2 * kvm nested virtualization support for s390x
4 * Copyright IBM Corp. 2016
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License (version 2 only)
8 * as published by the Free Software Foundation.
10 * Author(s): David Hildenbrand <dahi@linux.vnet.ibm.com>
12 #include <linux/vmalloc.h>
13 #include <linux/kvm_host.h>
14 #include <linux/bug.h>
15 #include <linux/list.h>
16 #include <linux/bitmap.h>
18 #include <asm/mmu_context.h>
26 struct kvm_s390_sie_block scb_s
; /* 0x0000 */
27 /* the pinned originial scb */
28 struct kvm_s390_sie_block
*scb_o
; /* 0x0200 */
29 /* the shadow gmap in use by the vsie_page */
30 struct gmap
*gmap
; /* 0x0208 */
31 __u8 reserved
[0x0700 - 0x0210]; /* 0x0210 */
32 struct kvm_s390_crypto_cb crycb
; /* 0x0700 */
33 __u8 fac
[S390_ARCH_FAC_LIST_SIZE_BYTE
]; /* 0x0800 */
36 /* trigger a validity icpt for the given scb */
37 static int set_validity_icpt(struct kvm_s390_sie_block
*scb
,
41 scb
->ipb
= ((__u32
) reason_code
) << 16;
42 scb
->icptcode
= ICPT_VALIDITY
;
46 /* mark the prefix as unmapped, this will block the VSIE */
47 static void prefix_unmapped(struct vsie_page
*vsie_page
)
49 atomic_or(PROG_REQUEST
, &vsie_page
->scb_s
.prog20
);
52 /* mark the prefix as unmapped and wait until the VSIE has been left */
53 static void prefix_unmapped_sync(struct vsie_page
*vsie_page
)
55 prefix_unmapped(vsie_page
);
56 if (vsie_page
->scb_s
.prog0c
& PROG_IN_SIE
)
57 atomic_or(CPUSTAT_STOP_INT
, &vsie_page
->scb_s
.cpuflags
);
58 while (vsie_page
->scb_s
.prog0c
& PROG_IN_SIE
)
62 /* mark the prefix as mapped, this will allow the VSIE to run */
63 static void prefix_mapped(struct vsie_page
*vsie_page
)
65 atomic_andnot(PROG_REQUEST
, &vsie_page
->scb_s
.prog20
);
68 /* test if the prefix is mapped into the gmap shadow */
69 static int prefix_is_mapped(struct vsie_page
*vsie_page
)
71 return !(atomic_read(&vsie_page
->scb_s
.prog20
) & PROG_REQUEST
);
74 /* copy the updated intervention request bits into the shadow scb */
75 static void update_intervention_requests(struct vsie_page
*vsie_page
)
77 const int bits
= CPUSTAT_STOP_INT
| CPUSTAT_IO_INT
| CPUSTAT_EXT_INT
;
80 cpuflags
= atomic_read(&vsie_page
->scb_o
->cpuflags
);
81 atomic_andnot(bits
, &vsie_page
->scb_s
.cpuflags
);
82 atomic_or(cpuflags
& bits
, &vsie_page
->scb_s
.cpuflags
);
85 /* shadow (filter and validate) the cpuflags */
86 static int prepare_cpuflags(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
88 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
89 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
90 int newflags
, cpuflags
= atomic_read(&scb_o
->cpuflags
);
92 /* we don't allow ESA/390 guests */
93 if (!(cpuflags
& CPUSTAT_ZARCH
))
94 return set_validity_icpt(scb_s
, 0x0001U
);
96 if (cpuflags
& (CPUSTAT_RRF
| CPUSTAT_MCDS
))
97 return set_validity_icpt(scb_s
, 0x0001U
);
98 else if (cpuflags
& (CPUSTAT_SLSV
| CPUSTAT_SLSR
))
99 return set_validity_icpt(scb_s
, 0x0007U
);
101 /* intervention requests will be set later */
102 newflags
= CPUSTAT_ZARCH
;
103 if (cpuflags
& CPUSTAT_GED
&& test_kvm_facility(vcpu
->kvm
, 8))
104 newflags
|= CPUSTAT_GED
;
105 if (cpuflags
& CPUSTAT_GED2
&& test_kvm_facility(vcpu
->kvm
, 78)) {
106 if (cpuflags
& CPUSTAT_GED
)
107 return set_validity_icpt(scb_s
, 0x0001U
);
108 newflags
|= CPUSTAT_GED2
;
111 atomic_set(&scb_s
->cpuflags
, newflags
);
116 * Create a shadow copy of the crycb block and setup key wrapping, if
117 * requested for guest 3 and enabled for guest 2.
119 * We only accept format-1 (no AP in g2), but convert it into format-2
120 * There is nothing to do for format-0.
122 * Returns: - 0 if shadowed or nothing to do
123 * - > 0 if control has to be given to guest 2
125 static int shadow_crycb(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
127 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
128 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
129 u32 crycb_addr
= scb_o
->crycbd
& 0x7ffffff8U
;
130 unsigned long *b1
, *b2
;
134 if (!(scb_o
->crycbd
& vcpu
->arch
.sie_block
->crycbd
& CRYCB_FORMAT1
))
136 /* format-1 is supported with message-security-assist extension 3 */
137 if (!test_kvm_facility(vcpu
->kvm
, 76))
139 /* we may only allow it if enabled for guest 2 */
140 ecb3_flags
= scb_o
->ecb3
& vcpu
->arch
.sie_block
->ecb3
&
141 (ECB3_AES
| ECB3_DEA
);
145 if ((crycb_addr
& PAGE_MASK
) != ((crycb_addr
+ 128) & PAGE_MASK
))
146 return set_validity_icpt(scb_s
, 0x003CU
);
147 else if (!crycb_addr
)
148 return set_validity_icpt(scb_s
, 0x0039U
);
150 /* copy only the wrapping keys */
151 if (read_guest_real(vcpu
, crycb_addr
+ 72, &vsie_page
->crycb
, 56))
152 return set_validity_icpt(scb_s
, 0x0035U
);
154 scb_s
->ecb3
|= ecb3_flags
;
155 scb_s
->crycbd
= ((__u32
)(__u64
) &vsie_page
->crycb
) | CRYCB_FORMAT1
|
158 /* xor both blocks in one run */
159 b1
= (unsigned long *) vsie_page
->crycb
.dea_wrapping_key_mask
;
160 b2
= (unsigned long *)
161 vcpu
->kvm
->arch
.crypto
.crycb
->dea_wrapping_key_mask
;
162 /* as 56%8 == 0, bitmap_xor won't overwrite any data */
163 bitmap_xor(b1
, b1
, b2
, BITS_PER_BYTE
* 56);
167 /* shadow (round up/down) the ibc to avoid validity icpt */
168 static void prepare_ibc(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
170 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
171 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
172 __u64 min_ibc
= (sclp
.ibc
>> 16) & 0x0fffU
;
175 /* ibc installed in g2 and requested for g3 */
176 if (vcpu
->kvm
->arch
.model
.ibc
&& (scb_o
->ibc
& 0x0fffU
)) {
177 scb_s
->ibc
= scb_o
->ibc
& 0x0fffU
;
178 /* takte care of the minimum ibc level of the machine */
179 if (scb_s
->ibc
< min_ibc
)
180 scb_s
->ibc
= min_ibc
;
181 /* take care of the maximum ibc level set for the guest */
182 if (scb_s
->ibc
> vcpu
->kvm
->arch
.model
.ibc
)
183 scb_s
->ibc
= vcpu
->kvm
->arch
.model
.ibc
;
187 /* unshadow the scb, copying parameters back to the real scb */
188 static void unshadow_scb(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
190 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
191 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
194 scb_o
->icptcode
= scb_s
->icptcode
;
195 scb_o
->icptstatus
= scb_s
->icptstatus
;
196 scb_o
->ipa
= scb_s
->ipa
;
197 scb_o
->ipb
= scb_s
->ipb
;
198 scb_o
->gbea
= scb_s
->gbea
;
201 scb_o
->cputm
= scb_s
->cputm
;
202 scb_o
->ckc
= scb_s
->ckc
;
203 scb_o
->todpr
= scb_s
->todpr
;
206 scb_o
->gpsw
= scb_s
->gpsw
;
207 scb_o
->gg14
= scb_s
->gg14
;
208 scb_o
->gg15
= scb_s
->gg15
;
209 memcpy(scb_o
->gcr
, scb_s
->gcr
, 128);
210 scb_o
->pp
= scb_s
->pp
;
212 /* interrupt intercept */
213 switch (scb_s
->icptcode
) {
217 memcpy((void *)((u64
)scb_o
+ 0xc0),
218 (void *)((u64
)scb_s
+ 0xc0), 0xf0 - 0xc0);
222 memcpy((void *)((u64
)scb_o
+ 0xc0),
223 (void *)((u64
)scb_s
+ 0xc0), 0xd0 - 0xc0);
227 if (scb_s
->ihcpu
!= 0xffffU
)
228 scb_o
->ihcpu
= scb_s
->ihcpu
;
232 * Setup the shadow scb by copying and checking the relevant parts of the g2
235 * Returns: - 0 if the scb has been shadowed
236 * - > 0 if control has to be given to guest 2
238 static int shadow_scb(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
240 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
241 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
242 bool had_tx
= scb_s
->ecb
& 0x10U
;
243 unsigned long new_mso
;
246 /* make sure we don't have any leftovers when reusing the scb */
255 rc
= prepare_cpuflags(vcpu
, vsie_page
);
260 scb_s
->cputm
= scb_o
->cputm
;
261 scb_s
->ckc
= scb_o
->ckc
;
262 scb_s
->todpr
= scb_o
->todpr
;
263 scb_s
->epoch
= scb_o
->epoch
;
266 scb_s
->gpsw
= scb_o
->gpsw
;
267 scb_s
->gg14
= scb_o
->gg14
;
268 scb_s
->gg15
= scb_o
->gg15
;
269 memcpy(scb_s
->gcr
, scb_o
->gcr
, 128);
270 scb_s
->pp
= scb_o
->pp
;
272 /* interception / execution handling */
273 scb_s
->gbea
= scb_o
->gbea
;
274 scb_s
->lctl
= scb_o
->lctl
;
275 scb_s
->svcc
= scb_o
->svcc
;
276 scb_s
->ictl
= scb_o
->ictl
;
278 * SKEY handling functions can't deal with false setting of PTE invalid
279 * bits. Therefore we cannot provide interpretation and would later
280 * have to provide own emulation handlers.
282 scb_s
->ictl
|= ICTL_ISKE
| ICTL_SSKE
| ICTL_RRBE
;
283 scb_s
->icpua
= scb_o
->icpua
;
285 new_mso
= scb_o
->mso
& 0xfffffffffff00000UL
;
286 /* if the hva of the prefix changes, we have to remap the prefix */
287 if (scb_s
->mso
!= new_mso
|| scb_s
->prefix
!= scb_o
->prefix
)
288 prefix_unmapped(vsie_page
);
289 /* SIE will do mso/msl validity and exception checks for us */
290 scb_s
->msl
= scb_o
->msl
& 0xfffffffffff00000UL
;
291 scb_s
->mso
= new_mso
;
292 scb_s
->prefix
= scb_o
->prefix
;
294 /* We have to definetly flush the tlb if this scb never ran */
295 if (scb_s
->ihcpu
!= 0xffffU
)
296 scb_s
->ihcpu
= scb_o
->ihcpu
;
298 /* MVPG and Protection Exception Interpretation are always available */
299 scb_s
->eca
|= scb_o
->eca
& 0x01002000U
;
300 /* Host-protection-interruption introduced with ESOP */
301 if (test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_ESOP
))
302 scb_s
->ecb
|= scb_o
->ecb
& 0x02U
;
303 /* transactional execution */
304 if (test_kvm_facility(vcpu
->kvm
, 73)) {
305 /* remap the prefix is tx is toggled on */
306 if ((scb_o
->ecb
& 0x10U
) && !had_tx
)
307 prefix_unmapped(vsie_page
);
308 scb_s
->ecb
|= scb_o
->ecb
& 0x10U
;
311 if (test_kvm_facility(vcpu
->kvm
, 129)) {
312 scb_s
->eca
|= scb_o
->eca
& 0x00020000U
;
313 scb_s
->ecd
|= scb_o
->ecd
& 0x20000000U
;
315 /* Run-time-Instrumentation */
316 if (test_kvm_facility(vcpu
->kvm
, 64))
317 scb_s
->ecb3
|= scb_o
->ecb3
& 0x01U
;
318 if (test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_SIIF
))
319 scb_s
->eca
|= scb_o
->eca
& 0x00000001U
;
321 prepare_ibc(vcpu
, vsie_page
);
322 rc
= shadow_crycb(vcpu
, vsie_page
);
325 unshadow_scb(vcpu
, vsie_page
);
329 void kvm_s390_vsie_gmap_notifier(struct gmap
*gmap
, unsigned long start
,
332 struct kvm
*kvm
= gmap
->private;
333 struct vsie_page
*cur
;
334 unsigned long prefix
;
338 if (!gmap_is_shadow(gmap
))
340 if (start
>= 1UL << 31)
341 /* We are only interested in prefix pages */
345 * Only new shadow blocks are added to the list during runtime,
346 * therefore we can safely reference them all the time.
348 for (i
= 0; i
< kvm
->arch
.vsie
.page_count
; i
++) {
349 page
= READ_ONCE(kvm
->arch
.vsie
.pages
[i
]);
352 cur
= page_to_virt(page
);
353 if (READ_ONCE(cur
->gmap
) != gmap
)
355 prefix
= cur
->scb_s
.prefix
<< GUEST_PREFIX_SHIFT
;
356 /* with mso/msl, the prefix lies at an offset */
357 prefix
+= cur
->scb_s
.mso
;
358 if (prefix
<= end
&& start
<= prefix
+ 2 * PAGE_SIZE
- 1)
359 prefix_unmapped_sync(cur
);
364 * Map the first prefix page and if tx is enabled also the second prefix page.
366 * The prefix will be protected, a gmap notifier will inform about unmaps.
367 * The shadow scb must not be executed until the prefix is remapped, this is
368 * guaranteed by properly handling PROG_REQUEST.
370 * Returns: - 0 on if successfully mapped or already mapped
371 * - > 0 if control has to be given to guest 2
372 * - -EAGAIN if the caller can retry immediately
373 * - -ENOMEM if out of memory
375 static int map_prefix(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
377 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
378 u64 prefix
= scb_s
->prefix
<< GUEST_PREFIX_SHIFT
;
381 if (prefix_is_mapped(vsie_page
))
384 /* mark it as mapped so we can catch any concurrent unmappers */
385 prefix_mapped(vsie_page
);
387 /* with mso/msl, the prefix lies at offset *mso* */
388 prefix
+= scb_s
->mso
;
390 rc
= kvm_s390_shadow_fault(vcpu
, vsie_page
->gmap
, prefix
);
391 if (!rc
&& (scb_s
->ecb
& 0x10U
))
392 rc
= kvm_s390_shadow_fault(vcpu
, vsie_page
->gmap
,
395 * We don't have to mprotect, we will be called for all unshadows.
396 * SIE will detect if protection applies and trigger a validity.
399 prefix_unmapped(vsie_page
);
400 if (rc
> 0 || rc
== -EFAULT
)
401 rc
= set_validity_icpt(scb_s
, 0x0037U
);
406 * Pin the guest page given by gpa and set hpa to the pinned host address.
407 * Will always be pinned writable.
409 * Returns: - 0 on success
410 * - -EINVAL if the gpa is not valid guest storage
411 * - -ENOMEM if out of memory
413 static int pin_guest_page(struct kvm
*kvm
, gpa_t gpa
, hpa_t
*hpa
)
419 hva
= gfn_to_hva(kvm
, gpa_to_gfn(gpa
));
420 if (kvm_is_error_hva(hva
))
422 rc
= get_user_pages_fast(hva
, 1, 1, &page
);
427 *hpa
= (hpa_t
) page_to_virt(page
) + (gpa
& ~PAGE_MASK
);
431 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
432 static void unpin_guest_page(struct kvm
*kvm
, gpa_t gpa
, hpa_t hpa
)
436 page
= virt_to_page(hpa
);
437 set_page_dirty_lock(page
);
439 /* mark the page always as dirty for migration */
440 mark_page_dirty(kvm
, gpa_to_gfn(gpa
));
443 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
444 static void unpin_blocks(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
446 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
447 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
451 hpa
= (u64
) scb_s
->scaoh
<< 32 | scb_s
->scaol
;
453 gpa
= scb_o
->scaol
& ~0xfUL
;
454 if (test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_64BSCAO
))
455 gpa
|= (u64
) scb_o
->scaoh
<< 32;
456 unpin_guest_page(vcpu
->kvm
, gpa
, hpa
);
463 gpa
= scb_o
->itdba
& ~0xffUL
;
464 unpin_guest_page(vcpu
->kvm
, gpa
, hpa
);
470 gpa
= scb_o
->gvrd
& ~0x1ffUL
;
471 unpin_guest_page(vcpu
->kvm
, gpa
, hpa
);
477 gpa
= scb_o
->riccbd
& ~0x3fUL
;
478 unpin_guest_page(vcpu
->kvm
, gpa
, hpa
);
484 * Instead of shadowing some blocks, we can simply forward them because the
485 * addresses in the scb are 64 bit long.
487 * This works as long as the data lies in one page. If blocks ever exceed one
488 * page, we have to fall back to shadowing.
490 * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
491 * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
493 * Returns: - 0 if all blocks were pinned.
494 * - > 0 if control has to be given to guest 2
495 * - -ENOMEM if out of memory
497 static int pin_blocks(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
499 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
500 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
505 gpa
= scb_o
->scaol
& ~0xfUL
;
506 if (test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_64BSCAO
))
507 gpa
|= (u64
) scb_o
->scaoh
<< 32;
509 if (!(gpa
& ~0x1fffUL
))
510 rc
= set_validity_icpt(scb_s
, 0x0038U
);
511 else if ((gpa
& ~0x1fffUL
) == kvm_s390_get_prefix(vcpu
))
512 rc
= set_validity_icpt(scb_s
, 0x0011U
);
513 else if ((gpa
& PAGE_MASK
) !=
514 ((gpa
+ sizeof(struct bsca_block
) - 1) & PAGE_MASK
))
515 rc
= set_validity_icpt(scb_s
, 0x003bU
);
517 rc
= pin_guest_page(vcpu
->kvm
, gpa
, &hpa
);
519 rc
= set_validity_icpt(scb_s
, 0x0034U
);
523 scb_s
->scaoh
= (u32
)((u64
)hpa
>> 32);
524 scb_s
->scaol
= (u32
)(u64
)hpa
;
527 gpa
= scb_o
->itdba
& ~0xffUL
;
528 if (gpa
&& (scb_s
->ecb
& 0x10U
)) {
529 if (!(gpa
& ~0x1fffU
)) {
530 rc
= set_validity_icpt(scb_s
, 0x0080U
);
533 /* 256 bytes cannot cross page boundaries */
534 rc
= pin_guest_page(vcpu
->kvm
, gpa
, &hpa
);
536 rc
= set_validity_icpt(scb_s
, 0x0080U
);
542 gpa
= scb_o
->gvrd
& ~0x1ffUL
;
543 if (gpa
&& (scb_s
->eca
& 0x00020000U
) &&
544 !(scb_s
->ecd
& 0x20000000U
)) {
545 if (!(gpa
& ~0x1fffUL
)) {
546 rc
= set_validity_icpt(scb_s
, 0x1310U
);
550 * 512 bytes vector registers cannot cross page boundaries
551 * if this block gets bigger, we have to shadow it.
553 rc
= pin_guest_page(vcpu
->kvm
, gpa
, &hpa
);
555 rc
= set_validity_icpt(scb_s
, 0x1310U
);
561 gpa
= scb_o
->riccbd
& ~0x3fUL
;
562 if (gpa
&& (scb_s
->ecb3
& 0x01U
)) {
563 if (!(gpa
& ~0x1fffUL
)) {
564 rc
= set_validity_icpt(scb_s
, 0x0043U
);
567 /* 64 bytes cannot cross page boundaries */
568 rc
= pin_guest_page(vcpu
->kvm
, gpa
, &hpa
);
570 rc
= set_validity_icpt(scb_s
, 0x0043U
);
571 /* Validity 0x0044 will be checked by SIE */
578 unpin_blocks(vcpu
, vsie_page
);
582 /* unpin the scb provided by guest 2, marking it as dirty */
583 static void unpin_scb(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
,
586 hpa_t hpa
= (hpa_t
) vsie_page
->scb_o
;
589 unpin_guest_page(vcpu
->kvm
, gpa
, hpa
);
590 vsie_page
->scb_o
= NULL
;
594 * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
596 * Returns: - 0 if the scb was pinned.
597 * - > 0 if control has to be given to guest 2
598 * - -ENOMEM if out of memory
600 static int pin_scb(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
,
606 rc
= pin_guest_page(vcpu
->kvm
, gpa
, &hpa
);
608 rc
= kvm_s390_inject_program_int(vcpu
, PGM_ADDRESSING
);
613 vsie_page
->scb_o
= (struct kvm_s390_sie_block
*) hpa
;
618 * Inject a fault into guest 2.
620 * Returns: - > 0 if control has to be given to guest 2
621 * < 0 if an error occurred during injection.
623 static int inject_fault(struct kvm_vcpu
*vcpu
, __u16 code
, __u64 vaddr
,
626 struct kvm_s390_pgm_info pgm
= {
629 /* 0-51: virtual address */
630 (vaddr
& 0xfffffffffffff000UL
) |
631 /* 52-53: store / fetch */
632 (((unsigned int) !write_flag
) + 1) << 10,
633 /* 62-63: asce id (alway primary == 0) */
634 .exc_access_id
= 0, /* always primary */
635 .op_access_id
= 0, /* not MVPG */
639 if (code
== PGM_PROTECTION
)
640 pgm
.trans_exc_code
|= 0x4UL
;
642 rc
= kvm_s390_inject_prog_irq(vcpu
, &pgm
);
647 * Handle a fault during vsie execution on a gmap shadow.
649 * Returns: - 0 if the fault was resolved
650 * - > 0 if control has to be given to guest 2
651 * - < 0 if an error occurred
653 static int handle_fault(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
657 if (current
->thread
.gmap_int_code
== PGM_PROTECTION
)
658 /* we can directly forward all protection exceptions */
659 return inject_fault(vcpu
, PGM_PROTECTION
,
660 current
->thread
.gmap_addr
, 1);
662 rc
= kvm_s390_shadow_fault(vcpu
, vsie_page
->gmap
,
663 current
->thread
.gmap_addr
);
665 rc
= inject_fault(vcpu
, rc
,
666 current
->thread
.gmap_addr
,
667 current
->thread
.gmap_write_flag
);
672 static inline void clear_vsie_icpt(struct vsie_page
*vsie_page
)
674 vsie_page
->scb_s
.icptcode
= 0;
677 /* rewind the psw and clear the vsie icpt, so we can retry execution */
678 static void retry_vsie_icpt(struct vsie_page
*vsie_page
)
680 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
681 int ilen
= insn_length(scb_s
->ipa
>> 8);
683 /* take care of EXECUTE instructions */
684 if (scb_s
->icptstatus
& 1) {
685 ilen
= (scb_s
->icptstatus
>> 4) & 0x6;
689 scb_s
->gpsw
.addr
= __rewind_psw(scb_s
->gpsw
, ilen
);
690 clear_vsie_icpt(vsie_page
);
694 * Try to shadow + enable the guest 2 provided facility list.
695 * Retry instruction execution if enabled for and provided by guest 2.
697 * Returns: - 0 if handled (retry or guest 2 icpt)
698 * - > 0 if control has to be given to guest 2
700 static int handle_stfle(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
702 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
703 __u32 fac
= vsie_page
->scb_o
->fac
& 0x7ffffff8U
;
705 if (fac
&& test_kvm_facility(vcpu
->kvm
, 7)) {
706 retry_vsie_icpt(vsie_page
);
707 if (read_guest_real(vcpu
, fac
, &vsie_page
->fac
,
708 sizeof(vsie_page
->fac
)))
709 return set_validity_icpt(scb_s
, 0x1090U
);
710 scb_s
->fac
= (__u32
)(__u64
) &vsie_page
->fac
;
716 * Run the vsie on a shadow scb and a shadow gmap, without any further
717 * sanity checks, handling SIE faults.
719 * Returns: - 0 everything went fine
720 * - > 0 if control has to be given to guest 2
721 * - < 0 if an error occurred
723 static int do_vsie_run(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
725 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
726 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
731 if (test_cpu_flag(CIF_MCCK_PENDING
))
734 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
739 rc
= sie64a(scb_s
, vcpu
->run
->s
.regs
.gprs
);
744 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
747 rc
= 0; /* we could still have an icpt */
748 else if (rc
== -EFAULT
)
749 return handle_fault(vcpu
, vsie_page
);
751 switch (scb_s
->icptcode
) {
753 if (scb_s
->ipa
== 0xb2b0)
754 rc
= handle_stfle(vcpu
, vsie_page
);
757 /* stop not requested by g2 - must have been a kick */
758 if (!(atomic_read(&scb_o
->cpuflags
) & CPUSTAT_STOP_INT
))
759 clear_vsie_icpt(vsie_page
);
762 if ((scb_s
->ipa
& 0xf000) != 0xf000)
763 scb_s
->ipa
+= 0x1000;
769 static void release_gmap_shadow(struct vsie_page
*vsie_page
)
772 gmap_put(vsie_page
->gmap
);
773 WRITE_ONCE(vsie_page
->gmap
, NULL
);
774 prefix_unmapped(vsie_page
);
777 static int acquire_gmap_shadow(struct kvm_vcpu
*vcpu
,
778 struct vsie_page
*vsie_page
)
785 asce
= vcpu
->arch
.sie_block
->gcr
[1];
786 cr0
.val
= vcpu
->arch
.sie_block
->gcr
[0];
787 edat
= cr0
.edat
&& test_kvm_facility(vcpu
->kvm
, 8);
788 edat
+= edat
&& test_kvm_facility(vcpu
->kvm
, 78);
791 * ASCE or EDAT could have changed since last icpt, or the gmap
792 * we're holding has been unshadowed. If the gmap is still valid,
793 * we can safely reuse it.
795 if (vsie_page
->gmap
&& gmap_shadow_valid(vsie_page
->gmap
, asce
, edat
))
798 /* release the old shadow - if any, and mark the prefix as unmapped */
799 release_gmap_shadow(vsie_page
);
800 gmap
= gmap_shadow(vcpu
->arch
.gmap
, asce
, edat
);
802 return PTR_ERR(gmap
);
803 gmap
->private = vcpu
->kvm
;
804 WRITE_ONCE(vsie_page
->gmap
, gmap
);
809 * Run the vsie on a shadowed scb, managing the gmap shadow, handling
810 * prefix pages and faults.
812 * Returns: - 0 if no errors occurred
813 * - > 0 if control has to be given to guest 2
814 * - -ENOMEM if out of memory
816 static int vsie_run(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
818 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
822 rc
= acquire_gmap_shadow(vcpu
, vsie_page
);
824 rc
= map_prefix(vcpu
, vsie_page
);
826 gmap_enable(vsie_page
->gmap
);
827 update_intervention_requests(vsie_page
);
828 rc
= do_vsie_run(vcpu
, vsie_page
);
829 gmap_enable(vcpu
->arch
.gmap
);
834 if (rc
|| scb_s
->icptcode
|| signal_pending(current
) ||
835 kvm_s390_vcpu_has_irq(vcpu
, 0))
841 * Addressing exceptions are always presentes as intercepts.
842 * As addressing exceptions are suppressing and our guest 3 PSW
843 * points at the responsible instruction, we have to
844 * forward the PSW and set the ilc. If we can't read guest 3
845 * instruction, we can use an arbitrary ilc. Let's always use
846 * ilen = 4 for now, so we can avoid reading in guest 3 virtual
847 * memory. (we could also fake the shadow so the hardware
850 scb_s
->icptcode
= ICPT_PROGI
;
851 scb_s
->iprcc
= PGM_ADDRESSING
;
853 scb_s
->gpsw
.addr
= __rewind_psw(scb_s
->gpsw
, 4);
859 * Get or create a vsie page for a scb address.
861 * Returns: - address of a vsie page (cached or new one)
862 * - NULL if the same scb address is already used by another VCPU
863 * - ERR_PTR(-ENOMEM) if out of memory
865 static struct vsie_page
*get_vsie_page(struct kvm
*kvm
, unsigned long addr
)
867 struct vsie_page
*vsie_page
;
872 page
= radix_tree_lookup(&kvm
->arch
.vsie
.addr_to_page
, addr
>> 9);
875 if (page_ref_inc_return(page
) == 2)
876 return page_to_virt(page
);
881 * We want at least #online_vcpus shadows, so every VCPU can execute
882 * the VSIE in parallel.
884 nr_vcpus
= atomic_read(&kvm
->online_vcpus
);
886 mutex_lock(&kvm
->arch
.vsie
.mutex
);
887 if (kvm
->arch
.vsie
.page_count
< nr_vcpus
) {
888 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
| GFP_DMA
);
890 mutex_unlock(&kvm
->arch
.vsie
.mutex
);
891 return ERR_PTR(-ENOMEM
);
894 kvm
->arch
.vsie
.pages
[kvm
->arch
.vsie
.page_count
] = page
;
895 kvm
->arch
.vsie
.page_count
++;
897 /* reuse an existing entry that belongs to nobody */
899 page
= kvm
->arch
.vsie
.pages
[kvm
->arch
.vsie
.next
];
900 if (page_ref_inc_return(page
) == 2)
903 kvm
->arch
.vsie
.next
++;
904 kvm
->arch
.vsie
.next
%= nr_vcpus
;
906 radix_tree_delete(&kvm
->arch
.vsie
.addr_to_page
, page
->index
>> 9);
909 /* double use of the same address */
910 if (radix_tree_insert(&kvm
->arch
.vsie
.addr_to_page
, addr
>> 9, page
)) {
912 mutex_unlock(&kvm
->arch
.vsie
.mutex
);
915 mutex_unlock(&kvm
->arch
.vsie
.mutex
);
917 vsie_page
= page_to_virt(page
);
918 memset(&vsie_page
->scb_s
, 0, sizeof(struct kvm_s390_sie_block
));
919 release_gmap_shadow(vsie_page
);
920 vsie_page
->scb_s
.ihcpu
= 0xffffU
;
924 /* put a vsie page acquired via get_vsie_page */
925 static void put_vsie_page(struct kvm
*kvm
, struct vsie_page
*vsie_page
)
927 struct page
*page
= pfn_to_page(__pa(vsie_page
) >> PAGE_SHIFT
);
932 int kvm_s390_handle_vsie(struct kvm_vcpu
*vcpu
)
934 struct vsie_page
*vsie_page
;
935 unsigned long scb_addr
;
938 vcpu
->stat
.instruction_sie
++;
939 if (!test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_SIEF2
))
941 if (vcpu
->arch
.sie_block
->gpsw
.mask
& PSW_MASK_PSTATE
)
942 return kvm_s390_inject_program_int(vcpu
, PGM_PRIVILEGED_OP
);
944 BUILD_BUG_ON(sizeof(struct vsie_page
) != 4096);
945 scb_addr
= kvm_s390_get_base_disp_s(vcpu
, NULL
);
947 /* 512 byte alignment */
948 if (unlikely(scb_addr
& 0x1ffUL
))
949 return kvm_s390_inject_program_int(vcpu
, PGM_SPECIFICATION
);
951 if (signal_pending(current
) || kvm_s390_vcpu_has_irq(vcpu
, 0))
954 vsie_page
= get_vsie_page(vcpu
->kvm
, scb_addr
);
955 if (IS_ERR(vsie_page
))
956 return PTR_ERR(vsie_page
);
958 /* double use of sie control block - simply do nothing */
961 rc
= pin_scb(vcpu
, vsie_page
, scb_addr
);
964 rc
= shadow_scb(vcpu
, vsie_page
);
967 rc
= pin_blocks(vcpu
, vsie_page
);
970 rc
= vsie_run(vcpu
, vsie_page
);
971 unpin_blocks(vcpu
, vsie_page
);
973 unshadow_scb(vcpu
, vsie_page
);
975 unpin_scb(vcpu
, vsie_page
, scb_addr
);
977 put_vsie_page(vcpu
->kvm
, vsie_page
);
979 return rc
< 0 ? rc
: 0;
982 /* Init the vsie data structures. To be called when a vm is initialized. */
983 void kvm_s390_vsie_init(struct kvm
*kvm
)
985 mutex_init(&kvm
->arch
.vsie
.mutex
);
986 INIT_RADIX_TREE(&kvm
->arch
.vsie
.addr_to_page
, GFP_KERNEL
);
989 /* Destroy the vsie data structures. To be called when a vm is destroyed. */
990 void kvm_s390_vsie_destroy(struct kvm
*kvm
)
992 struct vsie_page
*vsie_page
;
996 mutex_lock(&kvm
->arch
.vsie
.mutex
);
997 for (i
= 0; i
< kvm
->arch
.vsie
.page_count
; i
++) {
998 page
= kvm
->arch
.vsie
.pages
[i
];
999 kvm
->arch
.vsie
.pages
[i
] = NULL
;
1000 vsie_page
= page_to_virt(page
);
1001 release_gmap_shadow(vsie_page
);
1002 /* free the radix tree entry */
1003 radix_tree_delete(&kvm
->arch
.vsie
.addr_to_page
, page
->index
>> 9);
1006 kvm
->arch
.vsie
.page_count
= 0;
1007 mutex_unlock(&kvm
->arch
.vsie
.mutex
);