897 if (pfn == KVM_PFN_ERR_NEEDS_IO) {
4898 trace_kvm_s390_major_guest_pfault(vcpu);
4899 if (kvm_arch_setup_async_pf(vcpu))
4900 return 0;
4901 vcpu->stat.pfault_sync++;
4902 /* Could not setup async pfault, try again synchronously */
4903 flags &= ~FOLL_NOWAIT;
4904 goto try_again;
4905 }
4906 /* Any other error */
4907 if (is_error_pfn(pfn))
4908 return -EFAULT;
4909
4910 /* Success */
4911 mmap_read_lock(vcpu->arch.gmap->mm);
4912 /* Mark the userspace PTEs as young and/or dirty, to avoid page fault loops */
4913 rc = fixup_user_fault(vcpu->arch.gmap->mm, vmaddr, fault_flags, &unlocked);
4914 if (!rc)
4915 rc = __gmap_link(vcpu->arch.gmap, gaddr, vmaddr);
4916 scoped_guard(spinlock, &vcpu->kvm->mmu_lock) {
4917 kvm_release_faultin_page(vcpu->kvm, page, false, writable);
4918 }
4919 mmap_read_unlock(vcpu->arch.gmap->mm);
4920 return rc;
4921 }
4922
vcpu_dat_fault_handler(struct kvm_vcpu * vcpu,unsigned long gaddr,unsigned int flags)4923 static int vcpu_dat_fault_handler(struct kvm_vcpu *vcpu, unsigned long gaddr, unsigned int flags)
4924 {
4925 unsigned long gaddr_tmp;
4926 gfn_t gfn;
4927
4928 gfn = gpa_to_gfn(gaddr);
4929 if (kvm_is_ucontrol(vcpu->kvm)) {
4930 /*
4931 * This translates the per-vCPU guest address into a
4932 * fake guest address, which can then be used with the
4933 * fake memslots that are identity mapping userspace.
4934 * This allows ucontrol VMs to use the normal fault
4935 * resolution path, like normal VMs.
4936 */
4937 mmap_read_lock(vcpu->arch.gmap->mm);
4938 gaddr_tmp = __gmap_translate(vcpu->arch.gmap, gaddr);
4939 mmap_read_unlock(vcpu->arch.gmap->mm);
4940 if (gaddr_tmp == -EFAULT) {
4941 vcpu->run->exit_reason = KVM_EXIT_S390_UCONTROL;
4942 vcpu->run->s390_ucontrol.trans_exc_code = gaddr;
4943 vcpu->run->s390_ucontrol.pgm_code = PGM_SEGMENT_TRANSLATION;
4944 return -EREMOTE;
4945 }
4946 gfn = gpa_to_gfn(gaddr_tmp);
4947 }
4948 return __kvm_s390_handle_dat_fault(vcpu, gfn, gaddr, flags);
4949 }
4950
vcpu_post_run_handle_fault(struct kvm_vcpu * vcpu)4951 static int vcpu_post_run_handle_fault(struct kvm_vcpu *vcpu)
4952 {
4953 unsigned int flags = 0;
4954 unsigned long gaddr;
4955
4956 gaddr = current->thread.gmap_teid.addr * PAGE_SIZE;
4957 if (kvm_s390_cur_gmap_fault_is_write())
4958 flags = FAULT_FLAG_WRITE;
4959
4960 switch (current->thread.gmap_int_code & PGM_INT_CODE_MASK) {
4961 case 0:
4962 vcpu->stat.exit_null++;
4963 break;
4964 case PGM_NON_SECURE_STORAGE_ACCESS:
4965 kvm_s390_assert_primary_as(vcpu);
4966 /*
4967 * This is normal operation; a page belonging to a protected
4968 * guest has not been imported yet. Try to import the page into
4969 * the protected guest.
4970 */
4971 if (gmap_convert_to_secure(vcpu->arch.gmap, gaddr) == -EINVAL)
4972 send_sig(SIGSEGV, current, 0);
4973 break;
4974 case PGM_SECURE_STORAGE_ACCESS:
4975 case PGM_SECURE_STORAGE_VIOLATION:
4976 kvm_s390_assert_primary_as(vcpu);
4977 /*
4978 * This can happen after a reboot with asynchronous teardown;
4979 * the new guest (normal or protected) will run on top of the
4980 * previous protected guest. The old pages need to be destroyed
4981 * so the new guest can use them.
4982 */
4983 if (gmap_destroy_page(vcpu->arch.gmap, gaddr)) {
4984 /*
4985 * Either KVM messed up the secure guest mapping or the
4986 * same page is mapped into multiple secure guests.
4987 *
4988 * This exception is only triggered when a guest 2 is
4989 * running and can therefore never occur in kernel
4990 * context.
4991 */
4992 pr_warn_ratelimited("Secure storage violation (%x) in task: %s, pid %d\n",
4993 current->thread.gmap_int_code, current->comm,
4994 current->pid);
4995 send_sig(SIGSEGV, current, 0);
4996 }
4997 break;
4998 case PGM_PROTECTION:
4999 case PGM_SEGMENT_TRANSLATION:
5000 case PGM_PAGE_TRANSLATION:
5001 case PGM_ASCE_TYPE:
5002 case PGM_REGION_FIRST_TRANS:
5003 case PGM_REGION_SECOND_TRANS:
5004 case PGM_REGION_THIRD_TRANS:
5005 kvm_s390_assert_primary_as(vcpu);
5006 return vcpu_dat_fault_handler(vcpu, gaddr, flags);
5007 default:
5008 KVM_BUG(1, vcpu->kvm, "Unexpected program interrupt 0x%x, TEID 0x%016lx",
5009 current->thread.gmap_int_code, current->thread.gmap_teid.val);
5010 send_sig(SIGSEGV, current, 0);
5011 break;
5012 }
5013 return 0;
5014 }
5015
vcpu_post_run(struct kvm_vcpu * vcpu,int exit_reason)5016 static int vcpu_post_run(struct kvm_vcpu *vcpu, int exit_reason)
5017 {
5018 struct mcck_volatile_info *mcck_info;
5019 struct sie_page *sie_page;
5020 int rc;
5021
5022 VCPU_EVENT(vcpu, 6, "exit sie icptcode %d",
5023 vcpu->arch.sie_block->icptcode);
5024 trace_kvm_s390_sie_exit(vcpu, vcpu->arch.sie_block->icptcode);
5025
5026 if (guestdbg_enabled(vcpu))
5027 kvm_s390_restore_guest_per_regs(vcpu);
5028
5029 vcpu->run->s.regs.gprs[14] = vcpu->arch.sie_block->gg14;
5030 vcpu->run->s.regs.gprs[15] = vcpu->arch.sie_block->gg15;
5031
5032 if (exit_reason == -EINTR) {
5033 VCPU_EVENT(vcpu, 3, "%s", "machine check");
5034 sie_page = container_of(vcpu->arch.sie_block,
5035 struct sie_page, sie_block);
5036 mcck_info = &sie_page->mcck_info;
5037 kvm_s390_reinject_machine_check(vcpu, mcck_info);
5038 return 0;
5039 }
5040
5041 if (vcpu->arch.sie_block->icptcode > 0) {
5042 rc = kvm_handle_sie_intercept(vcpu);
5043
5044 if (rc != -EOPNOTSUPP)
5045 return rc;
5046 vcpu->run->exit_reason = KVM_EXIT_S390_SIEIC;
5047 vcpu->run->s390_sieic.icptcode = vcpu->arch.sie_block->icptcode;
5048 vcpu->run->s390_sieic.ipa = vcpu->arch.sie_block->ipa;
5049 vcpu->run->s390_sieic.ipb = vcpu->arch.sie_block->ipb;
5050 return -EREMOTE;
5051 }
5052
5053 return vcpu_post_run_handle_fault(vcpu);
5054 }
5055
5056 #define PSW_INT_MASK (PSW_MASK_EXT | PSW_MASK_IO | PSW_MASK_MCHECK)
__vcpu_run(struct kvm_vcpu * vcpu)5057 static int __vcpu_run(struct kvm_vcpu *vcpu)
5058 {
5059 int rc, exit_reason;
5060 struct sie_page *sie_page = (struct sie_page *)vcpu->arch.sie_block;
5061
5062 /*
5063 * We try to hold kvm->srcu during most of vcpu_run (except when run-
5064 * ning the guest), so that memslots (and other stuff) are protected
5065 */
5066 kvm_vcpu_srcu_read_lock(vcpu);
5067
5068 do {
5069 rc = vcpu_pre_run(vcpu);
5070 if (rc || guestdbg_exit_pending(vcpu))
5071 break;
5072
5073 kvm_vcpu_srcu_read_unlock(vcpu);
5074 /*
5075 * As PF_VCPU will be used in fault handler, between
5076 * guest_enter and guest_exit should be no uaccess.
5077 */
5078 local_irq_disable();
5079 guest_enter_irqoff();
5080 __disable_cpu_timer_accounting(vcpu);
5081 local_irq_enable();
5082 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5083 memcpy(sie_page->pv_grregs,
5084 vcpu->run->s.regs.gprs,
5085 sizeof(sie_page->pv_grregs));
5086 }
5087 exit_reason = sie64a(vcpu->arch.sie_block,
5088 vcpu->run->s.regs.gprs,
5089 vcpu->arch.gmap->asce);
5090 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5091 memcpy(vcpu->run->s.regs.gprs,
5092 sie_page->pv_grregs,
5093 sizeof(sie_page->pv_grregs));
5094 /*
5095 * We're not allowed to inject interrupts on intercepts
5096 * that leave the guest state in an "in-between" state
5097 * where the next SIE entry will do a continuation.
5098 * Fence interrupts in our "internal" PSW.
5099 */
5100 if (vcpu->arch.sie_block->icptcode == ICPT_PV_INSTR ||
5101 vcpu->arch.sie_block->icptcode == ICPT_PV_PREF) {
5102 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK;
5103 }
5104 }
5105 local_irq_disable();
5106 __enable_cpu_timer_accounting(vcpu);
5107 guest_exit_irqoff();
5108 local_irq_enable();
5109 kvm_vcpu_srcu_read_lock(vcpu);
5110
5111 rc = vcpu_post_run(vcpu, exit_reason);
5112 } while (!signal_pending(current) && !guestdbg_exit_pending(vcpu) && !rc);
5113
5114 kvm_vcpu_srcu_read_unlock(vcpu);
5115 return rc;
5116 }
5117
sync_regs_fmt2(struct kvm_vcpu * vcpu)5118 static void sync_regs_fmt2(struct kvm_vcpu *vcpu)
5119 {
5120 struct kvm_run *kvm_run = vcpu->run;
5121 struct runtime_instr_cb *riccb;
5122 struct gs_cb *gscb;
5123
5124 riccb = (struct runtime_instr_cb *) &kvm_run->s.regs.riccb;
5125 gscb = (struct gs_cb *) &kvm_run->s.regs.gscb;
5126 vcpu->arch.sie_block->gpsw.mask = kvm_run->psw_mask;
5127 vcpu->arch.sie_block->gpsw.addr = kvm_run->psw_addr;
5128 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
5129 vcpu->arch.sie_block->todpr = kvm_run->s.regs.todpr;
5130 vcpu->arch.sie_block->pp = kvm_run->s.regs.pp;
5131 vcpu->arch.sie_block->gbea = kvm_run->s.regs.gbea;
5132 }
5133 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PFAULT) {
5134 vcpu->arch.pfault_token = kvm_run->s.regs.pft;
5135 vcpu->arch.pfault_select = kvm_run->s.regs.pfs;
5136 vcpu->arch.pfault_compare = kvm_run->s.regs.pfc;
5137 if (vcpu->arch.pfault_token == KVM_S390_PFAULT_TOKEN_INVALID)
5138 kvm_clear_async_pf_completion_queue(vcpu);
5139 }
5140 if (kvm_run->kvm_dirty_regs & KVM_SYNC_DIAG318) {
5141 vcpu->arch.diag318_info.val = kvm_run->s.regs.diag318;
5142 vcpu->arch.sie_block->cpnc = vcpu->arch.diag318_info.cpnc;
5143 VCPU_EVENT(vcpu, 3, "setting cpnc to %d", vcpu->arch.diag318_info.cpnc);
5144 }
5145 /*
5146 * If userspace sets the riccb (e.g. after migration) to a valid state,
5147 * we should enable RI here instead of doing the lazy enablement.
5148 */
5149 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_RICCB) &&
5150 test_kvm_facility(vcpu->kvm, 64) &&
5151 riccb->v &&
5152 !(vcpu->arch.sie_block->ecb3 & ECB3_RI)) {
5153 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (sync_regs)");
5154 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
5155 }
5156 /*
5157 * If userspace sets the gscb (e.g. after migration) to non-zero,
5158 * we should enable GS here instead of doing the lazy enablement.
5159 */
5160 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_GSCB) &&
5161 test_kvm_facility(vcpu->kvm, 133) &&
5162 gscb->gssm &&
5163 !vcpu->arch.gs_enabled) {
5164 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (sync_regs)");
5165 vcpu->arch.sie_block->ecb |= ECB_GS;
5166 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
5167 vcpu->arch.gs_enabled = 1;
5168 }
5169 if ((kvm_run->kvm_dirty_regs & KVM_SYNC_BPBC) &&
5170 test_kvm_facility(vcpu->kvm, 82)) {
5171 vcpu->arch.sie_block->fpf &= ~FPF_BPBC;
5172 vcpu->arch.sie_block->fpf |= kvm_run->s.regs.bpbc ? FPF_BPBC : 0;
5173 }
5174 if (MACHINE_HAS_GS) {
5175 preempt_disable();
5176 local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT);
5177 if (current->thread.gs_cb) {
5178 vcpu->arch.host_gscb = current->thread.gs_cb;
5179 save_gs_cb(vcpu->arch.host_gscb);
5180 }
5181 if (vcpu->arch.gs_enabled) {
5182 current->thread.gs_cb = (struct gs_cb *)
5183 &vcpu->run->s.regs.gscb;
5184 restore_gs_cb(current->thread.gs_cb);
5185 }
5186 preempt_enable();
5187 }
5188 /* SIE will load etoken directly from SDNX and therefore kvm_run */
5189 }
5190
sync_regs(struct kvm_vcpu * vcpu)5191 static void sync_regs(struct kvm_vcpu *vcpu)
5192 {
5193 struct kvm_run *kvm_run = vcpu->run;
5194
5195 if (kvm_run->kvm_dirty_regs & KVM_SYNC_PREFIX)
5196 kvm_s390_set_prefix(vcpu, kvm_run->s.regs.prefix);
5197 if (kvm_run->kvm_dirty_regs & KVM_SYNC_CRS) {
5198 memcpy(&vcpu->arch.sie_block->gcr, &kvm_run->s.regs.crs, 128);
5199 /* some control register changes require a tlb flush */
5200 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5201 }
5202 if (kvm_run->kvm_dirty_regs & KVM_SYNC_ARCH0) {
5203 kvm_s390_set_cpu_timer(vcpu, kvm_run->s.regs.cputm);
5204 vcpu->arch.sie_block->ckc = kvm_run->s.regs.ckc;
5205 }
5206 save_access_regs(vcpu->arch.host_acrs);
5207 restore_access_regs(vcpu->run->s.regs.acrs);
5208 vcpu->arch.acrs_loaded = true;
5209 kvm_s390_fpu_load(vcpu->run);
5210 /* Sync fmt2 only data */
5211 if (likely(!kvm_s390_pv_cpu_is_protected(vcpu))) {
5212 sync_regs_fmt2(vcpu);
5213 } else {
5214 /*
5215 * In several places we have to modify our internal view to
5216 * not do things that are disallowed by the ultravisor. For
5217 * example we must not inject interrupts after specific exits
5218 * (e.g. 112 prefix page not secure). We do this by turning
5219 * off the machine check, external and I/O interrupt bits
5220 * of our PSW copy. To avoid getting validity intercepts, we
5221 * do only accept the condition code from userspace.
5222 */
5223 vcpu->arch.sie_block->gpsw.mask &= ~PSW_MASK_CC;
5224 vcpu->arch.sie_block->gpsw.mask |= kvm_run->psw_mask &
5225 PSW_MASK_CC;
5226 }
5227
5228 kvm_run->kvm_dirty_regs = 0;
5229 }
5230
store_regs_fmt2(struct kvm_vcpu * vcpu)5231 static void store_regs_fmt2(struct kvm_vcpu *vcpu)
5232 {
5233 struct kvm_run *kvm_run = vcpu->run;
5234
5235 kvm_run->s.regs.todpr = vcpu->arch.sie_block->todpr;
5236 kvm_run->s.regs.pp = vcpu->arch.sie_block->pp;
5237 kvm_run->s.regs.gbea = vcpu->arch.sie_block->gbea;
5238 kvm_run->s.regs.bpbc = (vcpu->arch.sie_block->fpf & FPF_BPBC) == FPF_BPBC;
5239 kvm_run->s.regs.diag318 = vcpu->arch.diag318_info.val;
5240 if (MACHINE_HAS_GS) {
5241 preempt_disable();
5242 local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT);
5243 if (vcpu->arch.gs_enabled)
5244 save_gs_cb(current->thread.gs_cb);
5245 current->thread.gs_cb = vcpu->arch.host_gscb;
5246 restore_gs_cb(vcpu->arch.host_gscb);
5247 if (!vcpu->arch.host_gscb)
5248 local_ctl_clear_bit(2, CR2_GUARDED_STORAGE_BIT);
5249 vcpu->arch.host_gscb = NULL;
5250 preempt_enable();
5251 }
5252 /* SIE will save etoken directly into SDNX and therefore kvm_run */
5253 }
5254
store_regs(struct kvm_vcpu * vcpu)5255 static void store_regs(struct kvm_vcpu *vcpu)
5256 {
5257 struct kvm_run *kvm_run = vcpu->run;
5258
5259 kvm_run->psw_mask = vcpu->arch.sie_block->gpsw.mask;
5260 kvm_run->psw_addr = vcpu->arch.sie_block->gpsw.addr;
5261 kvm_run->s.regs.prefix = kvm_s390_get_prefix(vcpu);
5262 memcpy(&kvm_run->s.regs.crs, &vcpu->arch.sie_block->gcr, 128);
5263 kvm_run->s.regs.cputm = kvm_s390_get_cpu_timer(vcpu);
5264 kvm_run->s.regs.ckc = vcpu->arch.sie_block->ckc;
5265 kvm_run->s.regs.pft = vcpu->arch.pfault_token;
5266 kvm_run->s.regs.pfs = vcpu->arch.pfault_select;
5267 kvm_run->s.regs.pfc = vcpu->arch.pfault_compare;
5268 save_access_regs(vcpu->run->s.regs.acrs);
5269 restore_access_regs(vcpu->arch.host_acrs);
5270 vcpu->arch.acrs_loaded = false;
5271 kvm_s390_fpu_store(vcpu->run);
5272 if (likely(!kvm_s390_pv_cpu_is_protected(vcpu)))
5273 store_regs_fmt2(vcpu);
5274 }
5275
kvm_arch_vcpu_ioctl_run(struct kvm_vcpu * vcpu)5276 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
5277 {
5278 struct kvm_run *kvm_run = vcpu->run;
5279 DECLARE_KERNEL_FPU_ONSTACK32(fpu);
5280 int rc;
5281
5282 /*
5283 * Running a VM while dumping always has the potential to
5284 * produce inconsistent dump data. But for PV vcpus a SIE
5285 * entry while dumping could also lead to a fatal validity
5286 * intercept which we absolutely want to avoid.
5287 */
5288 if (vcpu->kvm->arch.pv.dumping)
5289 return -EINVAL;
5290
5291 if (!vcpu->wants_to_run)
5292 return -EINTR;
5293
5294 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_S390_VALID_FIELDS ||
5295 kvm_run->kvm_dirty_regs & ~KVM_SYNC_S390_VALID_FIELDS)
5296 return -EINVAL;
5297
5298 vcpu_load(vcpu);
5299
5300 if (guestdbg_exit_pending(vcpu)) {
5301 kvm_s390_prepare_debug_exit(vcpu);
5302 rc = 0;
5303 goto out;
5304 }
5305
5306 kvm_sigset_activate(vcpu);
5307
5308 /*
5309 * no need to check the return value of vcpu_start as it can only have
5310 * an error for protvirt, but protvirt means user cpu state
5311 */
5312 if (!kvm_s390_user_cpu_state_ctrl(vcpu->kvm)) {
5313 kvm_s390_vcpu_start(vcpu);
5314 } else if (is_vcpu_stopped(vcpu)) {
5315 pr_err_ratelimited("can't run stopped vcpu %d\n",
5316 vcpu->vcpu_id);
5317 rc = -EINVAL;
5318 goto out;
5319 }
5320
5321 kernel_fpu_begin(&fpu, KERNEL_FPC | KERNEL_VXR);
5322 sync_regs(vcpu);
5323 enable_cpu_timer_accounting(vcpu);
5324
5325 might_fault();
5326 rc = __vcpu_run(vcpu);
5327
5328 if (signal_pending(current) && !rc) {
5329 kvm_run->exit_reason = KVM_EXIT_INTR;
5330 rc = -EINTR;
5331 }
5332
5333 if (guestdbg_exit_pending(vcpu) && !rc) {
5334 kvm_s390_prepare_debug_exit(vcpu);
5335 rc = 0;
5336 }
5337
5338 if (rc == -EREMOTE) {
5339 /* userspace support is needed, kvm_run has been prepared */
5340 rc = 0;
5341 }
5342
5343 disable_cpu_timer_accounting(vcpu);
5344 store_regs(vcpu);
5345 kernel_fpu_end(&fpu, KERNEL_FPC | KERNEL_VXR);
5346
5347 kvm_sigset_deactivate(vcpu);
5348
5349 vcpu->stat.exit_userspace++;
5350 out:
5351 vcpu_put(vcpu);
5352 return rc;
5353 }
5354
5355 /*
5356 * store status at address
5357 * we use have two special cases:
5358 * KVM_S390_STORE_STATUS_NOADDR: -> 0x1200 on 64 bit
5359 * KVM_S390_STORE_STATUS_PREFIXED: -> prefix
5360 */
kvm_s390_store_status_unloaded(struct kvm_vcpu * vcpu,unsigned long gpa)5361 int kvm_s390_store_status_unloaded(struct kvm_vcpu *vcpu, unsigned long gpa)
5362 {
5363 unsigned char archmode = 1;
5364 freg_t fprs[NUM_FPRS];
5365 unsigned int px;
5366 u64 clkcomp, cputm;
5367 int rc;
5368
5369 px = kvm_s390_get_prefix(vcpu);
5370 if (gpa == KVM_S390_STORE_STATUS_NOADDR) {
5371 if (write_guest_abs(vcpu, 163, &archmode, 1))
5372 return -EFAULT;
5373 gpa = 0;
5374 } else if (gpa == KVM_S390_STORE_STATUS_PREFIXED) {
5375 if (write_guest_real(vcpu, 163, &archmode, 1))
5376 return -EFAULT;
5377 gpa = px;
5378 } else
5379 gpa -= __LC_FPREGS_SAVE_AREA;
5380
5381 /* manually convert vector registers if necessary */
5382 if (cpu_has_vx()) {
5383 convert_vx_to_fp(fprs, (__vector128 *) vcpu->run->s.regs.vrs);
5384 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
5385 fprs, 128);
5386 } else {
5387 rc = write_guest_abs(vcpu, gpa + __LC_FPREGS_SAVE_AREA,
5388 vcpu->run->s.regs.fprs, 128);
5389 }
5390 rc |= write_guest_abs(vcpu, gpa + __LC_GPREGS_SAVE_AREA,
5391 vcpu->run->s.regs.gprs, 128);
5392 rc |= write_guest_abs(vcpu, gpa + __LC_PSW_SAVE_AREA,
5393 &vcpu->arch.sie_block->gpsw, 16);
5394 rc |= write_guest_abs(vcpu, gpa + __LC_PREFIX_SAVE_AREA,
5395 &px, 4);
5396 rc |= write_guest_abs(vcpu, gpa + __LC_FP_CREG_SAVE_AREA,
5397 &vcpu->run->s.regs.fpc, 4);
5398 rc |= write_guest_abs(vcpu, gpa + __LC_TOD_PROGREG_SAVE_AREA,
5399 &vcpu->arch.sie_block->todpr, 4);
5400 cputm = kvm_s390_get_cpu_timer(vcpu);
5401 rc |= write_guest_abs(vcpu, gpa + __LC_CPU_TIMER_SAVE_AREA,
5402 &cputm, 8);
5403 clkcomp = vcpu->arch.sie_block->ckc >> 8;
5404 rc |= write_guest_abs(vcpu, gpa + __LC_CLOCK_COMP_SAVE_AREA,
5405 &clkcomp, 8);
5406 rc |= write_guest_abs(vcpu, gpa + __LC_AREGS_SAVE_AREA,
5407 &vcpu->run->s.regs.acrs, 64);
5408 rc |= write_guest_abs(vcpu, gpa + __LC_CREGS_SAVE_AREA,
5409 &vcpu->arch.sie_block->gcr, 128);
5410 return rc ? -EFAULT : 0;
5411 }
5412
kvm_s390_vcpu_store_status(struct kvm_vcpu * vcpu,unsigned long addr)5413 int kvm_s390_vcpu_store_status(struct kvm_vcpu *vcpu, unsigned long addr)
5414 {
5415 /*
5416 * The guest FPRS and ACRS are in the host FPRS/ACRS due to the lazy
5417 * switch in the run ioctl. Let's update our copies before we save
5418 * it into the save area
5419 */
5420 kvm_s390_fpu_store(vcpu->run);
5421 save_access_regs(vcpu->run->s.regs.acrs);
5422
5423 return kvm_s390_store_status_unloaded(vcpu, addr);
5424 }
5425
__disable_ibs_on_vcpu(struct kvm_vcpu * vcpu)5426 static void __disable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
5427 {
5428 kvm_check_request(KVM_REQ_ENABLE_IBS, vcpu);
5429 kvm_s390_sync_request(KVM_REQ_DISABLE_IBS, vcpu);
5430 }
5431
__disable_ibs_on_all_vcpus(struct kvm * kvm)5432 static void __disable_ibs_on_all_vcpus(struct kvm *kvm)
5433 {
5434 unsigned long i;
5435 struct kvm_vcpu *vcpu;
5436
5437 kvm_for_each_vcpu(i, vcpu, kvm) {
5438 __disable_ibs_on_vcpu(vcpu);
5439 }
5440 }
5441
__enable_ibs_on_vcpu(struct kvm_vcpu * vcpu)5442 static void __enable_ibs_on_vcpu(struct kvm_vcpu *vcpu)
5443 {
5444 if (!sclp.has_ibs)
5445 return;
5446 kvm_check_request(KVM_REQ_DISABLE_IBS, vcpu);
5447 kvm_s390_sync_request(KVM_REQ_ENABLE_IBS, vcpu);
5448 }
5449
kvm_s390_vcpu_start(struct kvm_vcpu * vcpu)5450 int kvm_s390_vcpu_start(struct kvm_vcpu *vcpu)
5451 {
5452 int i, online_vcpus, r = 0, started_vcpus = 0;
5453
5454 if (!is_vcpu_stopped(vcpu))
5455 return 0;
5456
5457 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 1);
5458 /* Only one cpu at a time may enter/leave the STOPPED state. */
5459 spin_lock(&vcpu->kvm->arch.start_stop_lock);
5460 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
5461
5462 /* Let's tell the UV that we want to change into the operating state */
5463 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5464 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_OPR);
5465 if (r) {
5466 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5467 return r;
5468 }
5469 }
5470
5471 for (i = 0; i < online_vcpus; i++) {
5472 if (!is_vcpu_stopped(kvm_get_vcpu(vcpu->kvm, i)))
5473 started_vcpus++;
5474 }
5475
5476 if (started_vcpus == 0) {
5477 /* we're the only active VCPU -> speed it up */
5478 __enable_ibs_on_vcpu(vcpu);
5479 } else if (started_vcpus == 1) {
5480 /*
5481 * As we are starting a second VCPU, we have to disable
5482 * the IBS facility on all VCPUs to remove potentially
5483 * outstanding ENABLE requests.
5484 */
5485 __disable_ibs_on_all_vcpus(vcpu->kvm);
5486 }
5487
5488 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_STOPPED);
5489 /*
5490 * The real PSW might have changed due to a RESTART interpreted by the
5491 * ultravisor. We block all interrupts and let the next sie exit
5492 * refresh our view.
5493 */
5494 if (kvm_s390_pv_cpu_is_protected(vcpu))
5495 vcpu->arch.sie_block->gpsw.mask &= ~PSW_INT_MASK;
5496 /*
5497 * Another VCPU might have used IBS while we were offline.
5498 * Let's play safe and flush the VCPU at startup.
5499 */
5500 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5501 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5502 return 0;
5503 }
5504
kvm_s390_vcpu_stop(struct kvm_vcpu * vcpu)5505 int kvm_s390_vcpu_stop(struct kvm_vcpu *vcpu)
5506 {
5507 int i, online_vcpus, r = 0, started_vcpus = 0;
5508 struct kvm_vcpu *started_vcpu = NULL;
5509
5510 if (is_vcpu_stopped(vcpu))
5511 return 0;
5512
5513 trace_kvm_s390_vcpu_start_stop(vcpu->vcpu_id, 0);
5514 /* Only one cpu at a time may enter/leave the STOPPED state. */
5515 spin_lock(&vcpu->kvm->arch.start_stop_lock);
5516 online_vcpus = atomic_read(&vcpu->kvm->online_vcpus);
5517
5518 /* Let's tell the UV that we want to change into the stopped state */
5519 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5520 r = kvm_s390_pv_set_cpu_state(vcpu, PV_CPU_STATE_STP);
5521 if (r) {
5522 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5523 return r;
5524 }
5525 }
5526
5527 /*
5528 * Set the VCPU to STOPPED and THEN clear the interrupt flag,
5529 * now that the SIGP STOP and SIGP STOP AND STORE STATUS orders
5530 * have been fully processed. This will ensure that the VCPU
5531 * is kept BUSY if another VCPU is inquiring with SIGP SENSE.
5532 */
5533 kvm_s390_set_cpuflags(vcpu, CPUSTAT_STOPPED);
5534 kvm_s390_clear_stop_irq(vcpu);
5535
5536 __disable_ibs_on_vcpu(vcpu);
5537
5538 for (i = 0; i < online_vcpus; i++) {
5539 struct kvm_vcpu *tmp = kvm_get_vcpu(vcpu->kvm, i);
5540
5541 if (!is_vcpu_stopped(tmp)) {
5542 started_vcpus++;
5543 started_vcpu = tmp;
5544 }
5545 }
5546
5547 if (started_vcpus == 1) {
5548 /*
5549 * As we only have one VCPU left, we want to enable the
5550 * IBS facility for that VCPU to speed it up.
5551 */
5552 __enable_ibs_on_vcpu(started_vcpu);
5553 }
5554
5555 spin_unlock(&vcpu->kvm->arch.start_stop_lock);
5556 return 0;
5557 }
5558
kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu * vcpu,struct kvm_enable_cap * cap)5559 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5560 struct kvm_enable_cap *cap)
5561 {
5562 int r;
5563
5564 if (cap->flags)
5565 return -EINVAL;
5566
5567 switch (cap->cap) {
5568 case KVM_CAP_S390_CSS_SUPPORT:
5569 if (!vcpu->kvm->arch.css_support) {
5570 vcpu->kvm->arch.css_support = 1;
5571 VM_EVENT(vcpu->kvm, 3, "%s", "ENABLE: CSS support");
5572 trace_kvm_s390_enable_css(vcpu->kvm);
5573 }
5574 r = 0;
5575 break;
5576 default:
5577 r = -EINVAL;
5578 break;
5579 }
5580 return r;
5581 }
5582
kvm_s390_vcpu_sida_op(struct kvm_vcpu * vcpu,struct kvm_s390_mem_op * mop)5583 static long kvm_s390_vcpu_sida_op(struct kvm_vcpu *vcpu,
5584 struct kvm_s390_mem_op *mop)
5585 {
5586 void __user *uaddr = (void __user *)mop->buf;
5587 void *sida_addr;
5588 int r = 0;
5589
5590 if (mop->flags || !mop->size)
5591 return -EINVAL;
5592 if (mop->size + mop->sida_offset < mop->size)
5593 return -EINVAL;
5594 if (mop->size + mop->sida_offset > sida_size(vcpu->arch.sie_block))
5595 return -E2BIG;
5596 if (!kvm_s390_pv_cpu_is_protected(vcpu))
5597 return -EINVAL;
5598
5599 sida_addr = (char *)sida_addr(vcpu->arch.sie_block) + mop->sida_offset;
5600
5601 switch (mop->op) {
5602 case KVM_S390_MEMOP_SIDA_READ:
5603 if (copy_to_user(uaddr, sida_addr, mop->size))
5604 r = -EFAULT;
5605
5606 break;
5607 case KVM_S390_MEMOP_SIDA_WRITE:
5608 if (copy_from_user(sida_addr, uaddr, mop->size))
5609 r = -EFAULT;
5610 break;
5611 }
5612 return r;
5613 }
5614
kvm_s390_vcpu_mem_op(struct kvm_vcpu * vcpu,struct kvm_s390_mem_op * mop)5615 static long kvm_s390_vcpu_mem_op(struct kvm_vcpu *vcpu,
5616 struct kvm_s390_mem_op *mop)
5617 {
5618 void __user *uaddr = (void __user *)mop->buf;
5619 enum gacc_mode acc_mode;
5620 void *tmpbuf = NULL;
5621 int r;
5622
5623 r = mem_op_validate_common(mop, KVM_S390_MEMOP_F_INJECT_EXCEPTION |
5624 KVM_S390_MEMOP_F_CHECK_ONLY |
5625 KVM_S390_MEMOP_F_SKEY_PROTECTION);
5626 if (r)
5627 return r;
5628 if (mop->ar >= NUM_ACRS)
5629 return -EINVAL;
5630 if (kvm_s390_pv_cpu_is_protected(vcpu))
5631 return -EINVAL;
5632 if (!(mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY)) {
5633 tmpbuf = vmalloc(mop->size);
5634 if (!tmpbuf)
5635 return -ENOMEM;
5636 }
5637
5638 acc_mode = mop->op == KVM_S390_MEMOP_LOGICAL_READ ? GACC_FETCH : GACC_STORE;
5639 if (mop->flags & KVM_S390_MEMOP_F_CHECK_ONLY) {
5640 r = check_gva_range(vcpu, mop->gaddr, mop->ar, mop->size,
5641 acc_mode, mop->key);
5642 goto out_inject;
5643 }
5644 if (acc_mode == GACC_FETCH) {
5645 r = read_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf,
5646 mop->size, mop->key);
5647 if (r)
5648 goto out_inject;
5649 if (copy_to_user(uaddr, tmpbuf, mop->size)) {
5650 r = -EFAULT;
5651 goto out_free;
5652 }
5653 } else {
5654 if (copy_from_user(tmpbuf, uaddr, mop->size)) {
5655 r = -EFAULT;
5656 goto out_free;
5657 }
5658 r = write_guest_with_key(vcpu, mop->gaddr, mop->ar, tmpbuf,
5659 mop->size, mop->key);
5660 }
5661
5662 out_inject:
5663 if (r > 0 && (mop->flags & KVM_S390_MEMOP_F_INJECT_EXCEPTION) != 0)
5664 kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
5665
5666 out_free:
5667 vfree(tmpbuf);
5668 return r;
5669 }
5670
kvm_s390_vcpu_memsida_op(struct kvm_vcpu * vcpu,struct kvm_s390_mem_op * mop)5671 static long kvm_s390_vcpu_memsida_op(struct kvm_vcpu *vcpu,
5672 struct kvm_s390_mem_op *mop)
5673 {
5674 int r, srcu_idx;
5675
5676 srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5677
5678 switch (mop->op) {
5679 case KVM_S390_MEMOP_LOGICAL_READ:
5680 case KVM_S390_MEMOP_LOGICAL_WRITE:
5681 r = kvm_s390_vcpu_mem_op(vcpu, mop);
5682 break;
5683 case KVM_S390_MEMOP_SIDA_READ:
5684 case KVM_S390_MEMOP_SIDA_WRITE:
5685 /* we are locked against sida going away by the vcpu->mutex */
5686 r = kvm_s390_vcpu_sida_op(vcpu, mop);
5687 break;
5688 default:
5689 r = -EINVAL;
5690 }
5691
5692 srcu_read_unlock(&vcpu->kvm->srcu, srcu_idx);
5693 return r;
5694 }
5695
kvm_arch_vcpu_async_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)5696 long kvm_arch_vcpu_async_ioctl(struct file *filp,
5697 unsigned int ioctl, unsigned long arg)
5698 {
5699 struct kvm_vcpu *vcpu = filp->private_data;
5700 void __user *argp = (void __user *)arg;
5701 int rc;
5702
5703 switch (ioctl) {
5704 case KVM_S390_IRQ: {
5705 struct kvm_s390_irq s390irq;
5706
5707 if (copy_from_user(&s390irq, argp, sizeof(s390irq)))
5708 return -EFAULT;
5709 rc = kvm_s390_inject_vcpu(vcpu, &s390irq);
5710 break;
5711 }
5712 case KVM_S390_INTERRUPT: {
5713 struct kvm_s390_interrupt s390int;
5714 struct kvm_s390_irq s390irq = {};
5715
5716 if (copy_from_user(&s390int, argp, sizeof(s390int)))
5717 return -EFAULT;
5718 if (s390int_to_s390irq(&s390int, &s390irq))
5719 return -EINVAL;
5720 rc = kvm_s390_inject_vcpu(vcpu, &s390irq);
5721 break;
5722 }
5723 default:
5724 rc = -ENOIOCTLCMD;
5725 break;
5726 }
5727
5728 /*
5729 * To simplify single stepping of userspace-emulated instructions,
5730 * KVM_EXIT_S390_SIEIC exit sets KVM_GUESTDBG_EXIT_PENDING (see
5731 * should_handle_per_ifetch()). However, if userspace emulation injects
5732 * an interrupt, it needs to be cleared, so that KVM_EXIT_DEBUG happens
5733 * after (and not before) the interrupt delivery.
5734 */
5735 if (!rc)
5736 vcpu->guest_debug &= ~KVM_GUESTDBG_EXIT_PENDING;
5737
5738 return rc;
5739 }
5740
kvm_s390_handle_pv_vcpu_dump(struct kvm_vcpu * vcpu,struct kvm_pv_cmd * cmd)5741 static int kvm_s390_handle_pv_vcpu_dump(struct kvm_vcpu *vcpu,
5742 struct kvm_pv_cmd *cmd)
5743 {
5744 struct kvm_s390_pv_dmp dmp;
5745 void *data;
5746 int ret;
5747
5748 /* Dump initialization is a prerequisite */
5749 if (!vcpu->kvm->arch.pv.dumping)
5750 return -EINVAL;
5751
5752 if (copy_from_user(&dmp, (__u8 __user *)cmd->data, sizeof(dmp)))
5753 return -EFAULT;
5754
5755 /* We only handle this subcmd right now */
5756 if (dmp.subcmd != KVM_PV_DUMP_CPU)
5757 return -EINVAL;
5758
5759 /* CPU dump length is the same as create cpu storage donation. */
5760 if (dmp.buff_len != uv_info.guest_cpu_stor_len)
5761 return -EINVAL;
5762
5763 data = kvzalloc(uv_info.guest_cpu_stor_len, GFP_KERNEL);
5764 if (!data)
5765 return -ENOMEM;
5766
5767 ret = kvm_s390_pv_dump_cpu(vcpu, data, &cmd->rc, &cmd->rrc);
5768
5769 VCPU_EVENT(vcpu, 3, "PROTVIRT DUMP CPU %d rc %x rrc %x",
5770 vcpu->vcpu_id, cmd->rc, cmd->rrc);
5771
5772 if (ret)
5773 ret = -EINVAL;
5774
5775 /* On success copy over the dump data */
5776 if (!ret && copy_to_user((__u8 __user *)dmp.buff_addr, data, uv_info.guest_cpu_stor_len))
5777 ret = -EFAULT;
5778
5779 kvfree(data);
5780 return ret;
5781 }
5782
kvm_arch_vcpu_ioctl(struct file * filp,unsigned int ioctl,unsigned long arg)5783 long kvm_arch_vcpu_ioctl(struct file *filp,
5784 unsigned int ioctl, unsigned long arg)
5785 {
5786 struct kvm_vcpu *vcpu = filp->private_data;
5787 void __user *argp = (void __user *)arg;
5788 int idx;
5789 long r;
5790 u16 rc, rrc;
5791
5792 vcpu_load(vcpu);
5793
5794 switch (ioctl) {
5795 case KVM_S390_STORE_STATUS:
5796 idx = srcu_read_lock(&vcpu->kvm->srcu);
5797 r = kvm_s390_store_status_unloaded(vcpu, arg);
5798 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5799 break;
5800 case KVM_S390_SET_INITIAL_PSW: {
5801 psw_t psw;
5802
5803 r = -EFAULT;
5804 if (copy_from_user(&psw, argp, sizeof(psw)))
5805 break;
5806 r = kvm_arch_vcpu_ioctl_set_initial_psw(vcpu, psw);
5807 break;
5808 }
5809 case KVM_S390_CLEAR_RESET:
5810 r = 0;
5811 kvm_arch_vcpu_ioctl_clear_reset(vcpu);
5812 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5813 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5814 UVC_CMD_CPU_RESET_CLEAR, &rc, &rrc);
5815 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET CLEAR VCPU: rc %x rrc %x",
5816 rc, rrc);
5817 }
5818 break;
5819 case KVM_S390_INITIAL_RESET:
5820 r = 0;
5821 kvm_arch_vcpu_ioctl_initial_reset(vcpu);
5822 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5823 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5824 UVC_CMD_CPU_RESET_INITIAL,
5825 &rc, &rrc);
5826 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET INITIAL VCPU: rc %x rrc %x",
5827 rc, rrc);
5828 }
5829 break;
5830 case KVM_S390_NORMAL_RESET:
5831 r = 0;
5832 kvm_arch_vcpu_ioctl_normal_reset(vcpu);
5833 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
5834 r = uv_cmd_nodata(kvm_s390_pv_cpu_get_handle(vcpu),
5835 UVC_CMD_CPU_RESET, &rc, &rrc);
5836 VCPU_EVENT(vcpu, 3, "PROTVIRT RESET NORMAL VCPU: rc %x rrc %x",
5837 rc, rrc);
5838 }
5839 break;
5840 case KVM_SET_ONE_REG:
5841 case KVM_GET_ONE_REG: {
5842 struct kvm_one_reg reg;
5843 r = -EINVAL;
5844 if (kvm_s390_pv_cpu_is_protected(vcpu))
5845 break;
5846 r = -EFAULT;
5847 if (copy_from_user(®, argp, sizeof(reg)))
5848 break;
5849 if (ioctl == KVM_SET_ONE_REG)
5850 r = kvm_arch_vcpu_ioctl_set_one_reg(vcpu, ®);
5851 else
5852 r = kvm_arch_vcpu_ioctl_get_one_reg(vcpu, ®);
5853 break;
5854 }
5855 #ifdef CONFIG_KVM_S390_UCONTROL
5856 case KVM_S390_UCAS_MAP: {
5857 struct kvm_s390_ucas_mapping ucasmap;
5858
5859 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
5860 r = -EFAULT;
5861 break;
5862 }
5863
5864 if (!kvm_is_ucontrol(vcpu->kvm)) {
5865 r = -EINVAL;
5866 break;
5867 }
5868
5869 r = gmap_map_segment(vcpu->arch.gmap, ucasmap.user_addr,
5870 ucasmap.vcpu_addr, ucasmap.length);
5871 break;
5872 }
5873 case KVM_S390_UCAS_UNMAP: {
5874 struct kvm_s390_ucas_mapping ucasmap;
5875
5876 if (copy_from_user(&ucasmap, argp, sizeof(ucasmap))) {
5877 r = -EFAULT;
5878 break;
5879 }
5880
5881 if (!kvm_is_ucontrol(vcpu->kvm)) {
5882 r = -EINVAL;
5883 break;
5884 }
5885
5886 r = gmap_unmap_segment(vcpu->arch.gmap, ucasmap.vcpu_addr,
5887 ucasmap.length);
5888 break;
5889 }
5890 #endif
5891 case KVM_S390_VCPU_FAULT: {
5892 idx = srcu_read_lock(&vcpu->kvm->srcu);
5893 r = vcpu_dat_fault_handler(vcpu, arg, 0);
5894 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5895 break;
5896 }
5897 case KVM_ENABLE_CAP:
5898 {
5899 struct kvm_enable_cap cap;
5900 r = -EFAULT;
5901 if (copy_from_user(&cap, argp, sizeof(cap)))
5902 break;
5903 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5904 break;
5905 }
5906 case KVM_S390_MEM_OP: {
5907 struct kvm_s390_mem_op mem_op;
5908
5909 if (copy_from_user(&mem_op, argp, sizeof(mem_op)) == 0)
5910 r = kvm_s390_vcpu_memsida_op(vcpu, &mem_op);
5911 else
5912 r = -EFAULT;
5913 break;
5914 }
5915 case KVM_S390_SET_IRQ_STATE: {
5916 struct kvm_s390_irq_state irq_state;
5917
5918 r = -EFAULT;
5919 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
5920 break;
5921 if (irq_state.len > VCPU_IRQS_MAX_BUF ||
5922 irq_state.len == 0 ||
5923 irq_state.len % sizeof(struct kvm_s390_irq) > 0) {
5924 r = -EINVAL;
5925 break;
5926 }
5927 /* do not use irq_state.flags, it will break old QEMUs */
5928 r = kvm_s390_set_irq_state(vcpu,
5929 (void __user *) irq_state.buf,
5930 irq_state.len);
5931 break;
5932 }
5933 case KVM_S390_GET_IRQ_STATE: {
5934 struct kvm_s390_irq_state irq_state;
5935
5936 r = -EFAULT;
5937 if (copy_from_user(&irq_state, argp, sizeof(irq_state)))
5938 break;
5939 if (irq_state.len == 0) {
5940 r = -EINVAL;
5941 break;
5942 }
5943 /* do not use irq_state.flags, it will break old QEMUs */
5944 r = kvm_s390_get_irq_state(vcpu,
5945 (__u8 __user *) irq_state.buf,
5946 irq_state.len);
5947 break;
5948 }
5949 case KVM_S390_PV_CPU_COMMAND: {
5950 struct kvm_pv_cmd cmd;
5951
5952 r = -EINVAL;
5953 if (!is_prot_virt_host())
5954 break;
5955
5956 r = -EFAULT;
5957 if (copy_from_user(&cmd, argp, sizeof(cmd)))
5958 break;
5959
5960 r = -EINVAL;
5961 if (cmd.flags)
5962 break;
5963
5964 /* We only handle this cmd right now */
5965 if (cmd.cmd != KVM_PV_DUMP)
5966 break;
5967
5968 r = kvm_s390_handle_pv_vcpu_dump(vcpu, &cmd);
5969
5970 /* Always copy over UV rc / rrc data */
5971 if (copy_to_user((__u8 __user *)argp, &cmd.rc,
5972 sizeof(cmd.rc) + sizeof(cmd.rrc)))
5973 r = -EFAULT;
5974 break;
5975 }
5976 default:
5977 r = -ENOTTY;
5978 }
5979
5980 vcpu_put(vcpu);
5981 return r;
5982 }
5983
kvm_arch_vcpu_fault(struct kvm_vcpu * vcpu,struct vm_fault * vmf)5984 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5985 {
5986 #ifdef CONFIG_KVM_S390_UCONTROL
5987 if ((vmf->pgoff == KVM_S390_SIE_PAGE_OFFSET)
5988 && (kvm_is_ucontrol(vcpu->kvm))) {
5989 vmf->page = virt_to_page(vcpu->arch.sie_block);
5990 get_page(vmf->page);
5991 return 0;
5992 }
5993 #endif
5994 return VM_FAULT_SIGBUS;
5995 }
5996
kvm_arch_irqchip_in_kernel(struct kvm * kvm)5997 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
5998 {
5999 return true;
6000 }
6001
6002 /* Section: memory related */
kvm_arch_prepare_memory_region(struct kvm * kvm,const struct kvm_memory_slot * old,struct kvm_memory_slot * new,enum kvm_mr_change change)6003 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6004 const struct kvm_memory_slot *old,
6005 struct kvm_memory_slot *new,
6006 enum kvm_mr_change change)
6007 {
6008 gpa_t size;
6009
6010 if (kvm_is_ucontrol(kvm) && new->id < KVM_USER_MEM_SLOTS)
6011 return -EINVAL;
6012
6013 /* When we are protected, we should not change the memory slots */
6014 if (kvm_s390_pv_get_handle(kvm))
6015 return -EINVAL;
6016
6017 if (change != KVM_MR_DELETE && change != KVM_MR_FLAGS_ONLY) {
6018 /*
6019 * A few sanity checks. We can have memory slots which have to be
6020 * located/ended at a segment boundary (1MB). The memory in userland is
6021 * ok to be fragmented into various different vmas. It is okay to mmap()
6022 * and munmap() stuff in this slot after doing this call at any time
6023 */
6024
6025 if (new->userspace_addr & 0xffffful)
6026 return -EINVAL;
6027
6028 size = new->npages * PAGE_SIZE;
6029 if (size & 0xffffful)
6030 return -EINVAL;
6031
6032 if ((new->base_gfn * PAGE_SIZE) + size > kvm->arch.mem_limit)
6033 return -EINVAL;
6034 }
6035
6036 if (!kvm->arch.migration_mode)
6037 return 0;
6038
6039 /*
6040 * Turn off migration mode when:
6041 * - userspace creates a new memslot with dirty logging off,
6042 * - userspace modifies an existing memslot (MOVE or FLAGS_ONLY) and
6043 * dirty logging is turned off.
6044 * Migration mode expects dirty page logging being enabled to store
6045 * its dirty bitmap.
6046 */
6047 if (change != KVM_MR_DELETE &&
6048 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
6049 WARN(kvm_s390_vm_stop_migration(kvm),
6050 "Failed to stop migration mode");
6051
6052 return 0;
6053 }
6054
kvm_arch_commit_memory_region(struct kvm * kvm,struct kvm_memory_slot * old,const struct kvm_memory_slot * new,enum kvm_mr_change change)6055 void kvm_arch_commit_memory_region(struct kvm *kvm,
6056 struct kvm_memory_slot *old,
6057 const struct kvm_memory_slot *new,
6058 enum kvm_mr_change change)
6059 {
6060 int rc = 0;
6061
6062 if (kvm_is_ucontrol(kvm))
6063 return;
6064
6065 switch (change) {
6066 case KVM_MR_DELETE:
6067 rc = gmap_unmap_segment(kvm->arch.gmap, old->base_gfn * PAGE_SIZE,
6068 old->npages * PAGE_SIZE);
6069 break;
6070 case KVM_MR_MOVE:
6071 rc = gmap_unmap_segment(kvm->arch.gmap, old->base_gfn * PAGE_SIZE,
6072 old->npages * PAGE_SIZE);
6073 if (rc)
6074 break;
6075 fallthrough;
6076 case KVM_MR_CREATE:
6077 rc = gmap_map_segment(kvm->arch.gmap, new->userspace_addr,
6078 new->base_gfn * PAGE_SIZE,
6079 new->npages * PAGE_SIZE);
6080 break;
6081 case KVM_MR_FLAGS_ONLY:
6082 break;
6083 default:
6084 WARN(1, "Unknown KVM MR CHANGE: %d\n", change);
6085 }
6086 if (rc)
6087 pr_warn("failed to commit memory region\n");
6088 return;
6089 }
6090
nonhyp_mask(int i)6091 static inline unsigned long nonhyp_mask(int i)
6092 {
6093 unsigned int nonhyp_fai = (sclp.hmfai << i * 2) >> 30;
6094
6095 return 0x0000ffffffffffffUL >> (nonhyp_fai << 4);
6096 }
6097
kvm_s390_init(void)6098 static int __init kvm_s390_init(void)
6099 {
6100 int i, r;
6101
6102 if (!sclp.has_sief2) {
6103 pr_info("SIE is not available\n");
6104 return -ENODEV;
6105 }
6106
6107 if (nested && hpage) {
6108 pr_info("A KVM host that supports nesting cannot back its KVM guests with huge pages\n");
6109 return -EINVAL;
6110 }
6111
6112 for (i = 0; i < 16; i++)
6113 kvm_s390_fac_base[i] |=
6114 stfle_fac_list[i] & nonhyp_mask(i);
6115
6116 r = __kvm_s390_init();
6117 if (r)
6118 return r;
6119
6120 r = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
6121 if (r) {
6122 __kvm_s390_exit();
6123 return r;
6124 }
6125 return 0;
6126 }
6127
kvm_s390_exit(void)6128 static void __exit kvm_s390_exit(void)
6129 {
6130 kvm_exit();
6131
6132 __kvm_s390_exit();
6133 }
6134
6135 module_init(kvm_s390_init);
6136 module_exit(kvm_s390_exit);
6137
6138 /*
6139 * Enable autoloading of the kvm module.
6140 * Note that we add the module alias here instead of virt/kvm/kvm_main.c
6141 * since x86 takes a different approach.
6142 */
6143 #include <linux/miscdevice.h>
6144 MODULE_ALIAS_MISCDEV(KVM_MINOR);
6145 MODULE_ALIAS("devname:kvm");
6146