xref: /qemu/target/s390x/kvm/kvm.c (revision 947a38bd6f1301a97f3c9c7485e7fa118bbc40cf)
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 "qemu/error-report.h"
35 #include "qemu/timer.h"
36 #include "sysemu/sysemu.h"
37 #include "sysemu/hw_accel.h"
38 #include "hw/hw.h"
39 #include "sysemu/device_tree.h"
40 #include "qapi/qmp/qjson.h"
41 #include "exec/gdbstub.h"
42 #include "exec/address-spaces.h"
43 #include "trace.h"
44 #include "qapi-event.h"
45 #include "hw/s390x/s390-pci-inst.h"
46 #include "hw/s390x/s390-pci-bus.h"
47 #include "hw/s390x/ipl.h"
48 #include "hw/s390x/ebcdic.h"
49 #include "exec/memattrs.h"
50 #include "hw/s390x/s390-virtio-ccw.h"
51 #include "hw/s390x/s390-virtio-hcall.h"
52 
53 #ifndef DEBUG_KVM
54 #define DEBUG_KVM  0
55 #endif
56 
57 #define DPRINTF(fmt, ...) do {                \
58     if (DEBUG_KVM) {                          \
59         fprintf(stderr, fmt, ## __VA_ARGS__); \
60     }                                         \
61 } while (0);
62 
63 #define kvm_vm_check_mem_attr(s, attr) \
64     kvm_vm_check_attr(s, KVM_S390_VM_MEM_CTRL, attr)
65 
66 #define IPA0_DIAG                       0x8300
67 #define IPA0_SIGP                       0xae00
68 #define IPA0_B2                         0xb200
69 #define IPA0_B9                         0xb900
70 #define IPA0_EB                         0xeb00
71 #define IPA0_E3                         0xe300
72 
73 #define PRIV_B2_SCLP_CALL               0x20
74 #define PRIV_B2_CSCH                    0x30
75 #define PRIV_B2_HSCH                    0x31
76 #define PRIV_B2_MSCH                    0x32
77 #define PRIV_B2_SSCH                    0x33
78 #define PRIV_B2_STSCH                   0x34
79 #define PRIV_B2_TSCH                    0x35
80 #define PRIV_B2_TPI                     0x36
81 #define PRIV_B2_SAL                     0x37
82 #define PRIV_B2_RSCH                    0x38
83 #define PRIV_B2_STCRW                   0x39
84 #define PRIV_B2_STCPS                   0x3a
85 #define PRIV_B2_RCHP                    0x3b
86 #define PRIV_B2_SCHM                    0x3c
87 #define PRIV_B2_CHSC                    0x5f
88 #define PRIV_B2_SIGA                    0x74
89 #define PRIV_B2_XSCH                    0x76
90 
91 #define PRIV_EB_SQBS                    0x8a
92 #define PRIV_EB_PCISTB                  0xd0
93 #define PRIV_EB_SIC                     0xd1
94 
95 #define PRIV_B9_EQBS                    0x9c
96 #define PRIV_B9_CLP                     0xa0
97 #define PRIV_B9_PCISTG                  0xd0
98 #define PRIV_B9_PCILG                   0xd2
99 #define PRIV_B9_RPCIT                   0xd3
100 
101 #define PRIV_E3_MPCIFC                  0xd0
102 #define PRIV_E3_STPCIFC                 0xd4
103 
104 #define DIAG_TIMEREVENT                 0x288
105 #define DIAG_IPL                        0x308
106 #define DIAG_KVM_HYPERCALL              0x500
107 #define DIAG_KVM_BREAKPOINT             0x501
108 
109 #define ICPT_INSTRUCTION                0x04
110 #define ICPT_PROGRAM                    0x08
111 #define ICPT_EXT_INT                    0x14
112 #define ICPT_WAITPSW                    0x1c
113 #define ICPT_SOFT_INTERCEPT             0x24
114 #define ICPT_CPU_STOP                   0x28
115 #define ICPT_OPEREXC                    0x2c
116 #define ICPT_IO                         0x40
117 
118 #define NR_LOCAL_IRQS 32
119 /*
120  * Needs to be big enough to contain max_cpus emergency signals
121  * and in addition NR_LOCAL_IRQS interrupts
122  */
123 #define VCPU_IRQ_BUF_SIZE (sizeof(struct kvm_s390_irq) * \
124                            (max_cpus + NR_LOCAL_IRQS))
125 
126 static CPUWatchpoint hw_watchpoint;
127 /*
128  * We don't use a list because this structure is also used to transmit the
129  * hardware breakpoints to the kernel.
130  */
131 static struct kvm_hw_breakpoint *hw_breakpoints;
132 static int nb_hw_breakpoints;
133 
134 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
135     KVM_CAP_LAST_INFO
136 };
137 
138 static QemuMutex qemu_sigp_mutex;
139 
140 static int cap_sync_regs;
141 static int cap_async_pf;
142 static int cap_mem_op;
143 static int cap_s390_irq;
144 static int cap_ri;
145 static int cap_gs;
146 
147 static int active_cmma;
148 
149 static void *legacy_s390_alloc(size_t size, uint64_t *align);
150 
151 static int kvm_s390_query_mem_limit(uint64_t *memory_limit)
152 {
153     struct kvm_device_attr attr = {
154         .group = KVM_S390_VM_MEM_CTRL,
155         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
156         .addr = (uint64_t) memory_limit,
157     };
158 
159     return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
160 }
161 
162 int kvm_s390_set_mem_limit(uint64_t new_limit, uint64_t *hw_limit)
163 {
164     int rc;
165 
166     struct kvm_device_attr attr = {
167         .group = KVM_S390_VM_MEM_CTRL,
168         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
169         .addr = (uint64_t) &new_limit,
170     };
171 
172     if (!kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_LIMIT_SIZE)) {
173         return 0;
174     }
175 
176     rc = kvm_s390_query_mem_limit(hw_limit);
177     if (rc) {
178         return rc;
179     } else if (*hw_limit < new_limit) {
180         return -E2BIG;
181     }
182 
183     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
184 }
185 
186 int kvm_s390_cmma_active(void)
187 {
188     return active_cmma;
189 }
190 
191 static bool kvm_s390_cmma_available(void)
192 {
193     static bool initialized, value;
194 
195     if (!initialized) {
196         initialized = true;
197         value = kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_ENABLE_CMMA) &&
198                 kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_CLR_CMMA);
199     }
200     return value;
201 }
202 
203 void kvm_s390_cmma_reset(void)
204 {
205     int rc;
206     struct kvm_device_attr attr = {
207         .group = KVM_S390_VM_MEM_CTRL,
208         .attr = KVM_S390_VM_MEM_CLR_CMMA,
209     };
210 
211     if (!kvm_s390_cmma_active()) {
212         return;
213     }
214 
215     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
216     trace_kvm_clear_cmma(rc);
217 }
218 
219 static void kvm_s390_enable_cmma(void)
220 {
221     int rc;
222     struct kvm_device_attr attr = {
223         .group = KVM_S390_VM_MEM_CTRL,
224         .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
225     };
226 
227     if (mem_path) {
228         warn_report("CMM will not be enabled because it is not "
229                     "compatible with hugetlbfs.");
230         return;
231     }
232     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
233     active_cmma = !rc;
234     trace_kvm_enable_cmma(rc);
235 }
236 
237 static void kvm_s390_set_attr(uint64_t attr)
238 {
239     struct kvm_device_attr attribute = {
240         .group = KVM_S390_VM_CRYPTO,
241         .attr  = attr,
242     };
243 
244     int ret = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attribute);
245 
246     if (ret) {
247         error_report("Failed to set crypto device attribute %lu: %s",
248                      attr, strerror(-ret));
249     }
250 }
251 
252 static void kvm_s390_init_aes_kw(void)
253 {
254     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_AES_KW;
255 
256     if (object_property_get_bool(OBJECT(qdev_get_machine()), "aes-key-wrap",
257                                  NULL)) {
258             attr = KVM_S390_VM_CRYPTO_ENABLE_AES_KW;
259     }
260 
261     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
262             kvm_s390_set_attr(attr);
263     }
264 }
265 
266 static void kvm_s390_init_dea_kw(void)
267 {
268     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_DEA_KW;
269 
270     if (object_property_get_bool(OBJECT(qdev_get_machine()), "dea-key-wrap",
271                                  NULL)) {
272             attr = KVM_S390_VM_CRYPTO_ENABLE_DEA_KW;
273     }
274 
275     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
276             kvm_s390_set_attr(attr);
277     }
278 }
279 
280 void kvm_s390_crypto_reset(void)
281 {
282     if (s390_has_feat(S390_FEAT_MSA_EXT_3)) {
283         kvm_s390_init_aes_kw();
284         kvm_s390_init_dea_kw();
285     }
286 }
287 
288 int kvm_arch_init(MachineState *ms, KVMState *s)
289 {
290     MachineClass *mc = MACHINE_GET_CLASS(ms);
291 
292     mc->default_cpu_type = S390_CPU_TYPE_NAME("host");
293     cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
294     cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
295     cap_mem_op = kvm_check_extension(s, KVM_CAP_S390_MEM_OP);
296     cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
297 
298     if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
299         || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
300         phys_mem_set_alloc(legacy_s390_alloc);
301     }
302 
303     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_SIGP, 0);
304     kvm_vm_enable_cap(s, KVM_CAP_S390_VECTOR_REGISTERS, 0);
305     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_STSI, 0);
306     if (ri_allowed()) {
307         if (kvm_vm_enable_cap(s, KVM_CAP_S390_RI, 0) == 0) {
308             cap_ri = 1;
309         }
310     }
311     if (gs_allowed()) {
312         if (kvm_vm_enable_cap(s, KVM_CAP_S390_GS, 0) == 0) {
313             cap_gs = 1;
314         }
315     }
316 
317     /*
318      * The migration interface for ais was introduced with kernel 4.13
319      * but the capability itself had been active since 4.12. As migration
320      * support is considered necessary let's disable ais in the 2.10
321      * machine.
322      */
323     /* kvm_vm_enable_cap(s, KVM_CAP_S390_AIS, 0); */
324 
325     qemu_mutex_init(&qemu_sigp_mutex);
326 
327     return 0;
328 }
329 
330 int kvm_arch_irqchip_create(MachineState *ms, KVMState *s)
331 {
332     return 0;
333 }
334 
335 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
336 {
337     return cpu->cpu_index;
338 }
339 
340 int kvm_arch_init_vcpu(CPUState *cs)
341 {
342     S390CPU *cpu = S390_CPU(cs);
343     kvm_s390_set_cpu_state(cpu, cpu->env.cpu_state);
344     cpu->irqstate = g_malloc0(VCPU_IRQ_BUF_SIZE);
345     return 0;
346 }
347 
348 void kvm_s390_reset_vcpu(S390CPU *cpu)
349 {
350     CPUState *cs = CPU(cpu);
351 
352     /* The initial reset call is needed here to reset in-kernel
353      * vcpu data that we can't access directly from QEMU
354      * (i.e. with older kernels which don't support sync_regs/ONE_REG).
355      * Before this ioctl cpu_synchronize_state() is called in common kvm
356      * code (kvm-all) */
357     if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
358         error_report("Initial CPU reset failed on CPU %i", cs->cpu_index);
359     }
360 }
361 
362 static int can_sync_regs(CPUState *cs, int regs)
363 {
364     return cap_sync_regs && (cs->kvm_run->kvm_valid_regs & regs) == regs;
365 }
366 
367 int kvm_arch_put_registers(CPUState *cs, int level)
368 {
369     S390CPU *cpu = S390_CPU(cs);
370     CPUS390XState *env = &cpu->env;
371     struct kvm_sregs sregs;
372     struct kvm_regs regs;
373     struct kvm_fpu fpu = {};
374     int r;
375     int i;
376 
377     /* always save the PSW  and the GPRS*/
378     cs->kvm_run->psw_addr = env->psw.addr;
379     cs->kvm_run->psw_mask = env->psw.mask;
380 
381     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
382         for (i = 0; i < 16; i++) {
383             cs->kvm_run->s.regs.gprs[i] = env->regs[i];
384             cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
385         }
386     } else {
387         for (i = 0; i < 16; i++) {
388             regs.gprs[i] = env->regs[i];
389         }
390         r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
391         if (r < 0) {
392             return r;
393         }
394     }
395 
396     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
397         for (i = 0; i < 32; i++) {
398             cs->kvm_run->s.regs.vrs[i][0] = env->vregs[i][0].ll;
399             cs->kvm_run->s.regs.vrs[i][1] = env->vregs[i][1].ll;
400         }
401         cs->kvm_run->s.regs.fpc = env->fpc;
402         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_VRS;
403     } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
404         for (i = 0; i < 16; i++) {
405             cs->kvm_run->s.regs.fprs[i] = get_freg(env, i)->ll;
406         }
407         cs->kvm_run->s.regs.fpc = env->fpc;
408         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_FPRS;
409     } else {
410         /* Floating point */
411         for (i = 0; i < 16; i++) {
412             fpu.fprs[i] = get_freg(env, i)->ll;
413         }
414         fpu.fpc = env->fpc;
415 
416         r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
417         if (r < 0) {
418             return r;
419         }
420     }
421 
422     /* Do we need to save more than that? */
423     if (level == KVM_PUT_RUNTIME_STATE) {
424         return 0;
425     }
426 
427     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
428         cs->kvm_run->s.regs.cputm = env->cputm;
429         cs->kvm_run->s.regs.ckc = env->ckc;
430         cs->kvm_run->s.regs.todpr = env->todpr;
431         cs->kvm_run->s.regs.gbea = env->gbea;
432         cs->kvm_run->s.regs.pp = env->pp;
433         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ARCH0;
434     } else {
435         /*
436          * These ONE_REGS are not protected by a capability. As they are only
437          * necessary for migration we just trace a possible error, but don't
438          * return with an error return code.
439          */
440         kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
441         kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
442         kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
443         kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
444         kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
445     }
446 
447     if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
448         memcpy(cs->kvm_run->s.regs.riccb, env->riccb, 64);
449         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_RICCB;
450     }
451 
452     /* pfault parameters */
453     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
454         cs->kvm_run->s.regs.pft = env->pfault_token;
455         cs->kvm_run->s.regs.pfs = env->pfault_select;
456         cs->kvm_run->s.regs.pfc = env->pfault_compare;
457         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PFAULT;
458     } else if (cap_async_pf) {
459         r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
460         if (r < 0) {
461             return r;
462         }
463         r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
464         if (r < 0) {
465             return r;
466         }
467         r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
468         if (r < 0) {
469             return r;
470         }
471     }
472 
473     /* access registers and control registers*/
474     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
475         for (i = 0; i < 16; i++) {
476             cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
477             cs->kvm_run->s.regs.crs[i] = env->cregs[i];
478         }
479         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
480         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
481     } else {
482         for (i = 0; i < 16; i++) {
483             sregs.acrs[i] = env->aregs[i];
484             sregs.crs[i] = env->cregs[i];
485         }
486         r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
487         if (r < 0) {
488             return r;
489         }
490     }
491 
492     if (can_sync_regs(cs, KVM_SYNC_GSCB)) {
493         memcpy(cs->kvm_run->s.regs.gscb, env->gscb, 32);
494         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GSCB;
495     }
496 
497     /* Finally the prefix */
498     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
499         cs->kvm_run->s.regs.prefix = env->psa;
500         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
501     } else {
502         /* prefix is only supported via sync regs */
503     }
504     return 0;
505 }
506 
507 int kvm_arch_get_registers(CPUState *cs)
508 {
509     S390CPU *cpu = S390_CPU(cs);
510     CPUS390XState *env = &cpu->env;
511     struct kvm_sregs sregs;
512     struct kvm_regs regs;
513     struct kvm_fpu fpu;
514     int i, r;
515 
516     /* get the PSW */
517     env->psw.addr = cs->kvm_run->psw_addr;
518     env->psw.mask = cs->kvm_run->psw_mask;
519 
520     /* the GPRS */
521     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
522         for (i = 0; i < 16; i++) {
523             env->regs[i] = cs->kvm_run->s.regs.gprs[i];
524         }
525     } else {
526         r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
527         if (r < 0) {
528             return r;
529         }
530          for (i = 0; i < 16; i++) {
531             env->regs[i] = regs.gprs[i];
532         }
533     }
534 
535     /* The ACRS and CRS */
536     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
537         for (i = 0; i < 16; i++) {
538             env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
539             env->cregs[i] = cs->kvm_run->s.regs.crs[i];
540         }
541     } else {
542         r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
543         if (r < 0) {
544             return r;
545         }
546          for (i = 0; i < 16; i++) {
547             env->aregs[i] = sregs.acrs[i];
548             env->cregs[i] = sregs.crs[i];
549         }
550     }
551 
552     /* Floating point and vector registers */
553     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
554         for (i = 0; i < 32; i++) {
555             env->vregs[i][0].ll = cs->kvm_run->s.regs.vrs[i][0];
556             env->vregs[i][1].ll = cs->kvm_run->s.regs.vrs[i][1];
557         }
558         env->fpc = cs->kvm_run->s.regs.fpc;
559     } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
560         for (i = 0; i < 16; i++) {
561             get_freg(env, i)->ll = cs->kvm_run->s.regs.fprs[i];
562         }
563         env->fpc = cs->kvm_run->s.regs.fpc;
564     } else {
565         r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
566         if (r < 0) {
567             return r;
568         }
569         for (i = 0; i < 16; i++) {
570             get_freg(env, i)->ll = fpu.fprs[i];
571         }
572         env->fpc = fpu.fpc;
573     }
574 
575     /* The prefix */
576     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
577         env->psa = cs->kvm_run->s.regs.prefix;
578     }
579 
580     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
581         env->cputm = cs->kvm_run->s.regs.cputm;
582         env->ckc = cs->kvm_run->s.regs.ckc;
583         env->todpr = cs->kvm_run->s.regs.todpr;
584         env->gbea = cs->kvm_run->s.regs.gbea;
585         env->pp = cs->kvm_run->s.regs.pp;
586     } else {
587         /*
588          * These ONE_REGS are not protected by a capability. As they are only
589          * necessary for migration we just trace a possible error, but don't
590          * return with an error return code.
591          */
592         kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
593         kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
594         kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
595         kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
596         kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
597     }
598 
599     if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
600         memcpy(env->riccb, cs->kvm_run->s.regs.riccb, 64);
601     }
602 
603     if (can_sync_regs(cs, KVM_SYNC_GSCB)) {
604         memcpy(env->gscb, cs->kvm_run->s.regs.gscb, 32);
605     }
606 
607     /* pfault parameters */
608     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
609         env->pfault_token = cs->kvm_run->s.regs.pft;
610         env->pfault_select = cs->kvm_run->s.regs.pfs;
611         env->pfault_compare = cs->kvm_run->s.regs.pfc;
612     } else if (cap_async_pf) {
613         r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
614         if (r < 0) {
615             return r;
616         }
617         r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
618         if (r < 0) {
619             return r;
620         }
621         r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
622         if (r < 0) {
623             return r;
624         }
625     }
626 
627     return 0;
628 }
629 
630 int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_low)
631 {
632     int r;
633     struct kvm_device_attr attr = {
634         .group = KVM_S390_VM_TOD,
635         .attr = KVM_S390_VM_TOD_LOW,
636         .addr = (uint64_t)tod_low,
637     };
638 
639     r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
640     if (r) {
641         return r;
642     }
643 
644     attr.attr = KVM_S390_VM_TOD_HIGH;
645     attr.addr = (uint64_t)tod_high;
646     return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
647 }
648 
649 int kvm_s390_set_clock(uint8_t *tod_high, uint64_t *tod_low)
650 {
651     int r;
652 
653     struct kvm_device_attr attr = {
654         .group = KVM_S390_VM_TOD,
655         .attr = KVM_S390_VM_TOD_LOW,
656         .addr = (uint64_t)tod_low,
657     };
658 
659     r = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
660     if (r) {
661         return r;
662     }
663 
664     attr.attr = KVM_S390_VM_TOD_HIGH;
665     attr.addr = (uint64_t)tod_high;
666     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
667 }
668 
669 /**
670  * kvm_s390_mem_op:
671  * @addr:      the logical start address in guest memory
672  * @ar:        the access register number
673  * @hostbuf:   buffer in host memory. NULL = do only checks w/o copying
674  * @len:       length that should be transferred
675  * @is_write:  true = write, false = read
676  * Returns:    0 on success, non-zero if an exception or error occurred
677  *
678  * Use KVM ioctl to read/write from/to guest memory. An access exception
679  * is injected into the vCPU in case of translation errors.
680  */
681 int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, uint8_t ar, void *hostbuf,
682                     int len, bool is_write)
683 {
684     struct kvm_s390_mem_op mem_op = {
685         .gaddr = addr,
686         .flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION,
687         .size = len,
688         .op = is_write ? KVM_S390_MEMOP_LOGICAL_WRITE
689                        : KVM_S390_MEMOP_LOGICAL_READ,
690         .buf = (uint64_t)hostbuf,
691         .ar = ar,
692     };
693     int ret;
694 
695     if (!cap_mem_op) {
696         return -ENOSYS;
697     }
698     if (!hostbuf) {
699         mem_op.flags |= KVM_S390_MEMOP_F_CHECK_ONLY;
700     }
701 
702     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
703     if (ret < 0) {
704         error_printf("KVM_S390_MEM_OP failed: %s\n", strerror(-ret));
705     }
706     return ret;
707 }
708 
709 /*
710  * Legacy layout for s390:
711  * Older S390 KVM requires the topmost vma of the RAM to be
712  * smaller than an system defined value, which is at least 256GB.
713  * Larger systems have larger values. We put the guest between
714  * the end of data segment (system break) and this value. We
715  * use 32GB as a base to have enough room for the system break
716  * to grow. We also have to use MAP parameters that avoid
717  * read-only mapping of guest pages.
718  */
719 static void *legacy_s390_alloc(size_t size, uint64_t *align)
720 {
721     void *mem;
722 
723     mem = mmap((void *) 0x800000000ULL, size,
724                PROT_EXEC|PROT_READ|PROT_WRITE,
725                MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
726     return mem == MAP_FAILED ? NULL : mem;
727 }
728 
729 static uint8_t const *sw_bp_inst;
730 static uint8_t sw_bp_ilen;
731 
732 static void determine_sw_breakpoint_instr(void)
733 {
734         /* DIAG 501 is used for sw breakpoints with old kernels */
735         static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
736         /* Instruction 0x0000 is used for sw breakpoints with recent kernels */
737         static const uint8_t instr_0x0000[] = {0x00, 0x00};
738 
739         if (sw_bp_inst) {
740             return;
741         }
742         if (kvm_vm_enable_cap(kvm_state, KVM_CAP_S390_USER_INSTR0, 0)) {
743             sw_bp_inst = diag_501;
744             sw_bp_ilen = sizeof(diag_501);
745             DPRINTF("KVM: will use 4-byte sw breakpoints.\n");
746         } else {
747             sw_bp_inst = instr_0x0000;
748             sw_bp_ilen = sizeof(instr_0x0000);
749             DPRINTF("KVM: will use 2-byte sw breakpoints.\n");
750         }
751 }
752 
753 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
754 {
755     determine_sw_breakpoint_instr();
756 
757     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
758                             sw_bp_ilen, 0) ||
759         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)sw_bp_inst, sw_bp_ilen, 1)) {
760         return -EINVAL;
761     }
762     return 0;
763 }
764 
765 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
766 {
767     uint8_t t[MAX_ILEN];
768 
769     if (cpu_memory_rw_debug(cs, bp->pc, t, sw_bp_ilen, 0)) {
770         return -EINVAL;
771     } else if (memcmp(t, sw_bp_inst, sw_bp_ilen)) {
772         return -EINVAL;
773     } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
774                                    sw_bp_ilen, 1)) {
775         return -EINVAL;
776     }
777 
778     return 0;
779 }
780 
781 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
782                                                     int len, int type)
783 {
784     int n;
785 
786     for (n = 0; n < nb_hw_breakpoints; n++) {
787         if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
788             (hw_breakpoints[n].len == len || len == -1)) {
789             return &hw_breakpoints[n];
790         }
791     }
792 
793     return NULL;
794 }
795 
796 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
797 {
798     int size;
799 
800     if (find_hw_breakpoint(addr, len, type)) {
801         return -EEXIST;
802     }
803 
804     size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
805 
806     if (!hw_breakpoints) {
807         nb_hw_breakpoints = 0;
808         hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
809     } else {
810         hw_breakpoints =
811             (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
812     }
813 
814     if (!hw_breakpoints) {
815         nb_hw_breakpoints = 0;
816         return -ENOMEM;
817     }
818 
819     hw_breakpoints[nb_hw_breakpoints].addr = addr;
820     hw_breakpoints[nb_hw_breakpoints].len = len;
821     hw_breakpoints[nb_hw_breakpoints].type = type;
822 
823     nb_hw_breakpoints++;
824 
825     return 0;
826 }
827 
828 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
829                                   target_ulong len, int type)
830 {
831     switch (type) {
832     case GDB_BREAKPOINT_HW:
833         type = KVM_HW_BP;
834         break;
835     case GDB_WATCHPOINT_WRITE:
836         if (len < 1) {
837             return -EINVAL;
838         }
839         type = KVM_HW_WP_WRITE;
840         break;
841     default:
842         return -ENOSYS;
843     }
844     return insert_hw_breakpoint(addr, len, type);
845 }
846 
847 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
848                                   target_ulong len, int type)
849 {
850     int size;
851     struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
852 
853     if (bp == NULL) {
854         return -ENOENT;
855     }
856 
857     nb_hw_breakpoints--;
858     if (nb_hw_breakpoints > 0) {
859         /*
860          * In order to trim the array, move the last element to the position to
861          * be removed - if necessary.
862          */
863         if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
864             *bp = hw_breakpoints[nb_hw_breakpoints];
865         }
866         size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
867         hw_breakpoints =
868              (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
869     } else {
870         g_free(hw_breakpoints);
871         hw_breakpoints = NULL;
872     }
873 
874     return 0;
875 }
876 
877 void kvm_arch_remove_all_hw_breakpoints(void)
878 {
879     nb_hw_breakpoints = 0;
880     g_free(hw_breakpoints);
881     hw_breakpoints = NULL;
882 }
883 
884 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
885 {
886     int i;
887 
888     if (nb_hw_breakpoints > 0) {
889         dbg->arch.nr_hw_bp = nb_hw_breakpoints;
890         dbg->arch.hw_bp = hw_breakpoints;
891 
892         for (i = 0; i < nb_hw_breakpoints; ++i) {
893             hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
894                                                        hw_breakpoints[i].addr);
895         }
896         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
897     } else {
898         dbg->arch.nr_hw_bp = 0;
899         dbg->arch.hw_bp = NULL;
900     }
901 }
902 
903 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
904 {
905 }
906 
907 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
908 {
909     return MEMTXATTRS_UNSPECIFIED;
910 }
911 
912 int kvm_arch_process_async_events(CPUState *cs)
913 {
914     return cs->halted;
915 }
916 
917 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
918                                      struct kvm_s390_interrupt *interrupt)
919 {
920     int r = 0;
921 
922     interrupt->type = irq->type;
923     switch (irq->type) {
924     case KVM_S390_INT_VIRTIO:
925         interrupt->parm = irq->u.ext.ext_params;
926         /* fall through */
927     case KVM_S390_INT_PFAULT_INIT:
928     case KVM_S390_INT_PFAULT_DONE:
929         interrupt->parm64 = irq->u.ext.ext_params2;
930         break;
931     case KVM_S390_PROGRAM_INT:
932         interrupt->parm = irq->u.pgm.code;
933         break;
934     case KVM_S390_SIGP_SET_PREFIX:
935         interrupt->parm = irq->u.prefix.address;
936         break;
937     case KVM_S390_INT_SERVICE:
938         interrupt->parm = irq->u.ext.ext_params;
939         break;
940     case KVM_S390_MCHK:
941         interrupt->parm = irq->u.mchk.cr14;
942         interrupt->parm64 = irq->u.mchk.mcic;
943         break;
944     case KVM_S390_INT_EXTERNAL_CALL:
945         interrupt->parm = irq->u.extcall.code;
946         break;
947     case KVM_S390_INT_EMERGENCY:
948         interrupt->parm = irq->u.emerg.code;
949         break;
950     case KVM_S390_SIGP_STOP:
951     case KVM_S390_RESTART:
952         break; /* These types have no parameters */
953     case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
954         interrupt->parm = irq->u.io.subchannel_id << 16;
955         interrupt->parm |= irq->u.io.subchannel_nr;
956         interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
957         interrupt->parm64 |= irq->u.io.io_int_word;
958         break;
959     default:
960         r = -EINVAL;
961         break;
962     }
963     return r;
964 }
965 
966 static void inject_vcpu_irq_legacy(CPUState *cs, struct kvm_s390_irq *irq)
967 {
968     struct kvm_s390_interrupt kvmint = {};
969     int r;
970 
971     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
972     if (r < 0) {
973         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
974         exit(1);
975     }
976 
977     r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
978     if (r < 0) {
979         fprintf(stderr, "KVM failed to inject interrupt\n");
980         exit(1);
981     }
982 }
983 
984 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
985 {
986     CPUState *cs = CPU(cpu);
987     int r;
988 
989     if (cap_s390_irq) {
990         r = kvm_vcpu_ioctl(cs, KVM_S390_IRQ, irq);
991         if (!r) {
992             return;
993         }
994         error_report("KVM failed to inject interrupt %llx", irq->type);
995         exit(1);
996     }
997 
998     inject_vcpu_irq_legacy(cs, irq);
999 }
1000 
1001 static void __kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
1002 {
1003     struct kvm_s390_interrupt kvmint = {};
1004     int r;
1005 
1006     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
1007     if (r < 0) {
1008         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
1009         exit(1);
1010     }
1011 
1012     r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
1013     if (r < 0) {
1014         fprintf(stderr, "KVM failed to inject interrupt\n");
1015         exit(1);
1016     }
1017 }
1018 
1019 void kvm_s390_floating_interrupt(struct kvm_s390_irq *irq)
1020 {
1021     static bool use_flic = true;
1022     int r;
1023 
1024     if (use_flic) {
1025         r = kvm_s390_inject_flic(irq);
1026         if (r == -ENOSYS) {
1027             use_flic = false;
1028         }
1029         if (!r) {
1030             return;
1031         }
1032     }
1033     __kvm_s390_floating_interrupt(irq);
1034 }
1035 
1036 void kvm_s390_service_interrupt(uint32_t parm)
1037 {
1038     struct kvm_s390_irq irq = {
1039         .type = KVM_S390_INT_SERVICE,
1040         .u.ext.ext_params = parm,
1041     };
1042 
1043     kvm_s390_floating_interrupt(&irq);
1044 }
1045 
1046 void kvm_s390_program_interrupt(S390CPU *cpu, uint16_t code)
1047 {
1048     struct kvm_s390_irq irq = {
1049         .type = KVM_S390_PROGRAM_INT,
1050         .u.pgm.code = code,
1051     };
1052 
1053     kvm_s390_vcpu_interrupt(cpu, &irq);
1054 }
1055 
1056 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code)
1057 {
1058     struct kvm_s390_irq irq = {
1059         .type = KVM_S390_PROGRAM_INT,
1060         .u.pgm.code = code,
1061         .u.pgm.trans_exc_code = te_code,
1062         .u.pgm.exc_access_id = te_code & 3,
1063     };
1064 
1065     kvm_s390_vcpu_interrupt(cpu, &irq);
1066 }
1067 
1068 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
1069                                  uint16_t ipbh0)
1070 {
1071     CPUS390XState *env = &cpu->env;
1072     uint64_t sccb;
1073     uint32_t code;
1074     int r = 0;
1075 
1076     cpu_synchronize_state(CPU(cpu));
1077     sccb = env->regs[ipbh0 & 0xf];
1078     code = env->regs[(ipbh0 & 0xf0) >> 4];
1079 
1080     r = sclp_service_call(env, sccb, code);
1081     if (r < 0) {
1082         kvm_s390_program_interrupt(cpu, -r);
1083     } else {
1084         setcc(cpu, r);
1085     }
1086 
1087     return 0;
1088 }
1089 
1090 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1091 {
1092     CPUS390XState *env = &cpu->env;
1093     int rc = 0;
1094     uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
1095 
1096     cpu_synchronize_state(CPU(cpu));
1097 
1098     switch (ipa1) {
1099     case PRIV_B2_XSCH:
1100         ioinst_handle_xsch(cpu, env->regs[1]);
1101         break;
1102     case PRIV_B2_CSCH:
1103         ioinst_handle_csch(cpu, env->regs[1]);
1104         break;
1105     case PRIV_B2_HSCH:
1106         ioinst_handle_hsch(cpu, env->regs[1]);
1107         break;
1108     case PRIV_B2_MSCH:
1109         ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb);
1110         break;
1111     case PRIV_B2_SSCH:
1112         ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb);
1113         break;
1114     case PRIV_B2_STCRW:
1115         ioinst_handle_stcrw(cpu, run->s390_sieic.ipb);
1116         break;
1117     case PRIV_B2_STSCH:
1118         ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb);
1119         break;
1120     case PRIV_B2_TSCH:
1121         /* We should only get tsch via KVM_EXIT_S390_TSCH. */
1122         fprintf(stderr, "Spurious tsch intercept\n");
1123         break;
1124     case PRIV_B2_CHSC:
1125         ioinst_handle_chsc(cpu, run->s390_sieic.ipb);
1126         break;
1127     case PRIV_B2_TPI:
1128         /* This should have been handled by kvm already. */
1129         fprintf(stderr, "Spurious tpi intercept\n");
1130         break;
1131     case PRIV_B2_SCHM:
1132         ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
1133                            run->s390_sieic.ipb);
1134         break;
1135     case PRIV_B2_RSCH:
1136         ioinst_handle_rsch(cpu, env->regs[1]);
1137         break;
1138     case PRIV_B2_RCHP:
1139         ioinst_handle_rchp(cpu, env->regs[1]);
1140         break;
1141     case PRIV_B2_STCPS:
1142         /* We do not provide this instruction, it is suppressed. */
1143         break;
1144     case PRIV_B2_SAL:
1145         ioinst_handle_sal(cpu, env->regs[1]);
1146         break;
1147     case PRIV_B2_SIGA:
1148         /* Not provided, set CC = 3 for subchannel not operational */
1149         setcc(cpu, 3);
1150         break;
1151     case PRIV_B2_SCLP_CALL:
1152         rc = kvm_sclp_service_call(cpu, run, ipbh0);
1153         break;
1154     default:
1155         rc = -1;
1156         DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
1157         break;
1158     }
1159 
1160     return rc;
1161 }
1162 
1163 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run,
1164                                   uint8_t *ar)
1165 {
1166     CPUS390XState *env = &cpu->env;
1167     uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
1168     uint32_t base2 = run->s390_sieic.ipb >> 28;
1169     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1170                      ((run->s390_sieic.ipb & 0xff00) << 4);
1171 
1172     if (disp2 & 0x80000) {
1173         disp2 += 0xfff00000;
1174     }
1175     if (ar) {
1176         *ar = base2;
1177     }
1178 
1179     return (base2 ? env->regs[base2] : 0) +
1180            (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
1181 }
1182 
1183 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run,
1184                                   uint8_t *ar)
1185 {
1186     CPUS390XState *env = &cpu->env;
1187     uint32_t base2 = run->s390_sieic.ipb >> 28;
1188     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1189                      ((run->s390_sieic.ipb & 0xff00) << 4);
1190 
1191     if (disp2 & 0x80000) {
1192         disp2 += 0xfff00000;
1193     }
1194     if (ar) {
1195         *ar = base2;
1196     }
1197 
1198     return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
1199 }
1200 
1201 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
1202 {
1203     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1204 
1205     if (s390_has_feat(S390_FEAT_ZPCI)) {
1206         return clp_service_call(cpu, r2);
1207     } else {
1208         return -1;
1209     }
1210 }
1211 
1212 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
1213 {
1214     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1215     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1216 
1217     if (s390_has_feat(S390_FEAT_ZPCI)) {
1218         return pcilg_service_call(cpu, r1, r2);
1219     } else {
1220         return -1;
1221     }
1222 }
1223 
1224 static int kvm_pcistg_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 pcistg_service_call(cpu, r1, r2);
1231     } else {
1232         return -1;
1233     }
1234 }
1235 
1236 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1237 {
1238     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1239     uint64_t fiba;
1240     uint8_t ar;
1241 
1242     if (s390_has_feat(S390_FEAT_ZPCI)) {
1243         cpu_synchronize_state(CPU(cpu));
1244         fiba = get_base_disp_rxy(cpu, run, &ar);
1245 
1246         return stpcifc_service_call(cpu, r1, fiba, ar);
1247     } else {
1248         return -1;
1249     }
1250 }
1251 
1252 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
1253 {
1254     CPUS390XState *env = &cpu->env;
1255     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1256     uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1257     uint8_t isc;
1258     uint16_t mode;
1259     int r;
1260 
1261     cpu_synchronize_state(CPU(cpu));
1262     mode = env->regs[r1] & 0xffff;
1263     isc = (env->regs[r3] >> 27) & 0x7;
1264     r = css_do_sic(env, isc, mode);
1265     if (r) {
1266         kvm_s390_program_interrupt(cpu, -r);
1267     }
1268 
1269     return 0;
1270 }
1271 
1272 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
1273 {
1274     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1275     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1276 
1277     if (s390_has_feat(S390_FEAT_ZPCI)) {
1278         return rpcit_service_call(cpu, r1, r2);
1279     } else {
1280         return -1;
1281     }
1282 }
1283 
1284 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
1285 {
1286     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1287     uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1288     uint64_t gaddr;
1289     uint8_t ar;
1290 
1291     if (s390_has_feat(S390_FEAT_ZPCI)) {
1292         cpu_synchronize_state(CPU(cpu));
1293         gaddr = get_base_disp_rsy(cpu, run, &ar);
1294 
1295         return pcistb_service_call(cpu, r1, r3, gaddr, ar);
1296     } else {
1297         return -1;
1298     }
1299 }
1300 
1301 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1302 {
1303     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1304     uint64_t fiba;
1305     uint8_t ar;
1306 
1307     if (s390_has_feat(S390_FEAT_ZPCI)) {
1308         cpu_synchronize_state(CPU(cpu));
1309         fiba = get_base_disp_rxy(cpu, run, &ar);
1310 
1311         return mpcifc_service_call(cpu, r1, fiba, ar);
1312     } else {
1313         return -1;
1314     }
1315 }
1316 
1317 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1318 {
1319     int r = 0;
1320 
1321     switch (ipa1) {
1322     case PRIV_B9_CLP:
1323         r = kvm_clp_service_call(cpu, run);
1324         break;
1325     case PRIV_B9_PCISTG:
1326         r = kvm_pcistg_service_call(cpu, run);
1327         break;
1328     case PRIV_B9_PCILG:
1329         r = kvm_pcilg_service_call(cpu, run);
1330         break;
1331     case PRIV_B9_RPCIT:
1332         r = kvm_rpcit_service_call(cpu, run);
1333         break;
1334     case PRIV_B9_EQBS:
1335         /* just inject exception */
1336         r = -1;
1337         break;
1338     default:
1339         r = -1;
1340         DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
1341         break;
1342     }
1343 
1344     return r;
1345 }
1346 
1347 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1348 {
1349     int r = 0;
1350 
1351     switch (ipbl) {
1352     case PRIV_EB_PCISTB:
1353         r = kvm_pcistb_service_call(cpu, run);
1354         break;
1355     case PRIV_EB_SIC:
1356         r = kvm_sic_service_call(cpu, run);
1357         break;
1358     case PRIV_EB_SQBS:
1359         /* just inject exception */
1360         r = -1;
1361         break;
1362     default:
1363         r = -1;
1364         DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1365         break;
1366     }
1367 
1368     return r;
1369 }
1370 
1371 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1372 {
1373     int r = 0;
1374 
1375     switch (ipbl) {
1376     case PRIV_E3_MPCIFC:
1377         r = kvm_mpcifc_service_call(cpu, run);
1378         break;
1379     case PRIV_E3_STPCIFC:
1380         r = kvm_stpcifc_service_call(cpu, run);
1381         break;
1382     default:
1383         r = -1;
1384         DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1385         break;
1386     }
1387 
1388     return r;
1389 }
1390 
1391 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1392 {
1393     CPUS390XState *env = &cpu->env;
1394     int ret;
1395 
1396     cpu_synchronize_state(CPU(cpu));
1397     ret = s390_virtio_hypercall(env);
1398     if (ret == -EINVAL) {
1399         kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1400         return 0;
1401     }
1402 
1403     return ret;
1404 }
1405 
1406 static void kvm_handle_diag_288(S390CPU *cpu, struct kvm_run *run)
1407 {
1408     uint64_t r1, r3;
1409     int rc;
1410 
1411     cpu_synchronize_state(CPU(cpu));
1412     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1413     r3 = run->s390_sieic.ipa & 0x000f;
1414     rc = handle_diag_288(&cpu->env, r1, r3);
1415     if (rc) {
1416         kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1417     }
1418 }
1419 
1420 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1421 {
1422     uint64_t r1, r3;
1423 
1424     cpu_synchronize_state(CPU(cpu));
1425     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1426     r3 = run->s390_sieic.ipa & 0x000f;
1427     handle_diag_308(&cpu->env, r1, r3);
1428 }
1429 
1430 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1431 {
1432     CPUS390XState *env = &cpu->env;
1433     unsigned long pc;
1434 
1435     cpu_synchronize_state(CPU(cpu));
1436 
1437     pc = env->psw.addr - sw_bp_ilen;
1438     if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1439         env->psw.addr = pc;
1440         return EXCP_DEBUG;
1441     }
1442 
1443     return -ENOENT;
1444 }
1445 
1446 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1447 
1448 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1449 {
1450     int r = 0;
1451     uint16_t func_code;
1452 
1453     /*
1454      * For any diagnose call we support, bits 48-63 of the resulting
1455      * address specify the function code; the remainder is ignored.
1456      */
1457     func_code = decode_basedisp_rs(&cpu->env, ipb, NULL) & DIAG_KVM_CODE_MASK;
1458     switch (func_code) {
1459     case DIAG_TIMEREVENT:
1460         kvm_handle_diag_288(cpu, run);
1461         break;
1462     case DIAG_IPL:
1463         kvm_handle_diag_308(cpu, run);
1464         break;
1465     case DIAG_KVM_HYPERCALL:
1466         r = handle_hypercall(cpu, run);
1467         break;
1468     case DIAG_KVM_BREAKPOINT:
1469         r = handle_sw_breakpoint(cpu, run);
1470         break;
1471     default:
1472         DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1473         kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1474         break;
1475     }
1476 
1477     return r;
1478 }
1479 
1480 typedef struct SigpInfo {
1481     uint64_t param;
1482     int cc;
1483     uint64_t *status_reg;
1484 } SigpInfo;
1485 
1486 static void set_sigp_status(SigpInfo *si, uint64_t status)
1487 {
1488     *si->status_reg &= 0xffffffff00000000ULL;
1489     *si->status_reg |= status;
1490     si->cc = SIGP_CC_STATUS_STORED;
1491 }
1492 
1493 static void sigp_start(CPUState *cs, run_on_cpu_data arg)
1494 {
1495     S390CPU *cpu = S390_CPU(cs);
1496     SigpInfo *si = arg.host_ptr;
1497 
1498     if (s390_cpu_get_state(cpu) != CPU_STATE_STOPPED) {
1499         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1500         return;
1501     }
1502 
1503     s390_cpu_set_state(CPU_STATE_OPERATING, cpu);
1504     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1505 }
1506 
1507 static void sigp_stop(CPUState *cs, run_on_cpu_data arg)
1508 {
1509     S390CPU *cpu = S390_CPU(cs);
1510     SigpInfo *si = arg.host_ptr;
1511     struct kvm_s390_irq irq = {
1512         .type = KVM_S390_SIGP_STOP,
1513     };
1514 
1515     if (s390_cpu_get_state(cpu) != CPU_STATE_OPERATING) {
1516         si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1517         return;
1518     }
1519 
1520     /* disabled wait - sleeping in user space */
1521     if (cs->halted) {
1522         s390_cpu_set_state(CPU_STATE_STOPPED, cpu);
1523     } else {
1524         /* execute the stop function */
1525         cpu->env.sigp_order = SIGP_STOP;
1526         kvm_s390_vcpu_interrupt(cpu, &irq);
1527     }
1528     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1529 }
1530 
1531 #define ADTL_GS_OFFSET   1024 /* offset of GS data in adtl save area */
1532 #define ADTL_GS_MIN_SIZE 2048 /* minimal size of adtl save area for GS */
1533 static int do_store_adtl_status(S390CPU *cpu, hwaddr addr, hwaddr len)
1534 {
1535     hwaddr save = len;
1536     void *mem;
1537 
1538     mem = cpu_physical_memory_map(addr, &save, 1);
1539     if (!mem) {
1540         return -EFAULT;
1541     }
1542     if (save != len) {
1543         cpu_physical_memory_unmap(mem, len, 1, 0);
1544         return -EFAULT;
1545     }
1546 
1547     if (s390_has_feat(S390_FEAT_VECTOR)) {
1548         memcpy(mem, &cpu->env.vregs, 512);
1549     }
1550     if (s390_has_feat(S390_FEAT_GUARDED_STORAGE) && len >= ADTL_GS_MIN_SIZE) {
1551         memcpy(mem + ADTL_GS_OFFSET, &cpu->env.gscb, 32);
1552     }
1553 
1554     cpu_physical_memory_unmap(mem, len, 1, len);
1555 
1556     return 0;
1557 }
1558 
1559 struct sigp_save_area {
1560     uint64_t    fprs[16];                       /* 0x0000 */
1561     uint64_t    grs[16];                        /* 0x0080 */
1562     PSW         psw;                            /* 0x0100 */
1563     uint8_t     pad_0x0110[0x0118 - 0x0110];    /* 0x0110 */
1564     uint32_t    prefix;                         /* 0x0118 */
1565     uint32_t    fpc;                            /* 0x011c */
1566     uint8_t     pad_0x0120[0x0124 - 0x0120];    /* 0x0120 */
1567     uint32_t    todpr;                          /* 0x0124 */
1568     uint64_t    cputm;                          /* 0x0128 */
1569     uint64_t    ckc;                            /* 0x0130 */
1570     uint8_t     pad_0x0138[0x0140 - 0x0138];    /* 0x0138 */
1571     uint32_t    ars[16];                        /* 0x0140 */
1572     uint64_t    crs[16];                        /* 0x0384 */
1573 };
1574 QEMU_BUILD_BUG_ON(sizeof(struct sigp_save_area) != 512);
1575 
1576 #define KVM_S390_STORE_STATUS_DEF_ADDR offsetof(LowCore, floating_pt_save_area)
1577 static int kvm_s390_store_status(S390CPU *cpu, hwaddr addr, bool store_arch)
1578 {
1579     static const uint8_t ar_id = 1;
1580     struct sigp_save_area *sa;
1581     hwaddr len = sizeof(*sa);
1582     int i;
1583 
1584     sa = cpu_physical_memory_map(addr, &len, 1);
1585     if (!sa) {
1586         return -EFAULT;
1587     }
1588     if (len != sizeof(*sa)) {
1589         cpu_physical_memory_unmap(sa, len, 1, 0);
1590         return -EFAULT;
1591     }
1592 
1593     if (store_arch) {
1594         cpu_physical_memory_write(offsetof(LowCore, ar_access_id), &ar_id, 1);
1595     }
1596     for (i = 0; i < 16; ++i) {
1597         sa->fprs[i] = cpu_to_be64(get_freg(&cpu->env, i)->ll);
1598     }
1599     for (i = 0; i < 16; ++i) {
1600         sa->grs[i] = cpu_to_be64(cpu->env.regs[i]);
1601     }
1602     sa->psw.addr = cpu_to_be64(cpu->env.psw.addr);
1603     sa->psw.mask = cpu_to_be64(get_psw_mask(&cpu->env));
1604     sa->prefix = cpu_to_be32(cpu->env.psa);
1605     sa->fpc = cpu_to_be32(cpu->env.fpc);
1606     sa->todpr = cpu_to_be32(cpu->env.todpr);
1607     sa->cputm = cpu_to_be64(cpu->env.cputm);
1608     sa->ckc = cpu_to_be64(cpu->env.ckc >> 8);
1609     for (i = 0; i < 16; ++i) {
1610         sa->ars[i] = cpu_to_be32(cpu->env.aregs[i]);
1611     }
1612     for (i = 0; i < 16; ++i) {
1613         sa->ars[i] = cpu_to_be64(cpu->env.cregs[i]);
1614     }
1615 
1616     cpu_physical_memory_unmap(sa, len, 1, len);
1617 
1618     return 0;
1619 }
1620 
1621 static void sigp_stop_and_store_status(CPUState *cs, run_on_cpu_data arg)
1622 {
1623     S390CPU *cpu = S390_CPU(cs);
1624     SigpInfo *si = arg.host_ptr;
1625     struct kvm_s390_irq irq = {
1626         .type = KVM_S390_SIGP_STOP,
1627     };
1628 
1629     /* disabled wait - sleeping in user space */
1630     if (s390_cpu_get_state(cpu) == CPU_STATE_OPERATING && cs->halted) {
1631         s390_cpu_set_state(CPU_STATE_STOPPED, cpu);
1632     }
1633 
1634     switch (s390_cpu_get_state(cpu)) {
1635     case CPU_STATE_OPERATING:
1636         cpu->env.sigp_order = SIGP_STOP_STORE_STATUS;
1637         kvm_s390_vcpu_interrupt(cpu, &irq);
1638         /* store will be performed when handling the stop intercept */
1639         break;
1640     case CPU_STATE_STOPPED:
1641         /* already stopped, just store the status */
1642         cpu_synchronize_state(cs);
1643         kvm_s390_store_status(cpu, KVM_S390_STORE_STATUS_DEF_ADDR, true);
1644         break;
1645     }
1646     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1647 }
1648 
1649 static void sigp_store_status_at_address(CPUState *cs, run_on_cpu_data arg)
1650 {
1651     S390CPU *cpu = S390_CPU(cs);
1652     SigpInfo *si = arg.host_ptr;
1653     uint32_t address = si->param & 0x7ffffe00u;
1654 
1655     /* cpu has to be stopped */
1656     if (s390_cpu_get_state(cpu) != CPU_STATE_STOPPED) {
1657         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1658         return;
1659     }
1660 
1661     cpu_synchronize_state(cs);
1662 
1663     if (kvm_s390_store_status(cpu, address, false)) {
1664         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1665         return;
1666     }
1667     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1668 }
1669 
1670 #define ADTL_SAVE_LC_MASK  0xfUL
1671 static void sigp_store_adtl_status(CPUState *cs, run_on_cpu_data arg)
1672 {
1673     S390CPU *cpu = S390_CPU(cs);
1674     SigpInfo *si = arg.host_ptr;
1675     uint8_t lc = si->param & ADTL_SAVE_LC_MASK;
1676     hwaddr addr = si->param & ~ADTL_SAVE_LC_MASK;
1677     hwaddr len = 1UL << (lc ? lc : 10);
1678 
1679     if (!s390_has_feat(S390_FEAT_VECTOR) &&
1680         !s390_has_feat(S390_FEAT_GUARDED_STORAGE)) {
1681         set_sigp_status(si, SIGP_STAT_INVALID_ORDER);
1682         return;
1683     }
1684 
1685     /* cpu has to be stopped */
1686     if (s390_cpu_get_state(cpu) != CPU_STATE_STOPPED) {
1687         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1688         return;
1689     }
1690 
1691     /* address must be aligned to length */
1692     if (addr & (len - 1)) {
1693         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1694         return;
1695     }
1696 
1697     /* no GS: only lc == 0 is valid */
1698     if (!s390_has_feat(S390_FEAT_GUARDED_STORAGE) &&
1699         lc != 0) {
1700         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1701         return;
1702     }
1703 
1704     /* GS: 0, 10, 11, 12 are valid */
1705     if (s390_has_feat(S390_FEAT_GUARDED_STORAGE) &&
1706         lc != 0 &&
1707         lc != 10 &&
1708         lc != 11 &&
1709         lc != 12) {
1710         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1711         return;
1712     }
1713 
1714     cpu_synchronize_state(cs);
1715 
1716     if (do_store_adtl_status(cpu, addr, len)) {
1717         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1718         return;
1719     }
1720     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1721 }
1722 
1723 static void sigp_restart(CPUState *cs, run_on_cpu_data arg)
1724 {
1725     S390CPU *cpu = S390_CPU(cs);
1726     SigpInfo *si = arg.host_ptr;
1727     struct kvm_s390_irq irq = {
1728         .type = KVM_S390_RESTART,
1729     };
1730 
1731     switch (s390_cpu_get_state(cpu)) {
1732     case CPU_STATE_STOPPED:
1733         /* the restart irq has to be delivered prior to any other pending irq */
1734         cpu_synchronize_state(cs);
1735         do_restart_interrupt(&cpu->env);
1736         s390_cpu_set_state(CPU_STATE_OPERATING, cpu);
1737         break;
1738     case CPU_STATE_OPERATING:
1739         kvm_s390_vcpu_interrupt(cpu, &irq);
1740         break;
1741     }
1742     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1743 }
1744 
1745 int kvm_s390_cpu_restart(S390CPU *cpu)
1746 {
1747     SigpInfo si = {};
1748 
1749     run_on_cpu(CPU(cpu), sigp_restart, RUN_ON_CPU_HOST_PTR(&si));
1750     DPRINTF("DONE: KVM cpu restart: %p\n", &cpu->env);
1751     return 0;
1752 }
1753 
1754 static void sigp_initial_cpu_reset(CPUState *cs, run_on_cpu_data arg)
1755 {
1756     S390CPU *cpu = S390_CPU(cs);
1757     S390CPUClass *scc = S390_CPU_GET_CLASS(cpu);
1758     SigpInfo *si = arg.host_ptr;
1759 
1760     cpu_synchronize_state(cs);
1761     scc->initial_cpu_reset(cs);
1762     cpu_synchronize_post_reset(cs);
1763     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1764 }
1765 
1766 static void sigp_cpu_reset(CPUState *cs, run_on_cpu_data arg)
1767 {
1768     S390CPU *cpu = S390_CPU(cs);
1769     S390CPUClass *scc = S390_CPU_GET_CLASS(cpu);
1770     SigpInfo *si = arg.host_ptr;
1771 
1772     cpu_synchronize_state(cs);
1773     scc->cpu_reset(cs);
1774     cpu_synchronize_post_reset(cs);
1775     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1776 }
1777 
1778 static void sigp_set_prefix(CPUState *cs, run_on_cpu_data arg)
1779 {
1780     S390CPU *cpu = S390_CPU(cs);
1781     SigpInfo *si = arg.host_ptr;
1782     uint32_t addr = si->param & 0x7fffe000u;
1783 
1784     cpu_synchronize_state(cs);
1785 
1786     if (!address_space_access_valid(&address_space_memory, addr,
1787                                     sizeof(struct LowCore), false)) {
1788         set_sigp_status(si, SIGP_STAT_INVALID_PARAMETER);
1789         return;
1790     }
1791 
1792     /* cpu has to be stopped */
1793     if (s390_cpu_get_state(cpu) != CPU_STATE_STOPPED) {
1794         set_sigp_status(si, SIGP_STAT_INCORRECT_STATE);
1795         return;
1796     }
1797 
1798     cpu->env.psa = addr;
1799     cpu_synchronize_post_init(cs);
1800     si->cc = SIGP_CC_ORDER_CODE_ACCEPTED;
1801 }
1802 
1803 static int handle_sigp_single_dst(S390CPU *dst_cpu, uint8_t order,
1804                                   uint64_t param, uint64_t *status_reg)
1805 {
1806     SigpInfo si = {
1807         .param = param,
1808         .status_reg = status_reg,
1809     };
1810 
1811     /* cpu available? */
1812     if (dst_cpu == NULL) {
1813         return SIGP_CC_NOT_OPERATIONAL;
1814     }
1815 
1816     /* only resets can break pending orders */
1817     if (dst_cpu->env.sigp_order != 0 &&
1818         order != SIGP_CPU_RESET &&
1819         order != SIGP_INITIAL_CPU_RESET) {
1820         return SIGP_CC_BUSY;
1821     }
1822 
1823     switch (order) {
1824     case SIGP_START:
1825         run_on_cpu(CPU(dst_cpu), sigp_start, RUN_ON_CPU_HOST_PTR(&si));
1826         break;
1827     case SIGP_STOP:
1828         run_on_cpu(CPU(dst_cpu), sigp_stop, RUN_ON_CPU_HOST_PTR(&si));
1829         break;
1830     case SIGP_RESTART:
1831         run_on_cpu(CPU(dst_cpu), sigp_restart, RUN_ON_CPU_HOST_PTR(&si));
1832         break;
1833     case SIGP_STOP_STORE_STATUS:
1834         run_on_cpu(CPU(dst_cpu), sigp_stop_and_store_status, RUN_ON_CPU_HOST_PTR(&si));
1835         break;
1836     case SIGP_STORE_STATUS_ADDR:
1837         run_on_cpu(CPU(dst_cpu), sigp_store_status_at_address, RUN_ON_CPU_HOST_PTR(&si));
1838         break;
1839     case SIGP_STORE_ADTL_STATUS:
1840         run_on_cpu(CPU(dst_cpu), sigp_store_adtl_status, RUN_ON_CPU_HOST_PTR(&si));
1841         break;
1842     case SIGP_SET_PREFIX:
1843         run_on_cpu(CPU(dst_cpu), sigp_set_prefix, RUN_ON_CPU_HOST_PTR(&si));
1844         break;
1845     case SIGP_INITIAL_CPU_RESET:
1846         run_on_cpu(CPU(dst_cpu), sigp_initial_cpu_reset, RUN_ON_CPU_HOST_PTR(&si));
1847         break;
1848     case SIGP_CPU_RESET:
1849         run_on_cpu(CPU(dst_cpu), sigp_cpu_reset, RUN_ON_CPU_HOST_PTR(&si));
1850         break;
1851     default:
1852         DPRINTF("KVM: unknown SIGP: 0x%x\n", order);
1853         set_sigp_status(&si, SIGP_STAT_INVALID_ORDER);
1854     }
1855 
1856     return si.cc;
1857 }
1858 
1859 static int sigp_set_architecture(S390CPU *cpu, uint32_t param,
1860                                  uint64_t *status_reg)
1861 {
1862     CPUState *cur_cs;
1863     S390CPU *cur_cpu;
1864     bool all_stopped = true;
1865 
1866     CPU_FOREACH(cur_cs) {
1867         cur_cpu = S390_CPU(cur_cs);
1868 
1869         if (cur_cpu == cpu) {
1870             continue;
1871         }
1872         if (s390_cpu_get_state(cur_cpu) != CPU_STATE_STOPPED) {
1873             all_stopped = false;
1874         }
1875     }
1876 
1877     *status_reg &= 0xffffffff00000000ULL;
1878 
1879     /* Reject set arch order, with czam we're always in z/Arch mode. */
1880     *status_reg |= (all_stopped ? SIGP_STAT_INVALID_PARAMETER :
1881                     SIGP_STAT_INCORRECT_STATE);
1882     return SIGP_CC_STATUS_STORED;
1883 }
1884 
1885 static int handle_sigp(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1886 {
1887     CPUS390XState *env = &cpu->env;
1888     const uint8_t r1 = ipa1 >> 4;
1889     const uint8_t r3 = ipa1 & 0x0f;
1890     int ret;
1891     uint8_t order;
1892     uint64_t *status_reg;
1893     uint64_t param;
1894     S390CPU *dst_cpu = NULL;
1895 
1896     cpu_synchronize_state(CPU(cpu));
1897 
1898     /* get order code */
1899     order = decode_basedisp_rs(env, run->s390_sieic.ipb, NULL)
1900         & SIGP_ORDER_MASK;
1901     status_reg = &env->regs[r1];
1902     param = (r1 % 2) ? env->regs[r1] : env->regs[r1 + 1];
1903 
1904     if (qemu_mutex_trylock(&qemu_sigp_mutex)) {
1905         ret = SIGP_CC_BUSY;
1906         goto out;
1907     }
1908 
1909     switch (order) {
1910     case SIGP_SET_ARCH:
1911         ret = sigp_set_architecture(cpu, param, status_reg);
1912         break;
1913     default:
1914         /* all other sigp orders target a single vcpu */
1915         dst_cpu = s390_cpu_addr2state(env->regs[r3]);
1916         ret = handle_sigp_single_dst(dst_cpu, order, param, status_reg);
1917     }
1918     qemu_mutex_unlock(&qemu_sigp_mutex);
1919 
1920 out:
1921     trace_kvm_sigp_finished(order, CPU(cpu)->cpu_index,
1922                             dst_cpu ? CPU(dst_cpu)->cpu_index : -1, ret);
1923 
1924     if (ret >= 0) {
1925         setcc(cpu, ret);
1926         return 0;
1927     }
1928 
1929     return ret;
1930 }
1931 
1932 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1933 {
1934     unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1935     uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1936     int r = -1;
1937 
1938     DPRINTF("handle_instruction 0x%x 0x%x\n",
1939             run->s390_sieic.ipa, run->s390_sieic.ipb);
1940     switch (ipa0) {
1941     case IPA0_B2:
1942         r = handle_b2(cpu, run, ipa1);
1943         break;
1944     case IPA0_B9:
1945         r = handle_b9(cpu, run, ipa1);
1946         break;
1947     case IPA0_EB:
1948         r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1949         break;
1950     case IPA0_E3:
1951         r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1952         break;
1953     case IPA0_DIAG:
1954         r = handle_diag(cpu, run, run->s390_sieic.ipb);
1955         break;
1956     case IPA0_SIGP:
1957         r = handle_sigp(cpu, run, ipa1);
1958         break;
1959     }
1960 
1961     if (r < 0) {
1962         r = 0;
1963         kvm_s390_program_interrupt(cpu, PGM_OPERATION);
1964     }
1965 
1966     return r;
1967 }
1968 
1969 static bool is_special_wait_psw(CPUState *cs)
1970 {
1971     /* signal quiesce */
1972     return cs->kvm_run->psw_addr == 0xfffUL;
1973 }
1974 
1975 static void unmanageable_intercept(S390CPU *cpu, const char *str, int pswoffset)
1976 {
1977     CPUState *cs = CPU(cpu);
1978 
1979     error_report("Unmanageable %s! CPU%i new PSW: 0x%016lx:%016lx",
1980                  str, cs->cpu_index, ldq_phys(cs->as, cpu->env.psa + pswoffset),
1981                  ldq_phys(cs->as, cpu->env.psa + pswoffset + 8));
1982     s390_cpu_halt(cpu);
1983     qemu_system_guest_panicked(NULL);
1984 }
1985 
1986 /* try to detect pgm check loops */
1987 static int handle_oper_loop(S390CPU *cpu, struct kvm_run *run)
1988 {
1989     CPUState *cs = CPU(cpu);
1990     PSW oldpsw, newpsw;
1991 
1992     cpu_synchronize_state(cs);
1993     newpsw.mask = ldq_phys(cs->as, cpu->env.psa +
1994                            offsetof(LowCore, program_new_psw));
1995     newpsw.addr = ldq_phys(cs->as, cpu->env.psa +
1996                            offsetof(LowCore, program_new_psw) + 8);
1997     oldpsw.mask  = run->psw_mask;
1998     oldpsw.addr  = run->psw_addr;
1999     /*
2000      * Avoid endless loops of operation exceptions, if the pgm new
2001      * PSW will cause a new operation exception.
2002      * The heuristic checks if the pgm new psw is within 6 bytes before
2003      * the faulting psw address (with same DAT, AS settings) and the
2004      * new psw is not a wait psw and the fault was not triggered by
2005      * problem state. In that case go into crashed state.
2006      */
2007 
2008     if (oldpsw.addr - newpsw.addr <= 6 &&
2009         !(newpsw.mask & PSW_MASK_WAIT) &&
2010         !(oldpsw.mask & PSW_MASK_PSTATE) &&
2011         (newpsw.mask & PSW_MASK_ASC) == (oldpsw.mask & PSW_MASK_ASC) &&
2012         (newpsw.mask & PSW_MASK_DAT) == (oldpsw.mask & PSW_MASK_DAT)) {
2013         unmanageable_intercept(cpu, "operation exception loop",
2014                                offsetof(LowCore, program_new_psw));
2015         return EXCP_HALTED;
2016     }
2017     return 0;
2018 }
2019 
2020 static int handle_intercept(S390CPU *cpu)
2021 {
2022     CPUState *cs = CPU(cpu);
2023     struct kvm_run *run = cs->kvm_run;
2024     int icpt_code = run->s390_sieic.icptcode;
2025     int r = 0;
2026 
2027     DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
2028             (long)cs->kvm_run->psw_addr);
2029     switch (icpt_code) {
2030         case ICPT_INSTRUCTION:
2031             r = handle_instruction(cpu, run);
2032             break;
2033         case ICPT_PROGRAM:
2034             unmanageable_intercept(cpu, "program interrupt",
2035                                    offsetof(LowCore, program_new_psw));
2036             r = EXCP_HALTED;
2037             break;
2038         case ICPT_EXT_INT:
2039             unmanageable_intercept(cpu, "external interrupt",
2040                                    offsetof(LowCore, external_new_psw));
2041             r = EXCP_HALTED;
2042             break;
2043         case ICPT_WAITPSW:
2044             /* disabled wait, since enabled wait is handled in kernel */
2045             cpu_synchronize_state(cs);
2046             if (s390_cpu_halt(cpu) == 0) {
2047                 if (is_special_wait_psw(cs)) {
2048                     qemu_system_shutdown_request(SHUTDOWN_CAUSE_GUEST_SHUTDOWN);
2049                 } else {
2050                     qemu_system_guest_panicked(NULL);
2051                 }
2052             }
2053             r = EXCP_HALTED;
2054             break;
2055         case ICPT_CPU_STOP:
2056             if (s390_cpu_set_state(CPU_STATE_STOPPED, cpu) == 0) {
2057                 qemu_system_shutdown_request(SHUTDOWN_CAUSE_GUEST_SHUTDOWN);
2058             }
2059             if (cpu->env.sigp_order == SIGP_STOP_STORE_STATUS) {
2060                 kvm_s390_store_status(cpu, KVM_S390_STORE_STATUS_DEF_ADDR,
2061                                       true);
2062             }
2063             cpu->env.sigp_order = 0;
2064             r = EXCP_HALTED;
2065             break;
2066         case ICPT_OPEREXC:
2067             /* check for break points */
2068             r = handle_sw_breakpoint(cpu, run);
2069             if (r == -ENOENT) {
2070                 /* Then check for potential pgm check loops */
2071                 r = handle_oper_loop(cpu, run);
2072                 if (r == 0) {
2073                     kvm_s390_program_interrupt(cpu, PGM_OPERATION);
2074                 }
2075             }
2076             break;
2077         case ICPT_SOFT_INTERCEPT:
2078             fprintf(stderr, "KVM unimplemented icpt SOFT\n");
2079             exit(1);
2080             break;
2081         case ICPT_IO:
2082             fprintf(stderr, "KVM unimplemented icpt IO\n");
2083             exit(1);
2084             break;
2085         default:
2086             fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
2087             exit(1);
2088             break;
2089     }
2090 
2091     return r;
2092 }
2093 
2094 static int handle_tsch(S390CPU *cpu)
2095 {
2096     CPUState *cs = CPU(cpu);
2097     struct kvm_run *run = cs->kvm_run;
2098     int ret;
2099 
2100     cpu_synchronize_state(cs);
2101 
2102     ret = ioinst_handle_tsch(cpu, cpu->env.regs[1], run->s390_tsch.ipb);
2103     if (ret < 0) {
2104         /*
2105          * Failure.
2106          * If an I/O interrupt had been dequeued, we have to reinject it.
2107          */
2108         if (run->s390_tsch.dequeued) {
2109             kvm_s390_io_interrupt(run->s390_tsch.subchannel_id,
2110                                   run->s390_tsch.subchannel_nr,
2111                                   run->s390_tsch.io_int_parm,
2112                                   run->s390_tsch.io_int_word);
2113         }
2114         ret = 0;
2115     }
2116     return ret;
2117 }
2118 
2119 static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
2120 {
2121     struct sysib_322 sysib;
2122     int del;
2123 
2124     if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
2125         return;
2126     }
2127     /* Shift the stack of Extended Names to prepare for our own data */
2128     memmove(&sysib.ext_names[1], &sysib.ext_names[0],
2129             sizeof(sysib.ext_names[0]) * (sysib.count - 1));
2130     /* First virt level, that doesn't provide Ext Names delimits stack. It is
2131      * assumed it's not capable of managing Extended Names for lower levels.
2132      */
2133     for (del = 1; del < sysib.count; del++) {
2134         if (!sysib.vm[del].ext_name_encoding || !sysib.ext_names[del][0]) {
2135             break;
2136         }
2137     }
2138     if (del < sysib.count) {
2139         memset(sysib.ext_names[del], 0,
2140                sizeof(sysib.ext_names[0]) * (sysib.count - del));
2141     }
2142     /* Insert short machine name in EBCDIC, padded with blanks */
2143     if (qemu_name) {
2144         memset(sysib.vm[0].name, 0x40, sizeof(sysib.vm[0].name));
2145         ebcdic_put(sysib.vm[0].name, qemu_name, MIN(sizeof(sysib.vm[0].name),
2146                                                     strlen(qemu_name)));
2147     }
2148     sysib.vm[0].ext_name_encoding = 2; /* 2 = UTF-8 */
2149     memset(sysib.ext_names[0], 0, sizeof(sysib.ext_names[0]));
2150     /* If hypervisor specifies zero Extended Name in STSI322 SYSIB, it's
2151      * considered by s390 as not capable of providing any Extended Name.
2152      * Therefore if no name was specified on qemu invocation, we go with the
2153      * same "KVMguest" default, which KVM has filled into short name field.
2154      */
2155     if (qemu_name) {
2156         strncpy((char *)sysib.ext_names[0], qemu_name,
2157                 sizeof(sysib.ext_names[0]));
2158     } else {
2159         strcpy((char *)sysib.ext_names[0], "KVMguest");
2160     }
2161     /* Insert UUID */
2162     memcpy(sysib.vm[0].uuid, &qemu_uuid, sizeof(sysib.vm[0].uuid));
2163 
2164     s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
2165 }
2166 
2167 static int handle_stsi(S390CPU *cpu)
2168 {
2169     CPUState *cs = CPU(cpu);
2170     struct kvm_run *run = cs->kvm_run;
2171 
2172     switch (run->s390_stsi.fc) {
2173     case 3:
2174         if (run->s390_stsi.sel1 != 2 || run->s390_stsi.sel2 != 2) {
2175             return 0;
2176         }
2177         /* Only sysib 3.2.2 needs post-handling for now. */
2178         insert_stsi_3_2_2(cpu, run->s390_stsi.addr, run->s390_stsi.ar);
2179         return 0;
2180     default:
2181         return 0;
2182     }
2183 }
2184 
2185 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
2186 {
2187     CPUState *cs = CPU(cpu);
2188     struct kvm_run *run = cs->kvm_run;
2189 
2190     int ret = 0;
2191     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
2192 
2193     switch (arch_info->type) {
2194     case KVM_HW_WP_WRITE:
2195         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
2196             cs->watchpoint_hit = &hw_watchpoint;
2197             hw_watchpoint.vaddr = arch_info->addr;
2198             hw_watchpoint.flags = BP_MEM_WRITE;
2199             ret = EXCP_DEBUG;
2200         }
2201         break;
2202     case KVM_HW_BP:
2203         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
2204             ret = EXCP_DEBUG;
2205         }
2206         break;
2207     case KVM_SINGLESTEP:
2208         if (cs->singlestep_enabled) {
2209             ret = EXCP_DEBUG;
2210         }
2211         break;
2212     default:
2213         ret = -ENOSYS;
2214     }
2215 
2216     return ret;
2217 }
2218 
2219 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
2220 {
2221     S390CPU *cpu = S390_CPU(cs);
2222     int ret = 0;
2223 
2224     qemu_mutex_lock_iothread();
2225 
2226     switch (run->exit_reason) {
2227         case KVM_EXIT_S390_SIEIC:
2228             ret = handle_intercept(cpu);
2229             break;
2230         case KVM_EXIT_S390_RESET:
2231             s390_reipl_request();
2232             break;
2233         case KVM_EXIT_S390_TSCH:
2234             ret = handle_tsch(cpu);
2235             break;
2236         case KVM_EXIT_S390_STSI:
2237             ret = handle_stsi(cpu);
2238             break;
2239         case KVM_EXIT_DEBUG:
2240             ret = kvm_arch_handle_debug_exit(cpu);
2241             break;
2242         default:
2243             fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
2244             break;
2245     }
2246     qemu_mutex_unlock_iothread();
2247 
2248     if (ret == 0) {
2249         ret = EXCP_INTERRUPT;
2250     }
2251     return ret;
2252 }
2253 
2254 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
2255 {
2256     return true;
2257 }
2258 
2259 void kvm_s390_io_interrupt(uint16_t subchannel_id,
2260                            uint16_t subchannel_nr, uint32_t io_int_parm,
2261                            uint32_t io_int_word)
2262 {
2263     struct kvm_s390_irq irq = {
2264         .u.io.subchannel_id = subchannel_id,
2265         .u.io.subchannel_nr = subchannel_nr,
2266         .u.io.io_int_parm = io_int_parm,
2267         .u.io.io_int_word = io_int_word,
2268     };
2269 
2270     if (io_int_word & IO_INT_WORD_AI) {
2271         irq.type = KVM_S390_INT_IO(1, 0, 0, 0);
2272     } else {
2273         irq.type = KVM_S390_INT_IO(0, (subchannel_id & 0xff00) >> 8,
2274                                       (subchannel_id & 0x0006),
2275                                       subchannel_nr);
2276     }
2277     kvm_s390_floating_interrupt(&irq);
2278 }
2279 
2280 static uint64_t build_channel_report_mcic(void)
2281 {
2282     uint64_t mcic;
2283 
2284     /* subclass: indicate channel report pending */
2285     mcic = MCIC_SC_CP |
2286     /* subclass modifiers: none */
2287     /* storage errors: none */
2288     /* validity bits: no damage */
2289         MCIC_VB_WP | MCIC_VB_MS | MCIC_VB_PM | MCIC_VB_IA | MCIC_VB_FP |
2290         MCIC_VB_GR | MCIC_VB_CR | MCIC_VB_ST | MCIC_VB_AR | MCIC_VB_PR |
2291         MCIC_VB_FC | MCIC_VB_CT | MCIC_VB_CC;
2292     if (s390_has_feat(S390_FEAT_VECTOR)) {
2293         mcic |= MCIC_VB_VR;
2294     }
2295     if (s390_has_feat(S390_FEAT_GUARDED_STORAGE)) {
2296         mcic |= MCIC_VB_GS;
2297     }
2298     return mcic;
2299 }
2300 
2301 void kvm_s390_crw_mchk(void)
2302 {
2303     struct kvm_s390_irq irq = {
2304         .type = KVM_S390_MCHK,
2305         .u.mchk.cr14 = 1 << 28,
2306         .u.mchk.mcic = build_channel_report_mcic(),
2307     };
2308     kvm_s390_floating_interrupt(&irq);
2309 }
2310 
2311 void kvm_s390_enable_css_support(S390CPU *cpu)
2312 {
2313     int r;
2314 
2315     /* Activate host kernel channel subsystem support. */
2316     r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
2317     assert(r == 0);
2318 }
2319 
2320 void kvm_arch_init_irq_routing(KVMState *s)
2321 {
2322     /*
2323      * Note that while irqchip capabilities generally imply that cpustates
2324      * are handled in-kernel, it is not true for s390 (yet); therefore, we
2325      * have to override the common code kvm_halt_in_kernel_allowed setting.
2326      */
2327     if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
2328         kvm_gsi_routing_allowed = true;
2329         kvm_halt_in_kernel_allowed = false;
2330     }
2331 }
2332 
2333 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
2334                                     int vq, bool assign)
2335 {
2336     struct kvm_ioeventfd kick = {
2337         .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
2338         KVM_IOEVENTFD_FLAG_DATAMATCH,
2339         .fd = event_notifier_get_fd(notifier),
2340         .datamatch = vq,
2341         .addr = sch,
2342         .len = 8,
2343     };
2344     if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
2345         return -ENOSYS;
2346     }
2347     if (!assign) {
2348         kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
2349     }
2350     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
2351 }
2352 
2353 int kvm_s390_get_memslot_count(void)
2354 {
2355     return kvm_check_extension(kvm_state, KVM_CAP_NR_MEMSLOTS);
2356 }
2357 
2358 int kvm_s390_get_ri(void)
2359 {
2360     return cap_ri;
2361 }
2362 
2363 int kvm_s390_get_gs(void)
2364 {
2365     return cap_gs;
2366 }
2367 
2368 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
2369 {
2370     struct kvm_mp_state mp_state = {};
2371     int ret;
2372 
2373     /* the kvm part might not have been initialized yet */
2374     if (CPU(cpu)->kvm_state == NULL) {
2375         return 0;
2376     }
2377 
2378     switch (cpu_state) {
2379     case CPU_STATE_STOPPED:
2380         mp_state.mp_state = KVM_MP_STATE_STOPPED;
2381         break;
2382     case CPU_STATE_CHECK_STOP:
2383         mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
2384         break;
2385     case CPU_STATE_OPERATING:
2386         mp_state.mp_state = KVM_MP_STATE_OPERATING;
2387         break;
2388     case CPU_STATE_LOAD:
2389         mp_state.mp_state = KVM_MP_STATE_LOAD;
2390         break;
2391     default:
2392         error_report("Requested CPU state is not a valid S390 CPU state: %u",
2393                      cpu_state);
2394         exit(1);
2395     }
2396 
2397     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
2398     if (ret) {
2399         trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
2400                                        strerror(-ret));
2401     }
2402 
2403     return ret;
2404 }
2405 
2406 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu)
2407 {
2408     struct kvm_s390_irq_state irq_state;
2409     CPUState *cs = CPU(cpu);
2410     int32_t bytes;
2411 
2412     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2413         return;
2414     }
2415 
2416     irq_state.buf = (uint64_t) cpu->irqstate;
2417     irq_state.len = VCPU_IRQ_BUF_SIZE;
2418 
2419     bytes = kvm_vcpu_ioctl(cs, KVM_S390_GET_IRQ_STATE, &irq_state);
2420     if (bytes < 0) {
2421         cpu->irqstate_saved_size = 0;
2422         error_report("Migration of interrupt state failed");
2423         return;
2424     }
2425 
2426     cpu->irqstate_saved_size = bytes;
2427 }
2428 
2429 int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu)
2430 {
2431     CPUState *cs = CPU(cpu);
2432     struct kvm_s390_irq_state irq_state;
2433     int r;
2434 
2435     if (cpu->irqstate_saved_size == 0) {
2436         return 0;
2437     }
2438 
2439     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
2440         return -ENOSYS;
2441     }
2442 
2443     irq_state.buf = (uint64_t) cpu->irqstate;
2444     irq_state.len = cpu->irqstate_saved_size;
2445 
2446     r = kvm_vcpu_ioctl(cs, KVM_S390_SET_IRQ_STATE, &irq_state);
2447     if (r) {
2448         error_report("Setting interrupt state failed %d", r);
2449     }
2450     return r;
2451 }
2452 
2453 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
2454                              uint64_t address, uint32_t data, PCIDevice *dev)
2455 {
2456     S390PCIBusDevice *pbdev;
2457     uint32_t vec = data & ZPCI_MSI_VEC_MASK;
2458 
2459     if (!dev) {
2460         DPRINTF("add_msi_route no pci device\n");
2461         return -ENODEV;
2462     }
2463 
2464     pbdev = s390_pci_find_dev_by_target(s390_get_phb(), DEVICE(dev)->id);
2465     if (!pbdev) {
2466         DPRINTF("add_msi_route no zpci device\n");
2467         return -ENODEV;
2468     }
2469 
2470     route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
2471     route->flags = 0;
2472     route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
2473     route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
2474     route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
2475     route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset + vec;
2476     route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;
2477     return 0;
2478 }
2479 
2480 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
2481                                 int vector, PCIDevice *dev)
2482 {
2483     return 0;
2484 }
2485 
2486 int kvm_arch_release_virq_post(int virq)
2487 {
2488     return 0;
2489 }
2490 
2491 int kvm_arch_msi_data_to_gsi(uint32_t data)
2492 {
2493     abort();
2494 }
2495 
2496 static int query_cpu_subfunc(S390FeatBitmap features)
2497 {
2498     struct kvm_s390_vm_cpu_subfunc prop;
2499     struct kvm_device_attr attr = {
2500         .group = KVM_S390_VM_CPU_MODEL,
2501         .attr = KVM_S390_VM_CPU_MACHINE_SUBFUNC,
2502         .addr = (uint64_t) &prop,
2503     };
2504     int rc;
2505 
2506     rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2507     if (rc) {
2508         return  rc;
2509     }
2510 
2511     /*
2512      * We're going to add all subfunctions now, if the corresponding feature
2513      * is available that unlocks the query functions.
2514      */
2515     s390_add_from_feat_block(features, S390_FEAT_TYPE_PLO, prop.plo);
2516     if (test_bit(S390_FEAT_TOD_CLOCK_STEERING, features)) {
2517         s390_add_from_feat_block(features, S390_FEAT_TYPE_PTFF, prop.ptff);
2518     }
2519     if (test_bit(S390_FEAT_MSA, features)) {
2520         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMAC, prop.kmac);
2521         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMC, prop.kmc);
2522         s390_add_from_feat_block(features, S390_FEAT_TYPE_KM, prop.km);
2523         s390_add_from_feat_block(features, S390_FEAT_TYPE_KIMD, prop.kimd);
2524         s390_add_from_feat_block(features, S390_FEAT_TYPE_KLMD, prop.klmd);
2525     }
2526     if (test_bit(S390_FEAT_MSA_EXT_3, features)) {
2527         s390_add_from_feat_block(features, S390_FEAT_TYPE_PCKMO, prop.pckmo);
2528     }
2529     if (test_bit(S390_FEAT_MSA_EXT_4, features)) {
2530         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMCTR, prop.kmctr);
2531         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMF, prop.kmf);
2532         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMO, prop.kmo);
2533         s390_add_from_feat_block(features, S390_FEAT_TYPE_PCC, prop.pcc);
2534     }
2535     if (test_bit(S390_FEAT_MSA_EXT_5, features)) {
2536         s390_add_from_feat_block(features, S390_FEAT_TYPE_PPNO, prop.ppno);
2537     }
2538     if (test_bit(S390_FEAT_MSA_EXT_8, features)) {
2539         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMA, prop.kma);
2540     }
2541     return 0;
2542 }
2543 
2544 static int configure_cpu_subfunc(const S390FeatBitmap features)
2545 {
2546     struct kvm_s390_vm_cpu_subfunc prop = {};
2547     struct kvm_device_attr attr = {
2548         .group = KVM_S390_VM_CPU_MODEL,
2549         .attr = KVM_S390_VM_CPU_PROCESSOR_SUBFUNC,
2550         .addr = (uint64_t) &prop,
2551     };
2552 
2553     if (!kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2554                            KVM_S390_VM_CPU_PROCESSOR_SUBFUNC)) {
2555         /* hardware support might be missing, IBC will handle most of this */
2556         return 0;
2557     }
2558 
2559     s390_fill_feat_block(features, S390_FEAT_TYPE_PLO, prop.plo);
2560     if (test_bit(S390_FEAT_TOD_CLOCK_STEERING, features)) {
2561         s390_fill_feat_block(features, S390_FEAT_TYPE_PTFF, prop.ptff);
2562     }
2563     if (test_bit(S390_FEAT_MSA, features)) {
2564         s390_fill_feat_block(features, S390_FEAT_TYPE_KMAC, prop.kmac);
2565         s390_fill_feat_block(features, S390_FEAT_TYPE_KMC, prop.kmc);
2566         s390_fill_feat_block(features, S390_FEAT_TYPE_KM, prop.km);
2567         s390_fill_feat_block(features, S390_FEAT_TYPE_KIMD, prop.kimd);
2568         s390_fill_feat_block(features, S390_FEAT_TYPE_KLMD, prop.klmd);
2569     }
2570     if (test_bit(S390_FEAT_MSA_EXT_3, features)) {
2571         s390_fill_feat_block(features, S390_FEAT_TYPE_PCKMO, prop.pckmo);
2572     }
2573     if (test_bit(S390_FEAT_MSA_EXT_4, features)) {
2574         s390_fill_feat_block(features, S390_FEAT_TYPE_KMCTR, prop.kmctr);
2575         s390_fill_feat_block(features, S390_FEAT_TYPE_KMF, prop.kmf);
2576         s390_fill_feat_block(features, S390_FEAT_TYPE_KMO, prop.kmo);
2577         s390_fill_feat_block(features, S390_FEAT_TYPE_PCC, prop.pcc);
2578     }
2579     if (test_bit(S390_FEAT_MSA_EXT_5, features)) {
2580         s390_fill_feat_block(features, S390_FEAT_TYPE_PPNO, prop.ppno);
2581     }
2582     if (test_bit(S390_FEAT_MSA_EXT_8, features)) {
2583         s390_fill_feat_block(features, S390_FEAT_TYPE_KMA, prop.kma);
2584     }
2585     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2586 }
2587 
2588 static int kvm_to_feat[][2] = {
2589     { KVM_S390_VM_CPU_FEAT_ESOP, S390_FEAT_ESOP },
2590     { KVM_S390_VM_CPU_FEAT_SIEF2, S390_FEAT_SIE_F2 },
2591     { KVM_S390_VM_CPU_FEAT_64BSCAO , S390_FEAT_SIE_64BSCAO },
2592     { KVM_S390_VM_CPU_FEAT_SIIF, S390_FEAT_SIE_SIIF },
2593     { KVM_S390_VM_CPU_FEAT_GPERE, S390_FEAT_SIE_GPERE },
2594     { KVM_S390_VM_CPU_FEAT_GSLS, S390_FEAT_SIE_GSLS },
2595     { KVM_S390_VM_CPU_FEAT_IB, S390_FEAT_SIE_IB },
2596     { KVM_S390_VM_CPU_FEAT_CEI, S390_FEAT_SIE_CEI },
2597     { KVM_S390_VM_CPU_FEAT_IBS, S390_FEAT_SIE_IBS },
2598     { KVM_S390_VM_CPU_FEAT_SKEY, S390_FEAT_SIE_SKEY },
2599     { KVM_S390_VM_CPU_FEAT_CMMA, S390_FEAT_SIE_CMMA },
2600     { KVM_S390_VM_CPU_FEAT_PFMFI, S390_FEAT_SIE_PFMFI},
2601     { KVM_S390_VM_CPU_FEAT_SIGPIF, S390_FEAT_SIE_SIGPIF},
2602     { KVM_S390_VM_CPU_FEAT_KSS, S390_FEAT_SIE_KSS},
2603 };
2604 
2605 static int query_cpu_feat(S390FeatBitmap features)
2606 {
2607     struct kvm_s390_vm_cpu_feat prop;
2608     struct kvm_device_attr attr = {
2609         .group = KVM_S390_VM_CPU_MODEL,
2610         .attr = KVM_S390_VM_CPU_MACHINE_FEAT,
2611         .addr = (uint64_t) &prop,
2612     };
2613     int rc;
2614     int i;
2615 
2616     rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2617     if (rc) {
2618         return  rc;
2619     }
2620 
2621     for (i = 0; i < ARRAY_SIZE(kvm_to_feat); i++) {
2622         if (test_be_bit(kvm_to_feat[i][0], (uint8_t *) prop.feat)) {
2623             set_bit(kvm_to_feat[i][1], features);
2624         }
2625     }
2626     return 0;
2627 }
2628 
2629 static int configure_cpu_feat(const S390FeatBitmap features)
2630 {
2631     struct kvm_s390_vm_cpu_feat prop = {};
2632     struct kvm_device_attr attr = {
2633         .group = KVM_S390_VM_CPU_MODEL,
2634         .attr = KVM_S390_VM_CPU_PROCESSOR_FEAT,
2635         .addr = (uint64_t) &prop,
2636     };
2637     int i;
2638 
2639     for (i = 0; i < ARRAY_SIZE(kvm_to_feat); i++) {
2640         if (test_bit(kvm_to_feat[i][1], features)) {
2641             set_be_bit(kvm_to_feat[i][0], (uint8_t *) prop.feat);
2642         }
2643     }
2644     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2645 }
2646 
2647 bool kvm_s390_cpu_models_supported(void)
2648 {
2649     if (!cpu_model_allowed()) {
2650         /* compatibility machines interfere with the cpu model */
2651         return false;
2652     }
2653     return kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2654                              KVM_S390_VM_CPU_MACHINE) &&
2655            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2656                              KVM_S390_VM_CPU_PROCESSOR) &&
2657            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2658                              KVM_S390_VM_CPU_MACHINE_FEAT) &&
2659            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2660                              KVM_S390_VM_CPU_PROCESSOR_FEAT) &&
2661            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2662                              KVM_S390_VM_CPU_MACHINE_SUBFUNC);
2663 }
2664 
2665 void kvm_s390_get_host_cpu_model(S390CPUModel *model, Error **errp)
2666 {
2667     struct kvm_s390_vm_cpu_machine prop = {};
2668     struct kvm_device_attr attr = {
2669         .group = KVM_S390_VM_CPU_MODEL,
2670         .attr = KVM_S390_VM_CPU_MACHINE,
2671         .addr = (uint64_t) &prop,
2672     };
2673     uint16_t unblocked_ibc = 0, cpu_type = 0;
2674     int rc;
2675 
2676     memset(model, 0, sizeof(*model));
2677 
2678     if (!kvm_s390_cpu_models_supported()) {
2679         error_setg(errp, "KVM doesn't support CPU models");
2680         return;
2681     }
2682 
2683     /* query the basic cpu model properties */
2684     rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2685     if (rc) {
2686         error_setg(errp, "KVM: Error querying host CPU model: %d", rc);
2687         return;
2688     }
2689 
2690     cpu_type = cpuid_type(prop.cpuid);
2691     if (has_ibc(prop.ibc)) {
2692         model->lowest_ibc = lowest_ibc(prop.ibc);
2693         unblocked_ibc = unblocked_ibc(prop.ibc);
2694     }
2695     model->cpu_id = cpuid_id(prop.cpuid);
2696     model->cpu_id_format = cpuid_format(prop.cpuid);
2697     model->cpu_ver = 0xff;
2698 
2699     /* get supported cpu features indicated via STFL(E) */
2700     s390_add_from_feat_block(model->features, S390_FEAT_TYPE_STFL,
2701                              (uint8_t *) prop.fac_mask);
2702     /* dat-enhancement facility 2 has no bit but was introduced with stfle */
2703     if (test_bit(S390_FEAT_STFLE, model->features)) {
2704         set_bit(S390_FEAT_DAT_ENH_2, model->features);
2705     }
2706     /* get supported cpu features indicated e.g. via SCLP */
2707     rc = query_cpu_feat(model->features);
2708     if (rc) {
2709         error_setg(errp, "KVM: Error querying CPU features: %d", rc);
2710         return;
2711     }
2712     /* get supported cpu subfunctions indicated via query / test bit */
2713     rc = query_cpu_subfunc(model->features);
2714     if (rc) {
2715         error_setg(errp, "KVM: Error querying CPU subfunctions: %d", rc);
2716         return;
2717     }
2718 
2719     /* with cpu model support, CMM is only indicated if really available */
2720     if (kvm_s390_cmma_available()) {
2721         set_bit(S390_FEAT_CMM, model->features);
2722     } else {
2723         /* no cmm -> no cmm nt */
2724         clear_bit(S390_FEAT_CMM_NT, model->features);
2725     }
2726 
2727     /* We emulate a zPCI bus and AEN, therefore we don't need HW support */
2728     if (pci_available) {
2729         set_bit(S390_FEAT_ZPCI, model->features);
2730     }
2731     set_bit(S390_FEAT_ADAPTER_EVENT_NOTIFICATION, model->features);
2732 
2733     if (s390_known_cpu_type(cpu_type)) {
2734         /* we want the exact model, even if some features are missing */
2735         model->def = s390_find_cpu_def(cpu_type, ibc_gen(unblocked_ibc),
2736                                        ibc_ec_ga(unblocked_ibc), NULL);
2737     } else {
2738         /* model unknown, e.g. too new - search using features */
2739         model->def = s390_find_cpu_def(0, ibc_gen(unblocked_ibc),
2740                                        ibc_ec_ga(unblocked_ibc),
2741                                        model->features);
2742     }
2743     if (!model->def) {
2744         error_setg(errp, "KVM: host CPU model could not be identified");
2745         return;
2746     }
2747     /* strip of features that are not part of the maximum model */
2748     bitmap_and(model->features, model->features, model->def->full_feat,
2749                S390_FEAT_MAX);
2750 }
2751 
2752 void kvm_s390_apply_cpu_model(const S390CPUModel *model, Error **errp)
2753 {
2754     struct kvm_s390_vm_cpu_processor prop  = {
2755         .fac_list = { 0 },
2756     };
2757     struct kvm_device_attr attr = {
2758         .group = KVM_S390_VM_CPU_MODEL,
2759         .attr = KVM_S390_VM_CPU_PROCESSOR,
2760         .addr = (uint64_t) &prop,
2761     };
2762     int rc;
2763 
2764     if (!model) {
2765         /* compatibility handling if cpu models are disabled */
2766         if (kvm_s390_cmma_available()) {
2767             kvm_s390_enable_cmma();
2768         }
2769         return;
2770     }
2771     if (!kvm_s390_cpu_models_supported()) {
2772         error_setg(errp, "KVM doesn't support CPU models");
2773         return;
2774     }
2775     prop.cpuid = s390_cpuid_from_cpu_model(model);
2776     prop.ibc = s390_ibc_from_cpu_model(model);
2777     /* configure cpu features indicated via STFL(e) */
2778     s390_fill_feat_block(model->features, S390_FEAT_TYPE_STFL,
2779                          (uint8_t *) prop.fac_list);
2780     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2781     if (rc) {
2782         error_setg(errp, "KVM: Error configuring the CPU model: %d", rc);
2783         return;
2784     }
2785     /* configure cpu features indicated e.g. via SCLP */
2786     rc = configure_cpu_feat(model->features);
2787     if (rc) {
2788         error_setg(errp, "KVM: Error configuring CPU features: %d", rc);
2789         return;
2790     }
2791     /* configure cpu subfunctions indicated via query / test bit */
2792     rc = configure_cpu_subfunc(model->features);
2793     if (rc) {
2794         error_setg(errp, "KVM: Error configuring CPU subfunctions: %d", rc);
2795         return;
2796     }
2797     /* enable CMM via CMMA */
2798     if (test_bit(S390_FEAT_CMM, model->features)) {
2799         kvm_s390_enable_cmma();
2800     }
2801 }
2802