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>
17 #include <linux/sched/signal.h>
20 #include <asm/mmu_context.h>
28 struct kvm_s390_sie_block scb_s
; /* 0x0000 */
29 /* the pinned originial scb */
30 struct kvm_s390_sie_block
*scb_o
; /* 0x0200 */
31 /* the shadow gmap in use by the vsie_page */
32 struct gmap
*gmap
; /* 0x0208 */
33 /* address of the last reported fault to guest2 */
34 unsigned long fault_addr
; /* 0x0210 */
35 __u8 reserved
[0x0700 - 0x0218]; /* 0x0218 */
36 struct kvm_s390_crypto_cb crycb
; /* 0x0700 */
37 __u8 fac
[S390_ARCH_FAC_LIST_SIZE_BYTE
]; /* 0x0800 */
40 /* trigger a validity icpt for the given scb */
41 static int set_validity_icpt(struct kvm_s390_sie_block
*scb
,
45 scb
->ipb
= ((__u32
) reason_code
) << 16;
46 scb
->icptcode
= ICPT_VALIDITY
;
50 /* mark the prefix as unmapped, this will block the VSIE */
51 static void prefix_unmapped(struct vsie_page
*vsie_page
)
53 atomic_or(PROG_REQUEST
, &vsie_page
->scb_s
.prog20
);
56 /* mark the prefix as unmapped and wait until the VSIE has been left */
57 static void prefix_unmapped_sync(struct vsie_page
*vsie_page
)
59 prefix_unmapped(vsie_page
);
60 if (vsie_page
->scb_s
.prog0c
& PROG_IN_SIE
)
61 atomic_or(CPUSTAT_STOP_INT
, &vsie_page
->scb_s
.cpuflags
);
62 while (vsie_page
->scb_s
.prog0c
& PROG_IN_SIE
)
66 /* mark the prefix as mapped, this will allow the VSIE to run */
67 static void prefix_mapped(struct vsie_page
*vsie_page
)
69 atomic_andnot(PROG_REQUEST
, &vsie_page
->scb_s
.prog20
);
72 /* test if the prefix is mapped into the gmap shadow */
73 static int prefix_is_mapped(struct vsie_page
*vsie_page
)
75 return !(atomic_read(&vsie_page
->scb_s
.prog20
) & PROG_REQUEST
);
78 /* copy the updated intervention request bits into the shadow scb */
79 static void update_intervention_requests(struct vsie_page
*vsie_page
)
81 const int bits
= CPUSTAT_STOP_INT
| CPUSTAT_IO_INT
| CPUSTAT_EXT_INT
;
84 cpuflags
= atomic_read(&vsie_page
->scb_o
->cpuflags
);
85 atomic_andnot(bits
, &vsie_page
->scb_s
.cpuflags
);
86 atomic_or(cpuflags
& bits
, &vsie_page
->scb_s
.cpuflags
);
89 /* shadow (filter and validate) the cpuflags */
90 static int prepare_cpuflags(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
92 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
93 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
94 int newflags
, cpuflags
= atomic_read(&scb_o
->cpuflags
);
96 /* we don't allow ESA/390 guests */
97 if (!(cpuflags
& CPUSTAT_ZARCH
))
98 return set_validity_icpt(scb_s
, 0x0001U
);
100 if (cpuflags
& (CPUSTAT_RRF
| CPUSTAT_MCDS
))
101 return set_validity_icpt(scb_s
, 0x0001U
);
102 else if (cpuflags
& (CPUSTAT_SLSV
| CPUSTAT_SLSR
))
103 return set_validity_icpt(scb_s
, 0x0007U
);
105 /* intervention requests will be set later */
106 newflags
= CPUSTAT_ZARCH
;
107 if (cpuflags
& CPUSTAT_GED
&& test_kvm_facility(vcpu
->kvm
, 8))
108 newflags
|= CPUSTAT_GED
;
109 if (cpuflags
& CPUSTAT_GED2
&& test_kvm_facility(vcpu
->kvm
, 78)) {
110 if (cpuflags
& CPUSTAT_GED
)
111 return set_validity_icpt(scb_s
, 0x0001U
);
112 newflags
|= CPUSTAT_GED2
;
114 if (test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_GPERE
))
115 newflags
|= cpuflags
& CPUSTAT_P
;
116 if (test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_GSLS
))
117 newflags
|= cpuflags
& CPUSTAT_SM
;
118 if (test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_IBS
))
119 newflags
|= cpuflags
& CPUSTAT_IBS
;
121 atomic_set(&scb_s
->cpuflags
, newflags
);
126 * Create a shadow copy of the crycb block and setup key wrapping, if
127 * requested for guest 3 and enabled for guest 2.
129 * We only accept format-1 (no AP in g2), but convert it into format-2
130 * There is nothing to do for format-0.
132 * Returns: - 0 if shadowed or nothing to do
133 * - > 0 if control has to be given to guest 2
135 static int shadow_crycb(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
137 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
138 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
139 u32 crycb_addr
= scb_o
->crycbd
& 0x7ffffff8U
;
140 unsigned long *b1
, *b2
;
144 if (!(scb_o
->crycbd
& vcpu
->arch
.sie_block
->crycbd
& CRYCB_FORMAT1
))
146 /* format-1 is supported with message-security-assist extension 3 */
147 if (!test_kvm_facility(vcpu
->kvm
, 76))
149 /* we may only allow it if enabled for guest 2 */
150 ecb3_flags
= scb_o
->ecb3
& vcpu
->arch
.sie_block
->ecb3
&
151 (ECB3_AES
| ECB3_DEA
);
155 if ((crycb_addr
& PAGE_MASK
) != ((crycb_addr
+ 128) & PAGE_MASK
))
156 return set_validity_icpt(scb_s
, 0x003CU
);
157 else if (!crycb_addr
)
158 return set_validity_icpt(scb_s
, 0x0039U
);
160 /* copy only the wrapping keys */
161 if (read_guest_real(vcpu
, crycb_addr
+ 72, &vsie_page
->crycb
, 56))
162 return set_validity_icpt(scb_s
, 0x0035U
);
164 scb_s
->ecb3
|= ecb3_flags
;
165 scb_s
->crycbd
= ((__u32
)(__u64
) &vsie_page
->crycb
) | CRYCB_FORMAT1
|
168 /* xor both blocks in one run */
169 b1
= (unsigned long *) vsie_page
->crycb
.dea_wrapping_key_mask
;
170 b2
= (unsigned long *)
171 vcpu
->kvm
->arch
.crypto
.crycb
->dea_wrapping_key_mask
;
172 /* as 56%8 == 0, bitmap_xor won't overwrite any data */
173 bitmap_xor(b1
, b1
, b2
, BITS_PER_BYTE
* 56);
177 /* shadow (round up/down) the ibc to avoid validity icpt */
178 static void prepare_ibc(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
180 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
181 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
182 __u64 min_ibc
= (sclp
.ibc
>> 16) & 0x0fffU
;
185 /* ibc installed in g2 and requested for g3 */
186 if (vcpu
->kvm
->arch
.model
.ibc
&& (scb_o
->ibc
& 0x0fffU
)) {
187 scb_s
->ibc
= scb_o
->ibc
& 0x0fffU
;
188 /* takte care of the minimum ibc level of the machine */
189 if (scb_s
->ibc
< min_ibc
)
190 scb_s
->ibc
= min_ibc
;
191 /* take care of the maximum ibc level set for the guest */
192 if (scb_s
->ibc
> vcpu
->kvm
->arch
.model
.ibc
)
193 scb_s
->ibc
= vcpu
->kvm
->arch
.model
.ibc
;
197 /* unshadow the scb, copying parameters back to the real scb */
198 static void unshadow_scb(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
200 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
201 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
204 scb_o
->icptcode
= scb_s
->icptcode
;
205 scb_o
->icptstatus
= scb_s
->icptstatus
;
206 scb_o
->ipa
= scb_s
->ipa
;
207 scb_o
->ipb
= scb_s
->ipb
;
208 scb_o
->gbea
= scb_s
->gbea
;
211 scb_o
->cputm
= scb_s
->cputm
;
212 scb_o
->ckc
= scb_s
->ckc
;
213 scb_o
->todpr
= scb_s
->todpr
;
216 scb_o
->gpsw
= scb_s
->gpsw
;
217 scb_o
->gg14
= scb_s
->gg14
;
218 scb_o
->gg15
= scb_s
->gg15
;
219 memcpy(scb_o
->gcr
, scb_s
->gcr
, 128);
220 scb_o
->pp
= scb_s
->pp
;
222 /* interrupt intercept */
223 switch (scb_s
->icptcode
) {
227 memcpy((void *)((u64
)scb_o
+ 0xc0),
228 (void *)((u64
)scb_s
+ 0xc0), 0xf0 - 0xc0);
232 memcpy((void *)((u64
)scb_o
+ 0xc0),
233 (void *)((u64
)scb_s
+ 0xc0), 0xd0 - 0xc0);
237 if (scb_s
->ihcpu
!= 0xffffU
)
238 scb_o
->ihcpu
= scb_s
->ihcpu
;
242 * Setup the shadow scb by copying and checking the relevant parts of the g2
245 * Returns: - 0 if the scb has been shadowed
246 * - > 0 if control has to be given to guest 2
248 static int shadow_scb(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
250 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
251 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
252 bool had_tx
= scb_s
->ecb
& ECB_TE
;
253 unsigned long new_mso
= 0;
256 /* make sure we don't have any leftovers when reusing the scb */
265 rc
= prepare_cpuflags(vcpu
, vsie_page
);
270 scb_s
->cputm
= scb_o
->cputm
;
271 scb_s
->ckc
= scb_o
->ckc
;
272 scb_s
->todpr
= scb_o
->todpr
;
273 scb_s
->epoch
= scb_o
->epoch
;
276 scb_s
->gpsw
= scb_o
->gpsw
;
277 scb_s
->gg14
= scb_o
->gg14
;
278 scb_s
->gg15
= scb_o
->gg15
;
279 memcpy(scb_s
->gcr
, scb_o
->gcr
, 128);
280 scb_s
->pp
= scb_o
->pp
;
282 /* interception / execution handling */
283 scb_s
->gbea
= scb_o
->gbea
;
284 scb_s
->lctl
= scb_o
->lctl
;
285 scb_s
->svcc
= scb_o
->svcc
;
286 scb_s
->ictl
= scb_o
->ictl
;
288 * SKEY handling functions can't deal with false setting of PTE invalid
289 * bits. Therefore we cannot provide interpretation and would later
290 * have to provide own emulation handlers.
292 scb_s
->ictl
|= ICTL_ISKE
| ICTL_SSKE
| ICTL_RRBE
;
293 scb_s
->icpua
= scb_o
->icpua
;
295 if (!(atomic_read(&scb_s
->cpuflags
) & CPUSTAT_SM
))
296 new_mso
= scb_o
->mso
& 0xfffffffffff00000UL
;
297 /* if the hva of the prefix changes, we have to remap the prefix */
298 if (scb_s
->mso
!= new_mso
|| scb_s
->prefix
!= scb_o
->prefix
)
299 prefix_unmapped(vsie_page
);
300 /* SIE will do mso/msl validity and exception checks for us */
301 scb_s
->msl
= scb_o
->msl
& 0xfffffffffff00000UL
;
302 scb_s
->mso
= new_mso
;
303 scb_s
->prefix
= scb_o
->prefix
;
305 /* We have to definetly flush the tlb if this scb never ran */
306 if (scb_s
->ihcpu
!= 0xffffU
)
307 scb_s
->ihcpu
= scb_o
->ihcpu
;
309 /* MVPG and Protection Exception Interpretation are always available */
310 scb_s
->eca
|= scb_o
->eca
& (ECA_MVPGI
| ECA_PROTEXCI
);
311 /* Host-protection-interruption introduced with ESOP */
312 if (test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_ESOP
))
313 scb_s
->ecb
|= scb_o
->ecb
& ECB_HOSTPROTINT
;
314 /* transactional execution */
315 if (test_kvm_facility(vcpu
->kvm
, 73)) {
316 /* remap the prefix is tx is toggled on */
317 if ((scb_o
->ecb
& ECB_TE
) && !had_tx
)
318 prefix_unmapped(vsie_page
);
319 scb_s
->ecb
|= scb_o
->ecb
& ECB_TE
;
322 if (test_kvm_facility(vcpu
->kvm
, 129)) {
323 scb_s
->eca
|= scb_o
->eca
& ECA_VX
;
324 scb_s
->ecd
|= scb_o
->ecd
& ECD_HOSTREGMGMT
;
326 /* Run-time-Instrumentation */
327 if (test_kvm_facility(vcpu
->kvm
, 64))
328 scb_s
->ecb3
|= scb_o
->ecb3
& ECB3_RI
;
329 /* Instruction Execution Prevention */
330 if (test_kvm_facility(vcpu
->kvm
, 130))
331 scb_s
->ecb2
|= scb_o
->ecb2
& ECB2_IEP
;
332 if (test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_SIIF
))
333 scb_s
->eca
|= scb_o
->eca
& ECA_SII
;
334 if (test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_IB
))
335 scb_s
->eca
|= scb_o
->eca
& ECA_IB
;
336 if (test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_CEI
))
337 scb_s
->eca
|= scb_o
->eca
& ECA_CEI
;
339 prepare_ibc(vcpu
, vsie_page
);
340 rc
= shadow_crycb(vcpu
, vsie_page
);
343 unshadow_scb(vcpu
, vsie_page
);
347 void kvm_s390_vsie_gmap_notifier(struct gmap
*gmap
, unsigned long start
,
350 struct kvm
*kvm
= gmap
->private;
351 struct vsie_page
*cur
;
352 unsigned long prefix
;
356 if (!gmap_is_shadow(gmap
))
358 if (start
>= 1UL << 31)
359 /* We are only interested in prefix pages */
363 * Only new shadow blocks are added to the list during runtime,
364 * therefore we can safely reference them all the time.
366 for (i
= 0; i
< kvm
->arch
.vsie
.page_count
; i
++) {
367 page
= READ_ONCE(kvm
->arch
.vsie
.pages
[i
]);
370 cur
= page_to_virt(page
);
371 if (READ_ONCE(cur
->gmap
) != gmap
)
373 prefix
= cur
->scb_s
.prefix
<< GUEST_PREFIX_SHIFT
;
374 /* with mso/msl, the prefix lies at an offset */
375 prefix
+= cur
->scb_s
.mso
;
376 if (prefix
<= end
&& start
<= prefix
+ 2 * PAGE_SIZE
- 1)
377 prefix_unmapped_sync(cur
);
382 * Map the first prefix page and if tx is enabled also the second prefix page.
384 * The prefix will be protected, a gmap notifier will inform about unmaps.
385 * The shadow scb must not be executed until the prefix is remapped, this is
386 * guaranteed by properly handling PROG_REQUEST.
388 * Returns: - 0 on if successfully mapped or already mapped
389 * - > 0 if control has to be given to guest 2
390 * - -EAGAIN if the caller can retry immediately
391 * - -ENOMEM if out of memory
393 static int map_prefix(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
395 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
396 u64 prefix
= scb_s
->prefix
<< GUEST_PREFIX_SHIFT
;
399 if (prefix_is_mapped(vsie_page
))
402 /* mark it as mapped so we can catch any concurrent unmappers */
403 prefix_mapped(vsie_page
);
405 /* with mso/msl, the prefix lies at offset *mso* */
406 prefix
+= scb_s
->mso
;
408 rc
= kvm_s390_shadow_fault(vcpu
, vsie_page
->gmap
, prefix
);
409 if (!rc
&& (scb_s
->ecb
& ECB_TE
))
410 rc
= kvm_s390_shadow_fault(vcpu
, vsie_page
->gmap
,
413 * We don't have to mprotect, we will be called for all unshadows.
414 * SIE will detect if protection applies and trigger a validity.
417 prefix_unmapped(vsie_page
);
418 if (rc
> 0 || rc
== -EFAULT
)
419 rc
= set_validity_icpt(scb_s
, 0x0037U
);
424 * Pin the guest page given by gpa and set hpa to the pinned host address.
425 * Will always be pinned writable.
427 * Returns: - 0 on success
428 * - -EINVAL if the gpa is not valid guest storage
429 * - -ENOMEM if out of memory
431 static int pin_guest_page(struct kvm
*kvm
, gpa_t gpa
, hpa_t
*hpa
)
437 hva
= gfn_to_hva(kvm
, gpa_to_gfn(gpa
));
438 if (kvm_is_error_hva(hva
))
440 rc
= get_user_pages_fast(hva
, 1, 1, &page
);
445 *hpa
= (hpa_t
) page_to_virt(page
) + (gpa
& ~PAGE_MASK
);
449 /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */
450 static void unpin_guest_page(struct kvm
*kvm
, gpa_t gpa
, hpa_t hpa
)
454 page
= virt_to_page(hpa
);
455 set_page_dirty_lock(page
);
457 /* mark the page always as dirty for migration */
458 mark_page_dirty(kvm
, gpa_to_gfn(gpa
));
461 /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */
462 static void unpin_blocks(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
464 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
465 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
469 hpa
= (u64
) scb_s
->scaoh
<< 32 | scb_s
->scaol
;
471 gpa
= scb_o
->scaol
& ~0xfUL
;
472 if (test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_64BSCAO
))
473 gpa
|= (u64
) scb_o
->scaoh
<< 32;
474 unpin_guest_page(vcpu
->kvm
, gpa
, hpa
);
481 gpa
= scb_o
->itdba
& ~0xffUL
;
482 unpin_guest_page(vcpu
->kvm
, gpa
, hpa
);
488 gpa
= scb_o
->gvrd
& ~0x1ffUL
;
489 unpin_guest_page(vcpu
->kvm
, gpa
, hpa
);
495 gpa
= scb_o
->riccbd
& ~0x3fUL
;
496 unpin_guest_page(vcpu
->kvm
, gpa
, hpa
);
502 * Instead of shadowing some blocks, we can simply forward them because the
503 * addresses in the scb are 64 bit long.
505 * This works as long as the data lies in one page. If blocks ever exceed one
506 * page, we have to fall back to shadowing.
508 * As we reuse the sca, the vcpu pointers contained in it are invalid. We must
509 * therefore not enable any facilities that access these pointers (e.g. SIGPIF).
511 * Returns: - 0 if all blocks were pinned.
512 * - > 0 if control has to be given to guest 2
513 * - -ENOMEM if out of memory
515 static int pin_blocks(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
517 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
518 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
523 gpa
= scb_o
->scaol
& ~0xfUL
;
524 if (test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_64BSCAO
))
525 gpa
|= (u64
) scb_o
->scaoh
<< 32;
527 if (!(gpa
& ~0x1fffUL
))
528 rc
= set_validity_icpt(scb_s
, 0x0038U
);
529 else if ((gpa
& ~0x1fffUL
) == kvm_s390_get_prefix(vcpu
))
530 rc
= set_validity_icpt(scb_s
, 0x0011U
);
531 else if ((gpa
& PAGE_MASK
) !=
532 ((gpa
+ sizeof(struct bsca_block
) - 1) & PAGE_MASK
))
533 rc
= set_validity_icpt(scb_s
, 0x003bU
);
535 rc
= pin_guest_page(vcpu
->kvm
, gpa
, &hpa
);
537 rc
= set_validity_icpt(scb_s
, 0x0034U
);
541 scb_s
->scaoh
= (u32
)((u64
)hpa
>> 32);
542 scb_s
->scaol
= (u32
)(u64
)hpa
;
545 gpa
= scb_o
->itdba
& ~0xffUL
;
546 if (gpa
&& (scb_s
->ecb
& ECB_TE
)) {
547 if (!(gpa
& ~0x1fffU
)) {
548 rc
= set_validity_icpt(scb_s
, 0x0080U
);
551 /* 256 bytes cannot cross page boundaries */
552 rc
= pin_guest_page(vcpu
->kvm
, gpa
, &hpa
);
554 rc
= set_validity_icpt(scb_s
, 0x0080U
);
560 gpa
= scb_o
->gvrd
& ~0x1ffUL
;
561 if (gpa
&& (scb_s
->eca
& ECA_VX
) && !(scb_s
->ecd
& ECD_HOSTREGMGMT
)) {
562 if (!(gpa
& ~0x1fffUL
)) {
563 rc
= set_validity_icpt(scb_s
, 0x1310U
);
567 * 512 bytes vector registers cannot cross page boundaries
568 * if this block gets bigger, we have to shadow it.
570 rc
= pin_guest_page(vcpu
->kvm
, gpa
, &hpa
);
572 rc
= set_validity_icpt(scb_s
, 0x1310U
);
578 gpa
= scb_o
->riccbd
& ~0x3fUL
;
579 if (gpa
&& (scb_s
->ecb3
& ECB3_RI
)) {
580 if (!(gpa
& ~0x1fffUL
)) {
581 rc
= set_validity_icpt(scb_s
, 0x0043U
);
584 /* 64 bytes cannot cross page boundaries */
585 rc
= pin_guest_page(vcpu
->kvm
, gpa
, &hpa
);
587 rc
= set_validity_icpt(scb_s
, 0x0043U
);
588 /* Validity 0x0044 will be checked by SIE */
595 unpin_blocks(vcpu
, vsie_page
);
599 /* unpin the scb provided by guest 2, marking it as dirty */
600 static void unpin_scb(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
,
603 hpa_t hpa
= (hpa_t
) vsie_page
->scb_o
;
606 unpin_guest_page(vcpu
->kvm
, gpa
, hpa
);
607 vsie_page
->scb_o
= NULL
;
611 * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o.
613 * Returns: - 0 if the scb was pinned.
614 * - > 0 if control has to be given to guest 2
615 * - -ENOMEM if out of memory
617 static int pin_scb(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
,
623 rc
= pin_guest_page(vcpu
->kvm
, gpa
, &hpa
);
625 rc
= kvm_s390_inject_program_int(vcpu
, PGM_ADDRESSING
);
630 vsie_page
->scb_o
= (struct kvm_s390_sie_block
*) hpa
;
635 * Inject a fault into guest 2.
637 * Returns: - > 0 if control has to be given to guest 2
638 * < 0 if an error occurred during injection.
640 static int inject_fault(struct kvm_vcpu
*vcpu
, __u16 code
, __u64 vaddr
,
643 struct kvm_s390_pgm_info pgm
= {
646 /* 0-51: virtual address */
647 (vaddr
& 0xfffffffffffff000UL
) |
648 /* 52-53: store / fetch */
649 (((unsigned int) !write_flag
) + 1) << 10,
650 /* 62-63: asce id (alway primary == 0) */
651 .exc_access_id
= 0, /* always primary */
652 .op_access_id
= 0, /* not MVPG */
656 if (code
== PGM_PROTECTION
)
657 pgm
.trans_exc_code
|= 0x4UL
;
659 rc
= kvm_s390_inject_prog_irq(vcpu
, &pgm
);
664 * Handle a fault during vsie execution on a gmap shadow.
666 * Returns: - 0 if the fault was resolved
667 * - > 0 if control has to be given to guest 2
668 * - < 0 if an error occurred
670 static int handle_fault(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
674 if (current
->thread
.gmap_int_code
== PGM_PROTECTION
)
675 /* we can directly forward all protection exceptions */
676 return inject_fault(vcpu
, PGM_PROTECTION
,
677 current
->thread
.gmap_addr
, 1);
679 rc
= kvm_s390_shadow_fault(vcpu
, vsie_page
->gmap
,
680 current
->thread
.gmap_addr
);
682 rc
= inject_fault(vcpu
, rc
,
683 current
->thread
.gmap_addr
,
684 current
->thread
.gmap_write_flag
);
686 vsie_page
->fault_addr
= current
->thread
.gmap_addr
;
692 * Retry the previous fault that required guest 2 intervention. This avoids
693 * one superfluous SIE re-entry and direct exit.
695 * Will ignore any errors. The next SIE fault will do proper fault handling.
697 static void handle_last_fault(struct kvm_vcpu
*vcpu
,
698 struct vsie_page
*vsie_page
)
700 if (vsie_page
->fault_addr
)
701 kvm_s390_shadow_fault(vcpu
, vsie_page
->gmap
,
702 vsie_page
->fault_addr
);
703 vsie_page
->fault_addr
= 0;
706 static inline void clear_vsie_icpt(struct vsie_page
*vsie_page
)
708 vsie_page
->scb_s
.icptcode
= 0;
711 /* rewind the psw and clear the vsie icpt, so we can retry execution */
712 static void retry_vsie_icpt(struct vsie_page
*vsie_page
)
714 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
715 int ilen
= insn_length(scb_s
->ipa
>> 8);
717 /* take care of EXECUTE instructions */
718 if (scb_s
->icptstatus
& 1) {
719 ilen
= (scb_s
->icptstatus
>> 4) & 0x6;
723 scb_s
->gpsw
.addr
= __rewind_psw(scb_s
->gpsw
, ilen
);
724 clear_vsie_icpt(vsie_page
);
728 * Try to shadow + enable the guest 2 provided facility list.
729 * Retry instruction execution if enabled for and provided by guest 2.
731 * Returns: - 0 if handled (retry or guest 2 icpt)
732 * - > 0 if control has to be given to guest 2
734 static int handle_stfle(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
736 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
737 __u32 fac
= vsie_page
->scb_o
->fac
& 0x7ffffff8U
;
739 if (fac
&& test_kvm_facility(vcpu
->kvm
, 7)) {
740 retry_vsie_icpt(vsie_page
);
741 if (read_guest_real(vcpu
, fac
, &vsie_page
->fac
,
742 sizeof(vsie_page
->fac
)))
743 return set_validity_icpt(scb_s
, 0x1090U
);
744 scb_s
->fac
= (__u32
)(__u64
) &vsie_page
->fac
;
750 * Run the vsie on a shadow scb and a shadow gmap, without any further
751 * sanity checks, handling SIE faults.
753 * Returns: - 0 everything went fine
754 * - > 0 if control has to be given to guest 2
755 * - < 0 if an error occurred
757 static int do_vsie_run(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
759 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
760 struct kvm_s390_sie_block
*scb_o
= vsie_page
->scb_o
;
763 handle_last_fault(vcpu
, vsie_page
);
767 if (test_cpu_flag(CIF_MCCK_PENDING
))
770 srcu_read_unlock(&vcpu
->kvm
->srcu
, vcpu
->srcu_idx
);
772 guest_enter_irqoff();
775 rc
= sie64a(scb_s
, vcpu
->run
->s
.regs
.gprs
);
780 vcpu
->srcu_idx
= srcu_read_lock(&vcpu
->kvm
->srcu
);
783 rc
= 0; /* we could still have an icpt */
784 else if (rc
== -EFAULT
)
785 return handle_fault(vcpu
, vsie_page
);
787 switch (scb_s
->icptcode
) {
789 if (scb_s
->ipa
== 0xb2b0)
790 rc
= handle_stfle(vcpu
, vsie_page
);
793 /* stop not requested by g2 - must have been a kick */
794 if (!(atomic_read(&scb_o
->cpuflags
) & CPUSTAT_STOP_INT
))
795 clear_vsie_icpt(vsie_page
);
798 if ((scb_s
->ipa
& 0xf000) != 0xf000)
799 scb_s
->ipa
+= 0x1000;
805 static void release_gmap_shadow(struct vsie_page
*vsie_page
)
808 gmap_put(vsie_page
->gmap
);
809 WRITE_ONCE(vsie_page
->gmap
, NULL
);
810 prefix_unmapped(vsie_page
);
813 static int acquire_gmap_shadow(struct kvm_vcpu
*vcpu
,
814 struct vsie_page
*vsie_page
)
821 asce
= vcpu
->arch
.sie_block
->gcr
[1];
822 cr0
.val
= vcpu
->arch
.sie_block
->gcr
[0];
823 edat
= cr0
.edat
&& test_kvm_facility(vcpu
->kvm
, 8);
824 edat
+= edat
&& test_kvm_facility(vcpu
->kvm
, 78);
827 * ASCE or EDAT could have changed since last icpt, or the gmap
828 * we're holding has been unshadowed. If the gmap is still valid,
829 * we can safely reuse it.
831 if (vsie_page
->gmap
&& gmap_shadow_valid(vsie_page
->gmap
, asce
, edat
))
834 /* release the old shadow - if any, and mark the prefix as unmapped */
835 release_gmap_shadow(vsie_page
);
836 gmap
= gmap_shadow(vcpu
->arch
.gmap
, asce
, edat
);
838 return PTR_ERR(gmap
);
839 gmap
->private = vcpu
->kvm
;
840 WRITE_ONCE(vsie_page
->gmap
, gmap
);
845 * Register the shadow scb at the VCPU, e.g. for kicking out of vsie.
847 static void register_shadow_scb(struct kvm_vcpu
*vcpu
,
848 struct vsie_page
*vsie_page
)
850 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
852 WRITE_ONCE(vcpu
->arch
.vsie_block
, &vsie_page
->scb_s
);
854 * External calls have to lead to a kick of the vcpu and
855 * therefore the vsie -> Simulate Wait state.
857 atomic_or(CPUSTAT_WAIT
, &vcpu
->arch
.sie_block
->cpuflags
);
859 * We have to adjust the g3 epoch by the g2 epoch. The epoch will
860 * automatically be adjusted on tod clock changes via kvm_sync_clock.
863 scb_s
->epoch
+= vcpu
->kvm
->arch
.epoch
;
868 * Unregister a shadow scb from a VCPU.
870 static void unregister_shadow_scb(struct kvm_vcpu
*vcpu
)
872 atomic_andnot(CPUSTAT_WAIT
, &vcpu
->arch
.sie_block
->cpuflags
);
873 WRITE_ONCE(vcpu
->arch
.vsie_block
, NULL
);
877 * Run the vsie on a shadowed scb, managing the gmap shadow, handling
878 * prefix pages and faults.
880 * Returns: - 0 if no errors occurred
881 * - > 0 if control has to be given to guest 2
882 * - -ENOMEM if out of memory
884 static int vsie_run(struct kvm_vcpu
*vcpu
, struct vsie_page
*vsie_page
)
886 struct kvm_s390_sie_block
*scb_s
= &vsie_page
->scb_s
;
890 rc
= acquire_gmap_shadow(vcpu
, vsie_page
);
892 rc
= map_prefix(vcpu
, vsie_page
);
894 gmap_enable(vsie_page
->gmap
);
895 update_intervention_requests(vsie_page
);
896 rc
= do_vsie_run(vcpu
, vsie_page
);
897 gmap_enable(vcpu
->arch
.gmap
);
899 atomic_andnot(PROG_BLOCK_SIE
, &scb_s
->prog20
);
903 if (rc
|| scb_s
->icptcode
|| signal_pending(current
) ||
904 kvm_s390_vcpu_has_irq(vcpu
, 0))
910 * Addressing exceptions are always presentes as intercepts.
911 * As addressing exceptions are suppressing and our guest 3 PSW
912 * points at the responsible instruction, we have to
913 * forward the PSW and set the ilc. If we can't read guest 3
914 * instruction, we can use an arbitrary ilc. Let's always use
915 * ilen = 4 for now, so we can avoid reading in guest 3 virtual
916 * memory. (we could also fake the shadow so the hardware
919 scb_s
->icptcode
= ICPT_PROGI
;
920 scb_s
->iprcc
= PGM_ADDRESSING
;
922 scb_s
->gpsw
.addr
= __rewind_psw(scb_s
->gpsw
, 4);
928 * Get or create a vsie page for a scb address.
930 * Returns: - address of a vsie page (cached or new one)
931 * - NULL if the same scb address is already used by another VCPU
932 * - ERR_PTR(-ENOMEM) if out of memory
934 static struct vsie_page
*get_vsie_page(struct kvm
*kvm
, unsigned long addr
)
936 struct vsie_page
*vsie_page
;
941 page
= radix_tree_lookup(&kvm
->arch
.vsie
.addr_to_page
, addr
>> 9);
944 if (page_ref_inc_return(page
) == 2)
945 return page_to_virt(page
);
950 * We want at least #online_vcpus shadows, so every VCPU can execute
951 * the VSIE in parallel.
953 nr_vcpus
= atomic_read(&kvm
->online_vcpus
);
955 mutex_lock(&kvm
->arch
.vsie
.mutex
);
956 if (kvm
->arch
.vsie
.page_count
< nr_vcpus
) {
957 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
| GFP_DMA
);
959 mutex_unlock(&kvm
->arch
.vsie
.mutex
);
960 return ERR_PTR(-ENOMEM
);
963 kvm
->arch
.vsie
.pages
[kvm
->arch
.vsie
.page_count
] = page
;
964 kvm
->arch
.vsie
.page_count
++;
966 /* reuse an existing entry that belongs to nobody */
968 page
= kvm
->arch
.vsie
.pages
[kvm
->arch
.vsie
.next
];
969 if (page_ref_inc_return(page
) == 2)
972 kvm
->arch
.vsie
.next
++;
973 kvm
->arch
.vsie
.next
%= nr_vcpus
;
975 radix_tree_delete(&kvm
->arch
.vsie
.addr_to_page
, page
->index
>> 9);
978 /* double use of the same address */
979 if (radix_tree_insert(&kvm
->arch
.vsie
.addr_to_page
, addr
>> 9, page
)) {
981 mutex_unlock(&kvm
->arch
.vsie
.mutex
);
984 mutex_unlock(&kvm
->arch
.vsie
.mutex
);
986 vsie_page
= page_to_virt(page
);
987 memset(&vsie_page
->scb_s
, 0, sizeof(struct kvm_s390_sie_block
));
988 release_gmap_shadow(vsie_page
);
989 vsie_page
->fault_addr
= 0;
990 vsie_page
->scb_s
.ihcpu
= 0xffffU
;
994 /* put a vsie page acquired via get_vsie_page */
995 static void put_vsie_page(struct kvm
*kvm
, struct vsie_page
*vsie_page
)
997 struct page
*page
= pfn_to_page(__pa(vsie_page
) >> PAGE_SHIFT
);
1002 int kvm_s390_handle_vsie(struct kvm_vcpu
*vcpu
)
1004 struct vsie_page
*vsie_page
;
1005 unsigned long scb_addr
;
1008 vcpu
->stat
.instruction_sie
++;
1009 if (!test_kvm_cpu_feat(vcpu
->kvm
, KVM_S390_VM_CPU_FEAT_SIEF2
))
1011 if (vcpu
->arch
.sie_block
->gpsw
.mask
& PSW_MASK_PSTATE
)
1012 return kvm_s390_inject_program_int(vcpu
, PGM_PRIVILEGED_OP
);
1014 BUILD_BUG_ON(sizeof(struct vsie_page
) != 4096);
1015 scb_addr
= kvm_s390_get_base_disp_s(vcpu
, NULL
);
1017 /* 512 byte alignment */
1018 if (unlikely(scb_addr
& 0x1ffUL
))
1019 return kvm_s390_inject_program_int(vcpu
, PGM_SPECIFICATION
);
1021 if (signal_pending(current
) || kvm_s390_vcpu_has_irq(vcpu
, 0))
1024 vsie_page
= get_vsie_page(vcpu
->kvm
, scb_addr
);
1025 if (IS_ERR(vsie_page
))
1026 return PTR_ERR(vsie_page
);
1027 else if (!vsie_page
)
1028 /* double use of sie control block - simply do nothing */
1031 rc
= pin_scb(vcpu
, vsie_page
, scb_addr
);
1034 rc
= shadow_scb(vcpu
, vsie_page
);
1037 rc
= pin_blocks(vcpu
, vsie_page
);
1040 register_shadow_scb(vcpu
, vsie_page
);
1041 rc
= vsie_run(vcpu
, vsie_page
);
1042 unregister_shadow_scb(vcpu
);
1043 unpin_blocks(vcpu
, vsie_page
);
1045 unshadow_scb(vcpu
, vsie_page
);
1047 unpin_scb(vcpu
, vsie_page
, scb_addr
);
1049 put_vsie_page(vcpu
->kvm
, vsie_page
);
1051 return rc
< 0 ? rc
: 0;
1054 /* Init the vsie data structures. To be called when a vm is initialized. */
1055 void kvm_s390_vsie_init(struct kvm
*kvm
)
1057 mutex_init(&kvm
->arch
.vsie
.mutex
);
1058 INIT_RADIX_TREE(&kvm
->arch
.vsie
.addr_to_page
, GFP_KERNEL
);
1061 /* Destroy the vsie data structures. To be called when a vm is destroyed. */
1062 void kvm_s390_vsie_destroy(struct kvm
*kvm
)
1064 struct vsie_page
*vsie_page
;
1068 mutex_lock(&kvm
->arch
.vsie
.mutex
);
1069 for (i
= 0; i
< kvm
->arch
.vsie
.page_count
; i
++) {
1070 page
= kvm
->arch
.vsie
.pages
[i
];
1071 kvm
->arch
.vsie
.pages
[i
] = NULL
;
1072 vsie_page
= page_to_virt(page
);
1073 release_gmap_shadow(vsie_page
);
1074 /* free the radix tree entry */
1075 radix_tree_delete(&kvm
->arch
.vsie
.addr_to_page
, page
->index
>> 9);
1078 kvm
->arch
.vsie
.page_count
= 0;
1079 mutex_unlock(&kvm
->arch
.vsie
.mutex
);
1082 void kvm_s390_vsie_kick(struct kvm_vcpu
*vcpu
)
1084 struct kvm_s390_sie_block
*scb
= READ_ONCE(vcpu
->arch
.vsie_block
);
1087 * Even if the VCPU lets go of the shadow sie block reference, it is
1088 * still valid in the cache. So we can safely kick it.
1091 atomic_or(PROG_BLOCK_SIE
, &scb
->prog20
);
1092 if (scb
->prog0c
& PROG_IN_SIE
)
1093 atomic_or(CPUSTAT_STOP_INT
, &scb
->cpuflags
);