1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * handling privileged instructions
4 *
5 * Copyright IBM Corp. 2008, 2020
6 *
7 * Author(s): Carsten Otte <cotte@de.ibm.com>
8 * Christian Borntraeger <borntraeger@de.ibm.com>
9 */
10
11 #include <linux/kvm.h>
12 #include <linux/gfp.h>
13 #include <linux/errno.h>
14 #include <linux/mm_types.h>
15 #include <linux/pgtable.h>
16 #include <linux/io.h>
17 #include <asm/asm-offsets.h>
18 #include <asm/facility.h>
19 #include <asm/current.h>
20 #include <asm/debug.h>
21 #include <asm/ebcdic.h>
22 #include <asm/sysinfo.h>
23 #include <asm/page-states.h>
24 #include <asm/ptrace.h>
25 #include <asm/sclp.h>
26 #include <asm/ap.h>
27 #include <asm/gmap_helpers.h>
28 #include "gaccess.h"
29 #include "kvm-s390.h"
30 #include "trace.h"
31 #include "gmap.h"
32
handle_ri(struct kvm_vcpu * vcpu)33 static int handle_ri(struct kvm_vcpu *vcpu)
34 {
35 vcpu->stat.instruction_ri++;
36
37 if (test_kvm_facility(vcpu->kvm, 64)) {
38 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: RI (lazy)");
39 vcpu->arch.sie_block->ecb3 |= ECB3_RI;
40 kvm_s390_retry_instr(vcpu);
41 return 0;
42 } else
43 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
44 }
45
kvm_s390_handle_aa(struct kvm_vcpu * vcpu)46 int kvm_s390_handle_aa(struct kvm_vcpu *vcpu)
47 {
48 if ((vcpu->arch.sie_block->ipa & 0xf) <= 4)
49 return handle_ri(vcpu);
50 else
51 return -EOPNOTSUPP;
52 }
53
handle_gs(struct kvm_vcpu * vcpu)54 static int handle_gs(struct kvm_vcpu *vcpu)
55 {
56 vcpu->stat.instruction_gs++;
57
58 if (test_kvm_facility(vcpu->kvm, 133)) {
59 VCPU_EVENT(vcpu, 3, "%s", "ENABLE: GS (lazy)");
60 preempt_disable();
61 local_ctl_set_bit(2, CR2_GUARDED_STORAGE_BIT);
62 current->thread.gs_cb = (struct gs_cb *)&vcpu->run->s.regs.gscb;
63 restore_gs_cb(current->thread.gs_cb);
64 preempt_enable();
65 vcpu->arch.sie_block->ecb |= ECB_GS;
66 vcpu->arch.sie_block->ecd |= ECD_HOSTREGMGMT;
67 vcpu->arch.gs_enabled = 1;
68 kvm_s390_retry_instr(vcpu);
69 return 0;
70 } else
71 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
72 }
73
kvm_s390_handle_e3(struct kvm_vcpu * vcpu)74 int kvm_s390_handle_e3(struct kvm_vcpu *vcpu)
75 {
76 int code = vcpu->arch.sie_block->ipb & 0xff;
77
78 if (code == 0x49 || code == 0x4d)
79 return handle_gs(vcpu);
80 else
81 return -EOPNOTSUPP;
82 }
83 /* Handle SCK (SET CLOCK) interception */
handle_set_clock(struct kvm_vcpu * vcpu)84 static int handle_set_clock(struct kvm_vcpu *vcpu)
85 {
86 struct kvm_s390_vm_tod_clock gtod = { 0 };
87 int rc;
88 u8 ar;
89 u64 op2;
90
91 vcpu->stat.instruction_sck++;
92
93 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
94 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
95
96 op2 = kvm_s390_get_base_disp_s(vcpu, &ar);
97 if (op2 & 7) /* Operand must be on a doubleword boundary */
98 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
99 rc = read_guest(vcpu, op2, ar, >od.tod, sizeof(gtod.tod));
100 if (rc)
101 return kvm_s390_inject_prog_cond(vcpu, rc);
102
103 VCPU_EVENT(vcpu, 3, "SCK: setting guest TOD to 0x%llx", gtod.tod);
104 /*
105 * To set the TOD clock the kvm lock must be taken, but the vcpu lock
106 * is already held in handle_set_clock. The usual lock order is the
107 * opposite. As SCK is deprecated and should not be used in several
108 * cases, for example when the multiple epoch facility or TOD clock
109 * steering facility is installed (see Principles of Operation), a
110 * slow path can be used. If the lock can not be taken via try_lock,
111 * the instruction will be retried via -EAGAIN at a later point in
112 * time.
113 */
114 if (!kvm_s390_try_set_tod_clock(vcpu->kvm, >od)) {
115 kvm_s390_retry_instr(vcpu);
116 return -EAGAIN;
117 }
118
119 kvm_s390_set_psw_cc(vcpu, 0);
120 return 0;
121 }
122
handle_set_prefix(struct kvm_vcpu * vcpu)123 static int handle_set_prefix(struct kvm_vcpu *vcpu)
124 {
125 u64 operand2;
126 u32 address;
127 int rc;
128 u8 ar;
129
130 vcpu->stat.instruction_spx++;
131
132 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
133 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
134
135 operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
136
137 /* must be word boundary */
138 if (operand2 & 3)
139 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
140
141 /* get the value */
142 rc = read_guest(vcpu, operand2, ar, &address, sizeof(address));
143 if (rc)
144 return kvm_s390_inject_prog_cond(vcpu, rc);
145
146 address &= 0x7fffe000u;
147
148 /*
149 * Make sure the new value is valid memory. We only need to check the
150 * first page, since address is 8k aligned and memory pieces are always
151 * at least 1MB aligned and have at least a size of 1MB.
152 */
153 if (!kvm_is_gpa_in_memslot(vcpu->kvm, address))
154 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
155
156 kvm_s390_set_prefix(vcpu, address);
157 trace_kvm_s390_handle_prefix(vcpu, 1, address);
158 return 0;
159 }
160
handle_store_prefix(struct kvm_vcpu * vcpu)161 static int handle_store_prefix(struct kvm_vcpu *vcpu)
162 {
163 u64 operand2;
164 u32 address;
165 int rc;
166 u8 ar;
167
168 vcpu->stat.instruction_stpx++;
169
170 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
171 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
172
173 operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
174
175 /* must be word boundary */
176 if (operand2 & 3)
177 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
178
179 address = kvm_s390_get_prefix(vcpu);
180
181 /* get the value */
182 rc = write_guest(vcpu, operand2, ar, &address, sizeof(address));
183 if (rc)
184 return kvm_s390_inject_prog_cond(vcpu, rc);
185
186 VCPU_EVENT(vcpu, 3, "STPX: storing prefix 0x%x into 0x%llx", address, operand2);
187 trace_kvm_s390_handle_prefix(vcpu, 0, address);
188 return 0;
189 }
190
handle_store_cpu_address(struct kvm_vcpu * vcpu)191 static int handle_store_cpu_address(struct kvm_vcpu *vcpu)
192 {
193 u16 vcpu_id = vcpu->vcpu_id;
194 u64 ga;
195 int rc;
196 u8 ar;
197
198 vcpu->stat.instruction_stap++;
199
200 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
201 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
202
203 ga = kvm_s390_get_base_disp_s(vcpu, &ar);
204
205 if (ga & 1)
206 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
207
208 rc = write_guest(vcpu, ga, ar, &vcpu_id, sizeof(vcpu_id));
209 if (rc)
210 return kvm_s390_inject_prog_cond(vcpu, rc);
211
212 VCPU_EVENT(vcpu, 3, "STAP: storing cpu address (%u) to 0x%llx", vcpu_id, ga);
213 trace_kvm_s390_handle_stap(vcpu, ga);
214 return 0;
215 }
216
kvm_s390_skey_check_enable(struct kvm_vcpu * vcpu)217 int kvm_s390_skey_check_enable(struct kvm_vcpu *vcpu)
218 {
219 int rc;
220
221 trace_kvm_s390_skey_related_inst(vcpu);
222 /* Already enabled? */
223 if (vcpu->arch.skey_enabled)
224 return 0;
225
226 rc = gmap_enable_skeys(vcpu->arch.gmap);
227 VCPU_EVENT(vcpu, 3, "enabling storage keys for guest: %d", rc);
228 if (rc)
229 return rc;
230
231 if (kvm_s390_test_cpuflags(vcpu, CPUSTAT_KSS))
232 kvm_s390_clear_cpuflags(vcpu, CPUSTAT_KSS);
233 if (!vcpu->kvm->arch.use_skf)
234 vcpu->arch.sie_block->ictl |= ICTL_ISKE | ICTL_SSKE | ICTL_RRBE;
235 else
236 vcpu->arch.sie_block->ictl &= ~(ICTL_ISKE | ICTL_SSKE | ICTL_RRBE);
237 vcpu->arch.skey_enabled = true;
238 return 0;
239 }
240
try_handle_skey(struct kvm_vcpu * vcpu)241 static int try_handle_skey(struct kvm_vcpu *vcpu)
242 {
243 int rc;
244
245 rc = kvm_s390_skey_check_enable(vcpu);
246 if (rc)
247 return rc;
248 if (vcpu->kvm->arch.use_skf) {
249 /* with storage-key facility, SIE interprets it for us */
250 kvm_s390_retry_instr(vcpu);
251 VCPU_EVENT(vcpu, 4, "%s", "retrying storage key operation");
252 return -EAGAIN;
253 }
254 return 0;
255 }
256
handle_iske(struct kvm_vcpu * vcpu)257 static int handle_iske(struct kvm_vcpu *vcpu)
258 {
259 unsigned long gaddr;
260 int reg1, reg2;
261 union skey key;
262 int rc;
263
264 vcpu->stat.instruction_iske++;
265
266 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
267 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
268
269 rc = try_handle_skey(vcpu);
270 if (rc)
271 return rc != -EAGAIN ? rc : 0;
272
273 kvm_s390_get_regs_rre(vcpu, ®1, ®2);
274
275 gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
276 gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
277 gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
278 scoped_guard(read_lock, &vcpu->kvm->mmu_lock)
279 rc = dat_get_storage_key(vcpu->arch.gmap->asce, gpa_to_gfn(gaddr), &key);
280 if (rc > 0)
281 return kvm_s390_inject_program_int(vcpu, rc);
282 if (rc < 0)
283 return rc;
284 vcpu->run->s.regs.gprs[reg1] &= ~0xff;
285 vcpu->run->s.regs.gprs[reg1] |= key.skey;
286 return 0;
287 }
288
handle_rrbe(struct kvm_vcpu * vcpu)289 static int handle_rrbe(struct kvm_vcpu *vcpu)
290 {
291 unsigned long gaddr;
292 int reg1, reg2;
293 int rc;
294
295 vcpu->stat.instruction_rrbe++;
296
297 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
298 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
299
300 rc = try_handle_skey(vcpu);
301 if (rc)
302 return rc != -EAGAIN ? rc : 0;
303
304 kvm_s390_get_regs_rre(vcpu, ®1, ®2);
305
306 gaddr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
307 gaddr = kvm_s390_logical_to_effective(vcpu, gaddr);
308 gaddr = kvm_s390_real_to_abs(vcpu, gaddr);
309 scoped_guard(read_lock, &vcpu->kvm->mmu_lock)
310 rc = dat_reset_reference_bit(vcpu->arch.gmap->asce, gpa_to_gfn(gaddr));
311 if (rc > 0)
312 return kvm_s390_inject_program_int(vcpu, rc);
313 if (rc < 0)
314 return rc;
315 kvm_s390_set_psw_cc(vcpu, rc);
316 return 0;
317 }
318
319 #define SSKE_NQ 0x8
320 #define SSKE_MR 0x4
321 #define SSKE_MC 0x2
322 #define SSKE_MB 0x1
handle_sske(struct kvm_vcpu * vcpu)323 static int handle_sske(struct kvm_vcpu *vcpu)
324 {
325 unsigned char m3 = vcpu->arch.sie_block->ipb >> 28;
326 unsigned long start, end;
327 union skey key, oldkey;
328 int reg1, reg2;
329 int rc;
330
331 vcpu->stat.instruction_sske++;
332
333 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
334 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
335
336 rc = try_handle_skey(vcpu);
337 if (rc)
338 return rc != -EAGAIN ? rc : 0;
339
340 if (!test_kvm_facility(vcpu->kvm, 8))
341 m3 &= ~SSKE_MB;
342 if (!test_kvm_facility(vcpu->kvm, 10))
343 m3 &= ~(SSKE_MC | SSKE_MR);
344 if (!test_kvm_facility(vcpu->kvm, 14))
345 m3 &= ~SSKE_NQ;
346
347 kvm_s390_get_regs_rre(vcpu, ®1, ®2);
348
349 key.skey = vcpu->run->s.regs.gprs[reg1] & 0xfe;
350 start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
351 start = kvm_s390_logical_to_effective(vcpu, start);
352 if (m3 & SSKE_MB) {
353 /* start already designates an absolute address */
354 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
355 } else {
356 start = kvm_s390_real_to_abs(vcpu, start);
357 end = start + PAGE_SIZE;
358 }
359
360 while (start != end) {
361 scoped_guard(read_lock, &vcpu->kvm->mmu_lock) {
362 rc = dat_cond_set_storage_key(vcpu->arch.mc, vcpu->arch.gmap->asce,
363 gpa_to_gfn(start), key, &oldkey,
364 m3 & SSKE_NQ, m3 & SSKE_MR, m3 & SSKE_MC);
365 }
366 if (rc > 1)
367 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
368 if (rc == -ENOMEM) {
369 rc = kvm_s390_mmu_cache_topup(vcpu->arch.mc);
370 if (rc)
371 return rc;
372 continue;
373 }
374 if (rc < 0)
375 return rc;
376 start += PAGE_SIZE;
377 }
378
379 if (m3 & (SSKE_MC | SSKE_MR)) {
380 if (m3 & SSKE_MB) {
381 /* skey in reg1 is unpredictable */
382 kvm_s390_set_psw_cc(vcpu, 3);
383 } else {
384 kvm_s390_set_psw_cc(vcpu, rc);
385 vcpu->run->s.regs.gprs[reg1] &= ~0xff00UL;
386 vcpu->run->s.regs.gprs[reg1] |= (u64)oldkey.skey << 8;
387 }
388 }
389 if (m3 & SSKE_MB) {
390 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT)
391 vcpu->run->s.regs.gprs[reg2] &= ~PAGE_MASK;
392 else
393 vcpu->run->s.regs.gprs[reg2] &= ~0xfffff000UL;
394 end = kvm_s390_logical_to_effective(vcpu, end);
395 vcpu->run->s.regs.gprs[reg2] |= end;
396 }
397 return 0;
398 }
399
handle_ipte_interlock(struct kvm_vcpu * vcpu)400 static int handle_ipte_interlock(struct kvm_vcpu *vcpu)
401 {
402 vcpu->stat.instruction_ipte_interlock++;
403 if (psw_bits(vcpu->arch.sie_block->gpsw).pstate)
404 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
405 wait_event(vcpu->kvm->arch.ipte_wq, !ipte_lock_held(vcpu->kvm));
406 kvm_s390_retry_instr(vcpu);
407 VCPU_EVENT(vcpu, 4, "%s", "retrying ipte interlock operation");
408 return 0;
409 }
410
handle_test_block(struct kvm_vcpu * vcpu)411 static int handle_test_block(struct kvm_vcpu *vcpu)
412 {
413 gpa_t addr;
414 int reg2;
415
416 vcpu->stat.instruction_tb++;
417
418 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
419 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
420
421 kvm_s390_get_regs_rre(vcpu, NULL, ®2);
422 addr = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
423 addr = kvm_s390_logical_to_effective(vcpu, addr);
424 if (kvm_s390_check_low_addr_prot_real(vcpu, addr))
425 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
426 addr = kvm_s390_real_to_abs(vcpu, addr);
427
428 if (!kvm_is_gpa_in_memslot(vcpu->kvm, addr))
429 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
430 /*
431 * We don't expect errors on modern systems, and do not care
432 * about storage keys (yet), so let's just clear the page.
433 */
434 if (kvm_clear_guest(vcpu->kvm, addr, PAGE_SIZE))
435 return -EFAULT;
436 kvm_s390_set_psw_cc(vcpu, 0);
437 vcpu->run->s.regs.gprs[0] = 0;
438 return 0;
439 }
440
handle_tpi(struct kvm_vcpu * vcpu)441 static int handle_tpi(struct kvm_vcpu *vcpu)
442 {
443 struct kvm_s390_interrupt_info *inti;
444 unsigned long len;
445 u32 tpi_data[3];
446 int rc;
447 u64 addr;
448 u8 ar;
449
450 vcpu->stat.instruction_tpi++;
451
452 addr = kvm_s390_get_base_disp_s(vcpu, &ar);
453 if (addr & 3)
454 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
455
456 inti = kvm_s390_get_io_int(vcpu->kvm, vcpu->arch.sie_block->gcr[6], 0);
457 if (!inti) {
458 kvm_s390_set_psw_cc(vcpu, 0);
459 return 0;
460 }
461
462 tpi_data[0] = inti->io.subchannel_id << 16 | inti->io.subchannel_nr;
463 tpi_data[1] = inti->io.io_int_parm;
464 tpi_data[2] = inti->io.io_int_word;
465 if (addr) {
466 /*
467 * Store the two-word I/O interruption code into the
468 * provided area.
469 */
470 len = sizeof(tpi_data) - 4;
471 rc = write_guest(vcpu, addr, ar, &tpi_data, len);
472 if (rc) {
473 rc = kvm_s390_inject_prog_cond(vcpu, rc);
474 goto reinject_interrupt;
475 }
476 } else {
477 /*
478 * Store the three-word I/O interruption code into
479 * the appropriate lowcore area.
480 */
481 len = sizeof(tpi_data);
482 if (write_guest_lc(vcpu, __LC_SUBCHANNEL_ID, &tpi_data, len)) {
483 /* failed writes to the low core are not recoverable */
484 rc = -EFAULT;
485 goto reinject_interrupt;
486 }
487 }
488
489 /* irq was successfully handed to the guest */
490 kfree(inti);
491 kvm_s390_set_psw_cc(vcpu, 1);
492 return 0;
493 reinject_interrupt:
494 /*
495 * If we encounter a problem storing the interruption code, the
496 * instruction is suppressed from the guest's view: reinject the
497 * interrupt.
498 */
499 if (kvm_s390_reinject_io_int(vcpu->kvm, inti)) {
500 kfree(inti);
501 rc = -EFAULT;
502 }
503 /* don't set the cc, a pgm irq was injected or we drop to user space */
504 return rc ? -EFAULT : 0;
505 }
506
handle_tsch(struct kvm_vcpu * vcpu)507 static int handle_tsch(struct kvm_vcpu *vcpu)
508 {
509 struct kvm_s390_interrupt_info *inti = NULL;
510 const u64 isc_mask = 0xffUL << 24; /* all iscs set */
511
512 vcpu->stat.instruction_tsch++;
513
514 /* a valid schid has at least one bit set */
515 if (vcpu->run->s.regs.gprs[1])
516 inti = kvm_s390_get_io_int(vcpu->kvm, isc_mask,
517 vcpu->run->s.regs.gprs[1]);
518
519 /*
520 * Prepare exit to userspace.
521 * We indicate whether we dequeued a pending I/O interrupt
522 * so that userspace can re-inject it if the instruction gets
523 * a program check. While this may re-order the pending I/O
524 * interrupts, this is no problem since the priority is kept
525 * intact.
526 */
527 vcpu->run->exit_reason = KVM_EXIT_S390_TSCH;
528 vcpu->run->s390_tsch.dequeued = !!inti;
529 if (inti) {
530 vcpu->run->s390_tsch.subchannel_id = inti->io.subchannel_id;
531 vcpu->run->s390_tsch.subchannel_nr = inti->io.subchannel_nr;
532 vcpu->run->s390_tsch.io_int_parm = inti->io.io_int_parm;
533 vcpu->run->s390_tsch.io_int_word = inti->io.io_int_word;
534 }
535 vcpu->run->s390_tsch.ipb = vcpu->arch.sie_block->ipb;
536 kfree(inti);
537 return -EREMOTE;
538 }
539
handle_io_inst(struct kvm_vcpu * vcpu)540 static int handle_io_inst(struct kvm_vcpu *vcpu)
541 {
542 VCPU_EVENT(vcpu, 4, "%s", "I/O instruction");
543
544 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
545 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
546
547 if (vcpu->kvm->arch.css_support) {
548 /*
549 * Most I/O instructions will be handled by userspace.
550 * Exceptions are tpi and the interrupt portion of tsch.
551 */
552 if (vcpu->arch.sie_block->ipa == 0xb236)
553 return handle_tpi(vcpu);
554 if (vcpu->arch.sie_block->ipa == 0xb235)
555 return handle_tsch(vcpu);
556 /* Handle in userspace. */
557 vcpu->stat.instruction_io_other++;
558 return -EOPNOTSUPP;
559 } else {
560 /*
561 * Set condition code 3 to stop the guest from issuing channel
562 * I/O instructions.
563 */
564 kvm_s390_set_psw_cc(vcpu, 3);
565 return 0;
566 }
567 }
568
569 #if IS_ENABLED(CONFIG_VFIO_AP)
kvm_s390_is_gpa_in_memslot(struct kvm * kvm,gpa_t gpa)570 bool kvm_s390_is_gpa_in_memslot(struct kvm *kvm, gpa_t gpa)
571 {
572 return kvm_is_gpa_in_memslot(kvm, gpa);
573 }
574 EXPORT_SYMBOL_FOR_MODULES(kvm_s390_is_gpa_in_memslot, "vfio_ap");
575 #endif
576
577 /*
578 * handle_pqap: Handling pqap interception
579 * @vcpu: the vcpu having issue the pqap instruction
580 *
581 * We now support PQAP/AQIC instructions and we need to correctly
582 * answer the guest even if no dedicated driver's hook is available.
583 *
584 * The intercepting code calls a dedicated callback for this instruction
585 * if a driver did register one in the CRYPTO satellite of the
586 * SIE block.
587 *
588 * If no callback is available, the queues are not available, return this
589 * response code to the caller and set CC to 3.
590 * Else return the response code returned by the callback.
591 */
handle_pqap(struct kvm_vcpu * vcpu)592 static int handle_pqap(struct kvm_vcpu *vcpu)
593 {
594 struct ap_queue_status status = {};
595 crypto_hook pqap_hook;
596 unsigned long reg0;
597 int ret;
598 uint8_t fc;
599
600 /* Verify that the AP instruction are available */
601 if (!ap_instructions_available())
602 return -EOPNOTSUPP;
603 /* Verify that the guest is allowed to use AP instructions */
604 if (!(vcpu->arch.sie_block->eca & ECA_APIE))
605 return -EOPNOTSUPP;
606 /*
607 * The only possibly intercepted functions when AP instructions are
608 * available for the guest are AQIC and TAPQ with the t bit set
609 * since we do not set IC.3 (FIII) we currently will only intercept
610 * the AQIC function code.
611 * Note: running nested under z/VM can result in intercepts for other
612 * function codes, e.g. PQAP(QCI). We do not support this and bail out.
613 */
614 reg0 = vcpu->run->s.regs.gprs[0];
615 fc = (reg0 >> 24) & 0xff;
616 if (fc != 0x03)
617 return -EOPNOTSUPP;
618
619 /* PQAP instruction is allowed for guest kernel only */
620 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
621 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
622
623 /* Common PQAP instruction specification exceptions */
624 /* bits 41-47 must all be zeros */
625 if (reg0 & 0x007f0000UL)
626 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
627 /* APFT not install and T bit set */
628 if (!test_kvm_facility(vcpu->kvm, 15) && (reg0 & 0x00800000UL))
629 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
630 /* APXA not installed and APID greater 64 or APQI greater 16 */
631 if (!(vcpu->kvm->arch.crypto.crycbd & 0x02) && (reg0 & 0x0000c0f0UL))
632 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
633
634 /* AQIC function code specific exception */
635 /* facility 65 not present for AQIC function code */
636 if (!test_kvm_facility(vcpu->kvm, 65))
637 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
638
639 /*
640 * If the hook callback is registered, there will be a pointer to the
641 * hook function pointer in the kvm_s390_crypto structure. Lock the
642 * owner, retrieve the hook function pointer and call the hook.
643 */
644 down_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
645 if (vcpu->kvm->arch.crypto.pqap_hook) {
646 pqap_hook = *vcpu->kvm->arch.crypto.pqap_hook;
647 ret = pqap_hook(vcpu);
648 if (!ret) {
649 if (vcpu->run->s.regs.gprs[1] & 0x00ff0000)
650 kvm_s390_set_psw_cc(vcpu, 3);
651 else
652 kvm_s390_set_psw_cc(vcpu, 0);
653 }
654 up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
655 return ret;
656 }
657 up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
658 /*
659 * A vfio_driver must register a hook.
660 * No hook means no driver to enable the SIE CRYCB and no queues.
661 * We send this response to the guest.
662 */
663 status.response_code = 0x01;
664 memcpy(&vcpu->run->s.regs.gprs[1], &status, sizeof(status));
665 kvm_s390_set_psw_cc(vcpu, 3);
666 return 0;
667 }
668
handle_stfl(struct kvm_vcpu * vcpu)669 static int handle_stfl(struct kvm_vcpu *vcpu)
670 {
671 int rc;
672 unsigned int fac;
673
674 vcpu->stat.instruction_stfl++;
675
676 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
677 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
678
679 /*
680 * We need to shift the lower 32 facility bits (bit 0-31) from a u64
681 * into a u32 memory representation. They will remain bits 0-31.
682 */
683 fac = *vcpu->kvm->arch.model.fac_list >> 32;
684 rc = write_guest_lc(vcpu, offsetof(struct lowcore, stfl_fac_list),
685 &fac, sizeof(fac));
686 if (rc)
687 return rc;
688 VCPU_EVENT(vcpu, 3, "STFL: store facility list 0x%x", fac);
689 trace_kvm_s390_handle_stfl(vcpu, fac);
690 return 0;
691 }
692
693 #define PSW_MASK_ADDR_MODE (PSW_MASK_EA | PSW_MASK_BA)
694 #define PSW_MASK_UNASSIGNED 0xb80800fe7fffffffUL
695 #define PSW_ADDR_24 0x0000000000ffffffUL
696 #define PSW_ADDR_31 0x000000007fffffffUL
697
is_valid_psw(psw_t * psw)698 int is_valid_psw(psw_t *psw)
699 {
700 if (psw->mask & PSW_MASK_UNASSIGNED)
701 return 0;
702 if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_BA) {
703 if (psw->addr & ~PSW_ADDR_31)
704 return 0;
705 }
706 if (!(psw->mask & PSW_MASK_ADDR_MODE) && (psw->addr & ~PSW_ADDR_24))
707 return 0;
708 if ((psw->mask & PSW_MASK_ADDR_MODE) == PSW_MASK_EA)
709 return 0;
710 if (psw->addr & 1)
711 return 0;
712 return 1;
713 }
714
kvm_s390_handle_lpsw(struct kvm_vcpu * vcpu)715 int kvm_s390_handle_lpsw(struct kvm_vcpu *vcpu)
716 {
717 psw_t *gpsw = &vcpu->arch.sie_block->gpsw;
718 psw32_t new_psw;
719 u64 addr, iaddr;
720 int rc;
721 u8 ar;
722
723 vcpu->stat.instruction_lpsw++;
724
725 iaddr = gpsw->addr - kvm_s390_get_ilen(vcpu);
726 if (gpsw->mask & PSW_MASK_PSTATE)
727 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
728
729 addr = kvm_s390_get_base_disp_s(vcpu, &ar);
730 if (addr & 7)
731 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
732
733 rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
734 if (rc)
735 return kvm_s390_inject_prog_cond(vcpu, rc);
736 if (!(new_psw.mask & PSW32_MASK_BASE))
737 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
738 gpsw->mask = (new_psw.mask & ~PSW32_MASK_BASE) << 32;
739 gpsw->mask |= new_psw.addr & PSW32_ADDR_AMODE;
740 gpsw->addr = new_psw.addr & ~PSW32_ADDR_AMODE;
741 if (!is_valid_psw(gpsw))
742 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
743 vcpu->arch.sie_block->gbea = iaddr;
744 return 0;
745 }
746
handle_lpswe(struct kvm_vcpu * vcpu)747 static int handle_lpswe(struct kvm_vcpu *vcpu)
748 {
749 psw_t new_psw;
750 u64 addr, iaddr;
751 int rc;
752 u8 ar;
753
754 vcpu->stat.instruction_lpswe++;
755
756 iaddr = vcpu->arch.sie_block->gpsw.addr - kvm_s390_get_ilen(vcpu);
757 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
758 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
759
760 addr = kvm_s390_get_base_disp_s(vcpu, &ar);
761 if (addr & 7)
762 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
763 rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
764 if (rc)
765 return kvm_s390_inject_prog_cond(vcpu, rc);
766 vcpu->arch.sie_block->gpsw = new_psw;
767 if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
768 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
769 vcpu->arch.sie_block->gbea = iaddr;
770 return 0;
771 }
772
handle_lpswey(struct kvm_vcpu * vcpu)773 static int handle_lpswey(struct kvm_vcpu *vcpu)
774 {
775 psw_t new_psw;
776 u64 addr;
777 int rc;
778 u8 ar;
779
780 vcpu->stat.instruction_lpswey++;
781
782 if (!test_kvm_facility(vcpu->kvm, 193))
783 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
784
785 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
786 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
787
788 addr = kvm_s390_get_base_disp_siy(vcpu, &ar);
789 if (addr & 7)
790 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
791
792 rc = read_guest(vcpu, addr, ar, &new_psw, sizeof(new_psw));
793 if (rc)
794 return kvm_s390_inject_prog_cond(vcpu, rc);
795
796 vcpu->arch.sie_block->gpsw = new_psw;
797 if (!is_valid_psw(&vcpu->arch.sie_block->gpsw))
798 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
799
800 return 0;
801 }
802
handle_stidp(struct kvm_vcpu * vcpu)803 static int handle_stidp(struct kvm_vcpu *vcpu)
804 {
805 u64 stidp_data = vcpu->kvm->arch.model.cpuid;
806 u64 operand2;
807 int rc;
808 u8 ar;
809
810 vcpu->stat.instruction_stidp++;
811
812 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
813 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
814
815 operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
816
817 if (operand2 & 7)
818 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
819
820 rc = write_guest(vcpu, operand2, ar, &stidp_data, sizeof(stidp_data));
821 if (rc)
822 return kvm_s390_inject_prog_cond(vcpu, rc);
823
824 VCPU_EVENT(vcpu, 3, "STIDP: store cpu id 0x%llx", stidp_data);
825 return 0;
826 }
827
handle_stsi_3_2_2(struct kvm_vcpu * vcpu,struct sysinfo_3_2_2 * mem)828 static void handle_stsi_3_2_2(struct kvm_vcpu *vcpu, struct sysinfo_3_2_2 *mem)
829 {
830 int cpus = 0;
831 int n;
832
833 cpus = atomic_read(&vcpu->kvm->online_vcpus);
834
835 /* deal with other level 3 hypervisors */
836 if (stsi(mem, 3, 2, 2))
837 mem->count = 0;
838 if (mem->count < 8)
839 mem->count++;
840 for (n = mem->count - 1; n > 0 ; n--)
841 memcpy(&mem->vm[n], &mem->vm[n - 1], sizeof(mem->vm[0]));
842
843 memset(&mem->vm[0], 0, sizeof(mem->vm[0]));
844 mem->vm[0].cpus_total = cpus;
845 mem->vm[0].cpus_configured = cpus;
846 mem->vm[0].cpus_standby = 0;
847 mem->vm[0].cpus_reserved = 0;
848 mem->vm[0].caf = 1000;
849 memcpy(mem->vm[0].name, "KVMguest", 8);
850 ASCEBC(mem->vm[0].name, 8);
851 memcpy(mem->vm[0].cpi, "KVM/Linux ", 16);
852 ASCEBC(mem->vm[0].cpi, 16);
853 }
854
insert_stsi_usr_data(struct kvm_vcpu * vcpu,u64 addr,u8 ar,u8 fc,u8 sel1,u16 sel2)855 static void insert_stsi_usr_data(struct kvm_vcpu *vcpu, u64 addr, u8 ar,
856 u8 fc, u8 sel1, u16 sel2)
857 {
858 vcpu->run->exit_reason = KVM_EXIT_S390_STSI;
859 vcpu->run->s390_stsi.addr = addr;
860 vcpu->run->s390_stsi.ar = ar;
861 vcpu->run->s390_stsi.fc = fc;
862 vcpu->run->s390_stsi.sel1 = sel1;
863 vcpu->run->s390_stsi.sel2 = sel2;
864 }
865
handle_stsi(struct kvm_vcpu * vcpu)866 static int handle_stsi(struct kvm_vcpu *vcpu)
867 {
868 int fc = (vcpu->run->s.regs.gprs[0] & 0xf0000000) >> 28;
869 int sel1 = vcpu->run->s.regs.gprs[0] & 0xff;
870 int sel2 = vcpu->run->s.regs.gprs[1] & 0xffff;
871 unsigned long mem = 0;
872 u64 operand2;
873 int rc = 0;
874 u8 ar;
875
876 vcpu->stat.instruction_stsi++;
877 VCPU_EVENT(vcpu, 3, "STSI: fc: %u sel1: %u sel2: %u", fc, sel1, sel2);
878
879 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
880 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
881
882 /* Bailout forbidden function codes */
883 if (fc > 3 && fc != 15)
884 goto out_no_data;
885
886 /*
887 * fc 15 is provided only with
888 * - PTF/CPU topology support through facility 15
889 * - KVM_CAP_S390_USER_STSI
890 */
891 if (fc == 15 && (!test_kvm_facility(vcpu->kvm, 11) ||
892 !vcpu->kvm->arch.user_stsi))
893 goto out_no_data;
894
895 if (vcpu->run->s.regs.gprs[0] & 0x0fffff00
896 || vcpu->run->s.regs.gprs[1] & 0xffff0000)
897 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
898
899 if (fc == 0) {
900 vcpu->run->s.regs.gprs[0] = 3 << 28;
901 kvm_s390_set_psw_cc(vcpu, 0);
902 return 0;
903 }
904
905 operand2 = kvm_s390_get_base_disp_s(vcpu, &ar);
906
907 if (!kvm_s390_pv_cpu_is_protected(vcpu) && (operand2 & 0xfff))
908 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
909
910 switch (fc) {
911 case 1: /* same handling for 1 and 2 */
912 case 2:
913 mem = get_zeroed_page(GFP_KERNEL_ACCOUNT);
914 if (!mem)
915 goto out_no_data;
916 if (stsi((void *) mem, fc, sel1, sel2))
917 goto out_no_data;
918 break;
919 case 3:
920 if (sel1 != 2 || sel2 != 2)
921 goto out_no_data;
922 mem = get_zeroed_page(GFP_KERNEL_ACCOUNT);
923 if (!mem)
924 goto out_no_data;
925 handle_stsi_3_2_2(vcpu, (void *) mem);
926 break;
927 case 15: /* fc 15 is fully handled in userspace */
928 insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
929 trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
930 return -EREMOTE;
931 }
932 if (kvm_s390_pv_cpu_is_protected(vcpu)) {
933 memcpy(sida_addr(vcpu->arch.sie_block), (void *)mem, PAGE_SIZE);
934 rc = 0;
935 } else {
936 rc = write_guest(vcpu, operand2, ar, (void *)mem, PAGE_SIZE);
937 }
938 if (rc) {
939 rc = kvm_s390_inject_prog_cond(vcpu, rc);
940 goto out;
941 }
942 if (vcpu->kvm->arch.user_stsi) {
943 insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
944 rc = -EREMOTE;
945 }
946 trace_kvm_s390_handle_stsi(vcpu, fc, sel1, sel2, operand2);
947 free_page(mem);
948 kvm_s390_set_psw_cc(vcpu, 0);
949 vcpu->run->s.regs.gprs[0] = 0;
950 return rc;
951 out_no_data:
952 kvm_s390_set_psw_cc(vcpu, 3);
953 out:
954 free_page(mem);
955 return rc;
956 }
957
kvm_s390_handle_b2(struct kvm_vcpu * vcpu)958 int kvm_s390_handle_b2(struct kvm_vcpu *vcpu)
959 {
960 switch (vcpu->arch.sie_block->ipa & 0x00ff) {
961 case 0x02:
962 return handle_stidp(vcpu);
963 case 0x04:
964 return handle_set_clock(vcpu);
965 case 0x10:
966 return handle_set_prefix(vcpu);
967 case 0x11:
968 return handle_store_prefix(vcpu);
969 case 0x12:
970 return handle_store_cpu_address(vcpu);
971 case 0x14:
972 return kvm_s390_handle_vsie(vcpu);
973 case 0x21:
974 case 0x50:
975 return handle_ipte_interlock(vcpu);
976 case 0x29:
977 return handle_iske(vcpu);
978 case 0x2a:
979 return handle_rrbe(vcpu);
980 case 0x2b:
981 return handle_sske(vcpu);
982 case 0x2c:
983 return handle_test_block(vcpu);
984 case 0x30:
985 case 0x31:
986 case 0x32:
987 case 0x33:
988 case 0x34:
989 case 0x35:
990 case 0x36:
991 case 0x37:
992 case 0x38:
993 case 0x39:
994 case 0x3a:
995 case 0x3b:
996 case 0x3c:
997 case 0x5f:
998 case 0x74:
999 case 0x76:
1000 return handle_io_inst(vcpu);
1001 case 0x56:
1002 return handle_sthyi(vcpu);
1003 case 0x7d:
1004 return handle_stsi(vcpu);
1005 case 0xaf:
1006 return handle_pqap(vcpu);
1007 case 0xb1:
1008 return handle_stfl(vcpu);
1009 case 0xb2:
1010 return handle_lpswe(vcpu);
1011 default:
1012 return -EOPNOTSUPP;
1013 }
1014 }
1015
handle_epsw(struct kvm_vcpu * vcpu)1016 static int handle_epsw(struct kvm_vcpu *vcpu)
1017 {
1018 int reg1, reg2;
1019
1020 vcpu->stat.instruction_epsw++;
1021
1022 kvm_s390_get_regs_rre(vcpu, ®1, ®2);
1023
1024 /* This basically extracts the mask half of the psw. */
1025 vcpu->run->s.regs.gprs[reg1] &= 0xffffffff00000000UL;
1026 vcpu->run->s.regs.gprs[reg1] |= vcpu->arch.sie_block->gpsw.mask >> 32;
1027 if (reg2) {
1028 vcpu->run->s.regs.gprs[reg2] &= 0xffffffff00000000UL;
1029 vcpu->run->s.regs.gprs[reg2] |=
1030 vcpu->arch.sie_block->gpsw.mask & 0x00000000ffffffffUL;
1031 }
1032 return 0;
1033 }
1034
1035 #define PFMF_RESERVED 0xfffc0101UL
1036 #define PFMF_SK 0x00020000UL
1037 #define PFMF_CF 0x00010000UL
1038 #define PFMF_UI 0x00008000UL
1039 #define PFMF_FSC 0x00007000UL
1040 #define PFMF_NQ 0x00000800UL
1041 #define PFMF_MR 0x00000400UL
1042 #define PFMF_MC 0x00000200UL
1043 #define PFMF_KEY 0x000000feUL
1044
handle_pfmf(struct kvm_vcpu * vcpu)1045 static int handle_pfmf(struct kvm_vcpu *vcpu)
1046 {
1047 bool mr = false, mc = false, nq;
1048 int reg1, reg2;
1049 unsigned long start, end;
1050 union skey key;
1051
1052 vcpu->stat.instruction_pfmf++;
1053
1054 kvm_s390_get_regs_rre(vcpu, ®1, ®2);
1055
1056 if (!test_kvm_facility(vcpu->kvm, 8))
1057 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1058
1059 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1060 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1061
1062 if (vcpu->run->s.regs.gprs[reg1] & PFMF_RESERVED)
1063 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1064
1065 /* Only provide non-quiescing support if enabled for the guest */
1066 if (vcpu->run->s.regs.gprs[reg1] & PFMF_NQ &&
1067 !test_kvm_facility(vcpu->kvm, 14))
1068 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1069
1070 /* Only provide conditional-SSKE support if enabled for the guest */
1071 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK &&
1072 test_kvm_facility(vcpu->kvm, 10)) {
1073 mr = vcpu->run->s.regs.gprs[reg1] & PFMF_MR;
1074 mc = vcpu->run->s.regs.gprs[reg1] & PFMF_MC;
1075 }
1076
1077 nq = vcpu->run->s.regs.gprs[reg1] & PFMF_NQ;
1078 key.skey = vcpu->run->s.regs.gprs[reg1] & PFMF_KEY;
1079 start = vcpu->run->s.regs.gprs[reg2] & PAGE_MASK;
1080 start = kvm_s390_logical_to_effective(vcpu, start);
1081
1082 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1083 if (kvm_s390_check_low_addr_prot_real(vcpu, start))
1084 return kvm_s390_inject_prog_irq(vcpu, &vcpu->arch.pgm);
1085 }
1086
1087 switch (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1088 case 0x00000000:
1089 /* only 4k frames specify a real address */
1090 start = kvm_s390_real_to_abs(vcpu, start);
1091 end = (start + PAGE_SIZE) & ~(PAGE_SIZE - 1);
1092 break;
1093 case 0x00001000:
1094 end = (start + _SEGMENT_SIZE) & ~(_SEGMENT_SIZE - 1);
1095 break;
1096 case 0x00002000:
1097 /* only support 2G frame size if EDAT2 is available and we are
1098 not in 24-bit addressing mode */
1099 if (!test_kvm_facility(vcpu->kvm, 78) ||
1100 psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_24BIT)
1101 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1102 end = (start + _REGION3_SIZE) & ~(_REGION3_SIZE - 1);
1103 break;
1104 default:
1105 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1106 }
1107
1108 while (start != end) {
1109 if (vcpu->run->s.regs.gprs[reg1] & PFMF_CF) {
1110 if (kvm_clear_guest(vcpu->kvm, start, PAGE_SIZE))
1111 return kvm_s390_inject_program_int(vcpu, PGM_ADDRESSING);
1112 }
1113
1114 if (vcpu->run->s.regs.gprs[reg1] & PFMF_SK) {
1115 int rc = kvm_s390_skey_check_enable(vcpu);
1116
1117 if (rc)
1118 return rc;
1119 scoped_guard(read_lock, &vcpu->kvm->mmu_lock) {
1120 rc = dat_cond_set_storage_key(vcpu->arch.mc, vcpu->arch.gmap->asce,
1121 gpa_to_gfn(start), key,
1122 NULL, nq, mr, mc);
1123 }
1124 if (rc > 1)
1125 return kvm_s390_inject_program_int(vcpu, rc);
1126 if (rc == -ENOMEM) {
1127 rc = kvm_s390_mmu_cache_topup(vcpu->arch.mc);
1128 if (rc)
1129 return rc;
1130 continue;
1131 }
1132 if (rc < 0)
1133 return rc;
1134 }
1135 start += PAGE_SIZE;
1136 }
1137 if (vcpu->run->s.regs.gprs[reg1] & PFMF_FSC) {
1138 if (psw_bits(vcpu->arch.sie_block->gpsw).eaba == PSW_BITS_AMODE_64BIT) {
1139 vcpu->run->s.regs.gprs[reg2] = end;
1140 } else {
1141 vcpu->run->s.regs.gprs[reg2] &= ~0xffffffffUL;
1142 end = kvm_s390_logical_to_effective(vcpu, end);
1143 vcpu->run->s.regs.gprs[reg2] |= end;
1144 }
1145 }
1146 return 0;
1147 }
1148
1149 /*
1150 * Must be called with relevant read locks held (kvm->mm->mmap_lock, kvm->srcu)
1151 */
__do_essa(struct kvm_vcpu * vcpu,const int orc)1152 static inline int __do_essa(struct kvm_vcpu *vcpu, const int orc)
1153 {
1154 int r1, r2, nappended, entries;
1155 union essa_state state;
1156 unsigned long *cbrlo;
1157 unsigned long gfn;
1158 bool dirtied;
1159
1160 /*
1161 * We don't need to set SD.FPF.SK to 1 here, because if we have a
1162 * machine check here we either handle it or crash
1163 */
1164
1165 kvm_s390_get_regs_rre(vcpu, &r1, &r2);
1166 gfn = vcpu->run->s.regs.gprs[r2] >> PAGE_SHIFT;
1167 entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1168
1169 nappended = dat_perform_essa(vcpu->arch.gmap->asce, gfn, orc, &state, &dirtied);
1170 vcpu->run->s.regs.gprs[r1] = state.val;
1171 if (nappended < 0)
1172 return 0;
1173 /*
1174 * It is possible that all the normal 511 slots were full, in which case
1175 * we will now write in the 512th slot, which is reserved for host use.
1176 * In both cases we let the normal essa handling code process all the
1177 * slots, including the reserved one, if needed.
1178 */
1179 if (nappended > 0) {
1180 cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo & PAGE_MASK);
1181 cbrlo[entries] = gfn << PAGE_SHIFT;
1182 }
1183
1184 if (dirtied)
1185 atomic64_inc(&vcpu->kvm->arch.cmma_dirty_pages);
1186
1187 return nappended;
1188 }
1189
_essa_clear_cbrl(struct kvm_vcpu * vcpu,unsigned long * cbrl,int len)1190 static void _essa_clear_cbrl(struct kvm_vcpu *vcpu, unsigned long *cbrl, int len)
1191 {
1192 union crste *crstep;
1193 union pgste pgste;
1194 union pte *ptep;
1195 hva_t hva;
1196 int i;
1197
1198 lockdep_assert_held(&vcpu->kvm->mmu_lock);
1199
1200 for (i = 0; i < len; i++) {
1201 if (dat_entry_walk(NULL, gpa_to_gfn(cbrl[i]), vcpu->arch.gmap->asce,
1202 0, TABLE_TYPE_PAGE_TABLE, &crstep, &ptep))
1203 continue;
1204 if (!ptep || ptep->s.pr)
1205 continue;
1206 pgste = pgste_get_lock(ptep);
1207 if (pgste.usage == PGSTE_GPS_USAGE_UNUSED || pgste.zero) {
1208 hva = gpa_to_hva(vcpu->kvm, cbrl[i]);
1209 if (!kvm_is_error_hva(hva))
1210 gmap_helper_zap_one_page(vcpu->kvm->mm, hva);
1211 }
1212 pgste_set_unlock(ptep, pgste);
1213 }
1214 }
1215
handle_essa(struct kvm_vcpu * vcpu)1216 static int handle_essa(struct kvm_vcpu *vcpu)
1217 {
1218 lockdep_assert_held(&vcpu->kvm->srcu);
1219
1220 /* entries expected to be 1FF */
1221 int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
1222 unsigned long *cbrlo;
1223 int i, orc;
1224
1225 VCPU_EVENT(vcpu, 4, "ESSA: release %d pages", entries);
1226 vcpu->stat.instruction_essa++;
1227 if (!vcpu->kvm->arch.use_cmma)
1228 return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
1229
1230 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1231 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1232 /* Check for invalid operation request code */
1233 orc = (vcpu->arch.sie_block->ipb & 0xf0000000) >> 28;
1234 /* ORCs 0-6 are always valid */
1235 if (orc > (test_kvm_facility(vcpu->kvm, 147) ? ESSA_SET_STABLE_NODAT
1236 : ESSA_SET_STABLE_IF_RESIDENT))
1237 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1238
1239 if (!READ_ONCE(vcpu->kvm->arch.migration_mode)) {
1240 /*
1241 * CMMA is enabled in the KVM settings, but is disabled in
1242 * the SIE block and in the mm_context, and we are not doing
1243 * a migration. Enable CMMA in the mm_context.
1244 * Since we need to take a write lock to write to the context
1245 * to avoid races with storage keys handling, we check if the
1246 * value really needs to be written to; if the value is
1247 * already correct, we do nothing and avoid the lock.
1248 */
1249 set_bit(GMAP_FLAG_USES_CMM, &vcpu->arch.gmap->flags);
1250 /*
1251 * If we are here, we are supposed to have CMMA enabled in
1252 * the SIE block. Enabling CMMA works on a per-CPU basis,
1253 * while the context use_cmma flag is per process.
1254 * It's possible that the context flag is enabled and the
1255 * SIE flag is not, so we set the flag always; if it was
1256 * already set, nothing changes, otherwise we enable it
1257 * on this CPU too.
1258 */
1259 vcpu->arch.sie_block->ecb2 |= ECB2_CMMA;
1260 /* Retry the ESSA instruction */
1261 kvm_s390_retry_instr(vcpu);
1262 } else {
1263 scoped_guard(read_lock, &vcpu->kvm->mmu_lock)
1264 i = __do_essa(vcpu, orc);
1265 if (i < 0)
1266 return i;
1267 /* Account for the possible extra cbrl entry */
1268 entries += i;
1269 }
1270 /* reset nceo */
1271 vcpu->arch.sie_block->cbrlo &= PAGE_MASK;
1272 cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
1273
1274 mmap_read_lock(vcpu->kvm->mm);
1275 scoped_guard(read_lock, &vcpu->kvm->mmu_lock)
1276 _essa_clear_cbrl(vcpu, cbrlo, entries);
1277 mmap_read_unlock(vcpu->kvm->mm);
1278
1279 return 0;
1280 }
1281
kvm_s390_handle_b9(struct kvm_vcpu * vcpu)1282 int kvm_s390_handle_b9(struct kvm_vcpu *vcpu)
1283 {
1284 switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1285 case 0x8a:
1286 case 0x8e:
1287 case 0x8f:
1288 return handle_ipte_interlock(vcpu);
1289 case 0x8d:
1290 return handle_epsw(vcpu);
1291 case 0xab:
1292 return handle_essa(vcpu);
1293 case 0xaf:
1294 return handle_pfmf(vcpu);
1295 default:
1296 return -EOPNOTSUPP;
1297 }
1298 }
1299
kvm_s390_handle_lctl(struct kvm_vcpu * vcpu)1300 int kvm_s390_handle_lctl(struct kvm_vcpu *vcpu)
1301 {
1302 int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1303 int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1304 int reg, rc, nr_regs;
1305 u32 ctl_array[16];
1306 u64 ga;
1307 u8 ar;
1308
1309 vcpu->stat.instruction_lctl++;
1310
1311 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1312 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1313
1314 ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1315
1316 if (ga & 3)
1317 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1318
1319 VCPU_EVENT(vcpu, 4, "LCTL: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1320 trace_kvm_s390_handle_lctl(vcpu, 0, reg1, reg3, ga);
1321
1322 nr_regs = ((reg3 - reg1) & 0xf) + 1;
1323 rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1324 if (rc)
1325 return kvm_s390_inject_prog_cond(vcpu, rc);
1326 reg = reg1;
1327 nr_regs = 0;
1328 do {
1329 vcpu->arch.sie_block->gcr[reg] &= 0xffffffff00000000ul;
1330 vcpu->arch.sie_block->gcr[reg] |= ctl_array[nr_regs++];
1331 if (reg == reg3)
1332 break;
1333 reg = (reg + 1) % 16;
1334 } while (1);
1335 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1336 return 0;
1337 }
1338
kvm_s390_handle_stctl(struct kvm_vcpu * vcpu)1339 int kvm_s390_handle_stctl(struct kvm_vcpu *vcpu)
1340 {
1341 int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1342 int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1343 int reg, rc, nr_regs;
1344 u32 ctl_array[16];
1345 u64 ga;
1346 u8 ar;
1347
1348 vcpu->stat.instruction_stctl++;
1349
1350 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1351 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1352
1353 ga = kvm_s390_get_base_disp_rs(vcpu, &ar);
1354
1355 if (ga & 3)
1356 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1357
1358 VCPU_EVENT(vcpu, 4, "STCTL r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1359 trace_kvm_s390_handle_stctl(vcpu, 0, reg1, reg3, ga);
1360
1361 reg = reg1;
1362 nr_regs = 0;
1363 do {
1364 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1365 if (reg == reg3)
1366 break;
1367 reg = (reg + 1) % 16;
1368 } while (1);
1369 rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u32));
1370 return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1371 }
1372
handle_lctlg(struct kvm_vcpu * vcpu)1373 static int handle_lctlg(struct kvm_vcpu *vcpu)
1374 {
1375 int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1376 int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1377 int reg, rc, nr_regs;
1378 u64 ctl_array[16];
1379 u64 ga;
1380 u8 ar;
1381
1382 vcpu->stat.instruction_lctlg++;
1383
1384 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1385 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1386
1387 ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1388
1389 if (ga & 7)
1390 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1391
1392 VCPU_EVENT(vcpu, 4, "LCTLG: r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1393 trace_kvm_s390_handle_lctl(vcpu, 1, reg1, reg3, ga);
1394
1395 nr_regs = ((reg3 - reg1) & 0xf) + 1;
1396 rc = read_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1397 if (rc)
1398 return kvm_s390_inject_prog_cond(vcpu, rc);
1399 reg = reg1;
1400 nr_regs = 0;
1401 do {
1402 vcpu->arch.sie_block->gcr[reg] = ctl_array[nr_regs++];
1403 if (reg == reg3)
1404 break;
1405 reg = (reg + 1) % 16;
1406 } while (1);
1407 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1408 return 0;
1409 }
1410
handle_stctg(struct kvm_vcpu * vcpu)1411 static int handle_stctg(struct kvm_vcpu *vcpu)
1412 {
1413 int reg1 = (vcpu->arch.sie_block->ipa & 0x00f0) >> 4;
1414 int reg3 = vcpu->arch.sie_block->ipa & 0x000f;
1415 int reg, rc, nr_regs;
1416 u64 ctl_array[16];
1417 u64 ga;
1418 u8 ar;
1419
1420 vcpu->stat.instruction_stctg++;
1421
1422 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1423 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1424
1425 ga = kvm_s390_get_base_disp_rsy(vcpu, &ar);
1426
1427 if (ga & 7)
1428 return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
1429
1430 VCPU_EVENT(vcpu, 4, "STCTG r1:%d, r3:%d, addr: 0x%llx", reg1, reg3, ga);
1431 trace_kvm_s390_handle_stctl(vcpu, 1, reg1, reg3, ga);
1432
1433 reg = reg1;
1434 nr_regs = 0;
1435 do {
1436 ctl_array[nr_regs++] = vcpu->arch.sie_block->gcr[reg];
1437 if (reg == reg3)
1438 break;
1439 reg = (reg + 1) % 16;
1440 } while (1);
1441 rc = write_guest(vcpu, ga, ar, ctl_array, nr_regs * sizeof(u64));
1442 return rc ? kvm_s390_inject_prog_cond(vcpu, rc) : 0;
1443 }
1444
kvm_s390_handle_eb(struct kvm_vcpu * vcpu)1445 int kvm_s390_handle_eb(struct kvm_vcpu *vcpu)
1446 {
1447 switch (vcpu->arch.sie_block->ipb & 0x000000ff) {
1448 case 0x25:
1449 return handle_stctg(vcpu);
1450 case 0x2f:
1451 return handle_lctlg(vcpu);
1452 case 0x60:
1453 case 0x61:
1454 case 0x62:
1455 return handle_ri(vcpu);
1456 case 0x71:
1457 return handle_lpswey(vcpu);
1458 default:
1459 return -EOPNOTSUPP;
1460 }
1461 }
1462
handle_tprot(struct kvm_vcpu * vcpu)1463 static int handle_tprot(struct kvm_vcpu *vcpu)
1464 {
1465 u64 address, operand2;
1466 unsigned long gpa;
1467 u8 access_key;
1468 bool writable;
1469 int ret, cc;
1470 u8 ar;
1471
1472 vcpu->stat.instruction_tprot++;
1473
1474 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1475 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1476
1477 kvm_s390_get_base_disp_sse(vcpu, &address, &operand2, &ar, NULL);
1478 access_key = (operand2 & 0xf0) >> 4;
1479
1480 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1481 ipte_lock(vcpu->kvm);
1482
1483 ret = guest_translate_address_with_key(vcpu, address, ar, &gpa,
1484 GACC_STORE, access_key);
1485 if (ret == 0) {
1486 gfn_to_hva_prot(vcpu->kvm, gpa_to_gfn(gpa), &writable);
1487 } else if (ret == PGM_PROTECTION) {
1488 writable = false;
1489 /* Write protected? Try again with read-only... */
1490 ret = guest_translate_address_with_key(vcpu, address, ar, &gpa,
1491 GACC_FETCH, access_key);
1492 }
1493 if (ret >= 0) {
1494 cc = -1;
1495
1496 /* Fetching permitted; storing permitted */
1497 if (ret == 0 && writable)
1498 cc = 0;
1499 /* Fetching permitted; storing not permitted */
1500 else if (ret == 0 && !writable)
1501 cc = 1;
1502 /* Fetching not permitted; storing not permitted */
1503 else if (ret == PGM_PROTECTION)
1504 cc = 2;
1505 /* Translation not available */
1506 else if (ret != PGM_ADDRESSING && ret != PGM_TRANSLATION_SPEC)
1507 cc = 3;
1508
1509 if (cc != -1) {
1510 kvm_s390_set_psw_cc(vcpu, cc);
1511 ret = 0;
1512 } else {
1513 ret = kvm_s390_inject_program_int(vcpu, ret);
1514 }
1515 }
1516
1517 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_DAT)
1518 ipte_unlock(vcpu->kvm);
1519 return ret;
1520 }
1521
kvm_s390_handle_e5(struct kvm_vcpu * vcpu)1522 int kvm_s390_handle_e5(struct kvm_vcpu *vcpu)
1523 {
1524 switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1525 case 0x01:
1526 return handle_tprot(vcpu);
1527 default:
1528 return -EOPNOTSUPP;
1529 }
1530 }
1531
handle_sckpf(struct kvm_vcpu * vcpu)1532 static int handle_sckpf(struct kvm_vcpu *vcpu)
1533 {
1534 u32 value;
1535
1536 vcpu->stat.instruction_sckpf++;
1537
1538 if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
1539 return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
1540
1541 if (vcpu->run->s.regs.gprs[0] & 0x00000000ffff0000)
1542 return kvm_s390_inject_program_int(vcpu,
1543 PGM_SPECIFICATION);
1544
1545 value = vcpu->run->s.regs.gprs[0] & 0x000000000000ffff;
1546 vcpu->arch.sie_block->todpr = value;
1547
1548 return 0;
1549 }
1550
handle_ptff(struct kvm_vcpu * vcpu)1551 static int handle_ptff(struct kvm_vcpu *vcpu)
1552 {
1553 vcpu->stat.instruction_ptff++;
1554
1555 /* we don't emulate any control instructions yet */
1556 kvm_s390_set_psw_cc(vcpu, 3);
1557 return 0;
1558 }
1559
kvm_s390_handle_01(struct kvm_vcpu * vcpu)1560 int kvm_s390_handle_01(struct kvm_vcpu *vcpu)
1561 {
1562 switch (vcpu->arch.sie_block->ipa & 0x00ff) {
1563 case 0x04:
1564 return handle_ptff(vcpu);
1565 case 0x07:
1566 return handle_sckpf(vcpu);
1567 default:
1568 return -EOPNOTSUPP;
1569 }
1570 }
1571