Loading...
Note: File does not exist in v3.1.
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 "kvm-s390.h"
22#include "gaccess.h"
23
24struct vsie_page {
25 struct kvm_s390_sie_block scb_s; /* 0x0000 */
26 /*
27 * the backup info for machine check. ensure it's at
28 * the same offset as that in struct sie_page!
29 */
30 struct mcck_volatile_info mcck_info; /* 0x0200 */
31 /*
32 * The pinned original scb. Be aware that other VCPUs can modify
33 * it while we read from it. Values that are used for conditions or
34 * are reused conditionally, should be accessed via READ_ONCE.
35 */
36 struct kvm_s390_sie_block *scb_o; /* 0x0218 */
37 /* the shadow gmap in use by the vsie_page */
38 struct gmap *gmap; /* 0x0220 */
39 /* address of the last reported fault to guest2 */
40 unsigned long fault_addr; /* 0x0228 */
41 /* calculated guest addresses of satellite control blocks */
42 gpa_t sca_gpa; /* 0x0230 */
43 gpa_t itdba_gpa; /* 0x0238 */
44 gpa_t gvrd_gpa; /* 0x0240 */
45 gpa_t riccbd_gpa; /* 0x0248 */
46 gpa_t sdnx_gpa; /* 0x0250 */
47 __u8 reserved[0x0700 - 0x0258]; /* 0x0258 */
48 struct kvm_s390_crypto_cb crycb; /* 0x0700 */
49 __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */
50};
51
52/* trigger a validity icpt for the given scb */
53static int set_validity_icpt(struct kvm_s390_sie_block *scb,
54 __u16 reason_code)
55{
56 scb->ipa = 0x1000;
57 scb->ipb = ((__u32) reason_code) << 16;
58 scb->icptcode = ICPT_VALIDITY;
59 return 1;
60}
61
62/* mark the prefix as unmapped, this will block the VSIE */
63static void prefix_unmapped(struct vsie_page *vsie_page)
64{
65 atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20);
66}
67
68/* mark the prefix as unmapped and wait until the VSIE has been left */
69static void prefix_unmapped_sync(struct vsie_page *vsie_page)
70{
71 prefix_unmapped(vsie_page);
72 if (vsie_page->scb_s.prog0c & PROG_IN_SIE)
73 atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags);
74 while (vsie_page->scb_s.prog0c & PROG_IN_SIE)
75 cpu_relax();
76}
77
78/* mark the prefix as mapped, this will allow the VSIE to run */
79static void prefix_mapped(struct vsie_page *vsie_page)
80{
81 atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20);
82}
83
84/* test if the prefix is mapped into the gmap shadow */
85static int prefix_is_mapped(struct vsie_page *vsie_page)
86{
87 return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST);
88}
89
90/* copy the updated intervention request bits into the shadow scb */
91static void update_intervention_requests(struct vsie_page *vsie_page)
92{
93 const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT;
94 int cpuflags;
95
96 cpuflags = atomic_read(&vsie_page->scb_o->cpuflags);
97 atomic_andnot(bits, &vsie_page->scb_s.cpuflags);
98 atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags);
99}
100
101/* shadow (filter and validate) the cpuflags */
102static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
103{
104 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
105 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
106 int newflags, cpuflags = atomic_read(&scb_o->cpuflags);
107
108 /* we don't allow ESA/390 guests */
109 if (!(cpuflags & CPUSTAT_ZARCH))
110 return set_validity_icpt(scb_s, 0x0001U);
111
112 if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS))
113 return set_validity_icpt(scb_s, 0x0001U);
114 else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR))
115 return set_validity_icpt(scb_s, 0x0007U);
116
117 /* intervention requests will be set later */
118 newflags = CPUSTAT_ZARCH;
119 if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8))
120 newflags |= CPUSTAT_GED;
121 if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) {
122 if (cpuflags & CPUSTAT_GED)
123 return set_validity_icpt(scb_s, 0x0001U);
124 newflags |= CPUSTAT_GED2;
125 }
126 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE))
127 newflags |= cpuflags & CPUSTAT_P;
128 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS))
129 newflags |= cpuflags & CPUSTAT_SM;
130 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS))
131 newflags |= cpuflags & CPUSTAT_IBS;
132 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS))
133 newflags |= cpuflags & CPUSTAT_KSS;
134
135 atomic_set(&scb_s->cpuflags, newflags);
136 return 0;
137}
138/* Copy to APCB FORMAT1 from APCB FORMAT0 */
139static int setup_apcb10(struct kvm_vcpu *vcpu, struct kvm_s390_apcb1 *apcb_s,
140 unsigned long apcb_o, struct kvm_s390_apcb1 *apcb_h)
141{
142 struct kvm_s390_apcb0 tmp;
143
144 if (read_guest_real(vcpu, apcb_o, &tmp, sizeof(struct kvm_s390_apcb0)))
145 return -EFAULT;
146
147 apcb_s->apm[0] = apcb_h->apm[0] & tmp.apm[0];
148 apcb_s->aqm[0] = apcb_h->aqm[0] & tmp.aqm[0] & 0xffff000000000000UL;
149 apcb_s->adm[0] = apcb_h->adm[0] & tmp.adm[0] & 0xffff000000000000UL;
150
151 return 0;
152
153}
154
155/**
156 * setup_apcb00 - Copy to APCB FORMAT0 from APCB FORMAT0
157 * @vcpu: pointer to the virtual CPU
158 * @apcb_s: pointer to start of apcb in the shadow crycb
159 * @apcb_o: pointer to start of original apcb in the guest2
160 * @apcb_h: pointer to start of apcb in the guest1
161 *
162 * Returns 0 and -EFAULT on error reading guest apcb
163 */
164static int setup_apcb00(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
165 unsigned long apcb_o, unsigned long *apcb_h)
166{
167 if (read_guest_real(vcpu, apcb_o, apcb_s,
168 sizeof(struct kvm_s390_apcb0)))
169 return -EFAULT;
170
171 bitmap_and(apcb_s, apcb_s, apcb_h, sizeof(struct kvm_s390_apcb0));
172
173 return 0;
174}
175
176/**
177 * setup_apcb11 - Copy the FORMAT1 APCB from the guest to the shadow CRYCB
178 * @vcpu: pointer to the virtual CPU
179 * @apcb_s: pointer to start of apcb in the shadow crycb
180 * @apcb_o: pointer to start of original guest apcb
181 * @apcb_h: pointer to start of apcb in the host
182 *
183 * Returns 0 and -EFAULT on error reading guest apcb
184 */
185static int setup_apcb11(struct kvm_vcpu *vcpu, unsigned long *apcb_s,
186 unsigned long apcb_o,
187 unsigned long *apcb_h)
188{
189 if (read_guest_real(vcpu, apcb_o, apcb_s,
190 sizeof(struct kvm_s390_apcb1)))
191 return -EFAULT;
192
193 bitmap_and(apcb_s, apcb_s, apcb_h, sizeof(struct kvm_s390_apcb1));
194
195 return 0;
196}
197
198/**
199 * setup_apcb - Create a shadow copy of the apcb.
200 * @vcpu: pointer to the virtual CPU
201 * @crycb_s: pointer to shadow crycb
202 * @crycb_o: pointer to original guest crycb
203 * @crycb_h: pointer to the host crycb
204 * @fmt_o: format of the original guest crycb.
205 * @fmt_h: format of the host crycb.
206 *
207 * Checks the compatibility between the guest and host crycb and calls the
208 * appropriate copy function.
209 *
210 * Return 0 or an error number if the guest and host crycb are incompatible.
211 */
212static int setup_apcb(struct kvm_vcpu *vcpu, struct kvm_s390_crypto_cb *crycb_s,
213 const u32 crycb_o,
214 struct kvm_s390_crypto_cb *crycb_h,
215 int fmt_o, int fmt_h)
216{
217 struct kvm_s390_crypto_cb *crycb;
218
219 crycb = (struct kvm_s390_crypto_cb *) (unsigned long)crycb_o;
220
221 switch (fmt_o) {
222 case CRYCB_FORMAT2:
223 if ((crycb_o & PAGE_MASK) != ((crycb_o + 256) & PAGE_MASK))
224 return -EACCES;
225 if (fmt_h != CRYCB_FORMAT2)
226 return -EINVAL;
227 return setup_apcb11(vcpu, (unsigned long *)&crycb_s->apcb1,
228 (unsigned long) &crycb->apcb1,
229 (unsigned long *)&crycb_h->apcb1);
230 case CRYCB_FORMAT1:
231 switch (fmt_h) {
232 case CRYCB_FORMAT2:
233 return setup_apcb10(vcpu, &crycb_s->apcb1,
234 (unsigned long) &crycb->apcb0,
235 &crycb_h->apcb1);
236 case CRYCB_FORMAT1:
237 return setup_apcb00(vcpu,
238 (unsigned long *) &crycb_s->apcb0,
239 (unsigned long) &crycb->apcb0,
240 (unsigned long *) &crycb_h->apcb0);
241 }
242 break;
243 case CRYCB_FORMAT0:
244 if ((crycb_o & PAGE_MASK) != ((crycb_o + 32) & PAGE_MASK))
245 return -EACCES;
246
247 switch (fmt_h) {
248 case CRYCB_FORMAT2:
249 return setup_apcb10(vcpu, &crycb_s->apcb1,
250 (unsigned long) &crycb->apcb0,
251 &crycb_h->apcb1);
252 case CRYCB_FORMAT1:
253 case CRYCB_FORMAT0:
254 return setup_apcb00(vcpu,
255 (unsigned long *) &crycb_s->apcb0,
256 (unsigned long) &crycb->apcb0,
257 (unsigned long *) &crycb_h->apcb0);
258 }
259 }
260 return -EINVAL;
261}
262
263/**
264 * shadow_crycb - Create a shadow copy of the crycb block
265 * @vcpu: a pointer to the virtual CPU
266 * @vsie_page: a pointer to internal date used for the vSIE
267 *
268 * Create a shadow copy of the crycb block and setup key wrapping, if
269 * requested for guest 3 and enabled for guest 2.
270 *
271 * We accept format-1 or format-2, but we convert format-1 into format-2
272 * in the shadow CRYCB.
273 * Using format-2 enables the firmware to choose the right format when
274 * scheduling the SIE.
275 * There is nothing to do for format-0.
276 *
277 * This function centralize the issuing of set_validity_icpt() for all
278 * the subfunctions working on the crycb.
279 *
280 * Returns: - 0 if shadowed or nothing to do
281 * - > 0 if control has to be given to guest 2
282 */
283static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
284{
285 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
286 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
287 const uint32_t crycbd_o = READ_ONCE(scb_o->crycbd);
288 const u32 crycb_addr = crycbd_o & 0x7ffffff8U;
289 unsigned long *b1, *b2;
290 u8 ecb3_flags;
291 u32 ecd_flags;
292 int apie_h;
293 int apie_s;
294 int key_msk = test_kvm_facility(vcpu->kvm, 76);
295 int fmt_o = crycbd_o & CRYCB_FORMAT_MASK;
296 int fmt_h = vcpu->arch.sie_block->crycbd & CRYCB_FORMAT_MASK;
297 int ret = 0;
298
299 scb_s->crycbd = 0;
300
301 apie_h = vcpu->arch.sie_block->eca & ECA_APIE;
302 apie_s = apie_h & scb_o->eca;
303 if (!apie_s && (!key_msk || (fmt_o == CRYCB_FORMAT0)))
304 return 0;
305
306 if (!crycb_addr)
307 return set_validity_icpt(scb_s, 0x0039U);
308
309 if (fmt_o == CRYCB_FORMAT1)
310 if ((crycb_addr & PAGE_MASK) !=
311 ((crycb_addr + 128) & PAGE_MASK))
312 return set_validity_icpt(scb_s, 0x003CU);
313
314 if (apie_s) {
315 ret = setup_apcb(vcpu, &vsie_page->crycb, crycb_addr,
316 vcpu->kvm->arch.crypto.crycb,
317 fmt_o, fmt_h);
318 if (ret)
319 goto end;
320 scb_s->eca |= scb_o->eca & ECA_APIE;
321 }
322
323 /* we may only allow it if enabled for guest 2 */
324 ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 &
325 (ECB3_AES | ECB3_DEA);
326 ecd_flags = scb_o->ecd & vcpu->arch.sie_block->ecd & ECD_ECC;
327 if (!ecb3_flags && !ecd_flags)
328 goto end;
329
330 /* copy only the wrapping keys */
331 if (read_guest_real(vcpu, crycb_addr + 72,
332 vsie_page->crycb.dea_wrapping_key_mask, 56))
333 return set_validity_icpt(scb_s, 0x0035U);
334
335 scb_s->ecb3 |= ecb3_flags;
336 scb_s->ecd |= ecd_flags;
337
338 /* xor both blocks in one run */
339 b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask;
340 b2 = (unsigned long *)
341 vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask;
342 /* as 56%8 == 0, bitmap_xor won't overwrite any data */
343 bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56);
344end:
345 switch (ret) {
346 case -EINVAL:
347 return set_validity_icpt(scb_s, 0x0022U);
348 case -EFAULT:
349 return set_validity_icpt(scb_s, 0x0035U);
350 case -EACCES:
351 return set_validity_icpt(scb_s, 0x003CU);
352 }
353 scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT2;
354 return 0;
355}
356
357/* shadow (round up/down) the ibc to avoid validity icpt */
358static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
359{
360 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
361 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
362 /* READ_ONCE does not work on bitfields - use a temporary variable */
363 const uint32_t __new_ibc = scb_o->ibc;
364 const uint32_t new_ibc = READ_ONCE(__new_ibc) & 0x0fffU;
365 __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU;
366
367 scb_s->ibc = 0;
368 /* ibc installed in g2 and requested for g3 */
369 if (vcpu->kvm->arch.model.ibc && new_ibc) {
370 scb_s->ibc = new_ibc;
371 /* takte care of the minimum ibc level of the machine */
372 if (scb_s->ibc < min_ibc)
373 scb_s->ibc = min_ibc;
374 /* take care of the maximum ibc level set for the guest */
375 if (scb_s->ibc > vcpu->kvm->arch.model.ibc)
376 scb_s->ibc = vcpu->kvm->arch.model.ibc;
377 }
378}
379
380/* unshadow the scb, copying parameters back to the real scb */
381static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
382{
383 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
384 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
385
386 /* interception */
387 scb_o->icptcode = scb_s->icptcode;
388 scb_o->icptstatus = scb_s->icptstatus;
389 scb_o->ipa = scb_s->ipa;
390 scb_o->ipb = scb_s->ipb;
391 scb_o->gbea = scb_s->gbea;
392
393 /* timer */
394 scb_o->cputm = scb_s->cputm;
395 scb_o->ckc = scb_s->ckc;
396 scb_o->todpr = scb_s->todpr;
397
398 /* guest state */
399 scb_o->gpsw = scb_s->gpsw;
400 scb_o->gg14 = scb_s->gg14;
401 scb_o->gg15 = scb_s->gg15;
402 memcpy(scb_o->gcr, scb_s->gcr, 128);
403 scb_o->pp = scb_s->pp;
404
405 /* branch prediction */
406 if (test_kvm_facility(vcpu->kvm, 82)) {
407 scb_o->fpf &= ~FPF_BPBC;
408 scb_o->fpf |= scb_s->fpf & FPF_BPBC;
409 }
410
411 /* interrupt intercept */
412 switch (scb_s->icptcode) {
413 case ICPT_PROGI:
414 case ICPT_INSTPROGI:
415 case ICPT_EXTINT:
416 memcpy((void *)((u64)scb_o + 0xc0),
417 (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0);
418 break;
419 case ICPT_PARTEXEC:
420 /* MVPG only */
421 memcpy((void *)((u64)scb_o + 0xc0),
422 (void *)((u64)scb_s + 0xc0), 0xd0 - 0xc0);
423 break;
424 }
425
426 if (scb_s->ihcpu != 0xffffU)
427 scb_o->ihcpu = scb_s->ihcpu;
428}
429
430/*
431 * Setup the shadow scb by copying and checking the relevant parts of the g2
432 * provided scb.
433 *
434 * Returns: - 0 if the scb has been shadowed
435 * - > 0 if control has to be given to guest 2
436 */
437static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
438{
439 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
440 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
441 /* READ_ONCE does not work on bitfields - use a temporary variable */
442 const uint32_t __new_prefix = scb_o->prefix;
443 const uint32_t new_prefix = READ_ONCE(__new_prefix);
444 const bool wants_tx = READ_ONCE(scb_o->ecb) & ECB_TE;
445 bool had_tx = scb_s->ecb & ECB_TE;
446 unsigned long new_mso = 0;
447 int rc;
448
449 /* make sure we don't have any leftovers when reusing the scb */
450 scb_s->icptcode = 0;
451 scb_s->eca = 0;
452 scb_s->ecb = 0;
453 scb_s->ecb2 = 0;
454 scb_s->ecb3 = 0;
455 scb_s->ecd = 0;
456 scb_s->fac = 0;
457 scb_s->fpf = 0;
458
459 rc = prepare_cpuflags(vcpu, vsie_page);
460 if (rc)
461 goto out;
462
463 /* timer */
464 scb_s->cputm = scb_o->cputm;
465 scb_s->ckc = scb_o->ckc;
466 scb_s->todpr = scb_o->todpr;
467 scb_s->epoch = scb_o->epoch;
468
469 /* guest state */
470 scb_s->gpsw = scb_o->gpsw;
471 scb_s->gg14 = scb_o->gg14;
472 scb_s->gg15 = scb_o->gg15;
473 memcpy(scb_s->gcr, scb_o->gcr, 128);
474 scb_s->pp = scb_o->pp;
475
476 /* interception / execution handling */
477 scb_s->gbea = scb_o->gbea;
478 scb_s->lctl = scb_o->lctl;
479 scb_s->svcc = scb_o->svcc;
480 scb_s->ictl = scb_o->ictl;
481 /*
482 * SKEY handling functions can't deal with false setting of PTE invalid
483 * bits. Therefore we cannot provide interpretation and would later
484 * have to provide own emulation handlers.
485 */
486 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS))
487 scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
488
489 scb_s->icpua = scb_o->icpua;
490
491 if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM))
492 new_mso = READ_ONCE(scb_o->mso) & 0xfffffffffff00000UL;
493 /* if the hva of the prefix changes, we have to remap the prefix */
494 if (scb_s->mso != new_mso || scb_s->prefix != new_prefix)
495 prefix_unmapped(vsie_page);
496 /* SIE will do mso/msl validity and exception checks for us */
497 scb_s->msl = scb_o->msl & 0xfffffffffff00000UL;
498 scb_s->mso = new_mso;
499 scb_s->prefix = new_prefix;
500
501 /* We have to definetly flush the tlb if this scb never ran */
502 if (scb_s->ihcpu != 0xffffU)
503 scb_s->ihcpu = scb_o->ihcpu;
504
505 /* MVPG and Protection Exception Interpretation are always available */
506 scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI);
507 /* Host-protection-interruption introduced with ESOP */
508 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP))
509 scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT;
510 /* transactional execution */
511 if (test_kvm_facility(vcpu->kvm, 73) && wants_tx) {
512 /* remap the prefix is tx is toggled on */
513 if (!had_tx)
514 prefix_unmapped(vsie_page);
515 scb_s->ecb |= ECB_TE;
516 }
517 /* branch prediction */
518 if (test_kvm_facility(vcpu->kvm, 82))
519 scb_s->fpf |= scb_o->fpf & FPF_BPBC;
520 /* SIMD */
521 if (test_kvm_facility(vcpu->kvm, 129)) {
522 scb_s->eca |= scb_o->eca & ECA_VX;
523 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
524 }
525 /* Run-time-Instrumentation */
526 if (test_kvm_facility(vcpu->kvm, 64))
527 scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI;
528 /* Instruction Execution Prevention */
529 if (test_kvm_facility(vcpu->kvm, 130))
530 scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP;
531 /* Guarded Storage */
532 if (test_kvm_facility(vcpu->kvm, 133)) {
533 scb_s->ecb |= scb_o->ecb & ECB_GS;
534 scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT;
535 }
536 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF))
537 scb_s->eca |= scb_o->eca & ECA_SII;
538 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB))
539 scb_s->eca |= scb_o->eca & ECA_IB;
540 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI))
541 scb_s->eca |= scb_o->eca & ECA_CEI;
542 /* Epoch Extension */
543 if (test_kvm_facility(vcpu->kvm, 139))
544 scb_s->ecd |= scb_o->ecd & ECD_MEF;
545
546 /* etoken */
547 if (test_kvm_facility(vcpu->kvm, 156))
548 scb_s->ecd |= scb_o->ecd & ECD_ETOKENF;
549
550 scb_s->hpid = HPID_VSIE;
551 scb_s->cpnc = scb_o->cpnc;
552
553 prepare_ibc(vcpu, vsie_page);
554 rc = shadow_crycb(vcpu, vsie_page);
555out:
556 if (rc)
557 unshadow_scb(vcpu, vsie_page);
558 return rc;
559}
560
561void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start,
562 unsigned long end)
563{
564 struct kvm *kvm = gmap->private;
565 struct vsie_page *cur;
566 unsigned long prefix;
567 struct page *page;
568 int i;
569
570 if (!gmap_is_shadow(gmap))
571 return;
572 if (start >= 1UL << 31)
573 /* We are only interested in prefix pages */
574 return;
575
576 /*
577 * Only new shadow blocks are added to the list during runtime,
578 * therefore we can safely reference them all the time.
579 */
580 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
581 page = READ_ONCE(kvm->arch.vsie.pages[i]);
582 if (!page)
583 continue;
584 cur = page_to_virt(page);
585 if (READ_ONCE(cur->gmap) != gmap)
586 continue;
587 prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT;
588 /* with mso/msl, the prefix lies at an offset */
589 prefix += cur->scb_s.mso;
590 if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1)
591 prefix_unmapped_sync(cur);
592 }
593}
594
595/*
596 * Map the first prefix page and if tx is enabled also the second prefix page.
597 *
598 * The prefix will be protected, a gmap notifier will inform about unmaps.
599 * The shadow scb must not be executed until the prefix is remapped, this is
600 * guaranteed by properly handling PROG_REQUEST.
601 *
602 * Returns: - 0 on if successfully mapped or already mapped
603 * - > 0 if control has to be given to guest 2
604 * - -EAGAIN if the caller can retry immediately
605 * - -ENOMEM if out of memory
606 */
607static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
608{
609 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
610 u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT;
611 int rc;
612
613 if (prefix_is_mapped(vsie_page))
614 return 0;
615
616 /* mark it as mapped so we can catch any concurrent unmappers */
617 prefix_mapped(vsie_page);
618
619 /* with mso/msl, the prefix lies at offset *mso* */
620 prefix += scb_s->mso;
621
622 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix);
623 if (!rc && (scb_s->ecb & ECB_TE))
624 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
625 prefix + PAGE_SIZE);
626 /*
627 * We don't have to mprotect, we will be called for all unshadows.
628 * SIE will detect if protection applies and trigger a validity.
629 */
630 if (rc)
631 prefix_unmapped(vsie_page);
632 if (rc > 0 || rc == -EFAULT)
633 rc = set_validity_icpt(scb_s, 0x0037U);
634 return rc;
635}
636
637/*
638 * Pin the guest page given by gpa and set hpa to the pinned host address.
639 * Will always be pinned writable.
640 *
641 * Returns: - 0 on success
642 * - -EINVAL if the gpa is not valid guest storage
643 */
644static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa)
645{
646 struct page *page;
647
648 page = gfn_to_page(kvm, gpa_to_gfn(gpa));
649 if (is_error_page(page))
650 return -EINVAL;
651 *hpa = (hpa_t) page_to_virt(page) + (gpa & ~PAGE_MASK);
652 return 0;
653}
654
655/* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
656static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa)
657{
658 kvm_release_pfn_dirty(hpa >> PAGE_SHIFT);
659 /* mark the page always as dirty for migration */
660 mark_page_dirty(kvm, gpa_to_gfn(gpa));
661}
662
663/* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
664static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
665{
666 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
667 hpa_t hpa;
668
669 hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol;
670 if (hpa) {
671 unpin_guest_page(vcpu->kvm, vsie_page->sca_gpa, hpa);
672 vsie_page->sca_gpa = 0;
673 scb_s->scaol = 0;
674 scb_s->scaoh = 0;
675 }
676
677 hpa = scb_s->itdba;
678 if (hpa) {
679 unpin_guest_page(vcpu->kvm, vsie_page->itdba_gpa, hpa);
680 vsie_page->itdba_gpa = 0;
681 scb_s->itdba = 0;
682 }
683
684 hpa = scb_s->gvrd;
685 if (hpa) {
686 unpin_guest_page(vcpu->kvm, vsie_page->gvrd_gpa, hpa);
687 vsie_page->gvrd_gpa = 0;
688 scb_s->gvrd = 0;
689 }
690
691 hpa = scb_s->riccbd;
692 if (hpa) {
693 unpin_guest_page(vcpu->kvm, vsie_page->riccbd_gpa, hpa);
694 vsie_page->riccbd_gpa = 0;
695 scb_s->riccbd = 0;
696 }
697
698 hpa = scb_s->sdnxo;
699 if (hpa) {
700 unpin_guest_page(vcpu->kvm, vsie_page->sdnx_gpa, hpa);
701 vsie_page->sdnx_gpa = 0;
702 scb_s->sdnxo = 0;
703 }
704}
705
706/*
707 * Instead of shadowing some blocks, we can simply forward them because the
708 * addresses in the scb are 64 bit long.
709 *
710 * This works as long as the data lies in one page. If blocks ever exceed one
711 * page, we have to fall back to shadowing.
712 *
713 * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
714 * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
715 *
716 * Returns: - 0 if all blocks were pinned.
717 * - > 0 if control has to be given to guest 2
718 * - -ENOMEM if out of memory
719 */
720static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
721{
722 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
723 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
724 hpa_t hpa;
725 gpa_t gpa;
726 int rc = 0;
727
728 gpa = READ_ONCE(scb_o->scaol) & ~0xfUL;
729 if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO))
730 gpa |= (u64) READ_ONCE(scb_o->scaoh) << 32;
731 if (gpa) {
732 if (gpa < 2 * PAGE_SIZE)
733 rc = set_validity_icpt(scb_s, 0x0038U);
734 else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu))
735 rc = set_validity_icpt(scb_s, 0x0011U);
736 else if ((gpa & PAGE_MASK) !=
737 ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK))
738 rc = set_validity_icpt(scb_s, 0x003bU);
739 if (!rc) {
740 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
741 if (rc)
742 rc = set_validity_icpt(scb_s, 0x0034U);
743 }
744 if (rc)
745 goto unpin;
746 vsie_page->sca_gpa = gpa;
747 scb_s->scaoh = (u32)((u64)hpa >> 32);
748 scb_s->scaol = (u32)(u64)hpa;
749 }
750
751 gpa = READ_ONCE(scb_o->itdba) & ~0xffUL;
752 if (gpa && (scb_s->ecb & ECB_TE)) {
753 if (gpa < 2 * PAGE_SIZE) {
754 rc = set_validity_icpt(scb_s, 0x0080U);
755 goto unpin;
756 }
757 /* 256 bytes cannot cross page boundaries */
758 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
759 if (rc) {
760 rc = set_validity_icpt(scb_s, 0x0080U);
761 goto unpin;
762 }
763 vsie_page->itdba_gpa = gpa;
764 scb_s->itdba = hpa;
765 }
766
767 gpa = READ_ONCE(scb_o->gvrd) & ~0x1ffUL;
768 if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) {
769 if (gpa < 2 * PAGE_SIZE) {
770 rc = set_validity_icpt(scb_s, 0x1310U);
771 goto unpin;
772 }
773 /*
774 * 512 bytes vector registers cannot cross page boundaries
775 * if this block gets bigger, we have to shadow it.
776 */
777 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
778 if (rc) {
779 rc = set_validity_icpt(scb_s, 0x1310U);
780 goto unpin;
781 }
782 vsie_page->gvrd_gpa = gpa;
783 scb_s->gvrd = hpa;
784 }
785
786 gpa = READ_ONCE(scb_o->riccbd) & ~0x3fUL;
787 if (gpa && (scb_s->ecb3 & ECB3_RI)) {
788 if (gpa < 2 * PAGE_SIZE) {
789 rc = set_validity_icpt(scb_s, 0x0043U);
790 goto unpin;
791 }
792 /* 64 bytes cannot cross page boundaries */
793 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
794 if (rc) {
795 rc = set_validity_icpt(scb_s, 0x0043U);
796 goto unpin;
797 }
798 /* Validity 0x0044 will be checked by SIE */
799 vsie_page->riccbd_gpa = gpa;
800 scb_s->riccbd = hpa;
801 }
802 if (((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) ||
803 (scb_s->ecd & ECD_ETOKENF)) {
804 unsigned long sdnxc;
805
806 gpa = READ_ONCE(scb_o->sdnxo) & ~0xfUL;
807 sdnxc = READ_ONCE(scb_o->sdnxo) & 0xfUL;
808 if (!gpa || gpa < 2 * PAGE_SIZE) {
809 rc = set_validity_icpt(scb_s, 0x10b0U);
810 goto unpin;
811 }
812 if (sdnxc < 6 || sdnxc > 12) {
813 rc = set_validity_icpt(scb_s, 0x10b1U);
814 goto unpin;
815 }
816 if (gpa & ((1 << sdnxc) - 1)) {
817 rc = set_validity_icpt(scb_s, 0x10b2U);
818 goto unpin;
819 }
820 /* Due to alignment rules (checked above) this cannot
821 * cross page boundaries
822 */
823 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
824 if (rc) {
825 rc = set_validity_icpt(scb_s, 0x10b0U);
826 goto unpin;
827 }
828 vsie_page->sdnx_gpa = gpa;
829 scb_s->sdnxo = hpa | sdnxc;
830 }
831 return 0;
832unpin:
833 unpin_blocks(vcpu, vsie_page);
834 return rc;
835}
836
837/* unpin the scb provided by guest 2, marking it as dirty */
838static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
839 gpa_t gpa)
840{
841 hpa_t hpa = (hpa_t) vsie_page->scb_o;
842
843 if (hpa)
844 unpin_guest_page(vcpu->kvm, gpa, hpa);
845 vsie_page->scb_o = NULL;
846}
847
848/*
849 * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
850 *
851 * Returns: - 0 if the scb was pinned.
852 * - > 0 if control has to be given to guest 2
853 */
854static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page,
855 gpa_t gpa)
856{
857 hpa_t hpa;
858 int rc;
859
860 rc = pin_guest_page(vcpu->kvm, gpa, &hpa);
861 if (rc) {
862 rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
863 WARN_ON_ONCE(rc);
864 return 1;
865 }
866 vsie_page->scb_o = (struct kvm_s390_sie_block *) hpa;
867 return 0;
868}
869
870/*
871 * Inject a fault into guest 2.
872 *
873 * Returns: - > 0 if control has to be given to guest 2
874 * < 0 if an error occurred during injection.
875 */
876static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr,
877 bool write_flag)
878{
879 struct kvm_s390_pgm_info pgm = {
880 .code = code,
881 .trans_exc_code =
882 /* 0-51: virtual address */
883 (vaddr & 0xfffffffffffff000UL) |
884 /* 52-53: store / fetch */
885 (((unsigned int) !write_flag) + 1) << 10,
886 /* 62-63: asce id (alway primary == 0) */
887 .exc_access_id = 0, /* always primary */
888 .op_access_id = 0, /* not MVPG */
889 };
890 int rc;
891
892 if (code == PGM_PROTECTION)
893 pgm.trans_exc_code |= 0x4UL;
894
895 rc = kvm_s390_inject_prog_irq(vcpu, &pgm);
896 return rc ? rc : 1;
897}
898
899/*
900 * Handle a fault during vsie execution on a gmap shadow.
901 *
902 * Returns: - 0 if the fault was resolved
903 * - > 0 if control has to be given to guest 2
904 * - < 0 if an error occurred
905 */
906static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
907{
908 int rc;
909
910 if (current->thread.gmap_int_code == PGM_PROTECTION)
911 /* we can directly forward all protection exceptions */
912 return inject_fault(vcpu, PGM_PROTECTION,
913 current->thread.gmap_addr, 1);
914
915 rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
916 current->thread.gmap_addr);
917 if (rc > 0) {
918 rc = inject_fault(vcpu, rc,
919 current->thread.gmap_addr,
920 current->thread.gmap_write_flag);
921 if (rc >= 0)
922 vsie_page->fault_addr = current->thread.gmap_addr;
923 }
924 return rc;
925}
926
927/*
928 * Retry the previous fault that required guest 2 intervention. This avoids
929 * one superfluous SIE re-entry and direct exit.
930 *
931 * Will ignore any errors. The next SIE fault will do proper fault handling.
932 */
933static void handle_last_fault(struct kvm_vcpu *vcpu,
934 struct vsie_page *vsie_page)
935{
936 if (vsie_page->fault_addr)
937 kvm_s390_shadow_fault(vcpu, vsie_page->gmap,
938 vsie_page->fault_addr);
939 vsie_page->fault_addr = 0;
940}
941
942static inline void clear_vsie_icpt(struct vsie_page *vsie_page)
943{
944 vsie_page->scb_s.icptcode = 0;
945}
946
947/* rewind the psw and clear the vsie icpt, so we can retry execution */
948static void retry_vsie_icpt(struct vsie_page *vsie_page)
949{
950 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
951 int ilen = insn_length(scb_s->ipa >> 8);
952
953 /* take care of EXECUTE instructions */
954 if (scb_s->icptstatus & 1) {
955 ilen = (scb_s->icptstatus >> 4) & 0x6;
956 if (!ilen)
957 ilen = 4;
958 }
959 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen);
960 clear_vsie_icpt(vsie_page);
961}
962
963/*
964 * Try to shadow + enable the guest 2 provided facility list.
965 * Retry instruction execution if enabled for and provided by guest 2.
966 *
967 * Returns: - 0 if handled (retry or guest 2 icpt)
968 * - > 0 if control has to be given to guest 2
969 */
970static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
971{
972 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
973 __u32 fac = READ_ONCE(vsie_page->scb_o->fac) & 0x7ffffff8U;
974
975 if (fac && test_kvm_facility(vcpu->kvm, 7)) {
976 retry_vsie_icpt(vsie_page);
977 if (read_guest_real(vcpu, fac, &vsie_page->fac,
978 sizeof(vsie_page->fac)))
979 return set_validity_icpt(scb_s, 0x1090U);
980 scb_s->fac = (__u32)(__u64) &vsie_page->fac;
981 }
982 return 0;
983}
984
985/*
986 * Run the vsie on a shadow scb and a shadow gmap, without any further
987 * sanity checks, handling SIE faults.
988 *
989 * Returns: - 0 everything went fine
990 * - > 0 if control has to be given to guest 2
991 * - < 0 if an error occurred
992 */
993static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
994 __releases(vcpu->kvm->srcu)
995 __acquires(vcpu->kvm->srcu)
996{
997 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
998 struct kvm_s390_sie_block *scb_o = vsie_page->scb_o;
999 int guest_bp_isolation;
1000 int rc = 0;
1001
1002 handle_last_fault(vcpu, vsie_page);
1003
1004 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
1005
1006 /* save current guest state of bp isolation override */
1007 guest_bp_isolation = test_thread_flag(TIF_ISOLATE_BP_GUEST);
1008
1009 /*
1010 * The guest is running with BPBC, so we have to force it on for our
1011 * nested guest. This is done by enabling BPBC globally, so the BPBC
1012 * control in the SCB (which the nested guest can modify) is simply
1013 * ignored.
1014 */
1015 if (test_kvm_facility(vcpu->kvm, 82) &&
1016 vcpu->arch.sie_block->fpf & FPF_BPBC)
1017 set_thread_flag(TIF_ISOLATE_BP_GUEST);
1018
1019 local_irq_disable();
1020 guest_enter_irqoff();
1021 local_irq_enable();
1022
1023 /*
1024 * Simulate a SIE entry of the VCPU (see sie64a), so VCPU blocking
1025 * and VCPU requests also hinder the vSIE from running and lead
1026 * to an immediate exit. kvm_s390_vsie_kick() has to be used to
1027 * also kick the vSIE.
1028 */
1029 vcpu->arch.sie_block->prog0c |= PROG_IN_SIE;
1030 barrier();
1031 if (!kvm_s390_vcpu_sie_inhibited(vcpu))
1032 rc = sie64a(scb_s, vcpu->run->s.regs.gprs);
1033 barrier();
1034 vcpu->arch.sie_block->prog0c &= ~PROG_IN_SIE;
1035
1036 local_irq_disable();
1037 guest_exit_irqoff();
1038 local_irq_enable();
1039
1040 /* restore guest state for bp isolation override */
1041 if (!guest_bp_isolation)
1042 clear_thread_flag(TIF_ISOLATE_BP_GUEST);
1043
1044 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
1045
1046 if (rc == -EINTR) {
1047 VCPU_EVENT(vcpu, 3, "%s", "machine check");
1048 kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info);
1049 return 0;
1050 }
1051
1052 if (rc > 0)
1053 rc = 0; /* we could still have an icpt */
1054 else if (rc == -EFAULT)
1055 return handle_fault(vcpu, vsie_page);
1056
1057 switch (scb_s->icptcode) {
1058 case ICPT_INST:
1059 if (scb_s->ipa == 0xb2b0)
1060 rc = handle_stfle(vcpu, vsie_page);
1061 break;
1062 case ICPT_STOP:
1063 /* stop not requested by g2 - must have been a kick */
1064 if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT))
1065 clear_vsie_icpt(vsie_page);
1066 break;
1067 case ICPT_VALIDITY:
1068 if ((scb_s->ipa & 0xf000) != 0xf000)
1069 scb_s->ipa += 0x1000;
1070 break;
1071 }
1072 return rc;
1073}
1074
1075static void release_gmap_shadow(struct vsie_page *vsie_page)
1076{
1077 if (vsie_page->gmap)
1078 gmap_put(vsie_page->gmap);
1079 WRITE_ONCE(vsie_page->gmap, NULL);
1080 prefix_unmapped(vsie_page);
1081}
1082
1083static int acquire_gmap_shadow(struct kvm_vcpu *vcpu,
1084 struct vsie_page *vsie_page)
1085{
1086 unsigned long asce;
1087 union ctlreg0 cr0;
1088 struct gmap *gmap;
1089 int edat;
1090
1091 asce = vcpu->arch.sie_block->gcr[1];
1092 cr0.val = vcpu->arch.sie_block->gcr[0];
1093 edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8);
1094 edat += edat && test_kvm_facility(vcpu->kvm, 78);
1095
1096 /*
1097 * ASCE or EDAT could have changed since last icpt, or the gmap
1098 * we're holding has been unshadowed. If the gmap is still valid,
1099 * we can safely reuse it.
1100 */
1101 if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat))
1102 return 0;
1103
1104 /* release the old shadow - if any, and mark the prefix as unmapped */
1105 release_gmap_shadow(vsie_page);
1106 gmap = gmap_shadow(vcpu->arch.gmap, asce, edat);
1107 if (IS_ERR(gmap))
1108 return PTR_ERR(gmap);
1109 gmap->private = vcpu->kvm;
1110 WRITE_ONCE(vsie_page->gmap, gmap);
1111 return 0;
1112}
1113
1114/*
1115 * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
1116 */
1117static void register_shadow_scb(struct kvm_vcpu *vcpu,
1118 struct vsie_page *vsie_page)
1119{
1120 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1121
1122 WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s);
1123 /*
1124 * External calls have to lead to a kick of the vcpu and
1125 * therefore the vsie -> Simulate Wait state.
1126 */
1127 kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT);
1128 /*
1129 * We have to adjust the g3 epoch by the g2 epoch. The epoch will
1130 * automatically be adjusted on tod clock changes via kvm_sync_clock.
1131 */
1132 preempt_disable();
1133 scb_s->epoch += vcpu->kvm->arch.epoch;
1134
1135 if (scb_s->ecd & ECD_MEF) {
1136 scb_s->epdx += vcpu->kvm->arch.epdx;
1137 if (scb_s->epoch < vcpu->kvm->arch.epoch)
1138 scb_s->epdx += 1;
1139 }
1140
1141 preempt_enable();
1142}
1143
1144/*
1145 * Unregister a shadow scb from a VCPU.
1146 */
1147static void unregister_shadow_scb(struct kvm_vcpu *vcpu)
1148{
1149 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT);
1150 WRITE_ONCE(vcpu->arch.vsie_block, NULL);
1151}
1152
1153/*
1154 * Run the vsie on a shadowed scb, managing the gmap shadow, handling
1155 * prefix pages and faults.
1156 *
1157 * Returns: - 0 if no errors occurred
1158 * - > 0 if control has to be given to guest 2
1159 * - -ENOMEM if out of memory
1160 */
1161static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page)
1162{
1163 struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s;
1164 int rc = 0;
1165
1166 while (1) {
1167 rc = acquire_gmap_shadow(vcpu, vsie_page);
1168 if (!rc)
1169 rc = map_prefix(vcpu, vsie_page);
1170 if (!rc) {
1171 gmap_enable(vsie_page->gmap);
1172 update_intervention_requests(vsie_page);
1173 rc = do_vsie_run(vcpu, vsie_page);
1174 gmap_enable(vcpu->arch.gmap);
1175 }
1176 atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20);
1177
1178 if (rc == -EAGAIN)
1179 rc = 0;
1180 if (rc || scb_s->icptcode || signal_pending(current) ||
1181 kvm_s390_vcpu_has_irq(vcpu, 0) ||
1182 kvm_s390_vcpu_sie_inhibited(vcpu))
1183 break;
1184 cond_resched();
1185 }
1186
1187 if (rc == -EFAULT) {
1188 /*
1189 * Addressing exceptions are always presentes as intercepts.
1190 * As addressing exceptions are suppressing and our guest 3 PSW
1191 * points at the responsible instruction, we have to
1192 * forward the PSW and set the ilc. If we can't read guest 3
1193 * instruction, we can use an arbitrary ilc. Let's always use
1194 * ilen = 4 for now, so we can avoid reading in guest 3 virtual
1195 * memory. (we could also fake the shadow so the hardware
1196 * handles it).
1197 */
1198 scb_s->icptcode = ICPT_PROGI;
1199 scb_s->iprcc = PGM_ADDRESSING;
1200 scb_s->pgmilc = 4;
1201 scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4);
1202 rc = 1;
1203 }
1204 return rc;
1205}
1206
1207/*
1208 * Get or create a vsie page for a scb address.
1209 *
1210 * Returns: - address of a vsie page (cached or new one)
1211 * - NULL if the same scb address is already used by another VCPU
1212 * - ERR_PTR(-ENOMEM) if out of memory
1213 */
1214static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr)
1215{
1216 struct vsie_page *vsie_page;
1217 struct page *page;
1218 int nr_vcpus;
1219
1220 rcu_read_lock();
1221 page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9);
1222 rcu_read_unlock();
1223 if (page) {
1224 if (page_ref_inc_return(page) == 2)
1225 return page_to_virt(page);
1226 page_ref_dec(page);
1227 }
1228
1229 /*
1230 * We want at least #online_vcpus shadows, so every VCPU can execute
1231 * the VSIE in parallel.
1232 */
1233 nr_vcpus = atomic_read(&kvm->online_vcpus);
1234
1235 mutex_lock(&kvm->arch.vsie.mutex);
1236 if (kvm->arch.vsie.page_count < nr_vcpus) {
1237 page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA);
1238 if (!page) {
1239 mutex_unlock(&kvm->arch.vsie.mutex);
1240 return ERR_PTR(-ENOMEM);
1241 }
1242 page_ref_inc(page);
1243 kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page;
1244 kvm->arch.vsie.page_count++;
1245 } else {
1246 /* reuse an existing entry that belongs to nobody */
1247 while (true) {
1248 page = kvm->arch.vsie.pages[kvm->arch.vsie.next];
1249 if (page_ref_inc_return(page) == 2)
1250 break;
1251 page_ref_dec(page);
1252 kvm->arch.vsie.next++;
1253 kvm->arch.vsie.next %= nr_vcpus;
1254 }
1255 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1256 }
1257 page->index = addr;
1258 /* double use of the same address */
1259 if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) {
1260 page_ref_dec(page);
1261 mutex_unlock(&kvm->arch.vsie.mutex);
1262 return NULL;
1263 }
1264 mutex_unlock(&kvm->arch.vsie.mutex);
1265
1266 vsie_page = page_to_virt(page);
1267 memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block));
1268 release_gmap_shadow(vsie_page);
1269 vsie_page->fault_addr = 0;
1270 vsie_page->scb_s.ihcpu = 0xffffU;
1271 return vsie_page;
1272}
1273
1274/* put a vsie page acquired via get_vsie_page */
1275static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page)
1276{
1277 struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT);
1278
1279 page_ref_dec(page);
1280}
1281
1282int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu)
1283{
1284 struct vsie_page *vsie_page;
1285 unsigned long scb_addr;
1286 int rc;
1287
1288 vcpu->stat.instruction_sie++;
1289 if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2))
1290 return -EOPNOTSUPP;
1291 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1292 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1293
1294 BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE);
1295 scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL);
1296
1297 /* 512 byte alignment */
1298 if (unlikely(scb_addr & 0x1ffUL))
1299 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1300
1301 if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0) ||
1302 kvm_s390_vcpu_sie_inhibited(vcpu))
1303 return 0;
1304
1305 vsie_page = get_vsie_page(vcpu->kvm, scb_addr);
1306 if (IS_ERR(vsie_page))
1307 return PTR_ERR(vsie_page);
1308 else if (!vsie_page)
1309 /* double use of sie control block - simply do nothing */
1310 return 0;
1311
1312 rc = pin_scb(vcpu, vsie_page, scb_addr);
1313 if (rc)
1314 goto out_put;
1315 rc = shadow_scb(vcpu, vsie_page);
1316 if (rc)
1317 goto out_unpin_scb;
1318 rc = pin_blocks(vcpu, vsie_page);
1319 if (rc)
1320 goto out_unshadow;
1321 register_shadow_scb(vcpu, vsie_page);
1322 rc = vsie_run(vcpu, vsie_page);
1323 unregister_shadow_scb(vcpu);
1324 unpin_blocks(vcpu, vsie_page);
1325out_unshadow:
1326 unshadow_scb(vcpu, vsie_page);
1327out_unpin_scb:
1328 unpin_scb(vcpu, vsie_page, scb_addr);
1329out_put:
1330 put_vsie_page(vcpu->kvm, vsie_page);
1331
1332 return rc < 0 ? rc : 0;
1333}
1334
1335/* Init the vsie data structures. To be called when a vm is initialized. */
1336void kvm_s390_vsie_init(struct kvm *kvm)
1337{
1338 mutex_init(&kvm->arch.vsie.mutex);
1339 INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL);
1340}
1341
1342/* Destroy the vsie data structures. To be called when a vm is destroyed. */
1343void kvm_s390_vsie_destroy(struct kvm *kvm)
1344{
1345 struct vsie_page *vsie_page;
1346 struct page *page;
1347 int i;
1348
1349 mutex_lock(&kvm->arch.vsie.mutex);
1350 for (i = 0; i < kvm->arch.vsie.page_count; i++) {
1351 page = kvm->arch.vsie.pages[i];
1352 kvm->arch.vsie.pages[i] = NULL;
1353 vsie_page = page_to_virt(page);
1354 release_gmap_shadow(vsie_page);
1355 /* free the radix tree entry */
1356 radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9);
1357 __free_page(page);
1358 }
1359 kvm->arch.vsie.page_count = 0;
1360 mutex_unlock(&kvm->arch.vsie.mutex);
1361}
1362
1363void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu)
1364{
1365 struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block);
1366
1367 /*
1368 * Even if the VCPU lets go of the shadow sie block reference, it is
1369 * still valid in the cache. So we can safely kick it.
1370 */
1371 if (scb) {
1372 atomic_or(PROG_BLOCK_SIE, &scb->prog20);
1373 if (scb->prog0c & PROG_IN_SIE)
1374 atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags);
1375 }
1376}