Loading...
Note: File does not exist in v3.15.
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * kvm nested virtualization support for s390x
4 *
5 * Copyright IBM Corp. 2016, 2018
6 *
7 * Author(s): David Hildenbrand <dahi@linux.vnet.ibm.com>
8 */
9#include <linux/vmalloc.h>
10#include <linux/kvm_host.h>
11#include <linux/bug.h>
12#include <linux/list.h>
13#include <linux/bitmap.h>
14#include <linux/sched/signal.h>
15
16#include <asm/gmap.h>
17#include <asm/mmu_context.h>
18#include <asm/sclp.h>
19#include <asm/nmi.h>
20#include <asm/dis.h>
21#include <asm/fpu/api.h>
22#include "kvm-s390.h"
23#include "gaccess.h"
24
25struct vsie_page {
26 struct kvm_s390_sie_block scb_s; /* 0x0000 */
27 /*
28 * the backup info for machine check. ensure it's at
29 * the same offset as that in struct sie_page!
30 */
31 struct mcck_volatile_info mcck_info; /* 0x0200 */
32 /*
33 * The pinned original scb. Be aware that other VCPUs can modify
34 * it while we read from it. Values that are used for conditions or
35 * are reused conditionally, should be accessed via READ_ONCE.
36 */
37 struct kvm_s390_sie_block *scb_o; /* 0x0218 */
38 /* the shadow gmap in use by the vsie_page */
39 struct gmap *gmap; /* 0x0220 */
40 /* address of the last reported fault to guest2 */
41 unsigned long fault_addr; /* 0x0228 */
42 /* calculated guest addresses of satellite control blocks */
43 gpa_t sca_gpa; /* 0x0230 */
44 gpa_t itdba_gpa; /* 0x0238 */
45 gpa_t gvrd_gpa; /* 0x0240 */
46 gpa_t riccbd_gpa; /* 0x0248 */
47 gpa_t sdnx_gpa; /* 0x0250 */
48 __u8 reserved[0x0700 - 0x0258]; /* 0x0258 */
49 struct kvm_s390_crypto_cb crycb; /* 0x0700 */
50 __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */
51};
52
53/* trigger a validity icpt for the given scb */
54static int set_validity_icpt(struct kvm_s390_sie_block *scb,
55 __u16 reason_code)
56{
57 scb->ipa = 0x1000;
58 scb->ipb = ((__u32) reason_code) << 16;
59 scb->icptcode = ICPT_VALIDITY;
60 return 1;
61}
62
63/* mark the prefix as unmapped, this will block the VSIE */
64static void prefix_unmapped(struct vsie_page *vsie_page)
65{
66 atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20);
67}
68
69/* mark the prefix as unmapped and wait until the VSIE has been left */
70static void prefix_unmapped_sync(struct vsie_page *vsie_page)
71{
72 prefix_unmapped(vsie_page);
73 if (vsie_page->scb_s.prog0c & PROG_IN_SIE)
74 atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags);
75 while (vsie_page->scb_s.prog0c & PROG_IN_SIE)
76 cpu_relax();
77}
78
79/* mark the prefix as mapped, this will allow the VSIE to run */
80static void prefix_mapped(struct vsie_page *vsie_page)
81{
82 atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20);
83}
84
85/* test if the prefix is mapped into the gmap shadow */
86static int prefix_is_mapped(struct vsie_page *vsie_page)
87{
88 return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST);
89}
90
91/* copy the updated intervention request bits into the shadow scb */
92static void update_intervention_requests(struct vsie_page *vsie_page)
93{
94 const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT;
95 int cpuflags;
96
97 cpuflags = atomic_read(&vsie_page->scb_o->cpuflags);
98 atomic_andnot(bits, &vsie_page->scb_s.cpuflags);
99 atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags);
100}
101
102/* shadow (filter and validate) the cpuflags */
103static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
104{
105 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
106 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
107 int newflags, cpuflags = atomic_read(&scb_o->cpuflags);
108
109 /* we don't allow ESA/390 guests */
110 if (!(cpuflags & CPUSTAT_ZARCH))
111 return set_validity_icpt(scb_s, 0x0001U);
112
113 if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS))
114 return set_validity_icpt(scb_s, 0x0001U);
115 else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR))
116 return set_validity_icpt(scb_s, 0x0007U);
117
118 /* intervention requests will be set later */
119 newflags = CPUSTAT_ZARCH;
120 if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8))
121 newflags |= CPUSTAT_GED;
122 if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) {
123 if (cpuflags & CPUSTAT_GED)
124 return set_validity_icpt(scb_s, 0x0001U);
125 newflags |= CPUSTAT_GED2;
126 }
127 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE))
128 newflags |= cpuflags & CPUSTAT_P;
129 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS))
130 newflags |= cpuflags & CPUSTAT_SM;
131 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS))
132 newflags |= cpuflags & CPUSTAT_IBS;
133 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS))
134 newflags |= cpuflags & CPUSTAT_KSS;
135
136 atomic_set(&scb_s->cpuflags, newflags);
137 return 0;
138}
139/* Copy to APCB FORMAT1 from APCB FORMAT0 */
140static int setup_apcb10(struct kvm_vcpu *vcpu, struct kvm_s390_apcb1 *apcb_s,
141 unsigned long apcb_o, struct kvm_s390_apcb1 *apcb_h)
142{
143 struct kvm_s390_apcb0 tmp;
144
145 if (read_guest_real(vcpu, apcb_o, &tmp, sizeof(struct kvm_s390_apcb0)))
146 return -EFAULT;
147
148 apcb_s->apm[0] = apcb_h->apm[0] & tmp.apm[0];
149 apcb_s->aqm[0] = apcb_h->aqm[0] & tmp.aqm[0] & 0xffff000000000000UL;
150 apcb_s->adm[0] = apcb_h->adm[0] & tmp.adm[0] & 0xffff000000000000UL;
151
152 return 0;
153
154}
155
156/**
157 * setup_apcb00 - Copy to APCB FORMAT0 from APCB FORMAT0
158 * @vcpu: pointer to the virtual CPU
159 * @apcb_s: pointer to start of apcb in the shadow crycb
160 * @apcb_o: pointer to start of original apcb in the guest2
161 * @apcb_h: pointer to start of apcb in the guest1
162 *
163 * Returns 0 and -EFAULT on error reading guest apcb
164 */
165static int setup_apcb00(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
166 unsigned long apcb_o, unsigned long *apcb_h)
167{
168 if (read_guest_real(vcpu, apcb_o, apcb_s,
169 sizeof(struct kvm_s390_apcb0)))
170 return -EFAULT;
171
172 bitmap_and(apcb_s, apcb_s, apcb_h, sizeof(struct kvm_s390_apcb0));
173
174 return 0;
175}
176
177/**
178 * setup_apcb11 - Copy the FORMAT1 APCB from the guest to the shadow CRYCB
179 * @vcpu: pointer to the virtual CPU
180 * @apcb_s: pointer to start of apcb in the shadow crycb
181 * @apcb_o: pointer to start of original guest apcb
182 * @apcb_h: pointer to start of apcb in the host
183 *
184 * Returns 0 and -EFAULT on error reading guest apcb
185 */
186static int setup_apcb11(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
187 unsigned long apcb_o,
188 unsigned long *apcb_h)
189{
190 if (read_guest_real(vcpu, apcb_o, apcb_s,
191 sizeof(struct kvm_s390_apcb1)))
192 return -EFAULT;
193
194 bitmap_and(apcb_s, apcb_s, apcb_h, sizeof(struct kvm_s390_apcb1));
195
196 return 0;
197}
198
199/**
200 * setup_apcb - Create a shadow copy of the apcb.
201 * @vcpu: pointer to the virtual CPU
202 * @crycb_s: pointer to shadow crycb
203 * @crycb_o: pointer to original guest crycb
204 * @crycb_h: pointer to the host crycb
205 * @fmt_o: format of the original guest crycb.
206 * @fmt_h: format of the host crycb.
207 *
208 * Checks the compatibility between the guest and host crycb and calls the
209 * appropriate copy function.
210 *
211 * Return 0 or an error number if the guest and host crycb are incompatible.
212 */
213static int setup_apcb(struct kvm_vcpu *vcpu, struct kvm_s390_crypto_cb *crycb_s,
214 const u32 crycb_o,
215 struct kvm_s390_crypto_cb *crycb_h,
216 int fmt_o, int fmt_h)
217{
218 struct kvm_s390_crypto_cb *crycb;
219
220 crycb = (struct kvm_s390_crypto_cb *) (unsigned long)crycb_o;
221
222 switch (fmt_o) {
223 case CRYCB_FORMAT2:
224 if ((crycb_o & PAGE_MASK) != ((crycb_o + 256) & PAGE_MASK))
225 return -EACCES;
226 if (fmt_h != CRYCB_FORMAT2)
227 return -EINVAL;
228 return setup_apcb11(vcpu, (unsigned long *)&crycb_s->apcb1,
229 (unsigned long) &crycb->apcb1,
230 (unsigned long *)&crycb_h->apcb1);
231 case CRYCB_FORMAT1:
232 switch (fmt_h) {
233 case CRYCB_FORMAT2:
234 return setup_apcb10(vcpu, &crycb_s->apcb1,
235 (unsigned long) &crycb->apcb0,
236 &crycb_h->apcb1);
237 case CRYCB_FORMAT1:
238 return setup_apcb00(vcpu,
239 (unsigned long *) &crycb_s->apcb0,
240 (unsigned long) &crycb->apcb0,
241 (unsigned long *) &crycb_h->apcb0);
242 }
243 break;
244 case CRYCB_FORMAT0:
245 if ((crycb_o & PAGE_MASK) != ((crycb_o + 32) & PAGE_MASK))
246 return -EACCES;
247
248 switch (fmt_h) {
249 case CRYCB_FORMAT2:
250 return setup_apcb10(vcpu, &crycb_s->apcb1,
251 (unsigned long) &crycb->apcb0,
252 &crycb_h->apcb1);
253 case CRYCB_FORMAT1:
254 case CRYCB_FORMAT0:
255 return setup_apcb00(vcpu,
256 (unsigned long *) &crycb_s->apcb0,
257 (unsigned long) &crycb->apcb0,
258 (unsigned long *) &crycb_h->apcb0);
259 }
260 }
261 return -EINVAL;
262}
263
264/**
265 * shadow_crycb - Create a shadow copy of the crycb block
266 * @vcpu: a pointer to the virtual CPU
267 * @vsie_page: a pointer to internal date used for the vSIE
268 *
269 * Create a shadow copy of the crycb block and setup key wrapping, if
270 * requested for guest 3 and enabled for guest 2.
271 *
272 * We accept format-1 or format-2, but we convert format-1 into format-2
273 * in the shadow CRYCB.
274 * Using format-2 enables the firmware to choose the right format when
275 * scheduling the SIE.
276 * There is nothing to do for format-0.
277 *
278 * This function centralize the issuing of set_validity_icpt() for all
279 * the subfunctions working on the crycb.
280 *
281 * Returns: - 0 if shadowed or nothing to do
282 * - > 0 if control has to be given to guest 2
283 */
284static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
285{
286 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
287 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
288 const uint32_t crycbd_o = READ_ONCE(scb_o->crycbd);
289 const u32 crycb_addr = crycbd_o & 0x7ffffff8U;
290 unsigned long *b1, *b2;
291 u8 ecb3_flags;
292 u32 ecd_flags;
293 int apie_h;
294 int apie_s;
295 int key_msk = test_kvm_facility(vcpu->kvm, 76);
296 int fmt_o = crycbd_o & CRYCB_FORMAT_MASK;
297 int fmt_h = vcpu->arch.sie_block->crycbd & CRYCB_FORMAT_MASK;
298 int ret = 0;
299
300 scb_s->crycbd = 0;
301
302 apie_h = vcpu->arch.sie_block->eca & ECA_APIE;
303 apie_s = apie_h & scb_o->eca;
304 if (!apie_s && (!key_msk || (fmt_o == CRYCB_FORMAT0)))
305 return 0;
306
307 if (!crycb_addr)
308 return set_validity_icpt(scb_s, 0x0039U);
309
310 if (fmt_o == CRYCB_FORMAT1)
311 if ((crycb_addr & PAGE_MASK) !=
312 ((crycb_addr + 128) & PAGE_MASK))
313 return set_validity_icpt(scb_s, 0x003CU);
314
315 if (apie_s) {
316 ret = setup_apcb(vcpu, &vsie_page->crycb, crycb_addr,
317 vcpu->kvm->arch.crypto.crycb,
318 fmt_o, fmt_h);
319 if (ret)
320 goto end;
321 scb_s->eca |= scb_o->eca & ECA_APIE;
322 }
323
324 /* we may only allow it if enabled for guest 2 */
325 ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 &
326 (ECB3_AES | ECB3_DEA);
327 ecd_flags = scb_o->ecd & vcpu->arch.sie_block->ecd & ECD_ECC;
328 if (!ecb3_flags && !ecd_flags)
329 goto end;
330
331 /* copy only the wrapping keys */
332 if (read_guest_real(vcpu, crycb_addr + 72,
333 vsie_page->crycb.dea_wrapping_key_mask, 56))
334 return set_validity_icpt(scb_s, 0x0035U);
335
336 scb_s->ecb3 |= ecb3_flags;
337 scb_s->ecd |= ecd_flags;
338
339 /* xor both blocks in one run */
340 b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;
341 b2 = (unsigned long *)
342 vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask;
343 /* as 56%8 == 0, bitmap_xor won't overwrite any data */
344 bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56);
345end:
346 switch (ret) {
347 case -EINVAL:
348 return set_validity_icpt(scb_s, 0x0022U);
349 case -EFAULT:
350 return set_validity_icpt(scb_s, 0x0035U);
351 case -EACCES:
352 return set_validity_icpt(scb_s, 0x003CU);
353 }
354 scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT2;
355 return 0;
356}
357
358/* shadow (round up/down) the ibc to avoid validity icpt */
359static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
360{
361 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
362 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
363 /* READ_ONCE does not work on bitfields - use a temporary variable */
364 const uint32_t __new_ibc = scb_o->ibc;
365 const uint32_t new_ibc = READ_ONCE(__new_ibc) & 0x0fffU;
366 __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU;
367
368 scb_s->ibc = 0;
369 /* ibc installed in g2 and requested for g3 */
370 if (vcpu->kvm->arch.model.ibc && new_ibc) {
371 scb_s->ibc = new_ibc;
372 /* takte care of the minimum ibc level of the machine */
373 if (scb_s->ibc < min_ibc)
374 scb_s->ibc = min_ibc;
375 /* take care of the maximum ibc level set for the guest */
376 if (scb_s->ibc > vcpu->kvm->arch.model.ibc)
377 scb_s->ibc = vcpu->kvm->arch.model.ibc;
378 }
379}
380
381/* unshadow the scb, copying parameters back to the real scb */
382static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
383{
384 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
385 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
386
387 /* interception */
388 scb_o->icptcode = scb_s->icptcode;
389 scb_o->icptstatus = scb_s->icptstatus;
390 scb_o->ipa = scb_s->ipa;
391 scb_o->ipb = scb_s->ipb;
392 scb_o->gbea = scb_s->gbea;
393
394 /* timer */
395 scb_o->cputm = scb_s->cputm;
396 scb_o->ckc = scb_s->ckc;
397 scb_o->todpr = scb_s->todpr;
398
399 /* guest state */
400 scb_o->gpsw = scb_s->gpsw;
401 scb_o->gg14 = scb_s->gg14;
402 scb_o->gg15 = scb_s->gg15;
403 memcpy(scb_o->gcr, scb_s->gcr, 128);
404 scb_o->pp = scb_s->pp;
405
406 /* branch prediction */
407 if (test_kvm_facility(vcpu->kvm, 82)) {
408 scb_o->fpf &= ~FPF_BPBC;
409 scb_o->fpf |= scb_s->fpf & FPF_BPBC;
410 }
411
412 /* interrupt intercept */
413 switch (scb_s->icptcode) {
414 case ICPT_PROGI:
415 case ICPT_INSTPROGI:
416 case ICPT_EXTINT:
417 memcpy((void *)((u64)scb_o + 0xc0),
418 (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
419 break;
420 }
421
422 if (scb_s->ihcpu != 0xffffU)
423 scb_o->ihcpu = scb_s->ihcpu;
424}
425
426/*
427 * Setup the shadow scb by copying and checking the relevant parts of the g2
428 * provided scb.
429 *
430 * Returns: - 0 if the scb has been shadowed
431 * - > 0 if control has to be given to guest 2
432 */
433static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
434{
435 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
436 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
437 /* READ_ONCE does not work on bitfields - use a temporary variable */
438 const uint32_t __new_prefix = scb_o->prefix;
439 const uint32_t new_prefix = READ_ONCE(__new_prefix);
440 const bool wants_tx = READ_ONCE(scb_o->ecb) & ECB_TE;
441 bool had_tx = scb_s->ecb & ECB_TE;
442 unsigned long new_mso = 0;
443 int rc;
444
445 /* make sure we don't have any leftovers when reusing the scb */
446 scb_s->icptcode = 0;
447 scb_s->eca = 0;
448 scb_s->ecb = 0;
449 scb_s->ecb2 = 0;
450 scb_s->ecb3 = 0;
451 scb_s->ecd = 0;
452 scb_s->fac = 0;
453 scb_s->fpf = 0;
454
455 rc = prepare_cpuflags(vcpu, vsie_page);
456 if (rc)
457 goto out;
458
459 /* timer */
460 scb_s->cputm = scb_o->cputm;
461 scb_s->ckc = scb_o->ckc;
462 scb_s->todpr = scb_o->todpr;
463 scb_s->epoch = scb_o->epoch;
464
465 /* guest state */
466 scb_s->gpsw = scb_o->gpsw;
467 scb_s->gg14 = scb_o->gg14;
468 scb_s->gg15 = scb_o->gg15;
469 memcpy(scb_s->gcr, scb_o->gcr, 128);
470 scb_s->pp = scb_o->pp;
471
472 /* interception / execution handling */
473 scb_s->gbea = scb_o->gbea;
474 scb_s->lctl = scb_o->lctl;
475 scb_s->svcc = scb_o->svcc;
476 scb_s->ictl = scb_o->ictl;
477 /*
478 * SKEY handling functions can't deal with false setting of PTE invalid
479 * bits. Therefore we cannot provide interpretation and would later
480 * have to provide own emulation handlers.
481 */
482 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS))
483 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
484
485 scb_s->icpua = scb_o->icpua;
486
487 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM))
488 new_mso = READ_ONCE(scb_o->mso) & 0xfffffffffff00000UL;
489 /* if the hva of the prefix changes, we have to remap the prefix */
490 if (scb_s->mso != new_mso || scb_s->prefix != new_prefix)
491 prefix_unmapped(vsie_page);
492 /* SIE will do mso/msl validity and exception checks for us */
493 scb_s->msl = scb_o->msl & 0xfffffffffff00000UL;
494 scb_s->mso = new_mso;
495 scb_s->prefix = new_prefix;
496
497 /* We have to definetly flush the tlb if this scb never ran */
498 if (scb_s->ihcpu != 0xffffU)
499 scb_s->ihcpu = scb_o->ihcpu;
500
501 /* MVPG and Protection Exception Interpretation are always available */
502 scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI);
503 /* Host-protection-interruption introduced with ESOP */
504 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
505 scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT;
506 /*
507 * CPU Topology
508 * This facility only uses the utility field of the SCA and none of
509 * the cpu entries that are problematic with the other interpretation
510 * facilities so we can pass it through
511 */
512 if (test_kvm_facility(vcpu->kvm, 11))
513 scb_s->ecb |= scb_o->ecb & ECB_PTF;
514 /* transactional execution */
515 if (test_kvm_facility(vcpu->kvm, 73) && wants_tx) {
516 /* remap the prefix is tx is toggled on */
517 if (!had_tx)
518 prefix_unmapped(vsie_page);
519 scb_s->ecb |= ECB_TE;
520 }
521 /* specification exception interpretation */
522 scb_s->ecb |= scb_o->ecb & ECB_SPECI;
523 /* branch prediction */
524 if (test_kvm_facility(vcpu->kvm, 82))
525 scb_s->fpf |= scb_o->fpf & FPF_BPBC;
526 /* SIMD */
527 if (test_kvm_facility(vcpu->kvm, 129)) {
528 scb_s->eca |= scb_o->eca & ECA_VX;
529 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
530 }
531 /* Run-time-Instrumentation */
532 if (test_kvm_facility(vcpu->kvm, 64))
533 scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI;
534 /* Instruction Execution Prevention */
535 if (test_kvm_facility(vcpu->kvm, 130))
536 scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP;
537 /* Guarded Storage */
538 if (test_kvm_facility(vcpu->kvm, 133)) {
539 scb_s->ecb |= scb_o->ecb & ECB_GS;
540 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
541 }
542 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
543 scb_s->eca |= scb_o->eca & ECA_SII;
544 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
545 scb_s->eca |= scb_o->eca & ECA_IB;
546 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
547 scb_s->eca |= scb_o->eca & ECA_CEI;
548 /* Epoch Extension */
549 if (test_kvm_facility(vcpu->kvm, 139)) {
550 scb_s->ecd |= scb_o->ecd & ECD_MEF;
551 scb_s->epdx = scb_o->epdx;
552 }
553
554 /* etoken */
555 if (test_kvm_facility(vcpu->kvm, 156))
556 scb_s->ecd |= scb_o->ecd & ECD_ETOKENF;
557
558 scb_s->hpid = HPID_VSIE;
559 scb_s->cpnc = scb_o->cpnc;
560
561 prepare_ibc(vcpu, vsie_page);
562 rc = shadow_crycb(vcpu, vsie_page);
563out:
564 if (rc)
565 unshadow_scb(vcpu, vsie_page);
566 return rc;
567}
568
569void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
570 unsigned long end)
571{
572 struct kvm *kvm = gmap->private;
573 struct vsie_page *cur;
574 unsigned long prefix;
575 struct page *page;
576 int i;
577
578 if (!gmap_is_shadow(gmap))
579 return;
580 if (start >= 1UL << 31)
581 /* We are only interested in prefix pages */
582 return;
583
584 /*
585 * Only new shadow blocks are added to the list during runtime,
586 * therefore we can safely reference them all the time.
587 */
588 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
589 page = READ_ONCE(kvm->arch.vsie.pages[i]);
590 if (!page)
591 continue;
592 cur = page_to_virt(page);
593 if (READ_ONCE(cur->gmap) != gmap)
594 continue;
595 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
596 /* with mso/msl, the prefix lies at an offset */
597 prefix += cur->scb_s.mso;
598 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
599 prefix_unmapped_sync(cur);
600 }
601}
602
603/*
604 * Map the first prefix page and if tx is enabled also the second prefix page.
605 *
606 * The prefix will be protected, a gmap notifier will inform about unmaps.
607 * The shadow scb must not be executed until the prefix is remapped, this is
608 * guaranteed by properly handling PROG_REQUEST.
609 *
610 * Returns: - 0 on if successfully mapped or already mapped
611 * - > 0 if control has to be given to guest 2
612 * - -EAGAIN if the caller can retry immediately
613 * - -ENOMEM if out of memory
614 */
615static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
616{
617 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
618 u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
619 int rc;
620
621 if (prefix_is_mapped(vsie_page))
622 return 0;
623
624 /* mark it as mapped so we can catch any concurrent unmappers */
625 prefix_mapped(vsie_page);
626
627 /* with mso/msl, the prefix lies at offset *mso* */
628 prefix += scb_s->mso;
629
630 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix, NULL);
631 if (!rc && (scb_s->ecb & ECB_TE))
632 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
633 prefix + PAGE_SIZE, NULL);
634 /*
635 * We don't have to mprotect, we will be called for all unshadows.
636 * SIE will detect if protection applies and trigger a validity.
637 */
638 if (rc)
639 prefix_unmapped(vsie_page);
640 if (rc > 0 || rc == -EFAULT)
641 rc = set_validity_icpt(scb_s, 0x0037U);
642 return rc;
643}
644
645/*
646 * Pin the guest page given by gpa and set hpa to the pinned host address.
647 * Will always be pinned writable.
648 *
649 * Returns: - 0 on success
650 * - -EINVAL if the gpa is not valid guest storage
651 */
652static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
653{
654 struct page *page;
655
656 page = gfn_to_page(kvm, gpa_to_gfn(gpa));
657 if (is_error_page(page))
658 return -EINVAL;
659 *hpa = (hpa_t)page_to_phys(page) + (gpa & ~PAGE_MASK);
660 return 0;
661}
662
663/* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
664static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa)
665{
666 kvm_release_pfn_dirty(hpa >> PAGE_SHIFT);
667 /* mark the page always as dirty for migration */
668 mark_page_dirty(kvm, gpa_to_gfn(gpa));
669}
670
671/* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
672static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
673{
674 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
675 hpa_t hpa;
676
677 hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
678 if (hpa) {
679 unpin_guest_page(vcpu->kvm, vsie_page->sca_gpa, hpa);
680 vsie_page->sca_gpa = 0;
681 scb_s->scaol = 0;
682 scb_s->scaoh = 0;
683 }
684
685 hpa = scb_s->itdba;
686 if (hpa) {
687 unpin_guest_page(vcpu->kvm, vsie_page->itdba_gpa, hpa);
688 vsie_page->itdba_gpa = 0;
689 scb_s->itdba = 0;
690 }
691
692 hpa = scb_s->gvrd;
693 if (hpa) {
694 unpin_guest_page(vcpu->kvm, vsie_page->gvrd_gpa, hpa);
695 vsie_page->gvrd_gpa = 0;
696 scb_s->gvrd = 0;
697 }
698
699 hpa = scb_s->riccbd;
700 if (hpa) {
701 unpin_guest_page(vcpu->kvm, vsie_page->riccbd_gpa, hpa);
702 vsie_page->riccbd_gpa = 0;
703 scb_s->riccbd = 0;
704 }
705
706 hpa = scb_s->sdnxo;
707 if (hpa) {
708 unpin_guest_page(vcpu->kvm, vsie_page->sdnx_gpa, hpa);
709 vsie_page->sdnx_gpa = 0;
710 scb_s->sdnxo = 0;
711 }
712}
713
714/*
715 * Instead of shadowing some blocks, we can simply forward them because the
716 * addresses in the scb are 64 bit long.
717 *
718 * This works as long as the data lies in one page. If blocks ever exceed one
719 * page, we have to fall back to shadowing.
720 *
721 * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
722 * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
723 *
724 * Returns: - 0 if all blocks were pinned.
725 * - > 0 if control has to be given to guest 2
726 * - -ENOMEM if out of memory
727 */
728static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
729{
730 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
731 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
732 hpa_t hpa;
733 gpa_t gpa;
734 int rc = 0;
735
736 gpa = READ_ONCE(scb_o->scaol) & ~0xfUL;
737 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
738 gpa |= (u64) READ_ONCE(scb_o->scaoh) << 32;
739 if (gpa) {
740 if (gpa < 2 * PAGE_SIZE)
741 rc = set_validity_icpt(scb_s, 0x0038U);
742 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
743 rc = set_validity_icpt(scb_s, 0x0011U);
744 else if ((gpa & PAGE_MASK) !=
745 ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK))
746 rc = set_validity_icpt(scb_s, 0x003bU);
747 if (!rc) {
748 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
749 if (rc)
750 rc = set_validity_icpt(scb_s, 0x0034U);
751 }
752 if (rc)
753 goto unpin;
754 vsie_page->sca_gpa = gpa;
755 scb_s->scaoh = (u32)((u64)hpa >> 32);
756 scb_s->scaol = (u32)(u64)hpa;
757 }
758
759 gpa = READ_ONCE(scb_o->itdba) & ~0xffUL;
760 if (gpa && (scb_s->ecb & ECB_TE)) {
761 if (gpa < 2 * PAGE_SIZE) {
762 rc = set_validity_icpt(scb_s, 0x0080U);
763 goto unpin;
764 }
765 /* 256 bytes cannot cross page boundaries */
766 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
767 if (rc) {
768 rc = set_validity_icpt(scb_s, 0x0080U);
769 goto unpin;
770 }
771 vsie_page->itdba_gpa = gpa;
772 scb_s->itdba = hpa;
773 }
774
775 gpa = READ_ONCE(scb_o->gvrd) & ~0x1ffUL;
776 if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) {
777 if (gpa < 2 * PAGE_SIZE) {
778 rc = set_validity_icpt(scb_s, 0x1310U);
779 goto unpin;
780 }
781 /*
782 * 512 bytes vector registers cannot cross page boundaries
783 * if this block gets bigger, we have to shadow it.
784 */
785 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
786 if (rc) {
787 rc = set_validity_icpt(scb_s, 0x1310U);
788 goto unpin;
789 }
790 vsie_page->gvrd_gpa = gpa;
791 scb_s->gvrd = hpa;
792 }
793
794 gpa = READ_ONCE(scb_o->riccbd) & ~0x3fUL;
795 if (gpa && (scb_s->ecb3 & ECB3_RI)) {
796 if (gpa < 2 * PAGE_SIZE) {
797 rc = set_validity_icpt(scb_s, 0x0043U);
798 goto unpin;
799 }
800 /* 64 bytes cannot cross page boundaries */
801 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
802 if (rc) {
803 rc = set_validity_icpt(scb_s, 0x0043U);
804 goto unpin;
805 }
806 /* Validity 0x0044 will be checked by SIE */
807 vsie_page->riccbd_gpa = gpa;
808 scb_s->riccbd = hpa;
809 }
810 if (((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) ||
811 (scb_s->ecd & ECD_ETOKENF)) {
812 unsigned long sdnxc;
813
814 gpa = READ_ONCE(scb_o->sdnxo) & ~0xfUL;
815 sdnxc = READ_ONCE(scb_o->sdnxo) & 0xfUL;
816 if (!gpa || gpa < 2 * PAGE_SIZE) {
817 rc = set_validity_icpt(scb_s, 0x10b0U);
818 goto unpin;
819 }
820 if (sdnxc < 6 || sdnxc > 12) {
821 rc = set_validity_icpt(scb_s, 0x10b1U);
822 goto unpin;
823 }
824 if (gpa & ((1 << sdnxc) - 1)) {
825 rc = set_validity_icpt(scb_s, 0x10b2U);
826 goto unpin;
827 }
828 /* Due to alignment rules (checked above) this cannot
829 * cross page boundaries
830 */
831 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
832 if (rc) {
833 rc = set_validity_icpt(scb_s, 0x10b0U);
834 goto unpin;
835 }
836 vsie_page->sdnx_gpa = gpa;
837 scb_s->sdnxo = hpa | sdnxc;
838 }
839 return 0;
840unpin:
841 unpin_blocks(vcpu, vsie_page);
842 return rc;
843}
844
845/* unpin the scb provided by guest 2, marking it as dirty */
846static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
847 gpa_t gpa)
848{
849 hpa_t hpa = (hpa_t) vsie_page->scb_o;
850
851 if (hpa)
852 unpin_guest_page(vcpu->kvm, gpa, hpa);
853 vsie_page->scb_o = NULL;
854}
855
856/*
857 * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
858 *
859 * Returns: - 0 if the scb was pinned.
860 * - > 0 if control has to be given to guest 2
861 */
862static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
863 gpa_t gpa)
864{
865 hpa_t hpa;
866 int rc;
867
868 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
869 if (rc) {
870 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
871 WARN_ON_ONCE(rc);
872 return 1;
873 }
874 vsie_page->scb_o = phys_to_virt(hpa);
875 return 0;
876}
877
878/*
879 * Inject a fault into guest 2.
880 *
881 * Returns: - > 0 if control has to be given to guest 2
882 * < 0 if an error occurred during injection.
883 */
884static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
885 bool write_flag)
886{
887 struct kvm_s390_pgm_info pgm = {
888 .code = code,
889 .trans_exc_code =
890 /* 0-51: virtual address */
891 (vaddr & 0xfffffffffffff000UL) |
892 /* 52-53: store / fetch */
893 (((unsigned int) !write_flag) + 1) << 10,
894 /* 62-63: asce id (alway primary == 0) */
895 .exc_access_id = 0, /* always primary */
896 .op_access_id = 0, /* not MVPG */
897 };
898 int rc;
899
900 if (code == PGM_PROTECTION)
901 pgm.trans_exc_code |= 0x4UL;
902
903 rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
904 return rc ? rc : 1;
905}
906
907/*
908 * Handle a fault during vsie execution on a gmap shadow.
909 *
910 * Returns: - 0 if the fault was resolved
911 * - > 0 if control has to be given to guest 2
912 * - < 0 if an error occurred
913 */
914static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
915{
916 int rc;
917
918 if (current->thread.gmap_int_code == PGM_PROTECTION)
919 /* we can directly forward all protection exceptions */
920 return inject_fault(vcpu, PGM_PROTECTION,
921 current->thread.gmap_addr, 1);
922
923 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
924 current->thread.gmap_addr, NULL);
925 if (rc > 0) {
926 rc = inject_fault(vcpu, rc,
927 current->thread.gmap_addr,
928 current->thread.gmap_write_flag);
929 if (rc >= 0)
930 vsie_page->fault_addr = current->thread.gmap_addr;
931 }
932 return rc;
933}
934
935/*
936 * Retry the previous fault that required guest 2 intervention. This avoids
937 * one superfluous SIE re-entry and direct exit.
938 *
939 * Will ignore any errors. The next SIE fault will do proper fault handling.
940 */
941static void handle_last_fault(struct kvm_vcpu *vcpu,
942 struct vsie_page *vsie_page)
943{
944 if (vsie_page->fault_addr)
945 kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
946 vsie_page->fault_addr, NULL);
947 vsie_page->fault_addr = 0;
948}
949
950static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
951{
952 vsie_page->scb_s.icptcode = 0;
953}
954
955/* rewind the psw and clear the vsie icpt, so we can retry execution */
956static void retry_vsie_icpt(struct vsie_page *vsie_page)
957{
958 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
959 int ilen = insn_length(scb_s->ipa >> 8);
960
961 /* take care of EXECUTE instructions */
962 if (scb_s->icptstatus & 1) {
963 ilen = (scb_s->icptstatus >> 4) & 0x6;
964 if (!ilen)
965 ilen = 4;
966 }
967 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
968 clear_vsie_icpt(vsie_page);
969}
970
971/*
972 * Try to shadow + enable the guest 2 provided facility list.
973 * Retry instruction execution if enabled for and provided by guest 2.
974 *
975 * Returns: - 0 if handled (retry or guest 2 icpt)
976 * - > 0 if control has to be given to guest 2
977 */
978static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
979{
980 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
981 __u32 fac = READ_ONCE(vsie_page->scb_o->fac) & 0x7ffffff8U;
982
983 if (fac && test_kvm_facility(vcpu->kvm, 7)) {
984 retry_vsie_icpt(vsie_page);
985 if (read_guest_real(vcpu, fac, &vsie_page->fac,
986 sizeof(vsie_page->fac)))
987 return set_validity_icpt(scb_s, 0x1090U);
988 scb_s->fac = (__u32)(__u64) &vsie_page->fac;
989 }
990 return 0;
991}
992
993/*
994 * Get a register for a nested guest.
995 * @vcpu the vcpu of the guest
996 * @vsie_page the vsie_page for the nested guest
997 * @reg the register number, the upper 4 bits are ignored.
998 * returns: the value of the register.
999 */
1000static u64 vsie_get_register(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, u8 reg)
1001{
1002 /* no need to validate the parameter and/or perform error handling */
1003 reg &= 0xf;
1004 switch (reg) {
1005 case 15:
1006 return vsie_page->scb_s.gg15;
1007 case 14:
1008 return vsie_page->scb_s.gg14;
1009 default:
1010 return vcpu->run->s.regs.gprs[reg];
1011 }
1012}
1013
1014static int vsie_handle_mvpg(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1015{
1016 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1017 unsigned long pei_dest, pei_src, src, dest, mask, prefix;
1018 u64 *pei_block = &vsie_page->scb_o->mcic;
1019 int edat, rc_dest, rc_src;
1020 union ctlreg0 cr0;
1021
1022 cr0.val = vcpu->arch.sie_block->gcr[0];
1023 edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1024 mask = _kvm_s390_logical_to_effective(&scb_s->gpsw, PAGE_MASK);
1025 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
1026
1027 dest = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 20) & mask;
1028 dest = _kvm_s390_real_to_abs(prefix, dest) + scb_s->mso;
1029 src = vsie_get_register(vcpu, vsie_page, scb_s->ipb >> 16) & mask;
1030 src = _kvm_s390_real_to_abs(prefix, src) + scb_s->mso;
1031
1032 rc_dest = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, dest, &pei_dest);
1033 rc_src = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, src, &pei_src);
1034 /*
1035 * Either everything went well, or something non-critical went wrong
1036 * e.g. because of a race. In either case, simply retry.
1037 */
1038 if (rc_dest == -EAGAIN || rc_src == -EAGAIN || (!rc_dest && !rc_src)) {
1039 retry_vsie_icpt(vsie_page);
1040 return -EAGAIN;
1041 }
1042 /* Something more serious went wrong, propagate the error */
1043 if (rc_dest < 0)
1044 return rc_dest;
1045 if (rc_src < 0)
1046 return rc_src;
1047
1048 /* The only possible suppressing exception: just deliver it */
1049 if (rc_dest == PGM_TRANSLATION_SPEC || rc_src == PGM_TRANSLATION_SPEC) {
1050 clear_vsie_icpt(vsie_page);
1051 rc_dest = kvm_s390_inject_program_int(vcpu, PGM_TRANSLATION_SPEC);
1052 WARN_ON_ONCE(rc_dest);
1053 return 1;
1054 }
1055
1056 /*
1057 * Forward the PEI intercept to the guest if it was a page fault, or
1058 * also for segment and region table faults if EDAT applies.
1059 */
1060 if (edat) {
1061 rc_dest = rc_dest == PGM_ASCE_TYPE ? rc_dest : 0;
1062 rc_src = rc_src == PGM_ASCE_TYPE ? rc_src : 0;
1063 } else {
1064 rc_dest = rc_dest != PGM_PAGE_TRANSLATION ? rc_dest : 0;
1065 rc_src = rc_src != PGM_PAGE_TRANSLATION ? rc_src : 0;
1066 }
1067 if (!rc_dest && !rc_src) {
1068 pei_block[0] = pei_dest;
1069 pei_block[1] = pei_src;
1070 return 1;
1071 }
1072
1073 retry_vsie_icpt(vsie_page);
1074
1075 /*
1076 * The host has edat, and the guest does not, or it was an ASCE type
1077 * exception. The host needs to inject the appropriate DAT interrupts
1078 * into the guest.
1079 */
1080 if (rc_dest)
1081 return inject_fault(vcpu, rc_dest, dest, 1);
1082 return inject_fault(vcpu, rc_src, src, 0);
1083}
1084
1085/*
1086 * Run the vsie on a shadow scb and a shadow gmap, without any further
1087 * sanity checks, handling SIE faults.
1088 *
1089 * Returns: - 0 everything went fine
1090 * - > 0 if control has to be given to guest 2
1091 * - < 0 if an error occurred
1092 */
1093static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1094 __releases(vcpu->kvm->srcu)
1095 __acquires(vcpu->kvm->srcu)
1096{
1097 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1098 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
1099 int guest_bp_isolation;
1100 int rc = 0;
1101
1102 handle_last_fault(vcpu, vsie_page);
1103
1104 kvm_vcpu_srcu_read_unlock(vcpu);
1105
1106 /* save current guest state of bp isolation override */
1107 guest_bp_isolation = test_thread_flag(TIF_ISOLATE_BP_GUEST);
1108
1109 /*
1110 * The guest is running with BPBC, so we have to force it on for our
1111 * nested guest. This is done by enabling BPBC globally, so the BPBC
1112 * control in the SCB (which the nested guest can modify) is simply
1113 * ignored.
1114 */
1115 if (test_kvm_facility(vcpu->kvm, 82) &&
1116 vcpu->arch.sie_block->fpf & FPF_BPBC)
1117 set_thread_flag(TIF_ISOLATE_BP_GUEST);
1118
1119 local_irq_disable();
1120 guest_enter_irqoff();
1121 local_irq_enable();
1122
1123 /*
1124 * Simulate a SIE entry of the VCPU (see sie64a), so VCPU blocking
1125 * and VCPU requests also hinder the vSIE from running and lead
1126 * to an immediate exit. kvm_s390_vsie_kick() has to be used to
1127 * also kick the vSIE.
1128 */
1129 vcpu->arch.sie_block->prog0c |= PROG_IN_SIE;
1130 barrier();
1131 if (test_cpu_flag(CIF_FPU))
1132 load_fpu_regs();
1133 if (!kvm_s390_vcpu_sie_inhibited(vcpu))
1134 rc = sie64a(scb_s, vcpu->run->s.regs.gprs);
1135 barrier();
1136 vcpu->arch.sie_block->prog0c &= ~PROG_IN_SIE;
1137
1138 local_irq_disable();
1139 guest_exit_irqoff();
1140 local_irq_enable();
1141
1142 /* restore guest state for bp isolation override */
1143 if (!guest_bp_isolation)
1144 clear_thread_flag(TIF_ISOLATE_BP_GUEST);
1145
1146 kvm_vcpu_srcu_read_lock(vcpu);
1147
1148 if (rc == -EINTR) {
1149 VCPU_EVENT(vcpu, 3, "%s", "machine check");
1150 kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info);
1151 return 0;
1152 }
1153
1154 if (rc > 0)
1155 rc = 0; /* we could still have an icpt */
1156 else if (rc == -EFAULT)
1157 return handle_fault(vcpu, vsie_page);
1158
1159 switch (scb_s->icptcode) {
1160 case ICPT_INST:
1161 if (scb_s->ipa == 0xb2b0)
1162 rc = handle_stfle(vcpu, vsie_page);
1163 break;
1164 case ICPT_STOP:
1165 /* stop not requested by g2 - must have been a kick */
1166 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT))
1167 clear_vsie_icpt(vsie_page);
1168 break;
1169 case ICPT_VALIDITY:
1170 if ((scb_s->ipa & 0xf000) != 0xf000)
1171 scb_s->ipa += 0x1000;
1172 break;
1173 case ICPT_PARTEXEC:
1174 if (scb_s->ipa == 0xb254)
1175 rc = vsie_handle_mvpg(vcpu, vsie_page);
1176 break;
1177 }
1178 return rc;
1179}
1180
1181static void release_gmap_shadow(struct vsie_page *vsie_page)
1182{
1183 if (vsie_page->gmap)
1184 gmap_put(vsie_page->gmap);
1185 WRITE_ONCE(vsie_page->gmap, NULL);
1186 prefix_unmapped(vsie_page);
1187}
1188
1189static int acquire_gmap_shadow(struct kvm_vcpu *vcpu,
1190 struct vsie_page *vsie_page)
1191{
1192 unsigned long asce;
1193 union ctlreg0 cr0;
1194 struct gmap *gmap;
1195 int edat;
1196
1197 asce = vcpu->arch.sie_block->gcr[1];
1198 cr0.val = vcpu->arch.sie_block->gcr[0];
1199 edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1200 edat += edat && test_kvm_facility(vcpu->kvm, 78);
1201
1202 /*
1203 * ASCE or EDAT could have changed since last icpt, or the gmap
1204 * we're holding has been unshadowed. If the gmap is still valid,
1205 * we can safely reuse it.
1206 */
1207 if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat))
1208 return 0;
1209
1210 /* release the old shadow - if any, and mark the prefix as unmapped */
1211 release_gmap_shadow(vsie_page);
1212 gmap = gmap_shadow(vcpu->arch.gmap, asce, edat);
1213 if (IS_ERR(gmap))
1214 return PTR_ERR(gmap);
1215 gmap->private = vcpu->kvm;
1216 WRITE_ONCE(vsie_page->gmap, gmap);
1217 return 0;
1218}
1219
1220/*
1221 * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
1222 */
1223static void register_shadow_scb(struct kvm_vcpu *vcpu,
1224 struct vsie_page *vsie_page)
1225{
1226 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1227
1228 WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s);
1229 /*
1230 * External calls have to lead to a kick of the vcpu and
1231 * therefore the vsie -> Simulate Wait state.
1232 */
1233 kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
1234 /*
1235 * We have to adjust the g3 epoch by the g2 epoch. The epoch will
1236 * automatically be adjusted on tod clock changes via kvm_sync_clock.
1237 */
1238 preempt_disable();
1239 scb_s->epoch += vcpu->kvm->arch.epoch;
1240
1241 if (scb_s->ecd & ECD_MEF) {
1242 scb_s->epdx += vcpu->kvm->arch.epdx;
1243 if (scb_s->epoch < vcpu->kvm->arch.epoch)
1244 scb_s->epdx += 1;
1245 }
1246
1247 preempt_enable();
1248}
1249
1250/*
1251 * Unregister a shadow scb from a VCPU.
1252 */
1253static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
1254{
1255 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
1256 WRITE_ONCE(vcpu->arch.vsie_block, NULL);
1257}
1258
1259/*
1260 * Run the vsie on a shadowed scb, managing the gmap shadow, handling
1261 * prefix pages and faults.
1262 *
1263 * Returns: - 0 if no errors occurred
1264 * - > 0 if control has to be given to guest 2
1265 * - -ENOMEM if out of memory
1266 */
1267static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1268{
1269 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1270 int rc = 0;
1271
1272 while (1) {
1273 rc = acquire_gmap_shadow(vcpu, vsie_page);
1274 if (!rc)
1275 rc = map_prefix(vcpu, vsie_page);
1276 if (!rc) {
1277 gmap_enable(vsie_page->gmap);
1278 update_intervention_requests(vsie_page);
1279 rc = do_vsie_run(vcpu, vsie_page);
1280 gmap_enable(vcpu->arch.gmap);
1281 }
1282 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
1283
1284 if (rc == -EAGAIN)
1285 rc = 0;
1286 if (rc || scb_s->icptcode || signal_pending(current) ||
1287 kvm_s390_vcpu_has_irq(vcpu, 0) ||
1288 kvm_s390_vcpu_sie_inhibited(vcpu))
1289 break;
1290 cond_resched();
1291 }
1292
1293 if (rc == -EFAULT) {
1294 /*
1295 * Addressing exceptions are always presentes as intercepts.
1296 * As addressing exceptions are suppressing and our guest 3 PSW
1297 * points at the responsible instruction, we have to
1298 * forward the PSW and set the ilc. If we can't read guest 3
1299 * instruction, we can use an arbitrary ilc. Let's always use
1300 * ilen = 4 for now, so we can avoid reading in guest 3 virtual
1301 * memory. (we could also fake the shadow so the hardware
1302 * handles it).
1303 */
1304 scb_s->icptcode = ICPT_PROGI;
1305 scb_s->iprcc = PGM_ADDRESSING;
1306 scb_s->pgmilc = 4;
1307 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
1308 rc = 1;
1309 }
1310 return rc;
1311}
1312
1313/*
1314 * Get or create a vsie page for a scb address.
1315 *
1316 * Returns: - address of a vsie page (cached or new one)
1317 * - NULL if the same scb address is already used by another VCPU
1318 * - ERR_PTR(-ENOMEM) if out of memory
1319 */
1320static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
1321{
1322 struct vsie_page *vsie_page;
1323 struct page *page;
1324 int nr_vcpus;
1325
1326 rcu_read_lock();
1327 page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
1328 rcu_read_unlock();
1329 if (page) {
1330 if (page_ref_inc_return(page) == 2)
1331 return page_to_virt(page);
1332 page_ref_dec(page);
1333 }
1334
1335 /*
1336 * We want at least #online_vcpus shadows, so every VCPU can execute
1337 * the VSIE in parallel.
1338 */
1339 nr_vcpus = atomic_read(&kvm->online_vcpus);
1340
1341 mutex_lock(&kvm->arch.vsie.mutex);
1342 if (kvm->arch.vsie.page_count < nr_vcpus) {
1343 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO | GFP_DMA);
1344 if (!page) {
1345 mutex_unlock(&kvm->arch.vsie.mutex);
1346 return ERR_PTR(-ENOMEM);
1347 }
1348 page_ref_inc(page);
1349 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page;
1350 kvm->arch.vsie.page_count++;
1351 } else {
1352 /* reuse an existing entry that belongs to nobody */
1353 while (true) {
1354 page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
1355 if (page_ref_inc_return(page) == 2)
1356 break;
1357 page_ref_dec(page);
1358 kvm->arch.vsie.next++;
1359 kvm->arch.vsie.next %= nr_vcpus;
1360 }
1361 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1362 }
1363 page->index = addr;
1364 /* double use of the same address */
1365 if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) {
1366 page_ref_dec(page);
1367 mutex_unlock(&kvm->arch.vsie.mutex);
1368 return NULL;
1369 }
1370 mutex_unlock(&kvm->arch.vsie.mutex);
1371
1372 vsie_page = page_to_virt(page);
1373 memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
1374 release_gmap_shadow(vsie_page);
1375 vsie_page->fault_addr = 0;
1376 vsie_page->scb_s.ihcpu = 0xffffU;
1377 return vsie_page;
1378}
1379
1380/* put a vsie page acquired via get_vsie_page */
1381static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page)
1382{
1383 struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT);
1384
1385 page_ref_dec(page);
1386}
1387
1388int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1389{
1390 struct vsie_page *vsie_page;
1391 unsigned long scb_addr;
1392 int rc;
1393
1394 vcpu->stat.instruction_sie++;
1395 if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1396 return -EOPNOTSUPP;
1397 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1398 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1399
1400 BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE);
1401 scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1402
1403 /* 512 byte alignment */
1404 if (unlikely(scb_addr & 0x1ffUL))
1405 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1406
1407 if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0) ||
1408 kvm_s390_vcpu_sie_inhibited(vcpu))
1409 return 0;
1410
1411 vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1412 if (IS_ERR(vsie_page))
1413 return PTR_ERR(vsie_page);
1414 else if (!vsie_page)
1415 /* double use of sie control block - simply do nothing */
1416 return 0;
1417
1418 rc = pin_scb(vcpu, vsie_page, scb_addr);
1419 if (rc)
1420 goto out_put;
1421 rc = shadow_scb(vcpu, vsie_page);
1422 if (rc)
1423 goto out_unpin_scb;
1424 rc = pin_blocks(vcpu, vsie_page);
1425 if (rc)
1426 goto out_unshadow;
1427 register_shadow_scb(vcpu, vsie_page);
1428 rc = vsie_run(vcpu, vsie_page);
1429 unregister_shadow_scb(vcpu);
1430 unpin_blocks(vcpu, vsie_page);
1431out_unshadow:
1432 unshadow_scb(vcpu, vsie_page);
1433out_unpin_scb:
1434 unpin_scb(vcpu, vsie_page, scb_addr);
1435out_put:
1436 put_vsie_page(vcpu->kvm, vsie_page);
1437
1438 return rc < 0 ? rc : 0;
1439}
1440
1441/* Init the vsie data structures. To be called when a vm is initialized. */
1442void kvm_s390_vsie_init(struct kvm *kvm)
1443{
1444 mutex_init(&kvm->arch.vsie.mutex);
1445 INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL_ACCOUNT);
1446}
1447
1448/* Destroy the vsie data structures. To be called when a vm is destroyed. */
1449void kvm_s390_vsie_destroy(struct kvm *kvm)
1450{
1451 struct vsie_page *vsie_page;
1452 struct page *page;
1453 int i;
1454
1455 mutex_lock(&kvm->arch.vsie.mutex);
1456 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1457 page = kvm->arch.vsie.pages[i];
1458 kvm->arch.vsie.pages[i] = NULL;
1459 vsie_page = page_to_virt(page);
1460 release_gmap_shadow(vsie_page);
1461 /* free the radix tree entry */
1462 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1463 __free_page(page);
1464 }
1465 kvm->arch.vsie.page_count = 0;
1466 mutex_unlock(&kvm->arch.vsie.mutex);
1467}
1468
1469void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1470{
1471 struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1472
1473 /*
1474 * Even if the VCPU lets go of the shadow sie block reference, it is
1475 * still valid in the cache. So we can safely kick it.
1476 */
1477 if (scb) {
1478 atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1479 if (scb->prog0c & PROG_IN_SIE)
1480 atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);
1481 }
1482}