s390x/kvm: legacy_s390_alloc() only supports one allocation
[qemu.git] / target / s390x / kvm.c
1 /*
2 * QEMU S390x KVM implementation
3 *
4 * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
5 * Copyright IBM Corp. 2012
6 *
7 * This library is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2 of the License, or (at your option) any later version.
11 *
12 * This library is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * Contributions after 2012-10-29 are licensed under the terms of the
18 * GNU GPL, version 2 or (at your option) any later version.
19 *
20 * You should have received a copy of the GNU (Lesser) General Public
21 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
22 */
23
24 #include "qemu/osdep.h"
25 #include <sys/ioctl.h>
26
27 #include <linux/kvm.h>
28 #include <asm/ptrace.h>
29
30 #include "qemu-common.h"
31 #include "cpu.h"
32 #include "internal.h"
33 #include "kvm_s390x.h"
34 #include "qapi/error.h"
35 #include "qemu/error-report.h"
36 #include "qemu/timer.h"
37 #include "sysemu/sysemu.h"
38 #include "sysemu/hw_accel.h"
39 #include "hw/hw.h"
40 #include "sysemu/device_tree.h"
41 #include "exec/gdbstub.h"
42 #include "trace.h"
43 #include "hw/s390x/s390-pci-inst.h"
44 #include "hw/s390x/s390-pci-bus.h"
45 #include "hw/s390x/ipl.h"
46 #include "hw/s390x/ebcdic.h"
47 #include "exec/memattrs.h"
48 #include "hw/s390x/s390-virtio-ccw.h"
49 #include "hw/s390x/s390-virtio-hcall.h"
50
51 #ifndef DEBUG_KVM
52 #define DEBUG_KVM 0
53 #endif
54
55 #define DPRINTF(fmt, ...) do { \
56 if (DEBUG_KVM) { \
57 fprintf(stderr, fmt, ## __VA_ARGS__); \
58 } \
59 } while (0)
60
61 #define kvm_vm_check_mem_attr(s, attr) \
62 kvm_vm_check_attr(s, KVM_S390_VM_MEM_CTRL, attr)
63
64 #define IPA0_DIAG 0x8300
65 #define IPA0_SIGP 0xae00
66 #define IPA0_B2 0xb200
67 #define IPA0_B9 0xb900
68 #define IPA0_EB 0xeb00
69 #define IPA0_E3 0xe300
70
71 #define PRIV_B2_SCLP_CALL 0x20
72 #define PRIV_B2_CSCH 0x30
73 #define PRIV_B2_HSCH 0x31
74 #define PRIV_B2_MSCH 0x32
75 #define PRIV_B2_SSCH 0x33
76 #define PRIV_B2_STSCH 0x34
77 #define PRIV_B2_TSCH 0x35
78 #define PRIV_B2_TPI 0x36
79 #define PRIV_B2_SAL 0x37
80 #define PRIV_B2_RSCH 0x38
81 #define PRIV_B2_STCRW 0x39
82 #define PRIV_B2_STCPS 0x3a
83 #define PRIV_B2_RCHP 0x3b
84 #define PRIV_B2_SCHM 0x3c
85 #define PRIV_B2_CHSC 0x5f
86 #define PRIV_B2_SIGA 0x74
87 #define PRIV_B2_XSCH 0x76
88
89 #define PRIV_EB_SQBS 0x8a
90 #define PRIV_EB_PCISTB 0xd0
91 #define PRIV_EB_SIC 0xd1
92
93 #define PRIV_B9_EQBS 0x9c
94 #define PRIV_B9_CLP 0xa0
95 #define PRIV_B9_PCISTG 0xd0
96 #define PRIV_B9_PCILG 0xd2
97 #define PRIV_B9_RPCIT 0xd3
98
99 #define PRIV_E3_MPCIFC 0xd0
100 #define PRIV_E3_STPCIFC 0xd4
101
102 #define DIAG_TIMEREVENT 0x288
103 #define DIAG_IPL 0x308
104 #define DIAG_KVM_HYPERCALL 0x500
105 #define DIAG_KVM_BREAKPOINT 0x501
106
107 #define ICPT_INSTRUCTION 0x04
108 #define ICPT_PROGRAM 0x08
109 #define ICPT_EXT_INT 0x14
110 #define ICPT_WAITPSW 0x1c
111 #define ICPT_SOFT_INTERCEPT 0x24
112 #define ICPT_CPU_STOP 0x28
113 #define ICPT_OPEREXC 0x2c
114 #define ICPT_IO 0x40
115
116 #define NR_LOCAL_IRQS 32
117 /*
118 * Needs to be big enough to contain max_cpus emergency signals
119 * and in addition NR_LOCAL_IRQS interrupts
120 */
121 #define VCPU_IRQ_BUF_SIZE (sizeof(struct kvm_s390_irq) * \
122 (max_cpus + NR_LOCAL_IRQS))
123
124 static CPUWatchpoint hw_watchpoint;
125 /*
126 * We don't use a list because this structure is also used to transmit the
127 * hardware breakpoints to the kernel.
128 */
129 static struct kvm_hw_breakpoint *hw_breakpoints;
130 static int nb_hw_breakpoints;
131
132 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
133 KVM_CAP_LAST_INFO
134 };
135
136 static int cap_sync_regs;
137 static int cap_async_pf;
138 static int cap_mem_op;
139 static int cap_s390_irq;
140 static int cap_ri;
141 static int cap_gs;
142
143 static int active_cmma;
144
145 static void *legacy_s390_alloc(size_t size, uint64_t *align, bool shared);
146
147 static int kvm_s390_query_mem_limit(uint64_t *memory_limit)
148 {
149 struct kvm_device_attr attr = {
150 .group = KVM_S390_VM_MEM_CTRL,
151 .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
152 .addr = (uint64_t) memory_limit,
153 };
154
155 return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
156 }
157
158 int kvm_s390_set_mem_limit(uint64_t new_limit, uint64_t *hw_limit)
159 {
160 int rc;
161
162 struct kvm_device_attr attr = {
163 .group = KVM_S390_VM_MEM_CTRL,
164 .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
165 .addr = (uint64_t) &new_limit,
166 };
167
168 if (!kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_LIMIT_SIZE)) {
169 return 0;
170 }
171
172 rc = kvm_s390_query_mem_limit(hw_limit);
173 if (rc) {
174 return rc;
175 } else if (*hw_limit < new_limit) {
176 return -E2BIG;
177 }
178
179 return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
180 }
181
182 int kvm_s390_cmma_active(void)
183 {
184 return active_cmma;
185 }
186
187 static bool kvm_s390_cmma_available(void)
188 {
189 static bool initialized, value;
190
191 if (!initialized) {
192 initialized = true;
193 value = kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_ENABLE_CMMA) &&
194 kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_CLR_CMMA);
195 }
196 return value;
197 }
198
199 void kvm_s390_cmma_reset(void)
200 {
201 int rc;
202 struct kvm_device_attr attr = {
203 .group = KVM_S390_VM_MEM_CTRL,
204 .attr = KVM_S390_VM_MEM_CLR_CMMA,
205 };
206
207 if (!kvm_s390_cmma_active()) {
208 return;
209 }
210
211 rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
212 trace_kvm_clear_cmma(rc);
213 }
214
215 static void kvm_s390_enable_cmma(void)
216 {
217 int rc;
218 struct kvm_device_attr attr = {
219 .group = KVM_S390_VM_MEM_CTRL,
220 .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
221 };
222
223 if (mem_path) {
224 warn_report("CMM will not be enabled because it is not "
225 "compatible with hugetlbfs.");
226 return;
227 }
228 rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
229 active_cmma = !rc;
230 trace_kvm_enable_cmma(rc);
231 }
232
233 static void kvm_s390_set_attr(uint64_t attr)
234 {
235 struct kvm_device_attr attribute = {
236 .group = KVM_S390_VM_CRYPTO,
237 .attr = attr,
238 };
239
240 int ret = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attribute);
241
242 if (ret) {
243 error_report("Failed to set crypto device attribute %lu: %s",
244 attr, strerror(-ret));
245 }
246 }
247
248 static void kvm_s390_init_aes_kw(void)
249 {
250 uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_AES_KW;
251
252 if (object_property_get_bool(OBJECT(qdev_get_machine()), "aes-key-wrap",
253 NULL)) {
254 attr = KVM_S390_VM_CRYPTO_ENABLE_AES_KW;
255 }
256
257 if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
258 kvm_s390_set_attr(attr);
259 }
260 }
261
262 static void kvm_s390_init_dea_kw(void)
263 {
264 uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_DEA_KW;
265
266 if (object_property_get_bool(OBJECT(qdev_get_machine()), "dea-key-wrap",
267 NULL)) {
268 attr = KVM_S390_VM_CRYPTO_ENABLE_DEA_KW;
269 }
270
271 if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
272 kvm_s390_set_attr(attr);
273 }
274 }
275
276 void kvm_s390_crypto_reset(void)
277 {
278 if (s390_has_feat(S390_FEAT_MSA_EXT_3)) {
279 kvm_s390_init_aes_kw();
280 kvm_s390_init_dea_kw();
281 }
282 }
283
284 int kvm_arch_init(MachineState *ms, KVMState *s)
285 {
286 MachineClass *mc = MACHINE_GET_CLASS(ms);
287
288 mc->default_cpu_type = S390_CPU_TYPE_NAME("host");
289 cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
290 cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
291 cap_mem_op = kvm_check_extension(s, KVM_CAP_S390_MEM_OP);
292 cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
293
294 if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
295 || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
296 phys_mem_set_alloc(legacy_s390_alloc);
297 }
298
299 kvm_vm_enable_cap(s, KVM_CAP_S390_USER_SIGP, 0);
300 kvm_vm_enable_cap(s, KVM_CAP_S390_VECTOR_REGISTERS, 0);
301 kvm_vm_enable_cap(s, KVM_CAP_S390_USER_STSI, 0);
302 if (ri_allowed()) {
303 if (kvm_vm_enable_cap(s, KVM_CAP_S390_RI, 0) == 0) {
304 cap_ri = 1;
305 }
306 }
307 if (cpu_model_allowed()) {
308 if (kvm_vm_enable_cap(s, KVM_CAP_S390_GS, 0) == 0) {
309 cap_gs = 1;
310 }
311 }
312
313 /*
314 * The migration interface for ais was introduced with kernel 4.13
315 * but the capability itself had been active since 4.12. As migration
316 * support is considered necessary let's disable ais in the 2.10
317 * machine.
318 */
319 /* kvm_vm_enable_cap(s, KVM_CAP_S390_AIS, 0); */
320
321 return 0;
322 }
323
324 int kvm_arch_irqchip_create(MachineState *ms, KVMState *s)
325 {
326 return 0;
327 }
328
329 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
330 {
331 return cpu->cpu_index;
332 }
333
334 int kvm_arch_init_vcpu(CPUState *cs)
335 {
336 S390CPU *cpu = S390_CPU(cs);
337 kvm_s390_set_cpu_state(cpu, cpu->env.cpu_state);
338 cpu->irqstate = g_malloc0(VCPU_IRQ_BUF_SIZE);
339 return 0;
340 }
341
342 void kvm_s390_reset_vcpu(S390CPU *cpu)
343 {
344 CPUState *cs = CPU(cpu);
345
346 /* The initial reset call is needed here to reset in-kernel
347 * vcpu data that we can't access directly from QEMU
348 * (i.e. with older kernels which don't support sync_regs/ONE_REG).
349 * Before this ioctl cpu_synchronize_state() is called in common kvm
350 * code (kvm-all) */
351 if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
352 error_report("Initial CPU reset failed on CPU %i", cs->cpu_index);
353 }
354 }
355
356 static int can_sync_regs(CPUState *cs, int regs)
357 {
358 return cap_sync_regs && (cs->kvm_run->kvm_valid_regs & regs) == regs;
359 }
360
361 int kvm_arch_put_registers(CPUState *cs, int level)
362 {
363 S390CPU *cpu = S390_CPU(cs);
364 CPUS390XState *env = &cpu->env;
365 struct kvm_sregs sregs;
366 struct kvm_regs regs;
367 struct kvm_fpu fpu = {};
368 int r;
369 int i;
370
371 /* always save the PSW and the GPRS*/
372 cs->kvm_run->psw_addr = env->psw.addr;
373 cs->kvm_run->psw_mask = env->psw.mask;
374
375 if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
376 for (i = 0; i < 16; i++) {
377 cs->kvm_run->s.regs.gprs[i] = env->regs[i];
378 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
379 }
380 } else {
381 for (i = 0; i < 16; i++) {
382 regs.gprs[i] = env->regs[i];
383 }
384 r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
385 if (r < 0) {
386 return r;
387 }
388 }
389
390 if (can_sync_regs(cs, KVM_SYNC_VRS)) {
391 for (i = 0; i < 32; i++) {
392 cs->kvm_run->s.regs.vrs[i][0] = env->vregs[i][0].ll;
393 cs->kvm_run->s.regs.vrs[i][1] = env->vregs[i][1].ll;
394 }
395 cs->kvm_run->s.regs.fpc = env->fpc;
396 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_VRS;
397 } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
398 for (i = 0; i < 16; i++) {
399 cs->kvm_run->s.regs.fprs[i] = get_freg(env, i)->ll;
400 }
401 cs->kvm_run->s.regs.fpc = env->fpc;
402 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_FPRS;
403 } else {
404 /* Floating point */
405 for (i = 0; i < 16; i++) {
406 fpu.fprs[i] = get_freg(env, i)->ll;
407 }
408 fpu.fpc = env->fpc;
409
410 r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
411 if (r < 0) {
412 return r;
413 }
414 }
415
416 /* Do we need to save more than that? */
417 if (level == KVM_PUT_RUNTIME_STATE) {
418 return 0;
419 }
420
421 if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
422 cs->kvm_run->s.regs.cputm = env->cputm;
423 cs->kvm_run->s.regs.ckc = env->ckc;
424 cs->kvm_run->s.regs.todpr = env->todpr;
425 cs->kvm_run->s.regs.gbea = env->gbea;
426 cs->kvm_run->s.regs.pp = env->pp;
427 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ARCH0;
428 } else {
429 /*
430 * These ONE_REGS are not protected by a capability. As they are only
431 * necessary for migration we just trace a possible error, but don't
432 * return with an error return code.
433 */
434 kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
435 kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
436 kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
437 kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
438 kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
439 }
440
441 if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
442 memcpy(cs->kvm_run->s.regs.riccb, env->riccb, 64);
443 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_RICCB;
444 }
445
446 /* pfault parameters */
447 if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
448 cs->kvm_run->s.regs.pft = env->pfault_token;
449 cs->kvm_run->s.regs.pfs = env->pfault_select;
450 cs->kvm_run->s.regs.pfc = env->pfault_compare;
451 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PFAULT;
452 } else if (cap_async_pf) {
453 r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
454 if (r < 0) {
455 return r;
456 }
457 r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
458 if (r < 0) {
459 return r;
460 }
461 r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
462 if (r < 0) {
463 return r;
464 }
465 }
466
467 /* access registers and control registers*/
468 if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
469 for (i = 0; i < 16; i++) {
470 cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
471 cs->kvm_run->s.regs.crs[i] = env->cregs[i];
472 }
473 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
474 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
475 } else {
476 for (i = 0; i < 16; i++) {
477 sregs.acrs[i] = env->aregs[i];
478 sregs.crs[i] = env->cregs[i];
479 }
480 r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
481 if (r < 0) {
482 return r;
483 }
484 }
485
486 if (can_sync_regs(cs, KVM_SYNC_GSCB)) {
487 memcpy(cs->kvm_run->s.regs.gscb, env->gscb, 32);
488 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GSCB;
489 }
490
491 if (can_sync_regs(cs, KVM_SYNC_BPBC)) {
492 cs->kvm_run->s.regs.bpbc = env->bpbc;
493 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_BPBC;
494 }
495
496 /* Finally the prefix */
497 if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
498 cs->kvm_run->s.regs.prefix = env->psa;
499 cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
500 } else {
501 /* prefix is only supported via sync regs */
502 }
503 return 0;
504 }
505
506 int kvm_arch_get_registers(CPUState *cs)
507 {
508 S390CPU *cpu = S390_CPU(cs);
509 CPUS390XState *env = &cpu->env;
510 struct kvm_sregs sregs;
511 struct kvm_regs regs;
512 struct kvm_fpu fpu;
513 int i, r;
514
515 /* get the PSW */
516 env->psw.addr = cs->kvm_run->psw_addr;
517 env->psw.mask = cs->kvm_run->psw_mask;
518
519 /* the GPRS */
520 if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
521 for (i = 0; i < 16; i++) {
522 env->regs[i] = cs->kvm_run->s.regs.gprs[i];
523 }
524 } else {
525 r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
526 if (r < 0) {
527 return r;
528 }
529 for (i = 0; i < 16; i++) {
530 env->regs[i] = regs.gprs[i];
531 }
532 }
533
534 /* The ACRS and CRS */
535 if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
536 for (i = 0; i < 16; i++) {
537 env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
538 env->cregs[i] = cs->kvm_run->s.regs.crs[i];
539 }
540 } else {
541 r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
542 if (r < 0) {
543 return r;
544 }
545 for (i = 0; i < 16; i++) {
546 env->aregs[i] = sregs.acrs[i];
547 env->cregs[i] = sregs.crs[i];
548 }
549 }
550
551 /* Floating point and vector registers */
552 if (can_sync_regs(cs, KVM_SYNC_VRS)) {
553 for (i = 0; i < 32; i++) {
554 env->vregs[i][0].ll = cs->kvm_run->s.regs.vrs[i][0];
555 env->vregs[i][1].ll = cs->kvm_run->s.regs.vrs[i][1];
556 }
557 env->fpc = cs->kvm_run->s.regs.fpc;
558 } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
559 for (i = 0; i < 16; i++) {
560 get_freg(env, i)->ll = cs->kvm_run->s.regs.fprs[i];
561 }
562 env->fpc = cs->kvm_run->s.regs.fpc;
563 } else {
564 r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
565 if (r < 0) {
566 return r;
567 }
568 for (i = 0; i < 16; i++) {
569 get_freg(env, i)->ll = fpu.fprs[i];
570 }
571 env->fpc = fpu.fpc;
572 }
573
574 /* The prefix */
575 if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
576 env->psa = cs->kvm_run->s.regs.prefix;
577 }
578
579 if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
580 env->cputm = cs->kvm_run->s.regs.cputm;
581 env->ckc = cs->kvm_run->s.regs.ckc;
582 env->todpr = cs->kvm_run->s.regs.todpr;
583 env->gbea = cs->kvm_run->s.regs.gbea;
584 env->pp = cs->kvm_run->s.regs.pp;
585 } else {
586 /*
587 * These ONE_REGS are not protected by a capability. As they are only
588 * necessary for migration we just trace a possible error, but don't
589 * return with an error return code.
590 */
591 kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
592 kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
593 kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
594 kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
595 kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
596 }
597
598 if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
599 memcpy(env->riccb, cs->kvm_run->s.regs.riccb, 64);
600 }
601
602 if (can_sync_regs(cs, KVM_SYNC_GSCB)) {
603 memcpy(env->gscb, cs->kvm_run->s.regs.gscb, 32);
604 }
605
606 if (can_sync_regs(cs, KVM_SYNC_BPBC)) {
607 env->bpbc = cs->kvm_run->s.regs.bpbc;
608 }
609
610 /* pfault parameters */
611 if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
612 env->pfault_token = cs->kvm_run->s.regs.pft;
613 env->pfault_select = cs->kvm_run->s.regs.pfs;
614 env->pfault_compare = cs->kvm_run->s.regs.pfc;
615 } else if (cap_async_pf) {
616 r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
617 if (r < 0) {
618 return r;
619 }
620 r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
621 if (r < 0) {
622 return r;
623 }
624 r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
625 if (r < 0) {
626 return r;
627 }
628 }
629
630 return 0;
631 }
632
633 int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_low)
634 {
635 int r;
636 struct kvm_device_attr attr = {
637 .group = KVM_S390_VM_TOD,
638 .attr = KVM_S390_VM_TOD_LOW,
639 .addr = (uint64_t)tod_low,
640 };
641
642 r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
643 if (r) {
644 return r;
645 }
646
647 attr.attr = KVM_S390_VM_TOD_HIGH;
648 attr.addr = (uint64_t)tod_high;
649 return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
650 }
651
652 int kvm_s390_get_clock_ext(uint8_t *tod_high, uint64_t *tod_low)
653 {
654 int r;
655 struct kvm_s390_vm_tod_clock gtod;
656 struct kvm_device_attr attr = {
657 .group = KVM_S390_VM_TOD,
658 .attr = KVM_S390_VM_TOD_EXT,
659 .addr = (uint64_t)&gtod,
660 };
661
662 r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
663 *tod_high = gtod.epoch_idx;
664 *tod_low = gtod.tod;
665
666 return r;
667 }
668
669 int kvm_s390_set_clock(uint8_t tod_high, uint64_t tod_low)
670 {
671 int r;
672 struct kvm_device_attr attr = {
673 .group = KVM_S390_VM_TOD,
674 .attr = KVM_S390_VM_TOD_LOW,
675 .addr = (uint64_t)&tod_low,
676 };
677
678 r = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
679 if (r) {
680 return r;
681 }
682
683 attr.attr = KVM_S390_VM_TOD_HIGH;
684 attr.addr = (uint64_t)&tod_high;
685 return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
686 }
687
688 int kvm_s390_set_clock_ext(uint8_t tod_high, uint64_t tod_low)
689 {
690 struct kvm_s390_vm_tod_clock gtod = {
691 .epoch_idx = tod_high,
692 .tod = tod_low,
693 };
694 struct kvm_device_attr attr = {
695 .group = KVM_S390_VM_TOD,
696 .attr = KVM_S390_VM_TOD_EXT,
697 .addr = (uint64_t)&gtod,
698 };
699
700 return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
701 }
702
703 /**
704 * kvm_s390_mem_op:
705 * @addr: the logical start address in guest memory
706 * @ar: the access register number
707 * @hostbuf: buffer in host memory. NULL = do only checks w/o copying
708 * @len: length that should be transferred
709 * @is_write: true = write, false = read
710 * Returns: 0 on success, non-zero if an exception or error occurred
711 *
712 * Use KVM ioctl to read/write from/to guest memory. An access exception
713 * is injected into the vCPU in case of translation errors.
714 */
715 int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, uint8_t ar, void *hostbuf,
716 int len, bool is_write)
717 {
718 struct kvm_s390_mem_op mem_op = {
719 .gaddr = addr,
720 .flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION,
721 .size = len,
722 .op = is_write ? KVM_S390_MEMOP_LOGICAL_WRITE
723 : KVM_S390_MEMOP_LOGICAL_READ,
724 .buf = (uint64_t)hostbuf,
725 .ar = ar,
726 };
727 int ret;
728
729 if (!cap_mem_op) {
730 return -ENOSYS;
731 }
732 if (!hostbuf) {
733 mem_op.flags |= KVM_S390_MEMOP_F_CHECK_ONLY;
734 }
735
736 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
737 if (ret < 0) {
738 error_printf("KVM_S390_MEM_OP failed: %s\n", strerror(-ret));
739 }
740 return ret;
741 }
742
743 /*
744 * Legacy layout for s390:
745 * Older S390 KVM requires the topmost vma of the RAM to be
746 * smaller than an system defined value, which is at least 256GB.
747 * Larger systems have larger values. We put the guest between
748 * the end of data segment (system break) and this value. We
749 * use 32GB as a base to have enough room for the system break
750 * to grow. We also have to use MAP parameters that avoid
751 * read-only mapping of guest pages.
752 */
753 static void *legacy_s390_alloc(size_t size, uint64_t *align, bool shared)
754 {
755 static void *mem;
756
757 if (mem) {
758 /* we only support one allocation, which is enough for initial ram */
759 return NULL;
760 }
761
762 mem = mmap((void *) 0x800000000ULL, size,
763 PROT_EXEC|PROT_READ|PROT_WRITE,
764 MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
765 if (mem == MAP_FAILED) {
766 mem = NULL;
767 }
768 return mem;
769 }
770
771 static uint8_t const *sw_bp_inst;
772 static uint8_t sw_bp_ilen;
773
774 static void determine_sw_breakpoint_instr(void)
775 {
776 /* DIAG 501 is used for sw breakpoints with old kernels */
777 static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
778 /* Instruction 0x0000 is used for sw breakpoints with recent kernels */
779 static const uint8_t instr_0x0000[] = {0x00, 0x00};
780
781 if (sw_bp_inst) {
782 return;
783 }
784 if (kvm_vm_enable_cap(kvm_state, KVM_CAP_S390_USER_INSTR0, 0)) {
785 sw_bp_inst = diag_501;
786 sw_bp_ilen = sizeof(diag_501);
787 DPRINTF("KVM: will use 4-byte sw breakpoints.\n");
788 } else {
789 sw_bp_inst = instr_0x0000;
790 sw_bp_ilen = sizeof(instr_0x0000);
791 DPRINTF("KVM: will use 2-byte sw breakpoints.\n");
792 }
793 }
794
795 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
796 {
797 determine_sw_breakpoint_instr();
798
799 if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
800 sw_bp_ilen, 0) ||
801 cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)sw_bp_inst, sw_bp_ilen, 1)) {
802 return -EINVAL;
803 }
804 return 0;
805 }
806
807 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
808 {
809 uint8_t t[MAX_ILEN];
810
811 if (cpu_memory_rw_debug(cs, bp->pc, t, sw_bp_ilen, 0)) {
812 return -EINVAL;
813 } else if (memcmp(t, sw_bp_inst, sw_bp_ilen)) {
814 return -EINVAL;
815 } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
816 sw_bp_ilen, 1)) {
817 return -EINVAL;
818 }
819
820 return 0;
821 }
822
823 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
824 int len, int type)
825 {
826 int n;
827
828 for (n = 0; n < nb_hw_breakpoints; n++) {
829 if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
830 (hw_breakpoints[n].len == len || len == -1)) {
831 return &hw_breakpoints[n];
832 }
833 }
834
835 return NULL;
836 }
837
838 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
839 {
840 int size;
841
842 if (find_hw_breakpoint(addr, len, type)) {
843 return -EEXIST;
844 }
845
846 size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
847
848 if (!hw_breakpoints) {
849 nb_hw_breakpoints = 0;
850 hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
851 } else {
852 hw_breakpoints =
853 (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
854 }
855
856 if (!hw_breakpoints) {
857 nb_hw_breakpoints = 0;
858 return -ENOMEM;
859 }
860
861 hw_breakpoints[nb_hw_breakpoints].addr = addr;
862 hw_breakpoints[nb_hw_breakpoints].len = len;
863 hw_breakpoints[nb_hw_breakpoints].type = type;
864
865 nb_hw_breakpoints++;
866
867 return 0;
868 }
869
870 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
871 target_ulong len, int type)
872 {
873 switch (type) {
874 case GDB_BREAKPOINT_HW:
875 type = KVM_HW_BP;
876 break;
877 case GDB_WATCHPOINT_WRITE:
878 if (len < 1) {
879 return -EINVAL;
880 }
881 type = KVM_HW_WP_WRITE;
882 break;
883 default:
884 return -ENOSYS;
885 }
886 return insert_hw_breakpoint(addr, len, type);
887 }
888
889 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
890 target_ulong len, int type)
891 {
892 int size;
893 struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
894
895 if (bp == NULL) {
896 return -ENOENT;
897 }
898
899 nb_hw_breakpoints--;
900 if (nb_hw_breakpoints > 0) {
901 /*
902 * In order to trim the array, move the last element to the position to
903 * be removed - if necessary.
904 */
905 if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
906 *bp = hw_breakpoints[nb_hw_breakpoints];
907 }
908 size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
909 hw_breakpoints =
910 (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
911 } else {
912 g_free(hw_breakpoints);
913 hw_breakpoints = NULL;
914 }
915
916 return 0;
917 }
918
919 void kvm_arch_remove_all_hw_breakpoints(void)
920 {
921 nb_hw_breakpoints = 0;
922 g_free(hw_breakpoints);
923 hw_breakpoints = NULL;
924 }
925
926 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
927 {
928 int i;
929
930 if (nb_hw_breakpoints > 0) {
931 dbg->arch.nr_hw_bp = nb_hw_breakpoints;
932 dbg->arch.hw_bp = hw_breakpoints;
933
934 for (i = 0; i < nb_hw_breakpoints; ++i) {
935 hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
936 hw_breakpoints[i].addr);
937 }
938 dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
939 } else {
940 dbg->arch.nr_hw_bp = 0;
941 dbg->arch.hw_bp = NULL;
942 }
943 }
944
945 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
946 {
947 }
948
949 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
950 {
951 return MEMTXATTRS_UNSPECIFIED;
952 }
953
954 int kvm_arch_process_async_events(CPUState *cs)
955 {
956 return cs->halted;
957 }
958
959 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
960 struct kvm_s390_interrupt *interrupt)
961 {
962 int r = 0;
963
964 interrupt->type = irq->type;
965 switch (irq->type) {
966 case KVM_S390_INT_VIRTIO:
967 interrupt->parm = irq->u.ext.ext_params;
968 /* fall through */
969 case KVM_S390_INT_PFAULT_INIT:
970 case KVM_S390_INT_PFAULT_DONE:
971 interrupt->parm64 = irq->u.ext.ext_params2;
972 break;
973 case KVM_S390_PROGRAM_INT:
974 interrupt->parm = irq->u.pgm.code;
975 break;
976 case KVM_S390_SIGP_SET_PREFIX:
977 interrupt->parm = irq->u.prefix.address;
978 break;
979 case KVM_S390_INT_SERVICE:
980 interrupt->parm = irq->u.ext.ext_params;
981 break;
982 case KVM_S390_MCHK:
983 interrupt->parm = irq->u.mchk.cr14;
984 interrupt->parm64 = irq->u.mchk.mcic;
985 break;
986 case KVM_S390_INT_EXTERNAL_CALL:
987 interrupt->parm = irq->u.extcall.code;
988 break;
989 case KVM_S390_INT_EMERGENCY:
990 interrupt->parm = irq->u.emerg.code;
991 break;
992 case KVM_S390_SIGP_STOP:
993 case KVM_S390_RESTART:
994 break; /* These types have no parameters */
995 case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
996 interrupt->parm = irq->u.io.subchannel_id << 16;
997 interrupt->parm |= irq->u.io.subchannel_nr;
998 interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
999 interrupt->parm64 |= irq->u.io.io_int_word;
1000 break;
1001 default:
1002 r = -EINVAL;
1003 break;
1004 }
1005 return r;
1006 }
1007
1008 static void inject_vcpu_irq_legacy(CPUState *cs, struct kvm_s390_irq *irq)
1009 {
1010 struct kvm_s390_interrupt kvmint = {};
1011 int r;
1012
1013 r = s390_kvm_irq_to_interrupt(irq, &kvmint);
1014 if (r < 0) {
1015 fprintf(stderr, "%s called with bogus interrupt\n", __func__);
1016 exit(1);
1017 }
1018
1019 r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
1020 if (r < 0) {
1021 fprintf(stderr, "KVM failed to inject interrupt\n");
1022 exit(1);
1023 }
1024 }
1025
1026 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
1027 {
1028 CPUState *cs = CPU(cpu);
1029 int r;
1030
1031 if (cap_s390_irq) {
1032 r = kvm_vcpu_ioctl(cs, KVM_S390_IRQ, irq);
1033 if (!r) {
1034 return;
1035 }
1036 error_report("KVM failed to inject interrupt %llx", irq->type);
1037 exit(1);
1038 }
1039
1040 inject_vcpu_irq_legacy(cs, irq);
1041 }
1042
1043 void kvm_s390_floating_interrupt_legacy(struct kvm_s390_irq *irq)
1044 {
1045 struct kvm_s390_interrupt kvmint = {};
1046 int r;
1047
1048 r = s390_kvm_irq_to_interrupt(irq, &kvmint);
1049 if (r < 0) {
1050 fprintf(stderr, "%s called with bogus interrupt\n", __func__);
1051 exit(1);
1052 }
1053
1054 r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
1055 if (r < 0) {
1056 fprintf(stderr, "KVM failed to inject interrupt\n");
1057 exit(1);
1058 }
1059 }
1060
1061 void kvm_s390_program_interrupt(S390CPU *cpu, uint16_t code)
1062 {
1063 struct kvm_s390_irq irq = {
1064 .type = KVM_S390_PROGRAM_INT,
1065 .u.pgm.code = code,
1066 };
1067
1068 kvm_s390_vcpu_interrupt(cpu, &irq);
1069 }
1070
1071 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code)
1072 {
1073 struct kvm_s390_irq irq = {
1074 .type = KVM_S390_PROGRAM_INT,
1075 .u.pgm.code = code,
1076 .u.pgm.trans_exc_code = te_code,
1077 .u.pgm.exc_access_id = te_code & 3,
1078 };
1079
1080 kvm_s390_vcpu_interrupt(cpu, &irq);
1081 }
1082
1083 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
1084 uint16_t ipbh0)
1085 {
1086 CPUS390XState *env = &cpu->env;
1087 uint64_t sccb;
1088 uint32_t code;
1089 int r = 0;
1090
1091 sccb = env->regs[ipbh0 & 0xf];
1092 code = env->regs[(ipbh0 & 0xf0) >> 4];
1093
1094 r = sclp_service_call(env, sccb, code);
1095 if (r < 0) {
1096 kvm_s390_program_interrupt(cpu, -r);
1097 } else {
1098 setcc(cpu, r);
1099 }
1100
1101 return 0;
1102 }
1103
1104 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1105 {
1106 CPUS390XState *env = &cpu->env;
1107 int rc = 0;
1108 uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
1109
1110 switch (ipa1) {
1111 case PRIV_B2_XSCH:
1112 ioinst_handle_xsch(cpu, env->regs[1], RA_IGNORED);
1113 break;
1114 case PRIV_B2_CSCH:
1115 ioinst_handle_csch(cpu, env->regs[1], RA_IGNORED);
1116 break;
1117 case PRIV_B2_HSCH:
1118 ioinst_handle_hsch(cpu, env->regs[1], RA_IGNORED);
1119 break;
1120 case PRIV_B2_MSCH:
1121 ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb, RA_IGNORED);
1122 break;
1123 case PRIV_B2_SSCH:
1124 ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb, RA_IGNORED);
1125 break;
1126 case PRIV_B2_STCRW:
1127 ioinst_handle_stcrw(cpu, run->s390_sieic.ipb, RA_IGNORED);
1128 break;
1129 case PRIV_B2_STSCH:
1130 ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb, RA_IGNORED);
1131 break;
1132 case PRIV_B2_TSCH:
1133 /* We should only get tsch via KVM_EXIT_S390_TSCH. */
1134 fprintf(stderr, "Spurious tsch intercept\n");
1135 break;
1136 case PRIV_B2_CHSC:
1137 ioinst_handle_chsc(cpu, run->s390_sieic.ipb, RA_IGNORED);
1138 break;
1139 case PRIV_B2_TPI:
1140 /* This should have been handled by kvm already. */
1141 fprintf(stderr, "Spurious tpi intercept\n");
1142 break;
1143 case PRIV_B2_SCHM:
1144 ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
1145 run->s390_sieic.ipb, RA_IGNORED);
1146 break;
1147 case PRIV_B2_RSCH:
1148 ioinst_handle_rsch(cpu, env->regs[1], RA_IGNORED);
1149 break;
1150 case PRIV_B2_RCHP:
1151 ioinst_handle_rchp(cpu, env->regs[1], RA_IGNORED);
1152 break;
1153 case PRIV_B2_STCPS:
1154 /* We do not provide this instruction, it is suppressed. */
1155 break;
1156 case PRIV_B2_SAL:
1157 ioinst_handle_sal(cpu, env->regs[1], RA_IGNORED);
1158 break;
1159 case PRIV_B2_SIGA:
1160 /* Not provided, set CC = 3 for subchannel not operational */
1161 setcc(cpu, 3);
1162 break;
1163 case PRIV_B2_SCLP_CALL:
1164 rc = kvm_sclp_service_call(cpu, run, ipbh0);
1165 break;
1166 default:
1167 rc = -1;
1168 DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
1169 break;
1170 }
1171
1172 return rc;
1173 }
1174
1175 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run,
1176 uint8_t *ar)
1177 {
1178 CPUS390XState *env = &cpu->env;
1179 uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
1180 uint32_t base2 = run->s390_sieic.ipb >> 28;
1181 uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1182 ((run->s390_sieic.ipb & 0xff00) << 4);
1183
1184 if (disp2 & 0x80000) {
1185 disp2 += 0xfff00000;
1186 }
1187 if (ar) {
1188 *ar = base2;
1189 }
1190
1191 return (base2 ? env->regs[base2] : 0) +
1192 (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
1193 }
1194
1195 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run,
1196 uint8_t *ar)
1197 {
1198 CPUS390XState *env = &cpu->env;
1199 uint32_t base2 = run->s390_sieic.ipb >> 28;
1200 uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1201 ((run->s390_sieic.ipb & 0xff00) << 4);
1202
1203 if (disp2 & 0x80000) {
1204 disp2 += 0xfff00000;
1205 }
1206 if (ar) {
1207 *ar = base2;
1208 }
1209
1210 return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
1211 }
1212
1213 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
1214 {
1215 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1216
1217 if (s390_has_feat(S390_FEAT_ZPCI)) {
1218 return clp_service_call(cpu, r2, RA_IGNORED);
1219 } else {
1220 return -1;
1221 }
1222 }
1223
1224 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
1225 {
1226 uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1227 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1228
1229 if (s390_has_feat(S390_FEAT_ZPCI)) {
1230 return pcilg_service_call(cpu, r1, r2, RA_IGNORED);
1231 } else {
1232 return -1;
1233 }
1234 }
1235
1236 static int kvm_pcistg_service_call(S390CPU *cpu, struct kvm_run *run)
1237 {
1238 uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1239 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1240
1241 if (s390_has_feat(S390_FEAT_ZPCI)) {
1242 return pcistg_service_call(cpu, r1, r2, RA_IGNORED);
1243 } else {
1244 return -1;
1245 }
1246 }
1247
1248 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1249 {
1250 uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1251 uint64_t fiba;
1252 uint8_t ar;
1253
1254 if (s390_has_feat(S390_FEAT_ZPCI)) {
1255 fiba = get_base_disp_rxy(cpu, run, &ar);
1256
1257 return stpcifc_service_call(cpu, r1, fiba, ar, RA_IGNORED);
1258 } else {
1259 return -1;
1260 }
1261 }
1262
1263 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
1264 {
1265 CPUS390XState *env = &cpu->env;
1266 uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1267 uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1268 uint8_t isc;
1269 uint16_t mode;
1270 int r;
1271
1272 mode = env->regs[r1] & 0xffff;
1273 isc = (env->regs[r3] >> 27) & 0x7;
1274 r = css_do_sic(env, isc, mode);
1275 if (r) {
1276 kvm_s390_program_interrupt(cpu, -r);
1277 }
1278
1279 return 0;
1280 }
1281
1282 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
1283 {
1284 uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1285 uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1286
1287 if (s390_has_feat(S390_FEAT_ZPCI)) {
1288 return rpcit_service_call(cpu, r1, r2, RA_IGNORED);
1289 } else {
1290 return -1;
1291 }
1292 }
1293
1294 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
1295 {
1296 uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1297 uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1298 uint64_t gaddr;
1299 uint8_t ar;
1300
1301 if (s390_has_feat(S390_FEAT_ZPCI)) {
1302 gaddr = get_base_disp_rsy(cpu, run, &ar);
1303
1304 return pcistb_service_call(cpu, r1, r3, gaddr, ar, RA_IGNORED);
1305 } else {
1306 return -1;
1307 }
1308 }
1309
1310 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1311 {
1312 uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1313 uint64_t fiba;
1314 uint8_t ar;
1315
1316 if (s390_has_feat(S390_FEAT_ZPCI)) {
1317 fiba = get_base_disp_rxy(cpu, run, &ar);
1318
1319 return mpcifc_service_call(cpu, r1, fiba, ar, RA_IGNORED);
1320 } else {
1321 return -1;
1322 }
1323 }
1324
1325 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1326 {
1327 int r = 0;
1328
1329 switch (ipa1) {
1330 case PRIV_B9_CLP:
1331 r = kvm_clp_service_call(cpu, run);
1332 break;
1333 case PRIV_B9_PCISTG:
1334 r = kvm_pcistg_service_call(cpu, run);
1335 break;
1336 case PRIV_B9_PCILG:
1337 r = kvm_pcilg_service_call(cpu, run);
1338 break;
1339 case PRIV_B9_RPCIT:
1340 r = kvm_rpcit_service_call(cpu, run);
1341 break;
1342 case PRIV_B9_EQBS:
1343 /* just inject exception */
1344 r = -1;
1345 break;
1346 default:
1347 r = -1;
1348 DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
1349 break;
1350 }
1351
1352 return r;
1353 }
1354
1355 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1356 {
1357 int r = 0;
1358
1359 switch (ipbl) {
1360 case PRIV_EB_PCISTB:
1361 r = kvm_pcistb_service_call(cpu, run);
1362 break;
1363 case PRIV_EB_SIC:
1364 r = kvm_sic_service_call(cpu, run);
1365 break;
1366 case PRIV_EB_SQBS:
1367 /* just inject exception */
1368 r = -1;
1369 break;
1370 default:
1371 r = -1;
1372 DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1373 break;
1374 }
1375
1376 return r;
1377 }
1378
1379 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1380 {
1381 int r = 0;
1382
1383 switch (ipbl) {
1384 case PRIV_E3_MPCIFC:
1385 r = kvm_mpcifc_service_call(cpu, run);
1386 break;
1387 case PRIV_E3_STPCIFC:
1388 r = kvm_stpcifc_service_call(cpu, run);
1389 break;
1390 default:
1391 r = -1;
1392 DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1393 break;
1394 }
1395
1396 return r;
1397 }
1398
1399 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1400 {
1401 CPUS390XState *env = &cpu->env;
1402 int ret;
1403
1404 ret = s390_virtio_hypercall(env);
1405 if (ret == -EINVAL) {
1406 kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1407 return 0;
1408 }
1409
1410 return ret;
1411 }
1412
1413 static void kvm_handle_diag_288(S390CPU *cpu, struct kvm_run *run)
1414 {
1415 uint64_t r1, r3;
1416 int rc;
1417
1418 r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1419 r3 = run->s390_sieic.ipa & 0x000f;
1420 rc = handle_diag_288(&cpu->env, r1, r3);
1421 if (rc) {
1422 kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1423 }
1424 }
1425
1426 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1427 {
1428 uint64_t r1, r3;
1429
1430 r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1431 r3 = run->s390_sieic.ipa & 0x000f;
1432 handle_diag_308(&cpu->env, r1, r3, RA_IGNORED);
1433 }
1434
1435 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1436 {
1437 CPUS390XState *env = &cpu->env;
1438 unsigned long pc;
1439
1440 pc = env->psw.addr - sw_bp_ilen;
1441 if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1442 env->psw.addr = pc;
1443 return EXCP_DEBUG;
1444 }
1445
1446 return -ENOENT;
1447 }
1448
1449 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1450
1451 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1452 {
1453 int r = 0;
1454 uint16_t func_code;
1455
1456 /*
1457 * For any diagnose call we support, bits 48-63 of the resulting
1458 * address specify the function code; the remainder is ignored.
1459 */
1460 func_code = decode_basedisp_rs(&cpu->env, ipb, NULL) & DIAG_KVM_CODE_MASK;
1461 switch (func_code) {
1462 case DIAG_TIMEREVENT:
1463 kvm_handle_diag_288(cpu, run);
1464 break;
1465 case DIAG_IPL:
1466 kvm_handle_diag_308(cpu, run);
1467 break;
1468 case DIAG_KVM_HYPERCALL:
1469 r = handle_hypercall(cpu, run);
1470 break;
1471 case DIAG_KVM_BREAKPOINT:
1472 r = handle_sw_breakpoint(cpu, run);
1473 break;
1474 default:
1475 DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1476 kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1477 break;
1478 }
1479
1480 return r;
1481 }
1482
1483 static int kvm_s390_handle_sigp(S390CPU *cpu, uint8_t ipa1, uint32_t ipb)
1484 {
1485 CPUS390XState *env = &cpu->env;
1486 const uint8_t r1 = ipa1 >> 4;
1487 const uint8_t r3 = ipa1 & 0x0f;
1488 int ret;
1489 uint8_t order;
1490
1491 /* get order code */
1492 order = decode_basedisp_rs(env, ipb, NULL) & SIGP_ORDER_MASK;
1493
1494 ret = handle_sigp(env, order, r1, r3);
1495 setcc(cpu, ret);
1496 return 0;
1497 }
1498
1499 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1500 {
1501 unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1502 uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1503 int r = -1;
1504
1505 DPRINTF("handle_instruction 0x%x 0x%x\n",
1506 run->s390_sieic.ipa, run->s390_sieic.ipb);
1507 switch (ipa0) {
1508 case IPA0_B2:
1509 r = handle_b2(cpu, run, ipa1);
1510 break;
1511 case IPA0_B9:
1512 r = handle_b9(cpu, run, ipa1);
1513 break;
1514 case IPA0_EB:
1515 r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1516 break;
1517 case IPA0_E3:
1518 r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1519 break;
1520 case IPA0_DIAG:
1521 r = handle_diag(cpu, run, run->s390_sieic.ipb);
1522 break;
1523 case IPA0_SIGP:
1524 r = kvm_s390_handle_sigp(cpu, ipa1, run->s390_sieic.ipb);
1525 break;
1526 }
1527
1528 if (r < 0) {
1529 r = 0;
1530 kvm_s390_program_interrupt(cpu, PGM_OPERATION);
1531 }
1532
1533 return r;
1534 }
1535
1536 static void unmanageable_intercept(S390CPU *cpu, S390CrashReason reason,
1537 int pswoffset)
1538 {
1539 CPUState *cs = CPU(cpu);
1540
1541 s390_cpu_halt(cpu);
1542 cpu->env.crash_reason = reason;
1543 qemu_system_guest_panicked(cpu_get_crash_info(cs));
1544 }
1545
1546 /* try to detect pgm check loops */
1547 static int handle_oper_loop(S390CPU *cpu, struct kvm_run *run)
1548 {
1549 CPUState *cs = CPU(cpu);
1550 PSW oldpsw, newpsw;
1551
1552 newpsw.mask = ldq_phys(cs->as, cpu->env.psa +
1553 offsetof(LowCore, program_new_psw));
1554 newpsw.addr = ldq_phys(cs->as, cpu->env.psa +
1555 offsetof(LowCore, program_new_psw) + 8);
1556 oldpsw.mask = run->psw_mask;
1557 oldpsw.addr = run->psw_addr;
1558 /*
1559 * Avoid endless loops of operation exceptions, if the pgm new
1560 * PSW will cause a new operation exception.
1561 * The heuristic checks if the pgm new psw is within 6 bytes before
1562 * the faulting psw address (with same DAT, AS settings) and the
1563 * new psw is not a wait psw and the fault was not triggered by
1564 * problem state. In that case go into crashed state.
1565 */
1566
1567 if (oldpsw.addr - newpsw.addr <= 6 &&
1568 !(newpsw.mask & PSW_MASK_WAIT) &&
1569 !(oldpsw.mask & PSW_MASK_PSTATE) &&
1570 (newpsw.mask & PSW_MASK_ASC) == (oldpsw.mask & PSW_MASK_ASC) &&
1571 (newpsw.mask & PSW_MASK_DAT) == (oldpsw.mask & PSW_MASK_DAT)) {
1572 unmanageable_intercept(cpu, S390_CRASH_REASON_OPINT_LOOP,
1573 offsetof(LowCore, program_new_psw));
1574 return EXCP_HALTED;
1575 }
1576 return 0;
1577 }
1578
1579 static int handle_intercept(S390CPU *cpu)
1580 {
1581 CPUState *cs = CPU(cpu);
1582 struct kvm_run *run = cs->kvm_run;
1583 int icpt_code = run->s390_sieic.icptcode;
1584 int r = 0;
1585
1586 DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1587 (long)cs->kvm_run->psw_addr);
1588 switch (icpt_code) {
1589 case ICPT_INSTRUCTION:
1590 r = handle_instruction(cpu, run);
1591 break;
1592 case ICPT_PROGRAM:
1593 unmanageable_intercept(cpu, S390_CRASH_REASON_PGMINT_LOOP,
1594 offsetof(LowCore, program_new_psw));
1595 r = EXCP_HALTED;
1596 break;
1597 case ICPT_EXT_INT:
1598 unmanageable_intercept(cpu, S390_CRASH_REASON_EXTINT_LOOP,
1599 offsetof(LowCore, external_new_psw));
1600 r = EXCP_HALTED;
1601 break;
1602 case ICPT_WAITPSW:
1603 /* disabled wait, since enabled wait is handled in kernel */
1604 s390_handle_wait(cpu);
1605 r = EXCP_HALTED;
1606 break;
1607 case ICPT_CPU_STOP:
1608 do_stop_interrupt(&cpu->env);
1609 r = EXCP_HALTED;
1610 break;
1611 case ICPT_OPEREXC:
1612 /* check for break points */
1613 r = handle_sw_breakpoint(cpu, run);
1614 if (r == -ENOENT) {
1615 /* Then check for potential pgm check loops */
1616 r = handle_oper_loop(cpu, run);
1617 if (r == 0) {
1618 kvm_s390_program_interrupt(cpu, PGM_OPERATION);
1619 }
1620 }
1621 break;
1622 case ICPT_SOFT_INTERCEPT:
1623 fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1624 exit(1);
1625 break;
1626 case ICPT_IO:
1627 fprintf(stderr, "KVM unimplemented icpt IO\n");
1628 exit(1);
1629 break;
1630 default:
1631 fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1632 exit(1);
1633 break;
1634 }
1635
1636 return r;
1637 }
1638
1639 static int handle_tsch(S390CPU *cpu)
1640 {
1641 CPUState *cs = CPU(cpu);
1642 struct kvm_run *run = cs->kvm_run;
1643 int ret;
1644
1645 ret = ioinst_handle_tsch(cpu, cpu->env.regs[1], run->s390_tsch.ipb,
1646 RA_IGNORED);
1647 if (ret < 0) {
1648 /*
1649 * Failure.
1650 * If an I/O interrupt had been dequeued, we have to reinject it.
1651 */
1652 if (run->s390_tsch.dequeued) {
1653 s390_io_interrupt(run->s390_tsch.subchannel_id,
1654 run->s390_tsch.subchannel_nr,
1655 run->s390_tsch.io_int_parm,
1656 run->s390_tsch.io_int_word);
1657 }
1658 ret = 0;
1659 }
1660 return ret;
1661 }
1662
1663 static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
1664 {
1665 SysIB_322 sysib;
1666 int del;
1667
1668 if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
1669 return;
1670 }
1671 /* Shift the stack of Extended Names to prepare for our own data */
1672 memmove(&sysib.ext_names[1], &sysib.ext_names[0],
1673 sizeof(sysib.ext_names[0]) * (sysib.count - 1));
1674 /* First virt level, that doesn't provide Ext Names delimits stack. It is
1675 * assumed it's not capable of managing Extended Names for lower levels.
1676 */
1677 for (del = 1; del < sysib.count; del++) {
1678 if (!sysib.vm[del].ext_name_encoding || !sysib.ext_names[del][0]) {
1679 break;
1680 }
1681 }
1682 if (del < sysib.count) {
1683 memset(sysib.ext_names[del], 0,
1684 sizeof(sysib.ext_names[0]) * (sysib.count - del));
1685 }
1686 /* Insert short machine name in EBCDIC, padded with blanks */
1687 if (qemu_name) {
1688 memset(sysib.vm[0].name, 0x40, sizeof(sysib.vm[0].name));
1689 ebcdic_put(sysib.vm[0].name, qemu_name, MIN(sizeof(sysib.vm[0].name),
1690 strlen(qemu_name)));
1691 }
1692 sysib.vm[0].ext_name_encoding = 2; /* 2 = UTF-8 */
1693 memset(sysib.ext_names[0], 0, sizeof(sysib.ext_names[0]));
1694 /* If hypervisor specifies zero Extended Name in STSI322 SYSIB, it's
1695 * considered by s390 as not capable of providing any Extended Name.
1696 * Therefore if no name was specified on qemu invocation, we go with the
1697 * same "KVMguest" default, which KVM has filled into short name field.
1698 */
1699 if (qemu_name) {
1700 strncpy((char *)sysib.ext_names[0], qemu_name,
1701 sizeof(sysib.ext_names[0]));
1702 } else {
1703 strcpy((char *)sysib.ext_names[0], "KVMguest");
1704 }
1705 /* Insert UUID */
1706 memcpy(sysib.vm[0].uuid, &qemu_uuid, sizeof(sysib.vm[0].uuid));
1707
1708 s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
1709 }
1710
1711 static int handle_stsi(S390CPU *cpu)
1712 {
1713 CPUState *cs = CPU(cpu);
1714 struct kvm_run *run = cs->kvm_run;
1715
1716 switch (run->s390_stsi.fc) {
1717 case 3:
1718 if (run->s390_stsi.sel1 != 2 || run->s390_stsi.sel2 != 2) {
1719 return 0;
1720 }
1721 /* Only sysib 3.2.2 needs post-handling for now. */
1722 insert_stsi_3_2_2(cpu, run->s390_stsi.addr, run->s390_stsi.ar);
1723 return 0;
1724 default:
1725 return 0;
1726 }
1727 }
1728
1729 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
1730 {
1731 CPUState *cs = CPU(cpu);
1732 struct kvm_run *run = cs->kvm_run;
1733
1734 int ret = 0;
1735 struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1736
1737 switch (arch_info->type) {
1738 case KVM_HW_WP_WRITE:
1739 if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1740 cs->watchpoint_hit = &hw_watchpoint;
1741 hw_watchpoint.vaddr = arch_info->addr;
1742 hw_watchpoint.flags = BP_MEM_WRITE;
1743 ret = EXCP_DEBUG;
1744 }
1745 break;
1746 case KVM_HW_BP:
1747 if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1748 ret = EXCP_DEBUG;
1749 }
1750 break;
1751 case KVM_SINGLESTEP:
1752 if (cs->singlestep_enabled) {
1753 ret = EXCP_DEBUG;
1754 }
1755 break;
1756 default:
1757 ret = -ENOSYS;
1758 }
1759
1760 return ret;
1761 }
1762
1763 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1764 {
1765 S390CPU *cpu = S390_CPU(cs);
1766 int ret = 0;
1767
1768 qemu_mutex_lock_iothread();
1769
1770 kvm_cpu_synchronize_state(cs);
1771
1772 switch (run->exit_reason) {
1773 case KVM_EXIT_S390_SIEIC:
1774 ret = handle_intercept(cpu);
1775 break;
1776 case KVM_EXIT_S390_RESET:
1777 s390_ipl_reset_request(cs, S390_RESET_REIPL);
1778 break;
1779 case KVM_EXIT_S390_TSCH:
1780 ret = handle_tsch(cpu);
1781 break;
1782 case KVM_EXIT_S390_STSI:
1783 ret = handle_stsi(cpu);
1784 break;
1785 case KVM_EXIT_DEBUG:
1786 ret = kvm_arch_handle_debug_exit(cpu);
1787 break;
1788 default:
1789 fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
1790 break;
1791 }
1792 qemu_mutex_unlock_iothread();
1793
1794 if (ret == 0) {
1795 ret = EXCP_INTERRUPT;
1796 }
1797 return ret;
1798 }
1799
1800 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
1801 {
1802 return true;
1803 }
1804
1805 void kvm_s390_enable_css_support(S390CPU *cpu)
1806 {
1807 int r;
1808
1809 /* Activate host kernel channel subsystem support. */
1810 r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
1811 assert(r == 0);
1812 }
1813
1814 void kvm_arch_init_irq_routing(KVMState *s)
1815 {
1816 /*
1817 * Note that while irqchip capabilities generally imply that cpustates
1818 * are handled in-kernel, it is not true for s390 (yet); therefore, we
1819 * have to override the common code kvm_halt_in_kernel_allowed setting.
1820 */
1821 if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
1822 kvm_gsi_routing_allowed = true;
1823 kvm_halt_in_kernel_allowed = false;
1824 }
1825 }
1826
1827 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
1828 int vq, bool assign)
1829 {
1830 struct kvm_ioeventfd kick = {
1831 .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
1832 KVM_IOEVENTFD_FLAG_DATAMATCH,
1833 .fd = event_notifier_get_fd(notifier),
1834 .datamatch = vq,
1835 .addr = sch,
1836 .len = 8,
1837 };
1838 if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
1839 return -ENOSYS;
1840 }
1841 if (!assign) {
1842 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
1843 }
1844 return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
1845 }
1846
1847 int kvm_s390_get_ri(void)
1848 {
1849 return cap_ri;
1850 }
1851
1852 int kvm_s390_get_gs(void)
1853 {
1854 return cap_gs;
1855 }
1856
1857 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
1858 {
1859 struct kvm_mp_state mp_state = {};
1860 int ret;
1861
1862 /* the kvm part might not have been initialized yet */
1863 if (CPU(cpu)->kvm_state == NULL) {
1864 return 0;
1865 }
1866
1867 switch (cpu_state) {
1868 case S390_CPU_STATE_STOPPED:
1869 mp_state.mp_state = KVM_MP_STATE_STOPPED;
1870 break;
1871 case S390_CPU_STATE_CHECK_STOP:
1872 mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
1873 break;
1874 case S390_CPU_STATE_OPERATING:
1875 mp_state.mp_state = KVM_MP_STATE_OPERATING;
1876 break;
1877 case S390_CPU_STATE_LOAD:
1878 mp_state.mp_state = KVM_MP_STATE_LOAD;
1879 break;
1880 default:
1881 error_report("Requested CPU state is not a valid S390 CPU state: %u",
1882 cpu_state);
1883 exit(1);
1884 }
1885
1886 ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1887 if (ret) {
1888 trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
1889 strerror(-ret));
1890 }
1891
1892 return ret;
1893 }
1894
1895 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu)
1896 {
1897 struct kvm_s390_irq_state irq_state = {
1898 .buf = (uint64_t) cpu->irqstate,
1899 .len = VCPU_IRQ_BUF_SIZE,
1900 };
1901 CPUState *cs = CPU(cpu);
1902 int32_t bytes;
1903
1904 if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
1905 return;
1906 }
1907
1908 bytes = kvm_vcpu_ioctl(cs, KVM_S390_GET_IRQ_STATE, &irq_state);
1909 if (bytes < 0) {
1910 cpu->irqstate_saved_size = 0;
1911 error_report("Migration of interrupt state failed");
1912 return;
1913 }
1914
1915 cpu->irqstate_saved_size = bytes;
1916 }
1917
1918 int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu)
1919 {
1920 CPUState *cs = CPU(cpu);
1921 struct kvm_s390_irq_state irq_state = {
1922 .buf = (uint64_t) cpu->irqstate,
1923 .len = cpu->irqstate_saved_size,
1924 };
1925 int r;
1926
1927 if (cpu->irqstate_saved_size == 0) {
1928 return 0;
1929 }
1930
1931 if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
1932 return -ENOSYS;
1933 }
1934
1935 r = kvm_vcpu_ioctl(cs, KVM_S390_SET_IRQ_STATE, &irq_state);
1936 if (r) {
1937 error_report("Setting interrupt state failed %d", r);
1938 }
1939 return r;
1940 }
1941
1942 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1943 uint64_t address, uint32_t data, PCIDevice *dev)
1944 {
1945 S390PCIBusDevice *pbdev;
1946 uint32_t vec = data & ZPCI_MSI_VEC_MASK;
1947
1948 if (!dev) {
1949 DPRINTF("add_msi_route no pci device\n");
1950 return -ENODEV;
1951 }
1952
1953 pbdev = s390_pci_find_dev_by_target(s390_get_phb(), DEVICE(dev)->id);
1954 if (!pbdev) {
1955 DPRINTF("add_msi_route no zpci device\n");
1956 return -ENODEV;
1957 }
1958
1959 route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
1960 route->flags = 0;
1961 route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
1962 route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
1963 route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
1964 route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset + vec;
1965 route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;
1966 return 0;
1967 }
1968
1969 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
1970 int vector, PCIDevice *dev)
1971 {
1972 return 0;
1973 }
1974
1975 int kvm_arch_release_virq_post(int virq)
1976 {
1977 return 0;
1978 }
1979
1980 int kvm_arch_msi_data_to_gsi(uint32_t data)
1981 {
1982 abort();
1983 }
1984
1985 static int query_cpu_subfunc(S390FeatBitmap features)
1986 {
1987 struct kvm_s390_vm_cpu_subfunc prop;
1988 struct kvm_device_attr attr = {
1989 .group = KVM_S390_VM_CPU_MODEL,
1990 .attr = KVM_S390_VM_CPU_MACHINE_SUBFUNC,
1991 .addr = (uint64_t) &prop,
1992 };
1993 int rc;
1994
1995 rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
1996 if (rc) {
1997 return rc;
1998 }
1999
2000 /*
2001 * We're going to add all subfunctions now, if the corresponding feature
2002 * is available that unlocks the query functions.
2003 */
2004 s390_add_from_feat_block(features, S390_FEAT_TYPE_PLO, prop.plo);
2005 if (test_bit(S390_FEAT_TOD_CLOCK_STEERING, features)) {
2006 s390_add_from_feat_block(features, S390_FEAT_TYPE_PTFF, prop.ptff);
2007 }
2008 if (test_bit(S390_FEAT_MSA, features)) {
2009 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMAC, prop.kmac);
2010 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMC, prop.kmc);
2011 s390_add_from_feat_block(features, S390_FEAT_TYPE_KM, prop.km);
2012 s390_add_from_feat_block(features, S390_FEAT_TYPE_KIMD, prop.kimd);
2013 s390_add_from_feat_block(features, S390_FEAT_TYPE_KLMD, prop.klmd);
2014 }
2015 if (test_bit(S390_FEAT_MSA_EXT_3, features)) {
2016 s390_add_from_feat_block(features, S390_FEAT_TYPE_PCKMO, prop.pckmo);
2017 }
2018 if (test_bit(S390_FEAT_MSA_EXT_4, features)) {
2019 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMCTR, prop.kmctr);
2020 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMF, prop.kmf);
2021 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMO, prop.kmo);
2022 s390_add_from_feat_block(features, S390_FEAT_TYPE_PCC, prop.pcc);
2023 }
2024 if (test_bit(S390_FEAT_MSA_EXT_5, features)) {
2025 s390_add_from_feat_block(features, S390_FEAT_TYPE_PPNO, prop.ppno);
2026 }
2027 if (test_bit(S390_FEAT_MSA_EXT_8, features)) {
2028 s390_add_from_feat_block(features, S390_FEAT_TYPE_KMA, prop.kma);
2029 }
2030 return 0;
2031 }
2032
2033 static int configure_cpu_subfunc(const S390FeatBitmap features)
2034 {
2035 struct kvm_s390_vm_cpu_subfunc prop = {};
2036 struct kvm_device_attr attr = {
2037 .group = KVM_S390_VM_CPU_MODEL,
2038 .attr = KVM_S390_VM_CPU_PROCESSOR_SUBFUNC,
2039 .addr = (uint64_t) &prop,
2040 };
2041
2042 if (!kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2043 KVM_S390_VM_CPU_PROCESSOR_SUBFUNC)) {
2044 /* hardware support might be missing, IBC will handle most of this */
2045 return 0;
2046 }
2047
2048 s390_fill_feat_block(features, S390_FEAT_TYPE_PLO, prop.plo);
2049 if (test_bit(S390_FEAT_TOD_CLOCK_STEERING, features)) {
2050 s390_fill_feat_block(features, S390_FEAT_TYPE_PTFF, prop.ptff);
2051 }
2052 if (test_bit(S390_FEAT_MSA, features)) {
2053 s390_fill_feat_block(features, S390_FEAT_TYPE_KMAC, prop.kmac);
2054 s390_fill_feat_block(features, S390_FEAT_TYPE_KMC, prop.kmc);
2055 s390_fill_feat_block(features, S390_FEAT_TYPE_KM, prop.km);
2056 s390_fill_feat_block(features, S390_FEAT_TYPE_KIMD, prop.kimd);
2057 s390_fill_feat_block(features, S390_FEAT_TYPE_KLMD, prop.klmd);
2058 }
2059 if (test_bit(S390_FEAT_MSA_EXT_3, features)) {
2060 s390_fill_feat_block(features, S390_FEAT_TYPE_PCKMO, prop.pckmo);
2061 }
2062 if (test_bit(S390_FEAT_MSA_EXT_4, features)) {
2063 s390_fill_feat_block(features, S390_FEAT_TYPE_KMCTR, prop.kmctr);
2064 s390_fill_feat_block(features, S390_FEAT_TYPE_KMF, prop.kmf);
2065 s390_fill_feat_block(features, S390_FEAT_TYPE_KMO, prop.kmo);
2066 s390_fill_feat_block(features, S390_FEAT_TYPE_PCC, prop.pcc);
2067 }
2068 if (test_bit(S390_FEAT_MSA_EXT_5, features)) {
2069 s390_fill_feat_block(features, S390_FEAT_TYPE_PPNO, prop.ppno);
2070 }
2071 if (test_bit(S390_FEAT_MSA_EXT_8, features)) {
2072 s390_fill_feat_block(features, S390_FEAT_TYPE_KMA, prop.kma);
2073 }
2074 return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2075 }
2076
2077 static int kvm_to_feat[][2] = {
2078 { KVM_S390_VM_CPU_FEAT_ESOP, S390_FEAT_ESOP },
2079 { KVM_S390_VM_CPU_FEAT_SIEF2, S390_FEAT_SIE_F2 },
2080 { KVM_S390_VM_CPU_FEAT_64BSCAO , S390_FEAT_SIE_64BSCAO },
2081 { KVM_S390_VM_CPU_FEAT_SIIF, S390_FEAT_SIE_SIIF },
2082 { KVM_S390_VM_CPU_FEAT_GPERE, S390_FEAT_SIE_GPERE },
2083 { KVM_S390_VM_CPU_FEAT_GSLS, S390_FEAT_SIE_GSLS },
2084 { KVM_S390_VM_CPU_FEAT_IB, S390_FEAT_SIE_IB },
2085 { KVM_S390_VM_CPU_FEAT_CEI, S390_FEAT_SIE_CEI },
2086 { KVM_S390_VM_CPU_FEAT_IBS, S390_FEAT_SIE_IBS },
2087 { KVM_S390_VM_CPU_FEAT_SKEY, S390_FEAT_SIE_SKEY },
2088 { KVM_S390_VM_CPU_FEAT_CMMA, S390_FEAT_SIE_CMMA },
2089 { KVM_S390_VM_CPU_FEAT_PFMFI, S390_FEAT_SIE_PFMFI},
2090 { KVM_S390_VM_CPU_FEAT_SIGPIF, S390_FEAT_SIE_SIGPIF},
2091 { KVM_S390_VM_CPU_FEAT_KSS, S390_FEAT_SIE_KSS},
2092 };
2093
2094 static int query_cpu_feat(S390FeatBitmap features)
2095 {
2096 struct kvm_s390_vm_cpu_feat prop;
2097 struct kvm_device_attr attr = {
2098 .group = KVM_S390_VM_CPU_MODEL,
2099 .attr = KVM_S390_VM_CPU_MACHINE_FEAT,
2100 .addr = (uint64_t) &prop,
2101 };
2102 int rc;
2103 int i;
2104
2105 rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2106 if (rc) {
2107 return rc;
2108 }
2109
2110 for (i = 0; i < ARRAY_SIZE(kvm_to_feat); i++) {
2111 if (test_be_bit(kvm_to_feat[i][0], (uint8_t *) prop.feat)) {
2112 set_bit(kvm_to_feat[i][1], features);
2113 }
2114 }
2115 return 0;
2116 }
2117
2118 static int configure_cpu_feat(const S390FeatBitmap features)
2119 {
2120 struct kvm_s390_vm_cpu_feat prop = {};
2121 struct kvm_device_attr attr = {
2122 .group = KVM_S390_VM_CPU_MODEL,
2123 .attr = KVM_S390_VM_CPU_PROCESSOR_FEAT,
2124 .addr = (uint64_t) &prop,
2125 };
2126 int i;
2127
2128 for (i = 0; i < ARRAY_SIZE(kvm_to_feat); i++) {
2129 if (test_bit(kvm_to_feat[i][1], features)) {
2130 set_be_bit(kvm_to_feat[i][0], (uint8_t *) prop.feat);
2131 }
2132 }
2133 return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2134 }
2135
2136 bool kvm_s390_cpu_models_supported(void)
2137 {
2138 if (!cpu_model_allowed()) {
2139 /* compatibility machines interfere with the cpu model */
2140 return false;
2141 }
2142 return kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2143 KVM_S390_VM_CPU_MACHINE) &&
2144 kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2145 KVM_S390_VM_CPU_PROCESSOR) &&
2146 kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2147 KVM_S390_VM_CPU_MACHINE_FEAT) &&
2148 kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2149 KVM_S390_VM_CPU_PROCESSOR_FEAT) &&
2150 kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2151 KVM_S390_VM_CPU_MACHINE_SUBFUNC);
2152 }
2153
2154 void kvm_s390_get_host_cpu_model(S390CPUModel *model, Error **errp)
2155 {
2156 struct kvm_s390_vm_cpu_machine prop = {};
2157 struct kvm_device_attr attr = {
2158 .group = KVM_S390_VM_CPU_MODEL,
2159 .attr = KVM_S390_VM_CPU_MACHINE,
2160 .addr = (uint64_t) &prop,
2161 };
2162 uint16_t unblocked_ibc = 0, cpu_type = 0;
2163 int rc;
2164
2165 memset(model, 0, sizeof(*model));
2166
2167 if (!kvm_s390_cpu_models_supported()) {
2168 error_setg(errp, "KVM doesn't support CPU models");
2169 return;
2170 }
2171
2172 /* query the basic cpu model properties */
2173 rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2174 if (rc) {
2175 error_setg(errp, "KVM: Error querying host CPU model: %d", rc);
2176 return;
2177 }
2178
2179 cpu_type = cpuid_type(prop.cpuid);
2180 if (has_ibc(prop.ibc)) {
2181 model->lowest_ibc = lowest_ibc(prop.ibc);
2182 unblocked_ibc = unblocked_ibc(prop.ibc);
2183 }
2184 model->cpu_id = cpuid_id(prop.cpuid);
2185 model->cpu_id_format = cpuid_format(prop.cpuid);
2186 model->cpu_ver = 0xff;
2187
2188 /* get supported cpu features indicated via STFL(E) */
2189 s390_add_from_feat_block(model->features, S390_FEAT_TYPE_STFL,
2190 (uint8_t *) prop.fac_mask);
2191 /* dat-enhancement facility 2 has no bit but was introduced with stfle */
2192 if (test_bit(S390_FEAT_STFLE, model->features)) {
2193 set_bit(S390_FEAT_DAT_ENH_2, model->features);
2194 }
2195 /* get supported cpu features indicated e.g. via SCLP */
2196 rc = query_cpu_feat(model->features);
2197 if (rc) {
2198 error_setg(errp, "KVM: Error querying CPU features: %d", rc);
2199 return;
2200 }
2201 /* get supported cpu subfunctions indicated via query / test bit */
2202 rc = query_cpu_subfunc(model->features);
2203 if (rc) {
2204 error_setg(errp, "KVM: Error querying CPU subfunctions: %d", rc);
2205 return;
2206 }
2207
2208 /* PTFF subfunctions might be indicated although kernel support missing */
2209 if (!test_bit(S390_FEAT_MULTIPLE_EPOCH, model->features)) {
2210 clear_bit(S390_FEAT_PTFF_QSIE, model->features);
2211 clear_bit(S390_FEAT_PTFF_QTOUE, model->features);
2212 clear_bit(S390_FEAT_PTFF_STOE, model->features);
2213 clear_bit(S390_FEAT_PTFF_STOUE, model->features);
2214 }
2215
2216 /* with cpu model support, CMM is only indicated if really available */
2217 if (kvm_s390_cmma_available()) {
2218 set_bit(S390_FEAT_CMM, model->features);
2219 } else {
2220 /* no cmm -> no cmm nt */
2221 clear_bit(S390_FEAT_CMM_NT, model->features);
2222 }
2223
2224 /* bpb needs kernel support for migration, VSIE and reset */
2225 if (!kvm_check_extension(kvm_state, KVM_CAP_S390_BPB)) {
2226 clear_bit(S390_FEAT_BPB, model->features);
2227 }
2228
2229 /* We emulate a zPCI bus and AEN, therefore we don't need HW support */
2230 if (pci_available) {
2231 set_bit(S390_FEAT_ZPCI, model->features);
2232 }
2233 set_bit(S390_FEAT_ADAPTER_EVENT_NOTIFICATION, model->features);
2234
2235 if (s390_known_cpu_type(cpu_type)) {
2236 /* we want the exact model, even if some features are missing */
2237 model->def = s390_find_cpu_def(cpu_type, ibc_gen(unblocked_ibc),
2238 ibc_ec_ga(unblocked_ibc), NULL);
2239 } else {
2240 /* model unknown, e.g. too new - search using features */
2241 model->def = s390_find_cpu_def(0, ibc_gen(unblocked_ibc),
2242 ibc_ec_ga(unblocked_ibc),
2243 model->features);
2244 }
2245 if (!model->def) {
2246 error_setg(errp, "KVM: host CPU model could not be identified");
2247 return;
2248 }
2249 /* strip of features that are not part of the maximum model */
2250 bitmap_and(model->features, model->features, model->def->full_feat,
2251 S390_FEAT_MAX);
2252 }
2253
2254 void kvm_s390_apply_cpu_model(const S390CPUModel *model, Error **errp)
2255 {
2256 struct kvm_s390_vm_cpu_processor prop = {
2257 .fac_list = { 0 },
2258 };
2259 struct kvm_device_attr attr = {
2260 .group = KVM_S390_VM_CPU_MODEL,
2261 .attr = KVM_S390_VM_CPU_PROCESSOR,
2262 .addr = (uint64_t) &prop,
2263 };
2264 int rc;
2265
2266 if (!model) {
2267 /* compatibility handling if cpu models are disabled */
2268 if (kvm_s390_cmma_available()) {
2269 kvm_s390_enable_cmma();
2270 }
2271 return;
2272 }
2273 if (!kvm_s390_cpu_models_supported()) {
2274 error_setg(errp, "KVM doesn't support CPU models");
2275 return;
2276 }
2277 prop.cpuid = s390_cpuid_from_cpu_model(model);
2278 prop.ibc = s390_ibc_from_cpu_model(model);
2279 /* configure cpu features indicated via STFL(e) */
2280 s390_fill_feat_block(model->features, S390_FEAT_TYPE_STFL,
2281 (uint8_t *) prop.fac_list);
2282 rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2283 if (rc) {
2284 error_setg(errp, "KVM: Error configuring the CPU model: %d", rc);
2285 return;
2286 }
2287 /* configure cpu features indicated e.g. via SCLP */
2288 rc = configure_cpu_feat(model->features);
2289 if (rc) {
2290 error_setg(errp, "KVM: Error configuring CPU features: %d", rc);
2291 return;
2292 }
2293 /* configure cpu subfunctions indicated via query / test bit */
2294 rc = configure_cpu_subfunc(model->features);
2295 if (rc) {
2296 error_setg(errp, "KVM: Error configuring CPU subfunctions: %d", rc);
2297 return;
2298 }
2299 /* enable CMM via CMMA */
2300 if (test_bit(S390_FEAT_CMM, model->features)) {
2301 kvm_s390_enable_cmma();
2302 }
2303 }
2304
2305 void kvm_s390_restart_interrupt(S390CPU *cpu)
2306 {
2307 struct kvm_s390_irq irq = {
2308 .type = KVM_S390_RESTART,
2309 };
2310
2311 kvm_s390_vcpu_interrupt(cpu, &irq);
2312 }
2313
2314 void kvm_s390_stop_interrupt(S390CPU *cpu)
2315 {
2316 struct kvm_s390_irq irq = {
2317 .type = KVM_S390_SIGP_STOP,
2318 };
2319
2320 kvm_s390_vcpu_interrupt(cpu, &irq);
2321 }