Commit | Line | Data |
---|---|---|
d809aa23 | 1 | // SPDX-License-Identifier: GPL-2.0 |
a3508fbe DH |
2 | /* |
3 | * kvm nested virtualization support for s390x | |
4 | * | |
5 | * Copyright IBM Corp. 2016 | |
6 | * | |
a3508fbe DH |
7 | * Author(s): David Hildenbrand <dahi@linux.vnet.ibm.com> |
8 | */ | |
9 | #include <linux/vmalloc.h> | |
10 | #include <linux/kvm_host.h> | |
11 | #include <linux/bug.h> | |
12 | #include <linux/list.h> | |
13 | #include <linux/bitmap.h> | |
174cd4b1 IM |
14 | #include <linux/sched/signal.h> |
15 | ||
a3508fbe DH |
16 | #include <asm/gmap.h> |
17 | #include <asm/mmu_context.h> | |
18 | #include <asm/sclp.h> | |
19 | #include <asm/nmi.h> | |
66b630d5 | 20 | #include <asm/dis.h> |
a3508fbe DH |
21 | #include "kvm-s390.h" |
22 | #include "gaccess.h" | |
23 | ||
24 | struct vsie_page { | |
25 | struct kvm_s390_sie_block scb_s; /* 0x0000 */ | |
d52cd207 QH |
26 | /* |
27 | * the backup info for machine check. ensure it's at | |
28 | * the same offset as that in struct sie_page! | |
29 | */ | |
30 | struct mcck_volatile_info mcck_info; /* 0x0200 */ | |
b3ecd4aa DH |
31 | /* |
32 | * The pinned original scb. Be aware that other VCPUs can modify | |
33 | * it while we read from it. Values that are used for conditions or | |
34 | * are reused conditionally, should be accessed via READ_ONCE. | |
35 | */ | |
d52cd207 | 36 | struct kvm_s390_sie_block *scb_o; /* 0x0218 */ |
a3508fbe | 37 | /* the shadow gmap in use by the vsie_page */ |
d52cd207 | 38 | struct gmap *gmap; /* 0x0220 */ |
1b7029be | 39 | /* address of the last reported fault to guest2 */ |
d52cd207 | 40 | unsigned long fault_addr; /* 0x0228 */ |
15e5020e DH |
41 | /* calculated guest addresses of satellite control blocks */ |
42 | gpa_t sca_gpa; /* 0x0230 */ | |
43 | gpa_t itdba_gpa; /* 0x0238 */ | |
44 | gpa_t gvrd_gpa; /* 0x0240 */ | |
45 | gpa_t riccbd_gpa; /* 0x0248 */ | |
46 | gpa_t sdnx_gpa; /* 0x0250 */ | |
47 | __u8 reserved[0x0700 - 0x0258]; /* 0x0258 */ | |
bbeaa58b | 48 | struct kvm_s390_crypto_cb crycb; /* 0x0700 */ |
66b630d5 | 49 | __u8 fac[S390_ARCH_FAC_LIST_SIZE_BYTE]; /* 0x0800 */ |
1cae0255 | 50 | }; |
a3508fbe DH |
51 | |
52 | /* trigger a validity icpt for the given scb */ | |
53 | static int set_validity_icpt(struct kvm_s390_sie_block *scb, | |
54 | __u16 reason_code) | |
55 | { | |
56 | scb->ipa = 0x1000; | |
57 | scb->ipb = ((__u32) reason_code) << 16; | |
58 | scb->icptcode = ICPT_VALIDITY; | |
59 | return 1; | |
60 | } | |
61 | ||
62 | /* mark the prefix as unmapped, this will block the VSIE */ | |
63 | static void prefix_unmapped(struct vsie_page *vsie_page) | |
64 | { | |
65 | atomic_or(PROG_REQUEST, &vsie_page->scb_s.prog20); | |
66 | } | |
67 | ||
68 | /* mark the prefix as unmapped and wait until the VSIE has been left */ | |
69 | static void prefix_unmapped_sync(struct vsie_page *vsie_page) | |
70 | { | |
71 | prefix_unmapped(vsie_page); | |
72 | if (vsie_page->scb_s.prog0c & PROG_IN_SIE) | |
73 | atomic_or(CPUSTAT_STOP_INT, &vsie_page->scb_s.cpuflags); | |
74 | while (vsie_page->scb_s.prog0c & PROG_IN_SIE) | |
75 | cpu_relax(); | |
76 | } | |
77 | ||
78 | /* mark the prefix as mapped, this will allow the VSIE to run */ | |
79 | static void prefix_mapped(struct vsie_page *vsie_page) | |
80 | { | |
81 | atomic_andnot(PROG_REQUEST, &vsie_page->scb_s.prog20); | |
82 | } | |
83 | ||
06d68a6c DH |
84 | /* test if the prefix is mapped into the gmap shadow */ |
85 | static int prefix_is_mapped(struct vsie_page *vsie_page) | |
86 | { | |
87 | return !(atomic_read(&vsie_page->scb_s.prog20) & PROG_REQUEST); | |
88 | } | |
a3508fbe DH |
89 | |
90 | /* copy the updated intervention request bits into the shadow scb */ | |
91 | static void update_intervention_requests(struct vsie_page *vsie_page) | |
92 | { | |
93 | const int bits = CPUSTAT_STOP_INT | CPUSTAT_IO_INT | CPUSTAT_EXT_INT; | |
94 | int cpuflags; | |
95 | ||
96 | cpuflags = atomic_read(&vsie_page->scb_o->cpuflags); | |
97 | atomic_andnot(bits, &vsie_page->scb_s.cpuflags); | |
98 | atomic_or(cpuflags & bits, &vsie_page->scb_s.cpuflags); | |
99 | } | |
100 | ||
101 | /* shadow (filter and validate) the cpuflags */ | |
102 | static int prepare_cpuflags(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
103 | { | |
104 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
105 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
106 | int newflags, cpuflags = atomic_read(&scb_o->cpuflags); | |
107 | ||
108 | /* we don't allow ESA/390 guests */ | |
109 | if (!(cpuflags & CPUSTAT_ZARCH)) | |
110 | return set_validity_icpt(scb_s, 0x0001U); | |
111 | ||
112 | if (cpuflags & (CPUSTAT_RRF | CPUSTAT_MCDS)) | |
113 | return set_validity_icpt(scb_s, 0x0001U); | |
114 | else if (cpuflags & (CPUSTAT_SLSV | CPUSTAT_SLSR)) | |
115 | return set_validity_icpt(scb_s, 0x0007U); | |
116 | ||
117 | /* intervention requests will be set later */ | |
118 | newflags = CPUSTAT_ZARCH; | |
535ef81c DH |
119 | if (cpuflags & CPUSTAT_GED && test_kvm_facility(vcpu->kvm, 8)) |
120 | newflags |= CPUSTAT_GED; | |
121 | if (cpuflags & CPUSTAT_GED2 && test_kvm_facility(vcpu->kvm, 78)) { | |
122 | if (cpuflags & CPUSTAT_GED) | |
123 | return set_validity_icpt(scb_s, 0x0001U); | |
124 | newflags |= CPUSTAT_GED2; | |
125 | } | |
77d18f6d DH |
126 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GPERE)) |
127 | newflags |= cpuflags & CPUSTAT_P; | |
a1b7b9b2 DH |
128 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_GSLS)) |
129 | newflags |= cpuflags & CPUSTAT_SM; | |
7fd7f39d DH |
130 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IBS)) |
131 | newflags |= cpuflags & CPUSTAT_IBS; | |
730cd632 FA |
132 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_KSS)) |
133 | newflags |= cpuflags & CPUSTAT_KSS; | |
a3508fbe DH |
134 | |
135 | atomic_set(&scb_s->cpuflags, newflags); | |
136 | return 0; | |
137 | } | |
138 | ||
bbeaa58b DH |
139 | /* |
140 | * Create a shadow copy of the crycb block and setup key wrapping, if | |
141 | * requested for guest 3 and enabled for guest 2. | |
142 | * | |
143 | * We only accept format-1 (no AP in g2), but convert it into format-2 | |
144 | * There is nothing to do for format-0. | |
145 | * | |
146 | * Returns: - 0 if shadowed or nothing to do | |
147 | * - > 0 if control has to be given to guest 2 | |
148 | */ | |
149 | static int shadow_crycb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
150 | { | |
151 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
152 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
b3ecd4aa DH |
153 | const uint32_t crycbd_o = READ_ONCE(scb_o->crycbd); |
154 | const u32 crycb_addr = crycbd_o & 0x7ffffff8U; | |
bbeaa58b DH |
155 | unsigned long *b1, *b2; |
156 | u8 ecb3_flags; | |
157 | ||
158 | scb_s->crycbd = 0; | |
b3ecd4aa | 159 | if (!(crycbd_o & vcpu->arch.sie_block->crycbd & CRYCB_FORMAT1)) |
bbeaa58b DH |
160 | return 0; |
161 | /* format-1 is supported with message-security-assist extension 3 */ | |
162 | if (!test_kvm_facility(vcpu->kvm, 76)) | |
163 | return 0; | |
164 | /* we may only allow it if enabled for guest 2 */ | |
165 | ecb3_flags = scb_o->ecb3 & vcpu->arch.sie_block->ecb3 & | |
166 | (ECB3_AES | ECB3_DEA); | |
167 | if (!ecb3_flags) | |
168 | return 0; | |
169 | ||
170 | if ((crycb_addr & PAGE_MASK) != ((crycb_addr + 128) & PAGE_MASK)) | |
171 | return set_validity_icpt(scb_s, 0x003CU); | |
172 | else if (!crycb_addr) | |
173 | return set_validity_icpt(scb_s, 0x0039U); | |
174 | ||
175 | /* copy only the wrapping keys */ | |
176 | if (read_guest_real(vcpu, crycb_addr + 72, &vsie_page->crycb, 56)) | |
177 | return set_validity_icpt(scb_s, 0x0035U); | |
178 | ||
179 | scb_s->ecb3 |= ecb3_flags; | |
180 | scb_s->crycbd = ((__u32)(__u64) &vsie_page->crycb) | CRYCB_FORMAT1 | | |
181 | CRYCB_FORMAT2; | |
182 | ||
183 | /* xor both blocks in one run */ | |
184 | b1 = (unsigned long *) vsie_page->crycb.dea_wrapping_key_mask; | |
185 | b2 = (unsigned long *) | |
186 | vcpu->kvm->arch.crypto.crycb->dea_wrapping_key_mask; | |
187 | /* as 56%8 == 0, bitmap_xor won't overwrite any data */ | |
188 | bitmap_xor(b1, b1, b2, BITS_PER_BYTE * 56); | |
189 | return 0; | |
190 | } | |
191 | ||
3573602b DH |
192 | /* shadow (round up/down) the ibc to avoid validity icpt */ |
193 | static void prepare_ibc(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
194 | { | |
195 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
196 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
b3ecd4aa DH |
197 | /* READ_ONCE does not work on bitfields - use a temporary variable */ |
198 | const uint32_t __new_ibc = scb_o->ibc; | |
199 | const uint32_t new_ibc = READ_ONCE(__new_ibc) & 0x0fffU; | |
3573602b DH |
200 | __u64 min_ibc = (sclp.ibc >> 16) & 0x0fffU; |
201 | ||
202 | scb_s->ibc = 0; | |
203 | /* ibc installed in g2 and requested for g3 */ | |
b3ecd4aa DH |
204 | if (vcpu->kvm->arch.model.ibc && new_ibc) { |
205 | scb_s->ibc = new_ibc; | |
3573602b DH |
206 | /* takte care of the minimum ibc level of the machine */ |
207 | if (scb_s->ibc < min_ibc) | |
208 | scb_s->ibc = min_ibc; | |
209 | /* take care of the maximum ibc level set for the guest */ | |
210 | if (scb_s->ibc > vcpu->kvm->arch.model.ibc) | |
211 | scb_s->ibc = vcpu->kvm->arch.model.ibc; | |
212 | } | |
213 | } | |
214 | ||
a3508fbe DH |
215 | /* unshadow the scb, copying parameters back to the real scb */ |
216 | static void unshadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
217 | { | |
218 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
219 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
220 | ||
221 | /* interception */ | |
222 | scb_o->icptcode = scb_s->icptcode; | |
223 | scb_o->icptstatus = scb_s->icptstatus; | |
224 | scb_o->ipa = scb_s->ipa; | |
225 | scb_o->ipb = scb_s->ipb; | |
226 | scb_o->gbea = scb_s->gbea; | |
227 | ||
228 | /* timer */ | |
229 | scb_o->cputm = scb_s->cputm; | |
230 | scb_o->ckc = scb_s->ckc; | |
231 | scb_o->todpr = scb_s->todpr; | |
232 | ||
233 | /* guest state */ | |
234 | scb_o->gpsw = scb_s->gpsw; | |
235 | scb_o->gg14 = scb_s->gg14; | |
236 | scb_o->gg15 = scb_s->gg15; | |
237 | memcpy(scb_o->gcr, scb_s->gcr, 128); | |
238 | scb_o->pp = scb_s->pp; | |
239 | ||
35b3fde6 CB |
240 | /* branch prediction */ |
241 | if (test_kvm_facility(vcpu->kvm, 82)) { | |
242 | scb_o->fpf &= ~FPF_BPBC; | |
243 | scb_o->fpf |= scb_s->fpf & FPF_BPBC; | |
244 | } | |
245 | ||
a3508fbe DH |
246 | /* interrupt intercept */ |
247 | switch (scb_s->icptcode) { | |
248 | case ICPT_PROGI: | |
249 | case ICPT_INSTPROGI: | |
250 | case ICPT_EXTINT: | |
251 | memcpy((void *)((u64)scb_o + 0xc0), | |
252 | (void *)((u64)scb_s + 0xc0), 0xf0 - 0xc0); | |
253 | break; | |
254 | case ICPT_PARTEXEC: | |
255 | /* MVPG only */ | |
256 | memcpy((void *)((u64)scb_o + 0xc0), | |
257 | (void *)((u64)scb_s + 0xc0), 0xd0 - 0xc0); | |
258 | break; | |
259 | } | |
260 | ||
261 | if (scb_s->ihcpu != 0xffffU) | |
262 | scb_o->ihcpu = scb_s->ihcpu; | |
263 | } | |
264 | ||
265 | /* | |
266 | * Setup the shadow scb by copying and checking the relevant parts of the g2 | |
267 | * provided scb. | |
268 | * | |
269 | * Returns: - 0 if the scb has been shadowed | |
270 | * - > 0 if control has to be given to guest 2 | |
271 | */ | |
272 | static int shadow_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
273 | { | |
274 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
275 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
b3ecd4aa DH |
276 | /* READ_ONCE does not work on bitfields - use a temporary variable */ |
277 | const uint32_t __new_prefix = scb_o->prefix; | |
278 | const uint32_t new_prefix = READ_ONCE(__new_prefix); | |
279 | const bool wants_tx = READ_ONCE(scb_o->ecb) & ECB_TE; | |
0c9d8683 | 280 | bool had_tx = scb_s->ecb & ECB_TE; |
a1b7b9b2 | 281 | unsigned long new_mso = 0; |
a3508fbe DH |
282 | int rc; |
283 | ||
284 | /* make sure we don't have any leftovers when reusing the scb */ | |
285 | scb_s->icptcode = 0; | |
286 | scb_s->eca = 0; | |
287 | scb_s->ecb = 0; | |
288 | scb_s->ecb2 = 0; | |
289 | scb_s->ecb3 = 0; | |
290 | scb_s->ecd = 0; | |
66b630d5 | 291 | scb_s->fac = 0; |
35b3fde6 | 292 | scb_s->fpf = 0; |
a3508fbe DH |
293 | |
294 | rc = prepare_cpuflags(vcpu, vsie_page); | |
295 | if (rc) | |
296 | goto out; | |
297 | ||
298 | /* timer */ | |
299 | scb_s->cputm = scb_o->cputm; | |
300 | scb_s->ckc = scb_o->ckc; | |
301 | scb_s->todpr = scb_o->todpr; | |
302 | scb_s->epoch = scb_o->epoch; | |
303 | ||
304 | /* guest state */ | |
305 | scb_s->gpsw = scb_o->gpsw; | |
306 | scb_s->gg14 = scb_o->gg14; | |
307 | scb_s->gg15 = scb_o->gg15; | |
308 | memcpy(scb_s->gcr, scb_o->gcr, 128); | |
309 | scb_s->pp = scb_o->pp; | |
310 | ||
311 | /* interception / execution handling */ | |
312 | scb_s->gbea = scb_o->gbea; | |
313 | scb_s->lctl = scb_o->lctl; | |
314 | scb_s->svcc = scb_o->svcc; | |
315 | scb_s->ictl = scb_o->ictl; | |
316 | /* | |
317 | * SKEY handling functions can't deal with false setting of PTE invalid | |
318 | * bits. Therefore we cannot provide interpretation and would later | |
319 | * have to provide own emulation handlers. | |
320 | */ | |
730cd632 FA |
321 | if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_KSS)) |
322 | scb_s->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE; | |
323 | ||
a3508fbe DH |
324 | scb_s->icpua = scb_o->icpua; |
325 | ||
a1b7b9b2 | 326 | if (!(atomic_read(&scb_s->cpuflags) & CPUSTAT_SM)) |
b3ecd4aa | 327 | new_mso = READ_ONCE(scb_o->mso) & 0xfffffffffff00000UL; |
06d68a6c | 328 | /* if the hva of the prefix changes, we have to remap the prefix */ |
b3ecd4aa | 329 | if (scb_s->mso != new_mso || scb_s->prefix != new_prefix) |
06d68a6c | 330 | prefix_unmapped(vsie_page); |
a3508fbe DH |
331 | /* SIE will do mso/msl validity and exception checks for us */ |
332 | scb_s->msl = scb_o->msl & 0xfffffffffff00000UL; | |
06d68a6c | 333 | scb_s->mso = new_mso; |
b3ecd4aa | 334 | scb_s->prefix = new_prefix; |
a3508fbe DH |
335 | |
336 | /* We have to definetly flush the tlb if this scb never ran */ | |
337 | if (scb_s->ihcpu != 0xffffU) | |
338 | scb_s->ihcpu = scb_o->ihcpu; | |
339 | ||
340 | /* MVPG and Protection Exception Interpretation are always available */ | |
0c9d8683 | 341 | scb_s->eca |= scb_o->eca & (ECA_MVPGI | ECA_PROTEXCI); |
4ceafa90 DH |
342 | /* Host-protection-interruption introduced with ESOP */ |
343 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_ESOP)) | |
0c9d8683 | 344 | scb_s->ecb |= scb_o->ecb & ECB_HOSTPROTINT; |
166ecb3d | 345 | /* transactional execution */ |
b3ecd4aa | 346 | if (test_kvm_facility(vcpu->kvm, 73) && wants_tx) { |
166ecb3d | 347 | /* remap the prefix is tx is toggled on */ |
b3ecd4aa | 348 | if (!had_tx) |
166ecb3d | 349 | prefix_unmapped(vsie_page); |
b3ecd4aa | 350 | scb_s->ecb |= ECB_TE; |
166ecb3d | 351 | } |
35b3fde6 CB |
352 | /* branch prediction */ |
353 | if (test_kvm_facility(vcpu->kvm, 82)) | |
354 | scb_s->fpf |= scb_o->fpf & FPF_BPBC; | |
c9bc1eab DH |
355 | /* SIMD */ |
356 | if (test_kvm_facility(vcpu->kvm, 129)) { | |
0c9d8683 DH |
357 | scb_s->eca |= scb_o->eca & ECA_VX; |
358 | scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT; | |
c9bc1eab | 359 | } |
588438cb DH |
360 | /* Run-time-Instrumentation */ |
361 | if (test_kvm_facility(vcpu->kvm, 64)) | |
0c9d8683 | 362 | scb_s->ecb3 |= scb_o->ecb3 & ECB3_RI; |
cd1836f5 JF |
363 | /* Instruction Execution Prevention */ |
364 | if (test_kvm_facility(vcpu->kvm, 130)) | |
0c9d8683 | 365 | scb_s->ecb2 |= scb_o->ecb2 & ECB2_IEP; |
4e0b1ab7 FZ |
366 | /* Guarded Storage */ |
367 | if (test_kvm_facility(vcpu->kvm, 133)) { | |
368 | scb_s->ecb |= scb_o->ecb & ECB_GS; | |
369 | scb_s->ecd |= scb_o->ecd & ECD_HOSTREGMGMT; | |
370 | } | |
0615a326 | 371 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIIF)) |
0c9d8683 | 372 | scb_s->eca |= scb_o->eca & ECA_SII; |
5630a8e8 | 373 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_IB)) |
0c9d8683 | 374 | scb_s->eca |= scb_o->eca & ECA_IB; |
13ee3f67 | 375 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_CEI)) |
0c9d8683 | 376 | scb_s->eca |= scb_o->eca & ECA_CEI; |
8fa1696e CW |
377 | /* Epoch Extension */ |
378 | if (test_kvm_facility(vcpu->kvm, 139)) | |
379 | scb_s->ecd |= scb_o->ecd & ECD_MEF; | |
a3508fbe | 380 | |
3573602b | 381 | prepare_ibc(vcpu, vsie_page); |
bbeaa58b | 382 | rc = shadow_crycb(vcpu, vsie_page); |
a3508fbe DH |
383 | out: |
384 | if (rc) | |
385 | unshadow_scb(vcpu, vsie_page); | |
386 | return rc; | |
387 | } | |
388 | ||
389 | void kvm_s390_vsie_gmap_notifier(struct gmap *gmap, unsigned long start, | |
390 | unsigned long end) | |
391 | { | |
392 | struct kvm *kvm = gmap->private; | |
393 | struct vsie_page *cur; | |
394 | unsigned long prefix; | |
395 | struct page *page; | |
396 | int i; | |
397 | ||
398 | if (!gmap_is_shadow(gmap)) | |
399 | return; | |
400 | if (start >= 1UL << 31) | |
401 | /* We are only interested in prefix pages */ | |
402 | return; | |
403 | ||
404 | /* | |
405 | * Only new shadow blocks are added to the list during runtime, | |
406 | * therefore we can safely reference them all the time. | |
407 | */ | |
408 | for (i = 0; i < kvm->arch.vsie.page_count; i++) { | |
409 | page = READ_ONCE(kvm->arch.vsie.pages[i]); | |
410 | if (!page) | |
411 | continue; | |
412 | cur = page_to_virt(page); | |
413 | if (READ_ONCE(cur->gmap) != gmap) | |
414 | continue; | |
415 | prefix = cur->scb_s.prefix << GUEST_PREFIX_SHIFT; | |
416 | /* with mso/msl, the prefix lies at an offset */ | |
417 | prefix += cur->scb_s.mso; | |
166ecb3d | 418 | if (prefix <= end && start <= prefix + 2 * PAGE_SIZE - 1) |
a3508fbe DH |
419 | prefix_unmapped_sync(cur); |
420 | } | |
421 | } | |
422 | ||
423 | /* | |
166ecb3d | 424 | * Map the first prefix page and if tx is enabled also the second prefix page. |
a3508fbe DH |
425 | * |
426 | * The prefix will be protected, a gmap notifier will inform about unmaps. | |
427 | * The shadow scb must not be executed until the prefix is remapped, this is | |
428 | * guaranteed by properly handling PROG_REQUEST. | |
429 | * | |
430 | * Returns: - 0 on if successfully mapped or already mapped | |
431 | * - > 0 if control has to be given to guest 2 | |
432 | * - -EAGAIN if the caller can retry immediately | |
433 | * - -ENOMEM if out of memory | |
434 | */ | |
435 | static int map_prefix(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
436 | { | |
437 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
438 | u64 prefix = scb_s->prefix << GUEST_PREFIX_SHIFT; | |
439 | int rc; | |
440 | ||
06d68a6c DH |
441 | if (prefix_is_mapped(vsie_page)) |
442 | return 0; | |
443 | ||
a3508fbe DH |
444 | /* mark it as mapped so we can catch any concurrent unmappers */ |
445 | prefix_mapped(vsie_page); | |
446 | ||
447 | /* with mso/msl, the prefix lies at offset *mso* */ | |
448 | prefix += scb_s->mso; | |
449 | ||
450 | rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, prefix); | |
0c9d8683 | 451 | if (!rc && (scb_s->ecb & ECB_TE)) |
166ecb3d DH |
452 | rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, |
453 | prefix + PAGE_SIZE); | |
a3508fbe DH |
454 | /* |
455 | * We don't have to mprotect, we will be called for all unshadows. | |
456 | * SIE will detect if protection applies and trigger a validity. | |
457 | */ | |
458 | if (rc) | |
459 | prefix_unmapped(vsie_page); | |
460 | if (rc > 0 || rc == -EFAULT) | |
461 | rc = set_validity_icpt(scb_s, 0x0037U); | |
462 | return rc; | |
463 | } | |
464 | ||
465 | /* | |
466 | * Pin the guest page given by gpa and set hpa to the pinned host address. | |
467 | * Will always be pinned writable. | |
468 | * | |
469 | * Returns: - 0 on success | |
470 | * - -EINVAL if the gpa is not valid guest storage | |
a3508fbe DH |
471 | */ |
472 | static int pin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t *hpa) | |
473 | { | |
474 | struct page *page; | |
a3508fbe | 475 | |
f7a6509f DH |
476 | page = gfn_to_page(kvm, gpa_to_gfn(gpa)); |
477 | if (is_error_page(page)) | |
a3508fbe | 478 | return -EINVAL; |
a3508fbe DH |
479 | *hpa = (hpa_t) page_to_virt(page) + (gpa & ~PAGE_MASK); |
480 | return 0; | |
481 | } | |
482 | ||
483 | /* Unpins a page previously pinned via pin_guest_page, marking it as dirty. */ | |
484 | static void unpin_guest_page(struct kvm *kvm, gpa_t gpa, hpa_t hpa) | |
485 | { | |
f7a6509f | 486 | kvm_release_pfn_dirty(hpa >> PAGE_SHIFT); |
a3508fbe DH |
487 | /* mark the page always as dirty for migration */ |
488 | mark_page_dirty(kvm, gpa_to_gfn(gpa)); | |
489 | } | |
490 | ||
491 | /* unpin all blocks previously pinned by pin_blocks(), marking them dirty */ | |
492 | static void unpin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
493 | { | |
a3508fbe DH |
494 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; |
495 | hpa_t hpa; | |
a3508fbe DH |
496 | |
497 | hpa = (u64) scb_s->scaoh << 32 | scb_s->scaol; | |
498 | if (hpa) { | |
15e5020e DH |
499 | unpin_guest_page(vcpu->kvm, vsie_page->sca_gpa, hpa); |
500 | vsie_page->sca_gpa = 0; | |
a3508fbe DH |
501 | scb_s->scaol = 0; |
502 | scb_s->scaoh = 0; | |
503 | } | |
166ecb3d DH |
504 | |
505 | hpa = scb_s->itdba; | |
506 | if (hpa) { | |
15e5020e DH |
507 | unpin_guest_page(vcpu->kvm, vsie_page->itdba_gpa, hpa); |
508 | vsie_page->itdba_gpa = 0; | |
166ecb3d DH |
509 | scb_s->itdba = 0; |
510 | } | |
c9bc1eab DH |
511 | |
512 | hpa = scb_s->gvrd; | |
513 | if (hpa) { | |
15e5020e DH |
514 | unpin_guest_page(vcpu->kvm, vsie_page->gvrd_gpa, hpa); |
515 | vsie_page->gvrd_gpa = 0; | |
c9bc1eab DH |
516 | scb_s->gvrd = 0; |
517 | } | |
588438cb DH |
518 | |
519 | hpa = scb_s->riccbd; | |
520 | if (hpa) { | |
15e5020e DH |
521 | unpin_guest_page(vcpu->kvm, vsie_page->riccbd_gpa, hpa); |
522 | vsie_page->riccbd_gpa = 0; | |
588438cb DH |
523 | scb_s->riccbd = 0; |
524 | } | |
4e0b1ab7 FZ |
525 | |
526 | hpa = scb_s->sdnxo; | |
527 | if (hpa) { | |
15e5020e DH |
528 | unpin_guest_page(vcpu->kvm, vsie_page->sdnx_gpa, hpa); |
529 | vsie_page->sdnx_gpa = 0; | |
4e0b1ab7 FZ |
530 | scb_s->sdnxo = 0; |
531 | } | |
a3508fbe DH |
532 | } |
533 | ||
534 | /* | |
535 | * Instead of shadowing some blocks, we can simply forward them because the | |
536 | * addresses in the scb are 64 bit long. | |
537 | * | |
538 | * This works as long as the data lies in one page. If blocks ever exceed one | |
539 | * page, we have to fall back to shadowing. | |
540 | * | |
541 | * As we reuse the sca, the vcpu pointers contained in it are invalid. We must | |
542 | * therefore not enable any facilities that access these pointers (e.g. SIGPIF). | |
543 | * | |
544 | * Returns: - 0 if all blocks were pinned. | |
545 | * - > 0 if control has to be given to guest 2 | |
546 | * - -ENOMEM if out of memory | |
547 | */ | |
548 | static int pin_blocks(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
549 | { | |
550 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
551 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
552 | hpa_t hpa; | |
553 | gpa_t gpa; | |
554 | int rc = 0; | |
555 | ||
b3ecd4aa | 556 | gpa = READ_ONCE(scb_o->scaol) & ~0xfUL; |
19c439b5 | 557 | if (test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_64BSCAO)) |
b3ecd4aa | 558 | gpa |= (u64) READ_ONCE(scb_o->scaoh) << 32; |
a3508fbe DH |
559 | if (gpa) { |
560 | if (!(gpa & ~0x1fffUL)) | |
561 | rc = set_validity_icpt(scb_s, 0x0038U); | |
562 | else if ((gpa & ~0x1fffUL) == kvm_s390_get_prefix(vcpu)) | |
563 | rc = set_validity_icpt(scb_s, 0x0011U); | |
564 | else if ((gpa & PAGE_MASK) != | |
565 | ((gpa + sizeof(struct bsca_block) - 1) & PAGE_MASK)) | |
566 | rc = set_validity_icpt(scb_s, 0x003bU); | |
567 | if (!rc) { | |
568 | rc = pin_guest_page(vcpu->kvm, gpa, &hpa); | |
f7a6509f | 569 | if (rc) |
a3508fbe DH |
570 | rc = set_validity_icpt(scb_s, 0x0034U); |
571 | } | |
572 | if (rc) | |
573 | goto unpin; | |
15e5020e | 574 | vsie_page->sca_gpa = gpa; |
a3508fbe DH |
575 | scb_s->scaoh = (u32)((u64)hpa >> 32); |
576 | scb_s->scaol = (u32)(u64)hpa; | |
577 | } | |
166ecb3d | 578 | |
b3ecd4aa | 579 | gpa = READ_ONCE(scb_o->itdba) & ~0xffUL; |
0c9d8683 | 580 | if (gpa && (scb_s->ecb & ECB_TE)) { |
f4a551b7 | 581 | if (!(gpa & ~0x1fffUL)) { |
166ecb3d DH |
582 | rc = set_validity_icpt(scb_s, 0x0080U); |
583 | goto unpin; | |
584 | } | |
585 | /* 256 bytes cannot cross page boundaries */ | |
586 | rc = pin_guest_page(vcpu->kvm, gpa, &hpa); | |
f7a6509f | 587 | if (rc) { |
166ecb3d | 588 | rc = set_validity_icpt(scb_s, 0x0080U); |
166ecb3d | 589 | goto unpin; |
f7a6509f | 590 | } |
15e5020e | 591 | vsie_page->itdba_gpa = gpa; |
166ecb3d DH |
592 | scb_s->itdba = hpa; |
593 | } | |
c9bc1eab | 594 | |
b3ecd4aa | 595 | gpa = READ_ONCE(scb_o->gvrd) & ~0x1ffUL; |
0c9d8683 | 596 | if (gpa && (scb_s->eca & ECA_VX) && !(scb_s->ecd & ECD_HOSTREGMGMT)) { |
c9bc1eab DH |
597 | if (!(gpa & ~0x1fffUL)) { |
598 | rc = set_validity_icpt(scb_s, 0x1310U); | |
599 | goto unpin; | |
600 | } | |
601 | /* | |
602 | * 512 bytes vector registers cannot cross page boundaries | |
603 | * if this block gets bigger, we have to shadow it. | |
604 | */ | |
605 | rc = pin_guest_page(vcpu->kvm, gpa, &hpa); | |
f7a6509f | 606 | if (rc) { |
c9bc1eab | 607 | rc = set_validity_icpt(scb_s, 0x1310U); |
c9bc1eab | 608 | goto unpin; |
f7a6509f | 609 | } |
15e5020e | 610 | vsie_page->gvrd_gpa = gpa; |
c9bc1eab DH |
611 | scb_s->gvrd = hpa; |
612 | } | |
588438cb | 613 | |
b3ecd4aa | 614 | gpa = READ_ONCE(scb_o->riccbd) & ~0x3fUL; |
0c9d8683 | 615 | if (gpa && (scb_s->ecb3 & ECB3_RI)) { |
588438cb DH |
616 | if (!(gpa & ~0x1fffUL)) { |
617 | rc = set_validity_icpt(scb_s, 0x0043U); | |
618 | goto unpin; | |
619 | } | |
620 | /* 64 bytes cannot cross page boundaries */ | |
621 | rc = pin_guest_page(vcpu->kvm, gpa, &hpa); | |
f7a6509f | 622 | if (rc) { |
588438cb | 623 | rc = set_validity_icpt(scb_s, 0x0043U); |
588438cb | 624 | goto unpin; |
f7a6509f DH |
625 | } |
626 | /* Validity 0x0044 will be checked by SIE */ | |
15e5020e | 627 | vsie_page->riccbd_gpa = gpa; |
4d21cef3 | 628 | scb_s->riccbd = hpa; |
588438cb | 629 | } |
4e0b1ab7 FZ |
630 | if ((scb_s->ecb & ECB_GS) && !(scb_s->ecd & ECD_HOSTREGMGMT)) { |
631 | unsigned long sdnxc; | |
632 | ||
b3ecd4aa DH |
633 | gpa = READ_ONCE(scb_o->sdnxo) & ~0xfUL; |
634 | sdnxc = READ_ONCE(scb_o->sdnxo) & 0xfUL; | |
4e0b1ab7 FZ |
635 | if (!gpa || !(gpa & ~0x1fffUL)) { |
636 | rc = set_validity_icpt(scb_s, 0x10b0U); | |
637 | goto unpin; | |
638 | } | |
639 | if (sdnxc < 6 || sdnxc > 12) { | |
640 | rc = set_validity_icpt(scb_s, 0x10b1U); | |
641 | goto unpin; | |
642 | } | |
643 | if (gpa & ((1 << sdnxc) - 1)) { | |
644 | rc = set_validity_icpt(scb_s, 0x10b2U); | |
645 | goto unpin; | |
646 | } | |
647 | /* Due to alignment rules (checked above) this cannot | |
648 | * cross page boundaries | |
649 | */ | |
650 | rc = pin_guest_page(vcpu->kvm, gpa, &hpa); | |
f7a6509f | 651 | if (rc) { |
4e0b1ab7 | 652 | rc = set_validity_icpt(scb_s, 0x10b0U); |
4e0b1ab7 | 653 | goto unpin; |
f7a6509f | 654 | } |
15e5020e | 655 | vsie_page->sdnx_gpa = gpa; |
fe722d13 | 656 | scb_s->sdnxo = hpa | sdnxc; |
4e0b1ab7 | 657 | } |
a3508fbe DH |
658 | return 0; |
659 | unpin: | |
660 | unpin_blocks(vcpu, vsie_page); | |
661 | return rc; | |
662 | } | |
663 | ||
664 | /* unpin the scb provided by guest 2, marking it as dirty */ | |
665 | static void unpin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, | |
666 | gpa_t gpa) | |
667 | { | |
668 | hpa_t hpa = (hpa_t) vsie_page->scb_o; | |
669 | ||
670 | if (hpa) | |
671 | unpin_guest_page(vcpu->kvm, gpa, hpa); | |
672 | vsie_page->scb_o = NULL; | |
673 | } | |
674 | ||
675 | /* | |
676 | * Pin the scb at gpa provided by guest 2 at vsie_page->scb_o. | |
677 | * | |
678 | * Returns: - 0 if the scb was pinned. | |
679 | * - > 0 if control has to be given to guest 2 | |
a3508fbe DH |
680 | */ |
681 | static int pin_scb(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page, | |
682 | gpa_t gpa) | |
683 | { | |
684 | hpa_t hpa; | |
685 | int rc; | |
686 | ||
687 | rc = pin_guest_page(vcpu->kvm, gpa, &hpa); | |
f7a6509f | 688 | if (rc) { |
a3508fbe | 689 | rc = kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING); |
f7a6509f DH |
690 | WARN_ON_ONCE(rc); |
691 | return 1; | |
a3508fbe | 692 | } |
f7a6509f DH |
693 | vsie_page->scb_o = (struct kvm_s390_sie_block *) hpa; |
694 | return 0; | |
a3508fbe DH |
695 | } |
696 | ||
697 | /* | |
698 | * Inject a fault into guest 2. | |
699 | * | |
700 | * Returns: - > 0 if control has to be given to guest 2 | |
701 | * < 0 if an error occurred during injection. | |
702 | */ | |
703 | static int inject_fault(struct kvm_vcpu *vcpu, __u16 code, __u64 vaddr, | |
704 | bool write_flag) | |
705 | { | |
706 | struct kvm_s390_pgm_info pgm = { | |
707 | .code = code, | |
708 | .trans_exc_code = | |
709 | /* 0-51: virtual address */ | |
710 | (vaddr & 0xfffffffffffff000UL) | | |
711 | /* 52-53: store / fetch */ | |
712 | (((unsigned int) !write_flag) + 1) << 10, | |
713 | /* 62-63: asce id (alway primary == 0) */ | |
714 | .exc_access_id = 0, /* always primary */ | |
715 | .op_access_id = 0, /* not MVPG */ | |
716 | }; | |
717 | int rc; | |
718 | ||
719 | if (code == PGM_PROTECTION) | |
720 | pgm.trans_exc_code |= 0x4UL; | |
721 | ||
722 | rc = kvm_s390_inject_prog_irq(vcpu, &pgm); | |
723 | return rc ? rc : 1; | |
724 | } | |
725 | ||
726 | /* | |
727 | * Handle a fault during vsie execution on a gmap shadow. | |
728 | * | |
729 | * Returns: - 0 if the fault was resolved | |
730 | * - > 0 if control has to be given to guest 2 | |
731 | * - < 0 if an error occurred | |
732 | */ | |
733 | static int handle_fault(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
734 | { | |
735 | int rc; | |
736 | ||
737 | if (current->thread.gmap_int_code == PGM_PROTECTION) | |
738 | /* we can directly forward all protection exceptions */ | |
739 | return inject_fault(vcpu, PGM_PROTECTION, | |
740 | current->thread.gmap_addr, 1); | |
741 | ||
742 | rc = kvm_s390_shadow_fault(vcpu, vsie_page->gmap, | |
743 | current->thread.gmap_addr); | |
744 | if (rc > 0) { | |
745 | rc = inject_fault(vcpu, rc, | |
746 | current->thread.gmap_addr, | |
747 | current->thread.gmap_write_flag); | |
1b7029be DH |
748 | if (rc >= 0) |
749 | vsie_page->fault_addr = current->thread.gmap_addr; | |
a3508fbe DH |
750 | } |
751 | return rc; | |
752 | } | |
753 | ||
1b7029be DH |
754 | /* |
755 | * Retry the previous fault that required guest 2 intervention. This avoids | |
756 | * one superfluous SIE re-entry and direct exit. | |
757 | * | |
758 | * Will ignore any errors. The next SIE fault will do proper fault handling. | |
759 | */ | |
760 | static void handle_last_fault(struct kvm_vcpu *vcpu, | |
761 | struct vsie_page *vsie_page) | |
762 | { | |
763 | if (vsie_page->fault_addr) | |
764 | kvm_s390_shadow_fault(vcpu, vsie_page->gmap, | |
765 | vsie_page->fault_addr); | |
766 | vsie_page->fault_addr = 0; | |
767 | } | |
768 | ||
a3508fbe DH |
769 | static inline void clear_vsie_icpt(struct vsie_page *vsie_page) |
770 | { | |
771 | vsie_page->scb_s.icptcode = 0; | |
772 | } | |
773 | ||
66b630d5 DH |
774 | /* rewind the psw and clear the vsie icpt, so we can retry execution */ |
775 | static void retry_vsie_icpt(struct vsie_page *vsie_page) | |
776 | { | |
777 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
778 | int ilen = insn_length(scb_s->ipa >> 8); | |
779 | ||
780 | /* take care of EXECUTE instructions */ | |
781 | if (scb_s->icptstatus & 1) { | |
782 | ilen = (scb_s->icptstatus >> 4) & 0x6; | |
783 | if (!ilen) | |
784 | ilen = 4; | |
785 | } | |
786 | scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, ilen); | |
787 | clear_vsie_icpt(vsie_page); | |
788 | } | |
789 | ||
790 | /* | |
791 | * Try to shadow + enable the guest 2 provided facility list. | |
792 | * Retry instruction execution if enabled for and provided by guest 2. | |
793 | * | |
794 | * Returns: - 0 if handled (retry or guest 2 icpt) | |
795 | * - > 0 if control has to be given to guest 2 | |
796 | */ | |
797 | static int handle_stfle(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
798 | { | |
799 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
b3ecd4aa | 800 | __u32 fac = READ_ONCE(vsie_page->scb_o->fac) & 0x7ffffff8U; |
66b630d5 DH |
801 | |
802 | if (fac && test_kvm_facility(vcpu->kvm, 7)) { | |
803 | retry_vsie_icpt(vsie_page); | |
804 | if (read_guest_real(vcpu, fac, &vsie_page->fac, | |
805 | sizeof(vsie_page->fac))) | |
806 | return set_validity_icpt(scb_s, 0x1090U); | |
807 | scb_s->fac = (__u32)(__u64) &vsie_page->fac; | |
808 | } | |
809 | return 0; | |
810 | } | |
811 | ||
a3508fbe DH |
812 | /* |
813 | * Run the vsie on a shadow scb and a shadow gmap, without any further | |
814 | * sanity checks, handling SIE faults. | |
815 | * | |
816 | * Returns: - 0 everything went fine | |
817 | * - > 0 if control has to be given to guest 2 | |
818 | * - < 0 if an error occurred | |
819 | */ | |
820 | static int do_vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
821 | { | |
822 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
823 | struct kvm_s390_sie_block *scb_o = vsie_page->scb_o; | |
f315104a | 824 | int guest_bp_isolation; |
a3508fbe DH |
825 | int rc; |
826 | ||
1b7029be DH |
827 | handle_last_fault(vcpu, vsie_page); |
828 | ||
a3508fbe DH |
829 | if (need_resched()) |
830 | schedule(); | |
831 | if (test_cpu_flag(CIF_MCCK_PENDING)) | |
832 | s390_handle_mcck(); | |
833 | ||
834 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); | |
f315104a CB |
835 | |
836 | /* save current guest state of bp isolation override */ | |
837 | guest_bp_isolation = test_thread_flag(TIF_ISOLATE_BP_GUEST); | |
838 | ||
839 | /* | |
840 | * The guest is running with BPBC, so we have to force it on for our | |
841 | * nested guest. This is done by enabling BPBC globally, so the BPBC | |
842 | * control in the SCB (which the nested guest can modify) is simply | |
843 | * ignored. | |
844 | */ | |
845 | if (test_kvm_facility(vcpu->kvm, 82) && | |
846 | vcpu->arch.sie_block->fpf & FPF_BPBC) | |
847 | set_thread_flag(TIF_ISOLATE_BP_GUEST); | |
848 | ||
a3508fbe | 849 | local_irq_disable(); |
6edaa530 | 850 | guest_enter_irqoff(); |
a3508fbe DH |
851 | local_irq_enable(); |
852 | ||
853 | rc = sie64a(scb_s, vcpu->run->s.regs.gprs); | |
854 | ||
855 | local_irq_disable(); | |
6edaa530 | 856 | guest_exit_irqoff(); |
a3508fbe | 857 | local_irq_enable(); |
f315104a CB |
858 | |
859 | /* restore guest state for bp isolation override */ | |
860 | if (!guest_bp_isolation) | |
861 | clear_thread_flag(TIF_ISOLATE_BP_GUEST); | |
862 | ||
a3508fbe DH |
863 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
864 | ||
d52cd207 QH |
865 | if (rc == -EINTR) { |
866 | VCPU_EVENT(vcpu, 3, "%s", "machine check"); | |
c95c8953 | 867 | kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info); |
d52cd207 QH |
868 | return 0; |
869 | } | |
870 | ||
a3508fbe DH |
871 | if (rc > 0) |
872 | rc = 0; /* we could still have an icpt */ | |
873 | else if (rc == -EFAULT) | |
874 | return handle_fault(vcpu, vsie_page); | |
875 | ||
876 | switch (scb_s->icptcode) { | |
66b630d5 DH |
877 | case ICPT_INST: |
878 | if (scb_s->ipa == 0xb2b0) | |
879 | rc = handle_stfle(vcpu, vsie_page); | |
880 | break; | |
a3508fbe DH |
881 | case ICPT_STOP: |
882 | /* stop not requested by g2 - must have been a kick */ | |
883 | if (!(atomic_read(&scb_o->cpuflags) & CPUSTAT_STOP_INT)) | |
884 | clear_vsie_icpt(vsie_page); | |
885 | break; | |
886 | case ICPT_VALIDITY: | |
887 | if ((scb_s->ipa & 0xf000) != 0xf000) | |
888 | scb_s->ipa += 0x1000; | |
889 | break; | |
890 | } | |
891 | return rc; | |
892 | } | |
893 | ||
894 | static void release_gmap_shadow(struct vsie_page *vsie_page) | |
895 | { | |
896 | if (vsie_page->gmap) | |
897 | gmap_put(vsie_page->gmap); | |
898 | WRITE_ONCE(vsie_page->gmap, NULL); | |
06d68a6c | 899 | prefix_unmapped(vsie_page); |
a3508fbe DH |
900 | } |
901 | ||
902 | static int acquire_gmap_shadow(struct kvm_vcpu *vcpu, | |
903 | struct vsie_page *vsie_page) | |
904 | { | |
905 | unsigned long asce; | |
906 | union ctlreg0 cr0; | |
907 | struct gmap *gmap; | |
908 | int edat; | |
909 | ||
910 | asce = vcpu->arch.sie_block->gcr[1]; | |
911 | cr0.val = vcpu->arch.sie_block->gcr[0]; | |
912 | edat = cr0.edat && test_kvm_facility(vcpu->kvm, 8); | |
913 | edat += edat && test_kvm_facility(vcpu->kvm, 78); | |
914 | ||
06d68a6c DH |
915 | /* |
916 | * ASCE or EDAT could have changed since last icpt, or the gmap | |
917 | * we're holding has been unshadowed. If the gmap is still valid, | |
918 | * we can safely reuse it. | |
919 | */ | |
920 | if (vsie_page->gmap && gmap_shadow_valid(vsie_page->gmap, asce, edat)) | |
921 | return 0; | |
922 | ||
923 | /* release the old shadow - if any, and mark the prefix as unmapped */ | |
924 | release_gmap_shadow(vsie_page); | |
a3508fbe DH |
925 | gmap = gmap_shadow(vcpu->arch.gmap, asce, edat); |
926 | if (IS_ERR(gmap)) | |
927 | return PTR_ERR(gmap); | |
928 | gmap->private = vcpu->kvm; | |
929 | WRITE_ONCE(vsie_page->gmap, gmap); | |
930 | return 0; | |
931 | } | |
932 | ||
adbf1698 DH |
933 | /* |
934 | * Register the shadow scb at the VCPU, e.g. for kicking out of vsie. | |
935 | */ | |
936 | static void register_shadow_scb(struct kvm_vcpu *vcpu, | |
937 | struct vsie_page *vsie_page) | |
938 | { | |
91473b48 DH |
939 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; |
940 | ||
adbf1698 | 941 | WRITE_ONCE(vcpu->arch.vsie_block, &vsie_page->scb_s); |
b917ae57 DH |
942 | /* |
943 | * External calls have to lead to a kick of the vcpu and | |
944 | * therefore the vsie -> Simulate Wait state. | |
945 | */ | |
ef8f4f49 | 946 | kvm_s390_set_cpuflags(vcpu, CPUSTAT_WAIT); |
91473b48 DH |
947 | /* |
948 | * We have to adjust the g3 epoch by the g2 epoch. The epoch will | |
949 | * automatically be adjusted on tod clock changes via kvm_sync_clock. | |
950 | */ | |
951 | preempt_disable(); | |
952 | scb_s->epoch += vcpu->kvm->arch.epoch; | |
8fa1696e CW |
953 | |
954 | if (scb_s->ecd & ECD_MEF) { | |
955 | scb_s->epdx += vcpu->kvm->arch.epdx; | |
956 | if (scb_s->epoch < vcpu->kvm->arch.epoch) | |
957 | scb_s->epdx += 1; | |
958 | } | |
959 | ||
91473b48 | 960 | preempt_enable(); |
adbf1698 DH |
961 | } |
962 | ||
963 | /* | |
964 | * Unregister a shadow scb from a VCPU. | |
965 | */ | |
966 | static void unregister_shadow_scb(struct kvm_vcpu *vcpu) | |
967 | { | |
9daecfc6 | 968 | kvm_s390_clear_cpuflags(vcpu, CPUSTAT_WAIT); |
adbf1698 DH |
969 | WRITE_ONCE(vcpu->arch.vsie_block, NULL); |
970 | } | |
971 | ||
a3508fbe DH |
972 | /* |
973 | * Run the vsie on a shadowed scb, managing the gmap shadow, handling | |
974 | * prefix pages and faults. | |
975 | * | |
976 | * Returns: - 0 if no errors occurred | |
977 | * - > 0 if control has to be given to guest 2 | |
978 | * - -ENOMEM if out of memory | |
979 | */ | |
980 | static int vsie_run(struct kvm_vcpu *vcpu, struct vsie_page *vsie_page) | |
981 | { | |
982 | struct kvm_s390_sie_block *scb_s = &vsie_page->scb_s; | |
983 | int rc = 0; | |
984 | ||
985 | while (1) { | |
986 | rc = acquire_gmap_shadow(vcpu, vsie_page); | |
987 | if (!rc) | |
988 | rc = map_prefix(vcpu, vsie_page); | |
989 | if (!rc) { | |
990 | gmap_enable(vsie_page->gmap); | |
991 | update_intervention_requests(vsie_page); | |
992 | rc = do_vsie_run(vcpu, vsie_page); | |
993 | gmap_enable(vcpu->arch.gmap); | |
994 | } | |
adbf1698 | 995 | atomic_andnot(PROG_BLOCK_SIE, &scb_s->prog20); |
a3508fbe DH |
996 | |
997 | if (rc == -EAGAIN) | |
998 | rc = 0; | |
999 | if (rc || scb_s->icptcode || signal_pending(current) || | |
1000 | kvm_s390_vcpu_has_irq(vcpu, 0)) | |
1001 | break; | |
0b925159 | 1002 | } |
a3508fbe DH |
1003 | |
1004 | if (rc == -EFAULT) { | |
1005 | /* | |
1006 | * Addressing exceptions are always presentes as intercepts. | |
1007 | * As addressing exceptions are suppressing and our guest 3 PSW | |
1008 | * points at the responsible instruction, we have to | |
1009 | * forward the PSW and set the ilc. If we can't read guest 3 | |
1010 | * instruction, we can use an arbitrary ilc. Let's always use | |
1011 | * ilen = 4 for now, so we can avoid reading in guest 3 virtual | |
1012 | * memory. (we could also fake the shadow so the hardware | |
1013 | * handles it). | |
1014 | */ | |
1015 | scb_s->icptcode = ICPT_PROGI; | |
1016 | scb_s->iprcc = PGM_ADDRESSING; | |
1017 | scb_s->pgmilc = 4; | |
1018 | scb_s->gpsw.addr = __rewind_psw(scb_s->gpsw, 4); | |
1019 | } | |
1020 | return rc; | |
1021 | } | |
1022 | ||
1023 | /* | |
1024 | * Get or create a vsie page for a scb address. | |
1025 | * | |
1026 | * Returns: - address of a vsie page (cached or new one) | |
1027 | * - NULL if the same scb address is already used by another VCPU | |
1028 | * - ERR_PTR(-ENOMEM) if out of memory | |
1029 | */ | |
1030 | static struct vsie_page *get_vsie_page(struct kvm *kvm, unsigned long addr) | |
1031 | { | |
1032 | struct vsie_page *vsie_page; | |
1033 | struct page *page; | |
1034 | int nr_vcpus; | |
1035 | ||
1036 | rcu_read_lock(); | |
1037 | page = radix_tree_lookup(&kvm->arch.vsie.addr_to_page, addr >> 9); | |
1038 | rcu_read_unlock(); | |
1039 | if (page) { | |
1040 | if (page_ref_inc_return(page) == 2) | |
1041 | return page_to_virt(page); | |
1042 | page_ref_dec(page); | |
1043 | } | |
1044 | ||
1045 | /* | |
1046 | * We want at least #online_vcpus shadows, so every VCPU can execute | |
1047 | * the VSIE in parallel. | |
1048 | */ | |
1049 | nr_vcpus = atomic_read(&kvm->online_vcpus); | |
1050 | ||
1051 | mutex_lock(&kvm->arch.vsie.mutex); | |
1052 | if (kvm->arch.vsie.page_count < nr_vcpus) { | |
66b630d5 | 1053 | page = alloc_page(GFP_KERNEL | __GFP_ZERO | GFP_DMA); |
a3508fbe DH |
1054 | if (!page) { |
1055 | mutex_unlock(&kvm->arch.vsie.mutex); | |
1056 | return ERR_PTR(-ENOMEM); | |
1057 | } | |
1058 | page_ref_inc(page); | |
1059 | kvm->arch.vsie.pages[kvm->arch.vsie.page_count] = page; | |
1060 | kvm->arch.vsie.page_count++; | |
1061 | } else { | |
1062 | /* reuse an existing entry that belongs to nobody */ | |
1063 | while (true) { | |
1064 | page = kvm->arch.vsie.pages[kvm->arch.vsie.next]; | |
1065 | if (page_ref_inc_return(page) == 2) | |
1066 | break; | |
1067 | page_ref_dec(page); | |
1068 | kvm->arch.vsie.next++; | |
1069 | kvm->arch.vsie.next %= nr_vcpus; | |
1070 | } | |
1071 | radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9); | |
1072 | } | |
1073 | page->index = addr; | |
1074 | /* double use of the same address */ | |
1075 | if (radix_tree_insert(&kvm->arch.vsie.addr_to_page, addr >> 9, page)) { | |
1076 | page_ref_dec(page); | |
1077 | mutex_unlock(&kvm->arch.vsie.mutex); | |
1078 | return NULL; | |
1079 | } | |
1080 | mutex_unlock(&kvm->arch.vsie.mutex); | |
1081 | ||
1082 | vsie_page = page_to_virt(page); | |
1083 | memset(&vsie_page->scb_s, 0, sizeof(struct kvm_s390_sie_block)); | |
06d68a6c | 1084 | release_gmap_shadow(vsie_page); |
1b7029be | 1085 | vsie_page->fault_addr = 0; |
a3508fbe DH |
1086 | vsie_page->scb_s.ihcpu = 0xffffU; |
1087 | return vsie_page; | |
1088 | } | |
1089 | ||
1090 | /* put a vsie page acquired via get_vsie_page */ | |
1091 | static void put_vsie_page(struct kvm *kvm, struct vsie_page *vsie_page) | |
1092 | { | |
1093 | struct page *page = pfn_to_page(__pa(vsie_page) >> PAGE_SHIFT); | |
1094 | ||
1095 | page_ref_dec(page); | |
1096 | } | |
1097 | ||
1098 | int kvm_s390_handle_vsie(struct kvm_vcpu *vcpu) | |
1099 | { | |
1100 | struct vsie_page *vsie_page; | |
1101 | unsigned long scb_addr; | |
1102 | int rc; | |
1103 | ||
1104 | vcpu->stat.instruction_sie++; | |
1105 | if (!test_kvm_cpu_feat(vcpu->kvm, KVM_S390_VM_CPU_FEAT_SIEF2)) | |
1106 | return -EOPNOTSUPP; | |
1107 | if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE) | |
1108 | return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP); | |
1109 | ||
58cdf5eb | 1110 | BUILD_BUG_ON(sizeof(struct vsie_page) != PAGE_SIZE); |
a3508fbe DH |
1111 | scb_addr = kvm_s390_get_base_disp_s(vcpu, NULL); |
1112 | ||
1113 | /* 512 byte alignment */ | |
1114 | if (unlikely(scb_addr & 0x1ffUL)) | |
1115 | return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION); | |
1116 | ||
1117 | if (signal_pending(current) || kvm_s390_vcpu_has_irq(vcpu, 0)) | |
1118 | return 0; | |
1119 | ||
1120 | vsie_page = get_vsie_page(vcpu->kvm, scb_addr); | |
1121 | if (IS_ERR(vsie_page)) | |
1122 | return PTR_ERR(vsie_page); | |
1123 | else if (!vsie_page) | |
1124 | /* double use of sie control block - simply do nothing */ | |
1125 | return 0; | |
1126 | ||
1127 | rc = pin_scb(vcpu, vsie_page, scb_addr); | |
1128 | if (rc) | |
1129 | goto out_put; | |
1130 | rc = shadow_scb(vcpu, vsie_page); | |
1131 | if (rc) | |
1132 | goto out_unpin_scb; | |
1133 | rc = pin_blocks(vcpu, vsie_page); | |
1134 | if (rc) | |
1135 | goto out_unshadow; | |
adbf1698 | 1136 | register_shadow_scb(vcpu, vsie_page); |
a3508fbe | 1137 | rc = vsie_run(vcpu, vsie_page); |
adbf1698 | 1138 | unregister_shadow_scb(vcpu); |
a3508fbe DH |
1139 | unpin_blocks(vcpu, vsie_page); |
1140 | out_unshadow: | |
1141 | unshadow_scb(vcpu, vsie_page); | |
1142 | out_unpin_scb: | |
1143 | unpin_scb(vcpu, vsie_page, scb_addr); | |
1144 | out_put: | |
1145 | put_vsie_page(vcpu->kvm, vsie_page); | |
1146 | ||
1147 | return rc < 0 ? rc : 0; | |
1148 | } | |
1149 | ||
1150 | /* Init the vsie data structures. To be called when a vm is initialized. */ | |
1151 | void kvm_s390_vsie_init(struct kvm *kvm) | |
1152 | { | |
1153 | mutex_init(&kvm->arch.vsie.mutex); | |
1154 | INIT_RADIX_TREE(&kvm->arch.vsie.addr_to_page, GFP_KERNEL); | |
1155 | } | |
1156 | ||
1157 | /* Destroy the vsie data structures. To be called when a vm is destroyed. */ | |
1158 | void kvm_s390_vsie_destroy(struct kvm *kvm) | |
1159 | { | |
06d68a6c | 1160 | struct vsie_page *vsie_page; |
a3508fbe DH |
1161 | struct page *page; |
1162 | int i; | |
1163 | ||
1164 | mutex_lock(&kvm->arch.vsie.mutex); | |
1165 | for (i = 0; i < kvm->arch.vsie.page_count; i++) { | |
1166 | page = kvm->arch.vsie.pages[i]; | |
1167 | kvm->arch.vsie.pages[i] = NULL; | |
06d68a6c DH |
1168 | vsie_page = page_to_virt(page); |
1169 | release_gmap_shadow(vsie_page); | |
a3508fbe DH |
1170 | /* free the radix tree entry */ |
1171 | radix_tree_delete(&kvm->arch.vsie.addr_to_page, page->index >> 9); | |
1172 | __free_page(page); | |
1173 | } | |
1174 | kvm->arch.vsie.page_count = 0; | |
1175 | mutex_unlock(&kvm->arch.vsie.mutex); | |
1176 | } | |
adbf1698 DH |
1177 | |
1178 | void kvm_s390_vsie_kick(struct kvm_vcpu *vcpu) | |
1179 | { | |
1180 | struct kvm_s390_sie_block *scb = READ_ONCE(vcpu->arch.vsie_block); | |
1181 | ||
1182 | /* | |
1183 | * Even if the VCPU lets go of the shadow sie block reference, it is | |
1184 | * still valid in the cache. So we can safely kick it. | |
1185 | */ | |
1186 | if (scb) { | |
1187 | atomic_or(PROG_BLOCK_SIE, &scb->prog20); | |
1188 | if (scb->prog0c & PROG_IN_SIE) | |
1189 | atomic_or(CPUSTAT_STOP_INT, &scb->cpuflags); | |
1190 | } | |
1191 | } |