]>
Commit | Line | Data |
---|---|---|
a3508fbe DH |
1 | /* |
2 | * kvm nested virtualization support for s390x | |
3 | * | |
4 | * Copyright IBM Corp. 2016 | |
5 | * | |
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. | |
9 | * | |
10 | * Author(s): David Hildenbrand <dahi@linux.vnet.ibm.com> | |
11 | */ | |
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> | |
174cd4b1 IM |
17 | #include <linux/sched/signal.h> |
18 | ||
a3508fbe DH |
19 | #include <asm/gmap.h> |
20 | #include <asm/mmu_context.h> | |
21 | #include <asm/sclp.h> | |
22 | #include <asm/nmi.h> | |
66b630d5 | 23 | #include <asm/dis.h> |
a3508fbe DH |
24 | #include "kvm-s390.h" |
25 | #include "gaccess.h" | |
26 | ||
27 | struct vsie_page { | |
28 | struct kvm_s390_sie_block scb_s; /* 0x0000 */ | |
d52cd207 QH |
29 | /* |
30 | * the backup info for machine check. ensure it's at | |
31 | * the same offset as that in struct sie_page! | |
32 | */ | |
33 | struct mcck_volatile_info mcck_info; /* 0x0200 */ | |
a3508fbe | 34 | /* the pinned originial scb */ |
d52cd207 | 35 | struct kvm_s390_sie_block *scb_o; /* 0x0218 */ |
a3508fbe | 36 | /* the shadow gmap in use by the vsie_page */ |
d52cd207 | 37 | struct gmap *gmap; /* 0x0220 */ |
1b7029be | 38 | /* address of the last reported fault to guest2 */ |
d52cd207 QH |
39 | unsigned long fault_addr; /* 0x0228 */ |
40 | __u8 reserved[0x0700 - 0x0230]; /* 0x0230 */ | |
bbeaa58b | 41 | struct kvm_s390_crypto_cb crycb; /* 0x0700 */ |
66b630d5 | 42 | __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */ |
1cae0255 | 43 | }; |
a3508fbe DH |
44 | |
45 | /* trigger a validity icpt for the given scb */ | |
46 | static int set_validity_icpt(struct kvm_s390_sie_block *scb, | |
47 | __u16 reason_code) | |
48 | { | |
49 | scb->ipa = 0x1000; | |
50 | scb->ipb = ((__u32) reason_code) << 16; | |
51 | scb->icptcode = ICPT_VALIDITY; | |
52 | return 1; | |
53 | } | |
54 | ||
55 | /* mark the prefix as unmapped, this will block the VSIE */ | |
56 | static void prefix_unmapped(struct vsie_page *vsie_page) | |
57 | { | |
58 | atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20); | |
59 | } | |
60 | ||
61 | /* mark the prefix as unmapped and wait until the VSIE has been left */ | |
62 | static void prefix_unmapped_sync(struct vsie_page *vsie_page) | |
63 | { | |
64 | prefix_unmapped(vsie_page); | |
65 | if (vsie_page->scb_s.prog0c & PROG_IN_SIE) | |
66 | atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags); | |
67 | while (vsie_page->scb_s.prog0c & PROG_IN_SIE) | |
68 | cpu_relax(); | |
69 | } | |
70 | ||
71 | /* mark the prefix as mapped, this will allow the VSIE to run */ | |
72 | static void prefix_mapped(struct vsie_page *vsie_page) | |
73 | { | |
74 | atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20); | |
75 | } | |
76 | ||
06d68a6c DH |
77 | /* test if the prefix is mapped into the gmap shadow */ |
78 | static int prefix_is_mapped(struct vsie_page *vsie_page) | |
79 | { | |
80 | return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST); | |
81 | } | |
a3508fbe DH |
82 | |
83 | /* copy the updated intervention request bits into the shadow scb */ | |
84 | static void update_intervention_requests(struct vsie_page *vsie_page) | |
85 | { | |
86 | const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT; | |
87 | int cpuflags; | |
88 | ||
89 | cpuflags = atomic_read(&vsie_page->scb_o->cpuflags); | |
90 | atomic_andnot(bits, &vsie_page->scb_s.cpuflags); | |
91 | atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags); | |
92 | } | |
93 | ||
94 | /* shadow (filter and validate) the cpuflags */ | |
95 | static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
96 | { | |
97 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
98 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
99 | int newflags, cpuflags = atomic_read(&scb_o->cpuflags); | |
100 | ||
101 | /* we don't allow ESA/390 guests */ | |
102 | if (!(cpuflags & CPUSTAT_ZARCH)) | |
103 | return set_validity_icpt(scb_s, 0x0001U); | |
104 | ||
105 | if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS)) | |
106 | return set_validity_icpt(scb_s, 0x0001U); | |
107 | else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR)) | |
108 | return set_validity_icpt(scb_s, 0x0007U); | |
109 | ||
110 | /* intervention requests will be set later */ | |
111 | newflags = CPUSTAT_ZARCH; | |
535ef81c DH |
112 | if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8)) |
113 | newflags |= CPUSTAT_GED; | |
114 | if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) { | |
115 | if (cpuflags & CPUSTAT_GED) | |
116 | return set_validity_icpt(scb_s, 0x0001U); | |
117 | newflags |= CPUSTAT_GED2; | |
118 | } | |
77d18f6d DH |
119 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE)) |
120 | newflags |= cpuflags & CPUSTAT_P; | |
a1b7b9b2 DH |
121 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS)) |
122 | newflags |= cpuflags & CPUSTAT_SM; | |
7fd7f39d DH |
123 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS)) |
124 | newflags |= cpuflags & CPUSTAT_IBS; | |
730cd632 FA |
125 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS)) |
126 | newflags |= cpuflags & CPUSTAT_KSS; | |
a3508fbe DH |
127 | |
128 | atomic_set(&scb_s->cpuflags, newflags); | |
129 | return 0; | |
130 | } | |
131 | ||
bbeaa58b DH |
132 | /* |
133 | * Create a shadow copy of the crycb block and setup key wrapping, if | |
134 | * requested for guest 3 and enabled for guest 2. | |
135 | * | |
136 | * We only accept format-1 (no AP in g2), but convert it into format-2 | |
137 | * There is nothing to do for format-0. | |
138 | * | |
139 | * Returns: - 0 if shadowed or nothing to do | |
140 | * - > 0 if control has to be given to guest 2 | |
141 | */ | |
142 | static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
143 | { | |
144 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
145 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
146 | u32 crycb_addr = scb_o->crycbd & 0x7ffffff8U; | |
147 | unsigned long *b1, *b2; | |
148 | u8 ecb3_flags; | |
149 | ||
150 | scb_s->crycbd = 0; | |
151 | if (!(scb_o->crycbd & vcpu->arch.sie_block->crycbd & CRYCB_FORMAT1)) | |
152 | return 0; | |
153 | /* format-1 is supported with message-security-assist extension 3 */ | |
154 | if (!test_kvm_facility(vcpu->kvm, 76)) | |
155 | return 0; | |
156 | /* we may only allow it if enabled for guest 2 */ | |
157 | ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 & | |
158 | (ECB3_AES | ECB3_DEA); | |
159 | if (!ecb3_flags) | |
160 | return 0; | |
161 | ||
162 | if ((crycb_addr & PAGE_MASK) != ((crycb_addr + 128) & PAGE_MASK)) | |
163 | return set_validity_icpt(scb_s, 0x003CU); | |
164 | else if (!crycb_addr) | |
165 | return set_validity_icpt(scb_s, 0x0039U); | |
166 | ||
167 | /* copy only the wrapping keys */ | |
168 | if (read_guest_real(vcpu, crycb_addr + 72, &vsie_page->crycb, 56)) | |
169 | return set_validity_icpt(scb_s, 0x0035U); | |
170 | ||
171 | scb_s->ecb3 |= ecb3_flags; | |
172 | scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT1 | | |
173 | CRYCB_FORMAT2; | |
174 | ||
175 | /* xor both blocks in one run */ | |
176 | b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask; | |
177 | b2 = (unsigned long *) | |
178 | vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask; | |
179 | /* as 56%8 == 0, bitmap_xor won't overwrite any data */ | |
180 | bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56); | |
181 | return 0; | |
182 | } | |
183 | ||
3573602b DH |
184 | /* shadow (round up/down) the ibc to avoid validity icpt */ |
185 | static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
186 | { | |
187 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
188 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
189 | __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU; | |
190 | ||
191 | scb_s->ibc = 0; | |
192 | /* ibc installed in g2 and requested for g3 */ | |
193 | if (vcpu->kvm->arch.model.ibc && (scb_o->ibc & 0x0fffU)) { | |
194 | scb_s->ibc = scb_o->ibc & 0x0fffU; | |
195 | /* takte care of the minimum ibc level of the machine */ | |
196 | if (scb_s->ibc < min_ibc) | |
197 | scb_s->ibc = min_ibc; | |
198 | /* take care of the maximum ibc level set for the guest */ | |
199 | if (scb_s->ibc > vcpu->kvm->arch.model.ibc) | |
200 | scb_s->ibc = vcpu->kvm->arch.model.ibc; | |
201 | } | |
202 | } | |
203 | ||
a3508fbe DH |
204 | /* unshadow the scb, copying parameters back to the real scb */ |
205 | static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
206 | { | |
207 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
208 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
209 | ||
210 | /* interception */ | |
211 | scb_o->icptcode = scb_s->icptcode; | |
212 | scb_o->icptstatus = scb_s->icptstatus; | |
213 | scb_o->ipa = scb_s->ipa; | |
214 | scb_o->ipb = scb_s->ipb; | |
215 | scb_o->gbea = scb_s->gbea; | |
216 | ||
217 | /* timer */ | |
218 | scb_o->cputm = scb_s->cputm; | |
219 | scb_o->ckc = scb_s->ckc; | |
220 | scb_o->todpr = scb_s->todpr; | |
221 | ||
222 | /* guest state */ | |
223 | scb_o->gpsw = scb_s->gpsw; | |
224 | scb_o->gg14 = scb_s->gg14; | |
225 | scb_o->gg15 = scb_s->gg15; | |
226 | memcpy(scb_o->gcr, scb_s->gcr, 128); | |
227 | scb_o->pp = scb_s->pp; | |
228 | ||
22abd757 CB |
229 | /* branch prediction */ |
230 | if (test_kvm_facility(vcpu->kvm, 82)) { | |
231 | scb_o->fpf &= ~FPF_BPBC; | |
232 | scb_o->fpf |= scb_s->fpf & FPF_BPBC; | |
233 | } | |
234 | ||
a3508fbe DH |
235 | /* interrupt intercept */ |
236 | switch (scb_s->icptcode) { | |
237 | case ICPT_PROGI: | |
238 | case ICPT_INSTPROGI: | |
239 | case ICPT_EXTINT: | |
240 | memcpy((void *)((u64)scb_o + 0xc0), | |
241 | (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0); | |
242 | break; | |
243 | case ICPT_PARTEXEC: | |
244 | /* MVPG only */ | |
245 | memcpy((void *)((u64)scb_o + 0xc0), | |
246 | (void *)((u64)scb_s + 0xc0), 0xd0 - 0xc0); | |
247 | break; | |
248 | } | |
249 | ||
250 | if (scb_s->ihcpu != 0xffffU) | |
251 | scb_o->ihcpu = scb_s->ihcpu; | |
252 | } | |
253 | ||
254 | /* | |
255 | * Setup the shadow scb by copying and checking the relevant parts of the g2 | |
256 | * provided scb. | |
257 | * | |
258 | * Returns: - 0 if the scb has been shadowed | |
259 | * - > 0 if control has to be given to guest 2 | |
260 | */ | |
261 | static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
262 | { | |
263 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
264 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
0c9d8683 | 265 | bool had_tx = scb_s->ecb & ECB_TE; |
a1b7b9b2 | 266 | unsigned long new_mso = 0; |
a3508fbe DH |
267 | int rc; |
268 | ||
269 | /* make sure we don't have any leftovers when reusing the scb */ | |
270 | scb_s->icptcode = 0; | |
271 | scb_s->eca = 0; | |
272 | scb_s->ecb = 0; | |
273 | scb_s->ecb2 = 0; | |
274 | scb_s->ecb3 = 0; | |
275 | scb_s->ecd = 0; | |
66b630d5 | 276 | scb_s->fac = 0; |
22abd757 | 277 | scb_s->fpf = 0; |
a3508fbe DH |
278 | |
279 | rc = prepare_cpuflags(vcpu, vsie_page); | |
280 | if (rc) | |
281 | goto out; | |
282 | ||
283 | /* timer */ | |
284 | scb_s->cputm = scb_o->cputm; | |
285 | scb_s->ckc = scb_o->ckc; | |
286 | scb_s->todpr = scb_o->todpr; | |
287 | scb_s->epoch = scb_o->epoch; | |
288 | ||
289 | /* guest state */ | |
290 | scb_s->gpsw = scb_o->gpsw; | |
291 | scb_s->gg14 = scb_o->gg14; | |
292 | scb_s->gg15 = scb_o->gg15; | |
293 | memcpy(scb_s->gcr, scb_o->gcr, 128); | |
294 | scb_s->pp = scb_o->pp; | |
295 | ||
296 | /* interception / execution handling */ | |
297 | scb_s->gbea = scb_o->gbea; | |
298 | scb_s->lctl = scb_o->lctl; | |
299 | scb_s->svcc = scb_o->svcc; | |
300 | scb_s->ictl = scb_o->ictl; | |
301 | /* | |
302 | * SKEY handling functions can't deal with false setting of PTE invalid | |
303 | * bits. Therefore we cannot provide interpretation and would later | |
304 | * have to provide own emulation handlers. | |
305 | */ | |
730cd632 FA |
306 | if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS)) |
307 | scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE; | |
308 | ||
a3508fbe DH |
309 | scb_s->icpua = scb_o->icpua; |
310 | ||
a1b7b9b2 DH |
311 | if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM)) |
312 | new_mso = scb_o->mso & 0xfffffffffff00000UL; | |
06d68a6c DH |
313 | /* if the hva of the prefix changes, we have to remap the prefix */ |
314 | if (scb_s->mso != new_mso || scb_s->prefix != scb_o->prefix) | |
315 | prefix_unmapped(vsie_page); | |
a3508fbe DH |
316 | /* SIE will do mso/msl validity and exception checks for us */ |
317 | scb_s->msl = scb_o->msl & 0xfffffffffff00000UL; | |
06d68a6c | 318 | scb_s->mso = new_mso; |
a3508fbe DH |
319 | scb_s->prefix = scb_o->prefix; |
320 | ||
321 | /* We have to definetly flush the tlb if this scb never ran */ | |
322 | if (scb_s->ihcpu != 0xffffU) | |
323 | scb_s->ihcpu = scb_o->ihcpu; | |
324 | ||
325 | /* MVPG and Protection Exception Interpretation are always available */ | |
0c9d8683 | 326 | scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI); |
4ceafa90 DH |
327 | /* Host-protection-interruption introduced with ESOP */ |
328 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP)) | |
0c9d8683 | 329 | scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT; |
166ecb3d DH |
330 | /* transactional execution */ |
331 | if (test_kvm_facility(vcpu->kvm, 73)) { | |
332 | /* remap the prefix is tx is toggled on */ | |
0c9d8683 | 333 | if ((scb_o->ecb & ECB_TE) && !had_tx) |
166ecb3d | 334 | prefix_unmapped(vsie_page); |
0c9d8683 | 335 | scb_s->ecb |= scb_o->ecb & ECB_TE; |
166ecb3d | 336 | } |
22abd757 CB |
337 | /* branch prediction */ |
338 | if (test_kvm_facility(vcpu->kvm, 82)) | |
339 | scb_s->fpf |= scb_o->fpf & FPF_BPBC; | |
c9bc1eab DH |
340 | /* SIMD */ |
341 | if (test_kvm_facility(vcpu->kvm, 129)) { | |
0c9d8683 DH |
342 | scb_s->eca |= scb_o->eca & ECA_VX; |
343 | scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT; | |
c9bc1eab | 344 | } |
588438cb DH |
345 | /* Run-time-Instrumentation */ |
346 | if (test_kvm_facility(vcpu->kvm, 64)) | |
0c9d8683 | 347 | scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI; |
cd1836f5 JF |
348 | /* Instruction Execution Prevention */ |
349 | if (test_kvm_facility(vcpu->kvm, 130)) | |
0c9d8683 | 350 | scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP; |
4e0b1ab7 FZ |
351 | /* Guarded Storage */ |
352 | if (test_kvm_facility(vcpu->kvm, 133)) { | |
353 | scb_s->ecb |= scb_o->ecb & ECB_GS; | |
354 | scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT; | |
355 | } | |
0615a326 | 356 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF)) |
0c9d8683 | 357 | scb_s->eca |= scb_o->eca & ECA_SII; |
5630a8e8 | 358 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB)) |
0c9d8683 | 359 | scb_s->eca |= scb_o->eca & ECA_IB; |
13ee3f67 | 360 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI)) |
0c9d8683 | 361 | scb_s->eca |= scb_o->eca & ECA_CEI; |
a3508fbe | 362 | |
3573602b | 363 | prepare_ibc(vcpu, vsie_page); |
bbeaa58b | 364 | rc = shadow_crycb(vcpu, vsie_page); |
a3508fbe DH |
365 | out: |
366 | if (rc) | |
367 | unshadow_scb(vcpu, vsie_page); | |
368 | return rc; | |
369 | } | |
370 | ||
371 | void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start, | |
372 | unsigned long end) | |
373 | { | |
374 | struct kvm *kvm = gmap->private; | |
375 | struct vsie_page *cur; | |
376 | unsigned long prefix; | |
377 | struct page *page; | |
378 | int i; | |
379 | ||
380 | if (!gmap_is_shadow(gmap)) | |
381 | return; | |
382 | if (start >= 1UL << 31) | |
383 | /* We are only interested in prefix pages */ | |
384 | return; | |
385 | ||
386 | /* | |
387 | * Only new shadow blocks are added to the list during runtime, | |
388 | * therefore we can safely reference them all the time. | |
389 | */ | |
390 | for (i = 0; i < kvm->arch.vsie.page_count; i++) { | |
391 | page = READ_ONCE(kvm->arch.vsie.pages[i]); | |
392 | if (!page) | |
393 | continue; | |
394 | cur = page_to_virt(page); | |
395 | if (READ_ONCE(cur->gmap) != gmap) | |
396 | continue; | |
397 | prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT; | |
398 | /* with mso/msl, the prefix lies at an offset */ | |
399 | prefix += cur->scb_s.mso; | |
166ecb3d | 400 | if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1) |
a3508fbe DH |
401 | prefix_unmapped_sync(cur); |
402 | } | |
403 | } | |
404 | ||
405 | /* | |
166ecb3d | 406 | * Map the first prefix page and if tx is enabled also the second prefix page. |
a3508fbe DH |
407 | * |
408 | * The prefix will be protected, a gmap notifier will inform about unmaps. | |
409 | * The shadow scb must not be executed until the prefix is remapped, this is | |
410 | * guaranteed by properly handling PROG_REQUEST. | |
411 | * | |
412 | * Returns: - 0 on if successfully mapped or already mapped | |
413 | * - > 0 if control has to be given to guest 2 | |
414 | * - -EAGAIN if the caller can retry immediately | |
415 | * - -ENOMEM if out of memory | |
416 | */ | |
417 | static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
418 | { | |
419 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
420 | u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT; | |
421 | int rc; | |
422 | ||
06d68a6c DH |
423 | if (prefix_is_mapped(vsie_page)) |
424 | return 0; | |
425 | ||
a3508fbe DH |
426 | /* mark it as mapped so we can catch any concurrent unmappers */ |
427 | prefix_mapped(vsie_page); | |
428 | ||
429 | /* with mso/msl, the prefix lies at offset *mso* */ | |
430 | prefix += scb_s->mso; | |
431 | ||
432 | rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix); | |
0c9d8683 | 433 | if (!rc && (scb_s->ecb & ECB_TE)) |
166ecb3d DH |
434 | rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, |
435 | prefix + PAGE_SIZE); | |
a3508fbe DH |
436 | /* |
437 | * We don't have to mprotect, we will be called for all unshadows. | |
438 | * SIE will detect if protection applies and trigger a validity. | |
439 | */ | |
440 | if (rc) | |
441 | prefix_unmapped(vsie_page); | |
442 | if (rc > 0 || rc == -EFAULT) | |
443 | rc = set_validity_icpt(scb_s, 0x0037U); | |
444 | return rc; | |
445 | } | |
446 | ||
447 | /* | |
448 | * Pin the guest page given by gpa and set hpa to the pinned host address. | |
449 | * Will always be pinned writable. | |
450 | * | |
451 | * Returns: - 0 on success | |
452 | * - -EINVAL if the gpa is not valid guest storage | |
453 | * - -ENOMEM if out of memory | |
454 | */ | |
455 | static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa) | |
456 | { | |
457 | struct page *page; | |
458 | hva_t hva; | |
459 | int rc; | |
460 | ||
461 | hva = gfn_to_hva(kvm, gpa_to_gfn(gpa)); | |
462 | if (kvm_is_error_hva(hva)) | |
463 | return -EINVAL; | |
464 | rc = get_user_pages_fast(hva, 1, 1, &page); | |
465 | if (rc < 0) | |
466 | return rc; | |
467 | else if (rc != 1) | |
468 | return -ENOMEM; | |
469 | *hpa = (hpa_t) page_to_virt(page) + (gpa & ~PAGE_MASK); | |
470 | return 0; | |
471 | } | |
472 | ||
473 | /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */ | |
474 | static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa) | |
475 | { | |
476 | struct page *page; | |
477 | ||
478 | page = virt_to_page(hpa); | |
479 | set_page_dirty_lock(page); | |
480 | put_page(page); | |
481 | /* mark the page always as dirty for migration */ | |
482 | mark_page_dirty(kvm, gpa_to_gfn(gpa)); | |
483 | } | |
484 | ||
485 | /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */ | |
486 | static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
487 | { | |
488 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
489 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
490 | hpa_t hpa; | |
491 | gpa_t gpa; | |
492 | ||
493 | hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol; | |
494 | if (hpa) { | |
495 | gpa = scb_o->scaol & ~0xfUL; | |
19c439b5 DH |
496 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO)) |
497 | gpa |= (u64) scb_o->scaoh << 32; | |
a3508fbe DH |
498 | unpin_guest_page(vcpu->kvm, gpa, hpa); |
499 | scb_s->scaol = 0; | |
500 | scb_s->scaoh = 0; | |
501 | } | |
166ecb3d DH |
502 | |
503 | hpa = scb_s->itdba; | |
504 | if (hpa) { | |
505 | gpa = scb_o->itdba & ~0xffUL; | |
506 | unpin_guest_page(vcpu->kvm, gpa, hpa); | |
507 | scb_s->itdba = 0; | |
508 | } | |
c9bc1eab DH |
509 | |
510 | hpa = scb_s->gvrd; | |
511 | if (hpa) { | |
512 | gpa = scb_o->gvrd & ~0x1ffUL; | |
513 | unpin_guest_page(vcpu->kvm, gpa, hpa); | |
514 | scb_s->gvrd = 0; | |
515 | } | |
588438cb DH |
516 | |
517 | hpa = scb_s->riccbd; | |
518 | if (hpa) { | |
519 | gpa = scb_o->riccbd & ~0x3fUL; | |
520 | unpin_guest_page(vcpu->kvm, gpa, hpa); | |
521 | scb_s->riccbd = 0; | |
522 | } | |
4e0b1ab7 FZ |
523 | |
524 | hpa = scb_s->sdnxo; | |
525 | if (hpa) { | |
526 | gpa = scb_o->sdnxo; | |
527 | unpin_guest_page(vcpu->kvm, gpa, hpa); | |
528 | scb_s->sdnxo = 0; | |
529 | } | |
a3508fbe DH |
530 | } |
531 | ||
532 | /* | |
533 | * Instead of shadowing some blocks, we can simply forward them because the | |
534 | * addresses in the scb are 64 bit long. | |
535 | * | |
536 | * This works as long as the data lies in one page. If blocks ever exceed one | |
537 | * page, we have to fall back to shadowing. | |
538 | * | |
539 | * As we reuse the sca, the vcpu pointers contained in it are invalid. We must | |
540 | * therefore not enable any facilities that access these pointers (e.g. SIGPIF). | |
541 | * | |
542 | * Returns: - 0 if all blocks were pinned. | |
543 | * - > 0 if control has to be given to guest 2 | |
544 | * - -ENOMEM if out of memory | |
545 | */ | |
546 | static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
547 | { | |
548 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
549 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
550 | hpa_t hpa; | |
551 | gpa_t gpa; | |
552 | int rc = 0; | |
553 | ||
554 | gpa = scb_o->scaol & ~0xfUL; | |
19c439b5 DH |
555 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO)) |
556 | gpa |= (u64) scb_o->scaoh << 32; | |
a3508fbe DH |
557 | if (gpa) { |
558 | if (!(gpa & ~0x1fffUL)) | |
559 | rc = set_validity_icpt(scb_s, 0x0038U); | |
560 | else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu)) | |
561 | rc = set_validity_icpt(scb_s, 0x0011U); | |
562 | else if ((gpa & PAGE_MASK) != | |
563 | ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK)) | |
564 | rc = set_validity_icpt(scb_s, 0x003bU); | |
565 | if (!rc) { | |
566 | rc = pin_guest_page(vcpu->kvm, gpa, &hpa); | |
567 | if (rc == -EINVAL) | |
568 | rc = set_validity_icpt(scb_s, 0x0034U); | |
569 | } | |
570 | if (rc) | |
571 | goto unpin; | |
572 | scb_s->scaoh = (u32)((u64)hpa >> 32); | |
573 | scb_s->scaol = (u32)(u64)hpa; | |
574 | } | |
166ecb3d DH |
575 | |
576 | gpa = scb_o->itdba & ~0xffUL; | |
0c9d8683 | 577 | if (gpa && (scb_s->ecb & ECB_TE)) { |
166ecb3d DH |
578 | if (!(gpa & ~0x1fffU)) { |
579 | rc = set_validity_icpt(scb_s, 0x0080U); | |
580 | goto unpin; | |
581 | } | |
582 | /* 256 bytes cannot cross page boundaries */ | |
583 | rc = pin_guest_page(vcpu->kvm, gpa, &hpa); | |
584 | if (rc == -EINVAL) | |
585 | rc = set_validity_icpt(scb_s, 0x0080U); | |
586 | if (rc) | |
587 | goto unpin; | |
588 | scb_s->itdba = hpa; | |
589 | } | |
c9bc1eab DH |
590 | |
591 | gpa = scb_o->gvrd & ~0x1ffUL; | |
0c9d8683 | 592 | if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) { |
c9bc1eab DH |
593 | if (!(gpa & ~0x1fffUL)) { |
594 | rc = set_validity_icpt(scb_s, 0x1310U); | |
595 | goto unpin; | |
596 | } | |
597 | /* | |
598 | * 512 bytes vector registers cannot cross page boundaries | |
599 | * if this block gets bigger, we have to shadow it. | |
600 | */ | |
601 | rc = pin_guest_page(vcpu->kvm, gpa, &hpa); | |
602 | if (rc == -EINVAL) | |
603 | rc = set_validity_icpt(scb_s, 0x1310U); | |
604 | if (rc) | |
605 | goto unpin; | |
606 | scb_s->gvrd = hpa; | |
607 | } | |
588438cb DH |
608 | |
609 | gpa = scb_o->riccbd & ~0x3fUL; | |
0c9d8683 | 610 | if (gpa && (scb_s->ecb3 & ECB3_RI)) { |
588438cb DH |
611 | if (!(gpa & ~0x1fffUL)) { |
612 | rc = set_validity_icpt(scb_s, 0x0043U); | |
613 | goto unpin; | |
614 | } | |
615 | /* 64 bytes cannot cross page boundaries */ | |
616 | rc = pin_guest_page(vcpu->kvm, gpa, &hpa); | |
617 | if (rc == -EINVAL) | |
618 | rc = set_validity_icpt(scb_s, 0x0043U); | |
619 | /* Validity 0x0044 will be checked by SIE */ | |
620 | if (rc) | |
621 | goto unpin; | |
4d21cef3 | 622 | scb_s->riccbd = hpa; |
588438cb | 623 | } |
4e0b1ab7 FZ |
624 | if ((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) { |
625 | unsigned long sdnxc; | |
626 | ||
627 | gpa = scb_o->sdnxo & ~0xfUL; | |
628 | sdnxc = scb_o->sdnxo & 0xfUL; | |
629 | if (!gpa || !(gpa & ~0x1fffUL)) { | |
630 | rc = set_validity_icpt(scb_s, 0x10b0U); | |
631 | goto unpin; | |
632 | } | |
633 | if (sdnxc < 6 || sdnxc > 12) { | |
634 | rc = set_validity_icpt(scb_s, 0x10b1U); | |
635 | goto unpin; | |
636 | } | |
637 | if (gpa & ((1 << sdnxc) - 1)) { | |
638 | rc = set_validity_icpt(scb_s, 0x10b2U); | |
639 | goto unpin; | |
640 | } | |
641 | /* Due to alignment rules (checked above) this cannot | |
642 | * cross page boundaries | |
643 | */ | |
644 | rc = pin_guest_page(vcpu->kvm, gpa, &hpa); | |
645 | if (rc == -EINVAL) | |
646 | rc = set_validity_icpt(scb_s, 0x10b0U); | |
647 | if (rc) | |
648 | goto unpin; | |
fe722d13 | 649 | scb_s->sdnxo = hpa | sdnxc; |
4e0b1ab7 | 650 | } |
a3508fbe DH |
651 | return 0; |
652 | unpin: | |
653 | unpin_blocks(vcpu, vsie_page); | |
654 | return rc; | |
655 | } | |
656 | ||
657 | /* unpin the scb provided by guest 2, marking it as dirty */ | |
658 | static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, | |
659 | gpa_t gpa) | |
660 | { | |
661 | hpa_t hpa = (hpa_t) vsie_page->scb_o; | |
662 | ||
663 | if (hpa) | |
664 | unpin_guest_page(vcpu->kvm, gpa, hpa); | |
665 | vsie_page->scb_o = NULL; | |
666 | } | |
667 | ||
668 | /* | |
669 | * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o. | |
670 | * | |
671 | * Returns: - 0 if the scb was pinned. | |
672 | * - > 0 if control has to be given to guest 2 | |
673 | * - -ENOMEM if out of memory | |
674 | */ | |
675 | static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, | |
676 | gpa_t gpa) | |
677 | { | |
678 | hpa_t hpa; | |
679 | int rc; | |
680 | ||
681 | rc = pin_guest_page(vcpu->kvm, gpa, &hpa); | |
682 | if (rc == -EINVAL) { | |
683 | rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); | |
684 | if (!rc) | |
685 | rc = 1; | |
686 | } | |
687 | if (!rc) | |
688 | vsie_page->scb_o = (struct kvm_s390_sie_block *) hpa; | |
689 | return rc; | |
690 | } | |
691 | ||
692 | /* | |
693 | * Inject a fault into guest 2. | |
694 | * | |
695 | * Returns: - > 0 if control has to be given to guest 2 | |
696 | * < 0 if an error occurred during injection. | |
697 | */ | |
698 | static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr, | |
699 | bool write_flag) | |
700 | { | |
701 | struct kvm_s390_pgm_info pgm = { | |
702 | .code = code, | |
703 | .trans_exc_code = | |
704 | /* 0-51: virtual address */ | |
705 | (vaddr & 0xfffffffffffff000UL) | | |
706 | /* 52-53: store / fetch */ | |
707 | (((unsigned int) !write_flag) + 1) << 10, | |
708 | /* 62-63: asce id (alway primary == 0) */ | |
709 | .exc_access_id = 0, /* always primary */ | |
710 | .op_access_id = 0, /* not MVPG */ | |
711 | }; | |
712 | int rc; | |
713 | ||
714 | if (code == PGM_PROTECTION) | |
715 | pgm.trans_exc_code |= 0x4UL; | |
716 | ||
717 | rc = kvm_s390_inject_prog_irq(vcpu, &pgm); | |
718 | return rc ? rc : 1; | |
719 | } | |
720 | ||
721 | /* | |
722 | * Handle a fault during vsie execution on a gmap shadow. | |
723 | * | |
724 | * Returns: - 0 if the fault was resolved | |
725 | * - > 0 if control has to be given to guest 2 | |
726 | * - < 0 if an error occurred | |
727 | */ | |
728 | static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
729 | { | |
730 | int rc; | |
731 | ||
732 | if (current->thread.gmap_int_code == PGM_PROTECTION) | |
733 | /* we can directly forward all protection exceptions */ | |
734 | return inject_fault(vcpu, PGM_PROTECTION, | |
735 | current->thread.gmap_addr, 1); | |
736 | ||
737 | rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, | |
738 | current->thread.gmap_addr); | |
739 | if (rc > 0) { | |
740 | rc = inject_fault(vcpu, rc, | |
741 | current->thread.gmap_addr, | |
742 | current->thread.gmap_write_flag); | |
1b7029be DH |
743 | if (rc >= 0) |
744 | vsie_page->fault_addr = current->thread.gmap_addr; | |
a3508fbe DH |
745 | } |
746 | return rc; | |
747 | } | |
748 | ||
1b7029be DH |
749 | /* |
750 | * Retry the previous fault that required guest 2 intervention. This avoids | |
751 | * one superfluous SIE re-entry and direct exit. | |
752 | * | |
753 | * Will ignore any errors. The next SIE fault will do proper fault handling. | |
754 | */ | |
755 | static void handle_last_fault(struct kvm_vcpu *vcpu, | |
756 | struct vsie_page *vsie_page) | |
757 | { | |
758 | if (vsie_page->fault_addr) | |
759 | kvm_s390_shadow_fault(vcpu, vsie_page->gmap, | |
760 | vsie_page->fault_addr); | |
761 | vsie_page->fault_addr = 0; | |
762 | } | |
763 | ||
a3508fbe DH |
764 | static inline void clear_vsie_icpt(struct vsie_page *vsie_page) |
765 | { | |
766 | vsie_page->scb_s.icptcode = 0; | |
767 | } | |
768 | ||
66b630d5 DH |
769 | /* rewind the psw and clear the vsie icpt, so we can retry execution */ |
770 | static void retry_vsie_icpt(struct vsie_page *vsie_page) | |
771 | { | |
772 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
773 | int ilen = insn_length(scb_s->ipa >> 8); | |
774 | ||
775 | /* take care of EXECUTE instructions */ | |
776 | if (scb_s->icptstatus & 1) { | |
777 | ilen = (scb_s->icptstatus >> 4) & 0x6; | |
778 | if (!ilen) | |
779 | ilen = 4; | |
780 | } | |
781 | scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen); | |
782 | clear_vsie_icpt(vsie_page); | |
783 | } | |
784 | ||
785 | /* | |
786 | * Try to shadow + enable the guest 2 provided facility list. | |
787 | * Retry instruction execution if enabled for and provided by guest 2. | |
788 | * | |
789 | * Returns: - 0 if handled (retry or guest 2 icpt) | |
790 | * - > 0 if control has to be given to guest 2 | |
791 | */ | |
792 | static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
793 | { | |
794 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
795 | __u32 fac = vsie_page->scb_o->fac & 0x7ffffff8U; | |
796 | ||
797 | if (fac && test_kvm_facility(vcpu->kvm, 7)) { | |
798 | retry_vsie_icpt(vsie_page); | |
799 | if (read_guest_real(vcpu, fac, &vsie_page->fac, | |
800 | sizeof(vsie_page->fac))) | |
801 | return set_validity_icpt(scb_s, 0x1090U); | |
802 | scb_s->fac = (__u32)(__u64) &vsie_page->fac; | |
803 | } | |
804 | return 0; | |
805 | } | |
806 | ||
a3508fbe DH |
807 | /* |
808 | * Run the vsie on a shadow scb and a shadow gmap, without any further | |
809 | * sanity checks, handling SIE faults. | |
810 | * | |
811 | * Returns: - 0 everything went fine | |
812 | * - > 0 if control has to be given to guest 2 | |
813 | * - < 0 if an error occurred | |
814 | */ | |
815 | static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
816 | { | |
817 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
818 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
d52cd207 QH |
819 | struct mcck_volatile_info *mcck_info; |
820 | struct sie_page *sie_page; | |
a3508fbe DH |
821 | int rc; |
822 | ||
1b7029be DH |
823 | handle_last_fault(vcpu, vsie_page); |
824 | ||
a3508fbe DH |
825 | if (need_resched()) |
826 | schedule(); | |
827 | if (test_cpu_flag(CIF_MCCK_PENDING)) | |
828 | s390_handle_mcck(); | |
829 | ||
830 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); | |
831 | local_irq_disable(); | |
6edaa530 | 832 | guest_enter_irqoff(); |
a3508fbe DH |
833 | local_irq_enable(); |
834 | ||
835 | rc = sie64a(scb_s, vcpu->run->s.regs.gprs); | |
836 | ||
837 | local_irq_disable(); | |
6edaa530 | 838 | guest_exit_irqoff(); |
a3508fbe DH |
839 | local_irq_enable(); |
840 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); | |
841 | ||
d52cd207 QH |
842 | if (rc == -EINTR) { |
843 | VCPU_EVENT(vcpu, 3, "%s", "machine check"); | |
844 | sie_page = container_of(scb_s, struct sie_page, sie_block); | |
845 | mcck_info = &sie_page->mcck_info; | |
846 | kvm_s390_reinject_machine_check(vcpu, mcck_info); | |
847 | return 0; | |
848 | } | |
849 | ||
a3508fbe DH |
850 | if (rc > 0) |
851 | rc = 0; /* we could still have an icpt */ | |
852 | else if (rc == -EFAULT) | |
853 | return handle_fault(vcpu, vsie_page); | |
854 | ||
855 | switch (scb_s->icptcode) { | |
66b630d5 DH |
856 | case ICPT_INST: |
857 | if (scb_s->ipa == 0xb2b0) | |
858 | rc = handle_stfle(vcpu, vsie_page); | |
859 | break; | |
a3508fbe DH |
860 | case ICPT_STOP: |
861 | /* stop not requested by g2 - must have been a kick */ | |
862 | if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT)) | |
863 | clear_vsie_icpt(vsie_page); | |
864 | break; | |
865 | case ICPT_VALIDITY: | |
866 | if ((scb_s->ipa & 0xf000) != 0xf000) | |
867 | scb_s->ipa += 0x1000; | |
868 | break; | |
869 | } | |
870 | return rc; | |
871 | } | |
872 | ||
873 | static void release_gmap_shadow(struct vsie_page *vsie_page) | |
874 | { | |
875 | if (vsie_page->gmap) | |
876 | gmap_put(vsie_page->gmap); | |
877 | WRITE_ONCE(vsie_page->gmap, NULL); | |
06d68a6c | 878 | prefix_unmapped(vsie_page); |
a3508fbe DH |
879 | } |
880 | ||
881 | static int acquire_gmap_shadow(struct kvm_vcpu *vcpu, | |
882 | struct vsie_page *vsie_page) | |
883 | { | |
884 | unsigned long asce; | |
885 | union ctlreg0 cr0; | |
886 | struct gmap *gmap; | |
887 | int edat; | |
888 | ||
889 | asce = vcpu->arch.sie_block->gcr[1]; | |
890 | cr0.val = vcpu->arch.sie_block->gcr[0]; | |
891 | edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8); | |
892 | edat += edat && test_kvm_facility(vcpu->kvm, 78); | |
893 | ||
06d68a6c DH |
894 | /* |
895 | * ASCE or EDAT could have changed since last icpt, or the gmap | |
896 | * we're holding has been unshadowed. If the gmap is still valid, | |
897 | * we can safely reuse it. | |
898 | */ | |
899 | if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat)) | |
900 | return 0; | |
901 | ||
902 | /* release the old shadow - if any, and mark the prefix as unmapped */ | |
903 | release_gmap_shadow(vsie_page); | |
a3508fbe DH |
904 | gmap = gmap_shadow(vcpu->arch.gmap, asce, edat); |
905 | if (IS_ERR(gmap)) | |
906 | return PTR_ERR(gmap); | |
907 | gmap->private = vcpu->kvm; | |
908 | WRITE_ONCE(vsie_page->gmap, gmap); | |
909 | return 0; | |
910 | } | |
911 | ||
adbf1698 DH |
912 | /* |
913 | * Register the shadow scb at the VCPU, e.g. for kicking out of vsie. | |
914 | */ | |
915 | static void register_shadow_scb(struct kvm_vcpu *vcpu, | |
916 | struct vsie_page *vsie_page) | |
917 | { | |
91473b48 DH |
918 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; |
919 | ||
adbf1698 | 920 | WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s); |
b917ae57 DH |
921 | /* |
922 | * External calls have to lead to a kick of the vcpu and | |
923 | * therefore the vsie -> Simulate Wait state. | |
924 | */ | |
925 | atomic_or(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags); | |
91473b48 DH |
926 | /* |
927 | * We have to adjust the g3 epoch by the g2 epoch. The epoch will | |
928 | * automatically be adjusted on tod clock changes via kvm_sync_clock. | |
929 | */ | |
930 | preempt_disable(); | |
931 | scb_s->epoch += vcpu->kvm->arch.epoch; | |
932 | preempt_enable(); | |
adbf1698 DH |
933 | } |
934 | ||
935 | /* | |
936 | * Unregister a shadow scb from a VCPU. | |
937 | */ | |
938 | static void unregister_shadow_scb(struct kvm_vcpu *vcpu) | |
939 | { | |
b917ae57 | 940 | atomic_andnot(CPUSTAT_WAIT, &vcpu->arch.sie_block->cpuflags); |
adbf1698 DH |
941 | WRITE_ONCE(vcpu->arch.vsie_block, NULL); |
942 | } | |
943 | ||
a3508fbe DH |
944 | /* |
945 | * Run the vsie on a shadowed scb, managing the gmap shadow, handling | |
946 | * prefix pages and faults. | |
947 | * | |
948 | * Returns: - 0 if no errors occurred | |
949 | * - > 0 if control has to be given to guest 2 | |
950 | * - -ENOMEM if out of memory | |
951 | */ | |
952 | static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
953 | { | |
954 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
955 | int rc = 0; | |
956 | ||
957 | while (1) { | |
958 | rc = acquire_gmap_shadow(vcpu, vsie_page); | |
959 | if (!rc) | |
960 | rc = map_prefix(vcpu, vsie_page); | |
961 | if (!rc) { | |
962 | gmap_enable(vsie_page->gmap); | |
963 | update_intervention_requests(vsie_page); | |
964 | rc = do_vsie_run(vcpu, vsie_page); | |
965 | gmap_enable(vcpu->arch.gmap); | |
966 | } | |
adbf1698 | 967 | atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20); |
a3508fbe DH |
968 | |
969 | if (rc == -EAGAIN) | |
970 | rc = 0; | |
971 | if (rc || scb_s->icptcode || signal_pending(current) || | |
972 | kvm_s390_vcpu_has_irq(vcpu, 0)) | |
973 | break; | |
0b925159 | 974 | } |
a3508fbe DH |
975 | |
976 | if (rc == -EFAULT) { | |
977 | /* | |
978 | * Addressing exceptions are always presentes as intercepts. | |
979 | * As addressing exceptions are suppressing and our guest 3 PSW | |
980 | * points at the responsible instruction, we have to | |
981 | * forward the PSW and set the ilc. If we can't read guest 3 | |
982 | * instruction, we can use an arbitrary ilc. Let's always use | |
983 | * ilen = 4 for now, so we can avoid reading in guest 3 virtual | |
984 | * memory. (we could also fake the shadow so the hardware | |
985 | * handles it). | |
986 | */ | |
987 | scb_s->icptcode = ICPT_PROGI; | |
988 | scb_s->iprcc = PGM_ADDRESSING; | |
989 | scb_s->pgmilc = 4; | |
990 | scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4); | |
991 | } | |
992 | return rc; | |
993 | } | |
994 | ||
995 | /* | |
996 | * Get or create a vsie page for a scb address. | |
997 | * | |
998 | * Returns: - address of a vsie page (cached or new one) | |
999 | * - NULL if the same scb address is already used by another VCPU | |
1000 | * - ERR_PTR(-ENOMEM) if out of memory | |
1001 | */ | |
1002 | static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr) | |
1003 | { | |
1004 | struct vsie_page *vsie_page; | |
1005 | struct page *page; | |
1006 | int nr_vcpus; | |
1007 | ||
1008 | rcu_read_lock(); | |
1009 | page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9); | |
1010 | rcu_read_unlock(); | |
1011 | if (page) { | |
1012 | if (page_ref_inc_return(page) == 2) | |
1013 | return page_to_virt(page); | |
1014 | page_ref_dec(page); | |
1015 | } | |
1016 | ||
1017 | /* | |
1018 | * We want at least #online_vcpus shadows, so every VCPU can execute | |
1019 | * the VSIE in parallel. | |
1020 | */ | |
1021 | nr_vcpus = atomic_read(&kvm->online_vcpus); | |
1022 | ||
1023 | mutex_lock(&kvm->arch.vsie.mutex); | |
1024 | if (kvm->arch.vsie.page_count < nr_vcpus) { | |
66b630d5 | 1025 | page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA); |
a3508fbe DH |
1026 | if (!page) { |
1027 | mutex_unlock(&kvm->arch.vsie.mutex); | |
1028 | return ERR_PTR(-ENOMEM); | |
1029 | } | |
1030 | page_ref_inc(page); | |
1031 | kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page; | |
1032 | kvm->arch.vsie.page_count++; | |
1033 | } else { | |
1034 | /* reuse an existing entry that belongs to nobody */ | |
1035 | while (true) { | |
1036 | page = kvm->arch.vsie.pages[kvm->arch.vsie.next]; | |
1037 | if (page_ref_inc_return(page) == 2) | |
1038 | break; | |
1039 | page_ref_dec(page); | |
1040 | kvm->arch.vsie.next++; | |
1041 | kvm->arch.vsie.next %= nr_vcpus; | |
1042 | } | |
1043 | radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9); | |
1044 | } | |
1045 | page->index = addr; | |
1046 | /* double use of the same address */ | |
1047 | if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) { | |
1048 | page_ref_dec(page); | |
1049 | mutex_unlock(&kvm->arch.vsie.mutex); | |
1050 | return NULL; | |
1051 | } | |
1052 | mutex_unlock(&kvm->arch.vsie.mutex); | |
1053 | ||
1054 | vsie_page = page_to_virt(page); | |
1055 | memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block)); | |
06d68a6c | 1056 | release_gmap_shadow(vsie_page); |
1b7029be | 1057 | vsie_page->fault_addr = 0; |
a3508fbe DH |
1058 | vsie_page->scb_s.ihcpu = 0xffffU; |
1059 | return vsie_page; | |
1060 | } | |
1061 | ||
1062 | /* put a vsie page acquired via get_vsie_page */ | |
1063 | static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page) | |
1064 | { | |
1065 | struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT); | |
1066 | ||
1067 | page_ref_dec(page); | |
1068 | } | |
1069 | ||
1070 | int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu) | |
1071 | { | |
1072 | struct vsie_page *vsie_page; | |
1073 | unsigned long scb_addr; | |
1074 | int rc; | |
1075 | ||
1076 | vcpu->stat.instruction_sie++; | |
1077 | if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2)) | |
1078 | return -EOPNOTSUPP; | |
1079 | if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) | |
1080 | return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); | |
1081 | ||
1082 | BUILD_BUG_ON(sizeof(struct vsie_page) != 4096); | |
1083 | scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL); | |
1084 | ||
1085 | /* 512 byte alignment */ | |
1086 | if (unlikely(scb_addr & 0x1ffUL)) | |
1087 | return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); | |
1088 | ||
1089 | if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0)) | |
1090 | return 0; | |
1091 | ||
1092 | vsie_page = get_vsie_page(vcpu->kvm, scb_addr); | |
1093 | if (IS_ERR(vsie_page)) | |
1094 | return PTR_ERR(vsie_page); | |
1095 | else if (!vsie_page) | |
1096 | /* double use of sie control block - simply do nothing */ | |
1097 | return 0; | |
1098 | ||
1099 | rc = pin_scb(vcpu, vsie_page, scb_addr); | |
1100 | if (rc) | |
1101 | goto out_put; | |
1102 | rc = shadow_scb(vcpu, vsie_page); | |
1103 | if (rc) | |
1104 | goto out_unpin_scb; | |
1105 | rc = pin_blocks(vcpu, vsie_page); | |
1106 | if (rc) | |
1107 | goto out_unshadow; | |
adbf1698 | 1108 | register_shadow_scb(vcpu, vsie_page); |
a3508fbe | 1109 | rc = vsie_run(vcpu, vsie_page); |
adbf1698 | 1110 | unregister_shadow_scb(vcpu); |
a3508fbe DH |
1111 | unpin_blocks(vcpu, vsie_page); |
1112 | out_unshadow: | |
1113 | unshadow_scb(vcpu, vsie_page); | |
1114 | out_unpin_scb: | |
1115 | unpin_scb(vcpu, vsie_page, scb_addr); | |
1116 | out_put: | |
1117 | put_vsie_page(vcpu->kvm, vsie_page); | |
1118 | ||
1119 | return rc < 0 ? rc : 0; | |
1120 | } | |
1121 | ||
1122 | /* Init the vsie data structures. To be called when a vm is initialized. */ | |
1123 | void kvm_s390_vsie_init(struct kvm *kvm) | |
1124 | { | |
1125 | mutex_init(&kvm->arch.vsie.mutex); | |
1126 | INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL); | |
1127 | } | |
1128 | ||
1129 | /* Destroy the vsie data structures. To be called when a vm is destroyed. */ | |
1130 | void kvm_s390_vsie_destroy(struct kvm *kvm) | |
1131 | { | |
06d68a6c | 1132 | struct vsie_page *vsie_page; |
a3508fbe DH |
1133 | struct page *page; |
1134 | int i; | |
1135 | ||
1136 | mutex_lock(&kvm->arch.vsie.mutex); | |
1137 | for (i = 0; i < kvm->arch.vsie.page_count; i++) { | |
1138 | page = kvm->arch.vsie.pages[i]; | |
1139 | kvm->arch.vsie.pages[i] = NULL; | |
06d68a6c DH |
1140 | vsie_page = page_to_virt(page); |
1141 | release_gmap_shadow(vsie_page); | |
a3508fbe DH |
1142 | /* free the radix tree entry */ |
1143 | radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9); | |
1144 | __free_page(page); | |
1145 | } | |
1146 | kvm->arch.vsie.page_count = 0; | |
1147 | mutex_unlock(&kvm->arch.vsie.mutex); | |
1148 | } | |
adbf1698 DH |
1149 | |
1150 | void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu) | |
1151 | { | |
1152 | struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block); | |
1153 | ||
1154 | /* | |
1155 | * Even if the VCPU lets go of the shadow sie block reference, it is | |
1156 | * still valid in the cache. So we can safely kick it. | |
1157 | */ | |
1158 | if (scb) { | |
1159 | atomic_or(PROG_BLOCK_SIE, &scb->prog20); | |
1160 | if (scb->prog0c & PROG_IN_SIE) | |
1161 | atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags); | |
1162 | } | |
1163 | } |