xref: /qemu/target/s390x/kvm/kvm.c (revision 27e84d4ebd25b981ab27cb590fe06d1b0fcd06d2)
1 /*
2  * QEMU S390x KVM implementation
3  *
4  * Copyright (c) 2009 Alexander Graf <agraf@suse.de>
5  * Copyright IBM Corp. 2012
6  *
7  * This library is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2 of the License, or (at your option) any later version.
11  *
12  * This library is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * Lesser General Public License for more details.
16  *
17  * Contributions after 2012-10-29 are licensed under the terms of the
18  * GNU GPL, version 2 or (at your option) any later version.
19  *
20  * You should have received a copy of the GNU (Lesser) General Public
21  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
22  */
23 
24 #include "qemu/osdep.h"
25 #include <sys/ioctl.h>
26 
27 #include <linux/kvm.h>
28 #include <asm/ptrace.h>
29 
30 #include "qemu-common.h"
31 #include "cpu.h"
32 #include "internal.h"
33 #include "kvm_s390x.h"
34 #include "qapi/error.h"
35 #include "qemu/error-report.h"
36 #include "qemu/timer.h"
37 #include "sysemu/sysemu.h"
38 #include "sysemu/hw_accel.h"
39 #include "hw/hw.h"
40 #include "sysemu/device_tree.h"
41 #include "exec/gdbstub.h"
42 #include "trace.h"
43 #include "hw/s390x/s390-pci-inst.h"
44 #include "hw/s390x/s390-pci-bus.h"
45 #include "hw/s390x/ipl.h"
46 #include "hw/s390x/ebcdic.h"
47 #include "exec/memattrs.h"
48 #include "hw/s390x/s390-virtio-ccw.h"
49 #include "hw/s390x/s390-virtio-hcall.h"
50 
51 #ifndef DEBUG_KVM
52 #define DEBUG_KVM  0
53 #endif
54 
55 #define DPRINTF(fmt, ...) do {                \
56     if (DEBUG_KVM) {                          \
57         fprintf(stderr, fmt, ## __VA_ARGS__); \
58     }                                         \
59 } while (0)
60 
61 #define kvm_vm_check_mem_attr(s, attr) \
62     kvm_vm_check_attr(s, KVM_S390_VM_MEM_CTRL, attr)
63 
64 #define IPA0_DIAG                       0x8300
65 #define IPA0_SIGP                       0xae00
66 #define IPA0_B2                         0xb200
67 #define IPA0_B9                         0xb900
68 #define IPA0_EB                         0xeb00
69 #define IPA0_E3                         0xe300
70 
71 #define PRIV_B2_SCLP_CALL               0x20
72 #define PRIV_B2_CSCH                    0x30
73 #define PRIV_B2_HSCH                    0x31
74 #define PRIV_B2_MSCH                    0x32
75 #define PRIV_B2_SSCH                    0x33
76 #define PRIV_B2_STSCH                   0x34
77 #define PRIV_B2_TSCH                    0x35
78 #define PRIV_B2_TPI                     0x36
79 #define PRIV_B2_SAL                     0x37
80 #define PRIV_B2_RSCH                    0x38
81 #define PRIV_B2_STCRW                   0x39
82 #define PRIV_B2_STCPS                   0x3a
83 #define PRIV_B2_RCHP                    0x3b
84 #define PRIV_B2_SCHM                    0x3c
85 #define PRIV_B2_CHSC                    0x5f
86 #define PRIV_B2_SIGA                    0x74
87 #define PRIV_B2_XSCH                    0x76
88 
89 #define PRIV_EB_SQBS                    0x8a
90 #define PRIV_EB_PCISTB                  0xd0
91 #define PRIV_EB_SIC                     0xd1
92 
93 #define PRIV_B9_EQBS                    0x9c
94 #define PRIV_B9_CLP                     0xa0
95 #define PRIV_B9_PCISTG                  0xd0
96 #define PRIV_B9_PCILG                   0xd2
97 #define PRIV_B9_RPCIT                   0xd3
98 
99 #define PRIV_E3_MPCIFC                  0xd0
100 #define PRIV_E3_STPCIFC                 0xd4
101 
102 #define DIAG_TIMEREVENT                 0x288
103 #define DIAG_IPL                        0x308
104 #define DIAG_KVM_HYPERCALL              0x500
105 #define DIAG_KVM_BREAKPOINT             0x501
106 
107 #define ICPT_INSTRUCTION                0x04
108 #define ICPT_PROGRAM                    0x08
109 #define ICPT_EXT_INT                    0x14
110 #define ICPT_WAITPSW                    0x1c
111 #define ICPT_SOFT_INTERCEPT             0x24
112 #define ICPT_CPU_STOP                   0x28
113 #define ICPT_OPEREXC                    0x2c
114 #define ICPT_IO                         0x40
115 
116 #define NR_LOCAL_IRQS 32
117 /*
118  * Needs to be big enough to contain max_cpus emergency signals
119  * and in addition NR_LOCAL_IRQS interrupts
120  */
121 #define VCPU_IRQ_BUF_SIZE (sizeof(struct kvm_s390_irq) * \
122                            (max_cpus + NR_LOCAL_IRQS))
123 
124 static CPUWatchpoint hw_watchpoint;
125 /*
126  * We don't use a list because this structure is also used to transmit the
127  * hardware breakpoints to the kernel.
128  */
129 static struct kvm_hw_breakpoint *hw_breakpoints;
130 static int nb_hw_breakpoints;
131 
132 const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
133     KVM_CAP_LAST_INFO
134 };
135 
136 static int cap_sync_regs;
137 static int cap_async_pf;
138 static int cap_mem_op;
139 static int cap_s390_irq;
140 static int cap_ri;
141 static int cap_gs;
142 
143 static int active_cmma;
144 
145 static void *legacy_s390_alloc(size_t size, uint64_t *align, bool shared);
146 
147 static int kvm_s390_query_mem_limit(uint64_t *memory_limit)
148 {
149     struct kvm_device_attr attr = {
150         .group = KVM_S390_VM_MEM_CTRL,
151         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
152         .addr = (uint64_t) memory_limit,
153     };
154 
155     return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
156 }
157 
158 int kvm_s390_set_mem_limit(uint64_t new_limit, uint64_t *hw_limit)
159 {
160     int rc;
161 
162     struct kvm_device_attr attr = {
163         .group = KVM_S390_VM_MEM_CTRL,
164         .attr = KVM_S390_VM_MEM_LIMIT_SIZE,
165         .addr = (uint64_t) &new_limit,
166     };
167 
168     if (!kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_LIMIT_SIZE)) {
169         return 0;
170     }
171 
172     rc = kvm_s390_query_mem_limit(hw_limit);
173     if (rc) {
174         return rc;
175     } else if (*hw_limit < new_limit) {
176         return -E2BIG;
177     }
178 
179     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
180 }
181 
182 int kvm_s390_cmma_active(void)
183 {
184     return active_cmma;
185 }
186 
187 static bool kvm_s390_cmma_available(void)
188 {
189     static bool initialized, value;
190 
191     if (!initialized) {
192         initialized = true;
193         value = kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_ENABLE_CMMA) &&
194                 kvm_vm_check_mem_attr(kvm_state, KVM_S390_VM_MEM_CLR_CMMA);
195     }
196     return value;
197 }
198 
199 void kvm_s390_cmma_reset(void)
200 {
201     int rc;
202     struct kvm_device_attr attr = {
203         .group = KVM_S390_VM_MEM_CTRL,
204         .attr = KVM_S390_VM_MEM_CLR_CMMA,
205     };
206 
207     if (!kvm_s390_cmma_active()) {
208         return;
209     }
210 
211     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
212     trace_kvm_clear_cmma(rc);
213 }
214 
215 static void kvm_s390_enable_cmma(void)
216 {
217     int rc;
218     struct kvm_device_attr attr = {
219         .group = KVM_S390_VM_MEM_CTRL,
220         .attr = KVM_S390_VM_MEM_ENABLE_CMMA,
221     };
222 
223     if (mem_path) {
224         warn_report("CMM will not be enabled because it is not "
225                     "compatible with hugetlbfs.");
226         return;
227     }
228     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
229     active_cmma = !rc;
230     trace_kvm_enable_cmma(rc);
231 }
232 
233 static void kvm_s390_set_attr(uint64_t attr)
234 {
235     struct kvm_device_attr attribute = {
236         .group = KVM_S390_VM_CRYPTO,
237         .attr  = attr,
238     };
239 
240     int ret = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attribute);
241 
242     if (ret) {
243         error_report("Failed to set crypto device attribute %lu: %s",
244                      attr, strerror(-ret));
245     }
246 }
247 
248 static void kvm_s390_init_aes_kw(void)
249 {
250     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_AES_KW;
251 
252     if (object_property_get_bool(OBJECT(qdev_get_machine()), "aes-key-wrap",
253                                  NULL)) {
254             attr = KVM_S390_VM_CRYPTO_ENABLE_AES_KW;
255     }
256 
257     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
258             kvm_s390_set_attr(attr);
259     }
260 }
261 
262 static void kvm_s390_init_dea_kw(void)
263 {
264     uint64_t attr = KVM_S390_VM_CRYPTO_DISABLE_DEA_KW;
265 
266     if (object_property_get_bool(OBJECT(qdev_get_machine()), "dea-key-wrap",
267                                  NULL)) {
268             attr = KVM_S390_VM_CRYPTO_ENABLE_DEA_KW;
269     }
270 
271     if (kvm_vm_check_attr(kvm_state, KVM_S390_VM_CRYPTO, attr)) {
272             kvm_s390_set_attr(attr);
273     }
274 }
275 
276 void kvm_s390_crypto_reset(void)
277 {
278     if (s390_has_feat(S390_FEAT_MSA_EXT_3)) {
279         kvm_s390_init_aes_kw();
280         kvm_s390_init_dea_kw();
281     }
282 }
283 
284 int kvm_arch_init(MachineState *ms, KVMState *s)
285 {
286     MachineClass *mc = MACHINE_GET_CLASS(ms);
287 
288     mc->default_cpu_type = S390_CPU_TYPE_NAME("host");
289     cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
290     cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
291     cap_mem_op = kvm_check_extension(s, KVM_CAP_S390_MEM_OP);
292     cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
293 
294     if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
295         || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
296         phys_mem_set_alloc(legacy_s390_alloc);
297     }
298 
299     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_SIGP, 0);
300     kvm_vm_enable_cap(s, KVM_CAP_S390_VECTOR_REGISTERS, 0);
301     kvm_vm_enable_cap(s, KVM_CAP_S390_USER_STSI, 0);
302     if (ri_allowed()) {
303         if (kvm_vm_enable_cap(s, KVM_CAP_S390_RI, 0) == 0) {
304             cap_ri = 1;
305         }
306     }
307     if (cpu_model_allowed()) {
308         if (kvm_vm_enable_cap(s, KVM_CAP_S390_GS, 0) == 0) {
309             cap_gs = 1;
310         }
311     }
312 
313     /*
314      * The migration interface for ais was introduced with kernel 4.13
315      * but the capability itself had been active since 4.12. As migration
316      * support is considered necessary let's disable ais in the 2.10
317      * machine.
318      */
319     /* kvm_vm_enable_cap(s, KVM_CAP_S390_AIS, 0); */
320 
321     return 0;
322 }
323 
324 int kvm_arch_irqchip_create(MachineState *ms, KVMState *s)
325 {
326     return 0;
327 }
328 
329 unsigned long kvm_arch_vcpu_id(CPUState *cpu)
330 {
331     return cpu->cpu_index;
332 }
333 
334 int kvm_arch_init_vcpu(CPUState *cs)
335 {
336     S390CPU *cpu = S390_CPU(cs);
337     kvm_s390_set_cpu_state(cpu, cpu->env.cpu_state);
338     cpu->irqstate = g_malloc0(VCPU_IRQ_BUF_SIZE);
339     return 0;
340 }
341 
342 void kvm_s390_reset_vcpu(S390CPU *cpu)
343 {
344     CPUState *cs = CPU(cpu);
345 
346     /* The initial reset call is needed here to reset in-kernel
347      * vcpu data that we can't access directly from QEMU
348      * (i.e. with older kernels which don't support sync_regs/ONE_REG).
349      * Before this ioctl cpu_synchronize_state() is called in common kvm
350      * code (kvm-all) */
351     if (kvm_vcpu_ioctl(cs, KVM_S390_INITIAL_RESET, NULL)) {
352         error_report("Initial CPU reset failed on CPU %i", cs->cpu_index);
353     }
354 }
355 
356 static int can_sync_regs(CPUState *cs, int regs)
357 {
358     return cap_sync_regs && (cs->kvm_run->kvm_valid_regs & regs) == regs;
359 }
360 
361 int kvm_arch_put_registers(CPUState *cs, int level)
362 {
363     S390CPU *cpu = S390_CPU(cs);
364     CPUS390XState *env = &cpu->env;
365     struct kvm_sregs sregs;
366     struct kvm_regs regs;
367     struct kvm_fpu fpu = {};
368     int r;
369     int i;
370 
371     /* always save the PSW  and the GPRS*/
372     cs->kvm_run->psw_addr = env->psw.addr;
373     cs->kvm_run->psw_mask = env->psw.mask;
374 
375     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
376         for (i = 0; i < 16; i++) {
377             cs->kvm_run->s.regs.gprs[i] = env->regs[i];
378             cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GPRS;
379         }
380     } else {
381         for (i = 0; i < 16; i++) {
382             regs.gprs[i] = env->regs[i];
383         }
384         r = kvm_vcpu_ioctl(cs, KVM_SET_REGS, &regs);
385         if (r < 0) {
386             return r;
387         }
388     }
389 
390     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
391         for (i = 0; i < 32; i++) {
392             cs->kvm_run->s.regs.vrs[i][0] = env->vregs[i][0].ll;
393             cs->kvm_run->s.regs.vrs[i][1] = env->vregs[i][1].ll;
394         }
395         cs->kvm_run->s.regs.fpc = env->fpc;
396         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_VRS;
397     } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
398         for (i = 0; i < 16; i++) {
399             cs->kvm_run->s.regs.fprs[i] = get_freg(env, i)->ll;
400         }
401         cs->kvm_run->s.regs.fpc = env->fpc;
402         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_FPRS;
403     } else {
404         /* Floating point */
405         for (i = 0; i < 16; i++) {
406             fpu.fprs[i] = get_freg(env, i)->ll;
407         }
408         fpu.fpc = env->fpc;
409 
410         r = kvm_vcpu_ioctl(cs, KVM_SET_FPU, &fpu);
411         if (r < 0) {
412             return r;
413         }
414     }
415 
416     /* Do we need to save more than that? */
417     if (level == KVM_PUT_RUNTIME_STATE) {
418         return 0;
419     }
420 
421     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
422         cs->kvm_run->s.regs.cputm = env->cputm;
423         cs->kvm_run->s.regs.ckc = env->ckc;
424         cs->kvm_run->s.regs.todpr = env->todpr;
425         cs->kvm_run->s.regs.gbea = env->gbea;
426         cs->kvm_run->s.regs.pp = env->pp;
427         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ARCH0;
428     } else {
429         /*
430          * These ONE_REGS are not protected by a capability. As they are only
431          * necessary for migration we just trace a possible error, but don't
432          * return with an error return code.
433          */
434         kvm_set_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
435         kvm_set_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
436         kvm_set_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
437         kvm_set_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
438         kvm_set_one_reg(cs, KVM_REG_S390_PP, &env->pp);
439     }
440 
441     if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
442         memcpy(cs->kvm_run->s.regs.riccb, env->riccb, 64);
443         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_RICCB;
444     }
445 
446     /* pfault parameters */
447     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
448         cs->kvm_run->s.regs.pft = env->pfault_token;
449         cs->kvm_run->s.regs.pfs = env->pfault_select;
450         cs->kvm_run->s.regs.pfc = env->pfault_compare;
451         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PFAULT;
452     } else if (cap_async_pf) {
453         r = kvm_set_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
454         if (r < 0) {
455             return r;
456         }
457         r = kvm_set_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
458         if (r < 0) {
459             return r;
460         }
461         r = kvm_set_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
462         if (r < 0) {
463             return r;
464         }
465     }
466 
467     /* access registers and control registers*/
468     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
469         for (i = 0; i < 16; i++) {
470             cs->kvm_run->s.regs.acrs[i] = env->aregs[i];
471             cs->kvm_run->s.regs.crs[i] = env->cregs[i];
472         }
473         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ACRS;
474         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_CRS;
475     } else {
476         for (i = 0; i < 16; i++) {
477             sregs.acrs[i] = env->aregs[i];
478             sregs.crs[i] = env->cregs[i];
479         }
480         r = kvm_vcpu_ioctl(cs, KVM_SET_SREGS, &sregs);
481         if (r < 0) {
482             return r;
483         }
484     }
485 
486     if (can_sync_regs(cs, KVM_SYNC_GSCB)) {
487         memcpy(cs->kvm_run->s.regs.gscb, env->gscb, 32);
488         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_GSCB;
489     }
490 
491     if (can_sync_regs(cs, KVM_SYNC_BPBC)) {
492         cs->kvm_run->s.regs.bpbc = env->bpbc;
493         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_BPBC;
494     }
495 
496     if (can_sync_regs(cs, KVM_SYNC_ETOKEN)) {
497         cs->kvm_run->s.regs.etoken = env->etoken;
498         cs->kvm_run->s.regs.etoken_extension  = env->etoken_extension;
499         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_ETOKEN;
500     }
501 
502     /* Finally the prefix */
503     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
504         cs->kvm_run->s.regs.prefix = env->psa;
505         cs->kvm_run->kvm_dirty_regs |= KVM_SYNC_PREFIX;
506     } else {
507         /* prefix is only supported via sync regs */
508     }
509     return 0;
510 }
511 
512 int kvm_arch_get_registers(CPUState *cs)
513 {
514     S390CPU *cpu = S390_CPU(cs);
515     CPUS390XState *env = &cpu->env;
516     struct kvm_sregs sregs;
517     struct kvm_regs regs;
518     struct kvm_fpu fpu;
519     int i, r;
520 
521     /* get the PSW */
522     env->psw.addr = cs->kvm_run->psw_addr;
523     env->psw.mask = cs->kvm_run->psw_mask;
524 
525     /* the GPRS */
526     if (can_sync_regs(cs, KVM_SYNC_GPRS)) {
527         for (i = 0; i < 16; i++) {
528             env->regs[i] = cs->kvm_run->s.regs.gprs[i];
529         }
530     } else {
531         r = kvm_vcpu_ioctl(cs, KVM_GET_REGS, &regs);
532         if (r < 0) {
533             return r;
534         }
535          for (i = 0; i < 16; i++) {
536             env->regs[i] = regs.gprs[i];
537         }
538     }
539 
540     /* The ACRS and CRS */
541     if (can_sync_regs(cs, KVM_SYNC_ACRS | KVM_SYNC_CRS)) {
542         for (i = 0; i < 16; i++) {
543             env->aregs[i] = cs->kvm_run->s.regs.acrs[i];
544             env->cregs[i] = cs->kvm_run->s.regs.crs[i];
545         }
546     } else {
547         r = kvm_vcpu_ioctl(cs, KVM_GET_SREGS, &sregs);
548         if (r < 0) {
549             return r;
550         }
551          for (i = 0; i < 16; i++) {
552             env->aregs[i] = sregs.acrs[i];
553             env->cregs[i] = sregs.crs[i];
554         }
555     }
556 
557     /* Floating point and vector registers */
558     if (can_sync_regs(cs, KVM_SYNC_VRS)) {
559         for (i = 0; i < 32; i++) {
560             env->vregs[i][0].ll = cs->kvm_run->s.regs.vrs[i][0];
561             env->vregs[i][1].ll = cs->kvm_run->s.regs.vrs[i][1];
562         }
563         env->fpc = cs->kvm_run->s.regs.fpc;
564     } else if (can_sync_regs(cs, KVM_SYNC_FPRS)) {
565         for (i = 0; i < 16; i++) {
566             get_freg(env, i)->ll = cs->kvm_run->s.regs.fprs[i];
567         }
568         env->fpc = cs->kvm_run->s.regs.fpc;
569     } else {
570         r = kvm_vcpu_ioctl(cs, KVM_GET_FPU, &fpu);
571         if (r < 0) {
572             return r;
573         }
574         for (i = 0; i < 16; i++) {
575             get_freg(env, i)->ll = fpu.fprs[i];
576         }
577         env->fpc = fpu.fpc;
578     }
579 
580     /* The prefix */
581     if (can_sync_regs(cs, KVM_SYNC_PREFIX)) {
582         env->psa = cs->kvm_run->s.regs.prefix;
583     }
584 
585     if (can_sync_regs(cs, KVM_SYNC_ARCH0)) {
586         env->cputm = cs->kvm_run->s.regs.cputm;
587         env->ckc = cs->kvm_run->s.regs.ckc;
588         env->todpr = cs->kvm_run->s.regs.todpr;
589         env->gbea = cs->kvm_run->s.regs.gbea;
590         env->pp = cs->kvm_run->s.regs.pp;
591     } else {
592         /*
593          * These ONE_REGS are not protected by a capability. As they are only
594          * necessary for migration we just trace a possible error, but don't
595          * return with an error return code.
596          */
597         kvm_get_one_reg(cs, KVM_REG_S390_CPU_TIMER, &env->cputm);
598         kvm_get_one_reg(cs, KVM_REG_S390_CLOCK_COMP, &env->ckc);
599         kvm_get_one_reg(cs, KVM_REG_S390_TODPR, &env->todpr);
600         kvm_get_one_reg(cs, KVM_REG_S390_GBEA, &env->gbea);
601         kvm_get_one_reg(cs, KVM_REG_S390_PP, &env->pp);
602     }
603 
604     if (can_sync_regs(cs, KVM_SYNC_RICCB)) {
605         memcpy(env->riccb, cs->kvm_run->s.regs.riccb, 64);
606     }
607 
608     if (can_sync_regs(cs, KVM_SYNC_GSCB)) {
609         memcpy(env->gscb, cs->kvm_run->s.regs.gscb, 32);
610     }
611 
612     if (can_sync_regs(cs, KVM_SYNC_BPBC)) {
613         env->bpbc = cs->kvm_run->s.regs.bpbc;
614     }
615 
616     if (can_sync_regs(cs, KVM_SYNC_ETOKEN)) {
617         env->etoken = cs->kvm_run->s.regs.etoken;
618         env->etoken_extension = cs->kvm_run->s.regs.etoken_extension;
619     }
620 
621     /* pfault parameters */
622     if (can_sync_regs(cs, KVM_SYNC_PFAULT)) {
623         env->pfault_token = cs->kvm_run->s.regs.pft;
624         env->pfault_select = cs->kvm_run->s.regs.pfs;
625         env->pfault_compare = cs->kvm_run->s.regs.pfc;
626     } else if (cap_async_pf) {
627         r = kvm_get_one_reg(cs, KVM_REG_S390_PFTOKEN, &env->pfault_token);
628         if (r < 0) {
629             return r;
630         }
631         r = kvm_get_one_reg(cs, KVM_REG_S390_PFCOMPARE, &env->pfault_compare);
632         if (r < 0) {
633             return r;
634         }
635         r = kvm_get_one_reg(cs, KVM_REG_S390_PFSELECT, &env->pfault_select);
636         if (r < 0) {
637             return r;
638         }
639     }
640 
641     return 0;
642 }
643 
644 int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_low)
645 {
646     int r;
647     struct kvm_device_attr attr = {
648         .group = KVM_S390_VM_TOD,
649         .attr = KVM_S390_VM_TOD_LOW,
650         .addr = (uint64_t)tod_low,
651     };
652 
653     r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
654     if (r) {
655         return r;
656     }
657 
658     attr.attr = KVM_S390_VM_TOD_HIGH;
659     attr.addr = (uint64_t)tod_high;
660     return kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
661 }
662 
663 int kvm_s390_get_clock_ext(uint8_t *tod_high, uint64_t *tod_low)
664 {
665     int r;
666     struct kvm_s390_vm_tod_clock gtod;
667     struct kvm_device_attr attr = {
668         .group = KVM_S390_VM_TOD,
669         .attr = KVM_S390_VM_TOD_EXT,
670         .addr = (uint64_t)&gtod,
671     };
672 
673     r = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
674     *tod_high = gtod.epoch_idx;
675     *tod_low  = gtod.tod;
676 
677     return r;
678 }
679 
680 int kvm_s390_set_clock(uint8_t tod_high, uint64_t tod_low)
681 {
682     int r;
683     struct kvm_device_attr attr = {
684         .group = KVM_S390_VM_TOD,
685         .attr = KVM_S390_VM_TOD_LOW,
686         .addr = (uint64_t)&tod_low,
687     };
688 
689     r = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
690     if (r) {
691         return r;
692     }
693 
694     attr.attr = KVM_S390_VM_TOD_HIGH;
695     attr.addr = (uint64_t)&tod_high;
696     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
697 }
698 
699 int kvm_s390_set_clock_ext(uint8_t tod_high, uint64_t tod_low)
700 {
701     struct kvm_s390_vm_tod_clock gtod = {
702         .epoch_idx = tod_high,
703         .tod  = tod_low,
704     };
705     struct kvm_device_attr attr = {
706         .group = KVM_S390_VM_TOD,
707         .attr = KVM_S390_VM_TOD_EXT,
708         .addr = (uint64_t)&gtod,
709     };
710 
711     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
712 }
713 
714 /**
715  * kvm_s390_mem_op:
716  * @addr:      the logical start address in guest memory
717  * @ar:        the access register number
718  * @hostbuf:   buffer in host memory. NULL = do only checks w/o copying
719  * @len:       length that should be transferred
720  * @is_write:  true = write, false = read
721  * Returns:    0 on success, non-zero if an exception or error occurred
722  *
723  * Use KVM ioctl to read/write from/to guest memory. An access exception
724  * is injected into the vCPU in case of translation errors.
725  */
726 int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, uint8_t ar, void *hostbuf,
727                     int len, bool is_write)
728 {
729     struct kvm_s390_mem_op mem_op = {
730         .gaddr = addr,
731         .flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION,
732         .size = len,
733         .op = is_write ? KVM_S390_MEMOP_LOGICAL_WRITE
734                        : KVM_S390_MEMOP_LOGICAL_READ,
735         .buf = (uint64_t)hostbuf,
736         .ar = ar,
737     };
738     int ret;
739 
740     if (!cap_mem_op) {
741         return -ENOSYS;
742     }
743     if (!hostbuf) {
744         mem_op.flags |= KVM_S390_MEMOP_F_CHECK_ONLY;
745     }
746 
747     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
748     if (ret < 0) {
749         error_printf("KVM_S390_MEM_OP failed: %s\n", strerror(-ret));
750     }
751     return ret;
752 }
753 
754 /*
755  * Legacy layout for s390:
756  * Older S390 KVM requires the topmost vma of the RAM to be
757  * smaller than an system defined value, which is at least 256GB.
758  * Larger systems have larger values. We put the guest between
759  * the end of data segment (system break) and this value. We
760  * use 32GB as a base to have enough room for the system break
761  * to grow. We also have to use MAP parameters that avoid
762  * read-only mapping of guest pages.
763  */
764 static void *legacy_s390_alloc(size_t size, uint64_t *align, bool shared)
765 {
766     static void *mem;
767 
768     if (mem) {
769         /* we only support one allocation, which is enough for initial ram */
770         return NULL;
771     }
772 
773     mem = mmap((void *) 0x800000000ULL, size,
774                PROT_EXEC|PROT_READ|PROT_WRITE,
775                MAP_SHARED | MAP_ANONYMOUS | MAP_FIXED, -1, 0);
776     if (mem == MAP_FAILED) {
777         mem = NULL;
778     }
779     if (mem && align) {
780         *align = QEMU_VMALLOC_ALIGN;
781     }
782     return mem;
783 }
784 
785 static uint8_t const *sw_bp_inst;
786 static uint8_t sw_bp_ilen;
787 
788 static void determine_sw_breakpoint_instr(void)
789 {
790         /* DIAG 501 is used for sw breakpoints with old kernels */
791         static const uint8_t diag_501[] = {0x83, 0x24, 0x05, 0x01};
792         /* Instruction 0x0000 is used for sw breakpoints with recent kernels */
793         static const uint8_t instr_0x0000[] = {0x00, 0x00};
794 
795         if (sw_bp_inst) {
796             return;
797         }
798         if (kvm_vm_enable_cap(kvm_state, KVM_CAP_S390_USER_INSTR0, 0)) {
799             sw_bp_inst = diag_501;
800             sw_bp_ilen = sizeof(diag_501);
801             DPRINTF("KVM: will use 4-byte sw breakpoints.\n");
802         } else {
803             sw_bp_inst = instr_0x0000;
804             sw_bp_ilen = sizeof(instr_0x0000);
805             DPRINTF("KVM: will use 2-byte sw breakpoints.\n");
806         }
807 }
808 
809 int kvm_arch_insert_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
810 {
811     determine_sw_breakpoint_instr();
812 
813     if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
814                             sw_bp_ilen, 0) ||
815         cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)sw_bp_inst, sw_bp_ilen, 1)) {
816         return -EINVAL;
817     }
818     return 0;
819 }
820 
821 int kvm_arch_remove_sw_breakpoint(CPUState *cs, struct kvm_sw_breakpoint *bp)
822 {
823     uint8_t t[MAX_ILEN];
824 
825     if (cpu_memory_rw_debug(cs, bp->pc, t, sw_bp_ilen, 0)) {
826         return -EINVAL;
827     } else if (memcmp(t, sw_bp_inst, sw_bp_ilen)) {
828         return -EINVAL;
829     } else if (cpu_memory_rw_debug(cs, bp->pc, (uint8_t *)&bp->saved_insn,
830                                    sw_bp_ilen, 1)) {
831         return -EINVAL;
832     }
833 
834     return 0;
835 }
836 
837 static struct kvm_hw_breakpoint *find_hw_breakpoint(target_ulong addr,
838                                                     int len, int type)
839 {
840     int n;
841 
842     for (n = 0; n < nb_hw_breakpoints; n++) {
843         if (hw_breakpoints[n].addr == addr && hw_breakpoints[n].type == type &&
844             (hw_breakpoints[n].len == len || len == -1)) {
845             return &hw_breakpoints[n];
846         }
847     }
848 
849     return NULL;
850 }
851 
852 static int insert_hw_breakpoint(target_ulong addr, int len, int type)
853 {
854     int size;
855 
856     if (find_hw_breakpoint(addr, len, type)) {
857         return -EEXIST;
858     }
859 
860     size = (nb_hw_breakpoints + 1) * sizeof(struct kvm_hw_breakpoint);
861 
862     if (!hw_breakpoints) {
863         nb_hw_breakpoints = 0;
864         hw_breakpoints = (struct kvm_hw_breakpoint *)g_try_malloc(size);
865     } else {
866         hw_breakpoints =
867             (struct kvm_hw_breakpoint *)g_try_realloc(hw_breakpoints, size);
868     }
869 
870     if (!hw_breakpoints) {
871         nb_hw_breakpoints = 0;
872         return -ENOMEM;
873     }
874 
875     hw_breakpoints[nb_hw_breakpoints].addr = addr;
876     hw_breakpoints[nb_hw_breakpoints].len = len;
877     hw_breakpoints[nb_hw_breakpoints].type = type;
878 
879     nb_hw_breakpoints++;
880 
881     return 0;
882 }
883 
884 int kvm_arch_insert_hw_breakpoint(target_ulong addr,
885                                   target_ulong len, int type)
886 {
887     switch (type) {
888     case GDB_BREAKPOINT_HW:
889         type = KVM_HW_BP;
890         break;
891     case GDB_WATCHPOINT_WRITE:
892         if (len < 1) {
893             return -EINVAL;
894         }
895         type = KVM_HW_WP_WRITE;
896         break;
897     default:
898         return -ENOSYS;
899     }
900     return insert_hw_breakpoint(addr, len, type);
901 }
902 
903 int kvm_arch_remove_hw_breakpoint(target_ulong addr,
904                                   target_ulong len, int type)
905 {
906     int size;
907     struct kvm_hw_breakpoint *bp = find_hw_breakpoint(addr, len, type);
908 
909     if (bp == NULL) {
910         return -ENOENT;
911     }
912 
913     nb_hw_breakpoints--;
914     if (nb_hw_breakpoints > 0) {
915         /*
916          * In order to trim the array, move the last element to the position to
917          * be removed - if necessary.
918          */
919         if (bp != &hw_breakpoints[nb_hw_breakpoints]) {
920             *bp = hw_breakpoints[nb_hw_breakpoints];
921         }
922         size = nb_hw_breakpoints * sizeof(struct kvm_hw_breakpoint);
923         hw_breakpoints =
924              (struct kvm_hw_breakpoint *)g_realloc(hw_breakpoints, size);
925     } else {
926         g_free(hw_breakpoints);
927         hw_breakpoints = NULL;
928     }
929 
930     return 0;
931 }
932 
933 void kvm_arch_remove_all_hw_breakpoints(void)
934 {
935     nb_hw_breakpoints = 0;
936     g_free(hw_breakpoints);
937     hw_breakpoints = NULL;
938 }
939 
940 void kvm_arch_update_guest_debug(CPUState *cpu, struct kvm_guest_debug *dbg)
941 {
942     int i;
943 
944     if (nb_hw_breakpoints > 0) {
945         dbg->arch.nr_hw_bp = nb_hw_breakpoints;
946         dbg->arch.hw_bp = hw_breakpoints;
947 
948         for (i = 0; i < nb_hw_breakpoints; ++i) {
949             hw_breakpoints[i].phys_addr = s390_cpu_get_phys_addr_debug(cpu,
950                                                        hw_breakpoints[i].addr);
951         }
952         dbg->control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP;
953     } else {
954         dbg->arch.nr_hw_bp = 0;
955         dbg->arch.hw_bp = NULL;
956     }
957 }
958 
959 void kvm_arch_pre_run(CPUState *cpu, struct kvm_run *run)
960 {
961 }
962 
963 MemTxAttrs kvm_arch_post_run(CPUState *cs, struct kvm_run *run)
964 {
965     return MEMTXATTRS_UNSPECIFIED;
966 }
967 
968 int kvm_arch_process_async_events(CPUState *cs)
969 {
970     return cs->halted;
971 }
972 
973 static int s390_kvm_irq_to_interrupt(struct kvm_s390_irq *irq,
974                                      struct kvm_s390_interrupt *interrupt)
975 {
976     int r = 0;
977 
978     interrupt->type = irq->type;
979     switch (irq->type) {
980     case KVM_S390_INT_VIRTIO:
981         interrupt->parm = irq->u.ext.ext_params;
982         /* fall through */
983     case KVM_S390_INT_PFAULT_INIT:
984     case KVM_S390_INT_PFAULT_DONE:
985         interrupt->parm64 = irq->u.ext.ext_params2;
986         break;
987     case KVM_S390_PROGRAM_INT:
988         interrupt->parm = irq->u.pgm.code;
989         break;
990     case KVM_S390_SIGP_SET_PREFIX:
991         interrupt->parm = irq->u.prefix.address;
992         break;
993     case KVM_S390_INT_SERVICE:
994         interrupt->parm = irq->u.ext.ext_params;
995         break;
996     case KVM_S390_MCHK:
997         interrupt->parm = irq->u.mchk.cr14;
998         interrupt->parm64 = irq->u.mchk.mcic;
999         break;
1000     case KVM_S390_INT_EXTERNAL_CALL:
1001         interrupt->parm = irq->u.extcall.code;
1002         break;
1003     case KVM_S390_INT_EMERGENCY:
1004         interrupt->parm = irq->u.emerg.code;
1005         break;
1006     case KVM_S390_SIGP_STOP:
1007     case KVM_S390_RESTART:
1008         break; /* These types have no parameters */
1009     case KVM_S390_INT_IO_MIN...KVM_S390_INT_IO_MAX:
1010         interrupt->parm = irq->u.io.subchannel_id << 16;
1011         interrupt->parm |= irq->u.io.subchannel_nr;
1012         interrupt->parm64 = (uint64_t)irq->u.io.io_int_parm << 32;
1013         interrupt->parm64 |= irq->u.io.io_int_word;
1014         break;
1015     default:
1016         r = -EINVAL;
1017         break;
1018     }
1019     return r;
1020 }
1021 
1022 static void inject_vcpu_irq_legacy(CPUState *cs, struct kvm_s390_irq *irq)
1023 {
1024     struct kvm_s390_interrupt kvmint = {};
1025     int r;
1026 
1027     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
1028     if (r < 0) {
1029         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
1030         exit(1);
1031     }
1032 
1033     r = kvm_vcpu_ioctl(cs, KVM_S390_INTERRUPT, &kvmint);
1034     if (r < 0) {
1035         fprintf(stderr, "KVM failed to inject interrupt\n");
1036         exit(1);
1037     }
1038 }
1039 
1040 void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq)
1041 {
1042     CPUState *cs = CPU(cpu);
1043     int r;
1044 
1045     if (cap_s390_irq) {
1046         r = kvm_vcpu_ioctl(cs, KVM_S390_IRQ, irq);
1047         if (!r) {
1048             return;
1049         }
1050         error_report("KVM failed to inject interrupt %llx", irq->type);
1051         exit(1);
1052     }
1053 
1054     inject_vcpu_irq_legacy(cs, irq);
1055 }
1056 
1057 void kvm_s390_floating_interrupt_legacy(struct kvm_s390_irq *irq)
1058 {
1059     struct kvm_s390_interrupt kvmint = {};
1060     int r;
1061 
1062     r = s390_kvm_irq_to_interrupt(irq, &kvmint);
1063     if (r < 0) {
1064         fprintf(stderr, "%s called with bogus interrupt\n", __func__);
1065         exit(1);
1066     }
1067 
1068     r = kvm_vm_ioctl(kvm_state, KVM_S390_INTERRUPT, &kvmint);
1069     if (r < 0) {
1070         fprintf(stderr, "KVM failed to inject interrupt\n");
1071         exit(1);
1072     }
1073 }
1074 
1075 void kvm_s390_program_interrupt(S390CPU *cpu, uint16_t code)
1076 {
1077     struct kvm_s390_irq irq = {
1078         .type = KVM_S390_PROGRAM_INT,
1079         .u.pgm.code = code,
1080     };
1081 
1082     kvm_s390_vcpu_interrupt(cpu, &irq);
1083 }
1084 
1085 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code)
1086 {
1087     struct kvm_s390_irq irq = {
1088         .type = KVM_S390_PROGRAM_INT,
1089         .u.pgm.code = code,
1090         .u.pgm.trans_exc_code = te_code,
1091         .u.pgm.exc_access_id = te_code & 3,
1092     };
1093 
1094     kvm_s390_vcpu_interrupt(cpu, &irq);
1095 }
1096 
1097 static int kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
1098                                  uint16_t ipbh0)
1099 {
1100     CPUS390XState *env = &cpu->env;
1101     uint64_t sccb;
1102     uint32_t code;
1103     int r = 0;
1104 
1105     sccb = env->regs[ipbh0 & 0xf];
1106     code = env->regs[(ipbh0 & 0xf0) >> 4];
1107 
1108     r = sclp_service_call(env, sccb, code);
1109     if (r < 0) {
1110         kvm_s390_program_interrupt(cpu, -r);
1111     } else {
1112         setcc(cpu, r);
1113     }
1114 
1115     return 0;
1116 }
1117 
1118 static int handle_b2(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1119 {
1120     CPUS390XState *env = &cpu->env;
1121     int rc = 0;
1122     uint16_t ipbh0 = (run->s390_sieic.ipb & 0xffff0000) >> 16;
1123 
1124     switch (ipa1) {
1125     case PRIV_B2_XSCH:
1126         ioinst_handle_xsch(cpu, env->regs[1], RA_IGNORED);
1127         break;
1128     case PRIV_B2_CSCH:
1129         ioinst_handle_csch(cpu, env->regs[1], RA_IGNORED);
1130         break;
1131     case PRIV_B2_HSCH:
1132         ioinst_handle_hsch(cpu, env->regs[1], RA_IGNORED);
1133         break;
1134     case PRIV_B2_MSCH:
1135         ioinst_handle_msch(cpu, env->regs[1], run->s390_sieic.ipb, RA_IGNORED);
1136         break;
1137     case PRIV_B2_SSCH:
1138         ioinst_handle_ssch(cpu, env->regs[1], run->s390_sieic.ipb, RA_IGNORED);
1139         break;
1140     case PRIV_B2_STCRW:
1141         ioinst_handle_stcrw(cpu, run->s390_sieic.ipb, RA_IGNORED);
1142         break;
1143     case PRIV_B2_STSCH:
1144         ioinst_handle_stsch(cpu, env->regs[1], run->s390_sieic.ipb, RA_IGNORED);
1145         break;
1146     case PRIV_B2_TSCH:
1147         /* We should only get tsch via KVM_EXIT_S390_TSCH. */
1148         fprintf(stderr, "Spurious tsch intercept\n");
1149         break;
1150     case PRIV_B2_CHSC:
1151         ioinst_handle_chsc(cpu, run->s390_sieic.ipb, RA_IGNORED);
1152         break;
1153     case PRIV_B2_TPI:
1154         /* This should have been handled by kvm already. */
1155         fprintf(stderr, "Spurious tpi intercept\n");
1156         break;
1157     case PRIV_B2_SCHM:
1158         ioinst_handle_schm(cpu, env->regs[1], env->regs[2],
1159                            run->s390_sieic.ipb, RA_IGNORED);
1160         break;
1161     case PRIV_B2_RSCH:
1162         ioinst_handle_rsch(cpu, env->regs[1], RA_IGNORED);
1163         break;
1164     case PRIV_B2_RCHP:
1165         ioinst_handle_rchp(cpu, env->regs[1], RA_IGNORED);
1166         break;
1167     case PRIV_B2_STCPS:
1168         /* We do not provide this instruction, it is suppressed. */
1169         break;
1170     case PRIV_B2_SAL:
1171         ioinst_handle_sal(cpu, env->regs[1], RA_IGNORED);
1172         break;
1173     case PRIV_B2_SIGA:
1174         /* Not provided, set CC = 3 for subchannel not operational */
1175         setcc(cpu, 3);
1176         break;
1177     case PRIV_B2_SCLP_CALL:
1178         rc = kvm_sclp_service_call(cpu, run, ipbh0);
1179         break;
1180     default:
1181         rc = -1;
1182         DPRINTF("KVM: unhandled PRIV: 0xb2%x\n", ipa1);
1183         break;
1184     }
1185 
1186     return rc;
1187 }
1188 
1189 static uint64_t get_base_disp_rxy(S390CPU *cpu, struct kvm_run *run,
1190                                   uint8_t *ar)
1191 {
1192     CPUS390XState *env = &cpu->env;
1193     uint32_t x2 = (run->s390_sieic.ipa & 0x000f);
1194     uint32_t base2 = run->s390_sieic.ipb >> 28;
1195     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1196                      ((run->s390_sieic.ipb & 0xff00) << 4);
1197 
1198     if (disp2 & 0x80000) {
1199         disp2 += 0xfff00000;
1200     }
1201     if (ar) {
1202         *ar = base2;
1203     }
1204 
1205     return (base2 ? env->regs[base2] : 0) +
1206            (x2 ? env->regs[x2] : 0) + (long)(int)disp2;
1207 }
1208 
1209 static uint64_t get_base_disp_rsy(S390CPU *cpu, struct kvm_run *run,
1210                                   uint8_t *ar)
1211 {
1212     CPUS390XState *env = &cpu->env;
1213     uint32_t base2 = run->s390_sieic.ipb >> 28;
1214     uint32_t disp2 = ((run->s390_sieic.ipb & 0x0fff0000) >> 16) +
1215                      ((run->s390_sieic.ipb & 0xff00) << 4);
1216 
1217     if (disp2 & 0x80000) {
1218         disp2 += 0xfff00000;
1219     }
1220     if (ar) {
1221         *ar = base2;
1222     }
1223 
1224     return (base2 ? env->regs[base2] : 0) + (long)(int)disp2;
1225 }
1226 
1227 static int kvm_clp_service_call(S390CPU *cpu, struct kvm_run *run)
1228 {
1229     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1230 
1231     if (s390_has_feat(S390_FEAT_ZPCI)) {
1232         return clp_service_call(cpu, r2, RA_IGNORED);
1233     } else {
1234         return -1;
1235     }
1236 }
1237 
1238 static int kvm_pcilg_service_call(S390CPU *cpu, struct kvm_run *run)
1239 {
1240     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1241     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1242 
1243     if (s390_has_feat(S390_FEAT_ZPCI)) {
1244         return pcilg_service_call(cpu, r1, r2, RA_IGNORED);
1245     } else {
1246         return -1;
1247     }
1248 }
1249 
1250 static int kvm_pcistg_service_call(S390CPU *cpu, struct kvm_run *run)
1251 {
1252     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1253     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1254 
1255     if (s390_has_feat(S390_FEAT_ZPCI)) {
1256         return pcistg_service_call(cpu, r1, r2, RA_IGNORED);
1257     } else {
1258         return -1;
1259     }
1260 }
1261 
1262 static int kvm_stpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1263 {
1264     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1265     uint64_t fiba;
1266     uint8_t ar;
1267 
1268     if (s390_has_feat(S390_FEAT_ZPCI)) {
1269         fiba = get_base_disp_rxy(cpu, run, &ar);
1270 
1271         return stpcifc_service_call(cpu, r1, fiba, ar, RA_IGNORED);
1272     } else {
1273         return -1;
1274     }
1275 }
1276 
1277 static int kvm_sic_service_call(S390CPU *cpu, struct kvm_run *run)
1278 {
1279     CPUS390XState *env = &cpu->env;
1280     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1281     uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1282     uint8_t isc;
1283     uint16_t mode;
1284     int r;
1285 
1286     mode = env->regs[r1] & 0xffff;
1287     isc = (env->regs[r3] >> 27) & 0x7;
1288     r = css_do_sic(env, isc, mode);
1289     if (r) {
1290         kvm_s390_program_interrupt(cpu, -r);
1291     }
1292 
1293     return 0;
1294 }
1295 
1296 static int kvm_rpcit_service_call(S390CPU *cpu, struct kvm_run *run)
1297 {
1298     uint8_t r1 = (run->s390_sieic.ipb & 0x00f00000) >> 20;
1299     uint8_t r2 = (run->s390_sieic.ipb & 0x000f0000) >> 16;
1300 
1301     if (s390_has_feat(S390_FEAT_ZPCI)) {
1302         return rpcit_service_call(cpu, r1, r2, RA_IGNORED);
1303     } else {
1304         return -1;
1305     }
1306 }
1307 
1308 static int kvm_pcistb_service_call(S390CPU *cpu, struct kvm_run *run)
1309 {
1310     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1311     uint8_t r3 = run->s390_sieic.ipa & 0x000f;
1312     uint64_t gaddr;
1313     uint8_t ar;
1314 
1315     if (s390_has_feat(S390_FEAT_ZPCI)) {
1316         gaddr = get_base_disp_rsy(cpu, run, &ar);
1317 
1318         return pcistb_service_call(cpu, r1, r3, gaddr, ar, RA_IGNORED);
1319     } else {
1320         return -1;
1321     }
1322 }
1323 
1324 static int kvm_mpcifc_service_call(S390CPU *cpu, struct kvm_run *run)
1325 {
1326     uint8_t r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1327     uint64_t fiba;
1328     uint8_t ar;
1329 
1330     if (s390_has_feat(S390_FEAT_ZPCI)) {
1331         fiba = get_base_disp_rxy(cpu, run, &ar);
1332 
1333         return mpcifc_service_call(cpu, r1, fiba, ar, RA_IGNORED);
1334     } else {
1335         return -1;
1336     }
1337 }
1338 
1339 static int handle_b9(S390CPU *cpu, struct kvm_run *run, uint8_t ipa1)
1340 {
1341     int r = 0;
1342 
1343     switch (ipa1) {
1344     case PRIV_B9_CLP:
1345         r = kvm_clp_service_call(cpu, run);
1346         break;
1347     case PRIV_B9_PCISTG:
1348         r = kvm_pcistg_service_call(cpu, run);
1349         break;
1350     case PRIV_B9_PCILG:
1351         r = kvm_pcilg_service_call(cpu, run);
1352         break;
1353     case PRIV_B9_RPCIT:
1354         r = kvm_rpcit_service_call(cpu, run);
1355         break;
1356     case PRIV_B9_EQBS:
1357         /* just inject exception */
1358         r = -1;
1359         break;
1360     default:
1361         r = -1;
1362         DPRINTF("KVM: unhandled PRIV: 0xb9%x\n", ipa1);
1363         break;
1364     }
1365 
1366     return r;
1367 }
1368 
1369 static int handle_eb(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1370 {
1371     int r = 0;
1372 
1373     switch (ipbl) {
1374     case PRIV_EB_PCISTB:
1375         r = kvm_pcistb_service_call(cpu, run);
1376         break;
1377     case PRIV_EB_SIC:
1378         r = kvm_sic_service_call(cpu, run);
1379         break;
1380     case PRIV_EB_SQBS:
1381         /* just inject exception */
1382         r = -1;
1383         break;
1384     default:
1385         r = -1;
1386         DPRINTF("KVM: unhandled PRIV: 0xeb%x\n", ipbl);
1387         break;
1388     }
1389 
1390     return r;
1391 }
1392 
1393 static int handle_e3(S390CPU *cpu, struct kvm_run *run, uint8_t ipbl)
1394 {
1395     int r = 0;
1396 
1397     switch (ipbl) {
1398     case PRIV_E3_MPCIFC:
1399         r = kvm_mpcifc_service_call(cpu, run);
1400         break;
1401     case PRIV_E3_STPCIFC:
1402         r = kvm_stpcifc_service_call(cpu, run);
1403         break;
1404     default:
1405         r = -1;
1406         DPRINTF("KVM: unhandled PRIV: 0xe3%x\n", ipbl);
1407         break;
1408     }
1409 
1410     return r;
1411 }
1412 
1413 static int handle_hypercall(S390CPU *cpu, struct kvm_run *run)
1414 {
1415     CPUS390XState *env = &cpu->env;
1416     int ret;
1417 
1418     ret = s390_virtio_hypercall(env);
1419     if (ret == -EINVAL) {
1420         kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1421         return 0;
1422     }
1423 
1424     return ret;
1425 }
1426 
1427 static void kvm_handle_diag_288(S390CPU *cpu, struct kvm_run *run)
1428 {
1429     uint64_t r1, r3;
1430     int rc;
1431 
1432     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1433     r3 = run->s390_sieic.ipa & 0x000f;
1434     rc = handle_diag_288(&cpu->env, r1, r3);
1435     if (rc) {
1436         kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1437     }
1438 }
1439 
1440 static void kvm_handle_diag_308(S390CPU *cpu, struct kvm_run *run)
1441 {
1442     uint64_t r1, r3;
1443 
1444     r1 = (run->s390_sieic.ipa & 0x00f0) >> 4;
1445     r3 = run->s390_sieic.ipa & 0x000f;
1446     handle_diag_308(&cpu->env, r1, r3, RA_IGNORED);
1447 }
1448 
1449 static int handle_sw_breakpoint(S390CPU *cpu, struct kvm_run *run)
1450 {
1451     CPUS390XState *env = &cpu->env;
1452     unsigned long pc;
1453 
1454     pc = env->psw.addr - sw_bp_ilen;
1455     if (kvm_find_sw_breakpoint(CPU(cpu), pc)) {
1456         env->psw.addr = pc;
1457         return EXCP_DEBUG;
1458     }
1459 
1460     return -ENOENT;
1461 }
1462 
1463 #define DIAG_KVM_CODE_MASK 0x000000000000ffff
1464 
1465 static int handle_diag(S390CPU *cpu, struct kvm_run *run, uint32_t ipb)
1466 {
1467     int r = 0;
1468     uint16_t func_code;
1469 
1470     /*
1471      * For any diagnose call we support, bits 48-63 of the resulting
1472      * address specify the function code; the remainder is ignored.
1473      */
1474     func_code = decode_basedisp_rs(&cpu->env, ipb, NULL) & DIAG_KVM_CODE_MASK;
1475     switch (func_code) {
1476     case DIAG_TIMEREVENT:
1477         kvm_handle_diag_288(cpu, run);
1478         break;
1479     case DIAG_IPL:
1480         kvm_handle_diag_308(cpu, run);
1481         break;
1482     case DIAG_KVM_HYPERCALL:
1483         r = handle_hypercall(cpu, run);
1484         break;
1485     case DIAG_KVM_BREAKPOINT:
1486         r = handle_sw_breakpoint(cpu, run);
1487         break;
1488     default:
1489         DPRINTF("KVM: unknown DIAG: 0x%x\n", func_code);
1490         kvm_s390_program_interrupt(cpu, PGM_SPECIFICATION);
1491         break;
1492     }
1493 
1494     return r;
1495 }
1496 
1497 static int kvm_s390_handle_sigp(S390CPU *cpu, uint8_t ipa1, uint32_t ipb)
1498 {
1499     CPUS390XState *env = &cpu->env;
1500     const uint8_t r1 = ipa1 >> 4;
1501     const uint8_t r3 = ipa1 & 0x0f;
1502     int ret;
1503     uint8_t order;
1504 
1505     /* get order code */
1506     order = decode_basedisp_rs(env, ipb, NULL) & SIGP_ORDER_MASK;
1507 
1508     ret = handle_sigp(env, order, r1, r3);
1509     setcc(cpu, ret);
1510     return 0;
1511 }
1512 
1513 static int handle_instruction(S390CPU *cpu, struct kvm_run *run)
1514 {
1515     unsigned int ipa0 = (run->s390_sieic.ipa & 0xff00);
1516     uint8_t ipa1 = run->s390_sieic.ipa & 0x00ff;
1517     int r = -1;
1518 
1519     DPRINTF("handle_instruction 0x%x 0x%x\n",
1520             run->s390_sieic.ipa, run->s390_sieic.ipb);
1521     switch (ipa0) {
1522     case IPA0_B2:
1523         r = handle_b2(cpu, run, ipa1);
1524         break;
1525     case IPA0_B9:
1526         r = handle_b9(cpu, run, ipa1);
1527         break;
1528     case IPA0_EB:
1529         r = handle_eb(cpu, run, run->s390_sieic.ipb & 0xff);
1530         break;
1531     case IPA0_E3:
1532         r = handle_e3(cpu, run, run->s390_sieic.ipb & 0xff);
1533         break;
1534     case IPA0_DIAG:
1535         r = handle_diag(cpu, run, run->s390_sieic.ipb);
1536         break;
1537     case IPA0_SIGP:
1538         r = kvm_s390_handle_sigp(cpu, ipa1, run->s390_sieic.ipb);
1539         break;
1540     }
1541 
1542     if (r < 0) {
1543         r = 0;
1544         kvm_s390_program_interrupt(cpu, PGM_OPERATION);
1545     }
1546 
1547     return r;
1548 }
1549 
1550 static void unmanageable_intercept(S390CPU *cpu, S390CrashReason reason,
1551                                    int pswoffset)
1552 {
1553     CPUState *cs = CPU(cpu);
1554 
1555     s390_cpu_halt(cpu);
1556     cpu->env.crash_reason = reason;
1557     qemu_system_guest_panicked(cpu_get_crash_info(cs));
1558 }
1559 
1560 /* try to detect pgm check loops */
1561 static int handle_oper_loop(S390CPU *cpu, struct kvm_run *run)
1562 {
1563     CPUState *cs = CPU(cpu);
1564     PSW oldpsw, newpsw;
1565 
1566     newpsw.mask = ldq_phys(cs->as, cpu->env.psa +
1567                            offsetof(LowCore, program_new_psw));
1568     newpsw.addr = ldq_phys(cs->as, cpu->env.psa +
1569                            offsetof(LowCore, program_new_psw) + 8);
1570     oldpsw.mask  = run->psw_mask;
1571     oldpsw.addr  = run->psw_addr;
1572     /*
1573      * Avoid endless loops of operation exceptions, if the pgm new
1574      * PSW will cause a new operation exception.
1575      * The heuristic checks if the pgm new psw is within 6 bytes before
1576      * the faulting psw address (with same DAT, AS settings) and the
1577      * new psw is not a wait psw and the fault was not triggered by
1578      * problem state. In that case go into crashed state.
1579      */
1580 
1581     if (oldpsw.addr - newpsw.addr <= 6 &&
1582         !(newpsw.mask & PSW_MASK_WAIT) &&
1583         !(oldpsw.mask & PSW_MASK_PSTATE) &&
1584         (newpsw.mask & PSW_MASK_ASC) == (oldpsw.mask & PSW_MASK_ASC) &&
1585         (newpsw.mask & PSW_MASK_DAT) == (oldpsw.mask & PSW_MASK_DAT)) {
1586         unmanageable_intercept(cpu, S390_CRASH_REASON_OPINT_LOOP,
1587                                offsetof(LowCore, program_new_psw));
1588         return EXCP_HALTED;
1589     }
1590     return 0;
1591 }
1592 
1593 static int handle_intercept(S390CPU *cpu)
1594 {
1595     CPUState *cs = CPU(cpu);
1596     struct kvm_run *run = cs->kvm_run;
1597     int icpt_code = run->s390_sieic.icptcode;
1598     int r = 0;
1599 
1600     DPRINTF("intercept: 0x%x (at 0x%lx)\n", icpt_code,
1601             (long)cs->kvm_run->psw_addr);
1602     switch (icpt_code) {
1603         case ICPT_INSTRUCTION:
1604             r = handle_instruction(cpu, run);
1605             break;
1606         case ICPT_PROGRAM:
1607             unmanageable_intercept(cpu, S390_CRASH_REASON_PGMINT_LOOP,
1608                                    offsetof(LowCore, program_new_psw));
1609             r = EXCP_HALTED;
1610             break;
1611         case ICPT_EXT_INT:
1612             unmanageable_intercept(cpu, S390_CRASH_REASON_EXTINT_LOOP,
1613                                    offsetof(LowCore, external_new_psw));
1614             r = EXCP_HALTED;
1615             break;
1616         case ICPT_WAITPSW:
1617             /* disabled wait, since enabled wait is handled in kernel */
1618             s390_handle_wait(cpu);
1619             r = EXCP_HALTED;
1620             break;
1621         case ICPT_CPU_STOP:
1622             do_stop_interrupt(&cpu->env);
1623             r = EXCP_HALTED;
1624             break;
1625         case ICPT_OPEREXC:
1626             /* check for break points */
1627             r = handle_sw_breakpoint(cpu, run);
1628             if (r == -ENOENT) {
1629                 /* Then check for potential pgm check loops */
1630                 r = handle_oper_loop(cpu, run);
1631                 if (r == 0) {
1632                     kvm_s390_program_interrupt(cpu, PGM_OPERATION);
1633                 }
1634             }
1635             break;
1636         case ICPT_SOFT_INTERCEPT:
1637             fprintf(stderr, "KVM unimplemented icpt SOFT\n");
1638             exit(1);
1639             break;
1640         case ICPT_IO:
1641             fprintf(stderr, "KVM unimplemented icpt IO\n");
1642             exit(1);
1643             break;
1644         default:
1645             fprintf(stderr, "Unknown intercept code: %d\n", icpt_code);
1646             exit(1);
1647             break;
1648     }
1649 
1650     return r;
1651 }
1652 
1653 static int handle_tsch(S390CPU *cpu)
1654 {
1655     CPUState *cs = CPU(cpu);
1656     struct kvm_run *run = cs->kvm_run;
1657     int ret;
1658 
1659     ret = ioinst_handle_tsch(cpu, cpu->env.regs[1], run->s390_tsch.ipb,
1660                              RA_IGNORED);
1661     if (ret < 0) {
1662         /*
1663          * Failure.
1664          * If an I/O interrupt had been dequeued, we have to reinject it.
1665          */
1666         if (run->s390_tsch.dequeued) {
1667             s390_io_interrupt(run->s390_tsch.subchannel_id,
1668                               run->s390_tsch.subchannel_nr,
1669                               run->s390_tsch.io_int_parm,
1670                               run->s390_tsch.io_int_word);
1671         }
1672         ret = 0;
1673     }
1674     return ret;
1675 }
1676 
1677 static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
1678 {
1679     SysIB_322 sysib;
1680     int del;
1681 
1682     if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
1683         return;
1684     }
1685     /* Shift the stack of Extended Names to prepare for our own data */
1686     memmove(&sysib.ext_names[1], &sysib.ext_names[0],
1687             sizeof(sysib.ext_names[0]) * (sysib.count - 1));
1688     /* First virt level, that doesn't provide Ext Names delimits stack. It is
1689      * assumed it's not capable of managing Extended Names for lower levels.
1690      */
1691     for (del = 1; del < sysib.count; del++) {
1692         if (!sysib.vm[del].ext_name_encoding || !sysib.ext_names[del][0]) {
1693             break;
1694         }
1695     }
1696     if (del < sysib.count) {
1697         memset(sysib.ext_names[del], 0,
1698                sizeof(sysib.ext_names[0]) * (sysib.count - del));
1699     }
1700     /* Insert short machine name in EBCDIC, padded with blanks */
1701     if (qemu_name) {
1702         memset(sysib.vm[0].name, 0x40, sizeof(sysib.vm[0].name));
1703         ebcdic_put(sysib.vm[0].name, qemu_name, MIN(sizeof(sysib.vm[0].name),
1704                                                     strlen(qemu_name)));
1705     }
1706     sysib.vm[0].ext_name_encoding = 2; /* 2 = UTF-8 */
1707     memset(sysib.ext_names[0], 0, sizeof(sysib.ext_names[0]));
1708     /* If hypervisor specifies zero Extended Name in STSI322 SYSIB, it's
1709      * considered by s390 as not capable of providing any Extended Name.
1710      * Therefore if no name was specified on qemu invocation, we go with the
1711      * same "KVMguest" default, which KVM has filled into short name field.
1712      */
1713     if (qemu_name) {
1714         strncpy((char *)sysib.ext_names[0], qemu_name,
1715                 sizeof(sysib.ext_names[0]));
1716     } else {
1717         strcpy((char *)sysib.ext_names[0], "KVMguest");
1718     }
1719     /* Insert UUID */
1720     memcpy(sysib.vm[0].uuid, &qemu_uuid, sizeof(sysib.vm[0].uuid));
1721 
1722     s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
1723 }
1724 
1725 static int handle_stsi(S390CPU *cpu)
1726 {
1727     CPUState *cs = CPU(cpu);
1728     struct kvm_run *run = cs->kvm_run;
1729 
1730     switch (run->s390_stsi.fc) {
1731     case 3:
1732         if (run->s390_stsi.sel1 != 2 || run->s390_stsi.sel2 != 2) {
1733             return 0;
1734         }
1735         /* Only sysib 3.2.2 needs post-handling for now. */
1736         insert_stsi_3_2_2(cpu, run->s390_stsi.addr, run->s390_stsi.ar);
1737         return 0;
1738     default:
1739         return 0;
1740     }
1741 }
1742 
1743 static int kvm_arch_handle_debug_exit(S390CPU *cpu)
1744 {
1745     CPUState *cs = CPU(cpu);
1746     struct kvm_run *run = cs->kvm_run;
1747 
1748     int ret = 0;
1749     struct kvm_debug_exit_arch *arch_info = &run->debug.arch;
1750 
1751     switch (arch_info->type) {
1752     case KVM_HW_WP_WRITE:
1753         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1754             cs->watchpoint_hit = &hw_watchpoint;
1755             hw_watchpoint.vaddr = arch_info->addr;
1756             hw_watchpoint.flags = BP_MEM_WRITE;
1757             ret = EXCP_DEBUG;
1758         }
1759         break;
1760     case KVM_HW_BP:
1761         if (find_hw_breakpoint(arch_info->addr, -1, arch_info->type)) {
1762             ret = EXCP_DEBUG;
1763         }
1764         break;
1765     case KVM_SINGLESTEP:
1766         if (cs->singlestep_enabled) {
1767             ret = EXCP_DEBUG;
1768         }
1769         break;
1770     default:
1771         ret = -ENOSYS;
1772     }
1773 
1774     return ret;
1775 }
1776 
1777 int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
1778 {
1779     S390CPU *cpu = S390_CPU(cs);
1780     int ret = 0;
1781 
1782     qemu_mutex_lock_iothread();
1783 
1784     kvm_cpu_synchronize_state(cs);
1785 
1786     switch (run->exit_reason) {
1787         case KVM_EXIT_S390_SIEIC:
1788             ret = handle_intercept(cpu);
1789             break;
1790         case KVM_EXIT_S390_RESET:
1791             s390_ipl_reset_request(cs, S390_RESET_REIPL);
1792             break;
1793         case KVM_EXIT_S390_TSCH:
1794             ret = handle_tsch(cpu);
1795             break;
1796         case KVM_EXIT_S390_STSI:
1797             ret = handle_stsi(cpu);
1798             break;
1799         case KVM_EXIT_DEBUG:
1800             ret = kvm_arch_handle_debug_exit(cpu);
1801             break;
1802         default:
1803             fprintf(stderr, "Unknown KVM exit: %d\n", run->exit_reason);
1804             break;
1805     }
1806     qemu_mutex_unlock_iothread();
1807 
1808     if (ret == 0) {
1809         ret = EXCP_INTERRUPT;
1810     }
1811     return ret;
1812 }
1813 
1814 bool kvm_arch_stop_on_emulation_error(CPUState *cpu)
1815 {
1816     return true;
1817 }
1818 
1819 void kvm_s390_enable_css_support(S390CPU *cpu)
1820 {
1821     int r;
1822 
1823     /* Activate host kernel channel subsystem support. */
1824     r = kvm_vcpu_enable_cap(CPU(cpu), KVM_CAP_S390_CSS_SUPPORT, 0);
1825     assert(r == 0);
1826 }
1827 
1828 void kvm_arch_init_irq_routing(KVMState *s)
1829 {
1830     /*
1831      * Note that while irqchip capabilities generally imply that cpustates
1832      * are handled in-kernel, it is not true for s390 (yet); therefore, we
1833      * have to override the common code kvm_halt_in_kernel_allowed setting.
1834      */
1835     if (kvm_check_extension(s, KVM_CAP_IRQ_ROUTING)) {
1836         kvm_gsi_routing_allowed = true;
1837         kvm_halt_in_kernel_allowed = false;
1838     }
1839 }
1840 
1841 int kvm_s390_assign_subch_ioeventfd(EventNotifier *notifier, uint32_t sch,
1842                                     int vq, bool assign)
1843 {
1844     struct kvm_ioeventfd kick = {
1845         .flags = KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY |
1846         KVM_IOEVENTFD_FLAG_DATAMATCH,
1847         .fd = event_notifier_get_fd(notifier),
1848         .datamatch = vq,
1849         .addr = sch,
1850         .len = 8,
1851     };
1852     if (!kvm_check_extension(kvm_state, KVM_CAP_IOEVENTFD)) {
1853         return -ENOSYS;
1854     }
1855     if (!assign) {
1856         kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
1857     }
1858     return kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
1859 }
1860 
1861 int kvm_s390_get_ri(void)
1862 {
1863     return cap_ri;
1864 }
1865 
1866 int kvm_s390_get_gs(void)
1867 {
1868     return cap_gs;
1869 }
1870 
1871 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state)
1872 {
1873     struct kvm_mp_state mp_state = {};
1874     int ret;
1875 
1876     /* the kvm part might not have been initialized yet */
1877     if (CPU(cpu)->kvm_state == NULL) {
1878         return 0;
1879     }
1880 
1881     switch (cpu_state) {
1882     case S390_CPU_STATE_STOPPED:
1883         mp_state.mp_state = KVM_MP_STATE_STOPPED;
1884         break;
1885     case S390_CPU_STATE_CHECK_STOP:
1886         mp_state.mp_state = KVM_MP_STATE_CHECK_STOP;
1887         break;
1888     case S390_CPU_STATE_OPERATING:
1889         mp_state.mp_state = KVM_MP_STATE_OPERATING;
1890         break;
1891     case S390_CPU_STATE_LOAD:
1892         mp_state.mp_state = KVM_MP_STATE_LOAD;
1893         break;
1894     default:
1895         error_report("Requested CPU state is not a valid S390 CPU state: %u",
1896                      cpu_state);
1897         exit(1);
1898     }
1899 
1900     ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
1901     if (ret) {
1902         trace_kvm_failed_cpu_state_set(CPU(cpu)->cpu_index, cpu_state,
1903                                        strerror(-ret));
1904     }
1905 
1906     return ret;
1907 }
1908 
1909 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu)
1910 {
1911     struct kvm_s390_irq_state irq_state = {
1912         .buf = (uint64_t) cpu->irqstate,
1913         .len = VCPU_IRQ_BUF_SIZE,
1914     };
1915     CPUState *cs = CPU(cpu);
1916     int32_t bytes;
1917 
1918     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
1919         return;
1920     }
1921 
1922     bytes = kvm_vcpu_ioctl(cs, KVM_S390_GET_IRQ_STATE, &irq_state);
1923     if (bytes < 0) {
1924         cpu->irqstate_saved_size = 0;
1925         error_report("Migration of interrupt state failed");
1926         return;
1927     }
1928 
1929     cpu->irqstate_saved_size = bytes;
1930 }
1931 
1932 int kvm_s390_vcpu_interrupt_post_load(S390CPU *cpu)
1933 {
1934     CPUState *cs = CPU(cpu);
1935     struct kvm_s390_irq_state irq_state = {
1936         .buf = (uint64_t) cpu->irqstate,
1937         .len = cpu->irqstate_saved_size,
1938     };
1939     int r;
1940 
1941     if (cpu->irqstate_saved_size == 0) {
1942         return 0;
1943     }
1944 
1945     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_IRQ_STATE)) {
1946         return -ENOSYS;
1947     }
1948 
1949     r = kvm_vcpu_ioctl(cs, KVM_S390_SET_IRQ_STATE, &irq_state);
1950     if (r) {
1951         error_report("Setting interrupt state failed %d", r);
1952     }
1953     return r;
1954 }
1955 
1956 int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
1957                              uint64_t address, uint32_t data, PCIDevice *dev)
1958 {
1959     S390PCIBusDevice *pbdev;
1960     uint32_t vec = data & ZPCI_MSI_VEC_MASK;
1961 
1962     if (!dev) {
1963         DPRINTF("add_msi_route no pci device\n");
1964         return -ENODEV;
1965     }
1966 
1967     pbdev = s390_pci_find_dev_by_target(s390_get_phb(), DEVICE(dev)->id);
1968     if (!pbdev) {
1969         DPRINTF("add_msi_route no zpci device\n");
1970         return -ENODEV;
1971     }
1972 
1973     route->type = KVM_IRQ_ROUTING_S390_ADAPTER;
1974     route->flags = 0;
1975     route->u.adapter.summary_addr = pbdev->routes.adapter.summary_addr;
1976     route->u.adapter.ind_addr = pbdev->routes.adapter.ind_addr;
1977     route->u.adapter.summary_offset = pbdev->routes.adapter.summary_offset;
1978     route->u.adapter.ind_offset = pbdev->routes.adapter.ind_offset + vec;
1979     route->u.adapter.adapter_id = pbdev->routes.adapter.adapter_id;
1980     return 0;
1981 }
1982 
1983 int kvm_arch_add_msi_route_post(struct kvm_irq_routing_entry *route,
1984                                 int vector, PCIDevice *dev)
1985 {
1986     return 0;
1987 }
1988 
1989 int kvm_arch_release_virq_post(int virq)
1990 {
1991     return 0;
1992 }
1993 
1994 int kvm_arch_msi_data_to_gsi(uint32_t data)
1995 {
1996     abort();
1997 }
1998 
1999 static int query_cpu_subfunc(S390FeatBitmap features)
2000 {
2001     struct kvm_s390_vm_cpu_subfunc prop;
2002     struct kvm_device_attr attr = {
2003         .group = KVM_S390_VM_CPU_MODEL,
2004         .attr = KVM_S390_VM_CPU_MACHINE_SUBFUNC,
2005         .addr = (uint64_t) &prop,
2006     };
2007     int rc;
2008 
2009     rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2010     if (rc) {
2011         return  rc;
2012     }
2013 
2014     /*
2015      * We're going to add all subfunctions now, if the corresponding feature
2016      * is available that unlocks the query functions.
2017      */
2018     s390_add_from_feat_block(features, S390_FEAT_TYPE_PLO, prop.plo);
2019     if (test_bit(S390_FEAT_TOD_CLOCK_STEERING, features)) {
2020         s390_add_from_feat_block(features, S390_FEAT_TYPE_PTFF, prop.ptff);
2021     }
2022     if (test_bit(S390_FEAT_MSA, features)) {
2023         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMAC, prop.kmac);
2024         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMC, prop.kmc);
2025         s390_add_from_feat_block(features, S390_FEAT_TYPE_KM, prop.km);
2026         s390_add_from_feat_block(features, S390_FEAT_TYPE_KIMD, prop.kimd);
2027         s390_add_from_feat_block(features, S390_FEAT_TYPE_KLMD, prop.klmd);
2028     }
2029     if (test_bit(S390_FEAT_MSA_EXT_3, features)) {
2030         s390_add_from_feat_block(features, S390_FEAT_TYPE_PCKMO, prop.pckmo);
2031     }
2032     if (test_bit(S390_FEAT_MSA_EXT_4, features)) {
2033         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMCTR, prop.kmctr);
2034         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMF, prop.kmf);
2035         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMO, prop.kmo);
2036         s390_add_from_feat_block(features, S390_FEAT_TYPE_PCC, prop.pcc);
2037     }
2038     if (test_bit(S390_FEAT_MSA_EXT_5, features)) {
2039         s390_add_from_feat_block(features, S390_FEAT_TYPE_PPNO, prop.ppno);
2040     }
2041     if (test_bit(S390_FEAT_MSA_EXT_8, features)) {
2042         s390_add_from_feat_block(features, S390_FEAT_TYPE_KMA, prop.kma);
2043     }
2044     return 0;
2045 }
2046 
2047 static int configure_cpu_subfunc(const S390FeatBitmap features)
2048 {
2049     struct kvm_s390_vm_cpu_subfunc prop = {};
2050     struct kvm_device_attr attr = {
2051         .group = KVM_S390_VM_CPU_MODEL,
2052         .attr = KVM_S390_VM_CPU_PROCESSOR_SUBFUNC,
2053         .addr = (uint64_t) &prop,
2054     };
2055 
2056     if (!kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2057                            KVM_S390_VM_CPU_PROCESSOR_SUBFUNC)) {
2058         /* hardware support might be missing, IBC will handle most of this */
2059         return 0;
2060     }
2061 
2062     s390_fill_feat_block(features, S390_FEAT_TYPE_PLO, prop.plo);
2063     if (test_bit(S390_FEAT_TOD_CLOCK_STEERING, features)) {
2064         s390_fill_feat_block(features, S390_FEAT_TYPE_PTFF, prop.ptff);
2065     }
2066     if (test_bit(S390_FEAT_MSA, features)) {
2067         s390_fill_feat_block(features, S390_FEAT_TYPE_KMAC, prop.kmac);
2068         s390_fill_feat_block(features, S390_FEAT_TYPE_KMC, prop.kmc);
2069         s390_fill_feat_block(features, S390_FEAT_TYPE_KM, prop.km);
2070         s390_fill_feat_block(features, S390_FEAT_TYPE_KIMD, prop.kimd);
2071         s390_fill_feat_block(features, S390_FEAT_TYPE_KLMD, prop.klmd);
2072     }
2073     if (test_bit(S390_FEAT_MSA_EXT_3, features)) {
2074         s390_fill_feat_block(features, S390_FEAT_TYPE_PCKMO, prop.pckmo);
2075     }
2076     if (test_bit(S390_FEAT_MSA_EXT_4, features)) {
2077         s390_fill_feat_block(features, S390_FEAT_TYPE_KMCTR, prop.kmctr);
2078         s390_fill_feat_block(features, S390_FEAT_TYPE_KMF, prop.kmf);
2079         s390_fill_feat_block(features, S390_FEAT_TYPE_KMO, prop.kmo);
2080         s390_fill_feat_block(features, S390_FEAT_TYPE_PCC, prop.pcc);
2081     }
2082     if (test_bit(S390_FEAT_MSA_EXT_5, features)) {
2083         s390_fill_feat_block(features, S390_FEAT_TYPE_PPNO, prop.ppno);
2084     }
2085     if (test_bit(S390_FEAT_MSA_EXT_8, features)) {
2086         s390_fill_feat_block(features, S390_FEAT_TYPE_KMA, prop.kma);
2087     }
2088     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2089 }
2090 
2091 static int kvm_to_feat[][2] = {
2092     { KVM_S390_VM_CPU_FEAT_ESOP, S390_FEAT_ESOP },
2093     { KVM_S390_VM_CPU_FEAT_SIEF2, S390_FEAT_SIE_F2 },
2094     { KVM_S390_VM_CPU_FEAT_64BSCAO , S390_FEAT_SIE_64BSCAO },
2095     { KVM_S390_VM_CPU_FEAT_SIIF, S390_FEAT_SIE_SIIF },
2096     { KVM_S390_VM_CPU_FEAT_GPERE, S390_FEAT_SIE_GPERE },
2097     { KVM_S390_VM_CPU_FEAT_GSLS, S390_FEAT_SIE_GSLS },
2098     { KVM_S390_VM_CPU_FEAT_IB, S390_FEAT_SIE_IB },
2099     { KVM_S390_VM_CPU_FEAT_CEI, S390_FEAT_SIE_CEI },
2100     { KVM_S390_VM_CPU_FEAT_IBS, S390_FEAT_SIE_IBS },
2101     { KVM_S390_VM_CPU_FEAT_SKEY, S390_FEAT_SIE_SKEY },
2102     { KVM_S390_VM_CPU_FEAT_CMMA, S390_FEAT_SIE_CMMA },
2103     { KVM_S390_VM_CPU_FEAT_PFMFI, S390_FEAT_SIE_PFMFI},
2104     { KVM_S390_VM_CPU_FEAT_SIGPIF, S390_FEAT_SIE_SIGPIF},
2105     { KVM_S390_VM_CPU_FEAT_KSS, S390_FEAT_SIE_KSS},
2106 };
2107 
2108 static int query_cpu_feat(S390FeatBitmap features)
2109 {
2110     struct kvm_s390_vm_cpu_feat prop;
2111     struct kvm_device_attr attr = {
2112         .group = KVM_S390_VM_CPU_MODEL,
2113         .attr = KVM_S390_VM_CPU_MACHINE_FEAT,
2114         .addr = (uint64_t) &prop,
2115     };
2116     int rc;
2117     int i;
2118 
2119     rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2120     if (rc) {
2121         return  rc;
2122     }
2123 
2124     for (i = 0; i < ARRAY_SIZE(kvm_to_feat); i++) {
2125         if (test_be_bit(kvm_to_feat[i][0], (uint8_t *) prop.feat)) {
2126             set_bit(kvm_to_feat[i][1], features);
2127         }
2128     }
2129     return 0;
2130 }
2131 
2132 static int configure_cpu_feat(const S390FeatBitmap features)
2133 {
2134     struct kvm_s390_vm_cpu_feat prop = {};
2135     struct kvm_device_attr attr = {
2136         .group = KVM_S390_VM_CPU_MODEL,
2137         .attr = KVM_S390_VM_CPU_PROCESSOR_FEAT,
2138         .addr = (uint64_t) &prop,
2139     };
2140     int i;
2141 
2142     for (i = 0; i < ARRAY_SIZE(kvm_to_feat); i++) {
2143         if (test_bit(kvm_to_feat[i][1], features)) {
2144             set_be_bit(kvm_to_feat[i][0], (uint8_t *) prop.feat);
2145         }
2146     }
2147     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2148 }
2149 
2150 bool kvm_s390_cpu_models_supported(void)
2151 {
2152     if (!cpu_model_allowed()) {
2153         /* compatibility machines interfere with the cpu model */
2154         return false;
2155     }
2156     return kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2157                              KVM_S390_VM_CPU_MACHINE) &&
2158            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2159                              KVM_S390_VM_CPU_PROCESSOR) &&
2160            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2161                              KVM_S390_VM_CPU_MACHINE_FEAT) &&
2162            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2163                              KVM_S390_VM_CPU_PROCESSOR_FEAT) &&
2164            kvm_vm_check_attr(kvm_state, KVM_S390_VM_CPU_MODEL,
2165                              KVM_S390_VM_CPU_MACHINE_SUBFUNC);
2166 }
2167 
2168 void kvm_s390_get_host_cpu_model(S390CPUModel *model, Error **errp)
2169 {
2170     struct kvm_s390_vm_cpu_machine prop = {};
2171     struct kvm_device_attr attr = {
2172         .group = KVM_S390_VM_CPU_MODEL,
2173         .attr = KVM_S390_VM_CPU_MACHINE,
2174         .addr = (uint64_t) &prop,
2175     };
2176     uint16_t unblocked_ibc = 0, cpu_type = 0;
2177     int rc;
2178 
2179     memset(model, 0, sizeof(*model));
2180 
2181     if (!kvm_s390_cpu_models_supported()) {
2182         error_setg(errp, "KVM doesn't support CPU models");
2183         return;
2184     }
2185 
2186     /* query the basic cpu model properties */
2187     rc = kvm_vm_ioctl(kvm_state, KVM_GET_DEVICE_ATTR, &attr);
2188     if (rc) {
2189         error_setg(errp, "KVM: Error querying host CPU model: %d", rc);
2190         return;
2191     }
2192 
2193     cpu_type = cpuid_type(prop.cpuid);
2194     if (has_ibc(prop.ibc)) {
2195         model->lowest_ibc = lowest_ibc(prop.ibc);
2196         unblocked_ibc = unblocked_ibc(prop.ibc);
2197     }
2198     model->cpu_id = cpuid_id(prop.cpuid);
2199     model->cpu_id_format = cpuid_format(prop.cpuid);
2200     model->cpu_ver = 0xff;
2201 
2202     /* get supported cpu features indicated via STFL(E) */
2203     s390_add_from_feat_block(model->features, S390_FEAT_TYPE_STFL,
2204                              (uint8_t *) prop.fac_mask);
2205     /* dat-enhancement facility 2 has no bit but was introduced with stfle */
2206     if (test_bit(S390_FEAT_STFLE, model->features)) {
2207         set_bit(S390_FEAT_DAT_ENH_2, model->features);
2208     }
2209     /* get supported cpu features indicated e.g. via SCLP */
2210     rc = query_cpu_feat(model->features);
2211     if (rc) {
2212         error_setg(errp, "KVM: Error querying CPU features: %d", rc);
2213         return;
2214     }
2215     /* get supported cpu subfunctions indicated via query / test bit */
2216     rc = query_cpu_subfunc(model->features);
2217     if (rc) {
2218         error_setg(errp, "KVM: Error querying CPU subfunctions: %d", rc);
2219         return;
2220     }
2221 
2222     /* PTFF subfunctions might be indicated although kernel support missing */
2223     if (!test_bit(S390_FEAT_MULTIPLE_EPOCH, model->features)) {
2224         clear_bit(S390_FEAT_PTFF_QSIE, model->features);
2225         clear_bit(S390_FEAT_PTFF_QTOUE, model->features);
2226         clear_bit(S390_FEAT_PTFF_STOE, model->features);
2227         clear_bit(S390_FEAT_PTFF_STOUE, model->features);
2228     }
2229 
2230     /* with cpu model support, CMM is only indicated if really available */
2231     if (kvm_s390_cmma_available()) {
2232         set_bit(S390_FEAT_CMM, model->features);
2233     } else {
2234         /* no cmm -> no cmm nt */
2235         clear_bit(S390_FEAT_CMM_NT, model->features);
2236     }
2237 
2238     /* bpb needs kernel support for migration, VSIE and reset */
2239     if (!kvm_check_extension(kvm_state, KVM_CAP_S390_BPB)) {
2240         clear_bit(S390_FEAT_BPB, model->features);
2241     }
2242 
2243     /* We emulate a zPCI bus and AEN, therefore we don't need HW support */
2244     if (pci_available) {
2245         set_bit(S390_FEAT_ZPCI, model->features);
2246     }
2247     set_bit(S390_FEAT_ADAPTER_EVENT_NOTIFICATION, model->features);
2248 
2249     if (s390_known_cpu_type(cpu_type)) {
2250         /* we want the exact model, even if some features are missing */
2251         model->def = s390_find_cpu_def(cpu_type, ibc_gen(unblocked_ibc),
2252                                        ibc_ec_ga(unblocked_ibc), NULL);
2253     } else {
2254         /* model unknown, e.g. too new - search using features */
2255         model->def = s390_find_cpu_def(0, ibc_gen(unblocked_ibc),
2256                                        ibc_ec_ga(unblocked_ibc),
2257                                        model->features);
2258     }
2259     if (!model->def) {
2260         error_setg(errp, "KVM: host CPU model could not be identified");
2261         return;
2262     }
2263     /* strip of features that are not part of the maximum model */
2264     bitmap_and(model->features, model->features, model->def->full_feat,
2265                S390_FEAT_MAX);
2266 }
2267 
2268 void kvm_s390_apply_cpu_model(const S390CPUModel *model, Error **errp)
2269 {
2270     struct kvm_s390_vm_cpu_processor prop  = {
2271         .fac_list = { 0 },
2272     };
2273     struct kvm_device_attr attr = {
2274         .group = KVM_S390_VM_CPU_MODEL,
2275         .attr = KVM_S390_VM_CPU_PROCESSOR,
2276         .addr = (uint64_t) &prop,
2277     };
2278     int rc;
2279 
2280     if (!model) {
2281         /* compatibility handling if cpu models are disabled */
2282         if (kvm_s390_cmma_available()) {
2283             kvm_s390_enable_cmma();
2284         }
2285         return;
2286     }
2287     if (!kvm_s390_cpu_models_supported()) {
2288         error_setg(errp, "KVM doesn't support CPU models");
2289         return;
2290     }
2291     prop.cpuid = s390_cpuid_from_cpu_model(model);
2292     prop.ibc = s390_ibc_from_cpu_model(model);
2293     /* configure cpu features indicated via STFL(e) */
2294     s390_fill_feat_block(model->features, S390_FEAT_TYPE_STFL,
2295                          (uint8_t *) prop.fac_list);
2296     rc = kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
2297     if (rc) {
2298         error_setg(errp, "KVM: Error configuring the CPU model: %d", rc);
2299         return;
2300     }
2301     /* configure cpu features indicated e.g. via SCLP */
2302     rc = configure_cpu_feat(model->features);
2303     if (rc) {
2304         error_setg(errp, "KVM: Error configuring CPU features: %d", rc);
2305         return;
2306     }
2307     /* configure cpu subfunctions indicated via query / test bit */
2308     rc = configure_cpu_subfunc(model->features);
2309     if (rc) {
2310         error_setg(errp, "KVM: Error configuring CPU subfunctions: %d", rc);
2311         return;
2312     }
2313     /* enable CMM via CMMA */
2314     if (test_bit(S390_FEAT_CMM, model->features)) {
2315         kvm_s390_enable_cmma();
2316     }
2317 }
2318 
2319 void kvm_s390_restart_interrupt(S390CPU *cpu)
2320 {
2321     struct kvm_s390_irq irq = {
2322         .type = KVM_S390_RESTART,
2323     };
2324 
2325     kvm_s390_vcpu_interrupt(cpu, &irq);
2326 }
2327 
2328 void kvm_s390_stop_interrupt(S390CPU *cpu)
2329 {
2330     struct kvm_s390_irq irq = {
2331         .type = KVM_S390_SIGP_STOP,
2332     };
2333 
2334     kvm_s390_vcpu_interrupt(cpu, &irq);
2335 }
2336